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
72 * External virtual filesystem routines
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/kernel.h>
78 #include <sys/malloc.h>
79 #include <sys/mount.h>
81 #include <sys/vnode.h>
83 #include <sys/eventhandler.h>
84 #include <sys/kthread.h>
85 #include <sys/sysctl.h>
87 #include <machine/limits.h>
90 #include <sys/thread2.h>
91 #include <sys/sysref2.h>
94 #include <vm/vm_object.h>
96 struct mountscan_info {
97 TAILQ_ENTRY(mountscan_info) msi_entry;
99 struct mount *msi_node;
102 struct vmntvnodescan_info {
103 TAILQ_ENTRY(vmntvnodescan_info) entry;
111 static int vnlru_nowhere = 0;
112 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RD,
114 "Number of times the vnlru process ran without success");
117 static struct lwkt_token mntid_token;
118 static struct mount dummymount;
120 /* note: mountlist exported to pstat */
121 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
122 static TAILQ_HEAD(,mountscan_info) mountscan_list;
123 static struct lwkt_token mountlist_token;
125 static TAILQ_HEAD(,bio_ops) bio_ops_list = TAILQ_HEAD_INITIALIZER(bio_ops_list);
128 * Called from vfsinit()
133 lwkt_token_init(&mountlist_token, "mntlist");
134 lwkt_token_init(&mntid_token, "mntid");
135 TAILQ_INIT(&mountscan_list);
136 mount_init(&dummymount);
137 dummymount.mnt_flag |= MNT_RDONLY;
138 dummymount.mnt_kern_flag |= MNTK_ALL_MPSAFE;
142 * Support function called to remove a vnode from the mountlist and
143 * deal with side effects for scans in progress.
145 * Target mnt_token is held on call.
148 vremovevnodemnt(struct vnode *vp)
150 struct vmntvnodescan_info *info;
151 struct mount *mp = vp->v_mount;
153 TAILQ_FOREACH(info, &mp->mnt_vnodescan_list, entry) {
155 info->vp = TAILQ_NEXT(vp, v_nmntvnodes);
157 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
161 * Allocate a new vnode and associate it with a tag, mount point, and
164 * A VX locked and refd vnode is returned. The caller should setup the
165 * remaining fields and vx_put() or, if he wishes to leave a vref,
166 * vx_unlock() the vnode.
169 getnewvnode(enum vtagtype tag, struct mount *mp,
170 struct vnode **vpp, int lktimeout, int lkflags)
174 KKASSERT(mp != NULL);
176 vp = allocvnode(lktimeout, lkflags);
181 * By default the vnode is assigned the mount point's normal
184 vp->v_ops = &mp->mnt_vn_use_ops;
187 * Placing the vnode on the mount point's queue makes it visible.
188 * VNON prevents it from being messed with, however.
193 * A VX locked & refd vnode is returned.
200 * This function creates vnodes with special operations vectors. The
201 * mount point is optional.
203 * This routine is being phased out but is still used by vfs_conf to
204 * create vnodes for devices prior to the root mount (with mp == NULL).
207 getspecialvnode(enum vtagtype tag, struct mount *mp,
208 struct vop_ops **ops,
209 struct vnode **vpp, int lktimeout, int lkflags)
213 vp = allocvnode(lktimeout, lkflags);
222 * Placing the vnode on the mount point's queue makes it visible.
223 * VNON prevents it from being messed with, however.
228 * A VX locked & refd vnode is returned.
235 * Interlock against an unmount, return 0 on success, non-zero on failure.
237 * The passed flag may be 0 or LK_NOWAIT and is only used if an unmount
240 * If no unmount is in-progress LK_NOWAIT is ignored. No other flag bits
241 * are used. A shared locked will be obtained and the filesystem will not
242 * be unmountable until the lock is released.
245 vfs_busy(struct mount *mp, int flags)
249 atomic_add_int(&mp->mnt_refs, 1);
250 lwkt_gettoken(&mp->mnt_token);
251 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
252 if (flags & LK_NOWAIT) {
253 lwkt_reltoken(&mp->mnt_token);
254 atomic_add_int(&mp->mnt_refs, -1);
257 /* XXX not MP safe */
258 mp->mnt_kern_flag |= MNTK_MWAIT;
260 * Since all busy locks are shared except the exclusive
261 * lock granted when unmounting, the only place that a
262 * wakeup needs to be done is at the release of the
263 * exclusive lock at the end of dounmount.
265 tsleep((caddr_t)mp, 0, "vfs_busy", 0);
266 lwkt_reltoken(&mp->mnt_token);
267 atomic_add_int(&mp->mnt_refs, -1);
271 if (lockmgr(&mp->mnt_lock, lkflags))
272 panic("vfs_busy: unexpected lock failure");
273 lwkt_reltoken(&mp->mnt_token);
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", hz*5, 0);
332 lwkt_token_init(&mp->mnt_token, "permnt");
334 TAILQ_INIT(&mp->mnt_vnodescan_list);
335 TAILQ_INIT(&mp->mnt_nvnodelist);
336 TAILQ_INIT(&mp->mnt_reservedvnlist);
337 TAILQ_INIT(&mp->mnt_jlist);
338 mp->mnt_nvnodelistsize = 0;
340 mp->mnt_iosize_max = MAXPHYS;
344 * Lookup a mount point by filesystem identifier.
347 vfs_getvfs(fsid_t *fsid)
351 lwkt_gettoken(&mountlist_token);
352 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
353 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
354 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
358 lwkt_reltoken(&mountlist_token);
363 * Get a new unique fsid. Try to make its val[0] unique, since this value
364 * will be used to create fake device numbers for stat(). Also try (but
365 * not so hard) make its val[0] unique mod 2^16, since some emulators only
366 * support 16-bit device numbers. We end up with unique val[0]'s for the
367 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
369 * Keep in mind that several mounts may be running in parallel. Starting
370 * the search one past where the previous search terminated is both a
371 * micro-optimization and a defense against returning the same fsid to
375 vfs_getnewfsid(struct mount *mp)
377 static u_int16_t mntid_base;
381 lwkt_gettoken(&mntid_token);
382 mtype = mp->mnt_vfc->vfc_typenum;
383 tfsid.val[1] = mtype;
384 mtype = (mtype & 0xFF) << 24;
386 tfsid.val[0] = makeudev(255,
387 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
389 if (vfs_getvfs(&tfsid) == NULL)
392 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
393 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
394 lwkt_reltoken(&mntid_token);
398 * Set the FSID for a new mount point to the template. Adjust
399 * the FSID to avoid collisions.
402 vfs_setfsid(struct mount *mp, fsid_t *template)
406 bzero(&mp->mnt_stat.f_fsid, sizeof(mp->mnt_stat.f_fsid));
408 if (vfs_getvfs(template) == NULL)
413 mp->mnt_stat.f_fsid = *template;
418 * This routine is called when we have too many vnodes. It attempts
419 * to free <count> vnodes and will potentially free vnodes that still
420 * have VM backing store (VM backing store is typically the cause
421 * of a vnode blowout so we want to do this). Therefore, this operation
422 * is not considered cheap.
424 * A number of conditions may prevent a vnode from being reclaimed.
425 * the buffer cache may have references on the vnode, a directory
426 * vnode may still have references due to the namei cache representing
427 * underlying files, or the vnode may be in active use. It is not
428 * desireable to reuse such vnodes. These conditions may cause the
429 * number of vnodes to reach some minimum value regardless of what
430 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
434 * This is a quick non-blocking check to determine if the vnode is a good
435 * candidate for being (eventually) vgone()'d. Returns 0 if the vnode is
436 * not a good candidate, 1 if it is.
439 vmightfree(struct vnode *vp, int page_count, int pass)
441 if (vp->v_flag & VRECLAIMED)
444 if ((vp->v_flag & VFREE) && TAILQ_EMPTY(&vp->v_namecache))
447 if (sysref_isactive(&vp->v_sysref))
449 if (vp->v_object && vp->v_object->resident_page_count >= page_count)
453 * XXX horrible hack. Up to four passes will be taken. Each pass
454 * makes a larger set of vnodes eligible. For now what this really
455 * means is that we try to recycle files opened only once before
456 * recycling files opened multiple times.
458 switch(vp->v_flag & (VAGE0 | VAGE1)) {
478 * The vnode was found to be possibly vgone()able and the caller has locked it
479 * (thus the usecount should be 1 now). Determine if the vnode is actually
480 * vgone()able, doing some cleanups in the process. Returns 1 if the vnode
481 * can be vgone()'d, 0 otherwise.
483 * Note that v_auxrefs may be non-zero because (A) this vnode is not a leaf
484 * in the namecache topology and (B) this vnode has buffer cache bufs.
485 * We cannot remove vnodes with non-leaf namecache associations. We do a
486 * tentitive leaf check prior to attempting to flush out any buffers but the
487 * 'real' test when all is said in done is that v_auxrefs must become 0 for
488 * the vnode to be freeable.
490 * We could theoretically just unconditionally flush when v_auxrefs != 0,
491 * but flushing data associated with non-leaf nodes (which are always
492 * directories), just throws it away for no benefit. It is the buffer
493 * cache's responsibility to choose buffers to recycle from the cached
494 * data point of view.
497 visleaf(struct vnode *vp)
499 struct namecache *ncp;
501 spin_lock(&vp->v_spin);
502 TAILQ_FOREACH(ncp, &vp->v_namecache, nc_vnode) {
503 if (!TAILQ_EMPTY(&ncp->nc_list)) {
504 spin_unlock(&vp->v_spin);
508 spin_unlock(&vp->v_spin);
513 * Try to clean up the vnode to the point where it can be vgone()'d, returning
514 * 0 if it cannot be vgone()'d (or already has been), 1 if it can. Unlike
515 * vmightfree() this routine may flush the vnode and block. Vnodes marked
516 * VFREE are still candidates for vgone()ing because they may hold namecache
517 * resources and could be blocking the namecache directory hierarchy (and
518 * related vnodes) from being freed.
521 vtrytomakegoneable(struct vnode *vp, int page_count)
523 if (vp->v_flag & VRECLAIMED)
525 if (vp->v_sysref.refcnt > 1)
527 if (vp->v_object && vp->v_object->resident_page_count >= page_count)
529 if (vp->v_auxrefs && visleaf(vp)) {
530 vinvalbuf(vp, V_SAVE, 0, 0);
532 kprintf((vp->v_auxrefs ? "vrecycle: vp %p failed: %s\n" :
533 "vrecycle: vp %p succeeded: %s\n"), vp,
534 (TAILQ_FIRST(&vp->v_namecache) ?
535 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"));
540 * This sequence may seem a little strange, but we need to optimize
541 * the critical path a bit. We can't recycle vnodes with other
542 * references and because we are trying to recycle an otherwise
543 * perfectly fine vnode we have to invalidate the namecache in a
544 * way that avoids possible deadlocks (since the vnode lock is being
545 * held here). Finally, we have to check for other references one
546 * last time in case something snuck in during the inval.
548 if (vp->v_sysref.refcnt > 1 || vp->v_auxrefs != 0)
550 if (cache_inval_vp_nonblock(vp))
552 return (vp->v_sysref.refcnt <= 1 && vp->v_auxrefs == 0);
556 * Reclaim up to 1/10 of the vnodes associated with a mount point. Try
557 * to avoid vnodes which have lots of resident pages (we are trying to free
558 * vnodes, not memory).
560 * This routine is a callback from the mountlist scan. The mount point
561 * in question will be busied.
563 * NOTE: The 1/10 reclamation also ensures that the inactive data set
564 * (the vnodes being recycled by the one-time use) does not degenerate
565 * into too-small a set. This is important because once a vnode is
566 * marked as not being one-time-use (VAGE0/VAGE1 both 0) that vnode
567 * will not be destroyed EXCEPT by this mechanism. VM pages can still
568 * be cleaned/freed by the pageout daemon.
571 vlrureclaim(struct mount *mp, void *data)
573 struct vnlru_info *info = data;
579 int trigger_mult = vnlru_nowhere;
582 * Calculate the trigger point for the resident pages check. The
583 * minimum trigger value is approximately the number of pages in
584 * the system divded by the number of vnodes. However, due to
585 * various other system memory overheads unrelated to data caching
586 * it is a good idea to double the trigger (at least).
588 * trigger_mult starts at 0. If the recycler is having problems
589 * finding enough freeable vnodes it will increase trigger_mult.
590 * This should not happen in normal operation, even on machines with
591 * low amounts of memory, but extraordinary memory use by the system
592 * verses the amount of cached data can trigger it.
594 * (long) -> deal with 64 bit machines, intermediate overflow
596 usevnodes = desiredvnodes;
599 trigger = (long)vmstats.v_page_count * (trigger_mult + 2) / usevnodes;
602 lwkt_gettoken(&mp->mnt_token);
603 count = mp->mnt_nvnodelistsize / 10 + 1;
605 while (count && mp->mnt_syncer) {
607 * Next vnode. Use the special syncer vnode to placemark
608 * the LRU. This way the LRU code does not interfere with
611 vp = TAILQ_NEXT(mp->mnt_syncer, v_nmntvnodes);
612 TAILQ_REMOVE(&mp->mnt_nvnodelist, mp->mnt_syncer, v_nmntvnodes);
614 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp,
615 mp->mnt_syncer, v_nmntvnodes);
617 TAILQ_INSERT_HEAD(&mp->mnt_nvnodelist, mp->mnt_syncer,
619 vp = TAILQ_NEXT(mp->mnt_syncer, v_nmntvnodes);
627 * The VP will stick around while we hold mnt_token,
628 * at least until we block, so we can safely do an initial
629 * check, and then must check again after we lock the vnode.
631 if (vp->v_type == VNON || /* syncer or indeterminant */
632 !vmightfree(vp, trigger, info->pass) /* critical path opt */
639 * VX get the candidate vnode. If the VX get fails the
640 * vnode might still be on the mountlist. Our loop depends
641 * on us at least cycling the vnode to the end of the
644 if (vx_get_nonblock(vp) != 0) {
650 * Since we blocked locking the vp, make sure it is still
651 * a candidate for reclamation. That is, it has not already
652 * been reclaimed and only has our VX reference associated
655 if (vp->v_type == VNON || /* syncer or indeterminant */
656 (vp->v_flag & VRECLAIMED) ||
658 !vtrytomakegoneable(vp, trigger) /* critical path opt */
666 * All right, we are good, move the vp to the end of the
667 * mountlist and clean it out. The vget will have returned
668 * an error if the vnode was destroyed (VRECLAIMED set), so we
669 * do not have to check again. The vput() will move the
670 * vnode to the free list if the vgone() was successful.
672 KKASSERT(vp->v_mount == mp);
678 lwkt_reltoken(&mp->mnt_token);
683 * Attempt to recycle vnodes in a context that is always safe to block.
684 * Calling vlrurecycle() from the bowels of file system code has some
685 * interesting deadlock problems.
687 static struct thread *vnlruthread;
692 struct thread *td = curthread;
693 struct vnlru_info info;
696 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
700 kproc_suspend_loop();
703 * Do some opportunistic roving.
705 if (numvnodes > 100000)
706 vnode_free_rover_scan(50);
707 else if (numvnodes > 10000)
708 vnode_free_rover_scan(20);
710 vnode_free_rover_scan(5);
713 * Try to free some vnodes if we have too many
715 * (long) -> deal with 64 bit machines, intermediate overflow
717 if (numvnodes > desiredvnodes &&
718 freevnodes > desiredvnodes * 2 / 10) {
719 int count = numvnodes - desiredvnodes;
721 if (count > freevnodes / 100)
722 count = freevnodes / 100;
725 freesomevnodes(count);
729 * Do non-critical-path (more robust) cache cleaning,
730 * even if vnode counts are nominal, to try to avoid
731 * having to do it in the critical path.
736 * Nothing to do if most of our vnodes are already on
739 if (numvnodes - freevnodes <= (long)desiredvnodes * 9 / 10) {
740 tsleep(vnlruthread, 0, "vlruwt", hz);
745 * The pass iterates through the four combinations of
746 * VAGE0/VAGE1. We want to get rid of aged small files
751 while (done == 0 && info.pass < 4) {
752 done = mountlist_scan(vlrureclaim, &info,
758 * The vlrureclaim() call only processes 1/10 of the vnodes
759 * on each mount. If we couldn't find any repeat the loop
760 * at least enough times to cover all available vnodes before
761 * we start sleeping. Complain if the failure extends past
762 * 30 second, every 30 seconds.
766 if (vnlru_nowhere % 10 == 0)
767 tsleep(vnlruthread, 0, "vlrup", hz * 3);
768 if (vnlru_nowhere % 100 == 0)
769 kprintf("vnlru_proc: vnode recycler stopped working!\n");
770 if (vnlru_nowhere == 1000)
779 * MOUNTLIST FUNCTIONS
783 * mountlist_insert (MP SAFE)
785 * Add a new mount point to the mount list.
788 mountlist_insert(struct mount *mp, int how)
790 lwkt_gettoken(&mountlist_token);
791 if (how == MNTINS_FIRST)
792 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
794 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
795 lwkt_reltoken(&mountlist_token);
799 * mountlist_interlock (MP SAFE)
801 * Execute the specified interlock function with the mountlist token
802 * held. The function will be called in a serialized fashion verses
803 * other functions called through this mechanism.
806 mountlist_interlock(int (*callback)(struct mount *), struct mount *mp)
810 lwkt_gettoken(&mountlist_token);
811 error = callback(mp);
812 lwkt_reltoken(&mountlist_token);
817 * mountlist_boot_getfirst (DURING BOOT ONLY)
819 * This function returns the first mount on the mountlist, which is
820 * expected to be the root mount. Since no interlocks are obtained
821 * this function is only safe to use during booting.
825 mountlist_boot_getfirst(void)
827 return(TAILQ_FIRST(&mountlist));
831 * mountlist_remove (MP SAFE)
833 * Remove a node from the mountlist. If this node is the next scan node
834 * for any active mountlist scans, the active mountlist scan will be
835 * adjusted to skip the node, thus allowing removals during mountlist
839 mountlist_remove(struct mount *mp)
841 struct mountscan_info *msi;
843 lwkt_gettoken(&mountlist_token);
844 TAILQ_FOREACH(msi, &mountscan_list, msi_entry) {
845 if (msi->msi_node == mp) {
846 if (msi->msi_how & MNTSCAN_FORWARD)
847 msi->msi_node = TAILQ_NEXT(mp, mnt_list);
849 msi->msi_node = TAILQ_PREV(mp, mntlist, mnt_list);
852 TAILQ_REMOVE(&mountlist, mp, mnt_list);
853 lwkt_reltoken(&mountlist_token);
857 * mountlist_exists (MP SAFE)
859 * Checks if a node exists in the mountlist.
860 * This function is mainly used by VFS quota code to check if a
861 * cached nullfs struct mount pointer is still valid at use time
863 * FIXME: there is no warranty the mp passed to that function
864 * will be the same one used by VFS_ACCOUNT() later
867 mountlist_exists(struct mount *mp)
872 lwkt_gettoken(&mountlist_token);
873 TAILQ_FOREACH(lmp, &mountlist, mnt_list) {
879 lwkt_reltoken(&mountlist_token);
884 * mountlist_scan (MP SAFE)
886 * Safely scan the mount points on the mount list. Unless otherwise
887 * specified each mount point will be busied prior to the callback and
888 * unbusied afterwords. The callback may safely remove any mount point
889 * without interfering with the scan. If the current callback
890 * mount is removed the scanner will not attempt to unbusy it.
892 * If a mount node cannot be busied it is silently skipped.
894 * The callback return value is aggregated and a total is returned. A return
895 * value of < 0 is not aggregated and will terminate the scan.
897 * MNTSCAN_FORWARD - the mountlist is scanned in the forward direction
898 * MNTSCAN_REVERSE - the mountlist is scanned in reverse
899 * MNTSCAN_NOBUSY - the scanner will make the callback without busying
903 mountlist_scan(int (*callback)(struct mount *, void *), void *data, int how)
905 struct mountscan_info info;
910 lwkt_gettoken(&mountlist_token);
913 info.msi_node = NULL; /* paranoia */
914 TAILQ_INSERT_TAIL(&mountscan_list, &info, msi_entry);
918 if (how & MNTSCAN_FORWARD) {
919 info.msi_node = TAILQ_FIRST(&mountlist);
920 while ((mp = info.msi_node) != NULL) {
921 if (how & MNTSCAN_NOBUSY) {
922 count = callback(mp, data);
923 } else if (vfs_busy(mp, LK_NOWAIT) == 0) {
924 count = callback(mp, data);
925 if (mp == info.msi_node)
933 if (mp == info.msi_node)
934 info.msi_node = TAILQ_NEXT(mp, mnt_list);
936 } else if (how & MNTSCAN_REVERSE) {
937 info.msi_node = TAILQ_LAST(&mountlist, mntlist);
938 while ((mp = info.msi_node) != NULL) {
939 if (how & MNTSCAN_NOBUSY) {
940 count = callback(mp, data);
941 } else if (vfs_busy(mp, LK_NOWAIT) == 0) {
942 count = callback(mp, data);
943 if (mp == info.msi_node)
951 if (mp == info.msi_node)
952 info.msi_node = TAILQ_PREV(mp, mntlist, mnt_list);
955 TAILQ_REMOVE(&mountscan_list, &info, msi_entry);
956 lwkt_reltoken(&mountlist_token);
961 * MOUNT RELATED VNODE FUNCTIONS
964 static struct kproc_desc vnlru_kp = {
969 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
972 * Move a vnode from one mount queue to another.
977 insmntque(struct vnode *vp, struct mount *mp)
982 * Delete from old mount point vnode list, if on one.
984 if ((omp = vp->v_mount) != NULL) {
985 lwkt_gettoken(&omp->mnt_token);
986 KKASSERT(omp == vp->v_mount);
987 KASSERT(omp->mnt_nvnodelistsize > 0,
988 ("bad mount point vnode list size"));
990 omp->mnt_nvnodelistsize--;
991 lwkt_reltoken(&omp->mnt_token);
995 * Insert into list of vnodes for the new mount point, if available.
996 * The 'end' of the LRU list is the vnode prior to mp->mnt_syncer.
1002 lwkt_gettoken(&mp->mnt_token);
1004 if (mp->mnt_syncer) {
1005 TAILQ_INSERT_BEFORE(mp->mnt_syncer, vp, v_nmntvnodes);
1007 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1009 mp->mnt_nvnodelistsize++;
1010 lwkt_reltoken(&mp->mnt_token);
1015 * Scan the vnodes under a mount point and issue appropriate callbacks.
1017 * The fastfunc() callback is called with just the mountlist token held
1018 * (no vnode lock). It may not block and the vnode may be undergoing
1019 * modifications while the caller is processing it. The vnode will
1020 * not be entirely destroyed, however, due to the fact that the mountlist
1021 * token is held. A return value < 0 skips to the next vnode without calling
1022 * the slowfunc(), a return value > 0 terminates the loop.
1024 * The slowfunc() callback is called after the vnode has been successfully
1025 * locked based on passed flags. The vnode is skipped if it gets rearranged
1026 * or destroyed while blocking on the lock. A non-zero return value from
1027 * the slow function terminates the loop. The slow function is allowed to
1028 * arbitrarily block. The scanning code guarentees consistency of operation
1029 * even if the slow function deletes or moves the node, or blocks and some
1030 * other thread deletes or moves the node.
1032 * NOTE: We hold vmobj_token to prevent a VM object from being destroyed
1033 * out from under the fastfunc()'s vnode test. It will not prevent
1034 * v_object from getting NULL'd out but it will ensure that the
1035 * pointer (if we race) will remain stable.
1041 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
1042 int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data),
1045 struct vmntvnodescan_info info;
1048 int maxcount = mp->mnt_nvnodelistsize * 2;
1052 lwkt_gettoken(&mp->mnt_token);
1053 lwkt_gettoken(&vmobj_token);
1056 * If asked to do one pass stop after iterating available vnodes.
1057 * Under heavy loads new vnodes can be added while we are scanning,
1058 * so this isn't perfect. Create a slop factor of 2x.
1060 if (flags & VMSC_ONEPASS)
1061 stopcount = mp->mnt_nvnodelistsize;
1063 info.vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
1064 TAILQ_INSERT_TAIL(&mp->mnt_vnodescan_list, &info, entry);
1066 while ((vp = info.vp) != NULL) {
1067 if (--maxcount == 0) {
1068 kprintf("Warning: excessive fssync iteration\n");
1069 maxcount = mp->mnt_nvnodelistsize * 2;
1073 * Skip if visible but not ready, or special (e.g.
1076 if (vp->v_type == VNON)
1078 KKASSERT(vp->v_mount == mp);
1081 * Quick test. A negative return continues the loop without
1082 * calling the slow test. 0 continues onto the slow test.
1083 * A positive number aborts the loop.
1086 if ((r = fastfunc(mp, vp, data)) < 0) {
1095 * Get a vxlock on the vnode, retry if it has moved or isn't
1096 * in the mountlist where we expect it.
1101 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) {
1103 error = vget(vp, LK_EXCLUSIVE);
1105 case VMSC_GETVP|VMSC_NOWAIT:
1106 error = vget(vp, LK_EXCLUSIVE|LK_NOWAIT);
1119 * Do not call the slow function if the vnode is
1120 * invalid or if it was ripped out from under us
1121 * while we (potentially) blocked.
1123 if (info.vp == vp && vp->v_type != VNON)
1124 r = slowfunc(mp, vp, data);
1129 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) {
1131 case VMSC_GETVP|VMSC_NOWAIT:
1146 * Yield after some processing. Depending on the number
1147 * of vnodes, we might wind up running for a long time.
1148 * Because threads are not preemptable, time critical
1149 * userland processes might starve. Give them a chance
1152 if (++count == 10000) {
1153 /* We really want to yield a bit, so we simply sleep a tick */
1154 tsleep(mp, 0, "vnodescn", 1);
1159 * If doing one pass this decrements to zero. If it starts
1160 * at zero it is effectively unlimited for the purposes of
1163 if (--stopcount == 0)
1167 * Iterate. If the vnode was ripped out from under us
1168 * info.vp will already point to the next vnode, otherwise
1169 * we have to obtain the next valid vnode ourselves.
1172 info.vp = TAILQ_NEXT(vp, v_nmntvnodes);
1175 TAILQ_REMOVE(&mp->mnt_vnodescan_list, &info, entry);
1176 lwkt_reltoken(&vmobj_token);
1177 lwkt_reltoken(&mp->mnt_token);
1182 * Remove any vnodes in the vnode table belonging to mount point mp.
1184 * If FORCECLOSE is not specified, there should not be any active ones,
1185 * return error if any are found (nb: this is a user error, not a
1186 * system error). If FORCECLOSE is specified, detach any active vnodes
1189 * If WRITECLOSE is set, only flush out regular file vnodes open for
1192 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
1194 * `rootrefs' specifies the base reference count for the root vnode
1195 * of this filesystem. The root vnode is considered busy if its
1196 * v_sysref.refcnt exceeds this value. On a successful return, vflush()
1197 * will call vrele() on the root vnode exactly rootrefs times.
1198 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
1202 static int busyprt = 0; /* print out busy vnodes */
1203 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
1206 static int vflush_scan(struct mount *mp, struct vnode *vp, void *data);
1208 struct vflush_info {
1215 vflush(struct mount *mp, int rootrefs, int flags)
1217 struct thread *td = curthread; /* XXX */
1218 struct vnode *rootvp = NULL;
1220 struct vflush_info vflush_info;
1223 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
1224 ("vflush: bad args"));
1226 * Get the filesystem root vnode. We can vput() it
1227 * immediately, since with rootrefs > 0, it won't go away.
1229 if ((error = VFS_ROOT(mp, &rootvp)) != 0) {
1230 if ((flags & FORCECLOSE) == 0)
1233 /* continue anyway */
1239 vflush_info.busy = 0;
1240 vflush_info.flags = flags;
1241 vflush_info.td = td;
1242 vmntvnodescan(mp, VMSC_GETVX, NULL, vflush_scan, &vflush_info);
1244 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
1246 * If just the root vnode is busy, and if its refcount
1247 * is equal to `rootrefs', then go ahead and kill it.
1249 KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
1250 KASSERT(rootvp->v_sysref.refcnt >= rootrefs, ("vflush: rootrefs"));
1251 if (vflush_info.busy == 1 && rootvp->v_sysref.refcnt == rootrefs) {
1253 vgone_vxlocked(rootvp);
1255 vflush_info.busy = 0;
1258 if (vflush_info.busy)
1260 for (; rootrefs > 0; rootrefs--)
1266 * The scan callback is made with an VX locked vnode.
1269 vflush_scan(struct mount *mp, struct vnode *vp, void *data)
1271 struct vflush_info *info = data;
1273 int flags = info->flags;
1276 * Skip over a vnodes marked VSYSTEM.
1278 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
1283 * Do not force-close VCHR or VBLK vnodes
1285 if (vp->v_type == VCHR || vp->v_type == VBLK)
1286 flags &= ~(WRITECLOSE|FORCECLOSE);
1289 * If WRITECLOSE is set, flush out unlinked but still open
1290 * files (even if open only for reading) and regular file
1291 * vnodes open for writing.
1293 if ((flags & WRITECLOSE) &&
1294 (vp->v_type == VNON ||
1295 (VOP_GETATTR(vp, &vattr) == 0 &&
1296 vattr.va_nlink > 0)) &&
1297 (vp->v_writecount == 0 || vp->v_type != VREG)) {
1302 * If we are the only holder (refcnt of 1) or the vnode is in
1303 * termination (refcnt < 0), we can vgone the vnode.
1305 if (vp->v_sysref.refcnt <= 1) {
1311 * If FORCECLOSE is set, forcibly destroy the vnode and then move
1312 * it to a dummymount structure so vop_*() functions don't deref
1315 if (flags & FORCECLOSE) {
1318 if (vp->v_mount == NULL)
1319 insmntque(vp, &dummymount);
1323 if (vp->v_type == VCHR || vp->v_type == VBLK)
1324 kprintf("vflush: Warning, cannot destroy busy device vnode\n");
1327 vprint("vflush: busy vnode", vp);
1334 add_bio_ops(struct bio_ops *ops)
1336 TAILQ_INSERT_TAIL(&bio_ops_list, ops, entry);
1340 rem_bio_ops(struct bio_ops *ops)
1342 TAILQ_REMOVE(&bio_ops_list, ops, entry);
1346 * This calls the bio_ops io_sync function either for a mount point
1349 * WARNING: softdeps is weirdly coded and just isn't happy unless
1350 * io_sync is called with a NULL mount from the general syncing code.
1353 bio_ops_sync(struct mount *mp)
1355 struct bio_ops *ops;
1358 if ((ops = mp->mnt_bioops) != NULL)
1361 TAILQ_FOREACH(ops, &bio_ops_list, entry) {
1368 * Lookup a mount point by nch
1371 mount_get_by_nc(struct namecache *ncp)
1373 struct mount *mp = NULL;
1375 lwkt_gettoken(&mountlist_token);
1376 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1377 if (ncp == mp->mnt_ncmountpt.ncp)
1380 lwkt_reltoken(&mountlist_token);