2 * Copyright (c) 2004 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. All advertising materials mentioning features or use of this software
51 * must display the following acknowledgement:
52 * This product includes software developed by the University of
53 * California, Berkeley and its contributors.
54 * 4. Neither the name of the University nor the names of its contributors
55 * may be used to endorse or promote products derived from this software
56 * without specific prior written permission.
58 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
59 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
60 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
61 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
62 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
63 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
64 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
65 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
66 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
67 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
70 * $DragonFly: src/sys/kern/vfs_mount.c,v 1.2 2004/10/19 05:55:34 dillon Exp $
74 * External virtual filesystem routines
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/kernel.h>
81 #include <sys/malloc.h>
82 #include <sys/mount.h>
84 #include <sys/vnode.h>
86 #include <sys/eventhandler.h>
87 #include <sys/kthread.h>
88 #include <sys/sysctl.h>
90 #include <machine/limits.h>
93 #include <sys/thread2.h>
96 #include <vm/vm_object.h>
98 static int vnlru_nowhere = 0;
99 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
101 "Number of times the vnlru process ran without success");
104 static struct lwkt_token mntid_token;
106 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); /* mounted fs */
107 struct lwkt_token mountlist_token;
108 struct lwkt_token mntvnode_token;
112 * Called from vfsinit()
117 lwkt_token_init(&mountlist_token);
118 lwkt_token_init(&mntvnode_token);
119 lwkt_token_init(&mntid_token);
123 * Allocate a new vnode and associate it with a tag, mount point, and
126 * A VX locked and refd vnode is returned. The caller should setup the
127 * remaining fields and vx_put() or, if he wishes to leave a vref,
128 * vx_unlock() the vnode.
131 getnewvnode(enum vtagtype tag, struct mount *mp, struct vop_ops *ops,
132 struct vnode **vpp, int lktimeout, int lkflags)
136 vp = allocvnode(lktimeout, lkflags);
142 * Placing the vnode on the mount point's queue makes it visible.
143 * VNON prevents it from being messed with, however.
146 vfs_object_create(vp, curthread);
149 * A VX locked & refd vnode is returned.
156 * Mark a mount point as busy. Used to synchronize access and to delay
157 * unmounting. Interlock is not released on failure.
160 vfs_busy(struct mount *mp, int flags,
161 lwkt_tokref_t interlkp, struct thread *td)
165 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
166 if (flags & LK_NOWAIT)
168 mp->mnt_kern_flag |= MNTK_MWAIT;
170 * Since all busy locks are shared except the exclusive
171 * lock granted when unmounting, the only place that a
172 * wakeup needs to be done is at the release of the
173 * exclusive lock at the end of dounmount.
175 * note: interlkp is a serializer and thus can be safely
176 * held through any sleep
178 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
181 lkflags = LK_SHARED | LK_NOPAUSE;
183 lkflags |= LK_INTERLOCK;
184 if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
185 panic("vfs_busy: unexpected lock failure");
190 * Free a busy filesystem.
193 vfs_unbusy(struct mount *mp, struct thread *td)
195 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
199 * Lookup a filesystem type, and if found allocate and initialize
200 * a mount structure for it.
202 * Devname is usually updated by mount(8) after booting.
205 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
207 struct thread *td = curthread; /* XXX */
208 struct vfsconf *vfsp;
211 if (fstypename == NULL)
213 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
214 if (!strcmp(vfsp->vfc_name, fstypename))
219 mp = malloc((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK);
220 bzero((char *)mp, (u_long)sizeof(struct mount));
221 lockinit(&mp->mnt_lock, 0, "vfslock", VLKTIMEOUT, LK_NOPAUSE);
222 vfs_busy(mp, LK_NOWAIT, NULL, td);
223 TAILQ_INIT(&mp->mnt_nvnodelist);
224 TAILQ_INIT(&mp->mnt_reservedvnlist);
225 mp->mnt_nvnodelistsize = 0;
227 mp->mnt_op = vfsp->vfc_vfsops;
228 mp->mnt_flag = MNT_RDONLY;
229 mp->mnt_vnodecovered = NULLVP;
230 vfsp->vfc_refcount++;
231 mp->mnt_iosize_max = DFLTPHYS;
232 mp->mnt_stat.f_type = vfsp->vfc_typenum;
233 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
234 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
235 mp->mnt_stat.f_mntonname[0] = '/';
236 mp->mnt_stat.f_mntonname[1] = 0;
237 (void) copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
243 * Lookup a mount point by filesystem identifier.
246 vfs_getvfs(fsid_t *fsid)
251 lwkt_gettoken(&ilock, &mountlist_token);
252 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
253 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
254 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
258 lwkt_reltoken(&ilock);
263 * Get a new unique fsid. Try to make its val[0] unique, since this value
264 * will be used to create fake device numbers for stat(). Also try (but
265 * not so hard) make its val[0] unique mod 2^16, since some emulators only
266 * support 16-bit device numbers. We end up with unique val[0]'s for the
267 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
269 * Keep in mind that several mounts may be running in parallel. Starting
270 * the search one past where the previous search terminated is both a
271 * micro-optimization and a defense against returning the same fsid to
275 vfs_getnewfsid(struct mount *mp)
277 static u_int16_t mntid_base;
282 lwkt_gettoken(&ilock, &mntid_token);
283 mtype = mp->mnt_vfc->vfc_typenum;
284 tfsid.val[1] = mtype;
285 mtype = (mtype & 0xFF) << 24;
287 tfsid.val[0] = makeudev(255,
288 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
290 if (vfs_getvfs(&tfsid) == NULL)
293 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
294 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
295 lwkt_reltoken(&ilock);
299 * This routine is called when we have too many vnodes. It attempts
300 * to free <count> vnodes and will potentially free vnodes that still
301 * have VM backing store (VM backing store is typically the cause
302 * of a vnode blowout so we want to do this). Therefore, this operation
303 * is not considered cheap.
305 * A number of conditions may prevent a vnode from being reclaimed.
306 * the buffer cache may have references on the vnode, a directory
307 * vnode may still have references due to the namei cache representing
308 * underlying files, or the vnode may be in active use. It is not
309 * desireable to reuse such vnodes. These conditions may cause the
310 * number of vnodes to reach some minimum value regardless of what
311 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
315 * Return 0 if the vnode is not already on the free list, return 1 if the
316 * vnode, with some additional work could possibly be placed on the free list.
319 vmightfree(struct vnode *vp, int use_count, int page_count)
321 if (vp->v_flag & VFREE)
323 if (vp->v_usecount != use_count || vp->v_holdcnt)
325 if (vp->v_object && vp->v_object->resident_page_count >= page_count)
332 vlrureclaim(struct mount *mp)
342 * Calculate the trigger point, don't allow user
343 * screwups to blow us up. This prevents us from
344 * recycling vnodes with lots of resident pages. We
345 * aren't trying to free memory, we are trying to
348 usevnodes = desiredvnodes;
351 trigger = vmstats.v_page_count * 2 / usevnodes;
354 lwkt_gettoken(&ilock, &mntvnode_token);
355 count = mp->mnt_nvnodelistsize / 10 + 1;
356 while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
360 * The VP will stick around while we hold mntvnode_token,
361 * at least until we block, so we can safely do an initial
362 * check, and then must check again after we lock the vnode.
364 if (vp->v_type == VNON || /* XXX */
365 vp->v_type == VBAD || /* XXX */
366 !vmightfree(vp, 0, trigger) /* critical path opt */
368 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
369 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
375 * VX get the candidate vnode. If the VX get fails the
376 * vnode might still be on the mountlist. Our loop depends
377 * on us at least cycling the vnode to the end of the
380 if (vx_get_nonblock(vp) != 0) {
381 if (vp->v_mount == mp) {
382 TAILQ_REMOVE(&mp->mnt_nvnodelist,
384 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,
392 * Since we blocked locking the vp, make sure it is still
393 * a candidate for reclamation. That is, it has not already
394 * been reclaimed and only has our VX reference associated
397 if (vp->v_type == VNON || /* XXX */
398 vp->v_type == VBAD || /* XXX */
399 (vp->v_flag & VRECLAIMED) ||
401 !vmightfree(vp, 1, trigger) /* critical path opt */
403 if (vp->v_mount == mp) {
404 TAILQ_REMOVE(&mp->mnt_nvnodelist,
406 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,
415 * All right, we are good, move the vp to the end of the
416 * mountlist and clean it out. The vget will have returned
417 * an error if the vnode was destroyed (VRECLAIMED set), so we
418 * do not have to check again. The vput() will move the
419 * vnode to the free list if the vgone() was successful.
421 KKASSERT(vp->v_mount == mp);
422 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
423 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist,vp, v_nmntvnodes);
429 lwkt_reltoken(&ilock);
434 * Attempt to recycle vnodes in a context that is always safe to block.
435 * Calling vlrurecycle() from the bowels of file system code has some
436 * interesting deadlock problems.
438 static struct thread *vnlruthread;
439 static int vnlruproc_sig;
442 vnlru_proc_wait(void)
444 if (vnlruproc_sig == 0) {
445 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
448 tsleep(&vnlruproc_sig, 0, "vlruwk", hz);
454 struct mount *mp, *nmp;
458 struct thread *td = curthread;
460 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
465 kproc_suspend_loop();
466 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
468 wakeup(&vnlruproc_sig);
469 tsleep(td, 0, "vlruwt", hz);
474 lwkt_gettoken(&ilock, &mountlist_token);
475 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
476 if (vfs_busy(mp, LK_NOWAIT, &ilock, td)) {
477 nmp = TAILQ_NEXT(mp, mnt_list);
480 done += vlrureclaim(mp);
481 lwkt_gettokref(&ilock);
482 nmp = TAILQ_NEXT(mp, mnt_list);
485 lwkt_reltoken(&ilock);
488 tsleep(td, 0, "vlrup", hz * 3);
489 if (vnlru_nowhere % 10 == 0)
490 printf("vnlru_proc: vnode recycler stopped working!\n");
498 static struct kproc_desc vnlru_kp = {
503 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
506 * Move a vnode from one mount queue to another.
509 insmntque(struct vnode *vp, struct mount *mp)
513 lwkt_gettoken(&ilock, &mntvnode_token);
515 * Delete from old mount point vnode list, if on one.
517 if (vp->v_mount != NULL) {
518 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
519 ("bad mount point vnode list size"));
520 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
521 vp->v_mount->mnt_nvnodelistsize--;
524 * Insert into list of vnodes for the new mount point, if available.
526 if ((vp->v_mount = mp) == NULL) {
527 lwkt_reltoken(&ilock);
530 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
531 mp->mnt_nvnodelistsize++;
532 lwkt_reltoken(&ilock);
537 * Scan the vnodes under a mount point. The first function is called
538 * with just the mountlist token held (no vnode lock). The second
539 * function is called with the vnode VX locked.
545 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
546 int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data),
555 * Scan the vnodes on the mount's vnode list. Use a placemarker
557 pvp = allocvnode_placemarker();
559 lwkt_gettoken(&ilock, &mntvnode_token);
560 TAILQ_INSERT_HEAD(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
562 while ((vp = TAILQ_NEXT(pvp, v_nmntvnodes)) != NULL) {
564 * Move the placemarker and skip other placemarkers we
565 * encounter. The nothing can get in our way so the
566 * mount point on the vp must be valid.
568 TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
569 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, pvp, v_nmntvnodes);
570 if (vp->v_flag & VPLACEMARKER) /* another procs placemarker */
572 if (vp->v_type == VNON) /* visible but not ready */
574 KKASSERT(vp->v_mount == mp);
577 * Quick test. A negative return continues the loop without
578 * calling the slow test. 0 continues onto the slow test.
579 * A positive number aborts the loop.
582 if ((r = fastfunc(mp, vp, data)) < 0)
589 * Get a vxlock on the vnode, retry if it has moved or isn't
590 * in the mountlist where we expect it.
597 error = vget(vp, LK_EXCLUSIVE, curthread);
599 case VMSC_GETVP|VMSC_NOWAIT:
600 error = vget(vp, LK_EXCLUSIVE|LK_NOWAIT,
615 if (TAILQ_PREV(pvp, vnodelst, v_nmntvnodes) != vp)
617 if (vp->v_type == VNON)
619 r = slowfunc(mp, vp, data);
623 case VMSC_GETVP|VMSC_NOWAIT:
639 TAILQ_REMOVE(&mp->mnt_nvnodelist, pvp, v_nmntvnodes);
640 freevnode_placemarker(pvp);
641 lwkt_reltoken(&ilock);
646 * Remove any vnodes in the vnode table belonging to mount point mp.
648 * If FORCECLOSE is not specified, there should not be any active ones,
649 * return error if any are found (nb: this is a user error, not a
650 * system error). If FORCECLOSE is specified, detach any active vnodes
653 * If WRITECLOSE is set, only flush out regular file vnodes open for
656 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
658 * `rootrefs' specifies the base reference count for the root vnode
659 * of this filesystem. The root vnode is considered busy if its
660 * v_usecount exceeds this value. On a successful return, vflush()
661 * will call vrele() on the root vnode exactly rootrefs times.
662 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
666 static int busyprt = 0; /* print out busy vnodes */
667 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
670 static int vflush_scan(struct mount *mp, struct vnode *vp, void *data);
679 vflush(struct mount *mp, int rootrefs, int flags)
681 struct thread *td = curthread; /* XXX */
682 struct vnode *rootvp = NULL;
684 struct vflush_info vflush_info;
687 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
688 ("vflush: bad args"));
690 * Get the filesystem root vnode. We can vput() it
691 * immediately, since with rootrefs > 0, it won't go away.
693 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
698 vflush_info.busy = 0;
699 vflush_info.flags = flags;
701 vmntvnodescan(mp, VMSC_GETVX, NULL, vflush_scan, &vflush_info);
703 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
705 * If just the root vnode is busy, and if its refcount
706 * is equal to `rootrefs', then go ahead and kill it.
708 KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
709 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
710 if (vflush_info.busy == 1 && rootvp->v_usecount == rootrefs) {
711 if (vx_lock(rootvp) == 0) {
714 vflush_info.busy = 0;
718 if (vflush_info.busy)
720 for (; rootrefs > 0; rootrefs--)
726 * The scan callback is made with an VX locked vnode.
729 vflush_scan(struct mount *mp, struct vnode *vp, void *data)
731 struct vflush_info *info = data;
735 * Skip over a vnodes marked VSYSTEM.
737 if ((info->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
742 * If WRITECLOSE is set, flush out unlinked but still open
743 * files (even if open only for reading) and regular file
744 * vnodes open for writing.
746 if ((info->flags & WRITECLOSE) &&
747 (vp->v_type == VNON ||
748 (VOP_GETATTR(vp, &vattr, info->td) == 0 &&
749 vattr.va_nlink > 0)) &&
750 (vp->v_writecount == 0 || vp->v_type != VREG)) {
755 * With v_usecount == 0, all we need to do is clear out the
756 * vnode data structures and we are done.
758 if (vp->v_usecount == 1) {
764 * If FORCECLOSE is set, forcibly close the vnode. For block
765 * or character devices, revert to an anonymous device. For
766 * all other files, just kill them.
768 if (info->flags & FORCECLOSE) {
769 if (vp->v_type != VBLK && vp->v_type != VCHR) {
772 vclean(vp, 0, info->td);
773 vp->v_ops = spec_vnode_vops;
780 vprint("vflush: busy vnode", vp);