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("$FreeBSD$");
46 #include <sys/param.h>
47 #include <sys/systm.h>
50 #include <sys/condvar.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/sched.h>
69 #include <sys/sleepqueue.h>
71 #include <sys/sysctl.h>
72 #include <sys/syslog.h>
73 #include <sys/vmmeter.h>
74 #include <sys/vnode.h>
76 #include <machine/stdarg.h>
78 #include <security/mac/mac_framework.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_extern.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_kern.h>
96 static void delmntque(struct vnode *vp);
97 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
98 int slpflag, int slptimeo);
99 static void syncer_shutdown(void *arg, int howto);
100 static int vtryrecycle(struct vnode *vp);
101 static void vbusy(struct vnode *vp);
102 static void vinactive(struct vnode *, struct thread *);
103 static void v_incr_usecount(struct vnode *);
104 static void v_decr_usecount(struct vnode *);
105 static void v_decr_useonly(struct vnode *);
106 static void v_upgrade_usecount(struct vnode *);
107 static void vfree(struct vnode *);
108 static void vnlru_free(int);
109 static void vgonel(struct vnode *);
110 static void vfs_knllock(void *arg);
111 static void vfs_knlunlock(void *arg);
112 static void vfs_knl_assert_locked(void *arg);
113 static void vfs_knl_assert_unlocked(void *arg);
114 static void destroy_vpollinfo(struct vpollinfo *vi);
117 * Number of vnodes in existence. Increased whenever getnewvnode()
118 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
121 static unsigned long numvnodes;
123 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
124 "Number of vnodes in existence");
127 * Conversion tables for conversion from vnode types to inode formats
130 enum vtype iftovt_tab[16] = {
131 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
132 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
134 int vttoif_tab[10] = {
135 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
136 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
140 * List of vnodes that are ready for recycling.
142 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
145 * Free vnode target. Free vnodes may simply be files which have been stat'd
146 * but not read. This is somewhat common, and a small cache of such files
147 * should be kept to avoid recreation costs.
149 static u_long wantfreevnodes;
150 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
151 /* Number of vnodes in the free list. */
152 static u_long freevnodes;
153 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
154 "Number of vnodes in the free list");
156 static int vlru_allow_cache_src;
157 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
158 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
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,
167 "Number of calls to reassignbuf");
170 * Cache for the mount type id assigned to NFS. This is used for
171 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
173 int nfs_mount_type = -1;
175 /* To keep more than one thread at a time from running vfs_getnewfsid */
176 static struct mtx mntid_mtx;
179 * Lock for any access to the following:
184 static struct mtx vnode_free_list_mtx;
186 /* Publicly exported FS */
187 struct nfs_public nfs_pub;
189 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
190 static uma_zone_t vnode_zone;
191 static uma_zone_t vnodepoll_zone;
193 /* Set to 1 to print out reclaim of active vnodes */
197 * The workitem queue.
199 * It is useful to delay writes of file data and filesystem metadata
200 * for tens of seconds so that quickly created and deleted files need
201 * not waste disk bandwidth being created and removed. To realize this,
202 * we append vnodes to a "workitem" queue. When running with a soft
203 * updates implementation, most pending metadata dependencies should
204 * not wait for more than a few seconds. Thus, mounted on block devices
205 * are delayed only about a half the time that file data is delayed.
206 * Similarly, directory updates are more critical, so are only delayed
207 * about a third the time that file data is delayed. Thus, there are
208 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
209 * one each second (driven off the filesystem syncer process). The
210 * syncer_delayno variable indicates the next queue that is to be processed.
211 * Items that need to be processed soon are placed in this queue:
213 * syncer_workitem_pending[syncer_delayno]
215 * A delay of fifteen seconds is done by placing the request fifteen
216 * entries later in the queue:
218 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
221 static int syncer_delayno;
222 static long syncer_mask;
223 LIST_HEAD(synclist, bufobj);
224 static struct synclist *syncer_workitem_pending[2];
226 * The sync_mtx protects:
231 * syncer_workitem_pending
232 * syncer_worklist_len
235 static struct mtx sync_mtx;
236 static struct cv sync_wakeup;
238 #define SYNCER_MAXDELAY 32
239 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
240 static int syncdelay = 30; /* max time to delay syncing data */
241 static int filedelay = 30; /* time to delay syncing files */
242 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
243 "Time to delay syncing files (in seconds)");
244 static int dirdelay = 29; /* time to delay syncing directories */
245 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
246 "Time to delay syncing directories (in seconds)");
247 static int metadelay = 28; /* time to delay syncing metadata */
248 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
249 "Time to delay syncing metadata (in seconds)");
250 static int rushjob; /* number of slots to run ASAP */
251 static int stat_rush_requests; /* number of times I/O speeded up */
252 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
253 "Number of times I/O speeded up (rush requests)");
256 * When shutting down the syncer, run it at four times normal speed.
258 #define SYNCER_SHUTDOWN_SPEEDUP 4
259 static int sync_vnode_count;
260 static int syncer_worklist_len;
261 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
265 * Number of vnodes we want to exist at any one time. This is mostly used
266 * to size hash tables in vnode-related code. It is normally not used in
267 * getnewvnode(), as wantfreevnodes is normally nonzero.)
269 * XXX desiredvnodes is historical cruft and should not exist.
272 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
273 &desiredvnodes, 0, "Maximum number of vnodes");
274 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
275 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
276 static int vnlru_nowhere;
277 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
278 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
281 * Macros to control when a vnode is freed and recycled. All require
282 * the vnode interlock.
284 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
285 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
286 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
290 * Initialize the vnode management data structures.
292 * Reevaluate the following cap on the number of vnodes after the physical
293 * memory size exceeds 512GB. In the limit, as the physical memory size
294 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
296 #ifndef MAXVNODES_MAX
297 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
300 vntblinit(void *dummy __unused)
302 int physvnodes, virtvnodes;
305 * Desiredvnodes is a function of the physical memory size and the
306 * kernel's heap size. Generally speaking, it scales with the
307 * physical memory size. The ratio of desiredvnodes to physical pages
308 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
309 * marginal ratio of desiredvnodes to physical pages is one to
310 * sixteen. However, desiredvnodes is limited by the kernel's heap
311 * size. The memory required by desiredvnodes vnodes and vm objects
312 * may not exceed one seventh of the kernel's heap size.
314 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
315 cnt.v_page_count) / 16;
316 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
317 sizeof(struct vnode)));
318 desiredvnodes = min(physvnodes, virtvnodes);
319 if (desiredvnodes > MAXVNODES_MAX) {
321 printf("Reducing kern.maxvnodes %d -> %d\n",
322 desiredvnodes, MAXVNODES_MAX);
323 desiredvnodes = MAXVNODES_MAX;
325 wantfreevnodes = desiredvnodes / 4;
326 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
327 TAILQ_INIT(&vnode_free_list);
328 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
329 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
330 NULL, NULL, UMA_ALIGN_PTR, 0);
331 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
332 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
334 * Initialize the filesystem syncer.
336 syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
338 syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
340 syncer_maxdelay = syncer_mask + 1;
341 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
342 cv_init(&sync_wakeup, "syncer");
344 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
348 * Mark a mount point as busy. Used to synchronize access and to delay
349 * unmounting. Eventually, mountlist_mtx is not released on failure.
352 vfs_busy(struct mount *mp, int flags)
355 MPASS((flags & ~MBF_MASK) == 0);
356 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
361 * If mount point is currenly being unmounted, sleep until the
362 * mount point fate is decided. If thread doing the unmounting fails,
363 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
364 * that this mount point has survived the unmount attempt and vfs_busy
365 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
366 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
367 * about to be really destroyed. vfs_busy needs to release its
368 * reference on the mount point in this case and return with ENOENT,
369 * telling the caller that mount mount it tried to busy is no longer
372 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
373 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
376 CTR1(KTR_VFS, "%s: failed busying before sleeping",
380 if (flags & MBF_MNTLSTLOCK)
381 mtx_unlock(&mountlist_mtx);
382 mp->mnt_kern_flag |= MNTK_MWAIT;
383 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
384 if (flags & MBF_MNTLSTLOCK)
385 mtx_lock(&mountlist_mtx);
387 if (flags & MBF_MNTLSTLOCK)
388 mtx_unlock(&mountlist_mtx);
395 * Free a busy filesystem.
398 vfs_unbusy(struct mount *mp)
401 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
404 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
406 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
407 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
408 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
409 mp->mnt_kern_flag &= ~MNTK_DRAINING;
410 wakeup(&mp->mnt_lockref);
416 * Lookup a mount point by filesystem identifier.
419 vfs_getvfs(fsid_t *fsid)
423 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
424 mtx_lock(&mountlist_mtx);
425 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
426 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
427 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
429 mtx_unlock(&mountlist_mtx);
433 mtx_unlock(&mountlist_mtx);
434 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
435 return ((struct mount *) 0);
439 * Lookup a mount point by filesystem identifier, busying it before
443 vfs_busyfs(fsid_t *fsid)
448 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
449 mtx_lock(&mountlist_mtx);
450 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
451 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
452 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
453 error = vfs_busy(mp, MBF_MNTLSTLOCK);
455 mtx_unlock(&mountlist_mtx);
461 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
462 mtx_unlock(&mountlist_mtx);
463 return ((struct mount *) 0);
467 * Check if a user can access privileged mount options.
470 vfs_suser(struct mount *mp, struct thread *td)
475 * If the thread is jailed, but this is not a jail-friendly file
476 * system, deny immediately.
478 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
482 * If the file system was mounted outside the jail of the calling
483 * thread, deny immediately.
485 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
489 * If file system supports delegated administration, we don't check
490 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
491 * by the file system itself.
492 * If this is not the user that did original mount, we check for
493 * the PRIV_VFS_MOUNT_OWNER privilege.
495 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
496 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
497 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
504 * Get a new unique fsid. Try to make its val[0] unique, since this value
505 * will be used to create fake device numbers for stat(). Also try (but
506 * not so hard) make its val[0] unique mod 2^16, since some emulators only
507 * support 16-bit device numbers. We end up with unique val[0]'s for the
508 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
510 * Keep in mind that several mounts may be running in parallel. Starting
511 * the search one past where the previous search terminated is both a
512 * micro-optimization and a defense against returning the same fsid to
516 vfs_getnewfsid(struct mount *mp)
518 static u_int16_t mntid_base;
523 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
524 mtx_lock(&mntid_mtx);
525 mtype = mp->mnt_vfc->vfc_typenum;
526 tfsid.val[1] = mtype;
527 mtype = (mtype & 0xFF) << 24;
529 tfsid.val[0] = makedev(255,
530 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
532 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
536 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
537 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
538 mtx_unlock(&mntid_mtx);
542 * Knob to control the precision of file timestamps:
544 * 0 = seconds only; nanoseconds zeroed.
545 * 1 = seconds and nanoseconds, accurate within 1/HZ.
546 * 2 = seconds and nanoseconds, truncated to microseconds.
547 * >=3 = seconds and nanoseconds, maximum precision.
549 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
551 static int timestamp_precision = TSP_SEC;
552 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
553 ×tamp_precision, 0, "File timestamp precision (0: seconds, "
554 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
555 "3+: sec + ns (max. precision))");
558 * Get a current timestamp.
561 vfs_timestamp(struct timespec *tsp)
565 switch (timestamp_precision) {
567 tsp->tv_sec = time_second;
575 TIMEVAL_TO_TIMESPEC(&tv, tsp);
585 * Set vnode attributes to VNOVAL
588 vattr_null(struct vattr *vap)
592 vap->va_size = VNOVAL;
593 vap->va_bytes = VNOVAL;
594 vap->va_mode = VNOVAL;
595 vap->va_nlink = VNOVAL;
596 vap->va_uid = VNOVAL;
597 vap->va_gid = VNOVAL;
598 vap->va_fsid = VNOVAL;
599 vap->va_fileid = VNOVAL;
600 vap->va_blocksize = VNOVAL;
601 vap->va_rdev = VNOVAL;
602 vap->va_atime.tv_sec = VNOVAL;
603 vap->va_atime.tv_nsec = VNOVAL;
604 vap->va_mtime.tv_sec = VNOVAL;
605 vap->va_mtime.tv_nsec = VNOVAL;
606 vap->va_ctime.tv_sec = VNOVAL;
607 vap->va_ctime.tv_nsec = VNOVAL;
608 vap->va_birthtime.tv_sec = VNOVAL;
609 vap->va_birthtime.tv_nsec = VNOVAL;
610 vap->va_flags = VNOVAL;
611 vap->va_gen = VNOVAL;
616 * This routine is called when we have too many vnodes. It attempts
617 * to free <count> vnodes and will potentially free vnodes that still
618 * have VM backing store (VM backing store is typically the cause
619 * of a vnode blowout so we want to do this). Therefore, this operation
620 * is not considered cheap.
622 * A number of conditions may prevent a vnode from being reclaimed.
623 * the buffer cache may have references on the vnode, a directory
624 * vnode may still have references due to the namei cache representing
625 * underlying files, or the vnode may be in active use. It is not
626 * desireable to reuse such vnodes. These conditions may cause the
627 * number of vnodes to reach some minimum value regardless of what
628 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
631 vlrureclaim(struct mount *mp)
640 * Calculate the trigger point, don't allow user
641 * screwups to blow us up. This prevents us from
642 * recycling vnodes with lots of resident pages. We
643 * aren't trying to free memory, we are trying to
646 usevnodes = desiredvnodes;
649 trigger = cnt.v_page_count * 2 / usevnodes;
651 vn_start_write(NULL, &mp, V_WAIT);
653 count = mp->mnt_nvnodelistsize / 10 + 1;
655 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
656 while (vp != NULL && vp->v_type == VMARKER)
657 vp = TAILQ_NEXT(vp, v_nmntvnodes);
660 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
661 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
666 * If it's been deconstructed already, it's still
667 * referenced, or it exceeds the trigger, skip it.
669 if (vp->v_usecount ||
670 (!vlru_allow_cache_src &&
671 !LIST_EMPTY(&(vp)->v_cache_src)) ||
672 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
673 vp->v_object->resident_page_count > trigger)) {
679 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
681 goto next_iter_mntunlocked;
685 * v_usecount may have been bumped after VOP_LOCK() dropped
686 * the vnode interlock and before it was locked again.
688 * It is not necessary to recheck VI_DOOMED because it can
689 * only be set by another thread that holds both the vnode
690 * lock and vnode interlock. If another thread has the
691 * vnode lock before we get to VOP_LOCK() and obtains the
692 * vnode interlock after VOP_LOCK() drops the vnode
693 * interlock, the other thread will be unable to drop the
694 * vnode lock before our VOP_LOCK() call fails.
696 if (vp->v_usecount ||
697 (!vlru_allow_cache_src &&
698 !LIST_EMPTY(&(vp)->v_cache_src)) ||
699 (vp->v_object != NULL &&
700 vp->v_object->resident_page_count > trigger)) {
701 VOP_UNLOCK(vp, LK_INTERLOCK);
702 goto next_iter_mntunlocked;
704 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
705 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
710 next_iter_mntunlocked:
711 if ((count % 256) != 0)
715 if ((count % 256) != 0)
724 vn_finished_write(mp);
729 * Attempt to keep the free list at wantfreevnodes length.
732 vnlru_free(int count)
737 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
738 for (; count > 0; count--) {
739 vp = TAILQ_FIRST(&vnode_free_list);
741 * The list can be modified while the free_list_mtx
742 * has been dropped and vp could be NULL here.
746 VNASSERT(vp->v_op != NULL, vp,
747 ("vnlru_free: vnode already reclaimed."));
748 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
750 * Don't recycle if we can't get the interlock.
752 if (!VI_TRYLOCK(vp)) {
753 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
756 VNASSERT(VCANRECYCLE(vp), vp,
757 ("vp inconsistent on freelist"));
759 vp->v_iflag &= ~VI_FREE;
761 mtx_unlock(&vnode_free_list_mtx);
763 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
765 VFS_UNLOCK_GIANT(vfslocked);
767 * If the recycled succeeded this vdrop will actually free
768 * the vnode. If not it will simply place it back on
772 mtx_lock(&vnode_free_list_mtx);
776 * Attempt to recycle vnodes in a context that is always safe to block.
777 * Calling vlrurecycle() from the bowels of filesystem code has some
778 * interesting deadlock problems.
780 static struct proc *vnlruproc;
781 static int vnlruproc_sig;
786 struct mount *mp, *nmp;
788 struct proc *p = vnlruproc;
790 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
794 kproc_suspend_check(p);
795 mtx_lock(&vnode_free_list_mtx);
796 if (freevnodes > wantfreevnodes)
797 vnlru_free(freevnodes - wantfreevnodes);
798 if (numvnodes <= desiredvnodes * 9 / 10) {
800 wakeup(&vnlruproc_sig);
801 msleep(vnlruproc, &vnode_free_list_mtx,
802 PVFS|PDROP, "vlruwt", hz);
805 mtx_unlock(&vnode_free_list_mtx);
807 mtx_lock(&mountlist_mtx);
808 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
809 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
810 nmp = TAILQ_NEXT(mp, mnt_list);
813 vfslocked = VFS_LOCK_GIANT(mp);
814 done += vlrureclaim(mp);
815 VFS_UNLOCK_GIANT(vfslocked);
816 mtx_lock(&mountlist_mtx);
817 nmp = TAILQ_NEXT(mp, mnt_list);
820 mtx_unlock(&mountlist_mtx);
823 /* These messages are temporary debugging aids */
824 if (vnlru_nowhere < 5)
825 printf("vnlru process getting nowhere..\n");
826 else if (vnlru_nowhere == 5)
827 printf("vnlru process messages stopped.\n");
830 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
836 static struct kproc_desc vnlru_kp = {
841 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
845 * Routines having to do with the management of the vnode table.
849 vdestroy(struct vnode *vp)
853 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
854 mtx_lock(&vnode_free_list_mtx);
856 mtx_unlock(&vnode_free_list_mtx);
858 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
859 ("cleaned vnode still on the free list."));
860 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
861 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
862 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
863 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
864 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
865 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
866 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
867 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
868 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
869 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
870 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
871 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
874 mac_vnode_destroy(vp);
876 if (vp->v_pollinfo != NULL)
877 destroy_vpollinfo(vp->v_pollinfo);
879 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
882 lockdestroy(vp->v_vnlock);
883 mtx_destroy(&vp->v_interlock);
884 mtx_destroy(BO_MTX(bo));
885 uma_zfree(vnode_zone, vp);
889 * Try to recycle a freed vnode. We abort if anyone picks up a reference
890 * before we actually vgone(). This function must be called with the vnode
891 * held to prevent the vnode from being returned to the free list midway
895 vtryrecycle(struct vnode *vp)
899 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
900 VNASSERT(vp->v_holdcnt, vp,
901 ("vtryrecycle: Recycling vp %p without a reference.", vp));
903 * This vnode may found and locked via some other list, if so we
904 * can't recycle it yet.
906 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
908 "%s: impossible to recycle, vp %p lock is already held",
910 return (EWOULDBLOCK);
913 * Don't recycle if its filesystem is being suspended.
915 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
918 "%s: impossible to recycle, cannot start the write for %p",
923 * If we got this far, we need to acquire the interlock and see if
924 * anyone picked up this vnode from another list. If not, we will
925 * mark it with DOOMED via vgonel() so that anyone who does find it
929 if (vp->v_usecount) {
930 VOP_UNLOCK(vp, LK_INTERLOCK);
931 vn_finished_write(vnmp);
933 "%s: impossible to recycle, %p is already referenced",
937 if ((vp->v_iflag & VI_DOOMED) == 0)
939 VOP_UNLOCK(vp, LK_INTERLOCK);
940 vn_finished_write(vnmp);
945 * Return the next vnode from the free list.
948 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
951 struct vnode *vp = NULL;
954 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
955 mtx_lock(&vnode_free_list_mtx);
957 * Lend our context to reclaim vnodes if they've exceeded the max.
959 if (freevnodes > wantfreevnodes)
962 * Wait for available vnodes.
964 if (numvnodes > desiredvnodes) {
965 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
967 * File system is beeing suspended, we cannot risk a
968 * deadlock here, so allocate new vnode anyway.
970 if (freevnodes > wantfreevnodes)
971 vnlru_free(freevnodes - wantfreevnodes);
974 if (vnlruproc_sig == 0) {
975 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
978 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
980 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
981 if (numvnodes > desiredvnodes) {
982 mtx_unlock(&vnode_free_list_mtx);
989 mtx_unlock(&vnode_free_list_mtx);
990 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
994 vp->v_vnlock = &vp->v_lock;
995 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
997 * By default, don't allow shared locks unless filesystems
1000 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
1002 * Initialize bufobj.
1005 bo->__bo_vnode = vp;
1006 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
1007 bo->bo_ops = &buf_ops_bio;
1008 bo->bo_private = vp;
1009 TAILQ_INIT(&bo->bo_clean.bv_hd);
1010 TAILQ_INIT(&bo->bo_dirty.bv_hd);
1012 * Initialize namecache.
1014 LIST_INIT(&vp->v_cache_src);
1015 TAILQ_INIT(&vp->v_cache_dst);
1017 * Finalize various vnode identity bits.
1022 v_incr_usecount(vp);
1026 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1027 mac_vnode_associate_singlelabel(mp, vp);
1028 else if (mp == NULL && vops != &dead_vnodeops)
1029 printf("NULL mp in getnewvnode()\n");
1032 bo->bo_bsize = mp->mnt_stat.f_iosize;
1033 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1034 vp->v_vflag |= VV_NOKNOTE;
1042 * Delete from old mount point vnode list, if on one.
1045 delmntque(struct vnode *vp)
1054 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1055 ("bad mount point vnode list size"));
1056 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1057 mp->mnt_nvnodelistsize--;
1063 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1067 vp->v_op = &dead_vnodeops;
1068 /* XXX non mp-safe fs may still call insmntque with vnode
1070 if (!VOP_ISLOCKED(vp))
1071 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1077 * Insert into list of vnodes for the new mount point, if available.
1080 insmntque1(struct vnode *vp, struct mount *mp,
1081 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1085 KASSERT(vp->v_mount == NULL,
1086 ("insmntque: vnode already on per mount vnode list"));
1087 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1088 #ifdef DEBUG_VFS_LOCKS
1089 if (!VFS_NEEDSGIANT(mp))
1090 ASSERT_VOP_ELOCKED(vp,
1091 "insmntque: mp-safe fs and non-locked vp");
1094 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1095 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1096 mp->mnt_nvnodelistsize == 0)) {
1097 locked = VOP_ISLOCKED(vp);
1098 if (!locked || (locked == LK_EXCLUSIVE &&
1099 (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1108 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1109 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1110 ("neg mount point vnode list size"));
1111 mp->mnt_nvnodelistsize++;
1117 insmntque(struct vnode *vp, struct mount *mp)
1120 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1124 * Flush out and invalidate all buffers associated with a bufobj
1125 * Called with the underlying object locked.
1128 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1133 if (flags & V_SAVE) {
1134 error = bufobj_wwait(bo, slpflag, slptimeo);
1139 if (bo->bo_dirty.bv_cnt > 0) {
1141 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1144 * XXX We could save a lock/unlock if this was only
1145 * enabled under INVARIANTS
1148 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1149 panic("vinvalbuf: dirty bufs");
1153 * If you alter this loop please notice that interlock is dropped and
1154 * reacquired in flushbuflist. Special care is needed to ensure that
1155 * no race conditions occur from this.
1158 error = flushbuflist(&bo->bo_clean,
1159 flags, bo, slpflag, slptimeo);
1161 error = flushbuflist(&bo->bo_dirty,
1162 flags, bo, slpflag, slptimeo);
1163 if (error != 0 && error != EAGAIN) {
1167 } while (error != 0);
1170 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1171 * have write I/O in-progress but if there is a VM object then the
1172 * VM object can also have read-I/O in-progress.
1175 bufobj_wwait(bo, 0, 0);
1177 if (bo->bo_object != NULL) {
1178 VM_OBJECT_LOCK(bo->bo_object);
1179 vm_object_pip_wait(bo->bo_object, "bovlbx");
1180 VM_OBJECT_UNLOCK(bo->bo_object);
1183 } while (bo->bo_numoutput > 0);
1187 * Destroy the copy in the VM cache, too.
1189 if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) {
1190 VM_OBJECT_LOCK(bo->bo_object);
1191 vm_object_page_remove(bo->bo_object, 0, 0,
1192 (flags & V_SAVE) ? TRUE : FALSE);
1193 VM_OBJECT_UNLOCK(bo->bo_object);
1198 if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1199 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1200 panic("vinvalbuf: flush failed");
1207 * Flush out and invalidate all buffers associated with a vnode.
1208 * Called with the underlying object locked.
1211 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1214 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1215 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1216 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1220 * Flush out buffers on the specified list.
1224 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1227 struct buf *bp, *nbp;
1232 ASSERT_BO_LOCKED(bo);
1235 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1236 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1237 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1243 lblkno = nbp->b_lblkno;
1244 xflags = nbp->b_xflags &
1245 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1248 error = BUF_TIMELOCK(bp,
1249 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1250 "flushbuf", slpflag, slptimeo);
1253 return (error != ENOLCK ? error : EAGAIN);
1255 KASSERT(bp->b_bufobj == bo,
1256 ("bp %p wrong b_bufobj %p should be %p",
1257 bp, bp->b_bufobj, bo));
1258 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1264 * XXX Since there are no node locks for NFS, I
1265 * believe there is a slight chance that a delayed
1266 * write will occur while sleeping just above, so
1269 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1274 bp->b_flags |= B_ASYNC;
1277 return (EAGAIN); /* XXX: why not loop ? */
1282 bp->b_flags |= (B_INVAL | B_RELBUF);
1283 bp->b_flags &= ~B_ASYNC;
1287 (nbp->b_bufobj != bo ||
1288 nbp->b_lblkno != lblkno ||
1290 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1291 break; /* nbp invalid */
1297 * Truncate a file's buffer and pages to a specified length. This
1298 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1302 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1303 off_t length, int blksize)
1305 struct buf *bp, *nbp;
1310 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1311 vp, cred, blksize, (uintmax_t)length);
1314 * Round up to the *next* lbn.
1316 trunclbn = (length + blksize - 1) / blksize;
1318 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1325 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1326 if (bp->b_lblkno < trunclbn)
1329 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1330 BO_MTX(bo)) == ENOLCK)
1336 bp->b_flags |= (B_INVAL | B_RELBUF);
1337 bp->b_flags &= ~B_ASYNC;
1343 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1344 (nbp->b_vp != vp) ||
1345 (nbp->b_flags & B_DELWRI))) {
1351 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1352 if (bp->b_lblkno < trunclbn)
1355 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1356 BO_MTX(bo)) == ENOLCK)
1361 bp->b_flags |= (B_INVAL | B_RELBUF);
1362 bp->b_flags &= ~B_ASYNC;
1368 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1369 (nbp->b_vp != vp) ||
1370 (nbp->b_flags & B_DELWRI) == 0)) {
1379 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1380 if (bp->b_lblkno > 0)
1383 * Since we hold the vnode lock this should only
1384 * fail if we're racing with the buf daemon.
1387 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1388 BO_MTX(bo)) == ENOLCK) {
1391 VNASSERT((bp->b_flags & B_DELWRI), vp,
1392 ("buf(%p) on dirty queue without DELWRI", bp));
1403 bufobj_wwait(bo, 0, 0);
1405 vnode_pager_setsize(vp, length);
1411 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1414 * NOTE: We have to deal with the special case of a background bitmap
1415 * buffer, a situation where two buffers will have the same logical
1416 * block offset. We want (1) only the foreground buffer to be accessed
1417 * in a lookup and (2) must differentiate between the foreground and
1418 * background buffer in the splay tree algorithm because the splay
1419 * tree cannot normally handle multiple entities with the same 'index'.
1420 * We accomplish this by adding differentiating flags to the splay tree's
1425 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1428 struct buf *lefttreemax, *righttreemin, *y;
1432 lefttreemax = righttreemin = &dummy;
1434 if (lblkno < root->b_lblkno ||
1435 (lblkno == root->b_lblkno &&
1436 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1437 if ((y = root->b_left) == NULL)
1439 if (lblkno < y->b_lblkno) {
1441 root->b_left = y->b_right;
1444 if ((y = root->b_left) == NULL)
1447 /* Link into the new root's right tree. */
1448 righttreemin->b_left = root;
1449 righttreemin = root;
1450 } else if (lblkno > root->b_lblkno ||
1451 (lblkno == root->b_lblkno &&
1452 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1453 if ((y = root->b_right) == NULL)
1455 if (lblkno > y->b_lblkno) {
1457 root->b_right = y->b_left;
1460 if ((y = root->b_right) == NULL)
1463 /* Link into the new root's left tree. */
1464 lefttreemax->b_right = root;
1471 /* Assemble the new root. */
1472 lefttreemax->b_right = root->b_left;
1473 righttreemin->b_left = root->b_right;
1474 root->b_left = dummy.b_right;
1475 root->b_right = dummy.b_left;
1480 buf_vlist_remove(struct buf *bp)
1485 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1486 ASSERT_BO_LOCKED(bp->b_bufobj);
1487 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1488 (BX_VNDIRTY|BX_VNCLEAN),
1489 ("buf_vlist_remove: Buf %p is on two lists", bp));
1490 if (bp->b_xflags & BX_VNDIRTY)
1491 bv = &bp->b_bufobj->bo_dirty;
1493 bv = &bp->b_bufobj->bo_clean;
1494 if (bp != bv->bv_root) {
1495 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1496 KASSERT(root == bp, ("splay lookup failed in remove"));
1498 if (bp->b_left == NULL) {
1501 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1502 root->b_right = bp->b_right;
1505 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1507 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1511 * Add the buffer to the sorted clean or dirty block list using a
1512 * splay tree algorithm.
1514 * NOTE: xflags is passed as a constant, optimizing this inline function!
1517 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1522 ASSERT_BO_LOCKED(bo);
1523 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1524 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1525 bp->b_xflags |= xflags;
1526 if (xflags & BX_VNDIRTY)
1531 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1535 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1536 } else if (bp->b_lblkno < root->b_lblkno ||
1537 (bp->b_lblkno == root->b_lblkno &&
1538 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1539 bp->b_left = root->b_left;
1541 root->b_left = NULL;
1542 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1544 bp->b_right = root->b_right;
1546 root->b_right = NULL;
1547 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1554 * Lookup a buffer using the splay tree. Note that we specifically avoid
1555 * shadow buffers used in background bitmap writes.
1557 * This code isn't quite efficient as it could be because we are maintaining
1558 * two sorted lists and do not know which list the block resides in.
1560 * During a "make buildworld" the desired buffer is found at one of
1561 * the roots more than 60% of the time. Thus, checking both roots
1562 * before performing either splay eliminates unnecessary splays on the
1563 * first tree splayed.
1566 gbincore(struct bufobj *bo, daddr_t lblkno)
1570 ASSERT_BO_LOCKED(bo);
1571 if ((bp = bo->bo_clean.bv_root) != NULL &&
1572 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1574 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1575 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1577 if ((bp = bo->bo_clean.bv_root) != NULL) {
1578 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1579 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1582 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1583 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1584 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1591 * Associate a buffer with a vnode.
1594 bgetvp(struct vnode *vp, struct buf *bp)
1599 ASSERT_BO_LOCKED(bo);
1600 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1602 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1603 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1604 ("bgetvp: bp already attached! %p", bp));
1607 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1608 bp->b_flags |= B_NEEDSGIANT;
1612 * Insert onto list for new vnode.
1614 buf_vlist_add(bp, bo, BX_VNCLEAN);
1618 * Disassociate a buffer from a vnode.
1621 brelvp(struct buf *bp)
1626 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1627 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1630 * Delete from old vnode list, if on one.
1632 vp = bp->b_vp; /* XXX */
1635 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1636 buf_vlist_remove(bp);
1638 panic("brelvp: Buffer %p not on queue.", bp);
1639 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1640 bo->bo_flag &= ~BO_ONWORKLST;
1641 mtx_lock(&sync_mtx);
1642 LIST_REMOVE(bo, bo_synclist);
1643 syncer_worklist_len--;
1644 mtx_unlock(&sync_mtx);
1646 bp->b_flags &= ~B_NEEDSGIANT;
1648 bp->b_bufobj = NULL;
1654 * Add an item to the syncer work queue.
1657 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1661 ASSERT_BO_LOCKED(bo);
1663 mtx_lock(&sync_mtx);
1664 if (bo->bo_flag & BO_ONWORKLST)
1665 LIST_REMOVE(bo, bo_synclist);
1667 bo->bo_flag |= BO_ONWORKLST;
1668 syncer_worklist_len++;
1671 if (delay > syncer_maxdelay - 2)
1672 delay = syncer_maxdelay - 2;
1673 slot = (syncer_delayno + delay) & syncer_mask;
1675 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1677 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1679 mtx_unlock(&sync_mtx);
1683 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1687 mtx_lock(&sync_mtx);
1688 len = syncer_worklist_len - sync_vnode_count;
1689 mtx_unlock(&sync_mtx);
1690 error = SYSCTL_OUT(req, &len, sizeof(len));
1694 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1695 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1697 static struct proc *updateproc;
1698 static void sched_sync(void);
1699 static struct kproc_desc up_kp = {
1704 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1707 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1712 *bo = LIST_FIRST(slp);
1715 vp = (*bo)->__bo_vnode; /* XXX */
1716 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1719 * We use vhold in case the vnode does not
1720 * successfully sync. vhold prevents the vnode from
1721 * going away when we unlock the sync_mtx so that
1722 * we can acquire the vnode interlock.
1725 mtx_unlock(&sync_mtx);
1727 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1729 mtx_lock(&sync_mtx);
1730 return (*bo == LIST_FIRST(slp));
1732 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1733 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1735 vn_finished_write(mp);
1737 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1739 * Put us back on the worklist. The worklist
1740 * routine will remove us from our current
1741 * position and then add us back in at a later
1744 vn_syncer_add_to_worklist(*bo, syncdelay);
1748 mtx_lock(&sync_mtx);
1753 * System filesystem synchronizer daemon.
1758 struct synclist *gnext, *next;
1759 struct synclist *gslp, *slp;
1762 struct thread *td = curthread;
1764 int net_worklist_len;
1765 int syncer_final_iter;
1770 syncer_final_iter = 0;
1772 syncer_state = SYNCER_RUNNING;
1773 starttime = time_uptime;
1774 td->td_pflags |= TDP_NORUNNINGBUF;
1776 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1779 mtx_lock(&sync_mtx);
1781 if (syncer_state == SYNCER_FINAL_DELAY &&
1782 syncer_final_iter == 0) {
1783 mtx_unlock(&sync_mtx);
1784 kproc_suspend_check(td->td_proc);
1785 mtx_lock(&sync_mtx);
1787 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1788 if (syncer_state != SYNCER_RUNNING &&
1789 starttime != time_uptime) {
1791 printf("\nSyncing disks, vnodes remaining...");
1794 printf("%d ", net_worklist_len);
1796 starttime = time_uptime;
1799 * Push files whose dirty time has expired. Be careful
1800 * of interrupt race on slp queue.
1802 * Skip over empty worklist slots when shutting down.
1805 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1806 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1807 syncer_delayno += 1;
1808 if (syncer_delayno == syncer_maxdelay)
1810 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1811 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1813 * If the worklist has wrapped since the
1814 * it was emptied of all but syncer vnodes,
1815 * switch to the FINAL_DELAY state and run
1816 * for one more second.
1818 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1819 net_worklist_len == 0 &&
1820 last_work_seen == syncer_delayno) {
1821 syncer_state = SYNCER_FINAL_DELAY;
1822 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1824 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1825 LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1828 * Keep track of the last time there was anything
1829 * on the worklist other than syncer vnodes.
1830 * Return to the SHUTTING_DOWN state if any
1833 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1834 last_work_seen = syncer_delayno;
1835 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1836 syncer_state = SYNCER_SHUTTING_DOWN;
1837 while (!LIST_EMPTY(slp)) {
1838 error = sync_vnode(slp, &bo, td);
1840 LIST_REMOVE(bo, bo_synclist);
1841 LIST_INSERT_HEAD(next, bo, bo_synclist);
1845 if (!LIST_EMPTY(gslp)) {
1846 mtx_unlock(&sync_mtx);
1848 mtx_lock(&sync_mtx);
1849 while (!LIST_EMPTY(gslp)) {
1850 error = sync_vnode(gslp, &bo, td);
1852 LIST_REMOVE(bo, bo_synclist);
1853 LIST_INSERT_HEAD(gnext, bo,
1860 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1861 syncer_final_iter--;
1863 * The variable rushjob allows the kernel to speed up the
1864 * processing of the filesystem syncer process. A rushjob
1865 * value of N tells the filesystem syncer to process the next
1866 * N seconds worth of work on its queue ASAP. Currently rushjob
1867 * is used by the soft update code to speed up the filesystem
1868 * syncer process when the incore state is getting so far
1869 * ahead of the disk that the kernel memory pool is being
1870 * threatened with exhaustion.
1877 * Just sleep for a short period of time between
1878 * iterations when shutting down to allow some I/O
1881 * If it has taken us less than a second to process the
1882 * current work, then wait. Otherwise start right over
1883 * again. We can still lose time if any single round
1884 * takes more than two seconds, but it does not really
1885 * matter as we are just trying to generally pace the
1886 * filesystem activity.
1888 if (syncer_state != SYNCER_RUNNING ||
1889 time_uptime == starttime) {
1891 sched_prio(td, PPAUSE);
1894 if (syncer_state != SYNCER_RUNNING)
1895 cv_timedwait(&sync_wakeup, &sync_mtx,
1896 hz / SYNCER_SHUTDOWN_SPEEDUP);
1897 else if (time_uptime == starttime)
1898 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1903 * Request the syncer daemon to speed up its work.
1904 * We never push it to speed up more than half of its
1905 * normal turn time, otherwise it could take over the cpu.
1908 speedup_syncer(void)
1912 mtx_lock(&sync_mtx);
1913 if (rushjob < syncdelay / 2) {
1915 stat_rush_requests += 1;
1918 mtx_unlock(&sync_mtx);
1919 cv_broadcast(&sync_wakeup);
1924 * Tell the syncer to speed up its work and run though its work
1925 * list several times, then tell it to shut down.
1928 syncer_shutdown(void *arg, int howto)
1931 if (howto & RB_NOSYNC)
1933 mtx_lock(&sync_mtx);
1934 syncer_state = SYNCER_SHUTTING_DOWN;
1936 mtx_unlock(&sync_mtx);
1937 cv_broadcast(&sync_wakeup);
1938 kproc_shutdown(arg, howto);
1942 * Reassign a buffer from one vnode to another.
1943 * Used to assign file specific control information
1944 * (indirect blocks) to the vnode to which they belong.
1947 reassignbuf(struct buf *bp)
1960 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1961 bp, bp->b_vp, bp->b_flags);
1963 * B_PAGING flagged buffers cannot be reassigned because their vp
1964 * is not fully linked in.
1966 if (bp->b_flags & B_PAGING)
1967 panic("cannot reassign paging buffer");
1970 * Delete from old vnode list, if on one.
1973 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1974 buf_vlist_remove(bp);
1976 panic("reassignbuf: Buffer %p not on queue.", bp);
1978 * If dirty, put on list of dirty buffers; otherwise insert onto list
1981 if (bp->b_flags & B_DELWRI) {
1982 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1983 switch (vp->v_type) {
1993 vn_syncer_add_to_worklist(bo, delay);
1995 buf_vlist_add(bp, bo, BX_VNDIRTY);
1997 buf_vlist_add(bp, bo, BX_VNCLEAN);
1999 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2000 mtx_lock(&sync_mtx);
2001 LIST_REMOVE(bo, bo_synclist);
2002 syncer_worklist_len--;
2003 mtx_unlock(&sync_mtx);
2004 bo->bo_flag &= ~BO_ONWORKLST;
2009 bp = TAILQ_FIRST(&bv->bv_hd);
2010 KASSERT(bp == NULL || bp->b_bufobj == bo,
2011 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2012 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2013 KASSERT(bp == NULL || bp->b_bufobj == bo,
2014 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2016 bp = TAILQ_FIRST(&bv->bv_hd);
2017 KASSERT(bp == NULL || bp->b_bufobj == bo,
2018 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2019 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2020 KASSERT(bp == NULL || bp->b_bufobj == bo,
2021 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2027 * Increment the use and hold counts on the vnode, taking care to reference
2028 * the driver's usecount if this is a chardev. The vholdl() will remove
2029 * the vnode from the free list if it is presently free. Requires the
2030 * vnode interlock and returns with it held.
2033 v_incr_usecount(struct vnode *vp)
2036 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2038 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2040 vp->v_rdev->si_usecount++;
2047 * Turn a holdcnt into a use+holdcnt such that only one call to
2048 * v_decr_usecount is needed.
2051 v_upgrade_usecount(struct vnode *vp)
2054 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2056 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2058 vp->v_rdev->si_usecount++;
2064 * Decrement the vnode use and hold count along with the driver's usecount
2065 * if this is a chardev. The vdropl() below releases the vnode interlock
2066 * as it may free the vnode.
2069 v_decr_usecount(struct vnode *vp)
2072 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2073 VNASSERT(vp->v_usecount > 0, vp,
2074 ("v_decr_usecount: negative usecount"));
2075 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2077 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2079 vp->v_rdev->si_usecount--;
2086 * Decrement only the use count and driver use count. This is intended to
2087 * be paired with a follow on vdropl() to release the remaining hold count.
2088 * In this way we may vgone() a vnode with a 0 usecount without risk of
2089 * having it end up on a free list because the hold count is kept above 0.
2092 v_decr_useonly(struct vnode *vp)
2095 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2096 VNASSERT(vp->v_usecount > 0, vp,
2097 ("v_decr_useonly: negative usecount"));
2098 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2100 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2102 vp->v_rdev->si_usecount--;
2108 * Grab a particular vnode from the free list, increment its
2109 * reference count and lock it. VI_DOOMED is set if the vnode
2110 * is being destroyed. Only callers who specify LK_RETRY will
2111 * see doomed vnodes. If inactive processing was delayed in
2112 * vput try to do it here.
2115 vget(struct vnode *vp, int flags, struct thread *td)
2120 VFS_ASSERT_GIANT(vp->v_mount);
2121 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2122 ("vget: invalid lock operation"));
2123 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2125 if ((flags & LK_INTERLOCK) == 0)
2128 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2130 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2134 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2135 panic("vget: vn_lock failed to return ENOENT\n");
2137 /* Upgrade our holdcnt to a usecount. */
2138 v_upgrade_usecount(vp);
2140 * We don't guarantee that any particular close will
2141 * trigger inactive processing so just make a best effort
2142 * here at preventing a reference to a removed file. If
2143 * we don't succeed no harm is done.
2145 if (vp->v_iflag & VI_OWEINACT) {
2146 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2147 (flags & LK_NOWAIT) == 0)
2149 vp->v_iflag &= ~VI_OWEINACT;
2156 * Increase the reference count of a vnode.
2159 vref(struct vnode *vp)
2162 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2164 v_incr_usecount(vp);
2169 * Return reference count of a vnode.
2171 * The results of this call are only guaranteed when some mechanism other
2172 * than the VI lock is used to stop other processes from gaining references
2173 * to the vnode. This may be the case if the caller holds the only reference.
2174 * This is also useful when stale data is acceptable as race conditions may
2175 * be accounted for by some other means.
2178 vrefcnt(struct vnode *vp)
2183 usecnt = vp->v_usecount;
2189 #define VPUTX_VRELE 1
2190 #define VPUTX_VPUT 2
2191 #define VPUTX_VUNREF 3
2194 vputx(struct vnode *vp, int func)
2198 KASSERT(vp != NULL, ("vputx: null vp"));
2199 if (func == VPUTX_VUNREF)
2200 ASSERT_VOP_LOCKED(vp, "vunref");
2201 else if (func == VPUTX_VPUT)
2202 ASSERT_VOP_LOCKED(vp, "vput");
2204 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2205 VFS_ASSERT_GIANT(vp->v_mount);
2206 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2209 /* Skip this v_writecount check if we're going to panic below. */
2210 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2211 ("vputx: missed vn_close"));
2214 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2215 vp->v_usecount == 1)) {
2216 if (func == VPUTX_VPUT)
2218 v_decr_usecount(vp);
2222 if (vp->v_usecount != 1) {
2223 vprint("vputx: negative ref count", vp);
2224 panic("vputx: negative ref cnt");
2226 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2228 * We want to hold the vnode until the inactive finishes to
2229 * prevent vgone() races. We drop the use count here and the
2230 * hold count below when we're done.
2234 * We must call VOP_INACTIVE with the node locked. Mark
2235 * as VI_DOINGINACT to avoid recursion.
2237 vp->v_iflag |= VI_OWEINACT;
2240 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2244 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2245 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2251 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2255 if (vp->v_usecount > 0)
2256 vp->v_iflag &= ~VI_OWEINACT;
2258 if (vp->v_iflag & VI_OWEINACT)
2259 vinactive(vp, curthread);
2260 if (func != VPUTX_VUNREF)
2267 * Vnode put/release.
2268 * If count drops to zero, call inactive routine and return to freelist.
2271 vrele(struct vnode *vp)
2274 vputx(vp, VPUTX_VRELE);
2278 * Release an already locked vnode. This give the same effects as
2279 * unlock+vrele(), but takes less time and avoids releasing and
2280 * re-aquiring the lock (as vrele() acquires the lock internally.)
2283 vput(struct vnode *vp)
2286 vputx(vp, VPUTX_VPUT);
2290 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2293 vunref(struct vnode *vp)
2296 vputx(vp, VPUTX_VUNREF);
2300 * Somebody doesn't want the vnode recycled.
2303 vhold(struct vnode *vp)
2312 vholdl(struct vnode *vp)
2315 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2317 if (VSHOULDBUSY(vp))
2322 * Note that there is one less who cares about this vnode. vdrop() is the
2323 * opposite of vhold().
2326 vdrop(struct vnode *vp)
2334 * Drop the hold count of the vnode. If this is the last reference to
2335 * the vnode we will free it if it has been vgone'd otherwise it is
2336 * placed on the free list.
2339 vdropl(struct vnode *vp)
2342 ASSERT_VI_LOCKED(vp, "vdropl");
2343 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2344 if (vp->v_holdcnt <= 0)
2345 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2347 if (vp->v_holdcnt == 0) {
2348 if (vp->v_iflag & VI_DOOMED) {
2349 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__,
2360 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2361 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2362 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2363 * failed lock upgrade.
2366 vinactive(struct vnode *vp, struct thread *td)
2369 ASSERT_VOP_ELOCKED(vp, "vinactive");
2370 ASSERT_VI_LOCKED(vp, "vinactive");
2371 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2372 ("vinactive: recursed on VI_DOINGINACT"));
2373 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2374 vp->v_iflag |= VI_DOINGINACT;
2375 vp->v_iflag &= ~VI_OWEINACT;
2377 VOP_INACTIVE(vp, td);
2379 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2380 ("vinactive: lost VI_DOINGINACT"));
2381 vp->v_iflag &= ~VI_DOINGINACT;
2385 * Remove any vnodes in the vnode table belonging to mount point mp.
2387 * If FORCECLOSE is not specified, there should not be any active ones,
2388 * return error if any are found (nb: this is a user error, not a
2389 * system error). If FORCECLOSE is specified, detach any active vnodes
2392 * If WRITECLOSE is set, only flush out regular file vnodes open for
2395 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2397 * `rootrefs' specifies the base reference count for the root vnode
2398 * of this filesystem. The root vnode is considered busy if its
2399 * v_usecount exceeds this value. On a successful return, vflush(, td)
2400 * will call vrele() on the root vnode exactly rootrefs times.
2401 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2405 static int busyprt = 0; /* print out busy vnodes */
2406 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2410 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2412 struct vnode *vp, *mvp, *rootvp = NULL;
2414 int busy = 0, error;
2416 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2419 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2420 ("vflush: bad args"));
2422 * Get the filesystem root vnode. We can vput() it
2423 * immediately, since with rootrefs > 0, it won't go away.
2425 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2426 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2435 MNT_VNODE_FOREACH(vp, mp, mvp) {
2440 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2444 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2448 * Skip over a vnodes marked VV_SYSTEM.
2450 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2457 * If WRITECLOSE is set, flush out unlinked but still open
2458 * files (even if open only for reading) and regular file
2459 * vnodes open for writing.
2461 if (flags & WRITECLOSE) {
2462 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2465 if ((vp->v_type == VNON ||
2466 (error == 0 && vattr.va_nlink > 0)) &&
2467 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2476 * With v_usecount == 0, all we need to do is clear out the
2477 * vnode data structures and we are done.
2479 * If FORCECLOSE is set, forcibly close the vnode.
2481 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2482 VNASSERT(vp->v_usecount == 0 ||
2483 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2484 ("device VNODE %p is FORCECLOSED", vp));
2490 vprint("vflush: busy vnode", vp);
2498 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2500 * If just the root vnode is busy, and if its refcount
2501 * is equal to `rootrefs', then go ahead and kill it.
2504 KASSERT(busy > 0, ("vflush: not busy"));
2505 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2506 ("vflush: usecount %d < rootrefs %d",
2507 rootvp->v_usecount, rootrefs));
2508 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2509 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2511 VOP_UNLOCK(rootvp, 0);
2517 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2521 for (; rootrefs > 0; rootrefs--)
2527 * Recycle an unused vnode to the front of the free list.
2530 vrecycle(struct vnode *vp, struct thread *td)
2534 ASSERT_VOP_ELOCKED(vp, "vrecycle");
2535 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2538 if (vp->v_usecount == 0) {
2547 * Eliminate all activity associated with a vnode
2548 * in preparation for reuse.
2551 vgone(struct vnode *vp)
2559 * vgone, with the vp interlock held.
2562 vgonel(struct vnode *vp)
2569 ASSERT_VOP_ELOCKED(vp, "vgonel");
2570 ASSERT_VI_LOCKED(vp, "vgonel");
2571 VNASSERT(vp->v_holdcnt, vp,
2572 ("vgonel: vp %p has no reference.", vp));
2573 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2577 * Don't vgonel if we're already doomed.
2579 if (vp->v_iflag & VI_DOOMED)
2581 vp->v_iflag |= VI_DOOMED;
2583 * Check to see if the vnode is in use. If so, we have to call
2584 * VOP_CLOSE() and VOP_INACTIVE().
2586 active = vp->v_usecount;
2587 oweinact = (vp->v_iflag & VI_OWEINACT);
2590 * Clean out any buffers associated with the vnode.
2591 * If the flush fails, just toss the buffers.
2594 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2595 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2596 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2597 vinvalbuf(vp, 0, 0, 0);
2600 * If purging an active vnode, it must be closed and
2601 * deactivated before being reclaimed.
2604 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2605 if (oweinact || active) {
2607 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2612 * Reclaim the vnode.
2614 if (VOP_RECLAIM(vp, td))
2615 panic("vgone: cannot reclaim");
2617 vn_finished_secondary_write(mp);
2618 VNASSERT(vp->v_object == NULL, vp,
2619 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2621 * Clear the advisory locks and wake up waiting threads.
2623 lf_purgelocks(vp, &(vp->v_lockf));
2625 * Delete from old mount point vnode list.
2630 * Done with purge, reset to the standard lock and invalidate
2634 vp->v_vnlock = &vp->v_lock;
2635 vp->v_op = &dead_vnodeops;
2641 * Calculate the total number of references to a special device.
2644 vcount(struct vnode *vp)
2649 count = vp->v_rdev->si_usecount;
2655 * Same as above, but using the struct cdev *as argument
2658 count_dev(struct cdev *dev)
2663 count = dev->si_usecount;
2669 * Print out a description of a vnode.
2671 static char *typename[] =
2672 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2676 vn_printf(struct vnode *vp, const char *fmt, ...)
2679 char buf[256], buf2[16];
2685 printf("%p: ", (void *)vp);
2686 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2687 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2688 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2691 if (vp->v_vflag & VV_ROOT)
2692 strlcat(buf, "|VV_ROOT", sizeof(buf));
2693 if (vp->v_vflag & VV_ISTTY)
2694 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2695 if (vp->v_vflag & VV_NOSYNC)
2696 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2697 if (vp->v_vflag & VV_CACHEDLABEL)
2698 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2699 if (vp->v_vflag & VV_TEXT)
2700 strlcat(buf, "|VV_TEXT", sizeof(buf));
2701 if (vp->v_vflag & VV_COPYONWRITE)
2702 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2703 if (vp->v_vflag & VV_SYSTEM)
2704 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2705 if (vp->v_vflag & VV_PROCDEP)
2706 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2707 if (vp->v_vflag & VV_NOKNOTE)
2708 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2709 if (vp->v_vflag & VV_DELETED)
2710 strlcat(buf, "|VV_DELETED", sizeof(buf));
2711 if (vp->v_vflag & VV_MD)
2712 strlcat(buf, "|VV_MD", sizeof(buf));
2713 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2714 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2715 VV_NOKNOTE | VV_DELETED | VV_MD);
2717 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2718 strlcat(buf, buf2, sizeof(buf));
2720 if (vp->v_iflag & VI_MOUNT)
2721 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2722 if (vp->v_iflag & VI_AGE)
2723 strlcat(buf, "|VI_AGE", sizeof(buf));
2724 if (vp->v_iflag & VI_DOOMED)
2725 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2726 if (vp->v_iflag & VI_FREE)
2727 strlcat(buf, "|VI_FREE", sizeof(buf));
2728 if (vp->v_iflag & VI_DOINGINACT)
2729 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2730 if (vp->v_iflag & VI_OWEINACT)
2731 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2732 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2733 VI_DOINGINACT | VI_OWEINACT);
2735 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2736 strlcat(buf, buf2, sizeof(buf));
2738 printf(" flags (%s)\n", buf + 1);
2739 if (mtx_owned(VI_MTX(vp)))
2740 printf(" VI_LOCKed");
2741 if (vp->v_object != NULL)
2742 printf(" v_object %p ref %d pages %d\n",
2743 vp->v_object, vp->v_object->ref_count,
2744 vp->v_object->resident_page_count);
2746 lockmgr_printinfo(vp->v_vnlock);
2747 if (vp->v_data != NULL)
2753 * List all of the locked vnodes in the system.
2754 * Called when debugging the kernel.
2756 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2758 struct mount *mp, *nmp;
2762 * Note: because this is DDB, we can't obey the locking semantics
2763 * for these structures, which means we could catch an inconsistent
2764 * state and dereference a nasty pointer. Not much to be done
2767 db_printf("Locked vnodes\n");
2768 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2769 nmp = TAILQ_NEXT(mp, mnt_list);
2770 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2771 if (vp->v_type != VMARKER &&
2775 nmp = TAILQ_NEXT(mp, mnt_list);
2780 * Show details about the given vnode.
2782 DB_SHOW_COMMAND(vnode, db_show_vnode)
2788 vp = (struct vnode *)addr;
2789 vn_printf(vp, "vnode ");
2793 * Show details about the given mount point.
2795 DB_SHOW_COMMAND(mount, db_show_mount)
2805 /* No address given, print short info about all mount points. */
2806 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2807 db_printf("%p %s on %s (%s)\n", mp,
2808 mp->mnt_stat.f_mntfromname,
2809 mp->mnt_stat.f_mntonname,
2810 mp->mnt_stat.f_fstypename);
2814 db_printf("\nMore info: show mount <addr>\n");
2818 mp = (struct mount *)addr;
2819 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2820 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2823 flags = mp->mnt_flag;
2824 #define MNT_FLAG(flag) do { \
2825 if (flags & (flag)) { \
2826 if (buf[0] != '\0') \
2827 strlcat(buf, ", ", sizeof(buf)); \
2828 strlcat(buf, (#flag) + 4, sizeof(buf)); \
2832 MNT_FLAG(MNT_RDONLY);
2833 MNT_FLAG(MNT_SYNCHRONOUS);
2834 MNT_FLAG(MNT_NOEXEC);
2835 MNT_FLAG(MNT_NOSUID);
2836 MNT_FLAG(MNT_UNION);
2837 MNT_FLAG(MNT_ASYNC);
2838 MNT_FLAG(MNT_SUIDDIR);
2839 MNT_FLAG(MNT_SOFTDEP);
2840 MNT_FLAG(MNT_NOSYMFOLLOW);
2841 MNT_FLAG(MNT_GJOURNAL);
2842 MNT_FLAG(MNT_MULTILABEL);
2844 MNT_FLAG(MNT_NOATIME);
2845 MNT_FLAG(MNT_NOCLUSTERR);
2846 MNT_FLAG(MNT_NOCLUSTERW);
2847 MNT_FLAG(MNT_NFS4ACLS);
2848 MNT_FLAG(MNT_EXRDONLY);
2849 MNT_FLAG(MNT_EXPORTED);
2850 MNT_FLAG(MNT_DEFEXPORTED);
2851 MNT_FLAG(MNT_EXPORTANON);
2852 MNT_FLAG(MNT_EXKERB);
2853 MNT_FLAG(MNT_EXPUBLIC);
2854 MNT_FLAG(MNT_LOCAL);
2855 MNT_FLAG(MNT_QUOTA);
2856 MNT_FLAG(MNT_ROOTFS);
2858 MNT_FLAG(MNT_IGNORE);
2859 MNT_FLAG(MNT_UPDATE);
2860 MNT_FLAG(MNT_DELEXPORT);
2861 MNT_FLAG(MNT_RELOAD);
2862 MNT_FLAG(MNT_FORCE);
2863 MNT_FLAG(MNT_SNAPSHOT);
2864 MNT_FLAG(MNT_BYFSID);
2868 strlcat(buf, ", ", sizeof(buf));
2869 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2872 db_printf(" mnt_flag = %s\n", buf);
2875 flags = mp->mnt_kern_flag;
2876 #define MNT_KERN_FLAG(flag) do { \
2877 if (flags & (flag)) { \
2878 if (buf[0] != '\0') \
2879 strlcat(buf, ", ", sizeof(buf)); \
2880 strlcat(buf, (#flag) + 5, sizeof(buf)); \
2884 MNT_KERN_FLAG(MNTK_UNMOUNTF);
2885 MNT_KERN_FLAG(MNTK_ASYNC);
2886 MNT_KERN_FLAG(MNTK_SOFTDEP);
2887 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2888 MNT_KERN_FLAG(MNTK_UNMOUNT);
2889 MNT_KERN_FLAG(MNTK_MWAIT);
2890 MNT_KERN_FLAG(MNTK_SUSPEND);
2891 MNT_KERN_FLAG(MNTK_SUSPEND2);
2892 MNT_KERN_FLAG(MNTK_SUSPENDED);
2893 MNT_KERN_FLAG(MNTK_MPSAFE);
2894 MNT_KERN_FLAG(MNTK_NOKNOTE);
2895 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2896 #undef MNT_KERN_FLAG
2899 strlcat(buf, ", ", sizeof(buf));
2900 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2903 db_printf(" mnt_kern_flag = %s\n", buf);
2905 db_printf(" mnt_opt = ");
2906 opt = TAILQ_FIRST(mp->mnt_opt);
2908 db_printf("%s", opt->name);
2909 opt = TAILQ_NEXT(opt, link);
2910 while (opt != NULL) {
2911 db_printf(", %s", opt->name);
2912 opt = TAILQ_NEXT(opt, link);
2918 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
2919 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2920 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2921 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2922 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2923 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2924 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2925 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2926 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2927 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2928 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2929 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2931 db_printf(" mnt_cred = { uid=%u ruid=%u",
2932 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2933 if (jailed(mp->mnt_cred))
2934 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2936 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
2937 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
2938 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2939 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2940 db_printf(" mnt_noasync = %u\n", mp->mnt_noasync);
2941 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2942 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2943 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
2944 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2945 db_printf(" mnt_secondary_accwrites = %d\n",
2946 mp->mnt_secondary_accwrites);
2947 db_printf(" mnt_gjprovider = %s\n",
2948 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2951 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2952 if (vp->v_type != VMARKER) {
2953 vn_printf(vp, "vnode ");
2962 * Fill in a struct xvfsconf based on a struct vfsconf.
2965 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2968 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2969 xvfsp->vfc_typenum = vfsp->vfc_typenum;
2970 xvfsp->vfc_refcount = vfsp->vfc_refcount;
2971 xvfsp->vfc_flags = vfsp->vfc_flags;
2973 * These are unused in userland, we keep them
2974 * to not break binary compatibility.
2976 xvfsp->vfc_vfsops = NULL;
2977 xvfsp->vfc_next = NULL;
2981 * Top level filesystem related information gathering.
2984 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2986 struct vfsconf *vfsp;
2987 struct xvfsconf xvfsp;
2991 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2992 bzero(&xvfsp, sizeof(xvfsp));
2993 vfsconf2x(vfsp, &xvfsp);
2994 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
3001 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
3002 "S,xvfsconf", "List of all configured filesystems");
3004 #ifndef BURN_BRIDGES
3005 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3008 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3010 int *name = (int *)arg1 - 1; /* XXX */
3011 u_int namelen = arg2 + 1; /* XXX */
3012 struct vfsconf *vfsp;
3013 struct xvfsconf xvfsp;
3015 printf("WARNING: userland calling deprecated sysctl, "
3016 "please rebuild world\n");
3018 #if 1 || defined(COMPAT_PRELITE2)
3019 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3021 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3025 case VFS_MAXTYPENUM:
3028 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3031 return (ENOTDIR); /* overloaded */
3032 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3033 if (vfsp->vfc_typenum == name[2])
3036 return (EOPNOTSUPP);
3037 bzero(&xvfsp, sizeof(xvfsp));
3038 vfsconf2x(vfsp, &xvfsp);
3039 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3041 return (EOPNOTSUPP);
3044 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3045 vfs_sysctl, "Generic filesystem");
3047 #if 1 || defined(COMPAT_PRELITE2)
3050 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3053 struct vfsconf *vfsp;
3054 struct ovfsconf ovfs;
3056 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3057 bzero(&ovfs, sizeof(ovfs));
3058 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3059 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3060 ovfs.vfc_index = vfsp->vfc_typenum;
3061 ovfs.vfc_refcount = vfsp->vfc_refcount;
3062 ovfs.vfc_flags = vfsp->vfc_flags;
3063 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3070 #endif /* 1 || COMPAT_PRELITE2 */
3071 #endif /* !BURN_BRIDGES */
3073 #define KINFO_VNODESLOP 10
3076 * Dump vnode list (via sysctl).
3080 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3088 * Stale numvnodes access is not fatal here.
3091 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3093 /* Make an estimate */
3094 return (SYSCTL_OUT(req, 0, len));
3096 error = sysctl_wire_old_buffer(req, 0);
3099 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3101 mtx_lock(&mountlist_mtx);
3102 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3103 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3106 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3110 xvn[n].xv_size = sizeof *xvn;
3111 xvn[n].xv_vnode = vp;
3112 xvn[n].xv_id = 0; /* XXX compat */
3113 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3115 XV_COPY(writecount);
3121 xvn[n].xv_flag = vp->v_vflag;
3123 switch (vp->v_type) {
3130 if (vp->v_rdev == NULL) {
3134 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3137 xvn[n].xv_socket = vp->v_socket;
3140 xvn[n].xv_fifo = vp->v_fifoinfo;
3145 /* shouldn't happen? */
3153 mtx_lock(&mountlist_mtx);
3158 mtx_unlock(&mountlist_mtx);
3160 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3165 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3166 0, 0, sysctl_vnode, "S,xvnode", "");
3170 * Unmount all filesystems. The list is traversed in reverse order
3171 * of mounting to avoid dependencies.
3174 vfs_unmountall(void)
3180 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3181 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3185 * Since this only runs when rebooting, it is not interlocked.
3187 while(!TAILQ_EMPTY(&mountlist)) {
3188 mp = TAILQ_LAST(&mountlist, mntlist);
3189 error = dounmount(mp, MNT_FORCE, td);
3191 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3193 * XXX: Due to the way in which we mount the root
3194 * file system off of devfs, devfs will generate a
3195 * "busy" warning when we try to unmount it before
3196 * the root. Don't print a warning as a result in
3197 * order to avoid false positive errors that may
3198 * cause needless upset.
3200 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3201 printf("unmount of %s failed (",
3202 mp->mnt_stat.f_mntonname);
3206 printf("%d)\n", error);
3209 /* The unmount has removed mp from the mountlist */
3215 * perform msync on all vnodes under a mount point
3216 * the mount point must be locked.
3219 vfs_msync(struct mount *mp, int flags)
3221 struct vnode *vp, *mvp;
3222 struct vm_object *obj;
3224 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3226 MNT_VNODE_FOREACH(vp, mp, mvp) {
3229 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3230 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3233 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3235 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3243 VM_OBJECT_LOCK(obj);
3244 vm_object_page_clean(obj, 0, 0,
3246 OBJPC_SYNC : OBJPC_NOSYNC);
3247 VM_OBJECT_UNLOCK(obj);
3259 * Mark a vnode as free, putting it up for recycling.
3262 vfree(struct vnode *vp)
3265 ASSERT_VI_LOCKED(vp, "vfree");
3266 mtx_lock(&vnode_free_list_mtx);
3267 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3268 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3269 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3270 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3271 ("vfree: Freeing doomed vnode"));
3272 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3273 if (vp->v_iflag & VI_AGE) {
3274 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3276 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3279 vp->v_iflag &= ~VI_AGE;
3280 vp->v_iflag |= VI_FREE;
3281 mtx_unlock(&vnode_free_list_mtx);
3285 * Opposite of vfree() - mark a vnode as in use.
3288 vbusy(struct vnode *vp)
3290 ASSERT_VI_LOCKED(vp, "vbusy");
3291 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3292 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3293 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3295 mtx_lock(&vnode_free_list_mtx);
3296 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3298 vp->v_iflag &= ~(VI_FREE|VI_AGE);
3299 mtx_unlock(&vnode_free_list_mtx);
3303 destroy_vpollinfo(struct vpollinfo *vi)
3305 knlist_destroy(&vi->vpi_selinfo.si_note);
3306 mtx_destroy(&vi->vpi_lock);
3307 uma_zfree(vnodepoll_zone, vi);
3311 * Initalize per-vnode helper structure to hold poll-related state.
3314 v_addpollinfo(struct vnode *vp)
3316 struct vpollinfo *vi;
3318 if (vp->v_pollinfo != NULL)
3320 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3321 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3322 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3323 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3325 if (vp->v_pollinfo != NULL) {
3327 destroy_vpollinfo(vi);
3330 vp->v_pollinfo = vi;
3335 * Record a process's interest in events which might happen to
3336 * a vnode. Because poll uses the historic select-style interface
3337 * internally, this routine serves as both the ``check for any
3338 * pending events'' and the ``record my interest in future events''
3339 * functions. (These are done together, while the lock is held,
3340 * to avoid race conditions.)
3343 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3347 mtx_lock(&vp->v_pollinfo->vpi_lock);
3348 if (vp->v_pollinfo->vpi_revents & events) {
3350 * This leaves events we are not interested
3351 * in available for the other process which
3352 * which presumably had requested them
3353 * (otherwise they would never have been
3356 events &= vp->v_pollinfo->vpi_revents;
3357 vp->v_pollinfo->vpi_revents &= ~events;
3359 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3362 vp->v_pollinfo->vpi_events |= events;
3363 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3364 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3369 * Routine to create and manage a filesystem syncer vnode.
3371 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3372 static int sync_fsync(struct vop_fsync_args *);
3373 static int sync_inactive(struct vop_inactive_args *);
3374 static int sync_reclaim(struct vop_reclaim_args *);
3376 static struct vop_vector sync_vnodeops = {
3377 .vop_bypass = VOP_EOPNOTSUPP,
3378 .vop_close = sync_close, /* close */
3379 .vop_fsync = sync_fsync, /* fsync */
3380 .vop_inactive = sync_inactive, /* inactive */
3381 .vop_reclaim = sync_reclaim, /* reclaim */
3382 .vop_lock1 = vop_stdlock, /* lock */
3383 .vop_unlock = vop_stdunlock, /* unlock */
3384 .vop_islocked = vop_stdislocked, /* islocked */
3388 * Create a new filesystem syncer vnode for the specified mount point.
3391 vfs_allocate_syncvnode(struct mount *mp)
3395 static long start, incr, next;
3398 /* Allocate a new vnode */
3399 if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3400 mp->mnt_syncer = NULL;
3404 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3405 vp->v_vflag |= VV_FORCEINSMQ;
3406 error = insmntque(vp, mp);
3408 panic("vfs_allocate_syncvnode: insmntque failed");
3409 vp->v_vflag &= ~VV_FORCEINSMQ;
3412 * Place the vnode onto the syncer worklist. We attempt to
3413 * scatter them about on the list so that they will go off
3414 * at evenly distributed times even if all the filesystems
3415 * are mounted at once.
3418 if (next == 0 || next > syncer_maxdelay) {
3422 start = syncer_maxdelay / 2;
3423 incr = syncer_maxdelay;
3429 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3430 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3431 mtx_lock(&sync_mtx);
3433 mtx_unlock(&sync_mtx);
3435 mp->mnt_syncer = vp;
3440 * Do a lazy sync of the filesystem.
3443 sync_fsync(struct vop_fsync_args *ap)
3445 struct vnode *syncvp = ap->a_vp;
3446 struct mount *mp = syncvp->v_mount;
3451 * We only need to do something if this is a lazy evaluation.
3453 if (ap->a_waitfor != MNT_LAZY)
3457 * Move ourselves to the back of the sync list.
3459 bo = &syncvp->v_bufobj;
3461 vn_syncer_add_to_worklist(bo, syncdelay);
3465 * Walk the list of vnodes pushing all that are dirty and
3466 * not already on the sync list.
3468 mtx_lock(&mountlist_mtx);
3469 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3470 mtx_unlock(&mountlist_mtx);
3473 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3479 mp->mnt_kern_flag &= ~MNTK_ASYNC;
3481 vfs_msync(mp, MNT_NOWAIT);
3482 error = VFS_SYNC(mp, MNT_LAZY);
3485 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3486 mp->mnt_kern_flag |= MNTK_ASYNC;
3488 vn_finished_write(mp);
3494 * The syncer vnode is no referenced.
3497 sync_inactive(struct vop_inactive_args *ap)
3505 * The syncer vnode is no longer needed and is being decommissioned.
3507 * Modifications to the worklist must be protected by sync_mtx.
3510 sync_reclaim(struct vop_reclaim_args *ap)
3512 struct vnode *vp = ap->a_vp;
3517 vp->v_mount->mnt_syncer = NULL;
3518 if (bo->bo_flag & BO_ONWORKLST) {
3519 mtx_lock(&sync_mtx);
3520 LIST_REMOVE(bo, bo_synclist);
3521 syncer_worklist_len--;
3523 mtx_unlock(&sync_mtx);
3524 bo->bo_flag &= ~BO_ONWORKLST;
3532 * Check if vnode represents a disk device
3535 vn_isdisk(struct vnode *vp, int *errp)
3541 if (vp->v_type != VCHR)
3543 else if (vp->v_rdev == NULL)
3545 else if (vp->v_rdev->si_devsw == NULL)
3547 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3552 return (error == 0);
3556 * Common filesystem object access control check routine. Accepts a
3557 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3558 * and optional call-by-reference privused argument allowing vaccess()
3559 * to indicate to the caller whether privilege was used to satisfy the
3560 * request (obsoleted). Returns 0 on success, or an errno on failure.
3562 * The ifdef'd CAPABILITIES version is here for reference, but is not
3566 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3567 accmode_t accmode, struct ucred *cred, int *privused)
3569 accmode_t dac_granted;
3570 accmode_t priv_granted;
3572 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3573 ("invalid bit in accmode"));
3576 * Look for a normal, non-privileged way to access the file/directory
3577 * as requested. If it exists, go with that.
3580 if (privused != NULL)
3585 /* Check the owner. */
3586 if (cred->cr_uid == file_uid) {
3587 dac_granted |= VADMIN;
3588 if (file_mode & S_IXUSR)
3589 dac_granted |= VEXEC;
3590 if (file_mode & S_IRUSR)
3591 dac_granted |= VREAD;
3592 if (file_mode & S_IWUSR)
3593 dac_granted |= (VWRITE | VAPPEND);
3595 if ((accmode & dac_granted) == accmode)
3601 /* Otherwise, check the groups (first match) */
3602 if (groupmember(file_gid, cred)) {
3603 if (file_mode & S_IXGRP)
3604 dac_granted |= VEXEC;
3605 if (file_mode & S_IRGRP)
3606 dac_granted |= VREAD;
3607 if (file_mode & S_IWGRP)
3608 dac_granted |= (VWRITE | VAPPEND);
3610 if ((accmode & dac_granted) == accmode)
3616 /* Otherwise, check everyone else. */
3617 if (file_mode & S_IXOTH)
3618 dac_granted |= VEXEC;
3619 if (file_mode & S_IROTH)
3620 dac_granted |= VREAD;
3621 if (file_mode & S_IWOTH)
3622 dac_granted |= (VWRITE | VAPPEND);
3623 if ((accmode & dac_granted) == accmode)
3628 * Build a privilege mask to determine if the set of privileges
3629 * satisfies the requirements when combined with the granted mask
3630 * from above. For each privilege, if the privilege is required,
3631 * bitwise or the request type onto the priv_granted mask.
3637 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3638 * requests, instead of PRIV_VFS_EXEC.
3640 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3641 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3642 priv_granted |= VEXEC;
3644 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3645 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3646 priv_granted |= VEXEC;
3649 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3650 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3651 priv_granted |= VREAD;
3653 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3654 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3655 priv_granted |= (VWRITE | VAPPEND);
3657 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3658 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3659 priv_granted |= VADMIN;
3661 if ((accmode & (priv_granted | dac_granted)) == accmode) {
3662 /* XXX audit: privilege used */
3663 if (privused != NULL)
3668 return ((accmode & VADMIN) ? EPERM : EACCES);
3672 * Credential check based on process requesting service, and per-attribute
3676 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3677 struct thread *td, accmode_t accmode)
3681 * Kernel-invoked always succeeds.
3687 * Do not allow privileged processes in jail to directly manipulate
3688 * system attributes.
3690 switch (attrnamespace) {
3691 case EXTATTR_NAMESPACE_SYSTEM:
3692 /* Potentially should be: return (EPERM); */
3693 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3694 case EXTATTR_NAMESPACE_USER:
3695 return (VOP_ACCESS(vp, accmode, cred, td));
3701 #ifdef DEBUG_VFS_LOCKS
3703 * This only exists to supress warnings from unlocked specfs accesses. It is
3704 * no longer ok to have an unlocked VFS.
3706 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
3707 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
3709 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3710 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3711 "Drop into debugger on lock violation");
3713 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3714 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3715 0, "Check for interlock across VOPs");
3717 int vfs_badlock_print = 1; /* Print lock violations. */
3718 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3719 0, "Print lock violations");
3722 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3723 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3724 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3728 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3732 if (vfs_badlock_backtrace)
3735 if (vfs_badlock_print)
3736 printf("%s: %p %s\n", str, (void *)vp, msg);
3737 if (vfs_badlock_ddb)
3738 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3742 assert_vi_locked(struct vnode *vp, const char *str)
3745 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3746 vfs_badlock("interlock is not locked but should be", str, vp);
3750 assert_vi_unlocked(struct vnode *vp, const char *str)
3753 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3754 vfs_badlock("interlock is locked but should not be", str, vp);
3758 assert_vop_locked(struct vnode *vp, const char *str)
3761 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3762 vfs_badlock("is not locked but should be", str, vp);
3766 assert_vop_unlocked(struct vnode *vp, const char *str)
3769 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3770 vfs_badlock("is locked but should not be", str, vp);
3774 assert_vop_elocked(struct vnode *vp, const char *str)
3777 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3778 vfs_badlock("is not exclusive locked but should be", str, vp);
3783 assert_vop_elocked_other(struct vnode *vp, const char *str)
3786 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3787 vfs_badlock("is not exclusive locked by another thread",
3792 assert_vop_slocked(struct vnode *vp, const char *str)
3795 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3796 vfs_badlock("is not locked shared but should be", str, vp);
3799 #endif /* DEBUG_VFS_LOCKS */
3802 vop_rename_fail(struct vop_rename_args *ap)
3805 if (ap->a_tvp != NULL)
3807 if (ap->a_tdvp == ap->a_tvp)
3816 vop_rename_pre(void *ap)
3818 struct vop_rename_args *a = ap;
3820 #ifdef DEBUG_VFS_LOCKS
3822 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3823 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3824 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3825 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3827 /* Check the source (from). */
3828 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
3829 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
3830 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3831 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
3832 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3834 /* Check the target. */
3836 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3837 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3839 if (a->a_tdvp != a->a_fdvp)
3841 if (a->a_tvp != a->a_fvp)
3849 vop_strategy_pre(void *ap)
3851 #ifdef DEBUG_VFS_LOCKS
3852 struct vop_strategy_args *a;
3859 * Cluster ops lock their component buffers but not the IO container.
3861 if ((bp->b_flags & B_CLUSTER) != 0)
3864 if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
3865 if (vfs_badlock_print)
3867 "VOP_STRATEGY: bp is not locked but should be\n");
3868 if (vfs_badlock_ddb)
3869 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3875 vop_lookup_pre(void *ap)
3877 #ifdef DEBUG_VFS_LOCKS
3878 struct vop_lookup_args *a;
3883 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3884 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3889 vop_lookup_post(void *ap, int rc)
3891 #ifdef DEBUG_VFS_LOCKS
3892 struct vop_lookup_args *a;
3900 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3901 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3904 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3909 vop_lock_pre(void *ap)
3911 #ifdef DEBUG_VFS_LOCKS
3912 struct vop_lock1_args *a = ap;
3914 if ((a->a_flags & LK_INTERLOCK) == 0)
3915 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3917 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3922 vop_lock_post(void *ap, int rc)
3924 #ifdef DEBUG_VFS_LOCKS
3925 struct vop_lock1_args *a = ap;
3927 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3929 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3934 vop_unlock_pre(void *ap)
3936 #ifdef DEBUG_VFS_LOCKS
3937 struct vop_unlock_args *a = ap;
3939 if (a->a_flags & LK_INTERLOCK)
3940 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3941 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3946 vop_unlock_post(void *ap, int rc)
3948 #ifdef DEBUG_VFS_LOCKS
3949 struct vop_unlock_args *a = ap;
3951 if (a->a_flags & LK_INTERLOCK)
3952 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3957 vop_create_post(void *ap, int rc)
3959 struct vop_create_args *a = ap;
3962 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3966 vop_link_post(void *ap, int rc)
3968 struct vop_link_args *a = ap;
3971 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3972 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3977 vop_mkdir_post(void *ap, int rc)
3979 struct vop_mkdir_args *a = ap;
3982 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3986 vop_mknod_post(void *ap, int rc)
3988 struct vop_mknod_args *a = ap;
3991 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3995 vop_remove_post(void *ap, int rc)
3997 struct vop_remove_args *a = ap;
4000 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4001 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4006 vop_rename_post(void *ap, int rc)
4008 struct vop_rename_args *a = ap;
4011 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4012 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4013 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4015 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4017 if (a->a_tdvp != a->a_fdvp)
4019 if (a->a_tvp != a->a_fvp)
4027 vop_rmdir_post(void *ap, int rc)
4029 struct vop_rmdir_args *a = ap;
4032 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4033 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4038 vop_setattr_post(void *ap, int rc)
4040 struct vop_setattr_args *a = ap;
4043 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4047 vop_symlink_post(void *ap, int rc)
4049 struct vop_symlink_args *a = ap;
4052 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4055 static struct knlist fs_knlist;
4058 vfs_event_init(void *arg)
4060 knlist_init_mtx(&fs_knlist, NULL);
4062 /* XXX - correct order? */
4063 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4066 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
4069 KNOTE_UNLOCKED(&fs_knlist, event);
4072 static int filt_fsattach(struct knote *kn);
4073 static void filt_fsdetach(struct knote *kn);
4074 static int filt_fsevent(struct knote *kn, long hint);
4076 struct filterops fs_filtops =
4077 { 0, filt_fsattach, filt_fsdetach, filt_fsevent };
4080 filt_fsattach(struct knote *kn)
4083 kn->kn_flags |= EV_CLEAR;
4084 knlist_add(&fs_knlist, kn, 0);
4089 filt_fsdetach(struct knote *kn)
4092 knlist_remove(&fs_knlist, kn, 0);
4096 filt_fsevent(struct knote *kn, long hint)
4099 kn->kn_fflags |= hint;
4100 return (kn->kn_fflags != 0);
4104 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4110 error = SYSCTL_IN(req, &vc, sizeof(vc));
4113 if (vc.vc_vers != VFS_CTL_VERS1)
4115 mp = vfs_getvfs(&vc.vc_fsid);
4118 /* ensure that a specific sysctl goes to the right filesystem. */
4119 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4120 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4124 VCTLTOREQ(&vc, req);
4125 error = VFS_SYSCTL(mp, vc.vc_op, req);
4130 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
4134 * Function to initialize a va_filerev field sensibly.
4135 * XXX: Wouldn't a random number make a lot more sense ??
4138 init_va_filerev(void)
4143 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4146 static int filt_vfsread(struct knote *kn, long hint);
4147 static int filt_vfswrite(struct knote *kn, long hint);
4148 static int filt_vfsvnode(struct knote *kn, long hint);
4149 static void filt_vfsdetach(struct knote *kn);
4150 static struct filterops vfsread_filtops =
4151 { 1, NULL, filt_vfsdetach, filt_vfsread };
4152 static struct filterops vfswrite_filtops =
4153 { 1, NULL, filt_vfsdetach, filt_vfswrite };
4154 static struct filterops vfsvnode_filtops =
4155 { 1, NULL, filt_vfsdetach, filt_vfsvnode };
4158 vfs_knllock(void *arg)
4160 struct vnode *vp = arg;
4162 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4166 vfs_knlunlock(void *arg)
4168 struct vnode *vp = arg;
4174 vfs_knl_assert_locked(void *arg)
4176 #ifdef DEBUG_VFS_LOCKS
4177 struct vnode *vp = arg;
4179 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4184 vfs_knl_assert_unlocked(void *arg)
4186 #ifdef DEBUG_VFS_LOCKS
4187 struct vnode *vp = arg;
4189 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4194 vfs_kqfilter(struct vop_kqfilter_args *ap)
4196 struct vnode *vp = ap->a_vp;
4197 struct knote *kn = ap->a_kn;
4200 switch (kn->kn_filter) {
4202 kn->kn_fop = &vfsread_filtops;
4205 kn->kn_fop = &vfswrite_filtops;
4208 kn->kn_fop = &vfsvnode_filtops;
4214 kn->kn_hook = (caddr_t)vp;
4217 if (vp->v_pollinfo == NULL)
4219 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4220 knlist_add(knl, kn, 0);
4226 * Detach knote from vnode
4229 filt_vfsdetach(struct knote *kn)
4231 struct vnode *vp = (struct vnode *)kn->kn_hook;
4233 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4234 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4239 filt_vfsread(struct knote *kn, long hint)
4241 struct vnode *vp = (struct vnode *)kn->kn_hook;
4246 * filesystem is gone, so set the EOF flag and schedule
4247 * the knote for deletion.
4249 if (hint == NOTE_REVOKE) {
4251 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4256 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4260 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4261 res = (kn->kn_data != 0);
4268 filt_vfswrite(struct knote *kn, long hint)
4270 struct vnode *vp = (struct vnode *)kn->kn_hook;
4275 * filesystem is gone, so set the EOF flag and schedule
4276 * the knote for deletion.
4278 if (hint == NOTE_REVOKE)
4279 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4287 filt_vfsvnode(struct knote *kn, long hint)
4289 struct vnode *vp = (struct vnode *)kn->kn_hook;
4293 if (kn->kn_sfflags & hint)
4294 kn->kn_fflags |= hint;
4295 if (hint == NOTE_REVOKE) {
4296 kn->kn_flags |= EV_EOF;
4300 res = (kn->kn_fflags != 0);
4306 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4310 if (dp->d_reclen > ap->a_uio->uio_resid)
4311 return (ENAMETOOLONG);
4312 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4314 if (ap->a_ncookies != NULL) {
4315 if (ap->a_cookies != NULL)
4316 free(ap->a_cookies, M_TEMP);
4317 ap->a_cookies = NULL;
4318 *ap->a_ncookies = 0;
4322 if (ap->a_ncookies == NULL)
4325 KASSERT(ap->a_cookies,
4326 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4328 *ap->a_cookies = realloc(*ap->a_cookies,
4329 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4330 (*ap->a_cookies)[*ap->a_ncookies] = off;
4335 * Mark for update the access time of the file if the filesystem
4336 * supports VOP_MARKATIME. This functionality is used by execve and
4337 * mmap, so we want to avoid the I/O implied by directly setting
4338 * va_atime for the sake of efficiency.
4341 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4346 VFS_ASSERT_GIANT(mp);
4347 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4348 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4349 (void)VOP_MARKATIME(vp);
4353 * The purpose of this routine is to remove granularity from accmode_t,
4354 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4355 * VADMIN and VAPPEND.
4357 * If it returns 0, the caller is supposed to continue with the usual
4358 * access checks using 'accmode' as modified by this routine. If it
4359 * returns nonzero value, the caller is supposed to return that value
4362 * Note that after this routine runs, accmode may be zero.
4365 vfs_unixify_accmode(accmode_t *accmode)
4368 * There is no way to specify explicit "deny" rule using
4369 * file mode or POSIX.1e ACLs.
4371 if (*accmode & VEXPLICIT_DENY) {
4377 * None of these can be translated into usual access bits.
4378 * Also, the common case for NFSv4 ACLs is to not contain
4379 * either of these bits. Caller should check for VWRITE
4380 * on the containing directory instead.
4382 if (*accmode & (VDELETE_CHILD | VDELETE))
4385 if (*accmode & VADMIN_PERMS) {
4386 *accmode &= ~VADMIN_PERMS;
4391 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4392 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4394 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);