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("$FreeBSD$");
44 #include "opt_compat.h"
46 #include "opt_watchdog.h"
48 #include <sys/param.h>
49 #include <sys/systm.h>
52 #include <sys/condvar.h>
54 #include <sys/dirent.h>
55 #include <sys/event.h>
56 #include <sys/eventhandler.h>
57 #include <sys/extattr.h>
59 #include <sys/fcntl.h>
62 #include <sys/kernel.h>
63 #include <sys/kthread.h>
64 #include <sys/lockf.h>
65 #include <sys/malloc.h>
66 #include <sys/mount.h>
67 #include <sys/namei.h>
68 #include <sys/pctrie.h>
70 #include <sys/reboot.h>
71 #include <sys/rwlock.h>
72 #include <sys/sched.h>
73 #include <sys/sleepqueue.h>
76 #include <sys/sysctl.h>
77 #include <sys/syslog.h>
78 #include <sys/vmmeter.h>
79 #include <sys/vnode.h>
80 #include <sys/watchdog.h>
82 #include <machine/stdarg.h>
84 #include <security/mac/mac_framework.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_extern.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_page.h>
92 #include <vm/vm_kern.h>
99 static void delmntque(struct vnode *vp);
100 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
101 int slpflag, int slptimeo);
102 static void syncer_shutdown(void *arg, int howto);
103 static int vtryrecycle(struct vnode *vp);
104 static void v_incr_usecount(struct vnode *);
105 static void v_decr_usecount(struct vnode *);
106 static void v_decr_useonly(struct vnode *);
107 static void v_upgrade_usecount(struct vnode *);
108 static void 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 in vdropl() for VI_DOOMED vnode.
120 static unsigned long numvnodes;
122 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
123 "Number of vnodes in existence");
126 * Conversion tables for conversion from vnode types to inode formats
129 enum vtype iftovt_tab[16] = {
130 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
131 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
133 int vttoif_tab[10] = {
134 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
135 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
139 * List of vnodes that are ready for recycling.
141 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
144 * Free vnode target. Free vnodes may simply be files which have been stat'd
145 * but not read. This is somewhat common, and a small cache of such files
146 * should be kept to avoid recreation costs.
148 static u_long wantfreevnodes;
149 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
150 /* Number of vnodes in the free list. */
151 static u_long freevnodes;
152 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
153 "Number of vnodes in the free list");
155 static int vlru_allow_cache_src;
156 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
157 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
160 * Various variables used for debugging the new implementation of
162 * XXX these are probably of (very) limited utility now.
164 static int reassignbufcalls;
165 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
166 "Number of calls to reassignbuf");
169 * Cache for the mount type id assigned to NFS. This is used for
170 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
172 int nfs_mount_type = -1;
174 /* To keep more than one thread at a time from running vfs_getnewfsid */
175 static struct mtx mntid_mtx;
178 * Lock for any access to the following:
183 static struct mtx vnode_free_list_mtx;
185 /* Publicly exported FS */
186 struct nfs_public nfs_pub;
188 static uma_zone_t buf_trie_zone;
190 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
191 static uma_zone_t vnode_zone;
192 static uma_zone_t vnodepoll_zone;
195 * The workitem queue.
197 * It is useful to delay writes of file data and filesystem metadata
198 * for tens of seconds so that quickly created and deleted files need
199 * not waste disk bandwidth being created and removed. To realize this,
200 * we append vnodes to a "workitem" queue. When running with a soft
201 * updates implementation, most pending metadata dependencies should
202 * not wait for more than a few seconds. Thus, mounted on block devices
203 * are delayed only about a half the time that file data is delayed.
204 * Similarly, directory updates are more critical, so are only delayed
205 * about a third the time that file data is delayed. Thus, there are
206 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
207 * one each second (driven off the filesystem syncer process). The
208 * syncer_delayno variable indicates the next queue that is to be processed.
209 * Items that need to be processed soon are placed in this queue:
211 * syncer_workitem_pending[syncer_delayno]
213 * A delay of fifteen seconds is done by placing the request fifteen
214 * entries later in the queue:
216 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
219 static int syncer_delayno;
220 static long syncer_mask;
221 LIST_HEAD(synclist, bufobj);
222 static struct synclist *syncer_workitem_pending;
224 * The sync_mtx protects:
229 * syncer_workitem_pending
230 * syncer_worklist_len
233 static struct mtx sync_mtx;
234 static struct cv sync_wakeup;
236 #define SYNCER_MAXDELAY 32
237 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
238 static int syncdelay = 30; /* max time to delay syncing data */
239 static int filedelay = 30; /* time to delay syncing files */
240 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
241 "Time to delay syncing files (in seconds)");
242 static int dirdelay = 29; /* time to delay syncing directories */
243 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
244 "Time to delay syncing directories (in seconds)");
245 static int metadelay = 28; /* time to delay syncing metadata */
246 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
247 "Time to delay syncing metadata (in seconds)");
248 static int rushjob; /* number of slots to run ASAP */
249 static int stat_rush_requests; /* number of times I/O speeded up */
250 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
251 "Number of times I/O speeded up (rush requests)");
254 * When shutting down the syncer, run it at four times normal speed.
256 #define SYNCER_SHUTDOWN_SPEEDUP 4
257 static int sync_vnode_count;
258 static int syncer_worklist_len;
259 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
263 * Number of vnodes we want to exist at any one time. This is mostly used
264 * to size hash tables in vnode-related code. It is normally not used in
265 * getnewvnode(), as wantfreevnodes is normally nonzero.)
267 * XXX desiredvnodes is historical cruft and should not exist.
270 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
271 &desiredvnodes, 0, "Maximum number of vnodes");
272 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
273 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
274 static int vnlru_nowhere;
275 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
276 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
279 * Macros to control when a vnode is freed and recycled. All require
280 * the vnode interlock.
282 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
283 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
284 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
286 /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
290 * Support for the bufobj clean & dirty pctrie.
293 buf_trie_alloc(struct pctrie *ptree)
296 return uma_zalloc(buf_trie_zone, M_NOWAIT);
300 buf_trie_free(struct pctrie *ptree, void *node)
303 uma_zfree(buf_trie_zone, node);
305 PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
308 * Initialize the vnode management data structures.
310 * Reevaluate the following cap on the number of vnodes after the physical
311 * memory size exceeds 512GB. In the limit, as the physical memory size
312 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
314 #ifndef MAXVNODES_MAX
315 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
318 vntblinit(void *dummy __unused)
321 int physvnodes, virtvnodes;
324 * Desiredvnodes is a function of the physical memory size and the
325 * kernel's heap size. Generally speaking, it scales with the
326 * physical memory size. The ratio of desiredvnodes to physical pages
327 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
328 * marginal ratio of desiredvnodes to physical pages is one to
329 * sixteen. However, desiredvnodes is limited by the kernel's heap
330 * size. The memory required by desiredvnodes vnodes and vm objects
331 * may not exceed one seventh of the kernel's heap size.
333 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
334 cnt.v_page_count) / 16;
335 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
336 sizeof(struct vnode)));
337 desiredvnodes = min(physvnodes, virtvnodes);
338 if (desiredvnodes > MAXVNODES_MAX) {
340 printf("Reducing kern.maxvnodes %d -> %d\n",
341 desiredvnodes, MAXVNODES_MAX);
342 desiredvnodes = MAXVNODES_MAX;
344 wantfreevnodes = desiredvnodes / 4;
345 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
346 TAILQ_INIT(&vnode_free_list);
347 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
348 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
349 NULL, NULL, UMA_ALIGN_PTR, 0);
350 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
351 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
353 * Preallocate enough nodes to support one-per buf so that
354 * we can not fail an insert. reassignbuf() callers can not
355 * tolerate the insertion failure.
357 buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
358 NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR,
359 UMA_ZONE_NOFREE | UMA_ZONE_VM);
360 uma_prealloc(buf_trie_zone, nbuf);
362 * Initialize the filesystem syncer.
364 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
366 syncer_maxdelay = syncer_mask + 1;
367 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
368 cv_init(&sync_wakeup, "syncer");
369 for (i = 1; i <= sizeof(struct vnode); i <<= 1)
373 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
377 * Mark a mount point as busy. Used to synchronize access and to delay
378 * unmounting. Eventually, mountlist_mtx is not released on failure.
380 * vfs_busy() is a custom lock, it can block the caller.
381 * vfs_busy() only sleeps if the unmount is active on the mount point.
382 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
383 * vnode belonging to mp.
385 * Lookup uses vfs_busy() to traverse mount points.
387 * / vnode lock A / vnode lock (/var) D
388 * /var vnode lock B /log vnode lock(/var/log) E
389 * vfs_busy lock C vfs_busy lock F
391 * Within each file system, the lock order is C->A->B and F->D->E.
393 * When traversing across mounts, the system follows that lock order:
399 * The lookup() process for namei("/var") illustrates the process:
400 * VOP_LOOKUP() obtains B while A is held
401 * vfs_busy() obtains a shared lock on F while A and B are held
402 * vput() releases lock on B
403 * vput() releases lock on A
404 * VFS_ROOT() obtains lock on D while shared lock on F is held
405 * vfs_unbusy() releases shared lock on F
406 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
407 * Attempt to lock A (instead of vp_crossmp) while D is held would
408 * violate the global order, causing deadlocks.
410 * dounmount() locks B while F is drained.
413 vfs_busy(struct mount *mp, int flags)
416 MPASS((flags & ~MBF_MASK) == 0);
417 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
422 * If mount point is currenly being unmounted, sleep until the
423 * mount point fate is decided. If thread doing the unmounting fails,
424 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
425 * that this mount point has survived the unmount attempt and vfs_busy
426 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
427 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
428 * about to be really destroyed. vfs_busy needs to release its
429 * reference on the mount point in this case and return with ENOENT,
430 * telling the caller that mount mount it tried to busy is no longer
433 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
434 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
437 CTR1(KTR_VFS, "%s: failed busying before sleeping",
441 if (flags & MBF_MNTLSTLOCK)
442 mtx_unlock(&mountlist_mtx);
443 mp->mnt_kern_flag |= MNTK_MWAIT;
444 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
445 if (flags & MBF_MNTLSTLOCK)
446 mtx_lock(&mountlist_mtx);
449 if (flags & MBF_MNTLSTLOCK)
450 mtx_unlock(&mountlist_mtx);
457 * Free a busy filesystem.
460 vfs_unbusy(struct mount *mp)
463 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
466 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
468 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
469 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
470 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
471 mp->mnt_kern_flag &= ~MNTK_DRAINING;
472 wakeup(&mp->mnt_lockref);
478 * Lookup a mount point by filesystem identifier.
481 vfs_getvfs(fsid_t *fsid)
485 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
486 mtx_lock(&mountlist_mtx);
487 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
488 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
489 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
491 mtx_unlock(&mountlist_mtx);
495 mtx_unlock(&mountlist_mtx);
496 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
497 return ((struct mount *) 0);
501 * Lookup a mount point by filesystem identifier, busying it before
504 * To avoid congestion on mountlist_mtx, implement simple direct-mapped
505 * cache for popular filesystem identifiers. The cache is lockess, using
506 * the fact that struct mount's are never freed. In worst case we may
507 * get pointer to unmounted or even different filesystem, so we have to
508 * check what we got, and go slow way if so.
511 vfs_busyfs(fsid_t *fsid)
513 #define FSID_CACHE_SIZE 256
514 typedef struct mount * volatile vmp_t;
515 static vmp_t cache[FSID_CACHE_SIZE];
520 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
521 hash = fsid->val[0] ^ fsid->val[1];
522 hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
525 mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
526 mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
528 if (vfs_busy(mp, 0) != 0) {
532 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
533 mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
539 mtx_lock(&mountlist_mtx);
540 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
541 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
542 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
543 error = vfs_busy(mp, MBF_MNTLSTLOCK);
546 mtx_unlock(&mountlist_mtx);
553 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
554 mtx_unlock(&mountlist_mtx);
555 return ((struct mount *) 0);
559 * Check if a user can access privileged mount options.
562 vfs_suser(struct mount *mp, struct thread *td)
567 * If the thread is jailed, but this is not a jail-friendly file
568 * system, deny immediately.
570 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
574 * If the file system was mounted outside the jail of the calling
575 * thread, deny immediately.
577 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
581 * If file system supports delegated administration, we don't check
582 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
583 * by the file system itself.
584 * If this is not the user that did original mount, we check for
585 * the PRIV_VFS_MOUNT_OWNER privilege.
587 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
588 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
589 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
596 * Get a new unique fsid. Try to make its val[0] unique, since this value
597 * will be used to create fake device numbers for stat(). Also try (but
598 * not so hard) make its val[0] unique mod 2^16, since some emulators only
599 * support 16-bit device numbers. We end up with unique val[0]'s for the
600 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
602 * Keep in mind that several mounts may be running in parallel. Starting
603 * the search one past where the previous search terminated is both a
604 * micro-optimization and a defense against returning the same fsid to
608 vfs_getnewfsid(struct mount *mp)
610 static uint16_t mntid_base;
615 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
616 mtx_lock(&mntid_mtx);
617 mtype = mp->mnt_vfc->vfc_typenum;
618 tfsid.val[1] = mtype;
619 mtype = (mtype & 0xFF) << 24;
621 tfsid.val[0] = makedev(255,
622 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
624 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
628 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
629 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
630 mtx_unlock(&mntid_mtx);
634 * Knob to control the precision of file timestamps:
636 * 0 = seconds only; nanoseconds zeroed.
637 * 1 = seconds and nanoseconds, accurate within 1/HZ.
638 * 2 = seconds and nanoseconds, truncated to microseconds.
639 * >=3 = seconds and nanoseconds, maximum precision.
641 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
643 static int timestamp_precision = TSP_SEC;
644 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
645 ×tamp_precision, 0, "File timestamp precision (0: seconds, "
646 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
647 "3+: sec + ns (max. precision))");
650 * Get a current timestamp.
653 vfs_timestamp(struct timespec *tsp)
657 switch (timestamp_precision) {
659 tsp->tv_sec = time_second;
667 TIMEVAL_TO_TIMESPEC(&tv, tsp);
677 * Set vnode attributes to VNOVAL
680 vattr_null(struct vattr *vap)
684 vap->va_size = VNOVAL;
685 vap->va_bytes = VNOVAL;
686 vap->va_mode = VNOVAL;
687 vap->va_nlink = VNOVAL;
688 vap->va_uid = VNOVAL;
689 vap->va_gid = VNOVAL;
690 vap->va_fsid = VNOVAL;
691 vap->va_fileid = VNOVAL;
692 vap->va_blocksize = VNOVAL;
693 vap->va_rdev = VNOVAL;
694 vap->va_atime.tv_sec = VNOVAL;
695 vap->va_atime.tv_nsec = VNOVAL;
696 vap->va_mtime.tv_sec = VNOVAL;
697 vap->va_mtime.tv_nsec = VNOVAL;
698 vap->va_ctime.tv_sec = VNOVAL;
699 vap->va_ctime.tv_nsec = VNOVAL;
700 vap->va_birthtime.tv_sec = VNOVAL;
701 vap->va_birthtime.tv_nsec = VNOVAL;
702 vap->va_flags = VNOVAL;
703 vap->va_gen = VNOVAL;
708 * This routine is called when we have too many vnodes. It attempts
709 * to free <count> vnodes and will potentially free vnodes that still
710 * have VM backing store (VM backing store is typically the cause
711 * of a vnode blowout so we want to do this). Therefore, this operation
712 * is not considered cheap.
714 * A number of conditions may prevent a vnode from being reclaimed.
715 * the buffer cache may have references on the vnode, a directory
716 * vnode may still have references due to the namei cache representing
717 * underlying files, or the vnode may be in active use. It is not
718 * desireable to reuse such vnodes. These conditions may cause the
719 * number of vnodes to reach some minimum value regardless of what
720 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
723 vlrureclaim(struct mount *mp)
732 * Calculate the trigger point, don't allow user
733 * screwups to blow us up. This prevents us from
734 * recycling vnodes with lots of resident pages. We
735 * aren't trying to free memory, we are trying to
738 usevnodes = desiredvnodes;
741 trigger = cnt.v_page_count * 2 / usevnodes;
743 vn_start_write(NULL, &mp, V_WAIT);
745 count = mp->mnt_nvnodelistsize / 10 + 1;
747 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
748 while (vp != NULL && vp->v_type == VMARKER)
749 vp = TAILQ_NEXT(vp, v_nmntvnodes);
752 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
753 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
758 * If it's been deconstructed already, it's still
759 * referenced, or it exceeds the trigger, skip it.
761 if (vp->v_usecount ||
762 (!vlru_allow_cache_src &&
763 !LIST_EMPTY(&(vp)->v_cache_src)) ||
764 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
765 vp->v_object->resident_page_count > trigger)) {
771 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
773 goto next_iter_mntunlocked;
777 * v_usecount may have been bumped after VOP_LOCK() dropped
778 * the vnode interlock and before it was locked again.
780 * It is not necessary to recheck VI_DOOMED because it can
781 * only be set by another thread that holds both the vnode
782 * lock and vnode interlock. If another thread has the
783 * vnode lock before we get to VOP_LOCK() and obtains the
784 * vnode interlock after VOP_LOCK() drops the vnode
785 * interlock, the other thread will be unable to drop the
786 * vnode lock before our VOP_LOCK() call fails.
788 if (vp->v_usecount ||
789 (!vlru_allow_cache_src &&
790 !LIST_EMPTY(&(vp)->v_cache_src)) ||
791 (vp->v_object != NULL &&
792 vp->v_object->resident_page_count > trigger)) {
793 VOP_UNLOCK(vp, LK_INTERLOCK);
795 goto next_iter_mntunlocked;
797 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
798 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
803 next_iter_mntunlocked:
812 kern_yield(PRI_USER);
817 vn_finished_write(mp);
822 * Attempt to keep the free list at wantfreevnodes length.
825 vnlru_free(int count)
829 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
830 for (; count > 0; count--) {
831 vp = TAILQ_FIRST(&vnode_free_list);
833 * The list can be modified while the free_list_mtx
834 * has been dropped and vp could be NULL here.
838 VNASSERT(vp->v_op != NULL, vp,
839 ("vnlru_free: vnode already reclaimed."));
840 KASSERT((vp->v_iflag & VI_FREE) != 0,
841 ("Removing vnode not on freelist"));
842 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
843 ("Mangling active vnode"));
844 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
846 * Don't recycle if we can't get the interlock.
848 if (!VI_TRYLOCK(vp)) {
849 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
852 VNASSERT(VCANRECYCLE(vp), vp,
853 ("vp inconsistent on freelist"));
855 vp->v_iflag &= ~VI_FREE;
857 mtx_unlock(&vnode_free_list_mtx);
861 * If the recycled succeeded this vdrop will actually free
862 * the vnode. If not it will simply place it back on
866 mtx_lock(&vnode_free_list_mtx);
870 * Attempt to recycle vnodes in a context that is always safe to block.
871 * Calling vlrurecycle() from the bowels of filesystem code has some
872 * interesting deadlock problems.
874 static struct proc *vnlruproc;
875 static int vnlruproc_sig;
880 struct mount *mp, *nmp;
882 struct proc *p = vnlruproc;
884 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
888 kproc_suspend_check(p);
889 mtx_lock(&vnode_free_list_mtx);
890 if (freevnodes > wantfreevnodes)
891 vnlru_free(freevnodes - wantfreevnodes);
892 if (numvnodes <= desiredvnodes * 9 / 10) {
894 wakeup(&vnlruproc_sig);
895 msleep(vnlruproc, &vnode_free_list_mtx,
896 PVFS|PDROP, "vlruwt", hz);
899 mtx_unlock(&vnode_free_list_mtx);
901 mtx_lock(&mountlist_mtx);
902 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
903 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
904 nmp = TAILQ_NEXT(mp, mnt_list);
907 done += vlrureclaim(mp);
908 mtx_lock(&mountlist_mtx);
909 nmp = TAILQ_NEXT(mp, mnt_list);
912 mtx_unlock(&mountlist_mtx);
915 /* These messages are temporary debugging aids */
916 if (vnlru_nowhere < 5)
917 printf("vnlru process getting nowhere..\n");
918 else if (vnlru_nowhere == 5)
919 printf("vnlru process messages stopped.\n");
922 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
924 kern_yield(PRI_USER);
928 static struct kproc_desc vnlru_kp = {
933 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
937 * Routines having to do with the management of the vnode table.
941 * Try to recycle a freed vnode. We abort if anyone picks up a reference
942 * before we actually vgone(). This function must be called with the vnode
943 * held to prevent the vnode from being returned to the free list midway
947 vtryrecycle(struct vnode *vp)
951 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
952 VNASSERT(vp->v_holdcnt, vp,
953 ("vtryrecycle: Recycling vp %p without a reference.", vp));
955 * This vnode may found and locked via some other list, if so we
956 * can't recycle it yet.
958 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
960 "%s: impossible to recycle, vp %p lock is already held",
962 return (EWOULDBLOCK);
965 * Don't recycle if its filesystem is being suspended.
967 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
970 "%s: impossible to recycle, cannot start the write for %p",
975 * If we got this far, we need to acquire the interlock and see if
976 * anyone picked up this vnode from another list. If not, we will
977 * mark it with DOOMED via vgonel() so that anyone who does find it
981 if (vp->v_usecount) {
982 VOP_UNLOCK(vp, LK_INTERLOCK);
983 vn_finished_write(vnmp);
985 "%s: impossible to recycle, %p is already referenced",
989 if ((vp->v_iflag & VI_DOOMED) == 0)
991 VOP_UNLOCK(vp, LK_INTERLOCK);
992 vn_finished_write(vnmp);
997 * Wait for available vnodes.
1000 getnewvnode_wait(int suspended)
1003 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
1004 if (numvnodes > desiredvnodes) {
1007 * File system is beeing suspended, we cannot risk a
1008 * deadlock here, so allocate new vnode anyway.
1010 if (freevnodes > wantfreevnodes)
1011 vnlru_free(freevnodes - wantfreevnodes);
1014 if (vnlruproc_sig == 0) {
1015 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
1018 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
1021 return (numvnodes > desiredvnodes ? ENFILE : 0);
1025 getnewvnode_reserve(u_int count)
1030 /* First try to be quick and racy. */
1031 if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
1032 td->td_vp_reserv += count;
1035 atomic_subtract_long(&numvnodes, count);
1037 mtx_lock(&vnode_free_list_mtx);
1039 if (getnewvnode_wait(0) == 0) {
1042 atomic_add_long(&numvnodes, 1);
1045 mtx_unlock(&vnode_free_list_mtx);
1049 getnewvnode_drop_reserve(void)
1054 atomic_subtract_long(&numvnodes, td->td_vp_reserv);
1055 td->td_vp_reserv = 0;
1059 * Return the next vnode from the free list.
1062 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
1070 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
1073 if (td->td_vp_reserv > 0) {
1074 td->td_vp_reserv -= 1;
1077 mtx_lock(&vnode_free_list_mtx);
1079 * Lend our context to reclaim vnodes if they've exceeded the max.
1081 if (freevnodes > wantfreevnodes)
1083 error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1085 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1087 mtx_unlock(&vnode_free_list_mtx);
1091 atomic_add_long(&numvnodes, 1);
1092 mtx_unlock(&vnode_free_list_mtx);
1094 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1098 vp->v_vnlock = &vp->v_lock;
1099 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1101 * By default, don't allow shared locks unless filesystems
1104 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE);
1106 * Initialize bufobj.
1109 bo->__bo_vnode = vp;
1110 rw_init(BO_LOCKPTR(bo), "bufobj interlock");
1111 bo->bo_ops = &buf_ops_bio;
1112 bo->bo_private = vp;
1113 TAILQ_INIT(&bo->bo_clean.bv_hd);
1114 TAILQ_INIT(&bo->bo_dirty.bv_hd);
1116 * Initialize namecache.
1118 LIST_INIT(&vp->v_cache_src);
1119 TAILQ_INIT(&vp->v_cache_dst);
1121 * Finalize various vnode identity bits.
1126 v_incr_usecount(vp);
1130 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1131 mac_vnode_associate_singlelabel(mp, vp);
1132 else if (mp == NULL && vops != &dead_vnodeops)
1133 printf("NULL mp in getnewvnode()\n");
1136 bo->bo_bsize = mp->mnt_stat.f_iosize;
1137 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1138 vp->v_vflag |= VV_NOKNOTE;
1140 rangelock_init(&vp->v_rl);
1143 * For the filesystems which do not use vfs_hash_insert(),
1144 * still initialize v_hash to have vfs_hash_index() useful.
1145 * E.g., nullfs uses vfs_hash_index() on the lower vnode for
1148 vp->v_hash = (uintptr_t)vp >> vnsz2log;
1155 * Delete from old mount point vnode list, if on one.
1158 delmntque(struct vnode *vp)
1168 KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1169 ("Active vnode list size %d > Vnode list size %d",
1170 mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1171 active = vp->v_iflag & VI_ACTIVE;
1172 vp->v_iflag &= ~VI_ACTIVE;
1174 mtx_lock(&vnode_free_list_mtx);
1175 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1176 mp->mnt_activevnodelistsize--;
1177 mtx_unlock(&vnode_free_list_mtx);
1181 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1182 ("bad mount point vnode list size"));
1183 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1184 mp->mnt_nvnodelistsize--;
1190 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1194 vp->v_op = &dead_vnodeops;
1200 * Insert into list of vnodes for the new mount point, if available.
1203 insmntque1(struct vnode *vp, struct mount *mp,
1204 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1207 KASSERT(vp->v_mount == NULL,
1208 ("insmntque: vnode already on per mount vnode list"));
1209 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1210 ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
1213 * We acquire the vnode interlock early to ensure that the
1214 * vnode cannot be recycled by another process releasing a
1215 * holdcnt on it before we get it on both the vnode list
1216 * and the active vnode list. The mount mutex protects only
1217 * manipulation of the vnode list and the vnode freelist
1218 * mutex protects only manipulation of the active vnode list.
1219 * Hence the need to hold the vnode interlock throughout.
1223 if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1224 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1225 mp->mnt_nvnodelistsize == 0)) &&
1226 (vp->v_vflag & VV_FORCEINSMQ) == 0) {
1235 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1236 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1237 ("neg mount point vnode list size"));
1238 mp->mnt_nvnodelistsize++;
1239 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1240 ("Activating already active vnode"));
1241 vp->v_iflag |= VI_ACTIVE;
1242 mtx_lock(&vnode_free_list_mtx);
1243 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1244 mp->mnt_activevnodelistsize++;
1245 mtx_unlock(&vnode_free_list_mtx);
1252 insmntque(struct vnode *vp, struct mount *mp)
1255 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1259 * Flush out and invalidate all buffers associated with a bufobj
1260 * Called with the underlying object locked.
1263 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1268 if (flags & V_SAVE) {
1269 error = bufobj_wwait(bo, slpflag, slptimeo);
1274 if (bo->bo_dirty.bv_cnt > 0) {
1276 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1279 * XXX We could save a lock/unlock if this was only
1280 * enabled under INVARIANTS
1283 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1284 panic("vinvalbuf: dirty bufs");
1288 * If you alter this loop please notice that interlock is dropped and
1289 * reacquired in flushbuflist. Special care is needed to ensure that
1290 * no race conditions occur from this.
1293 error = flushbuflist(&bo->bo_clean,
1294 flags, bo, slpflag, slptimeo);
1295 if (error == 0 && !(flags & V_CLEANONLY))
1296 error = flushbuflist(&bo->bo_dirty,
1297 flags, bo, slpflag, slptimeo);
1298 if (error != 0 && error != EAGAIN) {
1302 } while (error != 0);
1305 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1306 * have write I/O in-progress but if there is a VM object then the
1307 * VM object can also have read-I/O in-progress.
1310 bufobj_wwait(bo, 0, 0);
1312 if (bo->bo_object != NULL) {
1313 VM_OBJECT_WLOCK(bo->bo_object);
1314 vm_object_pip_wait(bo->bo_object, "bovlbx");
1315 VM_OBJECT_WUNLOCK(bo->bo_object);
1318 } while (bo->bo_numoutput > 0);
1322 * Destroy the copy in the VM cache, too.
1324 if (bo->bo_object != NULL &&
1325 (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1326 VM_OBJECT_WLOCK(bo->bo_object);
1327 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1328 OBJPR_CLEANONLY : 0);
1329 VM_OBJECT_WUNLOCK(bo->bo_object);
1334 if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1335 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1336 panic("vinvalbuf: flush failed");
1343 * Flush out and invalidate all buffers associated with a vnode.
1344 * Called with the underlying object locked.
1347 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1350 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1351 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1352 if (vp->v_object != NULL && vp->v_object->handle != vp)
1354 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1358 * Flush out buffers on the specified list.
1362 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1365 struct buf *bp, *nbp;
1370 ASSERT_BO_WLOCKED(bo);
1373 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1374 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1375 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1381 lblkno = nbp->b_lblkno;
1382 xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
1385 error = BUF_TIMELOCK(bp,
1386 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo),
1387 "flushbuf", slpflag, slptimeo);
1390 return (error != ENOLCK ? error : EAGAIN);
1392 KASSERT(bp->b_bufobj == bo,
1393 ("bp %p wrong b_bufobj %p should be %p",
1394 bp, bp->b_bufobj, bo));
1395 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1401 * XXX Since there are no node locks for NFS, I
1402 * believe there is a slight chance that a delayed
1403 * write will occur while sleeping just above, so
1406 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1409 bp->b_flags |= B_ASYNC;
1412 return (EAGAIN); /* XXX: why not loop ? */
1415 bp->b_flags |= (B_INVAL | B_RELBUF);
1416 bp->b_flags &= ~B_ASYNC;
1420 (nbp->b_bufobj != bo ||
1421 nbp->b_lblkno != lblkno ||
1422 (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1423 break; /* nbp invalid */
1429 * Truncate a file's buffer and pages to a specified length. This
1430 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1434 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1436 struct buf *bp, *nbp;
1441 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1442 vp, cred, blksize, (uintmax_t)length);
1445 * Round up to the *next* lbn.
1447 trunclbn = (length + blksize - 1) / blksize;
1449 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1456 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1457 if (bp->b_lblkno < trunclbn)
1460 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1461 BO_LOCKPTR(bo)) == ENOLCK)
1465 bp->b_flags |= (B_INVAL | B_RELBUF);
1466 bp->b_flags &= ~B_ASYNC;
1472 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1473 (nbp->b_vp != vp) ||
1474 (nbp->b_flags & B_DELWRI))) {
1480 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1481 if (bp->b_lblkno < trunclbn)
1484 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1485 BO_LOCKPTR(bo)) == ENOLCK)
1488 bp->b_flags |= (B_INVAL | B_RELBUF);
1489 bp->b_flags &= ~B_ASYNC;
1495 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1496 (nbp->b_vp != vp) ||
1497 (nbp->b_flags & B_DELWRI) == 0)) {
1506 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1507 if (bp->b_lblkno > 0)
1510 * Since we hold the vnode lock this should only
1511 * fail if we're racing with the buf daemon.
1514 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1515 BO_LOCKPTR(bo)) == ENOLCK) {
1518 VNASSERT((bp->b_flags & B_DELWRI), vp,
1519 ("buf(%p) on dirty queue without DELWRI", bp));
1528 bufobj_wwait(bo, 0, 0);
1530 vnode_pager_setsize(vp, length);
1536 buf_vlist_remove(struct buf *bp)
1540 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1541 ASSERT_BO_WLOCKED(bp->b_bufobj);
1542 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1543 (BX_VNDIRTY|BX_VNCLEAN),
1544 ("buf_vlist_remove: Buf %p is on two lists", bp));
1545 if (bp->b_xflags & BX_VNDIRTY)
1546 bv = &bp->b_bufobj->bo_dirty;
1548 bv = &bp->b_bufobj->bo_clean;
1549 BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
1550 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1552 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1556 * Add the buffer to the sorted clean or dirty block list.
1558 * NOTE: xflags is passed as a constant, optimizing this inline function!
1561 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1567 ASSERT_BO_WLOCKED(bo);
1568 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1569 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1570 bp->b_xflags |= xflags;
1571 if (xflags & BX_VNDIRTY)
1577 * Keep the list ordered. Optimize empty list insertion. Assume
1578 * we tend to grow at the tail so lookup_le should usually be cheaper
1581 if (bv->bv_cnt == 0 ||
1582 bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
1583 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1584 else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
1585 TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
1587 TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
1588 error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
1590 panic("buf_vlist_add: Preallocated nodes insufficient.");
1595 * Lookup a buffer using the splay tree. Note that we specifically avoid
1596 * shadow buffers used in background bitmap writes.
1598 * This code isn't quite efficient as it could be because we are maintaining
1599 * two sorted lists and do not know which list the block resides in.
1601 * During a "make buildworld" the desired buffer is found at one of
1602 * the roots more than 60% of the time. Thus, checking both roots
1603 * before performing either splay eliminates unnecessary splays on the
1604 * first tree splayed.
1607 gbincore(struct bufobj *bo, daddr_t lblkno)
1611 ASSERT_BO_LOCKED(bo);
1612 bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
1615 return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
1619 * Associate a buffer with a vnode.
1622 bgetvp(struct vnode *vp, struct buf *bp)
1627 ASSERT_BO_WLOCKED(bo);
1628 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1630 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1631 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1632 ("bgetvp: bp already attached! %p", bp));
1638 * Insert onto list for new vnode.
1640 buf_vlist_add(bp, bo, BX_VNCLEAN);
1644 * Disassociate a buffer from a vnode.
1647 brelvp(struct buf *bp)
1652 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1653 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1656 * Delete from old vnode list, if on one.
1658 vp = bp->b_vp; /* XXX */
1661 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1662 buf_vlist_remove(bp);
1664 panic("brelvp: Buffer %p not on queue.", bp);
1665 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1666 bo->bo_flag &= ~BO_ONWORKLST;
1667 mtx_lock(&sync_mtx);
1668 LIST_REMOVE(bo, bo_synclist);
1669 syncer_worklist_len--;
1670 mtx_unlock(&sync_mtx);
1673 bp->b_bufobj = NULL;
1679 * Add an item to the syncer work queue.
1682 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1686 ASSERT_BO_WLOCKED(bo);
1688 mtx_lock(&sync_mtx);
1689 if (bo->bo_flag & BO_ONWORKLST)
1690 LIST_REMOVE(bo, bo_synclist);
1692 bo->bo_flag |= BO_ONWORKLST;
1693 syncer_worklist_len++;
1696 if (delay > syncer_maxdelay - 2)
1697 delay = syncer_maxdelay - 2;
1698 slot = (syncer_delayno + delay) & syncer_mask;
1700 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1701 mtx_unlock(&sync_mtx);
1705 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1709 mtx_lock(&sync_mtx);
1710 len = syncer_worklist_len - sync_vnode_count;
1711 mtx_unlock(&sync_mtx);
1712 error = SYSCTL_OUT(req, &len, sizeof(len));
1716 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1717 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1719 static struct proc *updateproc;
1720 static void sched_sync(void);
1721 static struct kproc_desc up_kp = {
1726 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1729 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1734 *bo = LIST_FIRST(slp);
1737 vp = (*bo)->__bo_vnode; /* XXX */
1738 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1741 * We use vhold in case the vnode does not
1742 * successfully sync. vhold prevents the vnode from
1743 * going away when we unlock the sync_mtx so that
1744 * we can acquire the vnode interlock.
1747 mtx_unlock(&sync_mtx);
1749 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1751 mtx_lock(&sync_mtx);
1752 return (*bo == LIST_FIRST(slp));
1754 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1755 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1757 vn_finished_write(mp);
1759 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1761 * Put us back on the worklist. The worklist
1762 * routine will remove us from our current
1763 * position and then add us back in at a later
1766 vn_syncer_add_to_worklist(*bo, syncdelay);
1770 mtx_lock(&sync_mtx);
1775 * System filesystem synchronizer daemon.
1780 struct synclist *next, *slp;
1783 struct thread *td = curthread;
1785 int net_worklist_len;
1786 int syncer_final_iter;
1791 syncer_final_iter = 0;
1793 syncer_state = SYNCER_RUNNING;
1794 starttime = time_uptime;
1795 td->td_pflags |= TDP_NORUNNINGBUF;
1797 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1800 mtx_lock(&sync_mtx);
1802 if (syncer_state == SYNCER_FINAL_DELAY &&
1803 syncer_final_iter == 0) {
1804 mtx_unlock(&sync_mtx);
1805 kproc_suspend_check(td->td_proc);
1806 mtx_lock(&sync_mtx);
1808 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1809 if (syncer_state != SYNCER_RUNNING &&
1810 starttime != time_uptime) {
1812 printf("\nSyncing disks, vnodes remaining...");
1815 printf("%d ", net_worklist_len);
1817 starttime = time_uptime;
1820 * Push files whose dirty time has expired. Be careful
1821 * of interrupt race on slp queue.
1823 * Skip over empty worklist slots when shutting down.
1826 slp = &syncer_workitem_pending[syncer_delayno];
1827 syncer_delayno += 1;
1828 if (syncer_delayno == syncer_maxdelay)
1830 next = &syncer_workitem_pending[syncer_delayno];
1832 * If the worklist has wrapped since the
1833 * it was emptied of all but syncer vnodes,
1834 * switch to the FINAL_DELAY state and run
1835 * for one more second.
1837 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1838 net_worklist_len == 0 &&
1839 last_work_seen == syncer_delayno) {
1840 syncer_state = SYNCER_FINAL_DELAY;
1841 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1843 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1844 syncer_worklist_len > 0);
1847 * Keep track of the last time there was anything
1848 * on the worklist other than syncer vnodes.
1849 * Return to the SHUTTING_DOWN state if any
1852 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1853 last_work_seen = syncer_delayno;
1854 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1855 syncer_state = SYNCER_SHUTTING_DOWN;
1856 while (!LIST_EMPTY(slp)) {
1857 error = sync_vnode(slp, &bo, td);
1859 LIST_REMOVE(bo, bo_synclist);
1860 LIST_INSERT_HEAD(next, bo, bo_synclist);
1864 if (first_printf == 0)
1865 wdog_kern_pat(WD_LASTVAL);
1868 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1869 syncer_final_iter--;
1871 * The variable rushjob allows the kernel to speed up the
1872 * processing of the filesystem syncer process. A rushjob
1873 * value of N tells the filesystem syncer to process the next
1874 * N seconds worth of work on its queue ASAP. Currently rushjob
1875 * is used by the soft update code to speed up the filesystem
1876 * syncer process when the incore state is getting so far
1877 * ahead of the disk that the kernel memory pool is being
1878 * threatened with exhaustion.
1885 * Just sleep for a short period of time between
1886 * iterations when shutting down to allow some I/O
1889 * If it has taken us less than a second to process the
1890 * current work, then wait. Otherwise start right over
1891 * again. We can still lose time if any single round
1892 * takes more than two seconds, but it does not really
1893 * matter as we are just trying to generally pace the
1894 * filesystem activity.
1896 if (syncer_state != SYNCER_RUNNING ||
1897 time_uptime == starttime) {
1899 sched_prio(td, PPAUSE);
1902 if (syncer_state != SYNCER_RUNNING)
1903 cv_timedwait(&sync_wakeup, &sync_mtx,
1904 hz / SYNCER_SHUTDOWN_SPEEDUP);
1905 else if (time_uptime == starttime)
1906 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1911 * Request the syncer daemon to speed up its work.
1912 * We never push it to speed up more than half of its
1913 * normal turn time, otherwise it could take over the cpu.
1916 speedup_syncer(void)
1920 mtx_lock(&sync_mtx);
1921 if (rushjob < syncdelay / 2) {
1923 stat_rush_requests += 1;
1926 mtx_unlock(&sync_mtx);
1927 cv_broadcast(&sync_wakeup);
1932 * Tell the syncer to speed up its work and run though its work
1933 * list several times, then tell it to shut down.
1936 syncer_shutdown(void *arg, int howto)
1939 if (howto & RB_NOSYNC)
1941 mtx_lock(&sync_mtx);
1942 syncer_state = SYNCER_SHUTTING_DOWN;
1944 mtx_unlock(&sync_mtx);
1945 cv_broadcast(&sync_wakeup);
1946 kproc_shutdown(arg, howto);
1950 * Reassign a buffer from one vnode to another.
1951 * Used to assign file specific control information
1952 * (indirect blocks) to the vnode to which they belong.
1955 reassignbuf(struct buf *bp)
1968 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1969 bp, bp->b_vp, bp->b_flags);
1971 * B_PAGING flagged buffers cannot be reassigned because their vp
1972 * is not fully linked in.
1974 if (bp->b_flags & B_PAGING)
1975 panic("cannot reassign paging buffer");
1978 * Delete from old vnode list, if on one.
1981 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1982 buf_vlist_remove(bp);
1984 panic("reassignbuf: Buffer %p not on queue.", bp);
1986 * If dirty, put on list of dirty buffers; otherwise insert onto list
1989 if (bp->b_flags & B_DELWRI) {
1990 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1991 switch (vp->v_type) {
2001 vn_syncer_add_to_worklist(bo, delay);
2003 buf_vlist_add(bp, bo, BX_VNDIRTY);
2005 buf_vlist_add(bp, bo, BX_VNCLEAN);
2007 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2008 mtx_lock(&sync_mtx);
2009 LIST_REMOVE(bo, bo_synclist);
2010 syncer_worklist_len--;
2011 mtx_unlock(&sync_mtx);
2012 bo->bo_flag &= ~BO_ONWORKLST;
2017 bp = TAILQ_FIRST(&bv->bv_hd);
2018 KASSERT(bp == NULL || bp->b_bufobj == bo,
2019 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2020 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2021 KASSERT(bp == NULL || bp->b_bufobj == bo,
2022 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2024 bp = TAILQ_FIRST(&bv->bv_hd);
2025 KASSERT(bp == NULL || bp->b_bufobj == bo,
2026 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2027 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2028 KASSERT(bp == NULL || bp->b_bufobj == bo,
2029 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2035 * Increment the use and hold counts on the vnode, taking care to reference
2036 * the driver's usecount if this is a chardev. The vholdl() will remove
2037 * the vnode from the free list if it is presently free. Requires the
2038 * vnode interlock and returns with it held.
2041 v_incr_usecount(struct vnode *vp)
2044 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2046 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2048 vp->v_rdev->si_usecount++;
2055 * Turn a holdcnt into a use+holdcnt such that only one call to
2056 * v_decr_usecount is needed.
2059 v_upgrade_usecount(struct vnode *vp)
2062 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2064 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2066 vp->v_rdev->si_usecount++;
2072 * Decrement the vnode use and hold count along with the driver's usecount
2073 * if this is a chardev. The vdropl() below releases the vnode interlock
2074 * as it may free the vnode.
2077 v_decr_usecount(struct vnode *vp)
2080 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2081 VNASSERT(vp->v_usecount > 0, vp,
2082 ("v_decr_usecount: negative usecount"));
2083 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2085 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2087 vp->v_rdev->si_usecount--;
2094 * Decrement only the use count and driver use count. This is intended to
2095 * be paired with a follow on vdropl() to release the remaining hold count.
2096 * In this way we may vgone() a vnode with a 0 usecount without risk of
2097 * having it end up on a free list because the hold count is kept above 0.
2100 v_decr_useonly(struct vnode *vp)
2103 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2104 VNASSERT(vp->v_usecount > 0, vp,
2105 ("v_decr_useonly: negative usecount"));
2106 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2108 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2110 vp->v_rdev->si_usecount--;
2116 * Grab a particular vnode from the free list, increment its
2117 * reference count and lock it. VI_DOOMED is set if the vnode
2118 * is being destroyed. Only callers who specify LK_RETRY will
2119 * see doomed vnodes. If inactive processing was delayed in
2120 * vput try to do it here.
2123 vget(struct vnode *vp, int flags, struct thread *td)
2128 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2129 ("vget: invalid lock operation"));
2130 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2132 if ((flags & LK_INTERLOCK) == 0)
2135 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2137 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2141 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2142 panic("vget: vn_lock failed to return ENOENT\n");
2144 /* Upgrade our holdcnt to a usecount. */
2145 v_upgrade_usecount(vp);
2147 * We don't guarantee that any particular close will
2148 * trigger inactive processing so just make a best effort
2149 * here at preventing a reference to a removed file. If
2150 * we don't succeed no harm is done.
2152 if (vp->v_iflag & VI_OWEINACT) {
2153 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2154 (flags & LK_NOWAIT) == 0)
2156 vp->v_iflag &= ~VI_OWEINACT;
2163 * Increase the reference count of a vnode.
2166 vref(struct vnode *vp)
2169 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2171 v_incr_usecount(vp);
2176 * Return reference count of a vnode.
2178 * The results of this call are only guaranteed when some mechanism other
2179 * than the VI lock is used to stop other processes from gaining references
2180 * to the vnode. This may be the case if the caller holds the only reference.
2181 * This is also useful when stale data is acceptable as race conditions may
2182 * be accounted for by some other means.
2185 vrefcnt(struct vnode *vp)
2190 usecnt = vp->v_usecount;
2196 #define VPUTX_VRELE 1
2197 #define VPUTX_VPUT 2
2198 #define VPUTX_VUNREF 3
2201 vputx(struct vnode *vp, int func)
2205 KASSERT(vp != NULL, ("vputx: null vp"));
2206 if (func == VPUTX_VUNREF)
2207 ASSERT_VOP_LOCKED(vp, "vunref");
2208 else if (func == VPUTX_VPUT)
2209 ASSERT_VOP_LOCKED(vp, "vput");
2211 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2212 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2215 /* Skip this v_writecount check if we're going to panic below. */
2216 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2217 ("vputx: missed vn_close"));
2220 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2221 vp->v_usecount == 1)) {
2222 if (func == VPUTX_VPUT)
2224 v_decr_usecount(vp);
2228 if (vp->v_usecount != 1) {
2229 vprint("vputx: negative ref count", vp);
2230 panic("vputx: negative ref cnt");
2232 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2234 * We want to hold the vnode until the inactive finishes to
2235 * prevent vgone() races. We drop the use count here and the
2236 * hold count below when we're done.
2240 * We must call VOP_INACTIVE with the node locked. Mark
2241 * as VI_DOINGINACT to avoid recursion.
2243 vp->v_iflag |= VI_OWEINACT;
2246 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2250 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2251 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2257 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2258 error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
2263 if (vp->v_usecount > 0)
2264 vp->v_iflag &= ~VI_OWEINACT;
2266 if (vp->v_iflag & VI_OWEINACT)
2267 vinactive(vp, curthread);
2268 if (func != VPUTX_VUNREF)
2275 * Vnode put/release.
2276 * If count drops to zero, call inactive routine and return to freelist.
2279 vrele(struct vnode *vp)
2282 vputx(vp, VPUTX_VRELE);
2286 * Release an already locked vnode. This give the same effects as
2287 * unlock+vrele(), but takes less time and avoids releasing and
2288 * re-aquiring the lock (as vrele() acquires the lock internally.)
2291 vput(struct vnode *vp)
2294 vputx(vp, VPUTX_VPUT);
2298 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2301 vunref(struct vnode *vp)
2304 vputx(vp, VPUTX_VUNREF);
2308 * Somebody doesn't want the vnode recycled.
2311 vhold(struct vnode *vp)
2320 * Increase the hold count and activate if this is the first reference.
2323 vholdl(struct vnode *vp)
2327 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2329 if (!VSHOULDBUSY(vp))
2331 ASSERT_VI_LOCKED(vp, "vholdl");
2332 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2333 VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2335 * Remove a vnode from the free list, mark it as in use,
2336 * and put it on the active list.
2338 mtx_lock(&vnode_free_list_mtx);
2339 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2341 vp->v_iflag &= ~(VI_FREE|VI_AGE);
2342 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2343 ("Activating already active vnode"));
2344 vp->v_iflag |= VI_ACTIVE;
2346 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2347 mp->mnt_activevnodelistsize++;
2348 mtx_unlock(&vnode_free_list_mtx);
2352 * Note that there is one less who cares about this vnode.
2353 * vdrop() is the opposite of vhold().
2356 vdrop(struct vnode *vp)
2364 * Drop the hold count of the vnode. If this is the last reference to
2365 * the vnode we place it on the free list unless it has been vgone'd
2366 * (marked VI_DOOMED) in which case we will free it.
2369 vdropl(struct vnode *vp)
2375 ASSERT_VI_LOCKED(vp, "vdropl");
2376 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2377 if (vp->v_holdcnt <= 0)
2378 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2380 if (vp->v_holdcnt > 0) {
2384 if ((vp->v_iflag & VI_DOOMED) == 0) {
2386 * Mark a vnode as free: remove it from its active list
2387 * and put it up for recycling on the freelist.
2389 VNASSERT(vp->v_op != NULL, vp,
2390 ("vdropl: vnode already reclaimed."));
2391 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2392 ("vnode already free"));
2393 VNASSERT(VSHOULDFREE(vp), vp,
2394 ("vdropl: freeing when we shouldn't"));
2395 active = vp->v_iflag & VI_ACTIVE;
2396 vp->v_iflag &= ~VI_ACTIVE;
2398 mtx_lock(&vnode_free_list_mtx);
2400 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2402 mp->mnt_activevnodelistsize--;
2404 if (vp->v_iflag & VI_AGE) {
2405 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
2407 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
2410 vp->v_iflag &= ~VI_AGE;
2411 vp->v_iflag |= VI_FREE;
2412 mtx_unlock(&vnode_free_list_mtx);
2417 * The vnode has been marked for destruction, so free it.
2419 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2420 atomic_subtract_long(&numvnodes, 1);
2422 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2423 ("cleaned vnode still on the free list."));
2424 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2425 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2426 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2427 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2428 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2429 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2430 VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
2431 ("clean blk trie not empty"));
2432 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2433 VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
2434 ("dirty blk trie not empty"));
2435 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2436 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2437 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2440 mac_vnode_destroy(vp);
2442 if (vp->v_pollinfo != NULL)
2443 destroy_vpollinfo(vp->v_pollinfo);
2445 /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2448 rangelock_destroy(&vp->v_rl);
2449 lockdestroy(vp->v_vnlock);
2450 mtx_destroy(&vp->v_interlock);
2451 rw_destroy(BO_LOCKPTR(bo));
2452 uma_zfree(vnode_zone, vp);
2456 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2457 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2458 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2459 * failed lock upgrade.
2462 vinactive(struct vnode *vp, struct thread *td)
2464 struct vm_object *obj;
2466 ASSERT_VOP_ELOCKED(vp, "vinactive");
2467 ASSERT_VI_LOCKED(vp, "vinactive");
2468 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2469 ("vinactive: recursed on VI_DOINGINACT"));
2470 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2471 vp->v_iflag |= VI_DOINGINACT;
2472 vp->v_iflag &= ~VI_OWEINACT;
2475 * Before moving off the active list, we must be sure that any
2476 * modified pages are on the vnode's dirty list since these will
2477 * no longer be checked once the vnode is on the inactive list.
2478 * Because the vnode vm object keeps a hold reference on the vnode
2479 * if there is at least one resident non-cached page, the vnode
2480 * cannot leave the active list without the page cleanup done.
2483 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2484 VM_OBJECT_WLOCK(obj);
2485 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2486 VM_OBJECT_WUNLOCK(obj);
2488 VOP_INACTIVE(vp, td);
2490 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2491 ("vinactive: lost VI_DOINGINACT"));
2492 vp->v_iflag &= ~VI_DOINGINACT;
2496 * Remove any vnodes in the vnode table belonging to mount point mp.
2498 * If FORCECLOSE is not specified, there should not be any active ones,
2499 * return error if any are found (nb: this is a user error, not a
2500 * system error). If FORCECLOSE is specified, detach any active vnodes
2503 * If WRITECLOSE is set, only flush out regular file vnodes open for
2506 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2508 * `rootrefs' specifies the base reference count for the root vnode
2509 * of this filesystem. The root vnode is considered busy if its
2510 * v_usecount exceeds this value. On a successful return, vflush(, td)
2511 * will call vrele() on the root vnode exactly rootrefs times.
2512 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2516 static int busyprt = 0; /* print out busy vnodes */
2517 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2521 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2523 struct vnode *vp, *mvp, *rootvp = NULL;
2525 int busy = 0, error;
2527 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2530 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2531 ("vflush: bad args"));
2533 * Get the filesystem root vnode. We can vput() it
2534 * immediately, since with rootrefs > 0, it won't go away.
2536 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2537 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2544 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2546 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2549 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2553 * Skip over a vnodes marked VV_SYSTEM.
2555 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2561 * If WRITECLOSE is set, flush out unlinked but still open
2562 * files (even if open only for reading) and regular file
2563 * vnodes open for writing.
2565 if (flags & WRITECLOSE) {
2566 if (vp->v_object != NULL) {
2567 VM_OBJECT_WLOCK(vp->v_object);
2568 vm_object_page_clean(vp->v_object, 0, 0, 0);
2569 VM_OBJECT_WUNLOCK(vp->v_object);
2571 error = VOP_FSYNC(vp, MNT_WAIT, td);
2575 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2578 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2581 if ((vp->v_type == VNON ||
2582 (error == 0 && vattr.va_nlink > 0)) &&
2583 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2591 * With v_usecount == 0, all we need to do is clear out the
2592 * vnode data structures and we are done.
2594 * If FORCECLOSE is set, forcibly close the vnode.
2596 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2597 VNASSERT(vp->v_usecount == 0 ||
2598 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2599 ("device VNODE %p is FORCECLOSED", vp));
2605 vprint("vflush: busy vnode", vp);
2611 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2613 * If just the root vnode is busy, and if its refcount
2614 * is equal to `rootrefs', then go ahead and kill it.
2617 KASSERT(busy > 0, ("vflush: not busy"));
2618 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2619 ("vflush: usecount %d < rootrefs %d",
2620 rootvp->v_usecount, rootrefs));
2621 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2622 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2624 VOP_UNLOCK(rootvp, 0);
2630 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2634 for (; rootrefs > 0; rootrefs--)
2640 * Recycle an unused vnode to the front of the free list.
2643 vrecycle(struct vnode *vp)
2647 ASSERT_VOP_ELOCKED(vp, "vrecycle");
2648 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2651 if (vp->v_usecount == 0) {
2660 * Eliminate all activity associated with a vnode
2661 * in preparation for reuse.
2664 vgone(struct vnode *vp)
2672 notify_lowervp_vfs_dummy(struct mount *mp __unused,
2673 struct vnode *lowervp __unused)
2678 * Notify upper mounts about reclaimed or unlinked vnode.
2681 vfs_notify_upper(struct vnode *vp, int event)
2683 static struct vfsops vgonel_vfsops = {
2684 .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
2685 .vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
2687 struct mount *mp, *ump, *mmp;
2694 if (TAILQ_EMPTY(&mp->mnt_uppers))
2697 mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
2698 mmp->mnt_op = &vgonel_vfsops;
2699 mmp->mnt_kern_flag |= MNTK_MARKER;
2701 mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
2702 for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
2703 if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
2704 ump = TAILQ_NEXT(ump, mnt_upper_link);
2707 TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
2710 case VFS_NOTIFY_UPPER_RECLAIM:
2711 VFS_RECLAIM_LOWERVP(ump, vp);
2713 case VFS_NOTIFY_UPPER_UNLINK:
2714 VFS_UNLINK_LOWERVP(ump, vp);
2717 KASSERT(0, ("invalid event %d", event));
2721 ump = TAILQ_NEXT(mmp, mnt_upper_link);
2722 TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
2725 mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
2726 if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
2727 mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
2728 wakeup(&mp->mnt_uppers);
2735 * vgone, with the vp interlock held.
2738 vgonel(struct vnode *vp)
2745 ASSERT_VOP_ELOCKED(vp, "vgonel");
2746 ASSERT_VI_LOCKED(vp, "vgonel");
2747 VNASSERT(vp->v_holdcnt, vp,
2748 ("vgonel: vp %p has no reference.", vp));
2749 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2753 * Don't vgonel if we're already doomed.
2755 if (vp->v_iflag & VI_DOOMED)
2757 vp->v_iflag |= VI_DOOMED;
2760 * Check to see if the vnode is in use. If so, we have to call
2761 * VOP_CLOSE() and VOP_INACTIVE().
2763 active = vp->v_usecount;
2764 oweinact = (vp->v_iflag & VI_OWEINACT);
2766 vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
2769 * Clean out any buffers associated with the vnode.
2770 * If the flush fails, just toss the buffers.
2773 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2774 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2775 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2776 vinvalbuf(vp, 0, 0, 0);
2779 * If purging an active vnode, it must be closed and
2780 * deactivated before being reclaimed.
2783 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2784 if (oweinact || active) {
2786 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2790 if (vp->v_type == VSOCK)
2791 vfs_unp_reclaim(vp);
2793 * Reclaim the vnode.
2795 if (VOP_RECLAIM(vp, td))
2796 panic("vgone: cannot reclaim");
2798 vn_finished_secondary_write(mp);
2799 VNASSERT(vp->v_object == NULL, vp,
2800 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2802 * Clear the advisory locks and wake up waiting threads.
2804 (void)VOP_ADVLOCKPURGE(vp);
2806 * Delete from old mount point vnode list.
2811 * Done with purge, reset to the standard lock and invalidate
2815 vp->v_vnlock = &vp->v_lock;
2816 vp->v_op = &dead_vnodeops;
2822 * Calculate the total number of references to a special device.
2825 vcount(struct vnode *vp)
2830 count = vp->v_rdev->si_usecount;
2836 * Same as above, but using the struct cdev *as argument
2839 count_dev(struct cdev *dev)
2844 count = dev->si_usecount;
2850 * Print out a description of a vnode.
2852 static char *typename[] =
2853 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2857 vn_printf(struct vnode *vp, const char *fmt, ...)
2860 char buf[256], buf2[16];
2866 printf("%p: ", (void *)vp);
2867 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2868 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2869 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2872 if (vp->v_vflag & VV_ROOT)
2873 strlcat(buf, "|VV_ROOT", sizeof(buf));
2874 if (vp->v_vflag & VV_ISTTY)
2875 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2876 if (vp->v_vflag & VV_NOSYNC)
2877 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2878 if (vp->v_vflag & VV_ETERNALDEV)
2879 strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
2880 if (vp->v_vflag & VV_CACHEDLABEL)
2881 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2882 if (vp->v_vflag & VV_TEXT)
2883 strlcat(buf, "|VV_TEXT", sizeof(buf));
2884 if (vp->v_vflag & VV_COPYONWRITE)
2885 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2886 if (vp->v_vflag & VV_SYSTEM)
2887 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2888 if (vp->v_vflag & VV_PROCDEP)
2889 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2890 if (vp->v_vflag & VV_NOKNOTE)
2891 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2892 if (vp->v_vflag & VV_DELETED)
2893 strlcat(buf, "|VV_DELETED", sizeof(buf));
2894 if (vp->v_vflag & VV_MD)
2895 strlcat(buf, "|VV_MD", sizeof(buf));
2896 if (vp->v_vflag & VV_FORCEINSMQ)
2897 strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
2898 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
2899 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2900 VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
2902 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2903 strlcat(buf, buf2, sizeof(buf));
2905 if (vp->v_iflag & VI_MOUNT)
2906 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2907 if (vp->v_iflag & VI_AGE)
2908 strlcat(buf, "|VI_AGE", sizeof(buf));
2909 if (vp->v_iflag & VI_DOOMED)
2910 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2911 if (vp->v_iflag & VI_FREE)
2912 strlcat(buf, "|VI_FREE", sizeof(buf));
2913 if (vp->v_iflag & VI_ACTIVE)
2914 strlcat(buf, "|VI_ACTIVE", sizeof(buf));
2915 if (vp->v_iflag & VI_DOINGINACT)
2916 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2917 if (vp->v_iflag & VI_OWEINACT)
2918 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2919 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2920 VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
2922 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2923 strlcat(buf, buf2, sizeof(buf));
2925 printf(" flags (%s)\n", buf + 1);
2926 if (mtx_owned(VI_MTX(vp)))
2927 printf(" VI_LOCKed");
2928 if (vp->v_object != NULL)
2929 printf(" v_object %p ref %d pages %d "
2930 "cleanbuf %d dirtybuf %d\n",
2931 vp->v_object, vp->v_object->ref_count,
2932 vp->v_object->resident_page_count,
2933 vp->v_bufobj.bo_dirty.bv_cnt,
2934 vp->v_bufobj.bo_clean.bv_cnt);
2936 lockmgr_printinfo(vp->v_vnlock);
2937 if (vp->v_data != NULL)
2943 * List all of the locked vnodes in the system.
2944 * Called when debugging the kernel.
2946 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2952 * Note: because this is DDB, we can't obey the locking semantics
2953 * for these structures, which means we could catch an inconsistent
2954 * state and dereference a nasty pointer. Not much to be done
2957 db_printf("Locked vnodes\n");
2958 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2959 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2960 if (vp->v_type != VMARKER && VOP_ISLOCKED(vp))
2967 * Show details about the given vnode.
2969 DB_SHOW_COMMAND(vnode, db_show_vnode)
2975 vp = (struct vnode *)addr;
2976 vn_printf(vp, "vnode ");
2980 * Show details about the given mount point.
2982 DB_SHOW_COMMAND(mount, db_show_mount)
2993 /* No address given, print short info about all mount points. */
2994 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2995 db_printf("%p %s on %s (%s)\n", mp,
2996 mp->mnt_stat.f_mntfromname,
2997 mp->mnt_stat.f_mntonname,
2998 mp->mnt_stat.f_fstypename);
3002 db_printf("\nMore info: show mount <addr>\n");
3006 mp = (struct mount *)addr;
3007 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
3008 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
3011 mflags = mp->mnt_flag;
3012 #define MNT_FLAG(flag) do { \
3013 if (mflags & (flag)) { \
3014 if (buf[0] != '\0') \
3015 strlcat(buf, ", ", sizeof(buf)); \
3016 strlcat(buf, (#flag) + 4, sizeof(buf)); \
3017 mflags &= ~(flag); \
3020 MNT_FLAG(MNT_RDONLY);
3021 MNT_FLAG(MNT_SYNCHRONOUS);
3022 MNT_FLAG(MNT_NOEXEC);
3023 MNT_FLAG(MNT_NOSUID);
3024 MNT_FLAG(MNT_NFS4ACLS);
3025 MNT_FLAG(MNT_UNION);
3026 MNT_FLAG(MNT_ASYNC);
3027 MNT_FLAG(MNT_SUIDDIR);
3028 MNT_FLAG(MNT_SOFTDEP);
3029 MNT_FLAG(MNT_NOSYMFOLLOW);
3030 MNT_FLAG(MNT_GJOURNAL);
3031 MNT_FLAG(MNT_MULTILABEL);
3033 MNT_FLAG(MNT_NOATIME);
3034 MNT_FLAG(MNT_NOCLUSTERR);
3035 MNT_FLAG(MNT_NOCLUSTERW);
3037 MNT_FLAG(MNT_EXRDONLY);
3038 MNT_FLAG(MNT_EXPORTED);
3039 MNT_FLAG(MNT_DEFEXPORTED);
3040 MNT_FLAG(MNT_EXPORTANON);
3041 MNT_FLAG(MNT_EXKERB);
3042 MNT_FLAG(MNT_EXPUBLIC);
3043 MNT_FLAG(MNT_LOCAL);
3044 MNT_FLAG(MNT_QUOTA);
3045 MNT_FLAG(MNT_ROOTFS);
3047 MNT_FLAG(MNT_IGNORE);
3048 MNT_FLAG(MNT_UPDATE);
3049 MNT_FLAG(MNT_DELEXPORT);
3050 MNT_FLAG(MNT_RELOAD);
3051 MNT_FLAG(MNT_FORCE);
3052 MNT_FLAG(MNT_SNAPSHOT);
3053 MNT_FLAG(MNT_BYFSID);
3057 strlcat(buf, ", ", sizeof(buf));
3058 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3059 "0x%016jx", mflags);
3061 db_printf(" mnt_flag = %s\n", buf);
3064 flags = mp->mnt_kern_flag;
3065 #define MNT_KERN_FLAG(flag) do { \
3066 if (flags & (flag)) { \
3067 if (buf[0] != '\0') \
3068 strlcat(buf, ", ", sizeof(buf)); \
3069 strlcat(buf, (#flag) + 5, sizeof(buf)); \
3073 MNT_KERN_FLAG(MNTK_UNMOUNTF);
3074 MNT_KERN_FLAG(MNTK_ASYNC);
3075 MNT_KERN_FLAG(MNTK_SOFTDEP);
3076 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3077 MNT_KERN_FLAG(MNTK_DRAINING);
3078 MNT_KERN_FLAG(MNTK_REFEXPIRE);
3079 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3080 MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3081 MNT_KERN_FLAG(MNTK_NO_IOPF);
3082 MNT_KERN_FLAG(MNTK_VGONE_UPPER);
3083 MNT_KERN_FLAG(MNTK_VGONE_WAITER);
3084 MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
3085 MNT_KERN_FLAG(MNTK_MARKER);
3086 MNT_KERN_FLAG(MNTK_NOASYNC);
3087 MNT_KERN_FLAG(MNTK_UNMOUNT);
3088 MNT_KERN_FLAG(MNTK_MWAIT);
3089 MNT_KERN_FLAG(MNTK_SUSPEND);
3090 MNT_KERN_FLAG(MNTK_SUSPEND2);
3091 MNT_KERN_FLAG(MNTK_SUSPENDED);
3092 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3093 MNT_KERN_FLAG(MNTK_NOKNOTE);
3094 #undef MNT_KERN_FLAG
3097 strlcat(buf, ", ", sizeof(buf));
3098 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3101 db_printf(" mnt_kern_flag = %s\n", buf);
3103 db_printf(" mnt_opt = ");
3104 opt = TAILQ_FIRST(mp->mnt_opt);
3106 db_printf("%s", opt->name);
3107 opt = TAILQ_NEXT(opt, link);
3108 while (opt != NULL) {
3109 db_printf(", %s", opt->name);
3110 opt = TAILQ_NEXT(opt, link);
3116 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
3117 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3118 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3119 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3120 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3121 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3122 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3123 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3124 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3125 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3126 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3127 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3129 db_printf(" mnt_cred = { uid=%u ruid=%u",
3130 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3131 if (jailed(mp->mnt_cred))
3132 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3134 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
3135 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
3136 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3137 db_printf(" mnt_activevnodelistsize = %d\n",
3138 mp->mnt_activevnodelistsize);
3139 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3140 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3141 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3142 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
3143 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3144 db_printf(" mnt_secondary_accwrites = %d\n",
3145 mp->mnt_secondary_accwrites);
3146 db_printf(" mnt_gjprovider = %s\n",
3147 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3149 db_printf("\n\nList of active vnodes\n");
3150 TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3151 if (vp->v_type != VMARKER) {
3152 vn_printf(vp, "vnode ");
3157 db_printf("\n\nList of inactive vnodes\n");
3158 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3159 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3160 vn_printf(vp, "vnode ");
3169 * Fill in a struct xvfsconf based on a struct vfsconf.
3172 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3174 struct xvfsconf xvfsp;
3176 bzero(&xvfsp, sizeof(xvfsp));
3177 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3178 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3179 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3180 xvfsp.vfc_flags = vfsp->vfc_flags;
3182 * These are unused in userland, we keep them
3183 * to not break binary compatibility.
3185 xvfsp.vfc_vfsops = NULL;
3186 xvfsp.vfc_next = NULL;
3187 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3190 #ifdef COMPAT_FREEBSD32
3192 uint32_t vfc_vfsops;
3193 char vfc_name[MFSNAMELEN];
3194 int32_t vfc_typenum;
3195 int32_t vfc_refcount;
3201 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3203 struct xvfsconf32 xvfsp;
3205 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3206 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3207 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3208 xvfsp.vfc_flags = vfsp->vfc_flags;
3209 xvfsp.vfc_vfsops = 0;
3211 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3216 * Top level filesystem related information gathering.
3219 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3221 struct vfsconf *vfsp;
3225 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3226 #ifdef COMPAT_FREEBSD32
3227 if (req->flags & SCTL_MASK32)
3228 error = vfsconf2x32(req, vfsp);
3231 error = vfsconf2x(req, vfsp);
3238 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3239 NULL, 0, sysctl_vfs_conflist,
3240 "S,xvfsconf", "List of all configured filesystems");
3242 #ifndef BURN_BRIDGES
3243 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3246 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3248 int *name = (int *)arg1 - 1; /* XXX */
3249 u_int namelen = arg2 + 1; /* XXX */
3250 struct vfsconf *vfsp;
3252 log(LOG_WARNING, "userland calling deprecated sysctl, "
3253 "please rebuild world\n");
3255 #if 1 || defined(COMPAT_PRELITE2)
3256 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3258 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3262 case VFS_MAXTYPENUM:
3265 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3268 return (ENOTDIR); /* overloaded */
3269 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3270 if (vfsp->vfc_typenum == name[2])
3273 return (EOPNOTSUPP);
3274 #ifdef COMPAT_FREEBSD32
3275 if (req->flags & SCTL_MASK32)
3276 return (vfsconf2x32(req, vfsp));
3279 return (vfsconf2x(req, vfsp));
3281 return (EOPNOTSUPP);
3284 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3285 vfs_sysctl, "Generic filesystem");
3287 #if 1 || defined(COMPAT_PRELITE2)
3290 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3293 struct vfsconf *vfsp;
3294 struct ovfsconf ovfs;
3296 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3297 bzero(&ovfs, sizeof(ovfs));
3298 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3299 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3300 ovfs.vfc_index = vfsp->vfc_typenum;
3301 ovfs.vfc_refcount = vfsp->vfc_refcount;
3302 ovfs.vfc_flags = vfsp->vfc_flags;
3303 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3310 #endif /* 1 || COMPAT_PRELITE2 */
3311 #endif /* !BURN_BRIDGES */
3313 #define KINFO_VNODESLOP 10
3316 * Dump vnode list (via sysctl).
3320 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3328 * Stale numvnodes access is not fatal here.
3331 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3333 /* Make an estimate */
3334 return (SYSCTL_OUT(req, 0, len));
3336 error = sysctl_wire_old_buffer(req, 0);
3339 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3341 mtx_lock(&mountlist_mtx);
3342 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3343 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3346 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3350 xvn[n].xv_size = sizeof *xvn;
3351 xvn[n].xv_vnode = vp;
3352 xvn[n].xv_id = 0; /* XXX compat */
3353 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3355 XV_COPY(writecount);
3361 xvn[n].xv_flag = vp->v_vflag;
3363 switch (vp->v_type) {
3370 if (vp->v_rdev == NULL) {
3374 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3377 xvn[n].xv_socket = vp->v_socket;
3380 xvn[n].xv_fifo = vp->v_fifoinfo;
3385 /* shouldn't happen? */
3393 mtx_lock(&mountlist_mtx);
3398 mtx_unlock(&mountlist_mtx);
3400 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3405 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3406 0, 0, sysctl_vnode, "S,xvnode", "");
3410 * Unmount all filesystems. The list is traversed in reverse order
3411 * of mounting to avoid dependencies.
3414 vfs_unmountall(void)
3420 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3424 * Since this only runs when rebooting, it is not interlocked.
3426 while(!TAILQ_EMPTY(&mountlist)) {
3427 mp = TAILQ_LAST(&mountlist, mntlist);
3428 error = dounmount(mp, MNT_FORCE, td);
3430 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3432 * XXX: Due to the way in which we mount the root
3433 * file system off of devfs, devfs will generate a
3434 * "busy" warning when we try to unmount it before
3435 * the root. Don't print a warning as a result in
3436 * order to avoid false positive errors that may
3437 * cause needless upset.
3439 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3440 printf("unmount of %s failed (",
3441 mp->mnt_stat.f_mntonname);
3445 printf("%d)\n", error);
3448 /* The unmount has removed mp from the mountlist */
3454 * perform msync on all vnodes under a mount point
3455 * the mount point must be locked.
3458 vfs_msync(struct mount *mp, int flags)
3460 struct vnode *vp, *mvp;
3461 struct vm_object *obj;
3463 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3464 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3466 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3467 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3469 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3471 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3478 VM_OBJECT_WLOCK(obj);
3479 vm_object_page_clean(obj, 0, 0,
3481 OBJPC_SYNC : OBJPC_NOSYNC);
3482 VM_OBJECT_WUNLOCK(obj);
3492 destroy_vpollinfo_free(struct vpollinfo *vi)
3495 knlist_destroy(&vi->vpi_selinfo.si_note);
3496 mtx_destroy(&vi->vpi_lock);
3497 uma_zfree(vnodepoll_zone, vi);
3501 destroy_vpollinfo(struct vpollinfo *vi)
3504 knlist_clear(&vi->vpi_selinfo.si_note, 1);
3505 seldrain(&vi->vpi_selinfo);
3506 destroy_vpollinfo_free(vi);
3510 * Initalize per-vnode helper structure to hold poll-related state.
3513 v_addpollinfo(struct vnode *vp)
3515 struct vpollinfo *vi;
3517 if (vp->v_pollinfo != NULL)
3519 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3520 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3521 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3522 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3524 if (vp->v_pollinfo != NULL) {
3526 destroy_vpollinfo_free(vi);
3529 vp->v_pollinfo = vi;
3534 * Record a process's interest in events which might happen to
3535 * a vnode. Because poll uses the historic select-style interface
3536 * internally, this routine serves as both the ``check for any
3537 * pending events'' and the ``record my interest in future events''
3538 * functions. (These are done together, while the lock is held,
3539 * to avoid race conditions.)
3542 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3546 mtx_lock(&vp->v_pollinfo->vpi_lock);
3547 if (vp->v_pollinfo->vpi_revents & events) {
3549 * This leaves events we are not interested
3550 * in available for the other process which
3551 * which presumably had requested them
3552 * (otherwise they would never have been
3555 events &= vp->v_pollinfo->vpi_revents;
3556 vp->v_pollinfo->vpi_revents &= ~events;
3558 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3561 vp->v_pollinfo->vpi_events |= events;
3562 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3563 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3568 * Routine to create and manage a filesystem syncer vnode.
3570 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3571 static int sync_fsync(struct vop_fsync_args *);
3572 static int sync_inactive(struct vop_inactive_args *);
3573 static int sync_reclaim(struct vop_reclaim_args *);
3575 static struct vop_vector sync_vnodeops = {
3576 .vop_bypass = VOP_EOPNOTSUPP,
3577 .vop_close = sync_close, /* close */
3578 .vop_fsync = sync_fsync, /* fsync */
3579 .vop_inactive = sync_inactive, /* inactive */
3580 .vop_reclaim = sync_reclaim, /* reclaim */
3581 .vop_lock1 = vop_stdlock, /* lock */
3582 .vop_unlock = vop_stdunlock, /* unlock */
3583 .vop_islocked = vop_stdislocked, /* islocked */
3587 * Create a new filesystem syncer vnode for the specified mount point.
3590 vfs_allocate_syncvnode(struct mount *mp)
3594 static long start, incr, next;
3597 /* Allocate a new vnode */
3598 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3600 panic("vfs_allocate_syncvnode: getnewvnode() failed");
3602 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3603 vp->v_vflag |= VV_FORCEINSMQ;
3604 error = insmntque(vp, mp);
3606 panic("vfs_allocate_syncvnode: insmntque() failed");
3607 vp->v_vflag &= ~VV_FORCEINSMQ;
3610 * Place the vnode onto the syncer worklist. We attempt to
3611 * scatter them about on the list so that they will go off
3612 * at evenly distributed times even if all the filesystems
3613 * are mounted at once.
3616 if (next == 0 || next > syncer_maxdelay) {
3620 start = syncer_maxdelay / 2;
3621 incr = syncer_maxdelay;
3627 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3628 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3629 mtx_lock(&sync_mtx);
3631 if (mp->mnt_syncer == NULL) {
3632 mp->mnt_syncer = vp;
3635 mtx_unlock(&sync_mtx);
3638 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3645 vfs_deallocate_syncvnode(struct mount *mp)
3649 mtx_lock(&sync_mtx);
3650 vp = mp->mnt_syncer;
3652 mp->mnt_syncer = NULL;
3653 mtx_unlock(&sync_mtx);
3659 * Do a lazy sync of the filesystem.
3662 sync_fsync(struct vop_fsync_args *ap)
3664 struct vnode *syncvp = ap->a_vp;
3665 struct mount *mp = syncvp->v_mount;
3670 * We only need to do something if this is a lazy evaluation.
3672 if (ap->a_waitfor != MNT_LAZY)
3676 * Move ourselves to the back of the sync list.
3678 bo = &syncvp->v_bufobj;
3680 vn_syncer_add_to_worklist(bo, syncdelay);
3684 * Walk the list of vnodes pushing all that are dirty and
3685 * not already on the sync list.
3687 if (vfs_busy(mp, MBF_NOWAIT) != 0)
3689 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3693 save = curthread_pflags_set(TDP_SYNCIO);
3694 vfs_msync(mp, MNT_NOWAIT);
3695 error = VFS_SYNC(mp, MNT_LAZY);
3696 curthread_pflags_restore(save);
3697 vn_finished_write(mp);
3703 * The syncer vnode is no referenced.
3706 sync_inactive(struct vop_inactive_args *ap)
3714 * The syncer vnode is no longer needed and is being decommissioned.
3716 * Modifications to the worklist must be protected by sync_mtx.
3719 sync_reclaim(struct vop_reclaim_args *ap)
3721 struct vnode *vp = ap->a_vp;
3726 mtx_lock(&sync_mtx);
3727 if (vp->v_mount->mnt_syncer == vp)
3728 vp->v_mount->mnt_syncer = NULL;
3729 if (bo->bo_flag & BO_ONWORKLST) {
3730 LIST_REMOVE(bo, bo_synclist);
3731 syncer_worklist_len--;
3733 bo->bo_flag &= ~BO_ONWORKLST;
3735 mtx_unlock(&sync_mtx);
3742 * Check if vnode represents a disk device
3745 vn_isdisk(struct vnode *vp, int *errp)
3751 if (vp->v_type != VCHR)
3753 else if (vp->v_rdev == NULL)
3755 else if (vp->v_rdev->si_devsw == NULL)
3757 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3762 return (error == 0);
3766 * Common filesystem object access control check routine. Accepts a
3767 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3768 * and optional call-by-reference privused argument allowing vaccess()
3769 * to indicate to the caller whether privilege was used to satisfy the
3770 * request (obsoleted). Returns 0 on success, or an errno on failure.
3773 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3774 accmode_t accmode, struct ucred *cred, int *privused)
3776 accmode_t dac_granted;
3777 accmode_t priv_granted;
3779 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3780 ("invalid bit in accmode"));
3781 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3782 ("VAPPEND without VWRITE"));
3785 * Look for a normal, non-privileged way to access the file/directory
3786 * as requested. If it exists, go with that.
3789 if (privused != NULL)
3794 /* Check the owner. */
3795 if (cred->cr_uid == file_uid) {
3796 dac_granted |= VADMIN;
3797 if (file_mode & S_IXUSR)
3798 dac_granted |= VEXEC;
3799 if (file_mode & S_IRUSR)
3800 dac_granted |= VREAD;
3801 if (file_mode & S_IWUSR)
3802 dac_granted |= (VWRITE | VAPPEND);
3804 if ((accmode & dac_granted) == accmode)
3810 /* Otherwise, check the groups (first match) */
3811 if (groupmember(file_gid, cred)) {
3812 if (file_mode & S_IXGRP)
3813 dac_granted |= VEXEC;
3814 if (file_mode & S_IRGRP)
3815 dac_granted |= VREAD;
3816 if (file_mode & S_IWGRP)
3817 dac_granted |= (VWRITE | VAPPEND);
3819 if ((accmode & dac_granted) == accmode)
3825 /* Otherwise, check everyone else. */
3826 if (file_mode & S_IXOTH)
3827 dac_granted |= VEXEC;
3828 if (file_mode & S_IROTH)
3829 dac_granted |= VREAD;
3830 if (file_mode & S_IWOTH)
3831 dac_granted |= (VWRITE | VAPPEND);
3832 if ((accmode & dac_granted) == accmode)
3837 * Build a privilege mask to determine if the set of privileges
3838 * satisfies the requirements when combined with the granted mask
3839 * from above. For each privilege, if the privilege is required,
3840 * bitwise or the request type onto the priv_granted mask.
3846 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3847 * requests, instead of PRIV_VFS_EXEC.
3849 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3850 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3851 priv_granted |= VEXEC;
3854 * Ensure that at least one execute bit is on. Otherwise,
3855 * a privileged user will always succeed, and we don't want
3856 * this to happen unless the file really is executable.
3858 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3859 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3860 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3861 priv_granted |= VEXEC;
3864 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3865 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3866 priv_granted |= VREAD;
3868 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3869 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3870 priv_granted |= (VWRITE | VAPPEND);
3872 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3873 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3874 priv_granted |= VADMIN;
3876 if ((accmode & (priv_granted | dac_granted)) == accmode) {
3877 /* XXX audit: privilege used */
3878 if (privused != NULL)
3883 return ((accmode & VADMIN) ? EPERM : EACCES);
3887 * Credential check based on process requesting service, and per-attribute
3891 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3892 struct thread *td, accmode_t accmode)
3896 * Kernel-invoked always succeeds.
3902 * Do not allow privileged processes in jail to directly manipulate
3903 * system attributes.
3905 switch (attrnamespace) {
3906 case EXTATTR_NAMESPACE_SYSTEM:
3907 /* Potentially should be: return (EPERM); */
3908 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3909 case EXTATTR_NAMESPACE_USER:
3910 return (VOP_ACCESS(vp, accmode, cred, td));
3916 #ifdef DEBUG_VFS_LOCKS
3918 * This only exists to supress warnings from unlocked specfs accesses. It is
3919 * no longer ok to have an unlocked VFS.
3921 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
3922 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
3924 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3925 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3926 "Drop into debugger on lock violation");
3928 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3929 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3930 0, "Check for interlock across VOPs");
3932 int vfs_badlock_print = 1; /* Print lock violations. */
3933 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3934 0, "Print lock violations");
3937 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3938 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3939 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3943 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3947 if (vfs_badlock_backtrace)
3950 if (vfs_badlock_print)
3951 printf("%s: %p %s\n", str, (void *)vp, msg);
3952 if (vfs_badlock_ddb)
3953 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3957 assert_vi_locked(struct vnode *vp, const char *str)
3960 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3961 vfs_badlock("interlock is not locked but should be", str, vp);
3965 assert_vi_unlocked(struct vnode *vp, const char *str)
3968 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3969 vfs_badlock("interlock is locked but should not be", str, vp);
3973 assert_vop_locked(struct vnode *vp, const char *str)
3977 if (!IGNORE_LOCK(vp)) {
3978 locked = VOP_ISLOCKED(vp);
3979 if (locked == 0 || locked == LK_EXCLOTHER)
3980 vfs_badlock("is not locked but should be", str, vp);
3985 assert_vop_unlocked(struct vnode *vp, const char *str)
3988 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3989 vfs_badlock("is locked but should not be", str, vp);
3993 assert_vop_elocked(struct vnode *vp, const char *str)
3996 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3997 vfs_badlock("is not exclusive locked but should be", str, vp);
4002 assert_vop_elocked_other(struct vnode *vp, const char *str)
4005 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
4006 vfs_badlock("is not exclusive locked by another thread",
4011 assert_vop_slocked(struct vnode *vp, const char *str)
4014 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
4015 vfs_badlock("is not locked shared but should be", str, vp);
4018 #endif /* DEBUG_VFS_LOCKS */
4021 vop_rename_fail(struct vop_rename_args *ap)
4024 if (ap->a_tvp != NULL)
4026 if (ap->a_tdvp == ap->a_tvp)
4035 vop_rename_pre(void *ap)
4037 struct vop_rename_args *a = ap;
4039 #ifdef DEBUG_VFS_LOCKS
4041 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4042 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4043 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4044 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4046 /* Check the source (from). */
4047 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4048 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4049 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4050 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4051 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4053 /* Check the target. */
4055 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4056 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4058 if (a->a_tdvp != a->a_fdvp)
4060 if (a->a_tvp != a->a_fvp)
4068 vop_strategy_pre(void *ap)
4070 #ifdef DEBUG_VFS_LOCKS
4071 struct vop_strategy_args *a;
4078 * Cluster ops lock their component buffers but not the IO container.
4080 if ((bp->b_flags & B_CLUSTER) != 0)
4083 if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4084 if (vfs_badlock_print)
4086 "VOP_STRATEGY: bp is not locked but should be\n");
4087 if (vfs_badlock_ddb)
4088 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4094 vop_lock_pre(void *ap)
4096 #ifdef DEBUG_VFS_LOCKS
4097 struct vop_lock1_args *a = ap;
4099 if ((a->a_flags & LK_INTERLOCK) == 0)
4100 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4102 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4107 vop_lock_post(void *ap, int rc)
4109 #ifdef DEBUG_VFS_LOCKS
4110 struct vop_lock1_args *a = ap;
4112 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4113 if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
4114 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4119 vop_unlock_pre(void *ap)
4121 #ifdef DEBUG_VFS_LOCKS
4122 struct vop_unlock_args *a = ap;
4124 if (a->a_flags & LK_INTERLOCK)
4125 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4126 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4131 vop_unlock_post(void *ap, int rc)
4133 #ifdef DEBUG_VFS_LOCKS
4134 struct vop_unlock_args *a = ap;
4136 if (a->a_flags & LK_INTERLOCK)
4137 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4142 vop_create_post(void *ap, int rc)
4144 struct vop_create_args *a = ap;
4147 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4151 vop_deleteextattr_post(void *ap, int rc)
4153 struct vop_deleteextattr_args *a = ap;
4156 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4160 vop_link_post(void *ap, int rc)
4162 struct vop_link_args *a = ap;
4165 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4166 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4171 vop_mkdir_post(void *ap, int rc)
4173 struct vop_mkdir_args *a = ap;
4176 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4180 vop_mknod_post(void *ap, int rc)
4182 struct vop_mknod_args *a = ap;
4185 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4189 vop_remove_post(void *ap, int rc)
4191 struct vop_remove_args *a = ap;
4194 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4195 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4200 vop_rename_post(void *ap, int rc)
4202 struct vop_rename_args *a = ap;
4205 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4206 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4207 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4209 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4211 if (a->a_tdvp != a->a_fdvp)
4213 if (a->a_tvp != a->a_fvp)
4221 vop_rmdir_post(void *ap, int rc)
4223 struct vop_rmdir_args *a = ap;
4226 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4227 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4232 vop_setattr_post(void *ap, int rc)
4234 struct vop_setattr_args *a = ap;
4237 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4241 vop_setextattr_post(void *ap, int rc)
4243 struct vop_setextattr_args *a = ap;
4246 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4250 vop_symlink_post(void *ap, int rc)
4252 struct vop_symlink_args *a = ap;
4255 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4258 static struct knlist fs_knlist;
4261 vfs_event_init(void *arg)
4263 knlist_init_mtx(&fs_knlist, NULL);
4265 /* XXX - correct order? */
4266 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4269 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4272 KNOTE_UNLOCKED(&fs_knlist, event);
4275 static int filt_fsattach(struct knote *kn);
4276 static void filt_fsdetach(struct knote *kn);
4277 static int filt_fsevent(struct knote *kn, long hint);
4279 struct filterops fs_filtops = {
4281 .f_attach = filt_fsattach,
4282 .f_detach = filt_fsdetach,
4283 .f_event = filt_fsevent
4287 filt_fsattach(struct knote *kn)
4290 kn->kn_flags |= EV_CLEAR;
4291 knlist_add(&fs_knlist, kn, 0);
4296 filt_fsdetach(struct knote *kn)
4299 knlist_remove(&fs_knlist, kn, 0);
4303 filt_fsevent(struct knote *kn, long hint)
4306 kn->kn_fflags |= hint;
4307 return (kn->kn_fflags != 0);
4311 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4317 error = SYSCTL_IN(req, &vc, sizeof(vc));
4320 if (vc.vc_vers != VFS_CTL_VERS1)
4322 mp = vfs_getvfs(&vc.vc_fsid);
4325 /* ensure that a specific sysctl goes to the right filesystem. */
4326 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4327 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4331 VCTLTOREQ(&vc, req);
4332 error = VFS_SYSCTL(mp, vc.vc_op, req);
4337 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4338 NULL, 0, sysctl_vfs_ctl, "",
4342 * Function to initialize a va_filerev field sensibly.
4343 * XXX: Wouldn't a random number make a lot more sense ??
4346 init_va_filerev(void)
4351 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4354 static int filt_vfsread(struct knote *kn, long hint);
4355 static int filt_vfswrite(struct knote *kn, long hint);
4356 static int filt_vfsvnode(struct knote *kn, long hint);
4357 static void filt_vfsdetach(struct knote *kn);
4358 static struct filterops vfsread_filtops = {
4360 .f_detach = filt_vfsdetach,
4361 .f_event = filt_vfsread
4363 static struct filterops vfswrite_filtops = {
4365 .f_detach = filt_vfsdetach,
4366 .f_event = filt_vfswrite
4368 static struct filterops vfsvnode_filtops = {
4370 .f_detach = filt_vfsdetach,
4371 .f_event = filt_vfsvnode
4375 vfs_knllock(void *arg)
4377 struct vnode *vp = arg;
4379 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4383 vfs_knlunlock(void *arg)
4385 struct vnode *vp = arg;
4391 vfs_knl_assert_locked(void *arg)
4393 #ifdef DEBUG_VFS_LOCKS
4394 struct vnode *vp = arg;
4396 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4401 vfs_knl_assert_unlocked(void *arg)
4403 #ifdef DEBUG_VFS_LOCKS
4404 struct vnode *vp = arg;
4406 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4411 vfs_kqfilter(struct vop_kqfilter_args *ap)
4413 struct vnode *vp = ap->a_vp;
4414 struct knote *kn = ap->a_kn;
4417 switch (kn->kn_filter) {
4419 kn->kn_fop = &vfsread_filtops;
4422 kn->kn_fop = &vfswrite_filtops;
4425 kn->kn_fop = &vfsvnode_filtops;
4431 kn->kn_hook = (caddr_t)vp;
4434 if (vp->v_pollinfo == NULL)
4436 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4438 knlist_add(knl, kn, 0);
4444 * Detach knote from vnode
4447 filt_vfsdetach(struct knote *kn)
4449 struct vnode *vp = (struct vnode *)kn->kn_hook;
4451 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4452 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4458 filt_vfsread(struct knote *kn, long hint)
4460 struct vnode *vp = (struct vnode *)kn->kn_hook;
4465 * filesystem is gone, so set the EOF flag and schedule
4466 * the knote for deletion.
4468 if (hint == NOTE_REVOKE) {
4470 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4475 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4479 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4480 res = (kn->kn_data != 0);
4487 filt_vfswrite(struct knote *kn, long hint)
4489 struct vnode *vp = (struct vnode *)kn->kn_hook;
4494 * filesystem is gone, so set the EOF flag and schedule
4495 * the knote for deletion.
4497 if (hint == NOTE_REVOKE)
4498 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4506 filt_vfsvnode(struct knote *kn, long hint)
4508 struct vnode *vp = (struct vnode *)kn->kn_hook;
4512 if (kn->kn_sfflags & hint)
4513 kn->kn_fflags |= hint;
4514 if (hint == NOTE_REVOKE) {
4515 kn->kn_flags |= EV_EOF;
4519 res = (kn->kn_fflags != 0);
4525 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4529 if (dp->d_reclen > ap->a_uio->uio_resid)
4530 return (ENAMETOOLONG);
4531 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4533 if (ap->a_ncookies != NULL) {
4534 if (ap->a_cookies != NULL)
4535 free(ap->a_cookies, M_TEMP);
4536 ap->a_cookies = NULL;
4537 *ap->a_ncookies = 0;
4541 if (ap->a_ncookies == NULL)
4544 KASSERT(ap->a_cookies,
4545 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4547 *ap->a_cookies = realloc(*ap->a_cookies,
4548 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4549 (*ap->a_cookies)[*ap->a_ncookies] = off;
4554 * Mark for update the access time of the file if the filesystem
4555 * supports VOP_MARKATIME. This functionality is used by execve and
4556 * mmap, so we want to avoid the I/O implied by directly setting
4557 * va_atime for the sake of efficiency.
4560 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4565 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4566 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4567 (void)VOP_MARKATIME(vp);
4571 * The purpose of this routine is to remove granularity from accmode_t,
4572 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4573 * VADMIN and VAPPEND.
4575 * If it returns 0, the caller is supposed to continue with the usual
4576 * access checks using 'accmode' as modified by this routine. If it
4577 * returns nonzero value, the caller is supposed to return that value
4580 * Note that after this routine runs, accmode may be zero.
4583 vfs_unixify_accmode(accmode_t *accmode)
4586 * There is no way to specify explicit "deny" rule using
4587 * file mode or POSIX.1e ACLs.
4589 if (*accmode & VEXPLICIT_DENY) {
4595 * None of these can be translated into usual access bits.
4596 * Also, the common case for NFSv4 ACLs is to not contain
4597 * either of these bits. Caller should check for VWRITE
4598 * on the containing directory instead.
4600 if (*accmode & (VDELETE_CHILD | VDELETE))
4603 if (*accmode & VADMIN_PERMS) {
4604 *accmode &= ~VADMIN_PERMS;
4609 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4610 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4612 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4618 * These are helper functions for filesystems to traverse all
4619 * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4621 * This interface replaces MNT_VNODE_FOREACH.
4624 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4627 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4632 kern_yield(PRI_USER);
4634 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4635 vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4636 while (vp != NULL && (vp->v_type == VMARKER ||
4637 (vp->v_iflag & VI_DOOMED) != 0))
4638 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4640 /* Check if we are done */
4642 __mnt_vnode_markerfree_all(mvp, mp);
4643 /* MNT_IUNLOCK(mp); -- done in above function */
4644 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4647 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4648 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4655 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4659 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4662 (*mvp)->v_type = VMARKER;
4664 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4665 while (vp != NULL && (vp->v_type == VMARKER ||
4666 (vp->v_iflag & VI_DOOMED) != 0))
4667 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4669 /* Check if we are done */
4673 free(*mvp, M_VNODE_MARKER);
4677 (*mvp)->v_mount = mp;
4678 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4686 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4694 mtx_assert(MNT_MTX(mp), MA_OWNED);
4696 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4697 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4700 free(*mvp, M_VNODE_MARKER);
4705 * These are helper functions for filesystems to traverse their
4706 * active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4709 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4712 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4717 free(*mvp, M_VNODE_MARKER);
4721 static struct vnode *
4722 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4724 struct vnode *vp, *nvp;
4726 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
4727 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4729 vp = TAILQ_NEXT(*mvp, v_actfreelist);
4730 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4731 while (vp != NULL) {
4732 if (vp->v_type == VMARKER) {
4733 vp = TAILQ_NEXT(vp, v_actfreelist);
4736 if (!VI_TRYLOCK(vp)) {
4737 if (mp_ncpus == 1 || should_yield()) {
4738 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4739 mtx_unlock(&vnode_free_list_mtx);
4741 mtx_lock(&vnode_free_list_mtx);
4746 KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
4747 KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
4748 ("alien vnode on the active list %p %p", vp, mp));
4749 if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
4751 nvp = TAILQ_NEXT(vp, v_actfreelist);
4756 /* Check if we are done */
4758 mtx_unlock(&vnode_free_list_mtx);
4759 mnt_vnode_markerfree_active(mvp, mp);
4762 TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4763 mtx_unlock(&vnode_free_list_mtx);
4764 ASSERT_VI_LOCKED(vp, "active iter");
4765 KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
4770 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4774 kern_yield(PRI_USER);
4775 mtx_lock(&vnode_free_list_mtx);
4776 return (mnt_vnode_next_active(mvp, mp));
4780 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4784 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4788 (*mvp)->v_type = VMARKER;
4789 (*mvp)->v_mount = mp;
4791 mtx_lock(&vnode_free_list_mtx);
4792 vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
4794 mtx_unlock(&vnode_free_list_mtx);
4795 mnt_vnode_markerfree_active(mvp, mp);
4798 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4799 return (mnt_vnode_next_active(mvp, mp));
4803 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4809 mtx_lock(&vnode_free_list_mtx);
4810 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4811 mtx_unlock(&vnode_free_list_mtx);
4812 mnt_vnode_markerfree_active(mvp, mp);