2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
38 * External virtual filesystem routines
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
47 #include <sys/param.h>
48 #include <sys/systm.h>
52 #include <sys/dirent.h>
53 #include <sys/event.h>
54 #include <sys/eventhandler.h>
55 #include <sys/extattr.h>
57 #include <sys/fcntl.h>
60 #include <sys/kernel.h>
61 #include <sys/kthread.h>
62 #include <sys/lockf.h>
63 #include <sys/malloc.h>
64 #include <sys/mount.h>
65 #include <sys/namei.h>
67 #include <sys/reboot.h>
68 #include <sys/sleepqueue.h>
70 #include <sys/sysctl.h>
71 #include <sys/syslog.h>
72 #include <sys/vmmeter.h>
73 #include <sys/vnode.h>
75 #include <machine/stdarg.h>
77 #include <security/mac/mac_framework.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_kern.h>
95 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure");
97 static void delmntque(struct vnode *vp);
98 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
99 int slpflag, int slptimeo);
100 static void syncer_shutdown(void *arg, int howto);
101 static int vtryrecycle(struct vnode *vp);
102 static void vbusy(struct vnode *vp);
103 static void vinactive(struct vnode *, struct thread *);
104 static void v_incr_usecount(struct vnode *);
105 static void v_decr_usecount(struct vnode *);
106 static void v_decr_useonly(struct vnode *);
107 static void v_upgrade_usecount(struct vnode *);
108 static void vfree(struct vnode *);
109 static void vnlru_free(int);
110 static void vdestroy(struct vnode *);
111 static void vgonel(struct vnode *);
112 static void vfs_knllock(void *arg);
113 static void vfs_knlunlock(void *arg);
114 static int vfs_knllocked(void *arg);
118 * Enable Giant pushdown based on whether or not the vm is mpsafe in this
119 * build. Without mpsafevm the buffer cache can not run Giant free.
122 TUNABLE_INT("debug.mpsafevfs", &mpsafe_vfs);
123 SYSCTL_INT(_debug, OID_AUTO, mpsafevfs, CTLFLAG_RD, &mpsafe_vfs, 0,
127 * Number of vnodes in existence. Increased whenever getnewvnode()
128 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
131 static unsigned long numvnodes;
133 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
136 * Conversion tables for conversion from vnode types to inode formats
139 enum vtype iftovt_tab[16] = {
140 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
141 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
143 int vttoif_tab[10] = {
144 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
145 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
149 * List of vnodes that are ready for recycling.
151 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
154 * Free vnode target. Free vnodes may simply be files which have been stat'd
155 * but not read. This is somewhat common, and a small cache of such files
156 * should be kept to avoid recreation costs.
158 static u_long wantfreevnodes;
159 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
160 /* Number of vnodes in the free list. */
161 static u_long freevnodes;
162 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
165 * Various variables used for debugging the new implementation of
167 * XXX these are probably of (very) limited utility now.
169 static int reassignbufcalls;
170 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
173 * Cache for the mount type id assigned to NFS. This is used for
174 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
176 int nfs_mount_type = -1;
178 /* To keep more than one thread at a time from running vfs_getnewfsid */
179 static struct mtx mntid_mtx;
182 * Lock for any access to the following:
187 static struct mtx vnode_free_list_mtx;
189 /* Publicly exported FS */
190 struct nfs_public nfs_pub;
192 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
193 static uma_zone_t vnode_zone;
194 static uma_zone_t vnodepoll_zone;
196 /* Set to 1 to print out reclaim of active vnodes */
200 * The workitem queue.
202 * It is useful to delay writes of file data and filesystem metadata
203 * for tens of seconds so that quickly created and deleted files need
204 * not waste disk bandwidth being created and removed. To realize this,
205 * we append vnodes to a "workitem" queue. When running with a soft
206 * updates implementation, most pending metadata dependencies should
207 * not wait for more than a few seconds. Thus, mounted on block devices
208 * are delayed only about a half the time that file data is delayed.
209 * Similarly, directory updates are more critical, so are only delayed
210 * about a third the time that file data is delayed. Thus, there are
211 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
212 * one each second (driven off the filesystem syncer process). The
213 * syncer_delayno variable indicates the next queue that is to be processed.
214 * Items that need to be processed soon are placed in this queue:
216 * syncer_workitem_pending[syncer_delayno]
218 * A delay of fifteen seconds is done by placing the request fifteen
219 * entries later in the queue:
221 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
224 static int syncer_delayno;
225 static long syncer_mask;
226 LIST_HEAD(synclist, bufobj);
227 static struct synclist *syncer_workitem_pending[2];
229 * The sync_mtx protects:
234 * syncer_workitem_pending
235 * syncer_worklist_len
238 static struct mtx sync_mtx;
240 #define SYNCER_MAXDELAY 32
241 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
242 static int syncdelay = 30; /* max time to delay syncing data */
243 static int filedelay = 30; /* time to delay syncing files */
244 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
245 static int dirdelay = 29; /* time to delay syncing directories */
246 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
247 static int metadelay = 28; /* time to delay syncing metadata */
248 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
249 static int rushjob; /* number of slots to run ASAP */
250 static int stat_rush_requests; /* number of times I/O speeded up */
251 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
254 * When shutting down the syncer, run it at four times normal speed.
256 #define SYNCER_SHUTDOWN_SPEEDUP 4
257 static int sync_vnode_count;
258 static int syncer_worklist_len;
259 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
263 * Number of vnodes we want to exist at any one time. This is mostly used
264 * to size hash tables in vnode-related code. It is normally not used in
265 * getnewvnode(), as wantfreevnodes is normally nonzero.)
267 * XXX desiredvnodes is historical cruft and should not exist.
270 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
271 &desiredvnodes, 0, "Maximum number of vnodes");
272 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
273 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
274 static int vnlru_nowhere;
275 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
276 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
279 * Macros to control when a vnode is freed and recycled. All require
280 * the vnode interlock.
282 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
283 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
284 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
288 * Initialize the vnode management data structures.
290 #ifndef MAXVNODES_MAX
291 #define MAXVNODES_MAX 100000
294 vntblinit(void *dummy __unused)
298 * Desiredvnodes is a function of the physical memory size and
299 * the kernel's heap size. Specifically, desiredvnodes scales
300 * in proportion to the physical memory size until two fifths
301 * of the kernel's heap size is consumed by vnodes and vm
304 desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
305 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
306 if (desiredvnodes > MAXVNODES_MAX) {
308 printf("Reducing kern.maxvnodes %d -> %d\n",
309 desiredvnodes, MAXVNODES_MAX);
310 desiredvnodes = MAXVNODES_MAX;
312 wantfreevnodes = desiredvnodes / 4;
313 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
314 TAILQ_INIT(&vnode_free_list);
315 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
316 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
317 NULL, NULL, UMA_ALIGN_PTR, 0);
318 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
319 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
321 * Initialize the filesystem syncer.
323 syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
325 syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
327 syncer_maxdelay = syncer_mask + 1;
328 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
330 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
334 * Mark a mount point as busy. Used to synchronize access and to delay
335 * unmounting. Interlock is not released on failure.
338 vfs_busy(struct mount *mp, int flags, struct mtx *interlkp,
345 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
346 if (flags & LK_NOWAIT) {
352 mtx_unlock(interlkp);
353 mp->mnt_kern_flag |= MNTK_MWAIT;
355 * Since all busy locks are shared except the exclusive
356 * lock granted when unmounting, the only place that a
357 * wakeup needs to be done is at the release of the
358 * exclusive lock at the end of dounmount.
360 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
368 mtx_unlock(interlkp);
369 lkflags = LK_SHARED | LK_INTERLOCK;
370 if (lockmgr(&mp->mnt_lock, lkflags, MNT_MTX(mp)))
371 panic("vfs_busy: unexpected lock failure");
376 * Free a busy filesystem.
379 vfs_unbusy(struct mount *mp, struct thread *td)
382 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL);
387 * Lookup a mount point by filesystem identifier.
390 vfs_getvfs(fsid_t *fsid)
394 mtx_lock(&mountlist_mtx);
395 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
396 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
397 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
399 mtx_unlock(&mountlist_mtx);
403 mtx_unlock(&mountlist_mtx);
404 return ((struct mount *) 0);
408 * Check if a user can access privileged mount options.
411 vfs_suser(struct mount *mp, struct thread *td)
416 * If the thread is jailed, but this is not a jail-friendly file
417 * system, deny immediately.
419 if (jailed(td->td_ucred) && !(mp->mnt_vfc->vfc_flags & VFCF_JAIL))
423 * If the file system was mounted outside a jail and a jailed thread
424 * tries to access it, deny immediately.
426 if (!jailed(mp->mnt_cred) && jailed(td->td_ucred))
430 * If the file system was mounted inside different jail that the jail of
431 * the calling thread, deny immediately.
433 if (jailed(mp->mnt_cred) && jailed(td->td_ucred) &&
434 mp->mnt_cred->cr_prison != td->td_ucred->cr_prison) {
438 if ((mp->mnt_flag & MNT_USER) == 0 ||
439 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
440 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
447 * Get a new unique fsid. Try to make its val[0] unique, since this value
448 * will be used to create fake device numbers for stat(). Also try (but
449 * not so hard) make its val[0] unique mod 2^16, since some emulators only
450 * support 16-bit device numbers. We end up with unique val[0]'s for the
451 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
453 * Keep in mind that several mounts may be running in parallel. Starting
454 * the search one past where the previous search terminated is both a
455 * micro-optimization and a defense against returning the same fsid to
459 vfs_getnewfsid(struct mount *mp)
461 static u_int16_t mntid_base;
466 mtx_lock(&mntid_mtx);
467 mtype = mp->mnt_vfc->vfc_typenum;
468 tfsid.val[1] = mtype;
469 mtype = (mtype & 0xFF) << 24;
471 tfsid.val[0] = makedev(255,
472 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
474 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
478 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
479 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
480 mtx_unlock(&mntid_mtx);
484 * Knob to control the precision of file timestamps:
486 * 0 = seconds only; nanoseconds zeroed.
487 * 1 = seconds and nanoseconds, accurate within 1/HZ.
488 * 2 = seconds and nanoseconds, truncated to microseconds.
489 * >=3 = seconds and nanoseconds, maximum precision.
491 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
493 static int timestamp_precision = TSP_SEC;
494 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
495 ×tamp_precision, 0, "");
498 * Get a current timestamp.
501 vfs_timestamp(struct timespec *tsp)
505 switch (timestamp_precision) {
507 tsp->tv_sec = time_second;
515 TIMEVAL_TO_TIMESPEC(&tv, tsp);
525 * Set vnode attributes to VNOVAL
528 vattr_null(struct vattr *vap)
532 vap->va_size = VNOVAL;
533 vap->va_bytes = VNOVAL;
534 vap->va_mode = VNOVAL;
535 vap->va_nlink = VNOVAL;
536 vap->va_uid = VNOVAL;
537 vap->va_gid = VNOVAL;
538 vap->va_fsid = VNOVAL;
539 vap->va_fileid = VNOVAL;
540 vap->va_blocksize = VNOVAL;
541 vap->va_rdev = VNOVAL;
542 vap->va_atime.tv_sec = VNOVAL;
543 vap->va_atime.tv_nsec = VNOVAL;
544 vap->va_mtime.tv_sec = VNOVAL;
545 vap->va_mtime.tv_nsec = VNOVAL;
546 vap->va_ctime.tv_sec = VNOVAL;
547 vap->va_ctime.tv_nsec = VNOVAL;
548 vap->va_birthtime.tv_sec = VNOVAL;
549 vap->va_birthtime.tv_nsec = VNOVAL;
550 vap->va_flags = VNOVAL;
551 vap->va_gen = VNOVAL;
556 * This routine is called when we have too many vnodes. It attempts
557 * to free <count> vnodes and will potentially free vnodes that still
558 * have VM backing store (VM backing store is typically the cause
559 * of a vnode blowout so we want to do this). Therefore, this operation
560 * is not considered cheap.
562 * A number of conditions may prevent a vnode from being reclaimed.
563 * the buffer cache may have references on the vnode, a directory
564 * vnode may still have references due to the namei cache representing
565 * underlying files, or the vnode may be in active use. It is not
566 * desireable to reuse such vnodes. These conditions may cause the
567 * number of vnodes to reach some minimum value regardless of what
568 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
571 vlrureclaim(struct mount *mp)
581 * Calculate the trigger point, don't allow user
582 * screwups to blow us up. This prevents us from
583 * recycling vnodes with lots of resident pages. We
584 * aren't trying to free memory, we are trying to
587 usevnodes = desiredvnodes;
590 trigger = cnt.v_page_count * 2 / usevnodes;
593 vn_start_write(NULL, &mp, V_WAIT);
595 count = mp->mnt_nvnodelistsize / 10 + 1;
597 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
598 while (vp != NULL && vp->v_type == VMARKER)
599 vp = TAILQ_NEXT(vp, v_nmntvnodes);
602 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
603 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
608 * If it's been deconstructed already, it's still
609 * referenced, or it exceeds the trigger, skip it.
611 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
612 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
613 vp->v_object->resident_page_count > trigger)) {
619 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
621 goto next_iter_mntunlocked;
625 * v_usecount may have been bumped after VOP_LOCK() dropped
626 * the vnode interlock and before it was locked again.
628 * It is not necessary to recheck VI_DOOMED because it can
629 * only be set by another thread that holds both the vnode
630 * lock and vnode interlock. If another thread has the
631 * vnode lock before we get to VOP_LOCK() and obtains the
632 * vnode interlock after VOP_LOCK() drops the vnode
633 * interlock, the other thread will be unable to drop the
634 * vnode lock before our VOP_LOCK() call fails.
636 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
637 (vp->v_object != NULL &&
638 vp->v_object->resident_page_count > trigger)) {
639 VOP_UNLOCK(vp, LK_INTERLOCK);
640 goto next_iter_mntunlocked;
642 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
643 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
648 next_iter_mntunlocked:
649 if ((count % 256) != 0)
653 if ((count % 256) != 0)
662 vn_finished_write(mp);
667 * Attempt to keep the free list at wantfreevnodes length.
670 vnlru_free(int count)
675 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
676 for (; count > 0; count--) {
677 vp = TAILQ_FIRST(&vnode_free_list);
679 * The list can be modified while the free_list_mtx
680 * has been dropped and vp could be NULL here.
684 VNASSERT(vp->v_op != NULL, vp,
685 ("vnlru_free: vnode already reclaimed."));
686 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
688 * Don't recycle if we can't get the interlock.
690 if (!VI_TRYLOCK(vp)) {
691 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
694 VNASSERT(VCANRECYCLE(vp), vp,
695 ("vp inconsistent on freelist"));
697 vp->v_iflag &= ~VI_FREE;
699 mtx_unlock(&vnode_free_list_mtx);
701 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
703 VFS_UNLOCK_GIANT(vfslocked);
705 * If the recycled succeeded this vdrop will actually free
706 * the vnode. If not it will simply place it back on
710 mtx_lock(&vnode_free_list_mtx);
714 * Attempt to recycle vnodes in a context that is always safe to block.
715 * Calling vlrurecycle() from the bowels of filesystem code has some
716 * interesting deadlock problems.
718 static struct proc *vnlruproc;
719 static int vnlruproc_sig;
724 struct mount *mp, *nmp;
726 struct proc *p = vnlruproc;
727 struct thread *td = curthread;
729 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
735 kproc_suspend_check(p);
736 mtx_lock(&vnode_free_list_mtx);
737 if (freevnodes > wantfreevnodes)
738 vnlru_free(freevnodes - wantfreevnodes);
739 if (numvnodes <= desiredvnodes * 9 / 10) {
741 wakeup(&vnlruproc_sig);
742 msleep(vnlruproc, &vnode_free_list_mtx,
743 PVFS|PDROP, "vlruwt", hz);
746 mtx_unlock(&vnode_free_list_mtx);
748 mtx_lock(&mountlist_mtx);
749 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
751 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
752 nmp = TAILQ_NEXT(mp, mnt_list);
755 if (!VFS_NEEDSGIANT(mp)) {
760 done += vlrureclaim(mp);
763 mtx_lock(&mountlist_mtx);
764 nmp = TAILQ_NEXT(mp, mnt_list);
767 mtx_unlock(&mountlist_mtx);
769 EVENTHANDLER_INVOKE(vfs_lowvnodes, desiredvnodes / 10);
771 /* These messages are temporary debugging aids */
772 if (vnlru_nowhere < 5)
773 printf("vnlru process getting nowhere..\n");
774 else if (vnlru_nowhere == 5)
775 printf("vnlru process messages stopped.\n");
778 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
784 static struct kproc_desc vnlru_kp = {
789 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
793 * Routines having to do with the management of the vnode table.
797 vdestroy(struct vnode *vp)
801 CTR1(KTR_VFS, "vdestroy vp %p", vp);
802 mtx_lock(&vnode_free_list_mtx);
804 mtx_unlock(&vnode_free_list_mtx);
806 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
807 ("cleaned vnode still on the free list."));
808 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
809 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
810 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
811 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
812 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
813 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
814 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
815 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
816 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
817 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
818 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
821 mac_vnode_destroy(vp);
823 if (vp->v_pollinfo != NULL) {
824 knlist_destroy(&vp->v_pollinfo->vpi_selinfo.si_note);
825 mtx_destroy(&vp->v_pollinfo->vpi_lock);
826 uma_zfree(vnodepoll_zone, vp->v_pollinfo);
829 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
832 lockdestroy(vp->v_vnlock);
833 mtx_destroy(&vp->v_interlock);
834 mtx_destroy(BO_MTX(bo));
835 uma_zfree(vnode_zone, vp);
839 * Try to recycle a freed vnode. We abort if anyone picks up a reference
840 * before we actually vgone(). This function must be called with the vnode
841 * held to prevent the vnode from being returned to the free list midway
845 vtryrecycle(struct vnode *vp)
849 CTR1(KTR_VFS, "vtryrecycle: trying vp %p", vp);
850 VNASSERT(vp->v_holdcnt, vp,
851 ("vtryrecycle: Recycling vp %p without a reference.", vp));
853 * This vnode may found and locked via some other list, if so we
854 * can't recycle it yet.
856 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0)
857 return (EWOULDBLOCK);
859 * Don't recycle if its filesystem is being suspended.
861 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
866 * If we got this far, we need to acquire the interlock and see if
867 * anyone picked up this vnode from another list. If not, we will
868 * mark it with DOOMED via vgonel() so that anyone who does find it
872 if (vp->v_usecount) {
873 VOP_UNLOCK(vp, LK_INTERLOCK);
874 vn_finished_write(vnmp);
877 if ((vp->v_iflag & VI_DOOMED) == 0)
879 VOP_UNLOCK(vp, LK_INTERLOCK);
880 vn_finished_write(vnmp);
881 CTR1(KTR_VFS, "vtryrecycle: recycled vp %p", vp);
886 * Return the next vnode from the free list.
889 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
892 struct vnode *vp = NULL;
895 mtx_lock(&vnode_free_list_mtx);
897 * Lend our context to reclaim vnodes if they've exceeded the max.
899 if (freevnodes > wantfreevnodes)
902 * Wait for available vnodes.
904 if (numvnodes > desiredvnodes) {
905 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
907 * File system is beeing suspended, we cannot risk a
908 * deadlock here, so allocate new vnode anyway.
910 if (freevnodes > wantfreevnodes)
911 vnlru_free(freevnodes - wantfreevnodes);
914 if (vnlruproc_sig == 0) {
915 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
918 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
920 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
921 if (numvnodes > desiredvnodes) {
922 mtx_unlock(&vnode_free_list_mtx);
929 mtx_unlock(&vnode_free_list_mtx);
930 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
934 vp->v_vnlock = &vp->v_lock;
935 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
937 * By default, don't allow shared locks unless filesystems
940 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
946 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
947 bo->bo_ops = &buf_ops_bio;
949 TAILQ_INIT(&bo->bo_clean.bv_hd);
950 TAILQ_INIT(&bo->bo_dirty.bv_hd);
952 * Initialize namecache.
954 LIST_INIT(&vp->v_cache_src);
955 TAILQ_INIT(&vp->v_cache_dst);
957 * Finalize various vnode identity bits.
966 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
967 mac_vnode_associate_singlelabel(mp, vp);
969 printf("NULL mp in getnewvnode()\n");
972 bo->bo_bsize = mp->mnt_stat.f_iosize;
973 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
974 vp->v_vflag |= VV_NOKNOTE;
977 CTR2(KTR_VFS, "getnewvnode: mp %p vp %p", mp, vp);
983 * Delete from old mount point vnode list, if on one.
986 delmntque(struct vnode *vp)
995 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
996 ("bad mount point vnode list size"));
997 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
998 mp->mnt_nvnodelistsize--;
1004 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1008 vp->v_op = &dead_vnodeops;
1009 /* XXX non mp-safe fs may still call insmntque with vnode
1011 if (!VOP_ISLOCKED(vp))
1012 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1018 * Insert into list of vnodes for the new mount point, if available.
1021 insmntque1(struct vnode *vp, struct mount *mp,
1022 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1025 KASSERT(vp->v_mount == NULL,
1026 ("insmntque: vnode already on per mount vnode list"));
1027 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1029 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1030 mp->mnt_nvnodelistsize == 0) {
1038 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1039 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1040 ("neg mount point vnode list size"));
1041 mp->mnt_nvnodelistsize++;
1047 insmntque(struct vnode *vp, struct mount *mp)
1050 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1054 * Flush out and invalidate all buffers associated with a bufobj
1055 * Called with the underlying object locked.
1058 bufobj_invalbuf(struct bufobj *bo, int flags, struct thread *td, int slpflag,
1064 if (flags & V_SAVE) {
1065 error = bufobj_wwait(bo, slpflag, slptimeo);
1070 if (bo->bo_dirty.bv_cnt > 0) {
1072 if ((error = BO_SYNC(bo, MNT_WAIT, td)) != 0)
1075 * XXX We could save a lock/unlock if this was only
1076 * enabled under INVARIANTS
1079 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1080 panic("vinvalbuf: dirty bufs");
1084 * If you alter this loop please notice that interlock is dropped and
1085 * reacquired in flushbuflist. Special care is needed to ensure that
1086 * no race conditions occur from this.
1089 error = flushbuflist(&bo->bo_clean,
1090 flags, bo, slpflag, slptimeo);
1092 error = flushbuflist(&bo->bo_dirty,
1093 flags, bo, slpflag, slptimeo);
1094 if (error != 0 && error != EAGAIN) {
1098 } while (error != 0);
1101 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1102 * have write I/O in-progress but if there is a VM object then the
1103 * VM object can also have read-I/O in-progress.
1106 bufobj_wwait(bo, 0, 0);
1108 if (bo->bo_object != NULL) {
1109 VM_OBJECT_LOCK(bo->bo_object);
1110 vm_object_pip_wait(bo->bo_object, "bovlbx");
1111 VM_OBJECT_UNLOCK(bo->bo_object);
1114 } while (bo->bo_numoutput > 0);
1118 * Destroy the copy in the VM cache, too.
1120 if (bo->bo_object != NULL) {
1121 VM_OBJECT_LOCK(bo->bo_object);
1122 vm_object_page_remove(bo->bo_object, 0, 0,
1123 (flags & V_SAVE) ? TRUE : FALSE);
1124 VM_OBJECT_UNLOCK(bo->bo_object);
1129 if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1130 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1131 panic("vinvalbuf: flush failed");
1138 * Flush out and invalidate all buffers associated with a vnode.
1139 * Called with the underlying object locked.
1142 vinvalbuf(struct vnode *vp, int flags, struct thread *td, int slpflag,
1146 CTR2(KTR_VFS, "vinvalbuf vp %p flags %d", vp, flags);
1147 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1148 return (bufobj_invalbuf(&vp->v_bufobj, flags, td, slpflag, slptimeo));
1152 * Flush out buffers on the specified list.
1156 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1159 struct buf *bp, *nbp;
1164 ASSERT_BO_LOCKED(bo);
1167 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1168 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1169 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1175 lblkno = nbp->b_lblkno;
1176 xflags = nbp->b_xflags &
1177 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1180 error = BUF_TIMELOCK(bp,
1181 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1182 "flushbuf", slpflag, slptimeo);
1185 return (error != ENOLCK ? error : EAGAIN);
1187 KASSERT(bp->b_bufobj == bo,
1188 ("bp %p wrong b_bufobj %p should be %p",
1189 bp, bp->b_bufobj, bo));
1190 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1196 * XXX Since there are no node locks for NFS, I
1197 * believe there is a slight chance that a delayed
1198 * write will occur while sleeping just above, so
1201 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1204 bp->b_flags |= B_ASYNC;
1207 return (EAGAIN); /* XXX: why not loop ? */
1210 bp->b_flags |= (B_INVAL | B_RELBUF);
1211 bp->b_flags &= ~B_ASYNC;
1215 (nbp->b_bufobj != bo ||
1216 nbp->b_lblkno != lblkno ||
1218 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1219 break; /* nbp invalid */
1225 * Truncate a file's buffer and pages to a specified length. This
1226 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1230 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1231 off_t length, int blksize)
1233 struct buf *bp, *nbp;
1238 CTR2(KTR_VFS, "vtruncbuf vp %p length %jd", vp, length);
1240 * Round up to the *next* lbn.
1242 trunclbn = (length + blksize - 1) / blksize;
1244 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1251 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1252 if (bp->b_lblkno < trunclbn)
1255 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1256 BO_MTX(bo)) == ENOLCK)
1260 bp->b_flags |= (B_INVAL | B_RELBUF);
1261 bp->b_flags &= ~B_ASYNC;
1266 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1267 (nbp->b_vp != vp) ||
1268 (nbp->b_flags & B_DELWRI))) {
1274 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1275 if (bp->b_lblkno < trunclbn)
1278 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1279 BO_MTX(bo)) == ENOLCK)
1282 bp->b_flags |= (B_INVAL | B_RELBUF);
1283 bp->b_flags &= ~B_ASYNC;
1287 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1288 (nbp->b_vp != vp) ||
1289 (nbp->b_flags & B_DELWRI) == 0)) {
1298 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1299 if (bp->b_lblkno > 0)
1302 * Since we hold the vnode lock this should only
1303 * fail if we're racing with the buf daemon.
1306 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1307 BO_MTX(bo)) == ENOLCK) {
1310 VNASSERT((bp->b_flags & B_DELWRI), vp,
1311 ("buf(%p) on dirty queue without DELWRI", bp));
1320 bufobj_wwait(bo, 0, 0);
1322 vnode_pager_setsize(vp, length);
1328 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1331 * NOTE: We have to deal with the special case of a background bitmap
1332 * buffer, a situation where two buffers will have the same logical
1333 * block offset. We want (1) only the foreground buffer to be accessed
1334 * in a lookup and (2) must differentiate between the foreground and
1335 * background buffer in the splay tree algorithm because the splay
1336 * tree cannot normally handle multiple entities with the same 'index'.
1337 * We accomplish this by adding differentiating flags to the splay tree's
1342 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1345 struct buf *lefttreemax, *righttreemin, *y;
1349 lefttreemax = righttreemin = &dummy;
1351 if (lblkno < root->b_lblkno ||
1352 (lblkno == root->b_lblkno &&
1353 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1354 if ((y = root->b_left) == NULL)
1356 if (lblkno < y->b_lblkno) {
1358 root->b_left = y->b_right;
1361 if ((y = root->b_left) == NULL)
1364 /* Link into the new root's right tree. */
1365 righttreemin->b_left = root;
1366 righttreemin = root;
1367 } else if (lblkno > root->b_lblkno ||
1368 (lblkno == root->b_lblkno &&
1369 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1370 if ((y = root->b_right) == NULL)
1372 if (lblkno > y->b_lblkno) {
1374 root->b_right = y->b_left;
1377 if ((y = root->b_right) == NULL)
1380 /* Link into the new root's left tree. */
1381 lefttreemax->b_right = root;
1388 /* Assemble the new root. */
1389 lefttreemax->b_right = root->b_left;
1390 righttreemin->b_left = root->b_right;
1391 root->b_left = dummy.b_right;
1392 root->b_right = dummy.b_left;
1397 buf_vlist_remove(struct buf *bp)
1402 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1403 ASSERT_BO_LOCKED(bp->b_bufobj);
1404 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1405 (BX_VNDIRTY|BX_VNCLEAN),
1406 ("buf_vlist_remove: Buf %p is on two lists", bp));
1407 if (bp->b_xflags & BX_VNDIRTY)
1408 bv = &bp->b_bufobj->bo_dirty;
1410 bv = &bp->b_bufobj->bo_clean;
1411 if (bp != bv->bv_root) {
1412 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1413 KASSERT(root == bp, ("splay lookup failed in remove"));
1415 if (bp->b_left == NULL) {
1418 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1419 root->b_right = bp->b_right;
1422 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1424 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1428 * Add the buffer to the sorted clean or dirty block list using a
1429 * splay tree algorithm.
1431 * NOTE: xflags is passed as a constant, optimizing this inline function!
1434 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1439 ASSERT_BO_LOCKED(bo);
1440 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1441 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1442 bp->b_xflags |= xflags;
1443 if (xflags & BX_VNDIRTY)
1448 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1452 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1453 } else if (bp->b_lblkno < root->b_lblkno ||
1454 (bp->b_lblkno == root->b_lblkno &&
1455 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1456 bp->b_left = root->b_left;
1458 root->b_left = NULL;
1459 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1461 bp->b_right = root->b_right;
1463 root->b_right = NULL;
1464 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1471 * Lookup a buffer using the splay tree. Note that we specifically avoid
1472 * shadow buffers used in background bitmap writes.
1474 * This code isn't quite efficient as it could be because we are maintaining
1475 * two sorted lists and do not know which list the block resides in.
1477 * During a "make buildworld" the desired buffer is found at one of
1478 * the roots more than 60% of the time. Thus, checking both roots
1479 * before performing either splay eliminates unnecessary splays on the
1480 * first tree splayed.
1483 gbincore(struct bufobj *bo, daddr_t lblkno)
1487 ASSERT_BO_LOCKED(bo);
1488 if ((bp = bo->bo_clean.bv_root) != NULL &&
1489 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1491 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1492 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1494 if ((bp = bo->bo_clean.bv_root) != NULL) {
1495 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1496 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1499 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1500 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1501 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1508 * Associate a buffer with a vnode.
1511 bgetvp(struct vnode *vp, struct buf *bp)
1516 ASSERT_BO_LOCKED(bo);
1517 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1519 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1520 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1521 ("bgetvp: bp already attached! %p", bp));
1524 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1525 bp->b_flags |= B_NEEDSGIANT;
1529 * Insert onto list for new vnode.
1531 buf_vlist_add(bp, bo, BX_VNCLEAN);
1535 * Disassociate a buffer from a vnode.
1538 brelvp(struct buf *bp)
1543 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1544 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1547 * Delete from old vnode list, if on one.
1549 vp = bp->b_vp; /* XXX */
1552 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1553 buf_vlist_remove(bp);
1555 panic("brelvp: Buffer %p not on queue.", bp);
1556 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1557 bo->bo_flag &= ~BO_ONWORKLST;
1558 mtx_lock(&sync_mtx);
1559 LIST_REMOVE(bo, bo_synclist);
1560 syncer_worklist_len--;
1561 mtx_unlock(&sync_mtx);
1563 bp->b_flags &= ~B_NEEDSGIANT;
1565 bp->b_bufobj = NULL;
1571 * Add an item to the syncer work queue.
1574 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1578 ASSERT_BO_LOCKED(bo);
1580 mtx_lock(&sync_mtx);
1581 if (bo->bo_flag & BO_ONWORKLST)
1582 LIST_REMOVE(bo, bo_synclist);
1584 bo->bo_flag |= BO_ONWORKLST;
1585 syncer_worklist_len++;
1588 if (delay > syncer_maxdelay - 2)
1589 delay = syncer_maxdelay - 2;
1590 slot = (syncer_delayno + delay) & syncer_mask;
1592 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1594 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1596 mtx_unlock(&sync_mtx);
1600 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1604 mtx_lock(&sync_mtx);
1605 len = syncer_worklist_len - sync_vnode_count;
1606 mtx_unlock(&sync_mtx);
1607 error = SYSCTL_OUT(req, &len, sizeof(len));
1611 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1612 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1614 static struct proc *updateproc;
1615 static void sched_sync(void);
1616 static struct kproc_desc up_kp = {
1621 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1624 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1629 *bo = LIST_FIRST(slp);
1632 vp = (*bo)->__bo_vnode; /* XXX */
1633 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1636 * We use vhold in case the vnode does not
1637 * successfully sync. vhold prevents the vnode from
1638 * going away when we unlock the sync_mtx so that
1639 * we can acquire the vnode interlock.
1642 mtx_unlock(&sync_mtx);
1644 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1646 mtx_lock(&sync_mtx);
1647 return (*bo == LIST_FIRST(slp));
1649 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1650 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1652 vn_finished_write(mp);
1654 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1656 * Put us back on the worklist. The worklist
1657 * routine will remove us from our current
1658 * position and then add us back in at a later
1661 vn_syncer_add_to_worklist(*bo, syncdelay);
1665 mtx_lock(&sync_mtx);
1670 * System filesystem synchronizer daemon.
1675 struct synclist *gnext, *next;
1676 struct synclist *gslp, *slp;
1679 struct thread *td = curthread;
1680 static int dummychan;
1682 int net_worklist_len;
1683 int syncer_final_iter;
1688 syncer_final_iter = 0;
1690 syncer_state = SYNCER_RUNNING;
1691 starttime = time_uptime;
1692 td->td_pflags |= TDP_NORUNNINGBUF;
1694 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1697 mtx_lock(&sync_mtx);
1699 if (syncer_state == SYNCER_FINAL_DELAY &&
1700 syncer_final_iter == 0) {
1701 mtx_unlock(&sync_mtx);
1702 kproc_suspend_check(td->td_proc);
1703 mtx_lock(&sync_mtx);
1705 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1706 if (syncer_state != SYNCER_RUNNING &&
1707 starttime != time_uptime) {
1709 printf("\nSyncing disks, vnodes remaining...");
1712 printf("%d ", net_worklist_len);
1714 starttime = time_uptime;
1717 * Push files whose dirty time has expired. Be careful
1718 * of interrupt race on slp queue.
1720 * Skip over empty worklist slots when shutting down.
1723 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1724 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1725 syncer_delayno += 1;
1726 if (syncer_delayno == syncer_maxdelay)
1728 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1729 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1731 * If the worklist has wrapped since the
1732 * it was emptied of all but syncer vnodes,
1733 * switch to the FINAL_DELAY state and run
1734 * for one more second.
1736 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1737 net_worklist_len == 0 &&
1738 last_work_seen == syncer_delayno) {
1739 syncer_state = SYNCER_FINAL_DELAY;
1740 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1742 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1743 LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1746 * Keep track of the last time there was anything
1747 * on the worklist other than syncer vnodes.
1748 * Return to the SHUTTING_DOWN state if any
1751 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1752 last_work_seen = syncer_delayno;
1753 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1754 syncer_state = SYNCER_SHUTTING_DOWN;
1755 while (!LIST_EMPTY(slp)) {
1756 error = sync_vnode(slp, &bo, td);
1758 LIST_REMOVE(bo, bo_synclist);
1759 LIST_INSERT_HEAD(next, bo, bo_synclist);
1763 if (!LIST_EMPTY(gslp)) {
1764 mtx_unlock(&sync_mtx);
1766 mtx_lock(&sync_mtx);
1767 while (!LIST_EMPTY(gslp)) {
1768 error = sync_vnode(gslp, &bo, td);
1770 LIST_REMOVE(bo, bo_synclist);
1771 LIST_INSERT_HEAD(gnext, bo,
1778 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1779 syncer_final_iter--;
1781 * The variable rushjob allows the kernel to speed up the
1782 * processing of the filesystem syncer process. A rushjob
1783 * value of N tells the filesystem syncer to process the next
1784 * N seconds worth of work on its queue ASAP. Currently rushjob
1785 * is used by the soft update code to speed up the filesystem
1786 * syncer process when the incore state is getting so far
1787 * ahead of the disk that the kernel memory pool is being
1788 * threatened with exhaustion.
1795 * Just sleep for a short period of time between
1796 * iterations when shutting down to allow some I/O
1799 * If it has taken us less than a second to process the
1800 * current work, then wait. Otherwise start right over
1801 * again. We can still lose time if any single round
1802 * takes more than two seconds, but it does not really
1803 * matter as we are just trying to generally pace the
1804 * filesystem activity.
1806 if (syncer_state != SYNCER_RUNNING)
1807 msleep(&dummychan, &sync_mtx, PPAUSE, "syncfnl",
1808 hz / SYNCER_SHUTDOWN_SPEEDUP);
1809 else if (time_uptime == starttime)
1810 msleep(&lbolt, &sync_mtx, PPAUSE, "syncer", 0);
1815 * Request the syncer daemon to speed up its work.
1816 * We never push it to speed up more than half of its
1817 * normal turn time, otherwise it could take over the cpu.
1820 speedup_syncer(void)
1825 td = FIRST_THREAD_IN_PROC(updateproc);
1826 mtx_lock(&sync_mtx);
1827 if (rushjob < syncdelay / 2) {
1829 stat_rush_requests += 1;
1832 mtx_unlock(&sync_mtx);
1833 sleepq_remove(td, &lbolt);
1838 * Tell the syncer to speed up its work and run though its work
1839 * list several times, then tell it to shut down.
1842 syncer_shutdown(void *arg, int howto)
1846 if (howto & RB_NOSYNC)
1848 td = FIRST_THREAD_IN_PROC(updateproc);
1849 mtx_lock(&sync_mtx);
1850 syncer_state = SYNCER_SHUTTING_DOWN;
1852 mtx_unlock(&sync_mtx);
1853 sleepq_remove(td, &lbolt);
1854 kproc_shutdown(arg, howto);
1858 * Reassign a buffer from one vnode to another.
1859 * Used to assign file specific control information
1860 * (indirect blocks) to the vnode to which they belong.
1863 reassignbuf(struct buf *bp)
1876 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1877 bp, bp->b_vp, bp->b_flags);
1879 * B_PAGING flagged buffers cannot be reassigned because their vp
1880 * is not fully linked in.
1882 if (bp->b_flags & B_PAGING)
1883 panic("cannot reassign paging buffer");
1886 * Delete from old vnode list, if on one.
1889 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1890 buf_vlist_remove(bp);
1892 panic("reassignbuf: Buffer %p not on queue.", bp);
1894 * If dirty, put on list of dirty buffers; otherwise insert onto list
1897 if (bp->b_flags & B_DELWRI) {
1898 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1899 switch (vp->v_type) {
1909 vn_syncer_add_to_worklist(bo, delay);
1911 buf_vlist_add(bp, bo, BX_VNDIRTY);
1913 buf_vlist_add(bp, bo, BX_VNCLEAN);
1915 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1916 mtx_lock(&sync_mtx);
1917 LIST_REMOVE(bo, bo_synclist);
1918 syncer_worklist_len--;
1919 mtx_unlock(&sync_mtx);
1920 bo->bo_flag &= ~BO_ONWORKLST;
1925 bp = TAILQ_FIRST(&bv->bv_hd);
1926 KASSERT(bp == NULL || bp->b_bufobj == bo,
1927 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1928 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1929 KASSERT(bp == NULL || bp->b_bufobj == bo,
1930 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1932 bp = TAILQ_FIRST(&bv->bv_hd);
1933 KASSERT(bp == NULL || bp->b_bufobj == bo,
1934 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1935 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1936 KASSERT(bp == NULL || bp->b_bufobj == bo,
1937 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1943 * Increment the use and hold counts on the vnode, taking care to reference
1944 * the driver's usecount if this is a chardev. The vholdl() will remove
1945 * the vnode from the free list if it is presently free. Requires the
1946 * vnode interlock and returns with it held.
1949 v_incr_usecount(struct vnode *vp)
1952 CTR3(KTR_VFS, "v_incr_usecount: vp %p holdcnt %d usecount %d\n",
1953 vp, vp->v_holdcnt, vp->v_usecount);
1955 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1957 vp->v_rdev->si_usecount++;
1964 * Turn a holdcnt into a use+holdcnt such that only one call to
1965 * v_decr_usecount is needed.
1968 v_upgrade_usecount(struct vnode *vp)
1971 CTR3(KTR_VFS, "v_upgrade_usecount: vp %p holdcnt %d usecount %d\n",
1972 vp, vp->v_holdcnt, vp->v_usecount);
1974 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1976 vp->v_rdev->si_usecount++;
1982 * Decrement the vnode use and hold count along with the driver's usecount
1983 * if this is a chardev. The vdropl() below releases the vnode interlock
1984 * as it may free the vnode.
1987 v_decr_usecount(struct vnode *vp)
1990 CTR3(KTR_VFS, "v_decr_usecount: vp %p holdcnt %d usecount %d\n",
1991 vp, vp->v_holdcnt, vp->v_usecount);
1992 ASSERT_VI_LOCKED(vp, __FUNCTION__);
1993 VNASSERT(vp->v_usecount > 0, vp,
1994 ("v_decr_usecount: negative usecount"));
1996 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1998 vp->v_rdev->si_usecount--;
2005 * Decrement only the use count and driver use count. This is intended to
2006 * be paired with a follow on vdropl() to release the remaining hold count.
2007 * In this way we may vgone() a vnode with a 0 usecount without risk of
2008 * having it end up on a free list because the hold count is kept above 0.
2011 v_decr_useonly(struct vnode *vp)
2014 CTR3(KTR_VFS, "v_decr_useonly: vp %p holdcnt %d usecount %d\n",
2015 vp, vp->v_holdcnt, vp->v_usecount);
2016 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2017 VNASSERT(vp->v_usecount > 0, vp,
2018 ("v_decr_useonly: negative usecount"));
2020 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2022 vp->v_rdev->si_usecount--;
2028 * Grab a particular vnode from the free list, increment its
2029 * reference count and lock it. VI_DOOMED is set if the vnode
2030 * is being destroyed. Only callers who specify LK_RETRY will
2031 * see doomed vnodes. If inactive processing was delayed in
2032 * vput try to do it here.
2035 vget(struct vnode *vp, int flags, struct thread *td)
2040 VFS_ASSERT_GIANT(vp->v_mount);
2041 if ((flags & LK_INTERLOCK) == 0)
2044 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2048 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2049 panic("vget: vn_lock failed to return ENOENT\n");
2051 /* Upgrade our holdcnt to a usecount. */
2052 v_upgrade_usecount(vp);
2054 * We don't guarantee that any particular close will
2055 * trigger inactive processing so just make a best effort
2056 * here at preventing a reference to a removed file. If
2057 * we don't succeed no harm is done.
2059 if (vp->v_iflag & VI_OWEINACT) {
2060 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2061 (flags & LK_NOWAIT) == 0)
2063 vp->v_iflag &= ~VI_OWEINACT;
2070 * Increase the reference count of a vnode.
2073 vref(struct vnode *vp)
2077 v_incr_usecount(vp);
2082 * Return reference count of a vnode.
2084 * The results of this call are only guaranteed when some mechanism other
2085 * than the VI lock is used to stop other processes from gaining references
2086 * to the vnode. This may be the case if the caller holds the only reference.
2087 * This is also useful when stale data is acceptable as race conditions may
2088 * be accounted for by some other means.
2091 vrefcnt(struct vnode *vp)
2096 usecnt = vp->v_usecount;
2104 * Vnode put/release.
2105 * If count drops to zero, call inactive routine and return to freelist.
2108 vrele(struct vnode *vp)
2110 struct thread *td = curthread; /* XXX */
2112 KASSERT(vp != NULL, ("vrele: null vp"));
2113 VFS_ASSERT_GIANT(vp->v_mount);
2117 /* Skip this v_writecount check if we're going to panic below. */
2118 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2119 ("vrele: missed vn_close"));
2121 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2122 vp->v_usecount == 1)) {
2123 v_decr_usecount(vp);
2126 if (vp->v_usecount != 1) {
2128 vprint("vrele: negative ref count", vp);
2131 panic("vrele: negative ref cnt");
2134 * We want to hold the vnode until the inactive finishes to
2135 * prevent vgone() races. We drop the use count here and the
2136 * hold count below when we're done.
2140 * We must call VOP_INACTIVE with the node locked. Mark
2141 * as VI_DOINGINACT to avoid recursion.
2143 vp->v_iflag |= VI_OWEINACT;
2144 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK) == 0) {
2146 if (vp->v_usecount > 0)
2147 vp->v_iflag &= ~VI_OWEINACT;
2148 if (vp->v_iflag & VI_OWEINACT)
2153 if (vp->v_usecount > 0)
2154 vp->v_iflag &= ~VI_OWEINACT;
2160 * Release an already locked vnode. This give the same effects as
2161 * unlock+vrele(), but takes less time and avoids releasing and
2162 * re-aquiring the lock (as vrele() acquires the lock internally.)
2165 vput(struct vnode *vp)
2167 struct thread *td = curthread; /* XXX */
2170 KASSERT(vp != NULL, ("vput: null vp"));
2171 ASSERT_VOP_LOCKED(vp, "vput");
2172 VFS_ASSERT_GIANT(vp->v_mount);
2174 /* Skip this v_writecount check if we're going to panic below. */
2175 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2176 ("vput: missed vn_close"));
2179 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2180 vp->v_usecount == 1)) {
2182 v_decr_usecount(vp);
2186 if (vp->v_usecount != 1) {
2188 vprint("vput: negative ref count", vp);
2190 panic("vput: negative ref cnt");
2193 * We want to hold the vnode until the inactive finishes to
2194 * prevent vgone() races. We drop the use count here and the
2195 * hold count below when we're done.
2198 vp->v_iflag |= VI_OWEINACT;
2199 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2200 error = VOP_LOCK(vp, LK_UPGRADE|LK_INTERLOCK|LK_NOWAIT);
2203 if (vp->v_usecount > 0)
2204 vp->v_iflag &= ~VI_OWEINACT;
2208 if (vp->v_usecount > 0)
2209 vp->v_iflag &= ~VI_OWEINACT;
2210 if (vp->v_iflag & VI_OWEINACT)
2218 * Somebody doesn't want the vnode recycled.
2221 vhold(struct vnode *vp)
2230 vholdl(struct vnode *vp)
2234 if (VSHOULDBUSY(vp))
2239 * Note that there is one less who cares about this vnode. vdrop() is the
2240 * opposite of vhold().
2243 vdrop(struct vnode *vp)
2251 * Drop the hold count of the vnode. If this is the last reference to
2252 * the vnode we will free it if it has been vgone'd otherwise it is
2253 * placed on the free list.
2256 vdropl(struct vnode *vp)
2259 ASSERT_VI_LOCKED(vp, "vdropl");
2260 if (vp->v_holdcnt <= 0)
2261 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2263 if (vp->v_holdcnt == 0) {
2264 if (vp->v_iflag & VI_DOOMED) {
2274 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2275 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2276 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2277 * failed lock upgrade.
2280 vinactive(struct vnode *vp, struct thread *td)
2283 ASSERT_VOP_LOCKED(vp, "vinactive");
2284 ASSERT_VI_LOCKED(vp, "vinactive");
2285 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2286 ("vinactive: recursed on VI_DOINGINACT"));
2287 vp->v_iflag |= VI_DOINGINACT;
2288 vp->v_iflag &= ~VI_OWEINACT;
2290 VOP_INACTIVE(vp, td);
2292 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2293 ("vinactive: lost VI_DOINGINACT"));
2294 vp->v_iflag &= ~VI_DOINGINACT;
2298 * Remove any vnodes in the vnode table belonging to mount point mp.
2300 * If FORCECLOSE is not specified, there should not be any active ones,
2301 * return error if any are found (nb: this is a user error, not a
2302 * system error). If FORCECLOSE is specified, detach any active vnodes
2305 * If WRITECLOSE is set, only flush out regular file vnodes open for
2308 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2310 * `rootrefs' specifies the base reference count for the root vnode
2311 * of this filesystem. The root vnode is considered busy if its
2312 * v_usecount exceeds this value. On a successful return, vflush(, td)
2313 * will call vrele() on the root vnode exactly rootrefs times.
2314 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2318 static int busyprt = 0; /* print out busy vnodes */
2319 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2323 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2325 struct vnode *vp, *mvp, *rootvp = NULL;
2327 int busy = 0, error;
2329 CTR1(KTR_VFS, "vflush: mp %p", mp);
2331 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2332 ("vflush: bad args"));
2334 * Get the filesystem root vnode. We can vput() it
2335 * immediately, since with rootrefs > 0, it won't go away.
2337 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp, td)) != 0)
2344 MNT_VNODE_FOREACH(vp, mp, mvp) {
2349 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2353 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2357 * Skip over a vnodes marked VV_SYSTEM.
2359 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2366 * If WRITECLOSE is set, flush out unlinked but still open
2367 * files (even if open only for reading) and regular file
2368 * vnodes open for writing.
2370 if (flags & WRITECLOSE) {
2371 error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2374 if ((vp->v_type == VNON ||
2375 (error == 0 && vattr.va_nlink > 0)) &&
2376 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2385 * With v_usecount == 0, all we need to do is clear out the
2386 * vnode data structures and we are done.
2388 * If FORCECLOSE is set, forcibly close the vnode.
2390 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2391 VNASSERT(vp->v_usecount == 0 ||
2392 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2393 ("device VNODE %p is FORCECLOSED", vp));
2399 vprint("vflush: busy vnode", vp);
2407 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2409 * If just the root vnode is busy, and if its refcount
2410 * is equal to `rootrefs', then go ahead and kill it.
2413 KASSERT(busy > 0, ("vflush: not busy"));
2414 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2415 ("vflush: usecount %d < rootrefs %d",
2416 rootvp->v_usecount, rootrefs));
2417 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2418 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2420 VOP_UNLOCK(rootvp, 0);
2427 for (; rootrefs > 0; rootrefs--)
2433 * Recycle an unused vnode to the front of the free list.
2436 vrecycle(struct vnode *vp, struct thread *td)
2440 ASSERT_VOP_LOCKED(vp, "vrecycle");
2443 if (vp->v_usecount == 0) {
2452 * Eliminate all activity associated with a vnode
2453 * in preparation for reuse.
2456 vgone(struct vnode *vp)
2464 * vgone, with the vp interlock held.
2467 vgonel(struct vnode *vp)
2474 CTR1(KTR_VFS, "vgonel: vp %p", vp);
2475 ASSERT_VOP_LOCKED(vp, "vgonel");
2476 ASSERT_VI_LOCKED(vp, "vgonel");
2477 VNASSERT(vp->v_holdcnt, vp,
2478 ("vgonel: vp %p has no reference.", vp));
2482 * Don't vgonel if we're already doomed.
2484 if (vp->v_iflag & VI_DOOMED)
2486 vp->v_iflag |= VI_DOOMED;
2488 * Check to see if the vnode is in use. If so, we have to call
2489 * VOP_CLOSE() and VOP_INACTIVE().
2491 active = vp->v_usecount;
2492 oweinact = (vp->v_iflag & VI_OWEINACT);
2495 * Clean out any buffers associated with the vnode.
2496 * If the flush fails, just toss the buffers.
2499 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2500 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2501 if (vinvalbuf(vp, V_SAVE, td, 0, 0) != 0)
2502 vinvalbuf(vp, 0, td, 0, 0);
2505 * If purging an active vnode, it must be closed and
2506 * deactivated before being reclaimed.
2509 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2510 if (oweinact || active) {
2512 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2517 * Reclaim the vnode.
2519 if (VOP_RECLAIM(vp, td))
2520 panic("vgone: cannot reclaim");
2522 vn_finished_secondary_write(mp);
2523 VNASSERT(vp->v_object == NULL, vp,
2524 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2526 * Clear the advisory locks and wake up waiting threads.
2528 lf_purgelocks(vp, &(vp->v_lockf));
2530 * Delete from old mount point vnode list.
2535 * Done with purge, reset to the standard lock and invalidate
2539 vp->v_vnlock = &vp->v_lock;
2540 vp->v_op = &dead_vnodeops;
2546 * Calculate the total number of references to a special device.
2549 vcount(struct vnode *vp)
2554 count = vp->v_rdev->si_usecount;
2560 * Same as above, but using the struct cdev *as argument
2563 count_dev(struct cdev *dev)
2568 count = dev->si_usecount;
2574 * Print out a description of a vnode.
2576 static char *typename[] =
2577 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2581 vn_printf(struct vnode *vp, const char *fmt, ...)
2584 char buf[256], buf2[16];
2590 printf("%p: ", (void *)vp);
2591 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2592 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2593 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2596 if (vp->v_vflag & VV_ROOT)
2597 strlcat(buf, "|VV_ROOT", sizeof(buf));
2598 if (vp->v_vflag & VV_ISTTY)
2599 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2600 if (vp->v_vflag & VV_NOSYNC)
2601 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2602 if (vp->v_vflag & VV_CACHEDLABEL)
2603 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2604 if (vp->v_vflag & VV_TEXT)
2605 strlcat(buf, "|VV_TEXT", sizeof(buf));
2606 if (vp->v_vflag & VV_COPYONWRITE)
2607 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2608 if (vp->v_vflag & VV_SYSTEM)
2609 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2610 if (vp->v_vflag & VV_PROCDEP)
2611 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2612 if (vp->v_vflag & VV_NOKNOTE)
2613 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2614 if (vp->v_vflag & VV_DELETED)
2615 strlcat(buf, "|VV_DELETED", sizeof(buf));
2616 if (vp->v_vflag & VV_MD)
2617 strlcat(buf, "|VV_MD", sizeof(buf));
2618 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2619 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2620 VV_NOKNOTE | VV_DELETED | VV_MD);
2622 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2623 strlcat(buf, buf2, sizeof(buf));
2625 if (vp->v_iflag & VI_MOUNT)
2626 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2627 if (vp->v_iflag & VI_AGE)
2628 strlcat(buf, "|VI_AGE", sizeof(buf));
2629 if (vp->v_iflag & VI_DOOMED)
2630 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2631 if (vp->v_iflag & VI_FREE)
2632 strlcat(buf, "|VI_FREE", sizeof(buf));
2633 if (vp->v_iflag & VI_OBJDIRTY)
2634 strlcat(buf, "|VI_OBJDIRTY", sizeof(buf));
2635 if (vp->v_iflag & VI_DOINGINACT)
2636 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2637 if (vp->v_iflag & VI_OWEINACT)
2638 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2639 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2640 VI_OBJDIRTY | VI_DOINGINACT | VI_OWEINACT);
2642 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2643 strlcat(buf, buf2, sizeof(buf));
2645 printf(" flags (%s)\n", buf + 1);
2646 if (mtx_owned(VI_MTX(vp)))
2647 printf(" VI_LOCKed");
2648 if (vp->v_object != NULL)
2649 printf(" v_object %p ref %d pages %d\n",
2650 vp->v_object, vp->v_object->ref_count,
2651 vp->v_object->resident_page_count);
2653 lockmgr_printinfo(vp->v_vnlock);
2655 if (vp->v_data != NULL)
2661 * List all of the locked vnodes in the system.
2662 * Called when debugging the kernel.
2664 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2666 struct mount *mp, *nmp;
2670 * Note: because this is DDB, we can't obey the locking semantics
2671 * for these structures, which means we could catch an inconsistent
2672 * state and dereference a nasty pointer. Not much to be done
2675 db_printf("Locked vnodes\n");
2676 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2677 nmp = TAILQ_NEXT(mp, mnt_list);
2678 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2679 if (vp->v_type != VMARKER &&
2683 nmp = TAILQ_NEXT(mp, mnt_list);
2688 * Show details about the given vnode.
2690 DB_SHOW_COMMAND(vnode, db_show_vnode)
2696 vp = (struct vnode *)addr;
2697 vn_printf(vp, "vnode ");
2701 * Show details about the given mount point.
2703 DB_SHOW_COMMAND(mount, db_show_mount)
2712 /* No address given, print short info about all mount points. */
2713 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2714 db_printf("%p %s on %s (%s)\n", mp,
2715 mp->mnt_stat.f_mntfromname,
2716 mp->mnt_stat.f_mntonname,
2717 mp->mnt_stat.f_fstypename);
2721 db_printf("\nMore info: show mount <addr>\n");
2725 mp = (struct mount *)addr;
2726 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2727 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2730 flags = mp->mnt_flag;
2731 #define MNT_FLAG(flag) do { \
2732 if (flags & (flag)) { \
2733 if (buf[0] != '\0') \
2734 strlcat(buf, ", ", sizeof(buf)); \
2735 strlcat(buf, (#flag) + 4, sizeof(buf)); \
2739 MNT_FLAG(MNT_RDONLY);
2740 MNT_FLAG(MNT_SYNCHRONOUS);
2741 MNT_FLAG(MNT_NOEXEC);
2742 MNT_FLAG(MNT_NOSUID);
2743 MNT_FLAG(MNT_UNION);
2744 MNT_FLAG(MNT_ASYNC);
2745 MNT_FLAG(MNT_SUIDDIR);
2746 MNT_FLAG(MNT_SOFTDEP);
2747 MNT_FLAG(MNT_NOSYMFOLLOW);
2748 MNT_FLAG(MNT_GJOURNAL);
2749 MNT_FLAG(MNT_MULTILABEL);
2751 MNT_FLAG(MNT_NOATIME);
2752 MNT_FLAG(MNT_NOCLUSTERR);
2753 MNT_FLAG(MNT_NOCLUSTERW);
2754 MNT_FLAG(MNT_EXRDONLY);
2755 MNT_FLAG(MNT_EXPORTED);
2756 MNT_FLAG(MNT_DEFEXPORTED);
2757 MNT_FLAG(MNT_EXPORTANON);
2758 MNT_FLAG(MNT_EXKERB);
2759 MNT_FLAG(MNT_EXPUBLIC);
2760 MNT_FLAG(MNT_LOCAL);
2761 MNT_FLAG(MNT_QUOTA);
2762 MNT_FLAG(MNT_ROOTFS);
2764 MNT_FLAG(MNT_IGNORE);
2765 MNT_FLAG(MNT_UPDATE);
2766 MNT_FLAG(MNT_DELEXPORT);
2767 MNT_FLAG(MNT_RELOAD);
2768 MNT_FLAG(MNT_FORCE);
2769 MNT_FLAG(MNT_SNAPSHOT);
2770 MNT_FLAG(MNT_BYFSID);
2774 strlcat(buf, ", ", sizeof(buf));
2775 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2778 db_printf(" mnt_flag = %s\n", buf);
2781 flags = mp->mnt_kern_flag;
2782 #define MNT_KERN_FLAG(flag) do { \
2783 if (flags & (flag)) { \
2784 if (buf[0] != '\0') \
2785 strlcat(buf, ", ", sizeof(buf)); \
2786 strlcat(buf, (#flag) + 5, sizeof(buf)); \
2790 MNT_KERN_FLAG(MNTK_UNMOUNTF);
2791 MNT_KERN_FLAG(MNTK_ASYNC);
2792 MNT_KERN_FLAG(MNTK_SOFTDEP);
2793 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2794 MNT_KERN_FLAG(MNTK_UNMOUNT);
2795 MNT_KERN_FLAG(MNTK_MWAIT);
2796 MNT_KERN_FLAG(MNTK_SUSPEND);
2797 MNT_KERN_FLAG(MNTK_SUSPEND2);
2798 MNT_KERN_FLAG(MNTK_SUSPENDED);
2799 MNT_KERN_FLAG(MNTK_MPSAFE);
2800 MNT_KERN_FLAG(MNTK_NOKNOTE);
2801 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2802 #undef MNT_KERN_FLAG
2805 strlcat(buf, ", ", sizeof(buf));
2806 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2809 db_printf(" mnt_kern_flag = %s\n", buf);
2812 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
2813 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2814 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2815 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2816 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2817 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2818 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2819 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2820 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2821 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2822 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2823 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2825 db_printf(" mnt_cred = { uid=%u ruid=%u",
2826 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2827 if (mp->mnt_cred->cr_prison != NULL)
2828 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2830 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
2831 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
2832 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2833 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2834 db_printf(" mnt_noasync = %u\n", mp->mnt_noasync);
2835 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2836 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2837 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
2838 db_printf(" mnt_markercnt = %d\n", mp->mnt_markercnt);
2839 db_printf(" mnt_holdcnt = %d\n", mp->mnt_holdcnt);
2840 db_printf(" mnt_holdcntwaiters = %d\n", mp->mnt_holdcntwaiters);
2841 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2842 db_printf(" mnt_secondary_accwrites = %d\n",
2843 mp->mnt_secondary_accwrites);
2844 db_printf(" mnt_gjprovider = %s\n",
2845 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2848 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2849 if (vp->v_type != VMARKER) {
2850 vn_printf(vp, "vnode ");
2859 * Fill in a struct xvfsconf based on a struct vfsconf.
2862 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2865 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2866 xvfsp->vfc_typenum = vfsp->vfc_typenum;
2867 xvfsp->vfc_refcount = vfsp->vfc_refcount;
2868 xvfsp->vfc_flags = vfsp->vfc_flags;
2870 * These are unused in userland, we keep them
2871 * to not break binary compatibility.
2873 xvfsp->vfc_vfsops = NULL;
2874 xvfsp->vfc_next = NULL;
2878 * Top level filesystem related information gathering.
2881 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2883 struct vfsconf *vfsp;
2884 struct xvfsconf xvfsp;
2888 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2889 bzero(&xvfsp, sizeof(xvfsp));
2890 vfsconf2x(vfsp, &xvfsp);
2891 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2898 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2899 "S,xvfsconf", "List of all configured filesystems");
2901 #ifndef BURN_BRIDGES
2902 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2905 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2907 int *name = (int *)arg1 - 1; /* XXX */
2908 u_int namelen = arg2 + 1; /* XXX */
2909 struct vfsconf *vfsp;
2910 struct xvfsconf xvfsp;
2912 printf("WARNING: userland calling deprecated sysctl, "
2913 "please rebuild world\n");
2915 #if 1 || defined(COMPAT_PRELITE2)
2916 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2918 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2922 case VFS_MAXTYPENUM:
2925 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2928 return (ENOTDIR); /* overloaded */
2929 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
2930 if (vfsp->vfc_typenum == name[2])
2933 return (EOPNOTSUPP);
2934 bzero(&xvfsp, sizeof(xvfsp));
2935 vfsconf2x(vfsp, &xvfsp);
2936 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2938 return (EOPNOTSUPP);
2941 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
2942 vfs_sysctl, "Generic filesystem");
2944 #if 1 || defined(COMPAT_PRELITE2)
2947 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2950 struct vfsconf *vfsp;
2951 struct ovfsconf ovfs;
2953 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2954 bzero(&ovfs, sizeof(ovfs));
2955 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2956 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2957 ovfs.vfc_index = vfsp->vfc_typenum;
2958 ovfs.vfc_refcount = vfsp->vfc_refcount;
2959 ovfs.vfc_flags = vfsp->vfc_flags;
2960 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2967 #endif /* 1 || COMPAT_PRELITE2 */
2968 #endif /* !BURN_BRIDGES */
2970 #define KINFO_VNODESLOP 10
2973 * Dump vnode list (via sysctl).
2977 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2980 struct thread *td = req->td;
2986 * Stale numvnodes access is not fatal here.
2989 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
2991 /* Make an estimate */
2992 return (SYSCTL_OUT(req, 0, len));
2994 error = sysctl_wire_old_buffer(req, 0);
2997 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
2999 mtx_lock(&mountlist_mtx);
3000 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3001 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
3004 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3008 xvn[n].xv_size = sizeof *xvn;
3009 xvn[n].xv_vnode = vp;
3010 xvn[n].xv_id = 0; /* XXX compat */
3011 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3013 XV_COPY(writecount);
3019 xvn[n].xv_flag = vp->v_vflag;
3021 switch (vp->v_type) {
3028 if (vp->v_rdev == NULL) {
3032 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3035 xvn[n].xv_socket = vp->v_socket;
3038 xvn[n].xv_fifo = vp->v_fifoinfo;
3043 /* shouldn't happen? */
3051 mtx_lock(&mountlist_mtx);
3056 mtx_unlock(&mountlist_mtx);
3058 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3063 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3064 0, 0, sysctl_vnode, "S,xvnode", "");
3068 * Unmount all filesystems. The list is traversed in reverse order
3069 * of mounting to avoid dependencies.
3072 vfs_unmountall(void)
3078 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3081 * Since this only runs when rebooting, it is not interlocked.
3083 while(!TAILQ_EMPTY(&mountlist)) {
3084 mp = TAILQ_LAST(&mountlist, mntlist);
3085 error = dounmount(mp, MNT_FORCE, td);
3087 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3089 * XXX: Due to the way in which we mount the root
3090 * file system off of devfs, devfs will generate a
3091 * "busy" warning when we try to unmount it before
3092 * the root. Don't print a warning as a result in
3093 * order to avoid false positive errors that may
3094 * cause needless upset.
3096 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3097 printf("unmount of %s failed (",
3098 mp->mnt_stat.f_mntonname);
3102 printf("%d)\n", error);
3105 /* The unmount has removed mp from the mountlist */
3111 * perform msync on all vnodes under a mount point
3112 * the mount point must be locked.
3115 vfs_msync(struct mount *mp, int flags)
3117 struct vnode *vp, *mvp;
3118 struct vm_object *obj;
3121 MNT_VNODE_FOREACH(vp, mp, mvp) {
3123 if ((vp->v_iflag & VI_OBJDIRTY) &&
3124 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3127 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3129 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3137 VM_OBJECT_LOCK(obj);
3138 vm_object_page_clean(obj, 0, 0,
3140 OBJPC_SYNC : OBJPC_NOSYNC);
3141 VM_OBJECT_UNLOCK(obj);
3153 * Mark a vnode as free, putting it up for recycling.
3156 vfree(struct vnode *vp)
3159 CTR1(KTR_VFS, "vfree vp %p", vp);
3160 ASSERT_VI_LOCKED(vp, "vfree");
3161 mtx_lock(&vnode_free_list_mtx);
3162 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3163 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3164 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3165 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3166 ("vfree: Freeing doomed vnode"));
3167 if (vp->v_iflag & VI_AGE) {
3168 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3170 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3173 vp->v_iflag &= ~VI_AGE;
3174 vp->v_iflag |= VI_FREE;
3175 mtx_unlock(&vnode_free_list_mtx);
3179 * Opposite of vfree() - mark a vnode as in use.
3182 vbusy(struct vnode *vp)
3184 CTR1(KTR_VFS, "vbusy vp %p", vp);
3185 ASSERT_VI_LOCKED(vp, "vbusy");
3186 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3187 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3189 mtx_lock(&vnode_free_list_mtx);
3190 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3192 vp->v_iflag &= ~(VI_FREE|VI_AGE);
3193 mtx_unlock(&vnode_free_list_mtx);
3197 * Initalize per-vnode helper structure to hold poll-related state.
3200 v_addpollinfo(struct vnode *vp)
3202 struct vpollinfo *vi;
3204 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3205 if (vp->v_pollinfo != NULL) {
3206 uma_zfree(vnodepoll_zone, vi);
3209 vp->v_pollinfo = vi;
3210 mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3211 knlist_init(&vp->v_pollinfo->vpi_selinfo.si_note, vp, vfs_knllock,
3212 vfs_knlunlock, vfs_knllocked);
3216 * Record a process's interest in events which might happen to
3217 * a vnode. Because poll uses the historic select-style interface
3218 * internally, this routine serves as both the ``check for any
3219 * pending events'' and the ``record my interest in future events''
3220 * functions. (These are done together, while the lock is held,
3221 * to avoid race conditions.)
3224 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3227 if (vp->v_pollinfo == NULL)
3229 mtx_lock(&vp->v_pollinfo->vpi_lock);
3230 if (vp->v_pollinfo->vpi_revents & events) {
3232 * This leaves events we are not interested
3233 * in available for the other process which
3234 * which presumably had requested them
3235 * (otherwise they would never have been
3238 events &= vp->v_pollinfo->vpi_revents;
3239 vp->v_pollinfo->vpi_revents &= ~events;
3241 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3244 vp->v_pollinfo->vpi_events |= events;
3245 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3246 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3251 * Routine to create and manage a filesystem syncer vnode.
3253 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3254 static int sync_fsync(struct vop_fsync_args *);
3255 static int sync_inactive(struct vop_inactive_args *);
3256 static int sync_reclaim(struct vop_reclaim_args *);
3258 static struct vop_vector sync_vnodeops = {
3259 .vop_bypass = VOP_EOPNOTSUPP,
3260 .vop_close = sync_close, /* close */
3261 .vop_fsync = sync_fsync, /* fsync */
3262 .vop_inactive = sync_inactive, /* inactive */
3263 .vop_reclaim = sync_reclaim, /* reclaim */
3264 .vop_lock1 = vop_stdlock, /* lock */
3265 .vop_unlock = vop_stdunlock, /* unlock */
3266 .vop_islocked = vop_stdislocked, /* islocked */
3270 * Create a new filesystem syncer vnode for the specified mount point.
3273 vfs_allocate_syncvnode(struct mount *mp)
3277 static long start, incr, next;
3280 /* Allocate a new vnode */
3281 if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3282 mp->mnt_syncer = NULL;
3286 error = insmntque(vp, mp);
3288 panic("vfs_allocate_syncvnode: insmntque failed");
3290 * Place the vnode onto the syncer worklist. We attempt to
3291 * scatter them about on the list so that they will go off
3292 * at evenly distributed times even if all the filesystems
3293 * are mounted at once.
3296 if (next == 0 || next > syncer_maxdelay) {
3300 start = syncer_maxdelay / 2;
3301 incr = syncer_maxdelay;
3307 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3308 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3309 mtx_lock(&sync_mtx);
3311 mtx_unlock(&sync_mtx);
3313 mp->mnt_syncer = vp;
3318 * Do a lazy sync of the filesystem.
3321 sync_fsync(struct vop_fsync_args *ap)
3323 struct vnode *syncvp = ap->a_vp;
3324 struct mount *mp = syncvp->v_mount;
3325 struct thread *td = ap->a_td;
3330 * We only need to do something if this is a lazy evaluation.
3332 if (ap->a_waitfor != MNT_LAZY)
3336 * Move ourselves to the back of the sync list.
3338 bo = &syncvp->v_bufobj;
3340 vn_syncer_add_to_worklist(bo, syncdelay);
3344 * Walk the list of vnodes pushing all that are dirty and
3345 * not already on the sync list.
3347 mtx_lock(&mountlist_mtx);
3348 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3349 mtx_unlock(&mountlist_mtx);
3352 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3358 mp->mnt_kern_flag &= ~MNTK_ASYNC;
3360 vfs_msync(mp, MNT_NOWAIT);
3361 error = VFS_SYNC(mp, MNT_LAZY, td);
3364 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3365 mp->mnt_kern_flag |= MNTK_ASYNC;
3367 vn_finished_write(mp);
3373 * The syncer vnode is no referenced.
3376 sync_inactive(struct vop_inactive_args *ap)
3384 * The syncer vnode is no longer needed and is being decommissioned.
3386 * Modifications to the worklist must be protected by sync_mtx.
3389 sync_reclaim(struct vop_reclaim_args *ap)
3391 struct vnode *vp = ap->a_vp;
3396 vp->v_mount->mnt_syncer = NULL;
3397 if (bo->bo_flag & BO_ONWORKLST) {
3398 mtx_lock(&sync_mtx);
3399 LIST_REMOVE(bo, bo_synclist);
3400 syncer_worklist_len--;
3402 mtx_unlock(&sync_mtx);
3403 bo->bo_flag &= ~BO_ONWORKLST;
3411 * Check if vnode represents a disk device
3414 vn_isdisk(struct vnode *vp, int *errp)
3420 if (vp->v_type != VCHR)
3422 else if (vp->v_rdev == NULL)
3424 else if (vp->v_rdev->si_devsw == NULL)
3426 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3431 return (error == 0);
3435 * Common filesystem object access control check routine. Accepts a
3436 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3437 * and optional call-by-reference privused argument allowing vaccess()
3438 * to indicate to the caller whether privilege was used to satisfy the
3439 * request (obsoleted). Returns 0 on success, or an errno on failure.
3441 * The ifdef'd CAPABILITIES version is here for reference, but is not
3445 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3446 mode_t acc_mode, struct ucred *cred, int *privused)
3449 mode_t priv_granted;
3452 * Look for a normal, non-privileged way to access the file/directory
3453 * as requested. If it exists, go with that.
3456 if (privused != NULL)
3461 /* Check the owner. */
3462 if (cred->cr_uid == file_uid) {
3463 dac_granted |= VADMIN;
3464 if (file_mode & S_IXUSR)
3465 dac_granted |= VEXEC;
3466 if (file_mode & S_IRUSR)
3467 dac_granted |= VREAD;
3468 if (file_mode & S_IWUSR)
3469 dac_granted |= (VWRITE | VAPPEND);
3471 if ((acc_mode & dac_granted) == acc_mode)
3477 /* Otherwise, check the groups (first match) */
3478 if (groupmember(file_gid, cred)) {
3479 if (file_mode & S_IXGRP)
3480 dac_granted |= VEXEC;
3481 if (file_mode & S_IRGRP)
3482 dac_granted |= VREAD;
3483 if (file_mode & S_IWGRP)
3484 dac_granted |= (VWRITE | VAPPEND);
3486 if ((acc_mode & dac_granted) == acc_mode)
3492 /* Otherwise, check everyone else. */
3493 if (file_mode & S_IXOTH)
3494 dac_granted |= VEXEC;
3495 if (file_mode & S_IROTH)
3496 dac_granted |= VREAD;
3497 if (file_mode & S_IWOTH)
3498 dac_granted |= (VWRITE | VAPPEND);
3499 if ((acc_mode & dac_granted) == acc_mode)
3504 * Build a privilege mask to determine if the set of privileges
3505 * satisfies the requirements when combined with the granted mask
3506 * from above. For each privilege, if the privilege is required,
3507 * bitwise or the request type onto the priv_granted mask.
3513 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3514 * requests, instead of PRIV_VFS_EXEC.
3516 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3517 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3518 priv_granted |= VEXEC;
3520 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3521 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3522 priv_granted |= VEXEC;
3525 if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3526 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3527 priv_granted |= VREAD;
3529 if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3530 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3531 priv_granted |= (VWRITE | VAPPEND);
3533 if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3534 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3535 priv_granted |= VADMIN;
3537 if ((acc_mode & (priv_granted | dac_granted)) == acc_mode) {
3538 /* XXX audit: privilege used */
3539 if (privused != NULL)
3544 return ((acc_mode & VADMIN) ? EPERM : EACCES);
3548 * Credential check based on process requesting service, and per-attribute
3552 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3553 struct thread *td, int access)
3557 * Kernel-invoked always succeeds.
3563 * Do not allow privileged processes in jail to directly manipulate
3564 * system attributes.
3566 switch (attrnamespace) {
3567 case EXTATTR_NAMESPACE_SYSTEM:
3568 /* Potentially should be: return (EPERM); */
3569 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3570 case EXTATTR_NAMESPACE_USER:
3571 return (VOP_ACCESS(vp, access, cred, td));
3577 #ifdef DEBUG_VFS_LOCKS
3579 * This only exists to supress warnings from unlocked specfs accesses. It is
3580 * no longer ok to have an unlocked VFS.
3582 #define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
3584 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3585 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3587 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3588 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3590 int vfs_badlock_print = 1; /* Print lock violations. */
3591 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3594 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3595 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3599 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3603 if (vfs_badlock_backtrace)
3606 if (vfs_badlock_print)
3607 printf("%s: %p %s\n", str, (void *)vp, msg);
3608 if (vfs_badlock_ddb)
3609 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3613 assert_vi_locked(struct vnode *vp, const char *str)
3616 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3617 vfs_badlock("interlock is not locked but should be", str, vp);
3621 assert_vi_unlocked(struct vnode *vp, const char *str)
3624 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3625 vfs_badlock("interlock is locked but should not be", str, vp);
3629 assert_vop_locked(struct vnode *vp, const char *str)
3632 if (vp && !IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3633 vfs_badlock("is not locked but should be", str, vp);
3637 assert_vop_unlocked(struct vnode *vp, const char *str)
3640 if (vp && !IGNORE_LOCK(vp) &&
3641 VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3642 vfs_badlock("is locked but should not be", str, vp);
3646 assert_vop_elocked(struct vnode *vp, const char *str)
3649 if (vp && !IGNORE_LOCK(vp) &&
3650 VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3651 vfs_badlock("is not exclusive locked but should be", str, vp);
3656 assert_vop_elocked_other(struct vnode *vp, const char *str)
3659 if (vp && !IGNORE_LOCK(vp) &&
3660 VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3661 vfs_badlock("is not exclusive locked by another thread",
3666 assert_vop_slocked(struct vnode *vp, const char *str)
3669 if (vp && !IGNORE_LOCK(vp) &&
3670 VOP_ISLOCKED(vp) != LK_SHARED)
3671 vfs_badlock("is not locked shared but should be", str, vp);
3674 #endif /* DEBUG_VFS_LOCKS */
3677 vop_rename_pre(void *ap)
3679 struct vop_rename_args *a = ap;
3681 #ifdef DEBUG_VFS_LOCKS
3683 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3684 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3685 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3686 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3688 /* Check the source (from). */
3689 if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp)
3690 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3691 if (a->a_tvp != a->a_fvp)
3692 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3694 /* Check the target. */
3696 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3697 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3699 if (a->a_tdvp != a->a_fdvp)
3701 if (a->a_tvp != a->a_fvp)
3709 vop_strategy_pre(void *ap)
3711 #ifdef DEBUG_VFS_LOCKS
3712 struct vop_strategy_args *a;
3719 * Cluster ops lock their component buffers but not the IO container.
3721 if ((bp->b_flags & B_CLUSTER) != 0)
3724 if (!BUF_ISLOCKED(bp)) {
3725 if (vfs_badlock_print)
3727 "VOP_STRATEGY: bp is not locked but should be\n");
3728 if (vfs_badlock_ddb)
3729 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3735 vop_lookup_pre(void *ap)
3737 #ifdef DEBUG_VFS_LOCKS
3738 struct vop_lookup_args *a;
3743 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3744 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3749 vop_lookup_post(void *ap, int rc)
3751 #ifdef DEBUG_VFS_LOCKS
3752 struct vop_lookup_args *a;
3760 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3761 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3764 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3769 vop_lock_pre(void *ap)
3771 #ifdef DEBUG_VFS_LOCKS
3772 struct vop_lock1_args *a = ap;
3774 if ((a->a_flags & LK_INTERLOCK) == 0)
3775 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3777 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3782 vop_lock_post(void *ap, int rc)
3784 #ifdef DEBUG_VFS_LOCKS
3785 struct vop_lock1_args *a = ap;
3787 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3789 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3794 vop_unlock_pre(void *ap)
3796 #ifdef DEBUG_VFS_LOCKS
3797 struct vop_unlock_args *a = ap;
3799 if (a->a_flags & LK_INTERLOCK)
3800 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3801 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3806 vop_unlock_post(void *ap, int rc)
3808 #ifdef DEBUG_VFS_LOCKS
3809 struct vop_unlock_args *a = ap;
3811 if (a->a_flags & LK_INTERLOCK)
3812 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3817 vop_create_post(void *ap, int rc)
3819 struct vop_create_args *a = ap;
3822 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3826 vop_link_post(void *ap, int rc)
3828 struct vop_link_args *a = ap;
3831 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3832 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3837 vop_mkdir_post(void *ap, int rc)
3839 struct vop_mkdir_args *a = ap;
3842 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3846 vop_mknod_post(void *ap, int rc)
3848 struct vop_mknod_args *a = ap;
3851 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3855 vop_remove_post(void *ap, int rc)
3857 struct vop_remove_args *a = ap;
3860 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3861 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3866 vop_rename_post(void *ap, int rc)
3868 struct vop_rename_args *a = ap;
3871 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3872 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3873 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3875 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3877 if (a->a_tdvp != a->a_fdvp)
3879 if (a->a_tvp != a->a_fvp)
3887 vop_rmdir_post(void *ap, int rc)
3889 struct vop_rmdir_args *a = ap;
3892 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3893 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3898 vop_setattr_post(void *ap, int rc)
3900 struct vop_setattr_args *a = ap;
3903 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
3907 vop_symlink_post(void *ap, int rc)
3909 struct vop_symlink_args *a = ap;
3912 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3915 static struct knlist fs_knlist;
3918 vfs_event_init(void *arg)
3920 knlist_init(&fs_knlist, NULL, NULL, NULL, NULL);
3922 /* XXX - correct order? */
3923 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
3926 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
3929 KNOTE_UNLOCKED(&fs_knlist, event);
3932 static int filt_fsattach(struct knote *kn);
3933 static void filt_fsdetach(struct knote *kn);
3934 static int filt_fsevent(struct knote *kn, long hint);
3936 struct filterops fs_filtops =
3937 { 0, filt_fsattach, filt_fsdetach, filt_fsevent };
3940 filt_fsattach(struct knote *kn)
3943 kn->kn_flags |= EV_CLEAR;
3944 knlist_add(&fs_knlist, kn, 0);
3949 filt_fsdetach(struct knote *kn)
3952 knlist_remove(&fs_knlist, kn, 0);
3956 filt_fsevent(struct knote *kn, long hint)
3959 kn->kn_fflags |= hint;
3960 return (kn->kn_fflags != 0);
3964 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
3970 error = SYSCTL_IN(req, &vc, sizeof(vc));
3973 if (vc.vc_vers != VFS_CTL_VERS1)
3975 mp = vfs_getvfs(&vc.vc_fsid);
3978 /* ensure that a specific sysctl goes to the right filesystem. */
3979 if (strcmp(vc.vc_fstypename, "*") != 0 &&
3980 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
3984 VCTLTOREQ(&vc, req);
3985 error = VFS_SYSCTL(mp, vc.vc_op, req);
3990 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
3994 * Function to initialize a va_filerev field sensibly.
3995 * XXX: Wouldn't a random number make a lot more sense ??
3998 init_va_filerev(void)
4003 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4006 static int filt_vfsread(struct knote *kn, long hint);
4007 static int filt_vfswrite(struct knote *kn, long hint);
4008 static int filt_vfsvnode(struct knote *kn, long hint);
4009 static void filt_vfsdetach(struct knote *kn);
4010 static struct filterops vfsread_filtops =
4011 { 1, NULL, filt_vfsdetach, filt_vfsread };
4012 static struct filterops vfswrite_filtops =
4013 { 1, NULL, filt_vfsdetach, filt_vfswrite };
4014 static struct filterops vfsvnode_filtops =
4015 { 1, NULL, filt_vfsdetach, filt_vfsvnode };
4018 vfs_knllock(void *arg)
4020 struct vnode *vp = arg;
4022 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4026 vfs_knlunlock(void *arg)
4028 struct vnode *vp = arg;
4034 vfs_knllocked(void *arg)
4036 struct vnode *vp = arg;
4038 return (VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
4042 vfs_kqfilter(struct vop_kqfilter_args *ap)
4044 struct vnode *vp = ap->a_vp;
4045 struct knote *kn = ap->a_kn;
4048 switch (kn->kn_filter) {
4050 kn->kn_fop = &vfsread_filtops;
4053 kn->kn_fop = &vfswrite_filtops;
4056 kn->kn_fop = &vfsvnode_filtops;
4062 kn->kn_hook = (caddr_t)vp;
4064 if (vp->v_pollinfo == NULL)
4066 if (vp->v_pollinfo == NULL)
4068 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4069 knlist_add(knl, kn, 0);
4075 * Detach knote from vnode
4078 filt_vfsdetach(struct knote *kn)
4080 struct vnode *vp = (struct vnode *)kn->kn_hook;
4082 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4083 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4088 filt_vfsread(struct knote *kn, long hint)
4090 struct vnode *vp = (struct vnode *)kn->kn_hook;
4094 * filesystem is gone, so set the EOF flag and schedule
4095 * the knote for deletion.
4097 if (hint == NOTE_REVOKE) {
4098 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4102 if (VOP_GETATTR(vp, &va, curthread->td_ucred, curthread))
4105 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4106 return (kn->kn_data != 0);
4111 filt_vfswrite(struct knote *kn, long hint)
4114 * filesystem is gone, so set the EOF flag and schedule
4115 * the knote for deletion.
4117 if (hint == NOTE_REVOKE)
4118 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4125 filt_vfsvnode(struct knote *kn, long hint)
4127 if (kn->kn_sfflags & hint)
4128 kn->kn_fflags |= hint;
4129 if (hint == NOTE_REVOKE) {
4130 kn->kn_flags |= EV_EOF;
4133 return (kn->kn_fflags != 0);
4137 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4141 if (dp->d_reclen > ap->a_uio->uio_resid)
4142 return (ENAMETOOLONG);
4143 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4145 if (ap->a_ncookies != NULL) {
4146 if (ap->a_cookies != NULL)
4147 free(ap->a_cookies, M_TEMP);
4148 ap->a_cookies = NULL;
4149 *ap->a_ncookies = 0;
4153 if (ap->a_ncookies == NULL)
4156 KASSERT(ap->a_cookies,
4157 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4159 *ap->a_cookies = realloc(*ap->a_cookies,
4160 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4161 (*ap->a_cookies)[*ap->a_ncookies] = off;
4166 * Mark for update the access time of the file if the filesystem
4167 * supports VA_MARK_ATIME. This functionality is used by execve
4168 * and mmap, so we want to avoid the synchronous I/O implied by
4169 * directly setting va_atime for the sake of efficiency.
4172 vfs_mark_atime(struct vnode *vp, struct thread *td)
4174 struct vattr atimeattr;
4176 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
4177 VATTR_NULL(&atimeattr);
4178 atimeattr.va_vaflags |= VA_MARK_ATIME;
4179 (void)VOP_SETATTR(vp, &atimeattr, td->td_ucred, td);