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/malloc.h>
63 #include <sys/mount.h>
64 #include <sys/namei.h>
66 #include <sys/reboot.h>
67 #include <sys/sleepqueue.h>
69 #include <sys/sysctl.h>
70 #include <sys/syslog.h>
71 #include <sys/vmmeter.h>
72 #include <sys/vnode.h>
74 #include <machine/stdarg.h>
76 #include <security/mac/mac_framework.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_extern.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_kern.h>
91 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure");
93 static void delmntque(struct vnode *vp);
94 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
95 int slpflag, int slptimeo);
96 static void syncer_shutdown(void *arg, int howto);
97 static int vtryrecycle(struct vnode *vp);
98 static void vbusy(struct vnode *vp);
99 static void vinactive(struct vnode *, struct thread *);
100 static void v_incr_usecount(struct vnode *);
101 static void v_decr_usecount(struct vnode *);
102 static void v_decr_useonly(struct vnode *);
103 static void v_upgrade_usecount(struct vnode *);
104 static void vfree(struct vnode *);
105 static void vnlru_free(int);
106 static void vdestroy(struct vnode *);
107 static void vgonel(struct vnode *);
108 static void vfs_knllock(void *arg);
109 static void vfs_knlunlock(void *arg);
110 static int vfs_knllocked(void *arg);
114 * Enable Giant pushdown based on whether or not the vm is mpsafe in this
115 * build. Without mpsafevm the buffer cache can not run Giant free.
118 TUNABLE_INT("debug.mpsafevfs", &mpsafe_vfs);
119 SYSCTL_INT(_debug, OID_AUTO, mpsafevfs, CTLFLAG_RD, &mpsafe_vfs, 0,
123 * Number of vnodes in existence. Increased whenever getnewvnode()
124 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
127 static unsigned long numvnodes;
129 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
132 * Conversion tables for conversion from vnode types to inode formats
135 enum vtype iftovt_tab[16] = {
136 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
137 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
139 int vttoif_tab[10] = {
140 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
141 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
145 * List of vnodes that are ready for recycling.
147 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
150 * Free vnode target. Free vnodes may simply be files which have been stat'd
151 * but not read. This is somewhat common, and a small cache of such files
152 * should be kept to avoid recreation costs.
154 static u_long wantfreevnodes;
155 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
156 /* Number of vnodes in the free list. */
157 static u_long freevnodes;
158 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
161 * Various variables used for debugging the new implementation of
163 * XXX these are probably of (very) limited utility now.
165 static int reassignbufcalls;
166 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
169 * Cache for the mount type id assigned to NFS. This is used for
170 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
172 int nfs_mount_type = -1;
174 /* To keep more than one thread at a time from running vfs_getnewfsid */
175 static struct mtx mntid_mtx;
178 * Lock for any access to the following:
183 static struct mtx vnode_free_list_mtx;
185 /* Publicly exported FS */
186 struct nfs_public nfs_pub;
188 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
189 static uma_zone_t vnode_zone;
190 static uma_zone_t vnodepoll_zone;
192 /* Set to 1 to print out reclaim of active vnodes */
196 * The workitem queue.
198 * It is useful to delay writes of file data and filesystem metadata
199 * for tens of seconds so that quickly created and deleted files need
200 * not waste disk bandwidth being created and removed. To realize this,
201 * we append vnodes to a "workitem" queue. When running with a soft
202 * updates implementation, most pending metadata dependencies should
203 * not wait for more than a few seconds. Thus, mounted on block devices
204 * are delayed only about a half the time that file data is delayed.
205 * Similarly, directory updates are more critical, so are only delayed
206 * about a third the time that file data is delayed. Thus, there are
207 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
208 * one each second (driven off the filesystem syncer process). The
209 * syncer_delayno variable indicates the next queue that is to be processed.
210 * Items that need to be processed soon are placed in this queue:
212 * syncer_workitem_pending[syncer_delayno]
214 * A delay of fifteen seconds is done by placing the request fifteen
215 * entries later in the queue:
217 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
220 static int syncer_delayno;
221 static long syncer_mask;
222 LIST_HEAD(synclist, bufobj);
223 static struct synclist *syncer_workitem_pending;
225 * The sync_mtx protects:
230 * syncer_workitem_pending
231 * syncer_worklist_len
234 static struct mtx sync_mtx;
236 #define SYNCER_MAXDELAY 32
237 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
238 static int syncdelay = 30; /* max time to delay syncing data */
239 static int filedelay = 30; /* time to delay syncing files */
240 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
241 static int dirdelay = 29; /* time to delay syncing directories */
242 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
243 static int metadelay = 28; /* time to delay syncing metadata */
244 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
245 static int rushjob; /* number of slots to run ASAP */
246 static int stat_rush_requests; /* number of times I/O speeded up */
247 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
250 * When shutting down the syncer, run it at four times normal speed.
252 #define SYNCER_SHUTDOWN_SPEEDUP 4
253 static int sync_vnode_count;
254 static int syncer_worklist_len;
255 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
259 * Number of vnodes we want to exist at any one time. This is mostly used
260 * to size hash tables in vnode-related code. It is normally not used in
261 * getnewvnode(), as wantfreevnodes is normally nonzero.)
263 * XXX desiredvnodes is historical cruft and should not exist.
266 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
267 &desiredvnodes, 0, "Maximum number of vnodes");
268 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
269 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
270 static int vnlru_nowhere;
271 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
272 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
275 * Macros to control when a vnode is freed and recycled. All require
276 * the vnode interlock.
278 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
279 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
280 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
284 * Initialize the vnode management data structures.
286 #ifndef MAXVNODES_MAX
287 #define MAXVNODES_MAX 100000
290 vntblinit(void *dummy __unused)
294 * Desiredvnodes is a function of the physical memory size and
295 * the kernel's heap size. Specifically, desiredvnodes scales
296 * in proportion to the physical memory size until two fifths
297 * of the kernel's heap size is consumed by vnodes and vm
300 desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
301 (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
302 if (desiredvnodes > MAXVNODES_MAX) {
304 printf("Reducing kern.maxvnodes %d -> %d\n",
305 desiredvnodes, MAXVNODES_MAX);
306 desiredvnodes = MAXVNODES_MAX;
308 wantfreevnodes = desiredvnodes / 4;
309 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
310 TAILQ_INIT(&vnode_free_list);
311 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
312 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
313 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
314 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
315 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
317 * Initialize the filesystem syncer.
319 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
321 syncer_maxdelay = syncer_mask + 1;
322 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
324 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
328 * Mark a mount point as busy. Used to synchronize access and to delay
329 * unmounting. Interlock is not released on failure.
332 vfs_busy(struct mount *mp, int flags, struct mtx *interlkp,
339 if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
340 if (flags & LK_NOWAIT) {
346 mtx_unlock(interlkp);
347 mp->mnt_kern_flag |= MNTK_MWAIT;
349 * Since all busy locks are shared except the exclusive
350 * lock granted when unmounting, the only place that a
351 * wakeup needs to be done is at the release of the
352 * exclusive lock at the end of dounmount.
354 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
362 mtx_unlock(interlkp);
363 lkflags = LK_SHARED | LK_INTERLOCK;
364 if (lockmgr(&mp->mnt_lock, lkflags, MNT_MTX(mp), td))
365 panic("vfs_busy: unexpected lock failure");
370 * Free a busy filesystem.
373 vfs_unbusy(struct mount *mp, struct thread *td)
376 lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
381 * Lookup a mount point by filesystem identifier.
384 vfs_getvfs(fsid_t *fsid)
388 mtx_lock(&mountlist_mtx);
389 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
390 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
391 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
393 mtx_unlock(&mountlist_mtx);
397 mtx_unlock(&mountlist_mtx);
398 return ((struct mount *) 0);
402 * Check if a user can access privileged mount options.
405 vfs_suser(struct mount *mp, struct thread *td)
410 * If the thread is jailed, but this is not a jail-friendly file
411 * system, deny immediately.
413 if (jailed(td->td_ucred) && !(mp->mnt_vfc->vfc_flags & VFCF_JAIL))
417 * If the file system was mounted outside a jail and a jailed thread
418 * tries to access it, deny immediately.
420 if (!jailed(mp->mnt_cred) && jailed(td->td_ucred))
424 * If the file system was mounted inside different jail that the jail of
425 * the calling thread, deny immediately.
427 if (jailed(mp->mnt_cred) && jailed(td->td_ucred) &&
428 mp->mnt_cred->cr_prison != td->td_ucred->cr_prison) {
432 if ((mp->mnt_flag & MNT_USER) == 0 ||
433 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
434 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
441 * Get a new unique fsid. Try to make its val[0] unique, since this value
442 * will be used to create fake device numbers for stat(). Also try (but
443 * not so hard) make its val[0] unique mod 2^16, since some emulators only
444 * support 16-bit device numbers. We end up with unique val[0]'s for the
445 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
447 * Keep in mind that several mounts may be running in parallel. Starting
448 * the search one past where the previous search terminated is both a
449 * micro-optimization and a defense against returning the same fsid to
453 vfs_getnewfsid(struct mount *mp)
455 static u_int16_t mntid_base;
460 mtx_lock(&mntid_mtx);
461 mtype = mp->mnt_vfc->vfc_typenum;
462 tfsid.val[1] = mtype;
463 mtype = (mtype & 0xFF) << 24;
465 tfsid.val[0] = makedev(255,
466 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
468 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
472 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
473 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
474 mtx_unlock(&mntid_mtx);
478 * Knob to control the precision of file timestamps:
480 * 0 = seconds only; nanoseconds zeroed.
481 * 1 = seconds and nanoseconds, accurate within 1/HZ.
482 * 2 = seconds and nanoseconds, truncated to microseconds.
483 * >=3 = seconds and nanoseconds, maximum precision.
485 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
487 static int timestamp_precision = TSP_SEC;
488 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
489 ×tamp_precision, 0, "");
492 * Get a current timestamp.
495 vfs_timestamp(struct timespec *tsp)
499 switch (timestamp_precision) {
501 tsp->tv_sec = time_second;
509 TIMEVAL_TO_TIMESPEC(&tv, tsp);
519 * Set vnode attributes to VNOVAL
522 vattr_null(struct vattr *vap)
526 vap->va_size = VNOVAL;
527 vap->va_bytes = VNOVAL;
528 vap->va_mode = VNOVAL;
529 vap->va_nlink = VNOVAL;
530 vap->va_uid = VNOVAL;
531 vap->va_gid = VNOVAL;
532 vap->va_fsid = VNOVAL;
533 vap->va_fileid = VNOVAL;
534 vap->va_blocksize = VNOVAL;
535 vap->va_rdev = VNOVAL;
536 vap->va_atime.tv_sec = VNOVAL;
537 vap->va_atime.tv_nsec = VNOVAL;
538 vap->va_mtime.tv_sec = VNOVAL;
539 vap->va_mtime.tv_nsec = VNOVAL;
540 vap->va_ctime.tv_sec = VNOVAL;
541 vap->va_ctime.tv_nsec = VNOVAL;
542 vap->va_birthtime.tv_sec = VNOVAL;
543 vap->va_birthtime.tv_nsec = VNOVAL;
544 vap->va_flags = VNOVAL;
545 vap->va_gen = VNOVAL;
550 * This routine is called when we have too many vnodes. It attempts
551 * to free <count> vnodes and will potentially free vnodes that still
552 * have VM backing store (VM backing store is typically the cause
553 * of a vnode blowout so we want to do this). Therefore, this operation
554 * is not considered cheap.
556 * A number of conditions may prevent a vnode from being reclaimed.
557 * the buffer cache may have references on the vnode, a directory
558 * vnode may still have references due to the namei cache representing
559 * underlying files, or the vnode may be in active use. It is not
560 * desireable to reuse such vnodes. These conditions may cause the
561 * number of vnodes to reach some minimum value regardless of what
562 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
565 vlrureclaim(struct mount *mp)
575 * Calculate the trigger point, don't allow user
576 * screwups to blow us up. This prevents us from
577 * recycling vnodes with lots of resident pages. We
578 * aren't trying to free memory, we are trying to
581 usevnodes = desiredvnodes;
584 trigger = cnt.v_page_count * 2 / usevnodes;
587 vn_start_write(NULL, &mp, V_WAIT);
589 count = mp->mnt_nvnodelistsize / 10 + 1;
591 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
592 while (vp != NULL && vp->v_type == VMARKER)
593 vp = TAILQ_NEXT(vp, v_nmntvnodes);
596 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
597 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
602 * If it's been deconstructed already, it's still
603 * referenced, or it exceeds the trigger, skip it.
605 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
606 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
607 vp->v_object->resident_page_count > trigger)) {
613 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT, td)) {
615 goto next_iter_mntunlocked;
619 * v_usecount may have been bumped after VOP_LOCK() dropped
620 * the vnode interlock and before it was locked again.
622 * It is not necessary to recheck VI_DOOMED because it can
623 * only be set by another thread that holds both the vnode
624 * lock and vnode interlock. If another thread has the
625 * vnode lock before we get to VOP_LOCK() and obtains the
626 * vnode interlock after VOP_LOCK() drops the vnode
627 * interlock, the other thread will be unable to drop the
628 * vnode lock before our VOP_LOCK() call fails.
630 if (vp->v_usecount || !LIST_EMPTY(&(vp)->v_cache_src) ||
631 (vp->v_object != NULL &&
632 vp->v_object->resident_page_count > trigger)) {
633 VOP_UNLOCK(vp, LK_INTERLOCK, td);
634 goto next_iter_mntunlocked;
636 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
637 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
639 VOP_UNLOCK(vp, 0, td);
642 next_iter_mntunlocked:
643 if ((count % 256) != 0)
647 if ((count % 256) != 0)
656 vn_finished_write(mp);
661 * Attempt to keep the free list at wantfreevnodes length.
664 vnlru_free(int count)
669 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
670 for (; count > 0; count--) {
671 vp = TAILQ_FIRST(&vnode_free_list);
673 * The list can be modified while the free_list_mtx
674 * has been dropped and vp could be NULL here.
678 VNASSERT(vp->v_op != NULL, vp,
679 ("vnlru_free: vnode already reclaimed."));
680 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
682 * Don't recycle if we can't get the interlock.
684 if (!VI_TRYLOCK(vp)) {
685 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
688 VNASSERT(VCANRECYCLE(vp), vp,
689 ("vp inconsistent on freelist"));
691 vp->v_iflag &= ~VI_FREE;
693 mtx_unlock(&vnode_free_list_mtx);
695 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
697 VFS_UNLOCK_GIANT(vfslocked);
699 * If the recycled succeeded this vdrop will actually free
700 * the vnode. If not it will simply place it back on
704 mtx_lock(&vnode_free_list_mtx);
708 * Attempt to recycle vnodes in a context that is always safe to block.
709 * Calling vlrurecycle() from the bowels of filesystem code has some
710 * interesting deadlock problems.
712 static struct proc *vnlruproc;
713 static int vnlruproc_sig;
718 struct mount *mp, *nmp;
720 struct proc *p = vnlruproc;
721 struct thread *td = FIRST_THREAD_IN_PROC(p);
725 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
729 kthread_suspend_check(p);
730 mtx_lock(&vnode_free_list_mtx);
731 if (freevnodes > wantfreevnodes)
732 vnlru_free(freevnodes - wantfreevnodes);
733 if (numvnodes <= desiredvnodes * 9 / 10) {
735 wakeup(&vnlruproc_sig);
736 msleep(vnlruproc, &vnode_free_list_mtx,
737 PVFS|PDROP, "vlruwt", hz);
740 mtx_unlock(&vnode_free_list_mtx);
742 mtx_lock(&mountlist_mtx);
743 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
745 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
746 nmp = TAILQ_NEXT(mp, mnt_list);
749 if (!VFS_NEEDSGIANT(mp)) {
754 done += vlrureclaim(mp);
757 mtx_lock(&mountlist_mtx);
758 nmp = TAILQ_NEXT(mp, mnt_list);
761 mtx_unlock(&mountlist_mtx);
763 EVENTHANDLER_INVOKE(vfs_lowvnodes, desiredvnodes / 10);
765 /* These messages are temporary debugging aids */
766 if (vnlru_nowhere < 5)
767 printf("vnlru process getting nowhere..\n");
768 else if (vnlru_nowhere == 5)
769 printf("vnlru process messages stopped.\n");
772 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
778 static struct kproc_desc vnlru_kp = {
783 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
786 * Routines having to do with the management of the vnode table.
790 vdestroy(struct vnode *vp)
794 CTR1(KTR_VFS, "vdestroy vp %p", vp);
795 mtx_lock(&vnode_free_list_mtx);
797 mtx_unlock(&vnode_free_list_mtx);
799 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
800 ("cleaned vnode still on the free list."));
801 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
802 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
803 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
804 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
805 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
806 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
807 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
808 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
809 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
810 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
811 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
814 mac_destroy_vnode(vp);
816 if (vp->v_pollinfo != NULL) {
817 knlist_destroy(&vp->v_pollinfo->vpi_selinfo.si_note);
818 mtx_destroy(&vp->v_pollinfo->vpi_lock);
819 uma_zfree(vnodepoll_zone, vp->v_pollinfo);
822 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
825 lockdestroy(vp->v_vnlock);
826 mtx_destroy(&vp->v_interlock);
827 uma_zfree(vnode_zone, vp);
831 * Try to recycle a freed vnode. We abort if anyone picks up a reference
832 * before we actually vgone(). This function must be called with the vnode
833 * held to prevent the vnode from being returned to the free list midway
837 vtryrecycle(struct vnode *vp)
839 struct thread *td = curthread;
842 CTR1(KTR_VFS, "vtryrecycle: trying vp %p", vp);
843 VNASSERT(vp->v_holdcnt, vp,
844 ("vtryrecycle: Recycling vp %p without a reference.", vp));
846 * This vnode may found and locked via some other list, if so we
847 * can't recycle it yet.
849 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
850 return (EWOULDBLOCK);
852 * Don't recycle if its filesystem is being suspended.
854 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
855 VOP_UNLOCK(vp, 0, td);
859 * If we got this far, we need to acquire the interlock and see if
860 * anyone picked up this vnode from another list. If not, we will
861 * mark it with DOOMED via vgonel() so that anyone who does find it
865 if (vp->v_usecount) {
866 VOP_UNLOCK(vp, LK_INTERLOCK, td);
867 vn_finished_write(vnmp);
870 if ((vp->v_iflag & VI_DOOMED) == 0)
872 VOP_UNLOCK(vp, LK_INTERLOCK, td);
873 vn_finished_write(vnmp);
874 CTR1(KTR_VFS, "vtryrecycle: recycled vp %p", vp);
879 * Return the next vnode from the free list.
882 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
885 struct vnode *vp = NULL;
888 mtx_lock(&vnode_free_list_mtx);
890 * Lend our context to reclaim vnodes if they've exceeded the max.
892 if (freevnodes > wantfreevnodes)
895 * Wait for available vnodes.
897 if (numvnodes > desiredvnodes) {
898 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
900 * File system is beeing suspended, we cannot risk a
901 * deadlock here, so allocate new vnode anyway.
903 if (freevnodes > wantfreevnodes)
904 vnlru_free(freevnodes - wantfreevnodes);
907 if (vnlruproc_sig == 0) {
908 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
911 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
913 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
914 if (numvnodes > desiredvnodes) {
915 mtx_unlock(&vnode_free_list_mtx);
922 mtx_unlock(&vnode_free_list_mtx);
923 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
927 vp->v_vnlock = &vp->v_lock;
928 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
930 * By default, don't allow shared locks unless filesystems
933 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
939 bo->bo_mtx = &vp->v_interlock;
940 bo->bo_ops = &buf_ops_bio;
942 TAILQ_INIT(&bo->bo_clean.bv_hd);
943 TAILQ_INIT(&bo->bo_dirty.bv_hd);
945 * Initialize namecache.
947 LIST_INIT(&vp->v_cache_src);
948 TAILQ_INIT(&vp->v_cache_dst);
950 * Finalize various vnode identity bits.
959 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
960 mac_associate_vnode_singlelabel(mp, vp);
962 printf("NULL mp in getnewvnode()\n");
965 bo->bo_bsize = mp->mnt_stat.f_iosize;
966 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
967 vp->v_vflag |= VV_NOKNOTE;
970 CTR2(KTR_VFS, "getnewvnode: mp %p vp %p", mp, vp);
976 * Delete from old mount point vnode list, if on one.
979 delmntque(struct vnode *vp)
988 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
989 ("bad mount point vnode list size"));
990 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
991 mp->mnt_nvnodelistsize--;
997 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1001 td = curthread; /* XXX ? */
1003 vp->v_op = &dead_vnodeops;
1004 /* XXX non mp-safe fs may still call insmntque with vnode
1006 if (!VOP_ISLOCKED(vp, td))
1007 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1013 * Insert into list of vnodes for the new mount point, if available.
1016 insmntque1(struct vnode *vp, struct mount *mp,
1017 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1020 KASSERT(vp->v_mount == NULL,
1021 ("insmntque: vnode already on per mount vnode list"));
1022 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1024 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 &&
1025 mp->mnt_nvnodelistsize == 0) {
1033 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1034 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1035 ("neg mount point vnode list size"));
1036 mp->mnt_nvnodelistsize++;
1042 insmntque(struct vnode *vp, struct mount *mp)
1045 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1049 * Flush out and invalidate all buffers associated with a bufobj
1050 * Called with the underlying object locked.
1053 bufobj_invalbuf(struct bufobj *bo, int flags, struct thread *td, int slpflag,
1059 if (flags & V_SAVE) {
1060 error = bufobj_wwait(bo, slpflag, slptimeo);
1065 if (bo->bo_dirty.bv_cnt > 0) {
1067 if ((error = BO_SYNC(bo, MNT_WAIT, td)) != 0)
1070 * XXX We could save a lock/unlock if this was only
1071 * enabled under INVARIANTS
1074 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1075 panic("vinvalbuf: dirty bufs");
1079 * If you alter this loop please notice that interlock is dropped and
1080 * reacquired in flushbuflist. Special care is needed to ensure that
1081 * no race conditions occur from this.
1084 error = flushbuflist(&bo->bo_clean,
1085 flags, bo, slpflag, slptimeo);
1087 error = flushbuflist(&bo->bo_dirty,
1088 flags, bo, slpflag, slptimeo);
1089 if (error != 0 && error != EAGAIN) {
1093 } while (error != 0);
1096 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1097 * have write I/O in-progress but if there is a VM object then the
1098 * VM object can also have read-I/O in-progress.
1101 bufobj_wwait(bo, 0, 0);
1103 if (bo->bo_object != NULL) {
1104 VM_OBJECT_LOCK(bo->bo_object);
1105 vm_object_pip_wait(bo->bo_object, "bovlbx");
1106 VM_OBJECT_UNLOCK(bo->bo_object);
1109 } while (bo->bo_numoutput > 0);
1113 * Destroy the copy in the VM cache, too.
1115 if (bo->bo_object != NULL) {
1116 VM_OBJECT_LOCK(bo->bo_object);
1117 vm_object_page_remove(bo->bo_object, 0, 0,
1118 (flags & V_SAVE) ? TRUE : FALSE);
1119 VM_OBJECT_UNLOCK(bo->bo_object);
1124 if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1125 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1126 panic("vinvalbuf: flush failed");
1133 * Flush out and invalidate all buffers associated with a vnode.
1134 * Called with the underlying object locked.
1137 vinvalbuf(struct vnode *vp, int flags, struct thread *td, int slpflag,
1141 CTR2(KTR_VFS, "vinvalbuf vp %p flags %d", vp, flags);
1142 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1143 return (bufobj_invalbuf(&vp->v_bufobj, flags, td, slpflag, slptimeo));
1147 * Flush out buffers on the specified list.
1151 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1154 struct buf *bp, *nbp;
1159 ASSERT_BO_LOCKED(bo);
1162 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1163 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1164 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1170 lblkno = nbp->b_lblkno;
1171 xflags = nbp->b_xflags &
1172 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1175 error = BUF_TIMELOCK(bp,
1176 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1177 "flushbuf", slpflag, slptimeo);
1180 return (error != ENOLCK ? error : EAGAIN);
1182 KASSERT(bp->b_bufobj == bo,
1183 ("bp %p wrong b_bufobj %p should be %p",
1184 bp, bp->b_bufobj, bo));
1185 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1191 * XXX Since there are no node locks for NFS, I
1192 * believe there is a slight chance that a delayed
1193 * write will occur while sleeping just above, so
1196 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1199 bp->b_flags |= B_ASYNC;
1202 return (EAGAIN); /* XXX: why not loop ? */
1205 bp->b_flags |= (B_INVAL | B_RELBUF);
1206 bp->b_flags &= ~B_ASYNC;
1210 (nbp->b_bufobj != bo ||
1211 nbp->b_lblkno != lblkno ||
1213 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1214 break; /* nbp invalid */
1220 * Truncate a file's buffer and pages to a specified length. This
1221 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1225 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1226 off_t length, int blksize)
1228 struct buf *bp, *nbp;
1233 CTR2(KTR_VFS, "vtruncbuf vp %p length %jd", vp, length);
1235 * Round up to the *next* lbn.
1237 trunclbn = (length + blksize - 1) / blksize;
1239 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1246 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1247 if (bp->b_lblkno < trunclbn)
1250 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1251 VI_MTX(vp)) == ENOLCK)
1255 bp->b_flags |= (B_INVAL | B_RELBUF);
1256 bp->b_flags &= ~B_ASYNC;
1261 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1262 (nbp->b_vp != vp) ||
1263 (nbp->b_flags & B_DELWRI))) {
1269 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1270 if (bp->b_lblkno < trunclbn)
1273 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1274 VI_MTX(vp)) == ENOLCK)
1277 bp->b_flags |= (B_INVAL | B_RELBUF);
1278 bp->b_flags &= ~B_ASYNC;
1282 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1283 (nbp->b_vp != vp) ||
1284 (nbp->b_flags & B_DELWRI) == 0)) {
1293 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1294 if (bp->b_lblkno > 0)
1297 * Since we hold the vnode lock this should only
1298 * fail if we're racing with the buf daemon.
1301 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1302 VI_MTX(vp)) == ENOLCK) {
1305 VNASSERT((bp->b_flags & B_DELWRI), vp,
1306 ("buf(%p) on dirty queue without DELWRI", bp));
1315 bufobj_wwait(bo, 0, 0);
1317 vnode_pager_setsize(vp, length);
1323 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1326 * NOTE: We have to deal with the special case of a background bitmap
1327 * buffer, a situation where two buffers will have the same logical
1328 * block offset. We want (1) only the foreground buffer to be accessed
1329 * in a lookup and (2) must differentiate between the foreground and
1330 * background buffer in the splay tree algorithm because the splay
1331 * tree cannot normally handle multiple entities with the same 'index'.
1332 * We accomplish this by adding differentiating flags to the splay tree's
1337 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1340 struct buf *lefttreemax, *righttreemin, *y;
1344 lefttreemax = righttreemin = &dummy;
1346 if (lblkno < root->b_lblkno ||
1347 (lblkno == root->b_lblkno &&
1348 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1349 if ((y = root->b_left) == NULL)
1351 if (lblkno < y->b_lblkno) {
1353 root->b_left = y->b_right;
1356 if ((y = root->b_left) == NULL)
1359 /* Link into the new root's right tree. */
1360 righttreemin->b_left = root;
1361 righttreemin = root;
1362 } else if (lblkno > root->b_lblkno ||
1363 (lblkno == root->b_lblkno &&
1364 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1365 if ((y = root->b_right) == NULL)
1367 if (lblkno > y->b_lblkno) {
1369 root->b_right = y->b_left;
1372 if ((y = root->b_right) == NULL)
1375 /* Link into the new root's left tree. */
1376 lefttreemax->b_right = root;
1383 /* Assemble the new root. */
1384 lefttreemax->b_right = root->b_left;
1385 righttreemin->b_left = root->b_right;
1386 root->b_left = dummy.b_right;
1387 root->b_right = dummy.b_left;
1392 buf_vlist_remove(struct buf *bp)
1397 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1398 ASSERT_BO_LOCKED(bp->b_bufobj);
1399 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1400 (BX_VNDIRTY|BX_VNCLEAN),
1401 ("buf_vlist_remove: Buf %p is on two lists", bp));
1402 if (bp->b_xflags & BX_VNDIRTY)
1403 bv = &bp->b_bufobj->bo_dirty;
1405 bv = &bp->b_bufobj->bo_clean;
1406 if (bp != bv->bv_root) {
1407 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1408 KASSERT(root == bp, ("splay lookup failed in remove"));
1410 if (bp->b_left == NULL) {
1413 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1414 root->b_right = bp->b_right;
1417 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1419 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1423 * Add the buffer to the sorted clean or dirty block list using a
1424 * splay tree algorithm.
1426 * NOTE: xflags is passed as a constant, optimizing this inline function!
1429 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1434 ASSERT_BO_LOCKED(bo);
1435 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1436 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1437 bp->b_xflags |= xflags;
1438 if (xflags & BX_VNDIRTY)
1443 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1447 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1448 } else if (bp->b_lblkno < root->b_lblkno ||
1449 (bp->b_lblkno == root->b_lblkno &&
1450 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1451 bp->b_left = root->b_left;
1453 root->b_left = NULL;
1454 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1456 bp->b_right = root->b_right;
1458 root->b_right = NULL;
1459 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1466 * Lookup a buffer using the splay tree. Note that we specifically avoid
1467 * shadow buffers used in background bitmap writes.
1469 * This code isn't quite efficient as it could be because we are maintaining
1470 * two sorted lists and do not know which list the block resides in.
1472 * During a "make buildworld" the desired buffer is found at one of
1473 * the roots more than 60% of the time. Thus, checking both roots
1474 * before performing either splay eliminates unnecessary splays on the
1475 * first tree splayed.
1478 gbincore(struct bufobj *bo, daddr_t lblkno)
1482 ASSERT_BO_LOCKED(bo);
1483 if ((bp = bo->bo_clean.bv_root) != NULL &&
1484 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1486 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1487 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1489 if ((bp = bo->bo_clean.bv_root) != NULL) {
1490 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1491 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1494 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1495 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1496 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1503 * Associate a buffer with a vnode.
1506 bgetvp(struct vnode *vp, struct buf *bp)
1509 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1511 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1512 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1513 ("bgetvp: bp already attached! %p", bp));
1515 ASSERT_VI_LOCKED(vp, "bgetvp");
1517 if (VFS_NEEDSGIANT(vp->v_mount) ||
1518 vp->v_bufobj.bo_flag & BO_NEEDSGIANT)
1519 bp->b_flags |= B_NEEDSGIANT;
1521 bp->b_bufobj = &vp->v_bufobj;
1523 * Insert onto list for new vnode.
1525 buf_vlist_add(bp, &vp->v_bufobj, BX_VNCLEAN);
1529 * Disassociate a buffer from a vnode.
1532 brelvp(struct buf *bp)
1537 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1538 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1541 * Delete from old vnode list, if on one.
1543 vp = bp->b_vp; /* XXX */
1546 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1547 buf_vlist_remove(bp);
1549 panic("brelvp: Buffer %p not on queue.", bp);
1550 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1551 bo->bo_flag &= ~BO_ONWORKLST;
1552 mtx_lock(&sync_mtx);
1553 LIST_REMOVE(bo, bo_synclist);
1554 syncer_worklist_len--;
1555 mtx_unlock(&sync_mtx);
1557 bp->b_flags &= ~B_NEEDSGIANT;
1559 bp->b_bufobj = NULL;
1564 * Add an item to the syncer work queue.
1567 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1571 ASSERT_BO_LOCKED(bo);
1573 mtx_lock(&sync_mtx);
1574 if (bo->bo_flag & BO_ONWORKLST)
1575 LIST_REMOVE(bo, bo_synclist);
1577 bo->bo_flag |= BO_ONWORKLST;
1578 syncer_worklist_len++;
1581 if (delay > syncer_maxdelay - 2)
1582 delay = syncer_maxdelay - 2;
1583 slot = (syncer_delayno + delay) & syncer_mask;
1585 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1586 mtx_unlock(&sync_mtx);
1590 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1594 mtx_lock(&sync_mtx);
1595 len = syncer_worklist_len - sync_vnode_count;
1596 mtx_unlock(&sync_mtx);
1597 error = SYSCTL_OUT(req, &len, sizeof(len));
1601 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1602 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1604 static struct proc *updateproc;
1605 static void sched_sync(void);
1606 static struct kproc_desc up_kp = {
1611 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
1614 sync_vnode(struct bufobj *bo, struct thread *td)
1619 vp = bo->__bo_vnode; /* XXX */
1620 if (VOP_ISLOCKED(vp, NULL) != 0)
1622 if (VI_TRYLOCK(vp) == 0)
1625 * We use vhold in case the vnode does not
1626 * successfully sync. vhold prevents the vnode from
1627 * going away when we unlock the sync_mtx so that
1628 * we can acquire the vnode interlock.
1631 mtx_unlock(&sync_mtx);
1633 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1635 mtx_lock(&sync_mtx);
1638 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
1639 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1640 VOP_UNLOCK(vp, 0, td);
1641 vn_finished_write(mp);
1643 if ((bo->bo_flag & BO_ONWORKLST) != 0) {
1645 * Put us back on the worklist. The worklist
1646 * routine will remove us from our current
1647 * position and then add us back in at a later
1650 vn_syncer_add_to_worklist(bo, syncdelay);
1653 mtx_lock(&sync_mtx);
1658 * System filesystem synchronizer daemon.
1663 struct synclist *next;
1664 struct synclist *slp;
1667 struct thread *td = FIRST_THREAD_IN_PROC(updateproc);
1668 static int dummychan;
1670 int net_worklist_len;
1671 int syncer_final_iter;
1677 syncer_final_iter = 0;
1679 syncer_state = SYNCER_RUNNING;
1680 starttime = time_uptime;
1681 td->td_pflags |= TDP_NORUNNINGBUF;
1683 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1686 mtx_lock(&sync_mtx);
1688 if (syncer_state == SYNCER_FINAL_DELAY &&
1689 syncer_final_iter == 0) {
1690 mtx_unlock(&sync_mtx);
1691 kthread_suspend_check(td->td_proc);
1692 mtx_lock(&sync_mtx);
1694 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1695 if (syncer_state != SYNCER_RUNNING &&
1696 starttime != time_uptime) {
1698 printf("\nSyncing disks, vnodes remaining...");
1701 printf("%d ", net_worklist_len);
1703 starttime = time_uptime;
1706 * Push files whose dirty time has expired. Be careful
1707 * of interrupt race on slp queue.
1709 * Skip over empty worklist slots when shutting down.
1712 slp = &syncer_workitem_pending[syncer_delayno];
1713 syncer_delayno += 1;
1714 if (syncer_delayno == syncer_maxdelay)
1716 next = &syncer_workitem_pending[syncer_delayno];
1718 * If the worklist has wrapped since the
1719 * it was emptied of all but syncer vnodes,
1720 * switch to the FINAL_DELAY state and run
1721 * for one more second.
1723 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1724 net_worklist_len == 0 &&
1725 last_work_seen == syncer_delayno) {
1726 syncer_state = SYNCER_FINAL_DELAY;
1727 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1729 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1730 syncer_worklist_len > 0);
1733 * Keep track of the last time there was anything
1734 * on the worklist other than syncer vnodes.
1735 * Return to the SHUTTING_DOWN state if any
1738 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1739 last_work_seen = syncer_delayno;
1740 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1741 syncer_state = SYNCER_SHUTTING_DOWN;
1742 while ((bo = LIST_FIRST(slp)) != NULL) {
1743 error = sync_vnode(bo, td);
1745 LIST_REMOVE(bo, bo_synclist);
1746 LIST_INSERT_HEAD(next, bo, bo_synclist);
1750 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1751 syncer_final_iter--;
1753 * The variable rushjob allows the kernel to speed up the
1754 * processing of the filesystem syncer process. A rushjob
1755 * value of N tells the filesystem syncer to process the next
1756 * N seconds worth of work on its queue ASAP. Currently rushjob
1757 * is used by the soft update code to speed up the filesystem
1758 * syncer process when the incore state is getting so far
1759 * ahead of the disk that the kernel memory pool is being
1760 * threatened with exhaustion.
1767 * Just sleep for a short period of time between
1768 * iterations when shutting down to allow some I/O
1771 * If it has taken us less than a second to process the
1772 * current work, then wait. Otherwise start right over
1773 * again. We can still lose time if any single round
1774 * takes more than two seconds, but it does not really
1775 * matter as we are just trying to generally pace the
1776 * filesystem activity.
1778 if (syncer_state != SYNCER_RUNNING)
1779 msleep(&dummychan, &sync_mtx, PPAUSE, "syncfnl",
1780 hz / SYNCER_SHUTDOWN_SPEEDUP);
1781 else if (time_uptime == starttime)
1782 msleep(&lbolt, &sync_mtx, PPAUSE, "syncer", 0);
1787 * Request the syncer daemon to speed up its work.
1788 * We never push it to speed up more than half of its
1789 * normal turn time, otherwise it could take over the cpu.
1792 speedup_syncer(void)
1797 td = FIRST_THREAD_IN_PROC(updateproc);
1798 sleepq_remove(td, &lbolt);
1799 mtx_lock(&sync_mtx);
1800 if (rushjob < syncdelay / 2) {
1802 stat_rush_requests += 1;
1805 mtx_unlock(&sync_mtx);
1810 * Tell the syncer to speed up its work and run though its work
1811 * list several times, then tell it to shut down.
1814 syncer_shutdown(void *arg, int howto)
1818 if (howto & RB_NOSYNC)
1820 td = FIRST_THREAD_IN_PROC(updateproc);
1821 sleepq_remove(td, &lbolt);
1822 mtx_lock(&sync_mtx);
1823 syncer_state = SYNCER_SHUTTING_DOWN;
1825 mtx_unlock(&sync_mtx);
1826 kproc_shutdown(arg, howto);
1830 * Reassign a buffer from one vnode to another.
1831 * Used to assign file specific control information
1832 * (indirect blocks) to the vnode to which they belong.
1835 reassignbuf(struct buf *bp)
1848 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1849 bp, bp->b_vp, bp->b_flags);
1851 * B_PAGING flagged buffers cannot be reassigned because their vp
1852 * is not fully linked in.
1854 if (bp->b_flags & B_PAGING)
1855 panic("cannot reassign paging buffer");
1858 * Delete from old vnode list, if on one.
1861 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1862 buf_vlist_remove(bp);
1864 panic("reassignbuf: Buffer %p not on queue.", bp);
1866 * If dirty, put on list of dirty buffers; otherwise insert onto list
1869 if (bp->b_flags & B_DELWRI) {
1870 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1871 switch (vp->v_type) {
1881 vn_syncer_add_to_worklist(bo, delay);
1883 buf_vlist_add(bp, bo, BX_VNDIRTY);
1885 buf_vlist_add(bp, bo, BX_VNCLEAN);
1887 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1888 mtx_lock(&sync_mtx);
1889 LIST_REMOVE(bo, bo_synclist);
1890 syncer_worklist_len--;
1891 mtx_unlock(&sync_mtx);
1892 bo->bo_flag &= ~BO_ONWORKLST;
1897 bp = TAILQ_FIRST(&bv->bv_hd);
1898 KASSERT(bp == NULL || bp->b_bufobj == bo,
1899 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1900 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1901 KASSERT(bp == NULL || bp->b_bufobj == bo,
1902 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1904 bp = TAILQ_FIRST(&bv->bv_hd);
1905 KASSERT(bp == NULL || bp->b_bufobj == bo,
1906 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1907 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1908 KASSERT(bp == NULL || bp->b_bufobj == bo,
1909 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1915 * Increment the use and hold counts on the vnode, taking care to reference
1916 * the driver's usecount if this is a chardev. The vholdl() will remove
1917 * the vnode from the free list if it is presently free. Requires the
1918 * vnode interlock and returns with it held.
1921 v_incr_usecount(struct vnode *vp)
1924 CTR3(KTR_VFS, "v_incr_usecount: vp %p holdcnt %d usecount %d\n",
1925 vp, vp->v_holdcnt, vp->v_usecount);
1927 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1929 vp->v_rdev->si_usecount++;
1936 * Turn a holdcnt into a use+holdcnt such that only one call to
1937 * v_decr_usecount is needed.
1940 v_upgrade_usecount(struct vnode *vp)
1943 CTR3(KTR_VFS, "v_upgrade_usecount: vp %p holdcnt %d usecount %d\n",
1944 vp, vp->v_holdcnt, vp->v_usecount);
1946 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1948 vp->v_rdev->si_usecount++;
1954 * Decrement the vnode use and hold count along with the driver's usecount
1955 * if this is a chardev. The vdropl() below releases the vnode interlock
1956 * as it may free the vnode.
1959 v_decr_usecount(struct vnode *vp)
1962 CTR3(KTR_VFS, "v_decr_usecount: vp %p holdcnt %d usecount %d\n",
1963 vp, vp->v_holdcnt, vp->v_usecount);
1964 ASSERT_VI_LOCKED(vp, __FUNCTION__);
1965 VNASSERT(vp->v_usecount > 0, vp,
1966 ("v_decr_usecount: negative usecount"));
1968 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1970 vp->v_rdev->si_usecount--;
1977 * Decrement only the use count and driver use count. This is intended to
1978 * be paired with a follow on vdropl() to release the remaining hold count.
1979 * In this way we may vgone() a vnode with a 0 usecount without risk of
1980 * having it end up on a free list because the hold count is kept above 0.
1983 v_decr_useonly(struct vnode *vp)
1986 CTR3(KTR_VFS, "v_decr_useonly: vp %p holdcnt %d usecount %d\n",
1987 vp, vp->v_holdcnt, vp->v_usecount);
1988 ASSERT_VI_LOCKED(vp, __FUNCTION__);
1989 VNASSERT(vp->v_usecount > 0, vp,
1990 ("v_decr_useonly: negative usecount"));
1992 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
1994 vp->v_rdev->si_usecount--;
2000 * Grab a particular vnode from the free list, increment its
2001 * reference count and lock it. The vnode lock bit is set if the
2002 * vnode is being eliminated in vgone. The process is awakened
2003 * when the transition is completed, and an error returned to
2004 * indicate that the vnode is no longer usable (possibly having
2005 * been changed to a new filesystem type).
2008 vget(struct vnode *vp, int flags, struct thread *td)
2017 VFS_ASSERT_GIANT(vp->v_mount);
2018 if ((flags & LK_INTERLOCK) == 0)
2021 * If the inactive call was deferred because vput() was called
2022 * with a shared lock, we have to do it here before another thread
2023 * gets a reference to data that should be dead.
2025 if (vp->v_iflag & VI_OWEINACT) {
2026 if (flags & LK_NOWAIT) {
2030 flags &= ~LK_TYPE_MASK;
2031 flags |= LK_EXCLUSIVE;
2035 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
2040 /* Upgrade our holdcnt to a usecount. */
2041 v_upgrade_usecount(vp);
2042 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2043 panic("vget: vn_lock failed to return ENOENT\n");
2045 if (vp->v_iflag & VI_OWEINACT)
2048 if ((oldflags & LK_TYPE_MASK) == 0)
2049 VOP_UNLOCK(vp, 0, td);
2056 * Increase the reference count of a vnode.
2059 vref(struct vnode *vp)
2063 v_incr_usecount(vp);
2068 * Return reference count of a vnode.
2070 * The results of this call are only guaranteed when some mechanism other
2071 * than the VI lock is used to stop other processes from gaining references
2072 * to the vnode. This may be the case if the caller holds the only reference.
2073 * This is also useful when stale data is acceptable as race conditions may
2074 * be accounted for by some other means.
2077 vrefcnt(struct vnode *vp)
2082 usecnt = vp->v_usecount;
2090 * Vnode put/release.
2091 * If count drops to zero, call inactive routine and return to freelist.
2094 vrele(struct vnode *vp)
2096 struct thread *td = curthread; /* XXX */
2098 KASSERT(vp != NULL, ("vrele: null vp"));
2099 VFS_ASSERT_GIANT(vp->v_mount);
2103 /* Skip this v_writecount check if we're going to panic below. */
2104 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2105 ("vrele: missed vn_close"));
2107 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2108 vp->v_usecount == 1)) {
2109 v_decr_usecount(vp);
2112 if (vp->v_usecount != 1) {
2114 vprint("vrele: negative ref count", vp);
2117 panic("vrele: negative ref cnt");
2120 * We want to hold the vnode until the inactive finishes to
2121 * prevent vgone() races. We drop the use count here and the
2122 * hold count below when we're done.
2126 * We must call VOP_INACTIVE with the node locked. Mark
2127 * as VI_DOINGINACT to avoid recursion.
2129 vp->v_iflag |= VI_OWEINACT;
2130 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
2132 if (vp->v_usecount > 0)
2133 vp->v_iflag &= ~VI_OWEINACT;
2134 if (vp->v_iflag & VI_OWEINACT)
2136 VOP_UNLOCK(vp, 0, td);
2139 if (vp->v_usecount > 0)
2140 vp->v_iflag &= ~VI_OWEINACT;
2146 * Release an already locked vnode. This give the same effects as
2147 * unlock+vrele(), but takes less time and avoids releasing and
2148 * re-aquiring the lock (as vrele() acquires the lock internally.)
2151 vput(struct vnode *vp)
2153 struct thread *td = curthread; /* XXX */
2156 KASSERT(vp != NULL, ("vput: null vp"));
2157 ASSERT_VOP_LOCKED(vp, "vput");
2158 VFS_ASSERT_GIANT(vp->v_mount);
2160 /* Skip this v_writecount check if we're going to panic below. */
2161 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2162 ("vput: missed vn_close"));
2165 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2166 vp->v_usecount == 1)) {
2167 VOP_UNLOCK(vp, 0, td);
2168 v_decr_usecount(vp);
2172 if (vp->v_usecount != 1) {
2174 vprint("vput: negative ref count", vp);
2176 panic("vput: negative ref cnt");
2179 * We want to hold the vnode until the inactive finishes to
2180 * prevent vgone() races. We drop the use count here and the
2181 * hold count below when we're done.
2184 vp->v_iflag |= VI_OWEINACT;
2185 if (VOP_ISLOCKED(vp, NULL) != LK_EXCLUSIVE) {
2186 error = VOP_LOCK(vp, LK_EXCLUPGRADE|LK_INTERLOCK|LK_NOWAIT, td);
2189 if (vp->v_usecount > 0)
2190 vp->v_iflag &= ~VI_OWEINACT;
2194 if (vp->v_usecount > 0)
2195 vp->v_iflag &= ~VI_OWEINACT;
2196 if (vp->v_iflag & VI_OWEINACT)
2198 VOP_UNLOCK(vp, 0, td);
2204 * Somebody doesn't want the vnode recycled.
2207 vhold(struct vnode *vp)
2216 vholdl(struct vnode *vp)
2220 if (VSHOULDBUSY(vp))
2225 * Note that there is one less who cares about this vnode. vdrop() is the
2226 * opposite of vhold().
2229 vdrop(struct vnode *vp)
2237 * Drop the hold count of the vnode. If this is the last reference to
2238 * the vnode we will free it if it has been vgone'd otherwise it is
2239 * placed on the free list.
2242 vdropl(struct vnode *vp)
2245 ASSERT_VI_LOCKED(vp, "vdropl");
2246 if (vp->v_holdcnt <= 0)
2247 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2249 if (vp->v_holdcnt == 0) {
2250 if (vp->v_iflag & VI_DOOMED) {
2260 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2261 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2262 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2263 * failed lock upgrade.
2266 vinactive(struct vnode *vp, struct thread *td)
2269 ASSERT_VOP_LOCKED(vp, "vinactive");
2270 ASSERT_VI_LOCKED(vp, "vinactive");
2271 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2272 ("vinactive: recursed on VI_DOINGINACT"));
2273 vp->v_iflag |= VI_DOINGINACT;
2274 vp->v_iflag &= ~VI_OWEINACT;
2276 VOP_INACTIVE(vp, td);
2278 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2279 ("vinactive: lost VI_DOINGINACT"));
2280 vp->v_iflag &= ~VI_DOINGINACT;
2284 * Remove any vnodes in the vnode table belonging to mount point mp.
2286 * If FORCECLOSE is not specified, there should not be any active ones,
2287 * return error if any are found (nb: this is a user error, not a
2288 * system error). If FORCECLOSE is specified, detach any active vnodes
2291 * If WRITECLOSE is set, only flush out regular file vnodes open for
2294 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2296 * `rootrefs' specifies the base reference count for the root vnode
2297 * of this filesystem. The root vnode is considered busy if its
2298 * v_usecount exceeds this value. On a successful return, vflush(, td)
2299 * will call vrele() on the root vnode exactly rootrefs times.
2300 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2304 static int busyprt = 0; /* print out busy vnodes */
2305 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2309 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2311 struct vnode *vp, *mvp, *rootvp = NULL;
2313 int busy = 0, error;
2315 CTR1(KTR_VFS, "vflush: mp %p", mp);
2317 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2318 ("vflush: bad args"));
2320 * Get the filesystem root vnode. We can vput() it
2321 * immediately, since with rootrefs > 0, it won't go away.
2323 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp, td)) != 0)
2330 MNT_VNODE_FOREACH(vp, mp, mvp) {
2335 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td);
2339 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2343 * Skip over a vnodes marked VV_SYSTEM.
2345 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2346 VOP_UNLOCK(vp, 0, td);
2352 * If WRITECLOSE is set, flush out unlinked but still open
2353 * files (even if open only for reading) and regular file
2354 * vnodes open for writing.
2356 if (flags & WRITECLOSE) {
2357 error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
2360 if ((vp->v_type == VNON ||
2361 (error == 0 && vattr.va_nlink > 0)) &&
2362 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2363 VOP_UNLOCK(vp, 0, td);
2371 * With v_usecount == 0, all we need to do is clear out the
2372 * vnode data structures and we are done.
2374 * If FORCECLOSE is set, forcibly close the vnode.
2376 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2377 VNASSERT(vp->v_usecount == 0 ||
2378 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2379 ("device VNODE %p is FORCECLOSED", vp));
2385 vprint("vflush: busy vnode", vp);
2388 VOP_UNLOCK(vp, 0, td);
2393 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2395 * If just the root vnode is busy, and if its refcount
2396 * is equal to `rootrefs', then go ahead and kill it.
2399 KASSERT(busy > 0, ("vflush: not busy"));
2400 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2401 ("vflush: usecount %d < rootrefs %d",
2402 rootvp->v_usecount, rootrefs));
2403 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2404 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK, td);
2406 VOP_UNLOCK(rootvp, 0, td);
2413 for (; rootrefs > 0; rootrefs--)
2419 * Recycle an unused vnode to the front of the free list.
2422 vrecycle(struct vnode *vp, struct thread *td)
2426 ASSERT_VOP_LOCKED(vp, "vrecycle");
2429 if (vp->v_usecount == 0) {
2438 * Eliminate all activity associated with a vnode
2439 * in preparation for reuse.
2442 vgone(struct vnode *vp)
2450 * vgone, with the vp interlock held.
2453 vgonel(struct vnode *vp)
2460 CTR1(KTR_VFS, "vgonel: vp %p", vp);
2461 ASSERT_VOP_LOCKED(vp, "vgonel");
2462 ASSERT_VI_LOCKED(vp, "vgonel");
2463 VNASSERT(vp->v_holdcnt, vp,
2464 ("vgonel: vp %p has no reference.", vp));
2468 * Don't vgonel if we're already doomed.
2470 if (vp->v_iflag & VI_DOOMED)
2472 vp->v_iflag |= VI_DOOMED;
2474 * Check to see if the vnode is in use. If so, we have to call
2475 * VOP_CLOSE() and VOP_INACTIVE().
2477 active = vp->v_usecount;
2478 oweinact = (vp->v_iflag & VI_OWEINACT);
2481 * Clean out any buffers associated with the vnode.
2482 * If the flush fails, just toss the buffers.
2485 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2486 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2487 if (vinvalbuf(vp, V_SAVE, td, 0, 0) != 0)
2488 vinvalbuf(vp, 0, td, 0, 0);
2491 * If purging an active vnode, it must be closed and
2492 * deactivated before being reclaimed.
2495 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2496 if (oweinact || active) {
2498 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2503 * Reclaim the vnode.
2505 if (VOP_RECLAIM(vp, td))
2506 panic("vgone: cannot reclaim");
2508 vn_finished_secondary_write(mp);
2509 VNASSERT(vp->v_object == NULL, vp,
2510 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2512 * Delete from old mount point vnode list.
2517 * Done with purge, reset to the standard lock and invalidate
2521 vp->v_vnlock = &vp->v_lock;
2522 vp->v_op = &dead_vnodeops;
2528 * Calculate the total number of references to a special device.
2531 vcount(struct vnode *vp)
2536 count = vp->v_rdev->si_usecount;
2542 * Same as above, but using the struct cdev *as argument
2545 count_dev(struct cdev *dev)
2550 count = dev->si_usecount;
2556 * Print out a description of a vnode.
2558 static char *typename[] =
2559 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2563 vn_printf(struct vnode *vp, const char *fmt, ...)
2566 char buf[256], buf2[16];
2572 printf("%p: ", (void *)vp);
2573 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2574 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2575 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2578 if (vp->v_vflag & VV_ROOT)
2579 strlcat(buf, "|VV_ROOT", sizeof(buf));
2580 if (vp->v_vflag & VV_ISTTY)
2581 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2582 if (vp->v_vflag & VV_NOSYNC)
2583 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2584 if (vp->v_vflag & VV_CACHEDLABEL)
2585 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2586 if (vp->v_vflag & VV_TEXT)
2587 strlcat(buf, "|VV_TEXT", sizeof(buf));
2588 if (vp->v_vflag & VV_COPYONWRITE)
2589 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2590 if (vp->v_vflag & VV_SYSTEM)
2591 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2592 if (vp->v_vflag & VV_PROCDEP)
2593 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2594 if (vp->v_vflag & VV_NOKNOTE)
2595 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2596 if (vp->v_vflag & VV_DELETED)
2597 strlcat(buf, "|VV_DELETED", sizeof(buf));
2598 if (vp->v_vflag & VV_MD)
2599 strlcat(buf, "|VV_MD", sizeof(buf));
2600 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2601 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2602 VV_NOKNOTE | VV_DELETED | VV_MD);
2604 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2605 strlcat(buf, buf2, sizeof(buf));
2607 if (vp->v_iflag & VI_MOUNT)
2608 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2609 if (vp->v_iflag & VI_AGE)
2610 strlcat(buf, "|VI_AGE", sizeof(buf));
2611 if (vp->v_iflag & VI_DOOMED)
2612 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2613 if (vp->v_iflag & VI_FREE)
2614 strlcat(buf, "|VI_FREE", sizeof(buf));
2615 if (vp->v_iflag & VI_OBJDIRTY)
2616 strlcat(buf, "|VI_OBJDIRTY", sizeof(buf));
2617 if (vp->v_iflag & VI_DOINGINACT)
2618 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2619 if (vp->v_iflag & VI_OWEINACT)
2620 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2621 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2622 VI_OBJDIRTY | VI_DOINGINACT | VI_OWEINACT);
2624 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2625 strlcat(buf, buf2, sizeof(buf));
2627 printf(" flags (%s)\n", buf + 1);
2628 if (mtx_owned(VI_MTX(vp)))
2629 printf(" VI_LOCKed");
2630 if (vp->v_object != NULL)
2631 printf(" v_object %p ref %d pages %d\n",
2632 vp->v_object, vp->v_object->ref_count,
2633 vp->v_object->resident_page_count);
2635 lockmgr_printinfo(vp->v_vnlock);
2637 if (vp->v_data != NULL)
2643 * List all of the locked vnodes in the system.
2644 * Called when debugging the kernel.
2646 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2648 struct mount *mp, *nmp;
2652 * Note: because this is DDB, we can't obey the locking semantics
2653 * for these structures, which means we could catch an inconsistent
2654 * state and dereference a nasty pointer. Not much to be done
2657 printf("Locked vnodes\n");
2658 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2659 nmp = TAILQ_NEXT(mp, mnt_list);
2660 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2661 if (vp->v_type != VMARKER && VOP_ISLOCKED(vp, NULL))
2664 nmp = TAILQ_NEXT(mp, mnt_list);
2669 * Show details about the given vnode.
2671 DB_SHOW_COMMAND(vnode, db_show_vnode)
2677 vp = (struct vnode *)addr;
2678 vn_printf(vp, "vnode ");
2683 * Fill in a struct xvfsconf based on a struct vfsconf.
2686 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2689 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2690 xvfsp->vfc_typenum = vfsp->vfc_typenum;
2691 xvfsp->vfc_refcount = vfsp->vfc_refcount;
2692 xvfsp->vfc_flags = vfsp->vfc_flags;
2694 * These are unused in userland, we keep them
2695 * to not break binary compatibility.
2697 xvfsp->vfc_vfsops = NULL;
2698 xvfsp->vfc_next = NULL;
2702 * Top level filesystem related information gathering.
2705 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2707 struct vfsconf *vfsp;
2708 struct xvfsconf xvfsp;
2712 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2713 bzero(&xvfsp, sizeof(xvfsp));
2714 vfsconf2x(vfsp, &xvfsp);
2715 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2722 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2723 "S,xvfsconf", "List of all configured filesystems");
2725 #ifndef BURN_BRIDGES
2726 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2729 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2731 int *name = (int *)arg1 - 1; /* XXX */
2732 u_int namelen = arg2 + 1; /* XXX */
2733 struct vfsconf *vfsp;
2734 struct xvfsconf xvfsp;
2736 printf("WARNING: userland calling deprecated sysctl, "
2737 "please rebuild world\n");
2739 #if 1 || defined(COMPAT_PRELITE2)
2740 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2742 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
2746 case VFS_MAXTYPENUM:
2749 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
2752 return (ENOTDIR); /* overloaded */
2753 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
2754 if (vfsp->vfc_typenum == name[2])
2757 return (EOPNOTSUPP);
2758 bzero(&xvfsp, sizeof(xvfsp));
2759 vfsconf2x(vfsp, &xvfsp);
2760 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
2762 return (EOPNOTSUPP);
2765 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
2766 vfs_sysctl, "Generic filesystem");
2768 #if 1 || defined(COMPAT_PRELITE2)
2771 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
2774 struct vfsconf *vfsp;
2775 struct ovfsconf ovfs;
2777 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2778 bzero(&ovfs, sizeof(ovfs));
2779 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
2780 strcpy(ovfs.vfc_name, vfsp->vfc_name);
2781 ovfs.vfc_index = vfsp->vfc_typenum;
2782 ovfs.vfc_refcount = vfsp->vfc_refcount;
2783 ovfs.vfc_flags = vfsp->vfc_flags;
2784 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
2791 #endif /* 1 || COMPAT_PRELITE2 */
2792 #endif /* !BURN_BRIDGES */
2794 #define KINFO_VNODESLOP 10
2797 * Dump vnode list (via sysctl).
2801 sysctl_vnode(SYSCTL_HANDLER_ARGS)
2804 struct thread *td = req->td;
2810 * Stale numvnodes access is not fatal here.
2813 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
2815 /* Make an estimate */
2816 return (SYSCTL_OUT(req, 0, len));
2818 error = sysctl_wire_old_buffer(req, 0);
2821 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
2823 mtx_lock(&mountlist_mtx);
2824 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2825 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
2828 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2832 xvn[n].xv_size = sizeof *xvn;
2833 xvn[n].xv_vnode = vp;
2834 xvn[n].xv_id = 0; /* XXX compat */
2835 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
2837 XV_COPY(writecount);
2843 xvn[n].xv_flag = vp->v_vflag;
2845 switch (vp->v_type) {
2852 if (vp->v_rdev == NULL) {
2856 xvn[n].xv_dev = dev2udev(vp->v_rdev);
2859 xvn[n].xv_socket = vp->v_socket;
2862 xvn[n].xv_fifo = vp->v_fifoinfo;
2867 /* shouldn't happen? */
2875 mtx_lock(&mountlist_mtx);
2880 mtx_unlock(&mountlist_mtx);
2882 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
2887 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
2888 0, 0, sysctl_vnode, "S,xvnode", "");
2892 * Unmount all filesystems. The list is traversed in reverse order
2893 * of mounting to avoid dependencies.
2896 vfs_unmountall(void)
2902 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
2905 * Since this only runs when rebooting, it is not interlocked.
2907 while(!TAILQ_EMPTY(&mountlist)) {
2908 mp = TAILQ_LAST(&mountlist, mntlist);
2909 error = dounmount(mp, MNT_FORCE, td);
2911 TAILQ_REMOVE(&mountlist, mp, mnt_list);
2913 * XXX: Due to the way in which we mount the root
2914 * file system off of devfs, devfs will generate a
2915 * "busy" warning when we try to unmount it before
2916 * the root. Don't print a warning as a result in
2917 * order to avoid false positive errors that may
2918 * cause needless upset.
2920 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
2921 printf("unmount of %s failed (",
2922 mp->mnt_stat.f_mntonname);
2926 printf("%d)\n", error);
2929 /* The unmount has removed mp from the mountlist */
2935 * perform msync on all vnodes under a mount point
2936 * the mount point must be locked.
2939 vfs_msync(struct mount *mp, int flags)
2941 struct vnode *vp, *mvp;
2942 struct vm_object *obj;
2945 MNT_VNODE_FOREACH(vp, mp, mvp) {
2947 if ((vp->v_iflag & VI_OBJDIRTY) &&
2948 (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
2951 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
2953 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
2961 VM_OBJECT_LOCK(obj);
2962 vm_object_page_clean(obj, 0, 0,
2964 OBJPC_SYNC : OBJPC_NOSYNC);
2965 VM_OBJECT_UNLOCK(obj);
2977 * Mark a vnode as free, putting it up for recycling.
2980 vfree(struct vnode *vp)
2983 CTR1(KTR_VFS, "vfree vp %p", vp);
2984 ASSERT_VI_LOCKED(vp, "vfree");
2985 mtx_lock(&vnode_free_list_mtx);
2986 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
2987 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
2988 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
2989 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
2990 ("vfree: Freeing doomed vnode"));
2991 if (vp->v_iflag & VI_AGE) {
2992 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
2994 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
2997 vp->v_iflag &= ~VI_AGE;
2998 vp->v_iflag |= VI_FREE;
2999 mtx_unlock(&vnode_free_list_mtx);
3003 * Opposite of vfree() - mark a vnode as in use.
3006 vbusy(struct vnode *vp)
3008 CTR1(KTR_VFS, "vbusy vp %p", vp);
3009 ASSERT_VI_LOCKED(vp, "vbusy");
3010 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3011 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3013 mtx_lock(&vnode_free_list_mtx);
3014 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3016 vp->v_iflag &= ~(VI_FREE|VI_AGE);
3017 mtx_unlock(&vnode_free_list_mtx);
3021 * Initalize per-vnode helper structure to hold poll-related state.
3024 v_addpollinfo(struct vnode *vp)
3026 struct vpollinfo *vi;
3028 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3029 if (vp->v_pollinfo != NULL) {
3030 uma_zfree(vnodepoll_zone, vi);
3033 vp->v_pollinfo = vi;
3034 mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3035 knlist_init(&vp->v_pollinfo->vpi_selinfo.si_note, vp, vfs_knllock,
3036 vfs_knlunlock, vfs_knllocked);
3040 * Record a process's interest in events which might happen to
3041 * a vnode. Because poll uses the historic select-style interface
3042 * internally, this routine serves as both the ``check for any
3043 * pending events'' and the ``record my interest in future events''
3044 * functions. (These are done together, while the lock is held,
3045 * to avoid race conditions.)
3048 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3051 if (vp->v_pollinfo == NULL)
3053 mtx_lock(&vp->v_pollinfo->vpi_lock);
3054 if (vp->v_pollinfo->vpi_revents & events) {
3056 * This leaves events we are not interested
3057 * in available for the other process which
3058 * which presumably had requested them
3059 * (otherwise they would never have been
3062 events &= vp->v_pollinfo->vpi_revents;
3063 vp->v_pollinfo->vpi_revents &= ~events;
3065 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3068 vp->v_pollinfo->vpi_events |= events;
3069 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3070 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3075 * Routine to create and manage a filesystem syncer vnode.
3077 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3078 static int sync_fsync(struct vop_fsync_args *);
3079 static int sync_inactive(struct vop_inactive_args *);
3080 static int sync_reclaim(struct vop_reclaim_args *);
3082 static struct vop_vector sync_vnodeops = {
3083 .vop_bypass = VOP_EOPNOTSUPP,
3084 .vop_close = sync_close, /* close */
3085 .vop_fsync = sync_fsync, /* fsync */
3086 .vop_inactive = sync_inactive, /* inactive */
3087 .vop_reclaim = sync_reclaim, /* reclaim */
3088 .vop_lock1 = vop_stdlock, /* lock */
3089 .vop_unlock = vop_stdunlock, /* unlock */
3090 .vop_islocked = vop_stdislocked, /* islocked */
3094 * Create a new filesystem syncer vnode for the specified mount point.
3097 vfs_allocate_syncvnode(struct mount *mp)
3100 static long start, incr, next;
3103 /* Allocate a new vnode */
3104 if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3105 mp->mnt_syncer = NULL;
3109 error = insmntque(vp, mp);
3111 panic("vfs_allocate_syncvnode: insmntque failed");
3113 * Place the vnode onto the syncer worklist. We attempt to
3114 * scatter them about on the list so that they will go off
3115 * at evenly distributed times even if all the filesystems
3116 * are mounted at once.
3119 if (next == 0 || next > syncer_maxdelay) {
3123 start = syncer_maxdelay / 2;
3124 incr = syncer_maxdelay;
3129 vn_syncer_add_to_worklist(&vp->v_bufobj,
3130 syncdelay > 0 ? next % syncdelay : 0);
3131 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3132 mtx_lock(&sync_mtx);
3134 mtx_unlock(&sync_mtx);
3136 mp->mnt_syncer = vp;
3141 * Do a lazy sync of the filesystem.
3144 sync_fsync(struct vop_fsync_args *ap)
3146 struct vnode *syncvp = ap->a_vp;
3147 struct mount *mp = syncvp->v_mount;
3148 struct thread *td = ap->a_td;
3153 * We only need to do something if this is a lazy evaluation.
3155 if (ap->a_waitfor != MNT_LAZY)
3159 * Move ourselves to the back of the sync list.
3161 bo = &syncvp->v_bufobj;
3163 vn_syncer_add_to_worklist(bo, syncdelay);
3167 * Walk the list of vnodes pushing all that are dirty and
3168 * not already on the sync list.
3170 mtx_lock(&mountlist_mtx);
3171 if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
3172 mtx_unlock(&mountlist_mtx);
3175 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3181 mp->mnt_kern_flag &= ~MNTK_ASYNC;
3183 vfs_msync(mp, MNT_NOWAIT);
3184 error = VFS_SYNC(mp, MNT_LAZY, td);
3187 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3188 mp->mnt_kern_flag |= MNTK_ASYNC;
3190 vn_finished_write(mp);
3196 * The syncer vnode is no referenced.
3199 sync_inactive(struct vop_inactive_args *ap)
3207 * The syncer vnode is no longer needed and is being decommissioned.
3209 * Modifications to the worklist must be protected by sync_mtx.
3212 sync_reclaim(struct vop_reclaim_args *ap)
3214 struct vnode *vp = ap->a_vp;
3219 vp->v_mount->mnt_syncer = NULL;
3220 if (bo->bo_flag & BO_ONWORKLST) {
3221 mtx_lock(&sync_mtx);
3222 LIST_REMOVE(bo, bo_synclist);
3223 syncer_worklist_len--;
3225 mtx_unlock(&sync_mtx);
3226 bo->bo_flag &= ~BO_ONWORKLST;
3234 * Check if vnode represents a disk device
3237 vn_isdisk(struct vnode *vp, int *errp)
3243 if (vp->v_type != VCHR)
3245 else if (vp->v_rdev == NULL)
3247 else if (vp->v_rdev->si_devsw == NULL)
3249 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3254 return (error == 0);
3258 * Common filesystem object access control check routine. Accepts a
3259 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3260 * and optional call-by-reference privused argument allowing vaccess()
3261 * to indicate to the caller whether privilege was used to satisfy the
3262 * request (obsoleted). Returns 0 on success, or an errno on failure.
3264 * The ifdef'd CAPABILITIES version is here for reference, but is not
3268 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3269 mode_t acc_mode, struct ucred *cred, int *privused)
3272 mode_t priv_granted;
3275 * Look for a normal, non-privileged way to access the file/directory
3276 * as requested. If it exists, go with that.
3279 if (privused != NULL)
3284 /* Check the owner. */
3285 if (cred->cr_uid == file_uid) {
3286 dac_granted |= VADMIN;
3287 if (file_mode & S_IXUSR)
3288 dac_granted |= VEXEC;
3289 if (file_mode & S_IRUSR)
3290 dac_granted |= VREAD;
3291 if (file_mode & S_IWUSR)
3292 dac_granted |= (VWRITE | VAPPEND);
3294 if ((acc_mode & dac_granted) == acc_mode)
3300 /* Otherwise, check the groups (first match) */
3301 if (groupmember(file_gid, cred)) {
3302 if (file_mode & S_IXGRP)
3303 dac_granted |= VEXEC;
3304 if (file_mode & S_IRGRP)
3305 dac_granted |= VREAD;
3306 if (file_mode & S_IWGRP)
3307 dac_granted |= (VWRITE | VAPPEND);
3309 if ((acc_mode & dac_granted) == acc_mode)
3315 /* Otherwise, check everyone else. */
3316 if (file_mode & S_IXOTH)
3317 dac_granted |= VEXEC;
3318 if (file_mode & S_IROTH)
3319 dac_granted |= VREAD;
3320 if (file_mode & S_IWOTH)
3321 dac_granted |= (VWRITE | VAPPEND);
3322 if ((acc_mode & dac_granted) == acc_mode)
3327 * Build a privilege mask to determine if the set of privileges
3328 * satisfies the requirements when combined with the granted mask
3329 * from above. For each privilege, if the privilege is required,
3330 * bitwise or the request type onto the priv_granted mask.
3336 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3337 * requests, instead of PRIV_VFS_EXEC.
3339 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3340 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3341 priv_granted |= VEXEC;
3343 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3344 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3345 priv_granted |= VEXEC;
3348 if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
3349 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3350 priv_granted |= VREAD;
3352 if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3353 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3354 priv_granted |= (VWRITE | VAPPEND);
3356 if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3357 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3358 priv_granted |= VADMIN;
3360 if ((acc_mode & (priv_granted | dac_granted)) == acc_mode) {
3361 /* XXX audit: privilege used */
3362 if (privused != NULL)
3367 return ((acc_mode & VADMIN) ? EPERM : EACCES);
3371 * Credential check based on process requesting service, and per-attribute
3375 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3376 struct thread *td, int access)
3380 * Kernel-invoked always succeeds.
3386 * Do not allow privileged processes in jail to directly manipulate
3387 * system attributes.
3389 switch (attrnamespace) {
3390 case EXTATTR_NAMESPACE_SYSTEM:
3391 /* Potentially should be: return (EPERM); */
3392 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3393 case EXTATTR_NAMESPACE_USER:
3394 return (VOP_ACCESS(vp, access, cred, td));
3400 #ifdef DEBUG_VFS_LOCKS
3402 * This only exists to supress warnings from unlocked specfs accesses. It is
3403 * no longer ok to have an unlocked VFS.
3405 #define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
3407 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3408 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3410 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3411 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3413 int vfs_badlock_print = 1; /* Print lock violations. */
3414 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3417 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3418 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3422 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3426 if (vfs_badlock_backtrace)
3429 if (vfs_badlock_print)
3430 printf("%s: %p %s\n", str, (void *)vp, msg);
3431 if (vfs_badlock_ddb)
3432 kdb_enter("lock violation");
3436 assert_vi_locked(struct vnode *vp, const char *str)
3439 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3440 vfs_badlock("interlock is not locked but should be", str, vp);
3444 assert_vi_unlocked(struct vnode *vp, const char *str)
3447 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3448 vfs_badlock("interlock is locked but should not be", str, vp);
3452 assert_vop_locked(struct vnode *vp, const char *str)
3455 if (vp && !IGNORE_LOCK(vp) && VOP_ISLOCKED(vp, NULL) == 0)
3456 vfs_badlock("is not locked but should be", str, vp);
3460 assert_vop_unlocked(struct vnode *vp, const char *str)
3463 if (vp && !IGNORE_LOCK(vp) &&
3464 VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
3465 vfs_badlock("is locked but should not be", str, vp);
3469 assert_vop_elocked(struct vnode *vp, const char *str)
3472 if (vp && !IGNORE_LOCK(vp) &&
3473 VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
3474 vfs_badlock("is not exclusive locked but should be", str, vp);
3479 assert_vop_elocked_other(struct vnode *vp, const char *str)
3482 if (vp && !IGNORE_LOCK(vp) &&
3483 VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
3484 vfs_badlock("is not exclusive locked by another thread",
3489 assert_vop_slocked(struct vnode *vp, const char *str)
3492 if (vp && !IGNORE_LOCK(vp) &&
3493 VOP_ISLOCKED(vp, curthread) != LK_SHARED)
3494 vfs_badlock("is not locked shared but should be", str, vp);
3497 #endif /* DEBUG_VFS_LOCKS */
3500 vop_rename_pre(void *ap)
3502 struct vop_rename_args *a = ap;
3504 #ifdef DEBUG_VFS_LOCKS
3506 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3507 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3508 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3509 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3511 /* Check the source (from). */
3512 if (a->a_tdvp != a->a_fdvp && a->a_tvp != a->a_fdvp)
3513 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3514 if (a->a_tvp != a->a_fvp)
3515 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3517 /* Check the target. */
3519 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3520 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3522 if (a->a_tdvp != a->a_fdvp)
3524 if (a->a_tvp != a->a_fvp)
3532 vop_strategy_pre(void *ap)
3534 #ifdef DEBUG_VFS_LOCKS
3535 struct vop_strategy_args *a;
3542 * Cluster ops lock their component buffers but not the IO container.
3544 if ((bp->b_flags & B_CLUSTER) != 0)
3547 if (BUF_REFCNT(bp) < 1) {
3548 if (vfs_badlock_print)
3550 "VOP_STRATEGY: bp is not locked but should be\n");
3551 if (vfs_badlock_ddb)
3552 kdb_enter("lock violation");
3558 vop_lookup_pre(void *ap)
3560 #ifdef DEBUG_VFS_LOCKS
3561 struct vop_lookup_args *a;
3566 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3567 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3572 vop_lookup_post(void *ap, int rc)
3574 #ifdef DEBUG_VFS_LOCKS
3575 struct vop_lookup_args *a;
3583 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3584 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3587 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3592 vop_lock_pre(void *ap)
3594 #ifdef DEBUG_VFS_LOCKS
3595 struct vop_lock1_args *a = ap;
3597 if ((a->a_flags & LK_INTERLOCK) == 0)
3598 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3600 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3605 vop_lock_post(void *ap, int rc)
3607 #ifdef DEBUG_VFS_LOCKS
3608 struct vop_lock1_args *a = ap;
3610 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3612 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3617 vop_unlock_pre(void *ap)
3619 #ifdef DEBUG_VFS_LOCKS
3620 struct vop_unlock_args *a = ap;
3622 if (a->a_flags & LK_INTERLOCK)
3623 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3624 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3629 vop_unlock_post(void *ap, int rc)
3631 #ifdef DEBUG_VFS_LOCKS
3632 struct vop_unlock_args *a = ap;
3634 if (a->a_flags & LK_INTERLOCK)
3635 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3640 vop_create_post(void *ap, int rc)
3642 struct vop_create_args *a = ap;
3645 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3649 vop_link_post(void *ap, int rc)
3651 struct vop_link_args *a = ap;
3654 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3655 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3660 vop_mkdir_post(void *ap, int rc)
3662 struct vop_mkdir_args *a = ap;
3665 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3669 vop_mknod_post(void *ap, int rc)
3671 struct vop_mknod_args *a = ap;
3674 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3678 vop_remove_post(void *ap, int rc)
3680 struct vop_remove_args *a = ap;
3683 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3684 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3689 vop_rename_post(void *ap, int rc)
3691 struct vop_rename_args *a = ap;
3694 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3695 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3696 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3698 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3700 if (a->a_tdvp != a->a_fdvp)
3702 if (a->a_tvp != a->a_fvp)
3710 vop_rmdir_post(void *ap, int rc)
3712 struct vop_rmdir_args *a = ap;
3715 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3716 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3721 vop_setattr_post(void *ap, int rc)
3723 struct vop_setattr_args *a = ap;
3726 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
3730 vop_symlink_post(void *ap, int rc)
3732 struct vop_symlink_args *a = ap;
3735 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3738 static struct knlist fs_knlist;
3741 vfs_event_init(void *arg)
3743 knlist_init(&fs_knlist, NULL, NULL, NULL, NULL);
3745 /* XXX - correct order? */
3746 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
3749 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
3752 KNOTE_UNLOCKED(&fs_knlist, event);
3755 static int filt_fsattach(struct knote *kn);
3756 static void filt_fsdetach(struct knote *kn);
3757 static int filt_fsevent(struct knote *kn, long hint);
3759 struct filterops fs_filtops =
3760 { 0, filt_fsattach, filt_fsdetach, filt_fsevent };
3763 filt_fsattach(struct knote *kn)
3766 kn->kn_flags |= EV_CLEAR;
3767 knlist_add(&fs_knlist, kn, 0);
3772 filt_fsdetach(struct knote *kn)
3775 knlist_remove(&fs_knlist, kn, 0);
3779 filt_fsevent(struct knote *kn, long hint)
3782 kn->kn_fflags |= hint;
3783 return (kn->kn_fflags != 0);
3787 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
3793 error = SYSCTL_IN(req, &vc, sizeof(vc));
3796 if (vc.vc_vers != VFS_CTL_VERS1)
3798 mp = vfs_getvfs(&vc.vc_fsid);
3801 /* ensure that a specific sysctl goes to the right filesystem. */
3802 if (strcmp(vc.vc_fstypename, "*") != 0 &&
3803 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
3807 VCTLTOREQ(&vc, req);
3808 error = VFS_SYSCTL(mp, vc.vc_op, req);
3813 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
3817 * Function to initialize a va_filerev field sensibly.
3818 * XXX: Wouldn't a random number make a lot more sense ??
3821 init_va_filerev(void)
3826 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
3829 static int filt_vfsread(struct knote *kn, long hint);
3830 static int filt_vfswrite(struct knote *kn, long hint);
3831 static int filt_vfsvnode(struct knote *kn, long hint);
3832 static void filt_vfsdetach(struct knote *kn);
3833 static struct filterops vfsread_filtops =
3834 { 1, NULL, filt_vfsdetach, filt_vfsread };
3835 static struct filterops vfswrite_filtops =
3836 { 1, NULL, filt_vfsdetach, filt_vfswrite };
3837 static struct filterops vfsvnode_filtops =
3838 { 1, NULL, filt_vfsdetach, filt_vfsvnode };
3841 vfs_knllock(void *arg)
3843 struct vnode *vp = arg;
3845 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
3849 vfs_knlunlock(void *arg)
3851 struct vnode *vp = arg;
3853 VOP_UNLOCK(vp, 0, curthread);
3857 vfs_knllocked(void *arg)
3859 struct vnode *vp = arg;
3861 return (VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE);
3865 vfs_kqfilter(struct vop_kqfilter_args *ap)
3867 struct vnode *vp = ap->a_vp;
3868 struct knote *kn = ap->a_kn;
3871 switch (kn->kn_filter) {
3873 kn->kn_fop = &vfsread_filtops;
3876 kn->kn_fop = &vfswrite_filtops;
3879 kn->kn_fop = &vfsvnode_filtops;
3885 kn->kn_hook = (caddr_t)vp;
3887 if (vp->v_pollinfo == NULL)
3889 if (vp->v_pollinfo == NULL)
3891 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
3892 knlist_add(knl, kn, 0);
3898 * Detach knote from vnode
3901 filt_vfsdetach(struct knote *kn)
3903 struct vnode *vp = (struct vnode *)kn->kn_hook;
3905 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
3906 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
3911 filt_vfsread(struct knote *kn, long hint)
3913 struct vnode *vp = (struct vnode *)kn->kn_hook;
3917 * filesystem is gone, so set the EOF flag and schedule
3918 * the knote for deletion.
3920 if (hint == NOTE_REVOKE) {
3921 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
3925 if (VOP_GETATTR(vp, &va, curthread->td_ucred, curthread))
3928 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
3929 return (kn->kn_data != 0);
3934 filt_vfswrite(struct knote *kn, long hint)
3937 * filesystem is gone, so set the EOF flag and schedule
3938 * the knote for deletion.
3940 if (hint == NOTE_REVOKE)
3941 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
3948 filt_vfsvnode(struct knote *kn, long hint)
3950 if (kn->kn_sfflags & hint)
3951 kn->kn_fflags |= hint;
3952 if (hint == NOTE_REVOKE) {
3953 kn->kn_flags |= EV_EOF;
3956 return (kn->kn_fflags != 0);
3960 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
3964 if (dp->d_reclen > ap->a_uio->uio_resid)
3965 return (ENAMETOOLONG);
3966 error = uiomove(dp, dp->d_reclen, ap->a_uio);
3968 if (ap->a_ncookies != NULL) {
3969 if (ap->a_cookies != NULL)
3970 free(ap->a_cookies, M_TEMP);
3971 ap->a_cookies = NULL;
3972 *ap->a_ncookies = 0;
3976 if (ap->a_ncookies == NULL)
3979 KASSERT(ap->a_cookies,
3980 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
3982 *ap->a_cookies = realloc(*ap->a_cookies,
3983 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
3984 (*ap->a_cookies)[*ap->a_ncookies] = off;
3989 * Mark for update the access time of the file if the filesystem
3990 * supports VA_MARK_ATIME. This functionality is used by execve
3991 * and mmap, so we want to avoid the synchronous I/O implied by
3992 * directly setting va_atime for the sake of efficiency.
3995 vfs_mark_atime(struct vnode *vp, struct thread *td)
3997 struct vattr atimeattr;
3999 if ((vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) {
4000 VATTR_NULL(&atimeattr);
4001 atimeattr.va_vaflags |= VA_MARK_ATIME;
4002 (void)VOP_SETATTR(vp, &atimeattr, td->td_ucred, td);