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>
88 #include <vm/vm_object.h>
90 struct mountscan_info {
91 TAILQ_ENTRY(mountscan_info) msi_entry;
93 struct mount *msi_node;
96 struct vmntvnodescan_info {
97 TAILQ_ENTRY(vmntvnodescan_info) entry;
105 static int vnlru_nowhere = 0;
106 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RD,
108 "Number of times the vnlru process ran without success");
111 static struct lwkt_token mntid_token;
112 static struct mount dummymount;
114 /* note: mountlist exported to pstat */
115 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
116 static TAILQ_HEAD(,mountscan_info) mountscan_list;
117 static struct lwkt_token mountlist_token;
119 static TAILQ_HEAD(,bio_ops) bio_ops_list = TAILQ_HEAD_INITIALIZER(bio_ops_list);
122 * Called from vfsinit()
127 lwkt_token_init(&mountlist_token, "mntlist");
128 lwkt_token_init(&mntid_token, "mntid");
129 TAILQ_INIT(&mountscan_list);
130 mount_init(&dummymount);
131 dummymount.mnt_flag |= MNT_RDONLY;
132 dummymount.mnt_kern_flag |= MNTK_ALL_MPSAFE;
136 * Support function called to remove a vnode from the mountlist and
137 * deal with side effects for scans in progress.
139 * Target mnt_token is held on call.
142 vremovevnodemnt(struct vnode *vp)
144 struct vmntvnodescan_info *info;
145 struct mount *mp = vp->v_mount;
147 TAILQ_FOREACH(info, &mp->mnt_vnodescan_list, entry) {
149 info->vp = TAILQ_NEXT(vp, v_nmntvnodes);
151 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
155 * Allocate a new vnode and associate it with a tag, mount point, and
158 * A VX locked and refd vnode is returned. The caller should setup the
159 * remaining fields and vx_put() or, if he wishes to leave a vref,
160 * vx_unlock() the vnode.
163 getnewvnode(enum vtagtype tag, struct mount *mp,
164 struct vnode **vpp, int lktimeout, int lkflags)
168 KKASSERT(mp != NULL);
170 vp = allocvnode(lktimeout, lkflags);
175 * By default the vnode is assigned the mount point's normal
178 vp->v_ops = &mp->mnt_vn_use_ops;
179 vp->v_pbuf_count = nswbuf_kva / NSWBUF_SPLIT;
182 * Placing the vnode on the mount point's queue makes it visible.
183 * VNON prevents it from being messed with, however.
188 * A VX locked & refd vnode is returned.
195 * This function creates vnodes with special operations vectors. The
196 * mount point is optional.
198 * This routine is being phased out but is still used by vfs_conf to
199 * create vnodes for devices prior to the root mount (with mp == NULL).
202 getspecialvnode(enum vtagtype tag, struct mount *mp,
203 struct vop_ops **ops,
204 struct vnode **vpp, int lktimeout, int lkflags)
208 vp = allocvnode(lktimeout, lkflags);
217 * Placing the vnode on the mount point's queue makes it visible.
218 * VNON prevents it from being messed with, however.
223 * A VX locked & refd vnode is returned.
230 * Interlock against an unmount, return 0 on success, non-zero on failure.
232 * The passed flag may be 0 or LK_NOWAIT and is only used if an unmount
235 * If no unmount is in-progress LK_NOWAIT is ignored. No other flag bits
236 * are used. A shared locked will be obtained and the filesystem will not
237 * be unmountable until the lock is released.
240 vfs_busy(struct mount *mp, int flags)
244 atomic_add_int(&mp->mnt_refs, 1);
245 lwkt_gettoken(&mp->mnt_token);
246 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
247 if (flags & LK_NOWAIT) {
248 lwkt_reltoken(&mp->mnt_token);
249 atomic_add_int(&mp->mnt_refs, -1);
252 /* XXX not MP safe */
253 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 * WARNING! mp can potentially go away once we release
264 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
265 lwkt_reltoken(&mp->mnt_token);
266 atomic_add_int(&mp->mnt_refs, -1);
270 if (lockmgr(&mp->mnt_lock, lkflags))
271 panic("vfs_busy: unexpected lock failure");
272 lwkt_reltoken(&mp->mnt_token);
277 * Free a busy filesystem.
279 * Once refs is decremented the mount point can potentially get ripped
280 * out from under us, but we want to clean up our refs before unlocking
281 * so do a hold/drop around the whole mess.
283 * This is not in the critical path (I hope).
286 vfs_unbusy(struct mount *mp)
289 atomic_add_int(&mp->mnt_refs, -1);
290 lockmgr(&mp->mnt_lock, LK_RELEASE);
295 * Lookup a filesystem type, and if found allocate and initialize
296 * a mount structure for it.
298 * Devname is usually updated by mount(8) after booting.
301 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
303 struct vfsconf *vfsp;
306 if (fstypename == NULL)
309 vfsp = vfsconf_find_by_name(fstypename);
312 mp = kmalloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
314 lockinit(&mp->mnt_lock, "vfslock", VLKTIMEOUT, 0);
318 mp->mnt_op = vfsp->vfc_vfsops;
319 mp->mnt_pbuf_count = nswbuf_kva / NSWBUF_SPLIT;
320 vfsp->vfc_refcount++;
321 mp->mnt_stat.f_type = vfsp->vfc_typenum;
322 mp->mnt_flag |= MNT_RDONLY;
323 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
324 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
325 copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
328 * Pre-set MPSAFE flags for VFS_MOUNT() call.
330 if (vfsp->vfc_flags & VFCF_MPSAFE)
331 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE;
339 * Basic mount structure initialization
342 mount_init(struct mount *mp)
344 lockinit(&mp->mnt_lock, "vfslock", hz*5, 0);
345 lwkt_token_init(&mp->mnt_token, "permnt");
347 TAILQ_INIT(&mp->mnt_vnodescan_list);
348 TAILQ_INIT(&mp->mnt_nvnodelist);
349 TAILQ_INIT(&mp->mnt_reservedvnlist);
350 TAILQ_INIT(&mp->mnt_jlist);
351 mp->mnt_nvnodelistsize = 0;
353 mp->mnt_hold = 1; /* hold for umount last drop */
354 mp->mnt_iosize_max = MAXPHYS;
355 vn_syncer_thr_create(mp);
359 mount_hold(struct mount *mp)
361 atomic_add_int(&mp->mnt_hold, 1);
365 mount_drop(struct mount *mp)
367 if (atomic_fetchadd_int(&mp->mnt_hold, -1) == 1) {
368 KKASSERT(mp->mnt_refs == 0);
374 * Lookup a mount point by filesystem identifier.
376 * If not NULL, the returned mp is held and the caller is expected to drop
377 * it via mount_drop().
380 vfs_getvfs(fsid_t *fsid)
384 lwkt_gettoken_shared(&mountlist_token);
385 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
386 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
387 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
392 lwkt_reltoken(&mountlist_token);
397 * Get a new unique fsid. Try to make its val[0] unique, since this value
398 * will be used to create fake device numbers for stat(). Also try (but
399 * not so hard) make its val[0] unique mod 2^16, since some emulators only
400 * support 16-bit device numbers. We end up with unique val[0]'s for the
401 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
403 * Keep in mind that several mounts may be running in parallel. Starting
404 * the search one past where the previous search terminated is both a
405 * micro-optimization and a defense against returning the same fsid to
409 vfs_getnewfsid(struct mount *mp)
411 static u_int16_t mntid_base;
416 lwkt_gettoken(&mntid_token);
417 mtype = mp->mnt_vfc->vfc_typenum;
418 tfsid.val[1] = mtype;
419 mtype = (mtype & 0xFF) << 24;
421 tfsid.val[0] = makeudev(255,
422 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
424 mptmp = vfs_getvfs(&tfsid);
429 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
430 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
431 lwkt_reltoken(&mntid_token);
435 * Set the FSID for a new mount point to the template. Adjust
436 * the FSID to avoid collisions.
439 vfs_setfsid(struct mount *mp, fsid_t *template)
444 bzero(&mp->mnt_stat.f_fsid, sizeof(mp->mnt_stat.f_fsid));
446 lwkt_gettoken(&mntid_token);
448 mptmp = vfs_getvfs(template);
455 mp->mnt_stat.f_fsid = *template;
456 lwkt_reltoken(&mntid_token);
462 * This routine is called when we have too many vnodes. It attempts
463 * to free <count> vnodes and will potentially free vnodes that still
464 * have VM backing store (VM backing store is typically the cause
465 * of a vnode blowout so we want to do this). Therefore, this operation
466 * is not considered cheap.
468 * A number of conditions may prevent a vnode from being reclaimed.
469 * the buffer cache may have references on the vnode, a directory
470 * vnode may still have references due to the namei cache representing
471 * underlying files, or the vnode may be in active use. It is not
472 * desireable to reuse such vnodes. These conditions may cause the
473 * number of vnodes to reach some minimum value regardless of what
474 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
478 * Attempt to recycle vnodes in a context that is always safe to block.
479 * Calling vlrurecycle() from the bowels of file system code has some
480 * interesting deadlock problems.
482 static struct thread *vnlruthread;
487 struct thread *td = curthread;
489 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
493 int ncachedandinactive;
495 kproc_suspend_loop();
498 * Try to free some vnodes if we have too many. Trigger based
499 * on potentially freeable vnodes but calculate the count
500 * based on total vnodes.
502 * (long) -> deal with 64 bit machines, intermediate overflow
504 synchronizevnodecount();
505 ncachedandinactive = countcachedandinactivevnodes();
506 if (numvnodes >= maxvnodes * 9 / 10 &&
507 ncachedandinactive >= maxvnodes * 5 / 10) {
508 int count = numvnodes - maxvnodes * 9 / 10;
510 if (count > (ncachedandinactive) / 100)
511 count = (ncachedandinactive) / 100;
514 freesomevnodes(count);
518 * Do non-critical-path (more robust) cache cleaning,
519 * even if vnode counts are nominal, to try to avoid
520 * having to do it in the critical path.
525 * Nothing to do if most of our vnodes are already on
528 synchronizevnodecount();
529 ncachedandinactive = countcachedandinactivevnodes();
530 if (numvnodes <= maxvnodes * 9 / 10 ||
531 ncachedandinactive <= maxvnodes * 5 / 10) {
532 tsleep(vnlruthread, 0, "vlruwt", hz);
539 * MOUNTLIST FUNCTIONS
543 * mountlist_insert (MP SAFE)
545 * Add a new mount point to the mount list.
548 mountlist_insert(struct mount *mp, int how)
550 lwkt_gettoken(&mountlist_token);
551 if (how == MNTINS_FIRST)
552 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
554 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
555 lwkt_reltoken(&mountlist_token);
559 * mountlist_interlock (MP SAFE)
561 * Execute the specified interlock function with the mountlist token
562 * held. The function will be called in a serialized fashion verses
563 * other functions called through this mechanism.
565 * The function is expected to be very short-lived.
568 mountlist_interlock(int (*callback)(struct mount *), struct mount *mp)
572 lwkt_gettoken(&mountlist_token);
573 error = callback(mp);
574 lwkt_reltoken(&mountlist_token);
579 * mountlist_boot_getfirst (DURING BOOT ONLY)
581 * This function returns the first mount on the mountlist, which is
582 * expected to be the root mount. Since no interlocks are obtained
583 * this function is only safe to use during booting.
587 mountlist_boot_getfirst(void)
589 return(TAILQ_FIRST(&mountlist));
593 * mountlist_remove (MP SAFE)
595 * Remove a node from the mountlist. If this node is the next scan node
596 * for any active mountlist scans, the active mountlist scan will be
597 * adjusted to skip the node, thus allowing removals during mountlist
601 mountlist_remove(struct mount *mp)
603 struct mountscan_info *msi;
605 lwkt_gettoken(&mountlist_token);
606 TAILQ_FOREACH(msi, &mountscan_list, msi_entry) {
607 if (msi->msi_node == mp) {
608 if (msi->msi_how & MNTSCAN_FORWARD)
609 msi->msi_node = TAILQ_NEXT(mp, mnt_list);
611 msi->msi_node = TAILQ_PREV(mp, mntlist,
615 TAILQ_REMOVE(&mountlist, mp, mnt_list);
616 lwkt_reltoken(&mountlist_token);
620 * mountlist_exists (MP SAFE)
622 * Checks if a node exists in the mountlist.
623 * This function is mainly used by VFS quota code to check if a
624 * cached nullfs struct mount pointer is still valid at use time
626 * FIXME: there is no warranty the mp passed to that function
627 * will be the same one used by VFS_ACCOUNT() later
630 mountlist_exists(struct mount *mp)
635 lwkt_gettoken_shared(&mountlist_token);
636 TAILQ_FOREACH(lmp, &mountlist, mnt_list) {
642 lwkt_reltoken(&mountlist_token);
650 * Safely scan the mount points on the mount list. Each mountpoint
651 * is held across the callback. The callback is responsible for
652 * acquiring any further tokens or locks.
654 * Unless otherwise specified each mount point will be busied prior to the
655 * callback and unbusied afterwords. The callback may safely remove any
656 * mount point without interfering with the scan. If the current callback
657 * mount is removed the scanner will not attempt to unbusy it.
659 * If a mount node cannot be busied it is silently skipped.
661 * The callback return value is aggregated and a total is returned. A return
662 * value of < 0 is not aggregated and will terminate the scan.
664 * MNTSCAN_FORWARD - the mountlist is scanned in the forward direction
665 * MNTSCAN_REVERSE - the mountlist is scanned in reverse
666 * MNTSCAN_NOBUSY - the scanner will make the callback without busying
669 * NOTE: mountlist_token is not held across the callback.
672 mountlist_scan(int (*callback)(struct mount *, void *), void *data, int how)
674 struct mountscan_info info;
679 lwkt_gettoken(&mountlist_token);
681 info.msi_node = NULL; /* paranoia */
682 TAILQ_INSERT_TAIL(&mountscan_list, &info, msi_entry);
683 lwkt_reltoken(&mountlist_token);
686 lwkt_gettoken_shared(&mountlist_token);
688 if (how & MNTSCAN_FORWARD) {
689 info.msi_node = TAILQ_FIRST(&mountlist);
690 while ((mp = info.msi_node) != NULL) {
692 if (how & MNTSCAN_NOBUSY) {
693 lwkt_reltoken(&mountlist_token);
694 count = callback(mp, data);
695 lwkt_gettoken_shared(&mountlist_token);
696 } else if (vfs_busy(mp, LK_NOWAIT) == 0) {
697 lwkt_reltoken(&mountlist_token);
698 count = callback(mp, data);
699 lwkt_gettoken_shared(&mountlist_token);
700 if (mp == info.msi_node)
709 if (mp == info.msi_node)
710 info.msi_node = TAILQ_NEXT(mp, mnt_list);
712 } else if (how & MNTSCAN_REVERSE) {
713 info.msi_node = TAILQ_LAST(&mountlist, mntlist);
714 while ((mp = info.msi_node) != NULL) {
716 if (how & MNTSCAN_NOBUSY) {
717 lwkt_reltoken(&mountlist_token);
718 count = callback(mp, data);
719 lwkt_gettoken_shared(&mountlist_token);
720 } else if (vfs_busy(mp, LK_NOWAIT) == 0) {
721 lwkt_reltoken(&mountlist_token);
722 count = callback(mp, data);
723 lwkt_gettoken_shared(&mountlist_token);
724 if (mp == info.msi_node)
733 if (mp == info.msi_node)
734 info.msi_node = TAILQ_PREV(mp, mntlist,
738 lwkt_reltoken(&mountlist_token);
740 lwkt_gettoken(&mountlist_token);
741 TAILQ_REMOVE(&mountscan_list, &info, msi_entry);
742 lwkt_reltoken(&mountlist_token);
748 * MOUNT RELATED VNODE FUNCTIONS
751 static struct kproc_desc vnlru_kp = {
756 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp);
759 * Move a vnode from one mount queue to another.
762 insmntque(struct vnode *vp, struct mount *mp)
767 * Delete from old mount point vnode list, if on one.
769 if ((omp = vp->v_mount) != NULL) {
770 lwkt_gettoken(&omp->mnt_token);
771 KKASSERT(omp == vp->v_mount);
772 KASSERT(omp->mnt_nvnodelistsize > 0,
773 ("bad mount point vnode list size"));
775 omp->mnt_nvnodelistsize--;
776 lwkt_reltoken(&omp->mnt_token);
780 * Insert into list of vnodes for the new mount point, if available.
781 * The 'end' of the LRU list is the vnode prior to mp->mnt_syncer.
787 lwkt_gettoken(&mp->mnt_token);
789 if (mp->mnt_syncer) {
790 TAILQ_INSERT_BEFORE(mp->mnt_syncer, vp, v_nmntvnodes);
792 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
794 mp->mnt_nvnodelistsize++;
795 lwkt_reltoken(&mp->mnt_token);
800 * Scan the vnodes under a mount point and issue appropriate callbacks.
802 * The fastfunc() callback is called with just the mountlist token held
803 * (no vnode lock). It may not block and the vnode may be undergoing
804 * modifications while the caller is processing it. The vnode will
805 * not be entirely destroyed, however, due to the fact that the mountlist
806 * token is held. A return value < 0 skips to the next vnode without calling
807 * the slowfunc(), a return value > 0 terminates the loop.
809 * WARNING! The fastfunc() should not indirect through vp->v_object, the vp
810 * data structure is unstable when called from fastfunc().
812 * The slowfunc() callback is called after the vnode has been successfully
813 * locked based on passed flags. The vnode is skipped if it gets rearranged
814 * or destroyed while blocking on the lock. A non-zero return value from
815 * the slow function terminates the loop. The slow function is allowed to
816 * arbitrarily block. The scanning code guarentees consistency of operation
817 * even if the slow function deletes or moves the node, or blocks and some
818 * other thread deletes or moves the node.
824 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
825 int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data),
828 struct vmntvnodescan_info info;
831 int maxcount = mp->mnt_nvnodelistsize * 2;
835 lwkt_gettoken(&mp->mnt_token);
838 * If asked to do one pass stop after iterating available vnodes.
839 * Under heavy loads new vnodes can be added while we are scanning,
840 * so this isn't perfect. Create a slop factor of 2x.
842 if (flags & VMSC_ONEPASS)
843 stopcount = mp->mnt_nvnodelistsize;
845 info.vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
846 TAILQ_INSERT_TAIL(&mp->mnt_vnodescan_list, &info, entry);
848 while ((vp = info.vp) != NULL) {
849 if (--maxcount == 0) {
850 kprintf("Warning: excessive fssync iteration\n");
851 maxcount = mp->mnt_nvnodelistsize * 2;
855 * Skip if visible but not ready, or special (e.g.
858 if (vp->v_type == VNON)
860 KKASSERT(vp->v_mount == mp);
863 * Quick test. A negative return continues the loop without
864 * calling the slow test. 0 continues onto the slow test.
865 * A positive number aborts the loop.
868 if ((r = fastfunc(mp, vp, data)) < 0) {
877 * Get a vxlock on the vnode, retry if it has moved or isn't
878 * in the mountlist where we expect it.
883 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) {
885 error = vget(vp, LK_EXCLUSIVE);
887 case VMSC_GETVP|VMSC_NOWAIT:
888 error = vget(vp, LK_EXCLUSIVE|LK_NOWAIT);
901 * Do not call the slow function if the vnode is
902 * invalid or if it was ripped out from under us
903 * while we (potentially) blocked.
905 if (info.vp == vp && vp->v_type != VNON)
906 r = slowfunc(mp, vp, data);
911 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) {
913 case VMSC_GETVP|VMSC_NOWAIT:
928 * Yield after some processing. Depending on the number
929 * of vnodes, we might wind up running for a long time.
930 * Because threads are not preemptable, time critical
931 * userland processes might starve. Give them a chance
934 if (++count == 10000) {
936 * We really want to yield a bit, so we simply
939 tsleep(mp, 0, "vnodescn", 1);
944 * If doing one pass this decrements to zero. If it starts
945 * at zero it is effectively unlimited for the purposes of
948 if (--stopcount == 0)
952 * Iterate. If the vnode was ripped out from under us
953 * info.vp will already point to the next vnode, otherwise
954 * we have to obtain the next valid vnode ourselves.
957 info.vp = TAILQ_NEXT(vp, v_nmntvnodes);
960 TAILQ_REMOVE(&mp->mnt_vnodescan_list, &info, entry);
961 lwkt_reltoken(&mp->mnt_token);
966 * Remove any vnodes in the vnode table belonging to mount point mp.
968 * If FORCECLOSE is not specified, there should not be any active ones,
969 * return error if any are found (nb: this is a user error, not a
970 * system error). If FORCECLOSE is specified, detach any active vnodes
973 * If WRITECLOSE is set, only flush out regular file vnodes open for
976 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
978 * `rootrefs' specifies the base reference count for the root vnode
979 * of this filesystem. The root vnode is considered busy if its
980 * v_refcnt exceeds this value. On a successful return, vflush()
981 * will call vrele() on the root vnode exactly rootrefs times.
982 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
985 static int debug_busyprt = 0; /* print out busy vnodes */
986 SYSCTL_INT(_vfs, OID_AUTO, debug_busyprt, CTLFLAG_RW, &debug_busyprt, 0, "");
988 static int vflush_scan(struct mount *mp, struct vnode *vp, void *data);
997 vflush(struct mount *mp, int rootrefs, int flags)
999 struct thread *td = curthread; /* XXX */
1000 struct vnode *rootvp = NULL;
1002 struct vflush_info vflush_info;
1005 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1006 ("vflush: bad args"));
1008 * Get the filesystem root vnode. We can vput() it
1009 * immediately, since with rootrefs > 0, it won't go away.
1011 if ((error = VFS_ROOT(mp, &rootvp)) != 0) {
1012 if ((flags & FORCECLOSE) == 0)
1015 /* continue anyway */
1021 vflush_info.busy = 0;
1022 vflush_info.flags = flags;
1023 vflush_info.td = td;
1024 vmntvnodescan(mp, VMSC_GETVX, NULL, vflush_scan, &vflush_info);
1026 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1028 * If just the root vnode is busy, and if its refcount
1029 * is equal to `rootrefs', then go ahead and kill it.
1031 KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
1032 KASSERT(VREFCNT(rootvp) >= rootrefs, ("vflush: rootrefs"));
1033 if (vflush_info.busy == 1 && VREFCNT(rootvp) == rootrefs) {
1035 vgone_vxlocked(rootvp);
1037 vflush_info.busy = 0;
1040 if (vflush_info.busy)
1042 for (; rootrefs > 0; rootrefs--)
1048 * The scan callback is made with an VX locked vnode.
1051 vflush_scan(struct mount *mp, struct vnode *vp, void *data)
1053 struct vflush_info *info = data;
1055 int flags = info->flags;
1058 * Generally speaking try to deactivate on 0 refs (catch-all)
1060 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE);
1063 * Skip over a vnodes marked VSYSTEM.
1065 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1070 * Do not force-close VCHR or VBLK vnodes
1072 if (vp->v_type == VCHR || vp->v_type == VBLK)
1073 flags &= ~(WRITECLOSE|FORCECLOSE);
1076 * If WRITECLOSE is set, flush out unlinked but still open
1077 * files (even if open only for reading) and regular file
1078 * vnodes open for writing.
1080 if ((flags & WRITECLOSE) &&
1081 (vp->v_type == VNON ||
1082 (VOP_GETATTR(vp, &vattr) == 0 &&
1083 vattr.va_nlink > 0)) &&
1084 (vp->v_writecount == 0 || vp->v_type != VREG)) {
1089 * If we are the only holder (refcnt of 1) or the vnode is in
1090 * termination (refcnt < 0), we can vgone the vnode.
1092 if (VREFCNT(vp) <= 1) {
1098 * If FORCECLOSE is set, forcibly destroy the vnode and then move
1099 * it to a dummymount structure so vop_*() functions don't deref
1102 if (flags & FORCECLOSE) {
1105 if (vp->v_mount == NULL)
1106 insmntque(vp, &dummymount);
1110 if (vp->v_type == VCHR || vp->v_type == VBLK)
1111 kprintf("vflush: Warning, cannot destroy busy device vnode\n");
1112 if (debug_busyprt) {
1113 const char *filename;
1115 spin_lock(&vp->v_spin);
1116 filename = TAILQ_FIRST(&vp->v_namecache) ?
1117 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?";
1118 spin_unlock(&vp->v_spin);
1119 kprintf("vflush: busy vnode (%p) %s\n", vp, filename);
1126 add_bio_ops(struct bio_ops *ops)
1128 TAILQ_INSERT_TAIL(&bio_ops_list, ops, entry);
1132 rem_bio_ops(struct bio_ops *ops)
1134 TAILQ_REMOVE(&bio_ops_list, ops, entry);
1138 * This calls the bio_ops io_sync function either for a mount point
1141 * WARNING: softdeps is weirdly coded and just isn't happy unless
1142 * io_sync is called with a NULL mount from the general syncing code.
1145 bio_ops_sync(struct mount *mp)
1147 struct bio_ops *ops;
1150 if ((ops = mp->mnt_bioops) != NULL)
1153 TAILQ_FOREACH(ops, &bio_ops_list, entry) {
1160 * Lookup a mount point by nch
1163 mount_get_by_nc(struct namecache *ncp)
1165 struct mount *mp = NULL;
1167 lwkt_gettoken_shared(&mountlist_token);
1168 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1169 if (ncp == mp->mnt_ncmountpt.ncp)
1172 lwkt_reltoken(&mountlist_token);