2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
38 * External virtual filesystem routines
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
46 #include <sys/param.h>
47 #include <sys/systm.h>
50 #include <sys/condvar.h>
52 #include <sys/dirent.h>
53 #include <sys/event.h>
54 #include <sys/eventhandler.h>
55 #include <sys/extattr.h>
57 #include <sys/fcntl.h>
60 #include <sys/kernel.h>
61 #include <sys/kthread.h>
62 #include <sys/lockf.h>
63 #include <sys/malloc.h>
64 #include <sys/mount.h>
65 #include <sys/namei.h>
67 #include <sys/reboot.h>
68 #include <sys/sched.h>
69 #include <sys/sleepqueue.h>
71 #include <sys/sysctl.h>
72 #include <sys/syslog.h>
73 #include <sys/vmmeter.h>
74 #include <sys/vnode.h>
76 #include <machine/stdarg.h>
78 #include <security/mac/mac_framework.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_extern.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_page.h>
86 #include <vm/vm_kern.h>
96 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure");
98 static void delmntque(struct vnode *vp);
99 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
100 int slpflag, int slptimeo);
101 static void syncer_shutdown(void *arg, int howto);
102 static int vtryrecycle(struct vnode *vp);
103 static void vbusy(struct vnode *vp);
104 static void vinactive(struct vnode *, struct thread *);
105 static void v_incr_usecount(struct vnode *);
106 static void v_decr_usecount(struct vnode *);
107 static void v_decr_useonly(struct vnode *);
108 static void v_upgrade_usecount(struct vnode *);
109 static void vfree(struct vnode *);
110 static void vnlru_free(int);
111 static void vgonel(struct vnode *);
112 static void vfs_knllock(void *arg);
113 static void vfs_knlunlock(void *arg);
114 static void vfs_knl_assert_locked(void *arg);
115 static void vfs_knl_assert_unlocked(void *arg);
116 static void destroy_vpollinfo(struct vpollinfo *vi);
119 * Number of vnodes in existence. Increased whenever getnewvnode()
120 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
123 static unsigned long numvnodes;
125 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
126 "Number of vnodes in existence");
129 * Conversion tables for conversion from vnode types to inode formats
132 enum vtype iftovt_tab[16] = {
133 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
134 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
136 int vttoif_tab[10] = {
137 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
138 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
142 * List of vnodes that are ready for recycling.
144 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
147 * Free vnode target. Free vnodes may simply be files which have been stat'd
148 * but not read. This is somewhat common, and a small cache of such files
149 * should be kept to avoid recreation costs.
151 static u_long wantfreevnodes;
152 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
153 /* Number of vnodes in the free list. */
154 static u_long freevnodes;
155 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
156 "Number of vnodes in the free list");
158 static int vlru_allow_cache_src;
159 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
160 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
163 * Various variables used for debugging the new implementation of
165 * XXX these are probably of (very) limited utility now.
167 static int reassignbufcalls;
168 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
169 "Number of calls to reassignbuf");
172 * Cache for the mount type id assigned to NFS. This is used for
173 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
175 int nfs_mount_type = -1;
177 /* To keep more than one thread at a time from running vfs_getnewfsid */
178 static struct mtx mntid_mtx;
181 * Lock for any access to the following:
186 static struct mtx vnode_free_list_mtx;
188 /* Publicly exported FS */
189 struct nfs_public nfs_pub;
191 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
192 static uma_zone_t vnode_zone;
193 static uma_zone_t vnodepoll_zone;
195 /* Set to 1 to print out reclaim of active vnodes */
199 * The workitem queue.
201 * It is useful to delay writes of file data and filesystem metadata
202 * for tens of seconds so that quickly created and deleted files need
203 * not waste disk bandwidth being created and removed. To realize this,
204 * we append vnodes to a "workitem" queue. When running with a soft
205 * updates implementation, most pending metadata dependencies should
206 * not wait for more than a few seconds. Thus, mounted on block devices
207 * are delayed only about a half the time that file data is delayed.
208 * Similarly, directory updates are more critical, so are only delayed
209 * about a third the time that file data is delayed. Thus, there are
210 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
211 * one each second (driven off the filesystem syncer process). The
212 * syncer_delayno variable indicates the next queue that is to be processed.
213 * Items that need to be processed soon are placed in this queue:
215 * syncer_workitem_pending[syncer_delayno]
217 * A delay of fifteen seconds is done by placing the request fifteen
218 * entries later in the queue:
220 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
223 static int syncer_delayno;
224 static long syncer_mask;
225 LIST_HEAD(synclist, bufobj);
226 static struct synclist *syncer_workitem_pending[2];
228 * The sync_mtx protects:
233 * syncer_workitem_pending
234 * syncer_worklist_len
237 static struct mtx sync_mtx;
238 static struct cv sync_wakeup;
240 #define SYNCER_MAXDELAY 32
241 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
242 static int syncdelay = 30; /* max time to delay syncing data */
243 static int filedelay = 30; /* time to delay syncing files */
244 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
245 "Time to delay syncing files (in seconds)");
246 static int dirdelay = 29; /* time to delay syncing directories */
247 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
248 "Time to delay syncing directories (in seconds)");
249 static int metadelay = 28; /* time to delay syncing metadata */
250 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
251 "Time to delay syncing metadata (in seconds)");
252 static int rushjob; /* number of slots to run ASAP */
253 static int stat_rush_requests; /* number of times I/O speeded up */
254 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
255 "Number of times I/O speeded up (rush requests)");
258 * When shutting down the syncer, run it at four times normal speed.
260 #define SYNCER_SHUTDOWN_SPEEDUP 4
261 static int sync_vnode_count;
262 static int syncer_worklist_len;
263 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
267 * Number of vnodes we want to exist at any one time. This is mostly used
268 * to size hash tables in vnode-related code. It is normally not used in
269 * getnewvnode(), as wantfreevnodes is normally nonzero.)
271 * XXX desiredvnodes is historical cruft and should not exist.
274 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
275 &desiredvnodes, 0, "Maximum number of vnodes");
276 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
277 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
278 static int vnlru_nowhere;
279 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
280 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
283 * Macros to control when a vnode is freed and recycled. All require
284 * the vnode interlock.
286 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
287 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
288 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
292 * Initialize the vnode management data structures.
294 * Reevaluate the following cap on the number of vnodes after the physical
295 * memory size exceeds 512GB. In the limit, as the physical memory size
296 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
298 #ifndef MAXVNODES_MAX
299 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
302 vntblinit(void *dummy __unused)
304 int physvnodes, virtvnodes;
307 * Desiredvnodes is a function of the physical memory size and the
308 * kernel's heap size. Generally speaking, it scales with the
309 * physical memory size. The ratio of desiredvnodes to physical pages
310 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
311 * marginal ratio of desiredvnodes to physical pages is one to
312 * sixteen. However, desiredvnodes is limited by the kernel's heap
313 * size. The memory required by desiredvnodes vnodes and vm objects
314 * may not exceed one seventh of the kernel's heap size.
316 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
317 cnt.v_page_count) / 16;
318 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
319 sizeof(struct vnode)));
320 desiredvnodes = min(physvnodes, virtvnodes);
321 if (desiredvnodes > MAXVNODES_MAX) {
323 printf("Reducing kern.maxvnodes %d -> %d\n",
324 desiredvnodes, MAXVNODES_MAX);
325 desiredvnodes = MAXVNODES_MAX;
327 wantfreevnodes = desiredvnodes / 4;
328 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
329 TAILQ_INIT(&vnode_free_list);
330 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
331 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
332 NULL, NULL, UMA_ALIGN_PTR, 0);
333 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
334 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
336 * Initialize the filesystem syncer.
338 syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
340 syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
342 syncer_maxdelay = syncer_mask + 1;
343 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
344 cv_init(&sync_wakeup, "syncer");
346 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
350 * Mark a mount point as busy. Used to synchronize access and to delay
351 * unmounting. Eventually, mountlist_mtx is not released on failure.
354 vfs_busy(struct mount *mp, int flags)
357 MPASS((flags & ~MBF_MASK) == 0);
358 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
363 * If mount point is currenly being unmounted, sleep until the
364 * mount point fate is decided. If thread doing the unmounting fails,
365 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
366 * that this mount point has survived the unmount attempt and vfs_busy
367 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
368 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
369 * about to be really destroyed. vfs_busy needs to release its
370 * reference on the mount point in this case and return with ENOENT,
371 * telling the caller that mount mount it tried to busy is no longer
374 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
375 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
378 CTR1(KTR_VFS, "%s: failed busying before sleeping",
382 if (flags & MBF_MNTLSTLOCK)
383 mtx_unlock(&mountlist_mtx);
384 mp->mnt_kern_flag |= MNTK_MWAIT;
385 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
386 if (flags & MBF_MNTLSTLOCK)
387 mtx_lock(&mountlist_mtx);
389 if (flags & MBF_MNTLSTLOCK)
390 mtx_unlock(&mountlist_mtx);
397 * Free a busy filesystem.
400 vfs_unbusy(struct mount *mp)
403 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
406 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
408 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
409 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
410 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
411 mp->mnt_kern_flag &= ~MNTK_DRAINING;
412 wakeup(&mp->mnt_lockref);
418 * Lookup a mount point by filesystem identifier.
421 vfs_getvfs(fsid_t *fsid)
425 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
426 mtx_lock(&mountlist_mtx);
427 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
428 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
429 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
431 mtx_unlock(&mountlist_mtx);
435 mtx_unlock(&mountlist_mtx);
436 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
437 return ((struct mount *) 0);
441 * Lookup a mount point by filesystem identifier, busying it before
445 vfs_busyfs(fsid_t *fsid)
450 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
451 mtx_lock(&mountlist_mtx);
452 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
453 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
454 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
455 error = vfs_busy(mp, MBF_MNTLSTLOCK);
457 mtx_unlock(&mountlist_mtx);
463 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
464 mtx_unlock(&mountlist_mtx);
465 return ((struct mount *) 0);
469 * Check if a user can access privileged mount options.
472 vfs_suser(struct mount *mp, struct thread *td)
477 * If the thread is jailed, but this is not a jail-friendly file
478 * system, deny immediately.
480 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
484 * If the file system was mounted outside the jail of the calling
485 * thread, deny immediately.
487 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
491 * If file system supports delegated administration, we don't check
492 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
493 * by the file system itself.
494 * If this is not the user that did original mount, we check for
495 * the PRIV_VFS_MOUNT_OWNER privilege.
497 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
498 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
499 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
506 * Get a new unique fsid. Try to make its val[0] unique, since this value
507 * will be used to create fake device numbers for stat(). Also try (but
508 * not so hard) make its val[0] unique mod 2^16, since some emulators only
509 * support 16-bit device numbers. We end up with unique val[0]'s for the
510 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
512 * Keep in mind that several mounts may be running in parallel. Starting
513 * the search one past where the previous search terminated is both a
514 * micro-optimization and a defense against returning the same fsid to
518 vfs_getnewfsid(struct mount *mp)
520 static u_int16_t mntid_base;
525 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
526 mtx_lock(&mntid_mtx);
527 mtype = mp->mnt_vfc->vfc_typenum;
528 tfsid.val[1] = mtype;
529 mtype = (mtype & 0xFF) << 24;
531 tfsid.val[0] = makedev(255,
532 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
534 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
538 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
539 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
540 mtx_unlock(&mntid_mtx);
544 * Knob to control the precision of file timestamps:
546 * 0 = seconds only; nanoseconds zeroed.
547 * 1 = seconds and nanoseconds, accurate within 1/HZ.
548 * 2 = seconds and nanoseconds, truncated to microseconds.
549 * >=3 = seconds and nanoseconds, maximum precision.
551 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
553 static int timestamp_precision = TSP_SEC;
554 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
555 ×tamp_precision, 0, "File timestamp precision (0: seconds, "
556 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
557 "3+: sec + ns (max. precision))");
560 * Get a current timestamp.
563 vfs_timestamp(struct timespec *tsp)
567 switch (timestamp_precision) {
569 tsp->tv_sec = time_second;
577 TIMEVAL_TO_TIMESPEC(&tv, tsp);
587 * Set vnode attributes to VNOVAL
590 vattr_null(struct vattr *vap)
594 vap->va_size = VNOVAL;
595 vap->va_bytes = VNOVAL;
596 vap->va_mode = VNOVAL;
597 vap->va_nlink = VNOVAL;
598 vap->va_uid = VNOVAL;
599 vap->va_gid = VNOVAL;
600 vap->va_fsid = VNOVAL;
601 vap->va_fileid = VNOVAL;
602 vap->va_blocksize = VNOVAL;
603 vap->va_rdev = VNOVAL;
604 vap->va_atime.tv_sec = VNOVAL;
605 vap->va_atime.tv_nsec = VNOVAL;
606 vap->va_mtime.tv_sec = VNOVAL;
607 vap->va_mtime.tv_nsec = VNOVAL;
608 vap->va_ctime.tv_sec = VNOVAL;
609 vap->va_ctime.tv_nsec = VNOVAL;
610 vap->va_birthtime.tv_sec = VNOVAL;
611 vap->va_birthtime.tv_nsec = VNOVAL;
612 vap->va_flags = VNOVAL;
613 vap->va_gen = VNOVAL;
618 * This routine is called when we have too many vnodes. It attempts
619 * to free <count> vnodes and will potentially free vnodes that still
620 * have VM backing store (VM backing store is typically the cause
621 * of a vnode blowout so we want to do this). Therefore, this operation
622 * is not considered cheap.
624 * A number of conditions may prevent a vnode from being reclaimed.
625 * the buffer cache may have references on the vnode, a directory
626 * vnode may still have references due to the namei cache representing
627 * underlying files, or the vnode may be in active use. It is not
628 * desireable to reuse such vnodes. These conditions may cause the
629 * number of vnodes to reach some minimum value regardless of what
630 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
633 vlrureclaim(struct mount *mp)
642 * Calculate the trigger point, don't allow user
643 * screwups to blow us up. This prevents us from
644 * recycling vnodes with lots of resident pages. We
645 * aren't trying to free memory, we are trying to
648 usevnodes = desiredvnodes;
651 trigger = cnt.v_page_count * 2 / usevnodes;
653 vn_start_write(NULL, &mp, V_WAIT);
655 count = mp->mnt_nvnodelistsize / 10 + 1;
657 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
658 while (vp != NULL && vp->v_type == VMARKER)
659 vp = TAILQ_NEXT(vp, v_nmntvnodes);
662 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
663 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
668 * If it's been deconstructed already, it's still
669 * referenced, or it exceeds the trigger, skip it.
671 if (vp->v_usecount ||
672 (!vlru_allow_cache_src &&
673 !LIST_EMPTY(&(vp)->v_cache_src)) ||
674 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
675 vp->v_object->resident_page_count > trigger)) {
681 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
683 goto next_iter_mntunlocked;
687 * v_usecount may have been bumped after VOP_LOCK() dropped
688 * the vnode interlock and before it was locked again.
690 * It is not necessary to recheck VI_DOOMED because it can
691 * only be set by another thread that holds both the vnode
692 * lock and vnode interlock. If another thread has the
693 * vnode lock before we get to VOP_LOCK() and obtains the
694 * vnode interlock after VOP_LOCK() drops the vnode
695 * interlock, the other thread will be unable to drop the
696 * vnode lock before our VOP_LOCK() call fails.
698 if (vp->v_usecount ||
699 (!vlru_allow_cache_src &&
700 !LIST_EMPTY(&(vp)->v_cache_src)) ||
701 (vp->v_object != NULL &&
702 vp->v_object->resident_page_count > trigger)) {
703 VOP_UNLOCK(vp, LK_INTERLOCK);
704 goto next_iter_mntunlocked;
706 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
707 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
712 next_iter_mntunlocked:
713 if ((count % 256) != 0)
717 if ((count % 256) != 0)
726 vn_finished_write(mp);
731 * Attempt to keep the free list at wantfreevnodes length.
734 vnlru_free(int count)
739 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
740 for (; count > 0; count--) {
741 vp = TAILQ_FIRST(&vnode_free_list);
743 * The list can be modified while the free_list_mtx
744 * has been dropped and vp could be NULL here.
748 VNASSERT(vp->v_op != NULL, vp,
749 ("vnlru_free: vnode already reclaimed."));
750 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
752 * Don't recycle if we can't get the interlock.
754 if (!VI_TRYLOCK(vp)) {
755 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
758 VNASSERT(VCANRECYCLE(vp), vp,
759 ("vp inconsistent on freelist"));
761 vp->v_iflag &= ~VI_FREE;
763 mtx_unlock(&vnode_free_list_mtx);
765 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
767 VFS_UNLOCK_GIANT(vfslocked);
769 * If the recycled succeeded this vdrop will actually free
770 * the vnode. If not it will simply place it back on
774 mtx_lock(&vnode_free_list_mtx);
778 * Attempt to recycle vnodes in a context that is always safe to block.
779 * Calling vlrurecycle() from the bowels of filesystem code has some
780 * interesting deadlock problems.
782 static struct proc *vnlruproc;
783 static int vnlruproc_sig;
788 struct mount *mp, *nmp;
790 struct proc *p = vnlruproc;
792 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
796 kproc_suspend_check(p);
797 mtx_lock(&vnode_free_list_mtx);
798 if (freevnodes > wantfreevnodes)
799 vnlru_free(freevnodes - wantfreevnodes);
800 if (numvnodes <= desiredvnodes * 9 / 10) {
802 wakeup(&vnlruproc_sig);
803 msleep(vnlruproc, &vnode_free_list_mtx,
804 PVFS|PDROP, "vlruwt", hz);
807 mtx_unlock(&vnode_free_list_mtx);
809 mtx_lock(&mountlist_mtx);
810 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
811 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
812 nmp = TAILQ_NEXT(mp, mnt_list);
815 vfslocked = VFS_LOCK_GIANT(mp);
816 done += vlrureclaim(mp);
817 VFS_UNLOCK_GIANT(vfslocked);
818 mtx_lock(&mountlist_mtx);
819 nmp = TAILQ_NEXT(mp, mnt_list);
822 mtx_unlock(&mountlist_mtx);
825 /* These messages are temporary debugging aids */
826 if (vnlru_nowhere < 5)
827 printf("vnlru process getting nowhere..\n");
828 else if (vnlru_nowhere == 5)
829 printf("vnlru process messages stopped.\n");
832 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
838 static struct kproc_desc vnlru_kp = {
843 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
847 * Routines having to do with the management of the vnode table.
851 vdestroy(struct vnode *vp)
855 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
856 mtx_lock(&vnode_free_list_mtx);
858 mtx_unlock(&vnode_free_list_mtx);
860 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
861 ("cleaned vnode still on the free list."));
862 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
863 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
864 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
865 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
866 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
867 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
868 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
869 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
870 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
871 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
872 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
873 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
876 mac_vnode_destroy(vp);
878 if (vp->v_pollinfo != NULL)
879 destroy_vpollinfo(vp->v_pollinfo);
881 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
884 lockdestroy(vp->v_vnlock);
885 mtx_destroy(&vp->v_interlock);
886 mtx_destroy(BO_MTX(bo));
887 uma_zfree(vnode_zone, vp);
891 * Try to recycle a freed vnode. We abort if anyone picks up a reference
892 * before we actually vgone(). This function must be called with the vnode
893 * held to prevent the vnode from being returned to the free list midway
897 vtryrecycle(struct vnode *vp)
901 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
902 VNASSERT(vp->v_holdcnt, vp,
903 ("vtryrecycle: Recycling vp %p without a reference.", vp));
905 * This vnode may found and locked via some other list, if so we
906 * can't recycle it yet.
908 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
910 "%s: impossible to recycle, vp %p lock is already held",
912 return (EWOULDBLOCK);
915 * Don't recycle if its filesystem is being suspended.
917 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
920 "%s: impossible to recycle, cannot start the write for %p",
925 * If we got this far, we need to acquire the interlock and see if
926 * anyone picked up this vnode from another list. If not, we will
927 * mark it with DOOMED via vgonel() so that anyone who does find it
931 if (vp->v_usecount) {
932 VOP_UNLOCK(vp, LK_INTERLOCK);
933 vn_finished_write(vnmp);
935 "%s: impossible to recycle, %p is already referenced",
939 if ((vp->v_iflag & VI_DOOMED) == 0)
941 VOP_UNLOCK(vp, LK_INTERLOCK);
942 vn_finished_write(vnmp);
947 * Return the next vnode from the free list.
950 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
953 struct vnode *vp = NULL;
956 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
957 mtx_lock(&vnode_free_list_mtx);
959 * Lend our context to reclaim vnodes if they've exceeded the max.
961 if (freevnodes > wantfreevnodes)
964 * Wait for available vnodes.
966 if (numvnodes > desiredvnodes) {
967 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
969 * File system is beeing suspended, we cannot risk a
970 * deadlock here, so allocate new vnode anyway.
972 if (freevnodes > wantfreevnodes)
973 vnlru_free(freevnodes - wantfreevnodes);
976 if (vnlruproc_sig == 0) {
977 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
980 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
982 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
983 if (numvnodes > desiredvnodes) {
984 mtx_unlock(&vnode_free_list_mtx);
991 mtx_unlock(&vnode_free_list_mtx);
992 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
996 vp->v_vnlock = &vp->v_lock;
997 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
999 * By default, don't allow shared locks unless filesystems
1002 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
1004 * Initialize bufobj.
1007 bo->__bo_vnode = vp;
1008 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
1009 bo->bo_ops = &buf_ops_bio;
1010 bo->bo_private = vp;
1011 TAILQ_INIT(&bo->bo_clean.bv_hd);
1012 TAILQ_INIT(&bo->bo_dirty.bv_hd);
1014 * Initialize namecache.
1016 LIST_INIT(&vp->v_cache_src);
1017 TAILQ_INIT(&vp->v_cache_dst);
1019 * Finalize various vnode identity bits.
1024 v_incr_usecount(vp);
1028 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1029 mac_vnode_associate_singlelabel(mp, vp);
1030 else if (mp == NULL && vops != &dead_vnodeops)
1031 printf("NULL mp in getnewvnode()\n");
1034 bo->bo_bsize = mp->mnt_stat.f_iosize;
1035 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1036 vp->v_vflag |= VV_NOKNOTE;
1044 * Delete from old mount point vnode list, if on one.
1047 delmntque(struct vnode *vp)
1056 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1057 ("bad mount point vnode list size"));
1058 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1059 mp->mnt_nvnodelistsize--;
1065 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1069 vp->v_op = &dead_vnodeops;
1070 /* XXX non mp-safe fs may still call insmntque with vnode
1072 if (!VOP_ISLOCKED(vp))
1073 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1079 * Insert into list of vnodes for the new mount point, if available.
1082 insmntque1(struct vnode *vp, struct mount *mp,
1083 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1087 KASSERT(vp->v_mount == NULL,
1088 ("insmntque: vnode already on per mount vnode list"));
1089 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1090 #ifdef DEBUG_VFS_LOCKS
1091 if (!VFS_NEEDSGIANT(mp))
1092 ASSERT_VOP_ELOCKED(vp,
1093 "insmntque: mp-safe fs and non-locked vp");
1096 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1097 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1098 mp->mnt_nvnodelistsize == 0)) {
1099 locked = VOP_ISLOCKED(vp);
1100 if (!locked || (locked == LK_EXCLUSIVE &&
1101 (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1110 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1111 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1112 ("neg mount point vnode list size"));
1113 mp->mnt_nvnodelistsize++;
1119 insmntque(struct vnode *vp, struct mount *mp)
1122 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1126 * Flush out and invalidate all buffers associated with a bufobj
1127 * Called with the underlying object locked.
1130 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1135 if (flags & V_SAVE) {
1136 error = bufobj_wwait(bo, slpflag, slptimeo);
1141 if (bo->bo_dirty.bv_cnt > 0) {
1143 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1146 * XXX We could save a lock/unlock if this was only
1147 * enabled under INVARIANTS
1150 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1151 panic("vinvalbuf: dirty bufs");
1155 * If you alter this loop please notice that interlock is dropped and
1156 * reacquired in flushbuflist. Special care is needed to ensure that
1157 * no race conditions occur from this.
1160 error = flushbuflist(&bo->bo_clean,
1161 flags, bo, slpflag, slptimeo);
1163 error = flushbuflist(&bo->bo_dirty,
1164 flags, bo, slpflag, slptimeo);
1165 if (error != 0 && error != EAGAIN) {
1169 } while (error != 0);
1172 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1173 * have write I/O in-progress but if there is a VM object then the
1174 * VM object can also have read-I/O in-progress.
1177 bufobj_wwait(bo, 0, 0);
1179 if (bo->bo_object != NULL) {
1180 VM_OBJECT_LOCK(bo->bo_object);
1181 vm_object_pip_wait(bo->bo_object, "bovlbx");
1182 VM_OBJECT_UNLOCK(bo->bo_object);
1185 } while (bo->bo_numoutput > 0);
1189 * Destroy the copy in the VM cache, too.
1191 if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) {
1192 VM_OBJECT_LOCK(bo->bo_object);
1193 vm_object_page_remove(bo->bo_object, 0, 0,
1194 (flags & V_SAVE) ? TRUE : FALSE);
1195 VM_OBJECT_UNLOCK(bo->bo_object);
1200 if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1201 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1202 panic("vinvalbuf: flush failed");
1209 * Flush out and invalidate all buffers associated with a vnode.
1210 * Called with the underlying object locked.
1213 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1216 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1217 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1218 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1222 * Flush out buffers on the specified list.
1226 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1229 struct buf *bp, *nbp;
1234 ASSERT_BO_LOCKED(bo);
1237 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1238 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1239 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1245 lblkno = nbp->b_lblkno;
1246 xflags = nbp->b_xflags &
1247 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1250 error = BUF_TIMELOCK(bp,
1251 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1252 "flushbuf", slpflag, slptimeo);
1255 return (error != ENOLCK ? error : EAGAIN);
1257 KASSERT(bp->b_bufobj == bo,
1258 ("bp %p wrong b_bufobj %p should be %p",
1259 bp, bp->b_bufobj, bo));
1260 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1266 * XXX Since there are no node locks for NFS, I
1267 * believe there is a slight chance that a delayed
1268 * write will occur while sleeping just above, so
1271 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1276 bp->b_flags |= B_ASYNC;
1279 return (EAGAIN); /* XXX: why not loop ? */
1284 bp->b_flags |= (B_INVAL | B_RELBUF);
1285 bp->b_flags &= ~B_ASYNC;
1289 (nbp->b_bufobj != bo ||
1290 nbp->b_lblkno != lblkno ||
1292 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1293 break; /* nbp invalid */
1299 * Truncate a file's buffer and pages to a specified length. This
1300 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1304 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1305 off_t length, int blksize)
1307 struct buf *bp, *nbp;
1312 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1313 vp, cred, blksize, (uintmax_t)length);
1316 * Round up to the *next* lbn.
1318 trunclbn = (length + blksize - 1) / blksize;
1320 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1327 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1328 if (bp->b_lblkno < trunclbn)
1331 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1332 BO_MTX(bo)) == ENOLCK)
1338 bp->b_flags |= (B_INVAL | B_RELBUF);
1339 bp->b_flags &= ~B_ASYNC;
1344 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1345 (nbp->b_vp != vp) ||
1346 (nbp->b_flags & B_DELWRI))) {
1352 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1353 if (bp->b_lblkno < trunclbn)
1356 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1357 BO_MTX(bo)) == ENOLCK)
1362 bp->b_flags |= (B_INVAL | B_RELBUF);
1363 bp->b_flags &= ~B_ASYNC;
1367 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1368 (nbp->b_vp != vp) ||
1369 (nbp->b_flags & B_DELWRI) == 0)) {
1378 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1379 if (bp->b_lblkno > 0)
1382 * Since we hold the vnode lock this should only
1383 * fail if we're racing with the buf daemon.
1386 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1387 BO_MTX(bo)) == ENOLCK) {
1390 VNASSERT((bp->b_flags & B_DELWRI), vp,
1391 ("buf(%p) on dirty queue without DELWRI", bp));
1402 bufobj_wwait(bo, 0, 0);
1404 vnode_pager_setsize(vp, length);
1410 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1413 * NOTE: We have to deal with the special case of a background bitmap
1414 * buffer, a situation where two buffers will have the same logical
1415 * block offset. We want (1) only the foreground buffer to be accessed
1416 * in a lookup and (2) must differentiate between the foreground and
1417 * background buffer in the splay tree algorithm because the splay
1418 * tree cannot normally handle multiple entities with the same 'index'.
1419 * We accomplish this by adding differentiating flags to the splay tree's
1424 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1427 struct buf *lefttreemax, *righttreemin, *y;
1431 lefttreemax = righttreemin = &dummy;
1433 if (lblkno < root->b_lblkno ||
1434 (lblkno == root->b_lblkno &&
1435 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1436 if ((y = root->b_left) == NULL)
1438 if (lblkno < y->b_lblkno) {
1440 root->b_left = y->b_right;
1443 if ((y = root->b_left) == NULL)
1446 /* Link into the new root's right tree. */
1447 righttreemin->b_left = root;
1448 righttreemin = root;
1449 } else if (lblkno > root->b_lblkno ||
1450 (lblkno == root->b_lblkno &&
1451 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1452 if ((y = root->b_right) == NULL)
1454 if (lblkno > y->b_lblkno) {
1456 root->b_right = y->b_left;
1459 if ((y = root->b_right) == NULL)
1462 /* Link into the new root's left tree. */
1463 lefttreemax->b_right = root;
1470 /* Assemble the new root. */
1471 lefttreemax->b_right = root->b_left;
1472 righttreemin->b_left = root->b_right;
1473 root->b_left = dummy.b_right;
1474 root->b_right = dummy.b_left;
1479 buf_vlist_remove(struct buf *bp)
1484 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1485 ASSERT_BO_LOCKED(bp->b_bufobj);
1486 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1487 (BX_VNDIRTY|BX_VNCLEAN),
1488 ("buf_vlist_remove: Buf %p is on two lists", bp));
1489 if (bp->b_xflags & BX_VNDIRTY)
1490 bv = &bp->b_bufobj->bo_dirty;
1492 bv = &bp->b_bufobj->bo_clean;
1493 if (bp != bv->bv_root) {
1494 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1495 KASSERT(root == bp, ("splay lookup failed in remove"));
1497 if (bp->b_left == NULL) {
1500 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1501 root->b_right = bp->b_right;
1504 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1506 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1510 * Add the buffer to the sorted clean or dirty block list using a
1511 * splay tree algorithm.
1513 * NOTE: xflags is passed as a constant, optimizing this inline function!
1516 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1521 ASSERT_BO_LOCKED(bo);
1522 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1523 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1524 bp->b_xflags |= xflags;
1525 if (xflags & BX_VNDIRTY)
1530 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1534 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1535 } else if (bp->b_lblkno < root->b_lblkno ||
1536 (bp->b_lblkno == root->b_lblkno &&
1537 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1538 bp->b_left = root->b_left;
1540 root->b_left = NULL;
1541 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1543 bp->b_right = root->b_right;
1545 root->b_right = NULL;
1546 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1553 * Lookup a buffer using the splay tree. Note that we specifically avoid
1554 * shadow buffers used in background bitmap writes.
1556 * This code isn't quite efficient as it could be because we are maintaining
1557 * two sorted lists and do not know which list the block resides in.
1559 * During a "make buildworld" the desired buffer is found at one of
1560 * the roots more than 60% of the time. Thus, checking both roots
1561 * before performing either splay eliminates unnecessary splays on the
1562 * first tree splayed.
1565 gbincore(struct bufobj *bo, daddr_t lblkno)
1569 ASSERT_BO_LOCKED(bo);
1570 if ((bp = bo->bo_clean.bv_root) != NULL &&
1571 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1573 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1574 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1576 if ((bp = bo->bo_clean.bv_root) != NULL) {
1577 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1578 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1581 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1582 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1583 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1590 * Associate a buffer with a vnode.
1593 bgetvp(struct vnode *vp, struct buf *bp)
1598 ASSERT_BO_LOCKED(bo);
1599 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1601 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1602 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1603 ("bgetvp: bp already attached! %p", bp));
1606 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1607 bp->b_flags |= B_NEEDSGIANT;
1611 * Insert onto list for new vnode.
1613 buf_vlist_add(bp, bo, BX_VNCLEAN);
1617 * Disassociate a buffer from a vnode.
1620 brelvp(struct buf *bp)
1625 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1626 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1629 * Delete from old vnode list, if on one.
1631 vp = bp->b_vp; /* XXX */
1634 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1635 buf_vlist_remove(bp);
1637 panic("brelvp: Buffer %p not on queue.", bp);
1638 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1639 bo->bo_flag &= ~BO_ONWORKLST;
1640 mtx_lock(&sync_mtx);
1641 LIST_REMOVE(bo, bo_synclist);
1642 syncer_worklist_len--;
1643 mtx_unlock(&sync_mtx);
1645 bp->b_flags &= ~B_NEEDSGIANT;
1647 bp->b_bufobj = NULL;
1653 * Add an item to the syncer work queue.
1656 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1660 ASSERT_BO_LOCKED(bo);
1662 mtx_lock(&sync_mtx);
1663 if (bo->bo_flag & BO_ONWORKLST)
1664 LIST_REMOVE(bo, bo_synclist);
1666 bo->bo_flag |= BO_ONWORKLST;
1667 syncer_worklist_len++;
1670 if (delay > syncer_maxdelay - 2)
1671 delay = syncer_maxdelay - 2;
1672 slot = (syncer_delayno + delay) & syncer_mask;
1674 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1676 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1678 mtx_unlock(&sync_mtx);
1682 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1686 mtx_lock(&sync_mtx);
1687 len = syncer_worklist_len - sync_vnode_count;
1688 mtx_unlock(&sync_mtx);
1689 error = SYSCTL_OUT(req, &len, sizeof(len));
1693 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1694 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1696 static struct proc *updateproc;
1697 static void sched_sync(void);
1698 static struct kproc_desc up_kp = {
1703 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1706 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1711 *bo = LIST_FIRST(slp);
1714 vp = (*bo)->__bo_vnode; /* XXX */
1715 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1718 * We use vhold in case the vnode does not
1719 * successfully sync. vhold prevents the vnode from
1720 * going away when we unlock the sync_mtx so that
1721 * we can acquire the vnode interlock.
1724 mtx_unlock(&sync_mtx);
1726 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1728 mtx_lock(&sync_mtx);
1729 return (*bo == LIST_FIRST(slp));
1731 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1732 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1734 vn_finished_write(mp);
1736 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1738 * Put us back on the worklist. The worklist
1739 * routine will remove us from our current
1740 * position and then add us back in at a later
1743 vn_syncer_add_to_worklist(*bo, syncdelay);
1747 mtx_lock(&sync_mtx);
1752 * System filesystem synchronizer daemon.
1757 struct synclist *gnext, *next;
1758 struct synclist *gslp, *slp;
1761 struct thread *td = curthread;
1763 int net_worklist_len;
1764 int syncer_final_iter;
1769 syncer_final_iter = 0;
1771 syncer_state = SYNCER_RUNNING;
1772 starttime = time_uptime;
1773 td->td_pflags |= TDP_NORUNNINGBUF;
1775 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1778 mtx_lock(&sync_mtx);
1780 if (syncer_state == SYNCER_FINAL_DELAY &&
1781 syncer_final_iter == 0) {
1782 mtx_unlock(&sync_mtx);
1783 kproc_suspend_check(td->td_proc);
1784 mtx_lock(&sync_mtx);
1786 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1787 if (syncer_state != SYNCER_RUNNING &&
1788 starttime != time_uptime) {
1790 printf("\nSyncing disks, vnodes remaining...");
1793 printf("%d ", net_worklist_len);
1795 starttime = time_uptime;
1798 * Push files whose dirty time has expired. Be careful
1799 * of interrupt race on slp queue.
1801 * Skip over empty worklist slots when shutting down.
1804 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1805 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1806 syncer_delayno += 1;
1807 if (syncer_delayno == syncer_maxdelay)
1809 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1810 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1812 * If the worklist has wrapped since the
1813 * it was emptied of all but syncer vnodes,
1814 * switch to the FINAL_DELAY state and run
1815 * for one more second.
1817 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1818 net_worklist_len == 0 &&
1819 last_work_seen == syncer_delayno) {
1820 syncer_state = SYNCER_FINAL_DELAY;
1821 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1823 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1824 LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1827 * Keep track of the last time there was anything
1828 * on the worklist other than syncer vnodes.
1829 * Return to the SHUTTING_DOWN state if any
1832 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1833 last_work_seen = syncer_delayno;
1834 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1835 syncer_state = SYNCER_SHUTTING_DOWN;
1836 while (!LIST_EMPTY(slp)) {
1837 error = sync_vnode(slp, &bo, td);
1839 LIST_REMOVE(bo, bo_synclist);
1840 LIST_INSERT_HEAD(next, bo, bo_synclist);
1844 if (!LIST_EMPTY(gslp)) {
1845 mtx_unlock(&sync_mtx);
1847 mtx_lock(&sync_mtx);
1848 while (!LIST_EMPTY(gslp)) {
1849 error = sync_vnode(gslp, &bo, td);
1851 LIST_REMOVE(bo, bo_synclist);
1852 LIST_INSERT_HEAD(gnext, bo,
1859 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1860 syncer_final_iter--;
1862 * The variable rushjob allows the kernel to speed up the
1863 * processing of the filesystem syncer process. A rushjob
1864 * value of N tells the filesystem syncer to process the next
1865 * N seconds worth of work on its queue ASAP. Currently rushjob
1866 * is used by the soft update code to speed up the filesystem
1867 * syncer process when the incore state is getting so far
1868 * ahead of the disk that the kernel memory pool is being
1869 * threatened with exhaustion.
1876 * Just sleep for a short period of time between
1877 * iterations when shutting down to allow some I/O
1880 * If it has taken us less than a second to process the
1881 * current work, then wait. Otherwise start right over
1882 * again. We can still lose time if any single round
1883 * takes more than two seconds, but it does not really
1884 * matter as we are just trying to generally pace the
1885 * filesystem activity.
1887 if (syncer_state != SYNCER_RUNNING ||
1888 time_uptime == starttime) {
1890 sched_prio(td, PPAUSE);
1893 if (syncer_state != SYNCER_RUNNING)
1894 cv_timedwait(&sync_wakeup, &sync_mtx,
1895 hz / SYNCER_SHUTDOWN_SPEEDUP);
1896 else if (time_uptime == starttime)
1897 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1902 * Request the syncer daemon to speed up its work.
1903 * We never push it to speed up more than half of its
1904 * normal turn time, otherwise it could take over the cpu.
1907 speedup_syncer(void)
1911 mtx_lock(&sync_mtx);
1912 if (rushjob < syncdelay / 2) {
1914 stat_rush_requests += 1;
1917 mtx_unlock(&sync_mtx);
1918 cv_broadcast(&sync_wakeup);
1923 * Tell the syncer to speed up its work and run though its work
1924 * list several times, then tell it to shut down.
1927 syncer_shutdown(void *arg, int howto)
1930 if (howto & RB_NOSYNC)
1932 mtx_lock(&sync_mtx);
1933 syncer_state = SYNCER_SHUTTING_DOWN;
1935 mtx_unlock(&sync_mtx);
1936 cv_broadcast(&sync_wakeup);
1937 kproc_shutdown(arg, howto);
1941 * Reassign a buffer from one vnode to another.
1942 * Used to assign file specific control information
1943 * (indirect blocks) to the vnode to which they belong.
1946 reassignbuf(struct buf *bp)
1959 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1960 bp, bp->b_vp, bp->b_flags);
1962 * B_PAGING flagged buffers cannot be reassigned because their vp
1963 * is not fully linked in.
1965 if (bp->b_flags & B_PAGING)
1966 panic("cannot reassign paging buffer");
1969 * Delete from old vnode list, if on one.
1972 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1973 buf_vlist_remove(bp);
1975 panic("reassignbuf: Buffer %p not on queue.", bp);
1977 * If dirty, put on list of dirty buffers; otherwise insert onto list
1980 if (bp->b_flags & B_DELWRI) {
1981 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1982 switch (vp->v_type) {
1992 vn_syncer_add_to_worklist(bo, delay);
1994 buf_vlist_add(bp, bo, BX_VNDIRTY);
1996 buf_vlist_add(bp, bo, BX_VNCLEAN);
1998 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1999 mtx_lock(&sync_mtx);
2000 LIST_REMOVE(bo, bo_synclist);
2001 syncer_worklist_len--;
2002 mtx_unlock(&sync_mtx);
2003 bo->bo_flag &= ~BO_ONWORKLST;
2008 bp = TAILQ_FIRST(&bv->bv_hd);
2009 KASSERT(bp == NULL || bp->b_bufobj == bo,
2010 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2011 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2012 KASSERT(bp == NULL || bp->b_bufobj == bo,
2013 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2015 bp = TAILQ_FIRST(&bv->bv_hd);
2016 KASSERT(bp == NULL || bp->b_bufobj == bo,
2017 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2018 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2019 KASSERT(bp == NULL || bp->b_bufobj == bo,
2020 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2026 * Increment the use and hold counts on the vnode, taking care to reference
2027 * the driver's usecount if this is a chardev. The vholdl() will remove
2028 * the vnode from the free list if it is presently free. Requires the
2029 * vnode interlock and returns with it held.
2032 v_incr_usecount(struct vnode *vp)
2035 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2037 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2039 vp->v_rdev->si_usecount++;
2046 * Turn a holdcnt into a use+holdcnt such that only one call to
2047 * v_decr_usecount is needed.
2050 v_upgrade_usecount(struct vnode *vp)
2053 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2055 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2057 vp->v_rdev->si_usecount++;
2063 * Decrement the vnode use and hold count along with the driver's usecount
2064 * if this is a chardev. The vdropl() below releases the vnode interlock
2065 * as it may free the vnode.
2068 v_decr_usecount(struct vnode *vp)
2071 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2072 VNASSERT(vp->v_usecount > 0, vp,
2073 ("v_decr_usecount: negative usecount"));
2074 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2076 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2078 vp->v_rdev->si_usecount--;
2085 * Decrement only the use count and driver use count. This is intended to
2086 * be paired with a follow on vdropl() to release the remaining hold count.
2087 * In this way we may vgone() a vnode with a 0 usecount without risk of
2088 * having it end up on a free list because the hold count is kept above 0.
2091 v_decr_useonly(struct vnode *vp)
2094 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2095 VNASSERT(vp->v_usecount > 0, vp,
2096 ("v_decr_useonly: negative usecount"));
2097 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2099 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2101 vp->v_rdev->si_usecount--;
2107 * Grab a particular vnode from the free list, increment its
2108 * reference count and lock it. VI_DOOMED is set if the vnode
2109 * is being destroyed. Only callers who specify LK_RETRY will
2110 * see doomed vnodes. If inactive processing was delayed in
2111 * vput try to do it here.
2114 vget(struct vnode *vp, int flags, struct thread *td)
2119 VFS_ASSERT_GIANT(vp->v_mount);
2120 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2121 ("vget: invalid lock operation"));
2122 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2124 if ((flags & LK_INTERLOCK) == 0)
2127 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2129 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2133 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2134 panic("vget: vn_lock failed to return ENOENT\n");
2136 /* Upgrade our holdcnt to a usecount. */
2137 v_upgrade_usecount(vp);
2139 * We don't guarantee that any particular close will
2140 * trigger inactive processing so just make a best effort
2141 * here at preventing a reference to a removed file. If
2142 * we don't succeed no harm is done.
2144 if (vp->v_iflag & VI_OWEINACT) {
2145 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2146 (flags & LK_NOWAIT) == 0)
2148 vp->v_iflag &= ~VI_OWEINACT;
2155 * Increase the reference count of a vnode.
2158 vref(struct vnode *vp)
2161 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2163 v_incr_usecount(vp);
2168 * Return reference count of a vnode.
2170 * The results of this call are only guaranteed when some mechanism other
2171 * than the VI lock is used to stop other processes from gaining references
2172 * to the vnode. This may be the case if the caller holds the only reference.
2173 * This is also useful when stale data is acceptable as race conditions may
2174 * be accounted for by some other means.
2177 vrefcnt(struct vnode *vp)
2182 usecnt = vp->v_usecount;
2188 #define VPUTX_VRELE 1
2189 #define VPUTX_VPUT 2
2190 #define VPUTX_VUNREF 3
2193 vputx(struct vnode *vp, int func)
2197 KASSERT(vp != NULL, ("vputx: null vp"));
2198 if (func == VPUTX_VUNREF)
2199 ASSERT_VOP_LOCKED(vp, "vunref");
2200 else if (func == VPUTX_VPUT)
2201 ASSERT_VOP_LOCKED(vp, "vput");
2203 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2204 VFS_ASSERT_GIANT(vp->v_mount);
2205 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2208 /* Skip this v_writecount check if we're going to panic below. */
2209 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2210 ("vputx: missed vn_close"));
2213 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2214 vp->v_usecount == 1)) {
2215 if (func == VPUTX_VPUT)
2217 v_decr_usecount(vp);
2221 if (vp->v_usecount != 1) {
2222 vprint("vputx: negative ref count", vp);
2223 panic("vputx: negative ref cnt");
2225 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2227 * We want to hold the vnode until the inactive finishes to
2228 * prevent vgone() races. We drop the use count here and the
2229 * hold count below when we're done.
2233 * We must call VOP_INACTIVE with the node locked. Mark
2234 * as VI_DOINGINACT to avoid recursion.
2236 vp->v_iflag |= VI_OWEINACT;
2239 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2243 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2244 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2250 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2254 if (vp->v_usecount > 0)
2255 vp->v_iflag &= ~VI_OWEINACT;
2257 if (vp->v_iflag & VI_OWEINACT)
2258 vinactive(vp, curthread);
2259 if (func != VPUTX_VUNREF)
2266 * Vnode put/release.
2267 * If count drops to zero, call inactive routine and return to freelist.
2270 vrele(struct vnode *vp)
2273 vputx(vp, VPUTX_VRELE);
2277 * Release an already locked vnode. This give the same effects as
2278 * unlock+vrele(), but takes less time and avoids releasing and
2279 * re-aquiring the lock (as vrele() acquires the lock internally.)
2282 vput(struct vnode *vp)
2285 vputx(vp, VPUTX_VPUT);
2289 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2292 vunref(struct vnode *vp)
2295 vputx(vp, VPUTX_VUNREF);
2299 * Somebody doesn't want the vnode recycled.
2302 vhold(struct vnode *vp)
2311 vholdl(struct vnode *vp)
2314 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2316 if (VSHOULDBUSY(vp))
2321 * Note that there is one less who cares about this vnode. vdrop() is the
2322 * opposite of vhold().
2325 vdrop(struct vnode *vp)
2333 * Drop the hold count of the vnode. If this is the last reference to
2334 * the vnode we will free it if it has been vgone'd otherwise it is
2335 * placed on the free list.
2338 vdropl(struct vnode *vp)
2341 ASSERT_VI_LOCKED(vp, "vdropl");
2342 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2343 if (vp->v_holdcnt <= 0)
2344 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2346 if (vp->v_holdcnt == 0) {
2347 if (vp->v_iflag & VI_DOOMED) {
2348 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__,
2359 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2360 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2361 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2362 * failed lock upgrade.
2365 vinactive(struct vnode *vp, struct thread *td)
2368 ASSERT_VOP_ELOCKED(vp, "vinactive");
2369 ASSERT_VI_LOCKED(vp, "vinactive");
2370 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2371 ("vinactive: recursed on VI_DOINGINACT"));
2372 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2373 vp->v_iflag |= VI_DOINGINACT;
2374 vp->v_iflag &= ~VI_OWEINACT;
2376 VOP_INACTIVE(vp, td);
2378 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2379 ("vinactive: lost VI_DOINGINACT"));
2380 vp->v_iflag &= ~VI_DOINGINACT;
2384 * Remove any vnodes in the vnode table belonging to mount point mp.
2386 * If FORCECLOSE is not specified, there should not be any active ones,
2387 * return error if any are found (nb: this is a user error, not a
2388 * system error). If FORCECLOSE is specified, detach any active vnodes
2391 * If WRITECLOSE is set, only flush out regular file vnodes open for
2394 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2396 * `rootrefs' specifies the base reference count for the root vnode
2397 * of this filesystem. The root vnode is considered busy if its
2398 * v_usecount exceeds this value. On a successful return, vflush(, td)
2399 * will call vrele() on the root vnode exactly rootrefs times.
2400 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2404 static int busyprt = 0; /* print out busy vnodes */
2405 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2409 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2411 struct vnode *vp, *mvp, *rootvp = NULL;
2413 int busy = 0, error;
2415 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2418 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2419 ("vflush: bad args"));
2421 * Get the filesystem root vnode. We can vput() it
2422 * immediately, since with rootrefs > 0, it won't go away.
2424 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2425 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2434 MNT_VNODE_FOREACH(vp, mp, mvp) {
2439 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2443 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2447 * Skip over a vnodes marked VV_SYSTEM.
2449 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2456 * If WRITECLOSE is set, flush out unlinked but still open
2457 * files (even if open only for reading) and regular file
2458 * vnodes open for writing.
2460 if (flags & WRITECLOSE) {
2461 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2464 if ((vp->v_type == VNON ||
2465 (error == 0 && vattr.va_nlink > 0)) &&
2466 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2475 * With v_usecount == 0, all we need to do is clear out the
2476 * vnode data structures and we are done.
2478 * If FORCECLOSE is set, forcibly close the vnode.
2480 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2481 VNASSERT(vp->v_usecount == 0 ||
2482 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2483 ("device VNODE %p is FORCECLOSED", vp));
2489 vprint("vflush: busy vnode", vp);
2497 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2499 * If just the root vnode is busy, and if its refcount
2500 * is equal to `rootrefs', then go ahead and kill it.
2503 KASSERT(busy > 0, ("vflush: not busy"));
2504 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2505 ("vflush: usecount %d < rootrefs %d",
2506 rootvp->v_usecount, rootrefs));
2507 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2508 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2510 VOP_UNLOCK(rootvp, 0);
2516 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2520 for (; rootrefs > 0; rootrefs--)
2526 * Recycle an unused vnode to the front of the free list.
2529 vrecycle(struct vnode *vp, struct thread *td)
2533 ASSERT_VOP_ELOCKED(vp, "vrecycle");
2534 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2537 if (vp->v_usecount == 0) {
2546 * Eliminate all activity associated with a vnode
2547 * in preparation for reuse.
2550 vgone(struct vnode *vp)
2558 * vgone, with the vp interlock held.
2561 vgonel(struct vnode *vp)
2568 ASSERT_VOP_ELOCKED(vp, "vgonel");
2569 ASSERT_VI_LOCKED(vp, "vgonel");
2570 VNASSERT(vp->v_holdcnt, vp,
2571 ("vgonel: vp %p has no reference.", vp));
2572 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2576 * Don't vgonel if we're already doomed.
2578 if (vp->v_iflag & VI_DOOMED)
2580 vp->v_iflag |= VI_DOOMED;
2582 * Check to see if the vnode is in use. If so, we have to call
2583 * VOP_CLOSE() and VOP_INACTIVE().
2585 active = vp->v_usecount;
2586 oweinact = (vp->v_iflag & VI_OWEINACT);
2589 * Clean out any buffers associated with the vnode.
2590 * If the flush fails, just toss the buffers.
2593 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2594 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2595 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2596 vinvalbuf(vp, 0, 0, 0);
2599 * If purging an active vnode, it must be closed and
2600 * deactivated before being reclaimed.
2603 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2604 if (oweinact || active) {
2606 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2611 * Reclaim the vnode.
2613 if (VOP_RECLAIM(vp, td))
2614 panic("vgone: cannot reclaim");
2616 vn_finished_secondary_write(mp);
2617 VNASSERT(vp->v_object == NULL, vp,
2618 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2620 * Clear the advisory locks and wake up waiting threads.
2622 lf_purgelocks(vp, &(vp->v_lockf));
2624 * Delete from old mount point vnode list.
2629 * Done with purge, reset to the standard lock and invalidate
2633 vp->v_vnlock = &vp->v_lock;
2634 vp->v_op = &dead_vnodeops;
2640 * Calculate the total number of references to a special device.
2643 vcount(struct vnode *vp)
2648 count = vp->v_rdev->si_usecount;
2654 * Same as above, but using the struct cdev *as argument
2657 count_dev(struct cdev *dev)
2662 count = dev->si_usecount;
2668 * Print out a description of a vnode.
2670 static char *typename[] =
2671 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2675 vn_printf(struct vnode *vp, const char *fmt, ...)
2678 char buf[256], buf2[16];
2684 printf("%p: ", (void *)vp);
2685 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2686 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2687 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2690 if (vp->v_vflag & VV_ROOT)
2691 strlcat(buf, "|VV_ROOT", sizeof(buf));
2692 if (vp->v_vflag & VV_ISTTY)
2693 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2694 if (vp->v_vflag & VV_NOSYNC)
2695 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2696 if (vp->v_vflag & VV_CACHEDLABEL)
2697 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2698 if (vp->v_vflag & VV_TEXT)
2699 strlcat(buf, "|VV_TEXT", sizeof(buf));
2700 if (vp->v_vflag & VV_COPYONWRITE)
2701 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2702 if (vp->v_vflag & VV_SYSTEM)
2703 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2704 if (vp->v_vflag & VV_PROCDEP)
2705 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2706 if (vp->v_vflag & VV_NOKNOTE)
2707 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2708 if (vp->v_vflag & VV_DELETED)
2709 strlcat(buf, "|VV_DELETED", sizeof(buf));
2710 if (vp->v_vflag & VV_MD)
2711 strlcat(buf, "|VV_MD", sizeof(buf));
2712 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2713 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2714 VV_NOKNOTE | VV_DELETED | VV_MD);
2716 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2717 strlcat(buf, buf2, sizeof(buf));
2719 if (vp->v_iflag & VI_MOUNT)
2720 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2721 if (vp->v_iflag & VI_AGE)
2722 strlcat(buf, "|VI_AGE", sizeof(buf));
2723 if (vp->v_iflag & VI_DOOMED)
2724 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2725 if (vp->v_iflag & VI_FREE)
2726 strlcat(buf, "|VI_FREE", sizeof(buf));
2727 if (vp->v_iflag & VI_DOINGINACT)
2728 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2729 if (vp->v_iflag & VI_OWEINACT)
2730 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2731 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2732 VI_DOINGINACT | VI_OWEINACT);
2734 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2735 strlcat(buf, buf2, sizeof(buf));
2737 printf(" flags (%s)\n", buf + 1);
2738 if (mtx_owned(VI_MTX(vp)))
2739 printf(" VI_LOCKed");
2740 if (vp->v_object != NULL)
2741 printf(" v_object %p ref %d pages %d\n",
2742 vp->v_object, vp->v_object->ref_count,
2743 vp->v_object->resident_page_count);
2745 lockmgr_printinfo(vp->v_vnlock);
2746 if (vp->v_data != NULL)
2752 * List all of the locked vnodes in the system.
2753 * Called when debugging the kernel.
2755 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2757 struct mount *mp, *nmp;
2761 * Note: because this is DDB, we can't obey the locking semantics
2762 * for these structures, which means we could catch an inconsistent
2763 * state and dereference a nasty pointer. Not much to be done
2766 db_printf("Locked vnodes\n");
2767 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2768 nmp = TAILQ_NEXT(mp, mnt_list);
2769 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2770 if (vp->v_type != VMARKER &&
2774 nmp = TAILQ_NEXT(mp, mnt_list);
2779 * Show details about the given vnode.
2781 DB_SHOW_COMMAND(vnode, db_show_vnode)
2787 vp = (struct vnode *)addr;
2788 vn_printf(vp, "vnode ");
2792 * Show details about the given mount point.
2794 DB_SHOW_COMMAND(mount, db_show_mount)
2804 /* No address given, print short info about all mount points. */
2805 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2806 db_printf("%p %s on %s (%s)\n", mp,
2807 mp->mnt_stat.f_mntfromname,
2808 mp->mnt_stat.f_mntonname,
2809 mp->mnt_stat.f_fstypename);
2813 db_printf("\nMore info: show mount <addr>\n");
2817 mp = (struct mount *)addr;
2818 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2819 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2822 flags = mp->mnt_flag;
2823 #define MNT_FLAG(flag) do { \
2824 if (flags & (flag)) { \
2825 if (buf[0] != '\0') \
2826 strlcat(buf, ", ", sizeof(buf)); \
2827 strlcat(buf, (#flag) + 4, sizeof(buf)); \
2831 MNT_FLAG(MNT_RDONLY);
2832 MNT_FLAG(MNT_SYNCHRONOUS);
2833 MNT_FLAG(MNT_NOEXEC);
2834 MNT_FLAG(MNT_NOSUID);
2835 MNT_FLAG(MNT_UNION);
2836 MNT_FLAG(MNT_ASYNC);
2837 MNT_FLAG(MNT_SUIDDIR);
2838 MNT_FLAG(MNT_SOFTDEP);
2839 MNT_FLAG(MNT_NOSYMFOLLOW);
2840 MNT_FLAG(MNT_GJOURNAL);
2841 MNT_FLAG(MNT_MULTILABEL);
2843 MNT_FLAG(MNT_NOATIME);
2844 MNT_FLAG(MNT_NOCLUSTERR);
2845 MNT_FLAG(MNT_NOCLUSTERW);
2846 MNT_FLAG(MNT_NFS4ACLS);
2847 MNT_FLAG(MNT_EXRDONLY);
2848 MNT_FLAG(MNT_EXPORTED);
2849 MNT_FLAG(MNT_DEFEXPORTED);
2850 MNT_FLAG(MNT_EXPORTANON);
2851 MNT_FLAG(MNT_EXKERB);
2852 MNT_FLAG(MNT_EXPUBLIC);
2853 MNT_FLAG(MNT_LOCAL);
2854 MNT_FLAG(MNT_QUOTA);
2855 MNT_FLAG(MNT_ROOTFS);
2857 MNT_FLAG(MNT_IGNORE);
2858 MNT_FLAG(MNT_UPDATE);
2859 MNT_FLAG(MNT_DELEXPORT);
2860 MNT_FLAG(MNT_RELOAD);
2861 MNT_FLAG(MNT_FORCE);
2862 MNT_FLAG(MNT_SNAPSHOT);
2863 MNT_FLAG(MNT_BYFSID);
2867 strlcat(buf, ", ", sizeof(buf));
2868 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2871 db_printf(" mnt_flag = %s\n", buf);
2874 flags = mp->mnt_kern_flag;
2875 #define MNT_KERN_FLAG(flag) do { \
2876 if (flags & (flag)) { \
2877 if (buf[0] != '\0') \
2878 strlcat(buf, ", ", sizeof(buf)); \
2879 strlcat(buf, (#flag) + 5, sizeof(buf)); \
2883 MNT_KERN_FLAG(MNTK_UNMOUNTF);
2884 MNT_KERN_FLAG(MNTK_ASYNC);
2885 MNT_KERN_FLAG(MNTK_SOFTDEP);
2886 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2887 MNT_KERN_FLAG(MNTK_UNMOUNT);
2888 MNT_KERN_FLAG(MNTK_MWAIT);
2889 MNT_KERN_FLAG(MNTK_SUSPEND);
2890 MNT_KERN_FLAG(MNTK_SUSPEND2);
2891 MNT_KERN_FLAG(MNTK_SUSPENDED);
2892 MNT_KERN_FLAG(MNTK_MPSAFE);
2893 MNT_KERN_FLAG(MNTK_NOKNOTE);
2894 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2895 #undef MNT_KERN_FLAG
2898 strlcat(buf, ", ", sizeof(buf));
2899 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2902 db_printf(" mnt_kern_flag = %s\n", buf);
2904 db_printf(" mnt_opt = ");
2905 opt = TAILQ_FIRST(mp->mnt_opt);
2907 db_printf("%s", opt->name);
2908 opt = TAILQ_NEXT(opt, link);
2909 while (opt != NULL) {
2910 db_printf(", %s", opt->name);
2911 opt = TAILQ_NEXT(opt, link);
2917 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
2918 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2919 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2920 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2921 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2922 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2923 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2924 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2925 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2926 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2927 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2928 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2930 db_printf(" mnt_cred = { uid=%u ruid=%u",
2931 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2932 if (jailed(mp->mnt_cred))
2933 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2935 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
2936 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
2937 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2938 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2939 db_printf(" mnt_noasync = %u\n", mp->mnt_noasync);
2940 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2941 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2942 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
2943 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2944 db_printf(" mnt_secondary_accwrites = %d\n",
2945 mp->mnt_secondary_accwrites);
2946 db_printf(" mnt_gjprovider = %s\n",
2947 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2950 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2951 if (vp->v_type != VMARKER) {
2952 vn_printf(vp, "vnode ");
2961 * Fill in a struct xvfsconf based on a struct vfsconf.
2964 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2967 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2968 xvfsp->vfc_typenum = vfsp->vfc_typenum;
2969 xvfsp->vfc_refcount = vfsp->vfc_refcount;
2970 xvfsp->vfc_flags = vfsp->vfc_flags;
2972 * These are unused in userland, we keep them
2973 * to not break binary compatibility.
2975 xvfsp->vfc_vfsops = NULL;
2976 xvfsp->vfc_next = NULL;
2980 * Top level filesystem related information gathering.
2983 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2985 struct vfsconf *vfsp;
2986 struct xvfsconf xvfsp;
2990 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2991 bzero(&xvfsp, sizeof(xvfsp));
2992 vfsconf2x(vfsp, &xvfsp);
2993 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
3000 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
3001 "S,xvfsconf", "List of all configured filesystems");
3003 #ifndef BURN_BRIDGES
3004 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3007 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3009 int *name = (int *)arg1 - 1; /* XXX */
3010 u_int namelen = arg2 + 1; /* XXX */
3011 struct vfsconf *vfsp;
3012 struct xvfsconf xvfsp;
3014 printf("WARNING: userland calling deprecated sysctl, "
3015 "please rebuild world\n");
3017 #if 1 || defined(COMPAT_PRELITE2)
3018 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3020 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3024 case VFS_MAXTYPENUM:
3027 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3030 return (ENOTDIR); /* overloaded */
3031 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3032 if (vfsp->vfc_typenum == name[2])
3035 return (EOPNOTSUPP);
3036 bzero(&xvfsp, sizeof(xvfsp));
3037 vfsconf2x(vfsp, &xvfsp);
3038 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3040 return (EOPNOTSUPP);
3043 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3044 vfs_sysctl, "Generic filesystem");
3046 #if 1 || defined(COMPAT_PRELITE2)
3049 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3052 struct vfsconf *vfsp;
3053 struct ovfsconf ovfs;
3055 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3056 bzero(&ovfs, sizeof(ovfs));
3057 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3058 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3059 ovfs.vfc_index = vfsp->vfc_typenum;
3060 ovfs.vfc_refcount = vfsp->vfc_refcount;
3061 ovfs.vfc_flags = vfsp->vfc_flags;
3062 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3069 #endif /* 1 || COMPAT_PRELITE2 */
3070 #endif /* !BURN_BRIDGES */
3072 #define KINFO_VNODESLOP 10
3075 * Dump vnode list (via sysctl).
3079 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3087 * Stale numvnodes access is not fatal here.
3090 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3092 /* Make an estimate */
3093 return (SYSCTL_OUT(req, 0, len));
3095 error = sysctl_wire_old_buffer(req, 0);
3098 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3100 mtx_lock(&mountlist_mtx);
3101 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3102 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3105 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3109 xvn[n].xv_size = sizeof *xvn;
3110 xvn[n].xv_vnode = vp;
3111 xvn[n].xv_id = 0; /* XXX compat */
3112 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3114 XV_COPY(writecount);
3120 xvn[n].xv_flag = vp->v_vflag;
3122 switch (vp->v_type) {
3129 if (vp->v_rdev == NULL) {
3133 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3136 xvn[n].xv_socket = vp->v_socket;
3139 xvn[n].xv_fifo = vp->v_fifoinfo;
3144 /* shouldn't happen? */
3152 mtx_lock(&mountlist_mtx);
3157 mtx_unlock(&mountlist_mtx);
3159 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3164 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3165 0, 0, sysctl_vnode, "S,xvnode", "");
3169 * Unmount all filesystems. The list is traversed in reverse order
3170 * of mounting to avoid dependencies.
3173 vfs_unmountall(void)
3179 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3180 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3184 * Since this only runs when rebooting, it is not interlocked.
3186 while(!TAILQ_EMPTY(&mountlist)) {
3187 mp = TAILQ_LAST(&mountlist, mntlist);
3188 error = dounmount(mp, MNT_FORCE, td);
3190 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3192 * XXX: Due to the way in which we mount the root
3193 * file system off of devfs, devfs will generate a
3194 * "busy" warning when we try to unmount it before
3195 * the root. Don't print a warning as a result in
3196 * order to avoid false positive errors that may
3197 * cause needless upset.
3199 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3200 printf("unmount of %s failed (",
3201 mp->mnt_stat.f_mntonname);
3205 printf("%d)\n", error);
3208 /* The unmount has removed mp from the mountlist */
3214 * perform msync on all vnodes under a mount point
3215 * the mount point must be locked.
3218 vfs_msync(struct mount *mp, int flags)
3220 struct vnode *vp, *mvp;
3221 struct vm_object *obj;
3223 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3225 MNT_VNODE_FOREACH(vp, mp, mvp) {
3228 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3229 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3232 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3234 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3242 VM_OBJECT_LOCK(obj);
3243 vm_object_page_clean(obj, 0, 0,
3245 OBJPC_SYNC : OBJPC_NOSYNC);
3246 VM_OBJECT_UNLOCK(obj);
3258 * Mark a vnode as free, putting it up for recycling.
3261 vfree(struct vnode *vp)
3264 ASSERT_VI_LOCKED(vp, "vfree");
3265 mtx_lock(&vnode_free_list_mtx);
3266 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3267 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3268 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3269 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3270 ("vfree: Freeing doomed vnode"));
3271 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3272 if (vp->v_iflag & VI_AGE) {
3273 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3275 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3278 vp->v_iflag &= ~VI_AGE;
3279 vp->v_iflag |= VI_FREE;
3280 mtx_unlock(&vnode_free_list_mtx);
3284 * Opposite of vfree() - mark a vnode as in use.
3287 vbusy(struct vnode *vp)
3289 ASSERT_VI_LOCKED(vp, "vbusy");
3290 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3291 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3292 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3294 mtx_lock(&vnode_free_list_mtx);
3295 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3297 vp->v_iflag &= ~(VI_FREE|VI_AGE);
3298 mtx_unlock(&vnode_free_list_mtx);
3302 destroy_vpollinfo(struct vpollinfo *vi)
3304 knlist_destroy(&vi->vpi_selinfo.si_note);
3305 mtx_destroy(&vi->vpi_lock);
3306 uma_zfree(vnodepoll_zone, vi);
3310 * Initalize per-vnode helper structure to hold poll-related state.
3313 v_addpollinfo(struct vnode *vp)
3315 struct vpollinfo *vi;
3317 if (vp->v_pollinfo != NULL)
3319 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3320 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3321 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3322 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3324 if (vp->v_pollinfo != NULL) {
3326 destroy_vpollinfo(vi);
3329 vp->v_pollinfo = vi;
3334 * Record a process's interest in events which might happen to
3335 * a vnode. Because poll uses the historic select-style interface
3336 * internally, this routine serves as both the ``check for any
3337 * pending events'' and the ``record my interest in future events''
3338 * functions. (These are done together, while the lock is held,
3339 * to avoid race conditions.)
3342 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3346 mtx_lock(&vp->v_pollinfo->vpi_lock);
3347 if (vp->v_pollinfo->vpi_revents & events) {
3349 * This leaves events we are not interested
3350 * in available for the other process which
3351 * which presumably had requested them
3352 * (otherwise they would never have been
3355 events &= vp->v_pollinfo->vpi_revents;
3356 vp->v_pollinfo->vpi_revents &= ~events;
3358 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3361 vp->v_pollinfo->vpi_events |= events;
3362 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3363 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3368 * Routine to create and manage a filesystem syncer vnode.
3370 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3371 static int sync_fsync(struct vop_fsync_args *);
3372 static int sync_inactive(struct vop_inactive_args *);
3373 static int sync_reclaim(struct vop_reclaim_args *);
3375 static struct vop_vector sync_vnodeops = {
3376 .vop_bypass = VOP_EOPNOTSUPP,
3377 .vop_close = sync_close, /* close */
3378 .vop_fsync = sync_fsync, /* fsync */
3379 .vop_inactive = sync_inactive, /* inactive */
3380 .vop_reclaim = sync_reclaim, /* reclaim */
3381 .vop_lock1 = vop_stdlock, /* lock */
3382 .vop_unlock = vop_stdunlock, /* unlock */
3383 .vop_islocked = vop_stdislocked, /* islocked */
3387 * Create a new filesystem syncer vnode for the specified mount point.
3390 vfs_allocate_syncvnode(struct mount *mp)
3394 static long start, incr, next;
3397 /* Allocate a new vnode */
3398 if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3399 mp->mnt_syncer = NULL;
3403 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3404 vp->v_vflag |= VV_FORCEINSMQ;
3405 error = insmntque(vp, mp);
3407 panic("vfs_allocate_syncvnode: insmntque failed");
3408 vp->v_vflag &= ~VV_FORCEINSMQ;
3411 * Place the vnode onto the syncer worklist. We attempt to
3412 * scatter them about on the list so that they will go off
3413 * at evenly distributed times even if all the filesystems
3414 * are mounted at once.
3417 if (next == 0 || next > syncer_maxdelay) {
3421 start = syncer_maxdelay / 2;
3422 incr = syncer_maxdelay;
3428 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3429 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3430 mtx_lock(&sync_mtx);
3432 mtx_unlock(&sync_mtx);
3434 mp->mnt_syncer = vp;
3439 * Do a lazy sync of the filesystem.
3442 sync_fsync(struct vop_fsync_args *ap)
3444 struct vnode *syncvp = ap->a_vp;
3445 struct mount *mp = syncvp->v_mount;
3450 * We only need to do something if this is a lazy evaluation.
3452 if (ap->a_waitfor != MNT_LAZY)
3456 * Move ourselves to the back of the sync list.
3458 bo = &syncvp->v_bufobj;
3460 vn_syncer_add_to_worklist(bo, syncdelay);
3464 * Walk the list of vnodes pushing all that are dirty and
3465 * not already on the sync list.
3467 mtx_lock(&mountlist_mtx);
3468 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3469 mtx_unlock(&mountlist_mtx);
3472 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3478 mp->mnt_kern_flag &= ~MNTK_ASYNC;
3480 vfs_msync(mp, MNT_NOWAIT);
3481 error = VFS_SYNC(mp, MNT_LAZY);
3484 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3485 mp->mnt_kern_flag |= MNTK_ASYNC;
3487 vn_finished_write(mp);
3493 * The syncer vnode is no referenced.
3496 sync_inactive(struct vop_inactive_args *ap)
3504 * The syncer vnode is no longer needed and is being decommissioned.
3506 * Modifications to the worklist must be protected by sync_mtx.
3509 sync_reclaim(struct vop_reclaim_args *ap)
3511 struct vnode *vp = ap->a_vp;
3516 vp->v_mount->mnt_syncer = NULL;
3517 if (bo->bo_flag & BO_ONWORKLST) {
3518 mtx_lock(&sync_mtx);
3519 LIST_REMOVE(bo, bo_synclist);
3520 syncer_worklist_len--;
3522 mtx_unlock(&sync_mtx);
3523 bo->bo_flag &= ~BO_ONWORKLST;
3531 * Check if vnode represents a disk device
3534 vn_isdisk(struct vnode *vp, int *errp)
3540 if (vp->v_type != VCHR)
3542 else if (vp->v_rdev == NULL)
3544 else if (vp->v_rdev->si_devsw == NULL)
3546 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3551 return (error == 0);
3555 * Common filesystem object access control check routine. Accepts a
3556 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3557 * and optional call-by-reference privused argument allowing vaccess()
3558 * to indicate to the caller whether privilege was used to satisfy the
3559 * request (obsoleted). Returns 0 on success, or an errno on failure.
3561 * The ifdef'd CAPABILITIES version is here for reference, but is not
3565 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3566 accmode_t accmode, struct ucred *cred, int *privused)
3568 accmode_t dac_granted;
3569 accmode_t priv_granted;
3571 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3572 ("invalid bit in accmode"));
3575 * Look for a normal, non-privileged way to access the file/directory
3576 * as requested. If it exists, go with that.
3579 if (privused != NULL)
3584 /* Check the owner. */
3585 if (cred->cr_uid == file_uid) {
3586 dac_granted |= VADMIN;
3587 if (file_mode & S_IXUSR)
3588 dac_granted |= VEXEC;
3589 if (file_mode & S_IRUSR)
3590 dac_granted |= VREAD;
3591 if (file_mode & S_IWUSR)
3592 dac_granted |= (VWRITE | VAPPEND);
3594 if ((accmode & dac_granted) == accmode)
3600 /* Otherwise, check the groups (first match) */
3601 if (groupmember(file_gid, cred)) {
3602 if (file_mode & S_IXGRP)
3603 dac_granted |= VEXEC;
3604 if (file_mode & S_IRGRP)
3605 dac_granted |= VREAD;
3606 if (file_mode & S_IWGRP)
3607 dac_granted |= (VWRITE | VAPPEND);
3609 if ((accmode & dac_granted) == accmode)
3615 /* Otherwise, check everyone else. */
3616 if (file_mode & S_IXOTH)
3617 dac_granted |= VEXEC;
3618 if (file_mode & S_IROTH)
3619 dac_granted |= VREAD;
3620 if (file_mode & S_IWOTH)
3621 dac_granted |= (VWRITE | VAPPEND);
3622 if ((accmode & dac_granted) == accmode)
3627 * Build a privilege mask to determine if the set of privileges
3628 * satisfies the requirements when combined with the granted mask
3629 * from above. For each privilege, if the privilege is required,
3630 * bitwise or the request type onto the priv_granted mask.
3636 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3637 * requests, instead of PRIV_VFS_EXEC.
3639 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3640 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3641 priv_granted |= VEXEC;
3643 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3644 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3645 priv_granted |= VEXEC;
3648 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3649 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3650 priv_granted |= VREAD;
3652 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3653 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3654 priv_granted |= (VWRITE | VAPPEND);
3656 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3657 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3658 priv_granted |= VADMIN;
3660 if ((accmode & (priv_granted | dac_granted)) == accmode) {
3661 /* XXX audit: privilege used */
3662 if (privused != NULL)
3667 return ((accmode & VADMIN) ? EPERM : EACCES);
3671 * Credential check based on process requesting service, and per-attribute
3675 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3676 struct thread *td, accmode_t accmode)
3680 * Kernel-invoked always succeeds.
3686 * Do not allow privileged processes in jail to directly manipulate
3687 * system attributes.
3689 switch (attrnamespace) {
3690 case EXTATTR_NAMESPACE_SYSTEM:
3691 /* Potentially should be: return (EPERM); */
3692 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3693 case EXTATTR_NAMESPACE_USER:
3694 return (VOP_ACCESS(vp, accmode, cred, td));
3700 #ifdef DEBUG_VFS_LOCKS
3702 * This only exists to supress warnings from unlocked specfs accesses. It is
3703 * no longer ok to have an unlocked VFS.
3705 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
3706 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
3708 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3709 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3710 "Drop into debugger on lock violation");
3712 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3713 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3714 0, "Check for interlock across VOPs");
3716 int vfs_badlock_print = 1; /* Print lock violations. */
3717 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3718 0, "Print lock violations");
3721 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3722 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3723 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3727 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3731 if (vfs_badlock_backtrace)
3734 if (vfs_badlock_print)
3735 printf("%s: %p %s\n", str, (void *)vp, msg);
3736 if (vfs_badlock_ddb)
3737 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3741 assert_vi_locked(struct vnode *vp, const char *str)
3744 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3745 vfs_badlock("interlock is not locked but should be", str, vp);
3749 assert_vi_unlocked(struct vnode *vp, const char *str)
3752 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3753 vfs_badlock("interlock is locked but should not be", str, vp);
3757 assert_vop_locked(struct vnode *vp, const char *str)
3760 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3761 vfs_badlock("is not locked but should be", str, vp);
3765 assert_vop_unlocked(struct vnode *vp, const char *str)
3768 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3769 vfs_badlock("is locked but should not be", str, vp);
3773 assert_vop_elocked(struct vnode *vp, const char *str)
3776 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3777 vfs_badlock("is not exclusive locked but should be", str, vp);
3782 assert_vop_elocked_other(struct vnode *vp, const char *str)
3785 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3786 vfs_badlock("is not exclusive locked by another thread",
3791 assert_vop_slocked(struct vnode *vp, const char *str)
3794 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3795 vfs_badlock("is not locked shared but should be", str, vp);
3798 #endif /* DEBUG_VFS_LOCKS */
3801 vop_rename_fail(struct vop_rename_args *ap)
3804 if (ap->a_tvp != NULL)
3806 if (ap->a_tdvp == ap->a_tvp)
3815 vop_rename_pre(void *ap)
3817 struct vop_rename_args *a = ap;
3819 #ifdef DEBUG_VFS_LOCKS
3821 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3822 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3823 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3824 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3826 /* Check the source (from). */
3827 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
3828 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
3829 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3830 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
3831 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3833 /* Check the target. */
3835 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3836 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3838 if (a->a_tdvp != a->a_fdvp)
3840 if (a->a_tvp != a->a_fvp)
3848 vop_strategy_pre(void *ap)
3850 #ifdef DEBUG_VFS_LOCKS
3851 struct vop_strategy_args *a;
3858 * Cluster ops lock their component buffers but not the IO container.
3860 if ((bp->b_flags & B_CLUSTER) != 0)
3863 if (!BUF_ISLOCKED(bp)) {
3864 if (vfs_badlock_print)
3866 "VOP_STRATEGY: bp is not locked but should be\n");
3867 if (vfs_badlock_ddb)
3868 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3874 vop_lookup_pre(void *ap)
3876 #ifdef DEBUG_VFS_LOCKS
3877 struct vop_lookup_args *a;
3882 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3883 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3888 vop_lookup_post(void *ap, int rc)
3890 #ifdef DEBUG_VFS_LOCKS
3891 struct vop_lookup_args *a;
3899 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3900 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3903 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3908 vop_lock_pre(void *ap)
3910 #ifdef DEBUG_VFS_LOCKS
3911 struct vop_lock1_args *a = ap;
3913 if ((a->a_flags & LK_INTERLOCK) == 0)
3914 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3916 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3921 vop_lock_post(void *ap, int rc)
3923 #ifdef DEBUG_VFS_LOCKS
3924 struct vop_lock1_args *a = ap;
3926 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3928 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3933 vop_unlock_pre(void *ap)
3935 #ifdef DEBUG_VFS_LOCKS
3936 struct vop_unlock_args *a = ap;
3938 if (a->a_flags & LK_INTERLOCK)
3939 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3940 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3945 vop_unlock_post(void *ap, int rc)
3947 #ifdef DEBUG_VFS_LOCKS
3948 struct vop_unlock_args *a = ap;
3950 if (a->a_flags & LK_INTERLOCK)
3951 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3956 vop_create_post(void *ap, int rc)
3958 struct vop_create_args *a = ap;
3961 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3965 vop_link_post(void *ap, int rc)
3967 struct vop_link_args *a = ap;
3970 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3971 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3976 vop_mkdir_post(void *ap, int rc)
3978 struct vop_mkdir_args *a = ap;
3981 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3985 vop_mknod_post(void *ap, int rc)
3987 struct vop_mknod_args *a = ap;
3990 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3994 vop_remove_post(void *ap, int rc)
3996 struct vop_remove_args *a = ap;
3999 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4000 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4005 vop_rename_post(void *ap, int rc)
4007 struct vop_rename_args *a = ap;
4010 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4011 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4012 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4014 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4016 if (a->a_tdvp != a->a_fdvp)
4018 if (a->a_tvp != a->a_fvp)
4026 vop_rmdir_post(void *ap, int rc)
4028 struct vop_rmdir_args *a = ap;
4031 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4032 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4037 vop_setattr_post(void *ap, int rc)
4039 struct vop_setattr_args *a = ap;
4042 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4046 vop_symlink_post(void *ap, int rc)
4048 struct vop_symlink_args *a = ap;
4051 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4054 static struct knlist fs_knlist;
4057 vfs_event_init(void *arg)
4059 knlist_init_mtx(&fs_knlist, NULL);
4061 /* XXX - correct order? */
4062 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4065 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
4068 KNOTE_UNLOCKED(&fs_knlist, event);
4071 static int filt_fsattach(struct knote *kn);
4072 static void filt_fsdetach(struct knote *kn);
4073 static int filt_fsevent(struct knote *kn, long hint);
4075 struct filterops fs_filtops =
4076 { 0, filt_fsattach, filt_fsdetach, filt_fsevent };
4079 filt_fsattach(struct knote *kn)
4082 kn->kn_flags |= EV_CLEAR;
4083 knlist_add(&fs_knlist, kn, 0);
4088 filt_fsdetach(struct knote *kn)
4091 knlist_remove(&fs_knlist, kn, 0);
4095 filt_fsevent(struct knote *kn, long hint)
4098 kn->kn_fflags |= hint;
4099 return (kn->kn_fflags != 0);
4103 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4109 error = SYSCTL_IN(req, &vc, sizeof(vc));
4112 if (vc.vc_vers != VFS_CTL_VERS1)
4114 mp = vfs_getvfs(&vc.vc_fsid);
4117 /* ensure that a specific sysctl goes to the right filesystem. */
4118 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4119 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4123 VCTLTOREQ(&vc, req);
4124 error = VFS_SYSCTL(mp, vc.vc_op, req);
4129 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
4133 * Function to initialize a va_filerev field sensibly.
4134 * XXX: Wouldn't a random number make a lot more sense ??
4137 init_va_filerev(void)
4142 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4145 static int filt_vfsread(struct knote *kn, long hint);
4146 static int filt_vfswrite(struct knote *kn, long hint);
4147 static int filt_vfsvnode(struct knote *kn, long hint);
4148 static void filt_vfsdetach(struct knote *kn);
4149 static struct filterops vfsread_filtops =
4150 { 1, NULL, filt_vfsdetach, filt_vfsread };
4151 static struct filterops vfswrite_filtops =
4152 { 1, NULL, filt_vfsdetach, filt_vfswrite };
4153 static struct filterops vfsvnode_filtops =
4154 { 1, NULL, filt_vfsdetach, filt_vfsvnode };
4157 vfs_knllock(void *arg)
4159 struct vnode *vp = arg;
4161 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4165 vfs_knlunlock(void *arg)
4167 struct vnode *vp = arg;
4173 vfs_knl_assert_locked(void *arg)
4175 #ifdef DEBUG_VFS_LOCKS
4176 struct vnode *vp = arg;
4178 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4183 vfs_knl_assert_unlocked(void *arg)
4185 #ifdef DEBUG_VFS_LOCKS
4186 struct vnode *vp = arg;
4188 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4193 vfs_kqfilter(struct vop_kqfilter_args *ap)
4195 struct vnode *vp = ap->a_vp;
4196 struct knote *kn = ap->a_kn;
4199 switch (kn->kn_filter) {
4201 kn->kn_fop = &vfsread_filtops;
4204 kn->kn_fop = &vfswrite_filtops;
4207 kn->kn_fop = &vfsvnode_filtops;
4213 kn->kn_hook = (caddr_t)vp;
4216 if (vp->v_pollinfo == NULL)
4218 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4219 knlist_add(knl, kn, 0);
4225 * Detach knote from vnode
4228 filt_vfsdetach(struct knote *kn)
4230 struct vnode *vp = (struct vnode *)kn->kn_hook;
4232 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4233 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4238 filt_vfsread(struct knote *kn, long hint)
4240 struct vnode *vp = (struct vnode *)kn->kn_hook;
4245 * filesystem is gone, so set the EOF flag and schedule
4246 * the knote for deletion.
4248 if (hint == NOTE_REVOKE) {
4250 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4255 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4259 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4260 res = (kn->kn_data != 0);
4267 filt_vfswrite(struct knote *kn, long hint)
4269 struct vnode *vp = (struct vnode *)kn->kn_hook;
4274 * filesystem is gone, so set the EOF flag and schedule
4275 * the knote for deletion.
4277 if (hint == NOTE_REVOKE)
4278 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4286 filt_vfsvnode(struct knote *kn, long hint)
4288 struct vnode *vp = (struct vnode *)kn->kn_hook;
4292 if (kn->kn_sfflags & hint)
4293 kn->kn_fflags |= hint;
4294 if (hint == NOTE_REVOKE) {
4295 kn->kn_flags |= EV_EOF;
4299 res = (kn->kn_fflags != 0);
4305 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4309 if (dp->d_reclen > ap->a_uio->uio_resid)
4310 return (ENAMETOOLONG);
4311 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4313 if (ap->a_ncookies != NULL) {
4314 if (ap->a_cookies != NULL)
4315 free(ap->a_cookies, M_TEMP);
4316 ap->a_cookies = NULL;
4317 *ap->a_ncookies = 0;
4321 if (ap->a_ncookies == NULL)
4324 KASSERT(ap->a_cookies,
4325 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4327 *ap->a_cookies = realloc(*ap->a_cookies,
4328 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4329 (*ap->a_cookies)[*ap->a_ncookies] = off;
4334 * Mark for update the access time of the file if the filesystem
4335 * supports VOP_MARKATIME. This functionality is used by execve and
4336 * mmap, so we want to avoid the I/O implied by directly setting
4337 * va_atime for the sake of efficiency.
4340 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4345 VFS_ASSERT_GIANT(mp);
4346 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4347 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4348 (void)VOP_MARKATIME(vp);
4352 * The purpose of this routine is to remove granularity from accmode_t,
4353 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4354 * VADMIN and VAPPEND.
4356 * If it returns 0, the caller is supposed to continue with the usual
4357 * access checks using 'accmode' as modified by this routine. If it
4358 * returns nonzero value, the caller is supposed to return that value
4361 * Note that after this routine runs, accmode may be zero.
4364 vfs_unixify_accmode(accmode_t *accmode)
4367 * There is no way to specify explicit "deny" rule using
4368 * file mode or POSIX.1e ACLs.
4370 if (*accmode & VEXPLICIT_DENY) {
4376 * None of these can be translated into usual access bits.
4377 * Also, the common case for NFSv4 ACLs is to not contain
4378 * either of these bits. Caller should check for VWRITE
4379 * on the containing directory instead.
4381 if (*accmode & (VDELETE_CHILD | VDELETE))
4384 if (*accmode & VADMIN_PERMS) {
4385 *accmode &= ~VADMIN_PERMS;
4390 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4391 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4393 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);