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 1.37 2008/09/17 21:44:18 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>
94 #include <sys/sysref2.h>
97 #include <vm/vm_object.h>
99 struct mountscan_info {
100 TAILQ_ENTRY(mountscan_info) msi_entry;
102 struct mount *msi_node;
105 struct vmntvnodescan_info {
106 TAILQ_ENTRY(vmntvnodescan_info) entry;
114 static int vnlru_nowhere = 0;
115 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RD,
117 "Number of times the vnlru process ran without success");
120 static struct lwkt_token mntid_token;
121 static struct mount dummymount;
123 /* note: mountlist exported to pstat */
124 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
125 static TAILQ_HEAD(,mountscan_info) mountscan_list;
126 static struct lwkt_token mountlist_token;
127 static TAILQ_HEAD(,vmntvnodescan_info) mntvnodescan_list;
128 struct lwkt_token mntvnode_token;
130 static TAILQ_HEAD(,bio_ops) bio_ops_list = TAILQ_HEAD_INITIALIZER(bio_ops_list);
133 * Called from vfsinit()
138 lwkt_token_init(&mountlist_token, "mntlist");
139 lwkt_token_init(&mntvnode_token, "mntvnode");
140 lwkt_token_init(&mntid_token, "mntid");
141 TAILQ_INIT(&mountscan_list);
142 TAILQ_INIT(&mntvnodescan_list);
143 mount_init(&dummymount);
144 dummymount.mnt_flag |= MNT_RDONLY;
145 dummymount.mnt_kern_flag |= MNTK_ALL_MPSAFE;
149 * Support function called with mntvnode_token held to remove a vnode
150 * from the mountlist. We must update any list scans which are in progress.
153 vremovevnodemnt(struct vnode *vp)
155 struct vmntvnodescan_info *info;
157 TAILQ_FOREACH(info, &mntvnodescan_list, entry) {
159 info->vp = TAILQ_NEXT(vp, v_nmntvnodes);
161 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
165 * Allocate a new vnode and associate it with a tag, mount point, and
168 * A VX locked and refd vnode is returned. The caller should setup the
169 * remaining fields and vx_put() or, if he wishes to leave a vref,
170 * vx_unlock() the vnode.
173 getnewvnode(enum vtagtype tag, struct mount *mp,
174 struct vnode **vpp, int lktimeout, int lkflags)
178 KKASSERT(mp != NULL);
180 vp = allocvnode(lktimeout, lkflags);
185 * By default the vnode is assigned the mount point's normal
188 vp->v_ops = &mp->mnt_vn_use_ops;
191 * Placing the vnode on the mount point's queue makes it visible.
192 * VNON prevents it from being messed with, however.
197 * A VX locked & refd vnode is returned.
204 * This function creates vnodes with special operations vectors. The
205 * mount point is optional.
207 * This routine is being phased out but is still used by vfs_conf to
208 * create vnodes for devices prior to the root mount (with mp == NULL).
211 getspecialvnode(enum vtagtype tag, struct mount *mp,
212 struct vop_ops **ops,
213 struct vnode **vpp, int lktimeout, int lkflags)
217 vp = allocvnode(lktimeout, lkflags);
226 * Placing the vnode on the mount point's queue makes it visible.
227 * VNON prevents it from being messed with, however.
232 * A VX locked & refd vnode is returned.
239 * Interlock against an unmount, return 0 on success, non-zero on failure.
241 * The passed flag may be 0 or LK_NOWAIT and is only used if an unmount
244 * If no unmount is in-progress LK_NOWAIT is ignored. No other flag bits
245 * are used. A shared locked will be obtained and the filesystem will not
246 * be unmountable until the lock is released.
249 vfs_busy(struct mount *mp, int flags)
253 atomic_add_int(&mp->mnt_refs, 1);
254 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
255 if (flags & LK_NOWAIT) {
256 atomic_add_int(&mp->mnt_refs, -1);
259 /* XXX not MP safe */
260 mp->mnt_kern_flag |= MNTK_MWAIT;
262 * Since all busy locks are shared except the exclusive
263 * lock granted when unmounting, the only place that a
264 * wakeup needs to be done is at the release of the
265 * exclusive lock at the end of dounmount.
267 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
268 atomic_add_int(&mp->mnt_refs, -1);
272 if (lockmgr(&mp->mnt_lock, lkflags))
273 panic("vfs_busy: unexpected lock failure");
278 * Free a busy filesystem.
280 * Decrement refs before releasing the lock so e.g. a pending umount
281 * doesn't give us an unexpected busy error.
284 vfs_unbusy(struct mount *mp)
286 atomic_add_int(&mp->mnt_refs, -1);
287 lockmgr(&mp->mnt_lock, LK_RELEASE);
291 * Lookup a filesystem type, and if found allocate and initialize
292 * a mount structure for it.
294 * Devname is usually updated by mount(8) after booting.
297 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
299 struct vfsconf *vfsp;
302 if (fstypename == NULL)
305 vfsp = vfsconf_find_by_name(fstypename);
308 mp = kmalloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
310 lockinit(&mp->mnt_lock, "vfslock", VLKTIMEOUT, 0);
314 mp->mnt_op = vfsp->vfc_vfsops;
315 vfsp->vfc_refcount++;
316 mp->mnt_stat.f_type = vfsp->vfc_typenum;
317 mp->mnt_flag |= MNT_RDONLY;
318 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
319 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
320 copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
326 * Basic mount structure initialization
329 mount_init(struct mount *mp)
331 lockinit(&mp->mnt_lock, "vfslock", 0, 0);
332 lwkt_token_init(&mp->mnt_token, "permnt");
334 TAILQ_INIT(&mp->mnt_nvnodelist);
335 TAILQ_INIT(&mp->mnt_reservedvnlist);
336 TAILQ_INIT(&mp->mnt_jlist);
337 mp->mnt_nvnodelistsize = 0;
339 mp->mnt_iosize_max = MAXPHYS;
343 * Lookup a mount point by filesystem identifier.
346 vfs_getvfs(fsid_t *fsid)
350 lwkt_gettoken(&mountlist_token);
351 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
352 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
353 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
357 lwkt_reltoken(&mountlist_token);
362 * Get a new unique fsid. Try to make its val[0] unique, since this value
363 * will be used to create fake device numbers for stat(). Also try (but
364 * not so hard) make its val[0] unique mod 2^16, since some emulators only
365 * support 16-bit device numbers. We end up with unique val[0]'s for the
366 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
368 * Keep in mind that several mounts may be running in parallel. Starting
369 * the search one past where the previous search terminated is both a
370 * micro-optimization and a defense against returning the same fsid to
374 vfs_getnewfsid(struct mount *mp)
376 static u_int16_t mntid_base;
380 lwkt_gettoken(&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(&mntid_token);
397 * Set the FSID for a new mount point to the template. Adjust
398 * the FSID to avoid collisions.
401 vfs_setfsid(struct mount *mp, fsid_t *template)
405 bzero(&mp->mnt_stat.f_fsid, sizeof(mp->mnt_stat.f_fsid));
407 if (vfs_getvfs(template) == NULL)
412 mp->mnt_stat.f_fsid = *template;
417 * This routine is called when we have too many vnodes. It attempts
418 * to free <count> vnodes and will potentially free vnodes that still
419 * have VM backing store (VM backing store is typically the cause
420 * of a vnode blowout so we want to do this). Therefore, this operation
421 * is not considered cheap.
423 * A number of conditions may prevent a vnode from being reclaimed.
424 * the buffer cache may have references on the vnode, a directory
425 * vnode may still have references due to the namei cache representing
426 * underlying files, or the vnode may be in active use. It is not
427 * desireable to reuse such vnodes. These conditions may cause the
428 * number of vnodes to reach some minimum value regardless of what
429 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
433 * This is a quick non-blocking check to determine if the vnode is a good
434 * candidate for being (eventually) vgone()'d. Returns 0 if the vnode is
435 * not a good candidate, 1 if it is.
438 vmightfree(struct vnode *vp, int page_count, int pass)
440 if (vp->v_flag & VRECLAIMED)
443 if ((vp->v_flag & VFREE) && TAILQ_EMPTY(&vp->v_namecache))
446 if (sysref_isactive(&vp->v_sysref))
448 if (vp->v_object && vp->v_object->resident_page_count >= page_count)
452 * XXX horrible hack. Up to four passes will be taken. Each pass
453 * makes a larger set of vnodes eligible. For now what this really
454 * means is that we try to recycle files opened only once before
455 * recycling files opened multiple times.
457 switch(vp->v_flag & (VAGE0 | VAGE1)) {
477 * The vnode was found to be possibly vgone()able and the caller has locked it
478 * (thus the usecount should be 1 now). Determine if the vnode is actually
479 * vgone()able, doing some cleanups in the process. Returns 1 if the vnode
480 * can be vgone()'d, 0 otherwise.
482 * Note that v_auxrefs may be non-zero because (A) this vnode is not a leaf
483 * in the namecache topology and (B) this vnode has buffer cache bufs.
484 * We cannot remove vnodes with non-leaf namecache associations. We do a
485 * tentitive leaf check prior to attempting to flush out any buffers but the
486 * 'real' test when all is said in done is that v_auxrefs must become 0 for
487 * the vnode to be freeable.
489 * We could theoretically just unconditionally flush when v_auxrefs != 0,
490 * but flushing data associated with non-leaf nodes (which are always
491 * directories), just throws it away for no benefit. It is the buffer
492 * cache's responsibility to choose buffers to recycle from the cached
493 * data point of view.
496 visleaf(struct vnode *vp)
498 struct namecache *ncp;
500 spin_lock(&vp->v_spin);
501 TAILQ_FOREACH(ncp, &vp->v_namecache, nc_vnode) {
502 if (!TAILQ_EMPTY(&ncp->nc_list)) {
503 spin_unlock(&vp->v_spin);
507 spin_unlock(&vp->v_spin);
512 * Try to clean up the vnode to the point where it can be vgone()'d, returning
513 * 0 if it cannot be vgone()'d (or already has been), 1 if it can. Unlike
514 * vmightfree() this routine may flush the vnode and block. Vnodes marked
515 * VFREE are still candidates for vgone()ing because they may hold namecache
516 * resources and could be blocking the namecache directory hierarchy (and
517 * related vnodes) from being freed.
520 vtrytomakegoneable(struct vnode *vp, int page_count)
522 if (vp->v_flag & VRECLAIMED)
524 if (vp->v_sysref.refcnt > 1)
526 if (vp->v_object && vp->v_object->resident_page_count >= page_count)
528 if (vp->v_auxrefs && visleaf(vp)) {
529 vinvalbuf(vp, V_SAVE, 0, 0);
531 kprintf((vp->v_auxrefs ? "vrecycle: vp %p failed: %s\n" :
532 "vrecycle: vp %p succeeded: %s\n"), vp,
533 (TAILQ_FIRST(&vp->v_namecache) ?
534 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"));
539 * This sequence may seem a little strange, but we need to optimize
540 * the critical path a bit. We can't recycle vnodes with other
541 * references and because we are trying to recycle an otherwise
542 * perfectly fine vnode we have to invalidate the namecache in a
543 * way that avoids possible deadlocks (since the vnode lock is being
544 * held here). Finally, we have to check for other references one
545 * last time in case something snuck in during the inval.
547 if (vp->v_sysref.refcnt > 1 || vp->v_auxrefs != 0)
549 if (cache_inval_vp_nonblock(vp))
551 return (vp->v_sysref.refcnt <= 1 && vp->v_auxrefs == 0);
555 * Reclaim up to 1/10 of the vnodes associated with a mount point. Try
556 * to avoid vnodes which have lots of resident pages (we are trying to free
557 * vnodes, not memory).
559 * This routine is a callback from the mountlist scan. The mount point
560 * in question will be busied.
562 * NOTE: The 1/10 reclamation also ensures that the inactive data set
563 * (the vnodes being recycled by the one-time use) does not degenerate
564 * into too-small a set. This is important because once a vnode is
565 * marked as not being one-time-use (VAGE0/VAGE1 both 0) that vnode
566 * will not be destroyed EXCEPT by this mechanism. VM pages can still
567 * be cleaned/freed by the pageout daemon.
570 vlrureclaim(struct mount *mp, void *data)
572 struct vnlru_info *info = data;
578 int trigger_mult = vnlru_nowhere;
581 * Calculate the trigger point for the resident pages check. The
582 * minimum trigger value is approximately the number of pages in
583 * the system divded by the number of vnodes. However, due to
584 * various other system memory overheads unrelated to data caching
585 * it is a good idea to double the trigger (at least).
587 * trigger_mult starts at 0. If the recycler is having problems
588 * finding enough freeable vnodes it will increase trigger_mult.
589 * This should not happen in normal operation, even on machines with
590 * low amounts of memory, but extraordinary memory use by the system
591 * verses the amount of cached data can trigger it.
593 usevnodes = desiredvnodes;
596 trigger = vmstats.v_page_count * (trigger_mult + 2) / usevnodes;
599 lwkt_gettoken(&mntvnode_token);
600 count = mp->mnt_nvnodelistsize / 10 + 1;
602 while (count && mp->mnt_syncer) {
604 * Next vnode. Use the special syncer vnode to placemark
605 * the LRU. This way the LRU code does not interfere with
608 vp = TAILQ_NEXT(mp->mnt_syncer, v_nmntvnodes);
609 TAILQ_REMOVE(&mp->mnt_nvnodelist, mp->mnt_syncer, v_nmntvnodes);
611 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp,
612 mp->mnt_syncer, v_nmntvnodes);
614 TAILQ_INSERT_HEAD(&mp->mnt_nvnodelist, mp->mnt_syncer,
616 vp = TAILQ_NEXT(mp->mnt_syncer, v_nmntvnodes);
624 * The VP will stick around while we hold mntvnode_token,
625 * at least until we block, so we can safely do an initial
626 * check, and then must check again after we lock the vnode.
628 if (vp->v_type == VNON || /* syncer or indeterminant */
629 !vmightfree(vp, trigger, info->pass) /* critical path opt */
636 * VX get the candidate vnode. If the VX get fails the
637 * vnode might still be on the mountlist. Our loop depends
638 * on us at least cycling the vnode to the end of the
641 if (vx_get_nonblock(vp) != 0) {
647 * Since we blocked locking the vp, make sure it is still
648 * a candidate for reclamation. That is, it has not already
649 * been reclaimed and only has our VX reference associated
652 if (vp->v_type == VNON || /* syncer or indeterminant */
653 (vp->v_flag & VRECLAIMED) ||
655 !vtrytomakegoneable(vp, trigger) /* critical path opt */
663 * All right, we are good, move the vp to the end of the
664 * mountlist and clean it out. The vget will have returned
665 * an error if the vnode was destroyed (VRECLAIMED set), so we
666 * do not have to check again. The vput() will move the
667 * vnode to the free list if the vgone() was successful.
669 KKASSERT(vp->v_mount == mp);
675 lwkt_reltoken(&mntvnode_token);
680 * Attempt to recycle vnodes in a context that is always safe to block.
681 * Calling vlrurecycle() from the bowels of file system code has some
682 * interesting deadlock problems.
684 static struct thread *vnlruthread;
685 static int vnlruproc_sig;
688 vnlru_proc_wait(void)
690 tsleep_interlock(&vnlruproc_sig, 0);
691 if (vnlruproc_sig == 0) {
692 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
695 tsleep(&vnlruproc_sig, PINTERLOCKED, "vlruwk", hz);
701 struct thread *td = curthread;
702 struct vnlru_info info;
705 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
709 kproc_suspend_loop();
712 * Try to free some vnodes if we have too many
714 if (numvnodes > desiredvnodes &&
715 freevnodes > desiredvnodes * 2 / 10) {
716 int count = numvnodes - desiredvnodes;
718 if (count > freevnodes / 100)
719 count = freevnodes / 100;
722 freesomevnodes(count);
726 * Nothing to do if most of our vnodes are already on
729 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
731 wakeup(&vnlruproc_sig);
732 tsleep(vnlruthread, 0, "vlruwt", hz);
738 * The pass iterates through the four combinations of
739 * VAGE0/VAGE1. We want to get rid of aged small files
744 while (done == 0 && info.pass < 4) {
745 done = mountlist_scan(vlrureclaim, &info,
751 * The vlrureclaim() call only processes 1/10 of the vnodes
752 * on each mount. If we couldn't find any repeat the loop
753 * at least enough times to cover all available vnodes before
754 * we start sleeping. Complain if the failure extends past
755 * 30 second, every 30 seconds.
759 if (vnlru_nowhere % 10 == 0)
760 tsleep(vnlruthread, 0, "vlrup", hz * 3);
761 if (vnlru_nowhere % 100 == 0)
762 kprintf("vnlru_proc: vnode recycler stopped working!\n");
763 if (vnlru_nowhere == 1000)
772 * MOUNTLIST FUNCTIONS
776 * mountlist_insert (MP SAFE)
778 * Add a new mount point to the mount list.
781 mountlist_insert(struct mount *mp, int how)
783 lwkt_gettoken(&mountlist_token);
784 if (how == MNTINS_FIRST)
785 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
787 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
788 lwkt_reltoken(&mountlist_token);
792 * mountlist_interlock (MP SAFE)
794 * Execute the specified interlock function with the mountlist token
795 * held. The function will be called in a serialized fashion verses
796 * other functions called through this mechanism.
799 mountlist_interlock(int (*callback)(struct mount *), struct mount *mp)
803 lwkt_gettoken(&mountlist_token);
804 error = callback(mp);
805 lwkt_reltoken(&mountlist_token);
810 * mountlist_boot_getfirst (DURING BOOT ONLY)
812 * This function returns the first mount on the mountlist, which is
813 * expected to be the root mount. Since no interlocks are obtained
814 * this function is only safe to use during booting.
818 mountlist_boot_getfirst(void)
820 return(TAILQ_FIRST(&mountlist));
824 * mountlist_remove (MP SAFE)
826 * Remove a node from the mountlist. If this node is the next scan node
827 * for any active mountlist scans, the active mountlist scan will be
828 * adjusted to skip the node, thus allowing removals during mountlist
832 mountlist_remove(struct mount *mp)
834 struct mountscan_info *msi;
836 lwkt_gettoken(&mountlist_token);
837 TAILQ_FOREACH(msi, &mountscan_list, msi_entry) {
838 if (msi->msi_node == mp) {
839 if (msi->msi_how & MNTSCAN_FORWARD)
840 msi->msi_node = TAILQ_NEXT(mp, mnt_list);
842 msi->msi_node = TAILQ_PREV(mp, mntlist, mnt_list);
845 TAILQ_REMOVE(&mountlist, mp, mnt_list);
846 lwkt_reltoken(&mountlist_token);
850 * mountlist_exists (MP SAFE)
852 * Checks if a node exists in the mountlist.
853 * This function is mainly used by VFS quota code to check if a
854 * cached nullfs struct mount pointer is still valid at use time
856 * FIXME: there is no warranty the mp passed to that function
857 * will be the same one used by VFS_ACCOUNT() later
860 mountlist_exists(struct mount *mp)
865 lwkt_gettoken(&mountlist_token);
866 TAILQ_FOREACH(lmp, &mountlist, mnt_list) {
872 lwkt_reltoken(&mountlist_token);
877 * mountlist_scan (MP SAFE)
879 * Safely scan the mount points on the mount list. Unless otherwise
880 * specified each mount point will be busied prior to the callback and
881 * unbusied afterwords. The callback may safely remove any mount point
882 * without interfering with the scan. If the current callback
883 * mount is removed the scanner will not attempt to unbusy it.
885 * If a mount node cannot be busied it is silently skipped.
887 * The callback return value is aggregated and a total is returned. A return
888 * value of < 0 is not aggregated and will terminate the scan.
890 * MNTSCAN_FORWARD - the mountlist is scanned in the forward direction
891 * MNTSCAN_REVERSE - the mountlist is scanned in reverse
892 * MNTSCAN_NOBUSY - the scanner will make the callback without busying
896 mountlist_scan(int (*callback)(struct mount *, void *), void *data, int how)
898 struct mountscan_info info;
903 lwkt_gettoken(&mountlist_token);
906 info.msi_node = NULL; /* paranoia */
907 TAILQ_INSERT_TAIL(&mountscan_list, &info, msi_entry);
911 if (how & MNTSCAN_FORWARD) {
912 info.msi_node = TAILQ_FIRST(&mountlist);
913 while ((mp = info.msi_node) != NULL) {
914 if (how & MNTSCAN_NOBUSY) {
915 count = callback(mp, data);
916 } else if (vfs_busy(mp, LK_NOWAIT) == 0) {
917 count = callback(mp, data);
918 if (mp == info.msi_node)
926 if (mp == info.msi_node)
927 info.msi_node = TAILQ_NEXT(mp, mnt_list);
929 } else if (how & MNTSCAN_REVERSE) {
930 info.msi_node = TAILQ_LAST(&mountlist, mntlist);
931 while ((mp = info.msi_node) != NULL) {
932 if (how & MNTSCAN_NOBUSY) {
933 count = callback(mp, data);
934 } else if (vfs_busy(mp, LK_NOWAIT) == 0) {
935 count = callback(mp, data);
936 if (mp == info.msi_node)
944 if (mp == info.msi_node)
945 info.msi_node = TAILQ_PREV(mp, mntlist, mnt_list);
948 TAILQ_REMOVE(&mountscan_list, &info, msi_entry);
949 lwkt_reltoken(&mountlist_token);
954 * MOUNT RELATED VNODE FUNCTIONS
957 static struct kproc_desc vnlru_kp = {
962 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
965 * Move a vnode from one mount queue to another.
970 insmntque(struct vnode *vp, struct mount *mp)
972 lwkt_gettoken(&mntvnode_token);
974 * Delete from old mount point vnode list, if on one.
976 if (vp->v_mount != NULL) {
977 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
978 ("bad mount point vnode list size"));
980 vp->v_mount->mnt_nvnodelistsize--;
983 * Insert into list of vnodes for the new mount point, if available.
984 * The 'end' of the LRU list is the vnode prior to mp->mnt_syncer.
986 if ((vp->v_mount = mp) == NULL) {
987 lwkt_reltoken(&mntvnode_token);
990 if (mp->mnt_syncer) {
991 TAILQ_INSERT_BEFORE(mp->mnt_syncer, vp, v_nmntvnodes);
993 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
995 mp->mnt_nvnodelistsize++;
996 lwkt_reltoken(&mntvnode_token);
1001 * Scan the vnodes under a mount point and issue appropriate callbacks.
1003 * The fastfunc() callback is called with just the mountlist token held
1004 * (no vnode lock). It may not block and the vnode may be undergoing
1005 * modifications while the caller is processing it. The vnode will
1006 * not be entirely destroyed, however, due to the fact that the mountlist
1007 * token is held. A return value < 0 skips to the next vnode without calling
1008 * the slowfunc(), a return value > 0 terminates the loop.
1010 * The slowfunc() callback is called after the vnode has been successfully
1011 * locked based on passed flags. The vnode is skipped if it gets rearranged
1012 * or destroyed while blocking on the lock. A non-zero return value from
1013 * the slow function terminates the loop. The slow function is allowed to
1014 * arbitrarily block. The scanning code guarentees consistency of operation
1015 * even if the slow function deletes or moves the node, or blocks and some
1016 * other thread deletes or moves the node.
1018 * NOTE: We hold vmobj_token to prevent a VM object from being destroyed
1019 * out from under the fastfunc()'s vnode test. It will not prevent
1020 * v_object from getting NULL'd out but it will ensure that the
1021 * pointer (if we race) will remain stable.
1027 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
1028 int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data),
1031 struct vmntvnodescan_info info;
1034 int maxcount = mp->mnt_nvnodelistsize * 2;
1038 lwkt_gettoken(&mntvnode_token);
1039 lwkt_gettoken(&vmobj_token);
1042 * If asked to do one pass stop after iterating available vnodes.
1043 * Under heavy loads new vnodes can be added while we are scanning,
1044 * so this isn't perfect. Create a slop factor of 2x.
1046 if (flags & VMSC_ONEPASS)
1047 stopcount = mp->mnt_nvnodelistsize;
1049 info.vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1050 TAILQ_INSERT_TAIL(&mntvnodescan_list, &info, entry);
1051 while ((vp = info.vp) != NULL) {
1052 if (--maxcount == 0) {
1053 kprintf("Warning: excessive fssync iteration\n");
1054 maxcount = mp->mnt_nvnodelistsize * 2;
1058 * Skip if visible but not ready, or special (e.g.
1061 if (vp->v_type == VNON)
1063 KKASSERT(vp->v_mount == mp);
1066 * Quick test. A negative return continues the loop without
1067 * calling the slow test. 0 continues onto the slow test.
1068 * A positive number aborts the loop.
1071 if ((r = fastfunc(mp, vp, data)) < 0) {
1080 * Get a vxlock on the vnode, retry if it has moved or isn't
1081 * in the mountlist where we expect it.
1086 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) {
1088 error = vget(vp, LK_EXCLUSIVE);
1090 case VMSC_GETVP|VMSC_NOWAIT:
1091 error = vget(vp, LK_EXCLUSIVE|LK_NOWAIT);
1104 * Do not call the slow function if the vnode is
1105 * invalid or if it was ripped out from under us
1106 * while we (potentially) blocked.
1108 if (info.vp == vp && vp->v_type != VNON)
1109 r = slowfunc(mp, vp, data);
1114 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) {
1116 case VMSC_GETVP|VMSC_NOWAIT:
1131 * Yield after some processing. Depending on the number
1132 * of vnodes, we might wind up running for a long time.
1133 * Because threads are not preemptable, time critical
1134 * userland processes might starve. Give them a chance
1137 if (++count == 10000) {
1138 /* We really want to yield a bit, so we simply sleep a tick */
1139 tsleep(mp, 0, "vnodescn", 1);
1144 * If doing one pass this decrements to zero. If it starts
1145 * at zero it is effectively unlimited for the purposes of
1148 if (--stopcount == 0)
1152 * Iterate. If the vnode was ripped out from under us
1153 * info.vp will already point to the next vnode, otherwise
1154 * we have to obtain the next valid vnode ourselves.
1157 info.vp = TAILQ_NEXT(vp, v_nmntvnodes);
1159 TAILQ_REMOVE(&mntvnodescan_list, &info, entry);
1160 lwkt_reltoken(&vmobj_token);
1161 lwkt_reltoken(&mntvnode_token);
1166 * Remove any vnodes in the vnode table belonging to mount point mp.
1168 * If FORCECLOSE is not specified, there should not be any active ones,
1169 * return error if any are found (nb: this is a user error, not a
1170 * system error). If FORCECLOSE is specified, detach any active vnodes
1173 * If WRITECLOSE is set, only flush out regular file vnodes open for
1176 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1178 * `rootrefs' specifies the base reference count for the root vnode
1179 * of this filesystem. The root vnode is considered busy if its
1180 * v_sysref.refcnt exceeds this value. On a successful return, vflush()
1181 * will call vrele() on the root vnode exactly rootrefs times.
1182 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1186 static int busyprt = 0; /* print out busy vnodes */
1187 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1190 static int vflush_scan(struct mount *mp, struct vnode *vp, void *data);
1192 struct vflush_info {
1199 vflush(struct mount *mp, int rootrefs, int flags)
1201 struct thread *td = curthread; /* XXX */
1202 struct vnode *rootvp = NULL;
1204 struct vflush_info vflush_info;
1207 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1208 ("vflush: bad args"));
1210 * Get the filesystem root vnode. We can vput() it
1211 * immediately, since with rootrefs > 0, it won't go away.
1213 if ((error = VFS_ROOT(mp, &rootvp)) != 0) {
1214 if ((flags & FORCECLOSE) == 0)
1217 /* continue anyway */
1223 vflush_info.busy = 0;
1224 vflush_info.flags = flags;
1225 vflush_info.td = td;
1226 vmntvnodescan(mp, VMSC_GETVX, NULL, vflush_scan, &vflush_info);
1228 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1230 * If just the root vnode is busy, and if its refcount
1231 * is equal to `rootrefs', then go ahead and kill it.
1233 KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
1234 KASSERT(rootvp->v_sysref.refcnt >= rootrefs, ("vflush: rootrefs"));
1235 if (vflush_info.busy == 1 && rootvp->v_sysref.refcnt == rootrefs) {
1237 vgone_vxlocked(rootvp);
1239 vflush_info.busy = 0;
1242 if (vflush_info.busy)
1244 for (; rootrefs > 0; rootrefs--)
1250 * The scan callback is made with an VX locked vnode.
1253 vflush_scan(struct mount *mp, struct vnode *vp, void *data)
1255 struct vflush_info *info = data;
1259 * Skip over a vnodes marked VSYSTEM.
1261 if ((info->flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1266 * If WRITECLOSE is set, flush out unlinked but still open
1267 * files (even if open only for reading) and regular file
1268 * vnodes open for writing.
1270 if ((info->flags & WRITECLOSE) &&
1271 (vp->v_type == VNON ||
1272 (VOP_GETATTR(vp, &vattr) == 0 &&
1273 vattr.va_nlink > 0)) &&
1274 (vp->v_writecount == 0 || vp->v_type != VREG)) {
1279 * If we are the only holder (refcnt of 1) or the vnode is in
1280 * termination (refcnt < 0), we can vgone the vnode.
1282 if (vp->v_sysref.refcnt <= 1) {
1288 * If FORCECLOSE is set, forcibly destroy the vnode and then move
1289 * it to a dummymount structure so vop_*() functions don't deref
1292 if (info->flags & FORCECLOSE) {
1295 if (vp->v_mount == NULL)
1296 insmntque(vp, &dummymount);
1302 vprint("vflush: busy vnode", vp);
1309 add_bio_ops(struct bio_ops *ops)
1311 TAILQ_INSERT_TAIL(&bio_ops_list, ops, entry);
1315 rem_bio_ops(struct bio_ops *ops)
1317 TAILQ_REMOVE(&bio_ops_list, ops, entry);
1321 * This calls the bio_ops io_sync function either for a mount point
1324 * WARNING: softdeps is weirdly coded and just isn't happy unless
1325 * io_sync is called with a NULL mount from the general syncing code.
1328 bio_ops_sync(struct mount *mp)
1330 struct bio_ops *ops;
1333 if ((ops = mp->mnt_bioops) != NULL)
1336 TAILQ_FOREACH(ops, &bio_ops_list, entry) {
1343 * Lookup a mount point by nch
1346 mount_get_by_nc(struct namecache *ncp)
1348 struct mount *mp = NULL;
1350 lwkt_gettoken(&mountlist_token);
1351 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1352 if (ncp == mp->mnt_ncmountpt.ncp)
1355 lwkt_reltoken(&mountlist_token);