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("$FreeBSD$");
45 #include "opt_watchdog.h"
47 #include <sys/param.h>
48 #include <sys/systm.h>
51 #include <sys/condvar.h>
53 #include <sys/dirent.h>
54 #include <sys/event.h>
55 #include <sys/eventhandler.h>
56 #include <sys/extattr.h>
58 #include <sys/fcntl.h>
61 #include <sys/kernel.h>
62 #include <sys/kthread.h>
63 #include <sys/lockf.h>
64 #include <sys/malloc.h>
65 #include <sys/mount.h>
66 #include <sys/namei.h>
68 #include <sys/reboot.h>
69 #include <sys/sched.h>
70 #include <sys/sleepqueue.h>
72 #include <sys/sysctl.h>
73 #include <sys/syslog.h>
74 #include <sys/vmmeter.h>
75 #include <sys/vnode.h>
77 #include <sys/watchdog.h>
80 #include <machine/stdarg.h>
82 #include <security/mac/mac_framework.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_extern.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_kern.h>
100 static void delmntque(struct vnode *vp);
101 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
102 int slpflag, int slptimeo);
103 static void syncer_shutdown(void *arg, int howto);
104 static int vtryrecycle(struct vnode *vp);
105 static void vbusy(struct vnode *vp);
106 static void vinactive(struct vnode *, struct thread *);
107 static void v_incr_usecount(struct vnode *);
108 static void v_decr_usecount(struct vnode *);
109 static void v_decr_useonly(struct vnode *);
110 static void v_upgrade_usecount(struct vnode *);
111 static void vfree(struct vnode *);
112 static void vnlru_free(int);
113 static void vgonel(struct vnode *);
114 static void vfs_knllock(void *arg);
115 static void vfs_knlunlock(void *arg);
116 static void vfs_knl_assert_locked(void *arg);
117 static void vfs_knl_assert_unlocked(void *arg);
118 static void destroy_vpollinfo(struct vpollinfo *vi);
121 * Number of vnodes in existence. Increased whenever getnewvnode()
122 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
125 static unsigned long numvnodes;
127 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
128 "Number of vnodes in existence");
131 * Conversion tables for conversion from vnode types to inode formats
134 enum vtype iftovt_tab[16] = {
135 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
136 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
138 int vttoif_tab[10] = {
139 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
140 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
144 * List of vnodes that are ready for recycling.
146 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
149 * Free vnode target. Free vnodes may simply be files which have been stat'd
150 * but not read. This is somewhat common, and a small cache of such files
151 * should be kept to avoid recreation costs.
153 static u_long wantfreevnodes;
154 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
155 /* Number of vnodes in the free list. */
156 static u_long freevnodes;
157 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
158 "Number of vnodes in the free list");
160 static int vlru_allow_cache_src;
161 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
162 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
165 * Various variables used for debugging the new implementation of
167 * XXX these are probably of (very) limited utility now.
169 static int reassignbufcalls;
170 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
171 "Number of calls to reassignbuf");
174 * Cache for the mount type id assigned to NFS. This is used for
175 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
177 int nfs_mount_type = -1;
179 /* To keep more than one thread at a time from running vfs_getnewfsid */
180 static struct mtx mntid_mtx;
183 * Lock for any access to the following:
188 static struct mtx vnode_free_list_mtx;
190 /* Publicly exported FS */
191 struct nfs_public nfs_pub;
193 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
194 static uma_zone_t vnode_zone;
195 static uma_zone_t vnodepoll_zone;
198 * The workitem queue.
200 * It is useful to delay writes of file data and filesystem metadata
201 * for tens of seconds so that quickly created and deleted files need
202 * not waste disk bandwidth being created and removed. To realize this,
203 * we append vnodes to a "workitem" queue. When running with a soft
204 * updates implementation, most pending metadata dependencies should
205 * not wait for more than a few seconds. Thus, mounted on block devices
206 * are delayed only about a half the time that file data is delayed.
207 * Similarly, directory updates are more critical, so are only delayed
208 * about a third the time that file data is delayed. Thus, there are
209 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
210 * one each second (driven off the filesystem syncer process). The
211 * syncer_delayno variable indicates the next queue that is to be processed.
212 * Items that need to be processed soon are placed in this queue:
214 * syncer_workitem_pending[syncer_delayno]
216 * A delay of fifteen seconds is done by placing the request fifteen
217 * entries later in the queue:
219 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
222 static int syncer_delayno;
223 static long syncer_mask;
224 LIST_HEAD(synclist, bufobj);
225 static struct synclist *syncer_workitem_pending[2];
227 * The sync_mtx protects:
232 * syncer_workitem_pending
233 * syncer_worklist_len
236 static struct mtx sync_mtx;
237 static struct cv sync_wakeup;
239 #define SYNCER_MAXDELAY 32
240 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
241 static int syncdelay = 30; /* max time to delay syncing data */
242 static int filedelay = 30; /* time to delay syncing files */
243 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
244 "Time to delay syncing files (in seconds)");
245 static int dirdelay = 29; /* time to delay syncing directories */
246 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
247 "Time to delay syncing directories (in seconds)");
248 static int metadelay = 28; /* time to delay syncing metadata */
249 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
250 "Time to delay syncing metadata (in seconds)");
251 static int rushjob; /* number of slots to run ASAP */
252 static int stat_rush_requests; /* number of times I/O speeded up */
253 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
254 "Number of times I/O speeded up (rush requests)");
257 * When shutting down the syncer, run it at four times normal speed.
259 #define SYNCER_SHUTDOWN_SPEEDUP 4
260 static int sync_vnode_count;
261 static int syncer_worklist_len;
262 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
266 * Number of vnodes we want to exist at any one time. This is mostly used
267 * to size hash tables in vnode-related code. It is normally not used in
268 * getnewvnode(), as wantfreevnodes is normally nonzero.)
270 * XXX desiredvnodes is historical cruft and should not exist.
273 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
274 &desiredvnodes, 0, "Maximum number of vnodes");
275 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
276 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
277 static int vnlru_nowhere;
278 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
279 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
282 * Macros to control when a vnode is freed and recycled. All require
283 * the vnode interlock.
285 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
286 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
287 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
291 * Initialize the vnode management data structures.
293 * Reevaluate the following cap on the number of vnodes after the physical
294 * memory size exceeds 512GB. In the limit, as the physical memory size
295 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
297 #ifndef MAXVNODES_MAX
298 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
301 vntblinit(void *dummy __unused)
303 int physvnodes, virtvnodes;
306 * Desiredvnodes is a function of the physical memory size and the
307 * kernel's heap size. Generally speaking, it scales with the
308 * physical memory size. The ratio of desiredvnodes to physical pages
309 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
310 * marginal ratio of desiredvnodes to physical pages is one to
311 * sixteen. However, desiredvnodes is limited by the kernel's heap
312 * size. The memory required by desiredvnodes vnodes and vm objects
313 * may not exceed one seventh of the kernel's heap size.
315 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
316 cnt.v_page_count) / 16;
317 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
318 sizeof(struct vnode)));
319 desiredvnodes = min(physvnodes, virtvnodes);
320 if (desiredvnodes > MAXVNODES_MAX) {
322 printf("Reducing kern.maxvnodes %d -> %d\n",
323 desiredvnodes, MAXVNODES_MAX);
324 desiredvnodes = MAXVNODES_MAX;
326 wantfreevnodes = desiredvnodes / 4;
327 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
328 TAILQ_INIT(&vnode_free_list);
329 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
330 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
331 NULL, NULL, UMA_ALIGN_PTR, 0);
332 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
333 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
335 * Initialize the filesystem syncer.
337 syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
339 syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
341 syncer_maxdelay = syncer_mask + 1;
342 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
343 cv_init(&sync_wakeup, "syncer");
345 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
349 * Mark a mount point as busy. Used to synchronize access and to delay
350 * unmounting. Eventually, mountlist_mtx is not released on failure.
352 * vfs_busy() is a custom lock, it can block the caller.
353 * vfs_busy() only sleeps if the unmount is active on the mount point.
354 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
355 * vnode belonging to mp.
357 * Lookup uses vfs_busy() to traverse mount points.
359 * / vnode lock A / vnode lock (/var) D
360 * /var vnode lock B /log vnode lock(/var/log) E
361 * vfs_busy lock C vfs_busy lock F
363 * Within each file system, the lock order is C->A->B and F->D->E.
365 * When traversing across mounts, the system follows that lock order:
371 * The lookup() process for namei("/var") illustrates the process:
372 * VOP_LOOKUP() obtains B while A is held
373 * vfs_busy() obtains a shared lock on F while A and B are held
374 * vput() releases lock on B
375 * vput() releases lock on A
376 * VFS_ROOT() obtains lock on D while shared lock on F is held
377 * vfs_unbusy() releases shared lock on F
378 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
379 * Attempt to lock A (instead of vp_crossmp) while D is held would
380 * violate the global order, causing deadlocks.
382 * dounmount() locks B while F is drained.
385 vfs_busy(struct mount *mp, int flags)
388 MPASS((flags & ~MBF_MASK) == 0);
389 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
394 * If mount point is currenly being unmounted, sleep until the
395 * mount point fate is decided. If thread doing the unmounting fails,
396 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
397 * that this mount point has survived the unmount attempt and vfs_busy
398 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
399 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
400 * about to be really destroyed. vfs_busy needs to release its
401 * reference on the mount point in this case and return with ENOENT,
402 * telling the caller that mount mount it tried to busy is no longer
405 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
406 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
409 CTR1(KTR_VFS, "%s: failed busying before sleeping",
413 if (flags & MBF_MNTLSTLOCK)
414 mtx_unlock(&mountlist_mtx);
415 mp->mnt_kern_flag |= MNTK_MWAIT;
416 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
417 if (flags & MBF_MNTLSTLOCK)
418 mtx_lock(&mountlist_mtx);
421 if (flags & MBF_MNTLSTLOCK)
422 mtx_unlock(&mountlist_mtx);
429 * Free a busy filesystem.
432 vfs_unbusy(struct mount *mp)
435 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
438 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
440 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
441 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
442 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
443 mp->mnt_kern_flag &= ~MNTK_DRAINING;
444 wakeup(&mp->mnt_lockref);
450 * Lookup a mount point by filesystem identifier.
453 vfs_getvfs(fsid_t *fsid)
457 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
458 mtx_lock(&mountlist_mtx);
459 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
460 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
461 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
463 mtx_unlock(&mountlist_mtx);
467 mtx_unlock(&mountlist_mtx);
468 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
469 return ((struct mount *) 0);
473 * Lookup a mount point by filesystem identifier, busying it before
477 vfs_busyfs(fsid_t *fsid)
482 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
483 mtx_lock(&mountlist_mtx);
484 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
485 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
486 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
487 error = vfs_busy(mp, MBF_MNTLSTLOCK);
489 mtx_unlock(&mountlist_mtx);
495 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
496 mtx_unlock(&mountlist_mtx);
497 return ((struct mount *) 0);
501 * Check if a user can access privileged mount options.
504 vfs_suser(struct mount *mp, struct thread *td)
509 * If the thread is jailed, but this is not a jail-friendly file
510 * system, deny immediately.
512 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
516 * If the file system was mounted outside the jail of the calling
517 * thread, deny immediately.
519 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
523 * If file system supports delegated administration, we don't check
524 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
525 * by the file system itself.
526 * If this is not the user that did original mount, we check for
527 * the PRIV_VFS_MOUNT_OWNER privilege.
529 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
530 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
531 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
538 * Get a new unique fsid. Try to make its val[0] unique, since this value
539 * will be used to create fake device numbers for stat(). Also try (but
540 * not so hard) make its val[0] unique mod 2^16, since some emulators only
541 * support 16-bit device numbers. We end up with unique val[0]'s for the
542 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
544 * Keep in mind that several mounts may be running in parallel. Starting
545 * the search one past where the previous search terminated is both a
546 * micro-optimization and a defense against returning the same fsid to
550 vfs_getnewfsid(struct mount *mp)
552 static uint16_t mntid_base;
557 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
558 mtx_lock(&mntid_mtx);
559 mtype = mp->mnt_vfc->vfc_typenum;
560 tfsid.val[1] = mtype;
561 mtype = (mtype & 0xFF) << 24;
563 tfsid.val[0] = makedev(255,
564 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
566 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
570 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
571 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
572 mtx_unlock(&mntid_mtx);
576 * Knob to control the precision of file timestamps:
578 * 0 = seconds only; nanoseconds zeroed.
579 * 1 = seconds and nanoseconds, accurate within 1/HZ.
580 * 2 = seconds and nanoseconds, truncated to microseconds.
581 * >=3 = seconds and nanoseconds, maximum precision.
583 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
585 static int timestamp_precision = TSP_SEC;
586 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
587 ×tamp_precision, 0, "File timestamp precision (0: seconds, "
588 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
589 "3+: sec + ns (max. precision))");
592 * Get a current timestamp.
595 vfs_timestamp(struct timespec *tsp)
599 switch (timestamp_precision) {
601 tsp->tv_sec = time_second;
609 TIMEVAL_TO_TIMESPEC(&tv, tsp);
619 * Set vnode attributes to VNOVAL
622 vattr_null(struct vattr *vap)
626 vap->va_size = VNOVAL;
627 vap->va_bytes = VNOVAL;
628 vap->va_mode = VNOVAL;
629 vap->va_nlink = VNOVAL;
630 vap->va_uid = VNOVAL;
631 vap->va_gid = VNOVAL;
632 vap->va_fsid = VNOVAL;
633 vap->va_fileid = VNOVAL;
634 vap->va_blocksize = VNOVAL;
635 vap->va_rdev = VNOVAL;
636 vap->va_atime.tv_sec = VNOVAL;
637 vap->va_atime.tv_nsec = VNOVAL;
638 vap->va_mtime.tv_sec = VNOVAL;
639 vap->va_mtime.tv_nsec = VNOVAL;
640 vap->va_ctime.tv_sec = VNOVAL;
641 vap->va_ctime.tv_nsec = VNOVAL;
642 vap->va_birthtime.tv_sec = VNOVAL;
643 vap->va_birthtime.tv_nsec = VNOVAL;
644 vap->va_flags = VNOVAL;
645 vap->va_gen = VNOVAL;
650 * This routine is called when we have too many vnodes. It attempts
651 * to free <count> vnodes and will potentially free vnodes that still
652 * have VM backing store (VM backing store is typically the cause
653 * of a vnode blowout so we want to do this). Therefore, this operation
654 * is not considered cheap.
656 * A number of conditions may prevent a vnode from being reclaimed.
657 * the buffer cache may have references on the vnode, a directory
658 * vnode may still have references due to the namei cache representing
659 * underlying files, or the vnode may be in active use. It is not
660 * desireable to reuse such vnodes. These conditions may cause the
661 * number of vnodes to reach some minimum value regardless of what
662 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
665 vlrureclaim(struct mount *mp)
674 * Calculate the trigger point, don't allow user
675 * screwups to blow us up. This prevents us from
676 * recycling vnodes with lots of resident pages. We
677 * aren't trying to free memory, we are trying to
680 usevnodes = desiredvnodes;
683 trigger = cnt.v_page_count * 2 / usevnodes;
685 vn_start_write(NULL, &mp, V_WAIT);
687 count = mp->mnt_nvnodelistsize / 10 + 1;
689 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
690 while (vp != NULL && vp->v_type == VMARKER)
691 vp = TAILQ_NEXT(vp, v_nmntvnodes);
694 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
695 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
700 * If it's been deconstructed already, it's still
701 * referenced, or it exceeds the trigger, skip it.
703 if (vp->v_usecount ||
704 (!vlru_allow_cache_src &&
705 !LIST_EMPTY(&(vp)->v_cache_src)) ||
706 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
707 vp->v_object->resident_page_count > trigger)) {
713 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
715 goto next_iter_mntunlocked;
719 * v_usecount may have been bumped after VOP_LOCK() dropped
720 * the vnode interlock and before it was locked again.
722 * It is not necessary to recheck VI_DOOMED because it can
723 * only be set by another thread that holds both the vnode
724 * lock and vnode interlock. If another thread has the
725 * vnode lock before we get to VOP_LOCK() and obtains the
726 * vnode interlock after VOP_LOCK() drops the vnode
727 * interlock, the other thread will be unable to drop the
728 * vnode lock before our VOP_LOCK() call fails.
730 if (vp->v_usecount ||
731 (!vlru_allow_cache_src &&
732 !LIST_EMPTY(&(vp)->v_cache_src)) ||
733 (vp->v_object != NULL &&
734 vp->v_object->resident_page_count > trigger)) {
735 VOP_UNLOCK(vp, LK_INTERLOCK);
736 goto next_iter_mntunlocked;
738 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
739 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
744 next_iter_mntunlocked:
753 kern_yield(PRI_UNCHANGED);
758 vn_finished_write(mp);
763 * Attempt to keep the free list at wantfreevnodes length.
766 vnlru_free(int count)
771 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
772 for (; count > 0; count--) {
773 vp = TAILQ_FIRST(&vnode_free_list);
775 * The list can be modified while the free_list_mtx
776 * has been dropped and vp could be NULL here.
780 VNASSERT(vp->v_op != NULL, vp,
781 ("vnlru_free: vnode already reclaimed."));
782 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
784 * Don't recycle if we can't get the interlock.
786 if (!VI_TRYLOCK(vp)) {
787 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
790 VNASSERT(VCANRECYCLE(vp), vp,
791 ("vp inconsistent on freelist"));
793 vp->v_iflag &= ~VI_FREE;
795 mtx_unlock(&vnode_free_list_mtx);
797 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
799 VFS_UNLOCK_GIANT(vfslocked);
801 * If the recycled succeeded this vdrop will actually free
802 * the vnode. If not it will simply place it back on
806 mtx_lock(&vnode_free_list_mtx);
810 * Attempt to recycle vnodes in a context that is always safe to block.
811 * Calling vlrurecycle() from the bowels of filesystem code has some
812 * interesting deadlock problems.
814 static struct proc *vnlruproc;
815 static int vnlruproc_sig;
820 struct mount *mp, *nmp;
822 struct proc *p = vnlruproc;
824 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
828 kproc_suspend_check(p);
829 mtx_lock(&vnode_free_list_mtx);
830 if (freevnodes > wantfreevnodes)
831 vnlru_free(freevnodes - wantfreevnodes);
832 if (numvnodes <= desiredvnodes * 9 / 10) {
834 wakeup(&vnlruproc_sig);
835 msleep(vnlruproc, &vnode_free_list_mtx,
836 PVFS|PDROP, "vlruwt", hz);
839 mtx_unlock(&vnode_free_list_mtx);
841 mtx_lock(&mountlist_mtx);
842 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
843 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
844 nmp = TAILQ_NEXT(mp, mnt_list);
847 vfslocked = VFS_LOCK_GIANT(mp);
848 done += vlrureclaim(mp);
849 VFS_UNLOCK_GIANT(vfslocked);
850 mtx_lock(&mountlist_mtx);
851 nmp = TAILQ_NEXT(mp, mnt_list);
854 mtx_unlock(&mountlist_mtx);
857 /* These messages are temporary debugging aids */
858 if (vnlru_nowhere < 5)
859 printf("vnlru process getting nowhere..\n");
860 else if (vnlru_nowhere == 5)
861 printf("vnlru process messages stopped.\n");
864 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
866 kern_yield(PRI_UNCHANGED);
870 static struct kproc_desc vnlru_kp = {
875 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
879 * Routines having to do with the management of the vnode table.
883 vdestroy(struct vnode *vp)
887 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
888 mtx_lock(&vnode_free_list_mtx);
890 mtx_unlock(&vnode_free_list_mtx);
892 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
893 ("cleaned vnode still on the free list."));
894 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
895 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
896 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
897 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
898 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
899 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
900 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
901 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
902 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
903 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
904 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
905 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
908 mac_vnode_destroy(vp);
910 if (vp->v_pollinfo != NULL)
911 destroy_vpollinfo(vp->v_pollinfo);
913 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
916 lockdestroy(vp->v_vnlock);
917 mtx_destroy(&vp->v_interlock);
918 mtx_destroy(BO_MTX(bo));
919 uma_zfree(vnode_zone, vp);
923 * Try to recycle a freed vnode. We abort if anyone picks up a reference
924 * before we actually vgone(). This function must be called with the vnode
925 * held to prevent the vnode from being returned to the free list midway
929 vtryrecycle(struct vnode *vp)
933 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
934 VNASSERT(vp->v_holdcnt, vp,
935 ("vtryrecycle: Recycling vp %p without a reference.", vp));
937 * This vnode may found and locked via some other list, if so we
938 * can't recycle it yet.
940 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
942 "%s: impossible to recycle, vp %p lock is already held",
944 return (EWOULDBLOCK);
947 * Don't recycle if its filesystem is being suspended.
949 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
952 "%s: impossible to recycle, cannot start the write for %p",
957 * If we got this far, we need to acquire the interlock and see if
958 * anyone picked up this vnode from another list. If not, we will
959 * mark it with DOOMED via vgonel() so that anyone who does find it
963 if (vp->v_usecount) {
964 VOP_UNLOCK(vp, LK_INTERLOCK);
965 vn_finished_write(vnmp);
967 "%s: impossible to recycle, %p is already referenced",
971 if ((vp->v_iflag & VI_DOOMED) == 0)
973 VOP_UNLOCK(vp, LK_INTERLOCK);
974 vn_finished_write(vnmp);
979 * Return the next vnode from the free list.
982 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
985 struct vnode *vp = NULL;
988 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
989 mtx_lock(&vnode_free_list_mtx);
991 * Lend our context to reclaim vnodes if they've exceeded the max.
993 if (freevnodes > wantfreevnodes)
996 * Wait for available vnodes.
998 if (numvnodes > desiredvnodes) {
999 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
1001 * File system is beeing suspended, we cannot risk a
1002 * deadlock here, so allocate new vnode anyway.
1004 if (freevnodes > wantfreevnodes)
1005 vnlru_free(freevnodes - wantfreevnodes);
1008 if (vnlruproc_sig == 0) {
1009 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
1012 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
1014 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1015 if (numvnodes > desiredvnodes) {
1016 mtx_unlock(&vnode_free_list_mtx);
1023 mtx_unlock(&vnode_free_list_mtx);
1024 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1028 vp->v_vnlock = &vp->v_lock;
1029 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1031 * By default, don't allow shared locks unless filesystems
1034 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
1036 * Initialize bufobj.
1039 bo->__bo_vnode = vp;
1040 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
1041 bo->bo_ops = &buf_ops_bio;
1042 bo->bo_private = vp;
1043 TAILQ_INIT(&bo->bo_clean.bv_hd);
1044 TAILQ_INIT(&bo->bo_dirty.bv_hd);
1046 * Initialize namecache.
1048 LIST_INIT(&vp->v_cache_src);
1049 TAILQ_INIT(&vp->v_cache_dst);
1051 * Finalize various vnode identity bits.
1056 v_incr_usecount(vp);
1060 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1061 mac_vnode_associate_singlelabel(mp, vp);
1062 else if (mp == NULL && vops != &dead_vnodeops)
1063 printf("NULL mp in getnewvnode()\n");
1066 bo->bo_bsize = mp->mnt_stat.f_iosize;
1067 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1068 vp->v_vflag |= VV_NOKNOTE;
1076 * Delete from old mount point vnode list, if on one.
1079 delmntque(struct vnode *vp)
1088 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1089 ("bad mount point vnode list size"));
1090 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1091 mp->mnt_nvnodelistsize--;
1097 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1101 vp->v_op = &dead_vnodeops;
1102 /* XXX non mp-safe fs may still call insmntque with vnode
1104 if (!VOP_ISLOCKED(vp))
1105 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1111 * Insert into list of vnodes for the new mount point, if available.
1114 insmntque1(struct vnode *vp, struct mount *mp,
1115 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1119 KASSERT(vp->v_mount == NULL,
1120 ("insmntque: vnode already on per mount vnode list"));
1121 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1122 #ifdef DEBUG_VFS_LOCKS
1123 if (!VFS_NEEDSGIANT(mp))
1124 ASSERT_VOP_ELOCKED(vp,
1125 "insmntque: mp-safe fs and non-locked vp");
1128 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1129 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1130 mp->mnt_nvnodelistsize == 0)) {
1131 locked = VOP_ISLOCKED(vp);
1132 if (!locked || (locked == LK_EXCLUSIVE &&
1133 (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1142 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1143 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1144 ("neg mount point vnode list size"));
1145 mp->mnt_nvnodelistsize++;
1151 insmntque(struct vnode *vp, struct mount *mp)
1154 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1158 * Flush out and invalidate all buffers associated with a bufobj
1159 * Called with the underlying object locked.
1162 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1167 if (flags & V_SAVE) {
1168 error = bufobj_wwait(bo, slpflag, slptimeo);
1173 if (bo->bo_dirty.bv_cnt > 0) {
1175 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1178 * XXX We could save a lock/unlock if this was only
1179 * enabled under INVARIANTS
1182 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1183 panic("vinvalbuf: dirty bufs");
1187 * If you alter this loop please notice that interlock is dropped and
1188 * reacquired in flushbuflist. Special care is needed to ensure that
1189 * no race conditions occur from this.
1192 error = flushbuflist(&bo->bo_clean,
1193 flags, bo, slpflag, slptimeo);
1194 if (error == 0 && !(flags & V_CLEANONLY))
1195 error = flushbuflist(&bo->bo_dirty,
1196 flags, bo, slpflag, slptimeo);
1197 if (error != 0 && error != EAGAIN) {
1201 } while (error != 0);
1204 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1205 * have write I/O in-progress but if there is a VM object then the
1206 * VM object can also have read-I/O in-progress.
1209 bufobj_wwait(bo, 0, 0);
1211 if (bo->bo_object != NULL) {
1212 VM_OBJECT_LOCK(bo->bo_object);
1213 vm_object_pip_wait(bo->bo_object, "bovlbx");
1214 VM_OBJECT_UNLOCK(bo->bo_object);
1217 } while (bo->bo_numoutput > 0);
1221 * Destroy the copy in the VM cache, too.
1223 if (bo->bo_object != NULL &&
1224 (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1225 VM_OBJECT_LOCK(bo->bo_object);
1226 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1227 OBJPR_CLEANONLY : 0);
1228 VM_OBJECT_UNLOCK(bo->bo_object);
1233 if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1234 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1235 panic("vinvalbuf: flush failed");
1242 * Flush out and invalidate all buffers associated with a vnode.
1243 * Called with the underlying object locked.
1246 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1249 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1250 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1251 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1255 * Flush out buffers on the specified list.
1259 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1262 struct buf *bp, *nbp;
1267 ASSERT_BO_LOCKED(bo);
1270 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1271 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1272 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1278 lblkno = nbp->b_lblkno;
1279 xflags = nbp->b_xflags &
1280 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1283 error = BUF_TIMELOCK(bp,
1284 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1285 "flushbuf", slpflag, slptimeo);
1288 return (error != ENOLCK ? error : EAGAIN);
1290 KASSERT(bp->b_bufobj == bo,
1291 ("bp %p wrong b_bufobj %p should be %p",
1292 bp, bp->b_bufobj, bo));
1293 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1299 * XXX Since there are no node locks for NFS, I
1300 * believe there is a slight chance that a delayed
1301 * write will occur while sleeping just above, so
1304 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1309 bp->b_flags |= B_ASYNC;
1312 return (EAGAIN); /* XXX: why not loop ? */
1317 bp->b_flags |= (B_INVAL | B_RELBUF);
1318 bp->b_flags &= ~B_ASYNC;
1322 (nbp->b_bufobj != bo ||
1323 nbp->b_lblkno != lblkno ||
1325 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1326 break; /* nbp invalid */
1332 * Truncate a file's buffer and pages to a specified length. This
1333 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1337 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1338 off_t length, int blksize)
1340 struct buf *bp, *nbp;
1345 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1346 vp, cred, blksize, (uintmax_t)length);
1349 * Round up to the *next* lbn.
1351 trunclbn = (length + blksize - 1) / blksize;
1353 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1360 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1361 if (bp->b_lblkno < trunclbn)
1364 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1365 BO_MTX(bo)) == ENOLCK)
1371 bp->b_flags |= (B_INVAL | B_RELBUF);
1372 bp->b_flags &= ~B_ASYNC;
1378 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1379 (nbp->b_vp != vp) ||
1380 (nbp->b_flags & B_DELWRI))) {
1386 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1387 if (bp->b_lblkno < trunclbn)
1390 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1391 BO_MTX(bo)) == ENOLCK)
1396 bp->b_flags |= (B_INVAL | B_RELBUF);
1397 bp->b_flags &= ~B_ASYNC;
1403 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1404 (nbp->b_vp != vp) ||
1405 (nbp->b_flags & B_DELWRI) == 0)) {
1414 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1415 if (bp->b_lblkno > 0)
1418 * Since we hold the vnode lock this should only
1419 * fail if we're racing with the buf daemon.
1422 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1423 BO_MTX(bo)) == ENOLCK) {
1426 VNASSERT((bp->b_flags & B_DELWRI), vp,
1427 ("buf(%p) on dirty queue without DELWRI", bp));
1438 bufobj_wwait(bo, 0, 0);
1440 vnode_pager_setsize(vp, length);
1446 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1449 * NOTE: We have to deal with the special case of a background bitmap
1450 * buffer, a situation where two buffers will have the same logical
1451 * block offset. We want (1) only the foreground buffer to be accessed
1452 * in a lookup and (2) must differentiate between the foreground and
1453 * background buffer in the splay tree algorithm because the splay
1454 * tree cannot normally handle multiple entities with the same 'index'.
1455 * We accomplish this by adding differentiating flags to the splay tree's
1460 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1463 struct buf *lefttreemax, *righttreemin, *y;
1467 lefttreemax = righttreemin = &dummy;
1469 if (lblkno < root->b_lblkno ||
1470 (lblkno == root->b_lblkno &&
1471 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1472 if ((y = root->b_left) == NULL)
1474 if (lblkno < y->b_lblkno) {
1476 root->b_left = y->b_right;
1479 if ((y = root->b_left) == NULL)
1482 /* Link into the new root's right tree. */
1483 righttreemin->b_left = root;
1484 righttreemin = root;
1485 } else if (lblkno > root->b_lblkno ||
1486 (lblkno == root->b_lblkno &&
1487 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1488 if ((y = root->b_right) == NULL)
1490 if (lblkno > y->b_lblkno) {
1492 root->b_right = y->b_left;
1495 if ((y = root->b_right) == NULL)
1498 /* Link into the new root's left tree. */
1499 lefttreemax->b_right = root;
1506 /* Assemble the new root. */
1507 lefttreemax->b_right = root->b_left;
1508 righttreemin->b_left = root->b_right;
1509 root->b_left = dummy.b_right;
1510 root->b_right = dummy.b_left;
1515 buf_vlist_remove(struct buf *bp)
1520 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1521 ASSERT_BO_LOCKED(bp->b_bufobj);
1522 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1523 (BX_VNDIRTY|BX_VNCLEAN),
1524 ("buf_vlist_remove: Buf %p is on two lists", bp));
1525 if (bp->b_xflags & BX_VNDIRTY)
1526 bv = &bp->b_bufobj->bo_dirty;
1528 bv = &bp->b_bufobj->bo_clean;
1529 if (bp != bv->bv_root) {
1530 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1531 KASSERT(root == bp, ("splay lookup failed in remove"));
1533 if (bp->b_left == NULL) {
1536 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1537 root->b_right = bp->b_right;
1540 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1542 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1546 * Add the buffer to the sorted clean or dirty block list using a
1547 * splay tree algorithm.
1549 * NOTE: xflags is passed as a constant, optimizing this inline function!
1552 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1557 ASSERT_BO_LOCKED(bo);
1558 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1559 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1560 bp->b_xflags |= xflags;
1561 if (xflags & BX_VNDIRTY)
1566 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1570 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1571 } else if (bp->b_lblkno < root->b_lblkno ||
1572 (bp->b_lblkno == root->b_lblkno &&
1573 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1574 bp->b_left = root->b_left;
1576 root->b_left = NULL;
1577 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1579 bp->b_right = root->b_right;
1581 root->b_right = NULL;
1582 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1589 * Lookup a buffer using the splay tree. Note that we specifically avoid
1590 * shadow buffers used in background bitmap writes.
1592 * This code isn't quite efficient as it could be because we are maintaining
1593 * two sorted lists and do not know which list the block resides in.
1595 * During a "make buildworld" the desired buffer is found at one of
1596 * the roots more than 60% of the time. Thus, checking both roots
1597 * before performing either splay eliminates unnecessary splays on the
1598 * first tree splayed.
1601 gbincore(struct bufobj *bo, daddr_t lblkno)
1605 ASSERT_BO_LOCKED(bo);
1606 if ((bp = bo->bo_clean.bv_root) != NULL &&
1607 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1609 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1610 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1612 if ((bp = bo->bo_clean.bv_root) != NULL) {
1613 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1614 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1617 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1618 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1619 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1626 * Associate a buffer with a vnode.
1629 bgetvp(struct vnode *vp, struct buf *bp)
1634 ASSERT_BO_LOCKED(bo);
1635 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1637 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1638 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1639 ("bgetvp: bp already attached! %p", bp));
1642 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1643 bp->b_flags |= B_NEEDSGIANT;
1647 * Insert onto list for new vnode.
1649 buf_vlist_add(bp, bo, BX_VNCLEAN);
1653 * Disassociate a buffer from a vnode.
1656 brelvp(struct buf *bp)
1661 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1662 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1665 * Delete from old vnode list, if on one.
1667 vp = bp->b_vp; /* XXX */
1670 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1671 buf_vlist_remove(bp);
1673 panic("brelvp: Buffer %p not on queue.", bp);
1674 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1675 bo->bo_flag &= ~BO_ONWORKLST;
1676 mtx_lock(&sync_mtx);
1677 LIST_REMOVE(bo, bo_synclist);
1678 syncer_worklist_len--;
1679 mtx_unlock(&sync_mtx);
1681 bp->b_flags &= ~B_NEEDSGIANT;
1683 bp->b_bufobj = NULL;
1689 * Add an item to the syncer work queue.
1692 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1696 ASSERT_BO_LOCKED(bo);
1698 mtx_lock(&sync_mtx);
1699 if (bo->bo_flag & BO_ONWORKLST)
1700 LIST_REMOVE(bo, bo_synclist);
1702 bo->bo_flag |= BO_ONWORKLST;
1703 syncer_worklist_len++;
1706 if (delay > syncer_maxdelay - 2)
1707 delay = syncer_maxdelay - 2;
1708 slot = (syncer_delayno + delay) & syncer_mask;
1710 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1712 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1714 mtx_unlock(&sync_mtx);
1718 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1722 mtx_lock(&sync_mtx);
1723 len = syncer_worklist_len - sync_vnode_count;
1724 mtx_unlock(&sync_mtx);
1725 error = SYSCTL_OUT(req, &len, sizeof(len));
1729 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1730 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1732 static struct proc *updateproc;
1733 static void sched_sync(void);
1734 static struct kproc_desc up_kp = {
1739 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1742 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1747 *bo = LIST_FIRST(slp);
1750 vp = (*bo)->__bo_vnode; /* XXX */
1751 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1754 * We use vhold in case the vnode does not
1755 * successfully sync. vhold prevents the vnode from
1756 * going away when we unlock the sync_mtx so that
1757 * we can acquire the vnode interlock.
1760 mtx_unlock(&sync_mtx);
1762 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1764 mtx_lock(&sync_mtx);
1765 return (*bo == LIST_FIRST(slp));
1767 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1768 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1770 vn_finished_write(mp);
1772 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1774 * Put us back on the worklist. The worklist
1775 * routine will remove us from our current
1776 * position and then add us back in at a later
1779 vn_syncer_add_to_worklist(*bo, syncdelay);
1783 mtx_lock(&sync_mtx);
1788 * System filesystem synchronizer daemon.
1793 struct synclist *gnext, *next;
1794 struct synclist *gslp, *slp;
1797 struct thread *td = curthread;
1799 int net_worklist_len;
1800 int syncer_final_iter;
1805 syncer_final_iter = 0;
1807 syncer_state = SYNCER_RUNNING;
1808 starttime = time_uptime;
1809 td->td_pflags |= TDP_NORUNNINGBUF;
1811 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1814 mtx_lock(&sync_mtx);
1816 if (syncer_state == SYNCER_FINAL_DELAY &&
1817 syncer_final_iter == 0) {
1818 mtx_unlock(&sync_mtx);
1819 kproc_suspend_check(td->td_proc);
1820 mtx_lock(&sync_mtx);
1822 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1823 if (syncer_state != SYNCER_RUNNING &&
1824 starttime != time_uptime) {
1826 printf("\nSyncing disks, vnodes remaining...");
1829 printf("%d ", net_worklist_len);
1831 starttime = time_uptime;
1834 * Push files whose dirty time has expired. Be careful
1835 * of interrupt race on slp queue.
1837 * Skip over empty worklist slots when shutting down.
1840 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1841 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1842 syncer_delayno += 1;
1843 if (syncer_delayno == syncer_maxdelay)
1845 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1846 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1848 * If the worklist has wrapped since the
1849 * it was emptied of all but syncer vnodes,
1850 * switch to the FINAL_DELAY state and run
1851 * for one more second.
1853 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1854 net_worklist_len == 0 &&
1855 last_work_seen == syncer_delayno) {
1856 syncer_state = SYNCER_FINAL_DELAY;
1857 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1859 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1860 LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1863 * Keep track of the last time there was anything
1864 * on the worklist other than syncer vnodes.
1865 * Return to the SHUTTING_DOWN state if any
1868 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1869 last_work_seen = syncer_delayno;
1870 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1871 syncer_state = SYNCER_SHUTTING_DOWN;
1872 while (!LIST_EMPTY(slp)) {
1873 error = sync_vnode(slp, &bo, td);
1875 LIST_REMOVE(bo, bo_synclist);
1876 LIST_INSERT_HEAD(next, bo, bo_synclist);
1880 if (first_printf == 0)
1881 wdog_kern_pat(WD_LASTVAL);
1884 if (!LIST_EMPTY(gslp)) {
1885 mtx_unlock(&sync_mtx);
1887 mtx_lock(&sync_mtx);
1888 while (!LIST_EMPTY(gslp)) {
1889 error = sync_vnode(gslp, &bo, td);
1891 LIST_REMOVE(bo, bo_synclist);
1892 LIST_INSERT_HEAD(gnext, bo,
1899 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1900 syncer_final_iter--;
1902 * The variable rushjob allows the kernel to speed up the
1903 * processing of the filesystem syncer process. A rushjob
1904 * value of N tells the filesystem syncer to process the next
1905 * N seconds worth of work on its queue ASAP. Currently rushjob
1906 * is used by the soft update code to speed up the filesystem
1907 * syncer process when the incore state is getting so far
1908 * ahead of the disk that the kernel memory pool is being
1909 * threatened with exhaustion.
1916 * Just sleep for a short period of time between
1917 * iterations when shutting down to allow some I/O
1920 * If it has taken us less than a second to process the
1921 * current work, then wait. Otherwise start right over
1922 * again. We can still lose time if any single round
1923 * takes more than two seconds, but it does not really
1924 * matter as we are just trying to generally pace the
1925 * filesystem activity.
1927 if (syncer_state != SYNCER_RUNNING ||
1928 time_uptime == starttime) {
1930 sched_prio(td, PPAUSE);
1933 if (syncer_state != SYNCER_RUNNING)
1934 cv_timedwait(&sync_wakeup, &sync_mtx,
1935 hz / SYNCER_SHUTDOWN_SPEEDUP);
1936 else if (time_uptime == starttime)
1937 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1942 * Request the syncer daemon to speed up its work.
1943 * We never push it to speed up more than half of its
1944 * normal turn time, otherwise it could take over the cpu.
1947 speedup_syncer(void)
1951 mtx_lock(&sync_mtx);
1952 if (rushjob < syncdelay / 2) {
1954 stat_rush_requests += 1;
1957 mtx_unlock(&sync_mtx);
1958 cv_broadcast(&sync_wakeup);
1963 * Tell the syncer to speed up its work and run though its work
1964 * list several times, then tell it to shut down.
1967 syncer_shutdown(void *arg, int howto)
1970 if (howto & RB_NOSYNC)
1972 mtx_lock(&sync_mtx);
1973 syncer_state = SYNCER_SHUTTING_DOWN;
1975 mtx_unlock(&sync_mtx);
1976 cv_broadcast(&sync_wakeup);
1977 kproc_shutdown(arg, howto);
1981 * Reassign a buffer from one vnode to another.
1982 * Used to assign file specific control information
1983 * (indirect blocks) to the vnode to which they belong.
1986 reassignbuf(struct buf *bp)
1999 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
2000 bp, bp->b_vp, bp->b_flags);
2002 * B_PAGING flagged buffers cannot be reassigned because their vp
2003 * is not fully linked in.
2005 if (bp->b_flags & B_PAGING)
2006 panic("cannot reassign paging buffer");
2009 * Delete from old vnode list, if on one.
2012 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2013 buf_vlist_remove(bp);
2015 panic("reassignbuf: Buffer %p not on queue.", bp);
2017 * If dirty, put on list of dirty buffers; otherwise insert onto list
2020 if (bp->b_flags & B_DELWRI) {
2021 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2022 switch (vp->v_type) {
2032 vn_syncer_add_to_worklist(bo, delay);
2034 buf_vlist_add(bp, bo, BX_VNDIRTY);
2036 buf_vlist_add(bp, bo, BX_VNCLEAN);
2038 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2039 mtx_lock(&sync_mtx);
2040 LIST_REMOVE(bo, bo_synclist);
2041 syncer_worklist_len--;
2042 mtx_unlock(&sync_mtx);
2043 bo->bo_flag &= ~BO_ONWORKLST;
2048 bp = TAILQ_FIRST(&bv->bv_hd);
2049 KASSERT(bp == NULL || bp->b_bufobj == bo,
2050 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2051 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2052 KASSERT(bp == NULL || bp->b_bufobj == bo,
2053 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2055 bp = TAILQ_FIRST(&bv->bv_hd);
2056 KASSERT(bp == NULL || bp->b_bufobj == bo,
2057 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2058 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2059 KASSERT(bp == NULL || bp->b_bufobj == bo,
2060 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2066 * Increment the use and hold counts on the vnode, taking care to reference
2067 * the driver's usecount if this is a chardev. The vholdl() will remove
2068 * the vnode from the free list if it is presently free. Requires the
2069 * vnode interlock and returns with it held.
2072 v_incr_usecount(struct vnode *vp)
2075 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2077 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2079 vp->v_rdev->si_usecount++;
2086 * Turn a holdcnt into a use+holdcnt such that only one call to
2087 * v_decr_usecount is needed.
2090 v_upgrade_usecount(struct vnode *vp)
2093 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2095 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2097 vp->v_rdev->si_usecount++;
2103 * Decrement the vnode use and hold count along with the driver's usecount
2104 * if this is a chardev. The vdropl() below releases the vnode interlock
2105 * as it may free the vnode.
2108 v_decr_usecount(struct vnode *vp)
2111 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2112 VNASSERT(vp->v_usecount > 0, vp,
2113 ("v_decr_usecount: negative usecount"));
2114 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2116 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2118 vp->v_rdev->si_usecount--;
2125 * Decrement only the use count and driver use count. This is intended to
2126 * be paired with a follow on vdropl() to release the remaining hold count.
2127 * In this way we may vgone() a vnode with a 0 usecount without risk of
2128 * having it end up on a free list because the hold count is kept above 0.
2131 v_decr_useonly(struct vnode *vp)
2134 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2135 VNASSERT(vp->v_usecount > 0, vp,
2136 ("v_decr_useonly: negative usecount"));
2137 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2139 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2141 vp->v_rdev->si_usecount--;
2147 * Grab a particular vnode from the free list, increment its
2148 * reference count and lock it. VI_DOOMED is set if the vnode
2149 * is being destroyed. Only callers who specify LK_RETRY will
2150 * see doomed vnodes. If inactive processing was delayed in
2151 * vput try to do it here.
2154 vget(struct vnode *vp, int flags, struct thread *td)
2159 VFS_ASSERT_GIANT(vp->v_mount);
2160 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2161 ("vget: invalid lock operation"));
2162 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2164 if ((flags & LK_INTERLOCK) == 0)
2167 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2169 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2173 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2174 panic("vget: vn_lock failed to return ENOENT\n");
2176 /* Upgrade our holdcnt to a usecount. */
2177 v_upgrade_usecount(vp);
2179 * We don't guarantee that any particular close will
2180 * trigger inactive processing so just make a best effort
2181 * here at preventing a reference to a removed file. If
2182 * we don't succeed no harm is done.
2184 if (vp->v_iflag & VI_OWEINACT) {
2185 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2186 (flags & LK_NOWAIT) == 0)
2188 vp->v_iflag &= ~VI_OWEINACT;
2195 * Increase the reference count of a vnode.
2198 vref(struct vnode *vp)
2201 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2203 v_incr_usecount(vp);
2208 * Return reference count of a vnode.
2210 * The results of this call are only guaranteed when some mechanism other
2211 * than the VI lock is used to stop other processes from gaining references
2212 * to the vnode. This may be the case if the caller holds the only reference.
2213 * This is also useful when stale data is acceptable as race conditions may
2214 * be accounted for by some other means.
2217 vrefcnt(struct vnode *vp)
2222 usecnt = vp->v_usecount;
2228 #define VPUTX_VRELE 1
2229 #define VPUTX_VPUT 2
2230 #define VPUTX_VUNREF 3
2233 vputx(struct vnode *vp, int func)
2237 KASSERT(vp != NULL, ("vputx: null vp"));
2238 if (func == VPUTX_VUNREF)
2239 ASSERT_VOP_LOCKED(vp, "vunref");
2240 else if (func == VPUTX_VPUT)
2241 ASSERT_VOP_LOCKED(vp, "vput");
2243 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2244 VFS_ASSERT_GIANT(vp->v_mount);
2245 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2248 /* Skip this v_writecount check if we're going to panic below. */
2249 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2250 ("vputx: missed vn_close"));
2253 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2254 vp->v_usecount == 1)) {
2255 if (func == VPUTX_VPUT)
2257 v_decr_usecount(vp);
2261 if (vp->v_usecount != 1) {
2262 vprint("vputx: negative ref count", vp);
2263 panic("vputx: negative ref cnt");
2265 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2267 * We want to hold the vnode until the inactive finishes to
2268 * prevent vgone() races. We drop the use count here and the
2269 * hold count below when we're done.
2273 * We must call VOP_INACTIVE with the node locked. Mark
2274 * as VI_DOINGINACT to avoid recursion.
2276 vp->v_iflag |= VI_OWEINACT;
2279 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2283 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2284 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2290 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2294 if (vp->v_usecount > 0)
2295 vp->v_iflag &= ~VI_OWEINACT;
2297 if (vp->v_iflag & VI_OWEINACT)
2298 vinactive(vp, curthread);
2299 if (func != VPUTX_VUNREF)
2306 * Vnode put/release.
2307 * If count drops to zero, call inactive routine and return to freelist.
2310 vrele(struct vnode *vp)
2313 vputx(vp, VPUTX_VRELE);
2317 * Release an already locked vnode. This give the same effects as
2318 * unlock+vrele(), but takes less time and avoids releasing and
2319 * re-aquiring the lock (as vrele() acquires the lock internally.)
2322 vput(struct vnode *vp)
2325 vputx(vp, VPUTX_VPUT);
2329 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2332 vunref(struct vnode *vp)
2335 vputx(vp, VPUTX_VUNREF);
2339 * Somebody doesn't want the vnode recycled.
2342 vhold(struct vnode *vp)
2351 vholdl(struct vnode *vp)
2354 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2356 if (VSHOULDBUSY(vp))
2361 * Note that there is one less who cares about this vnode. vdrop() is the
2362 * opposite of vhold().
2365 vdrop(struct vnode *vp)
2373 * Drop the hold count of the vnode. If this is the last reference to
2374 * the vnode we will free it if it has been vgone'd otherwise it is
2375 * placed on the free list.
2378 vdropl(struct vnode *vp)
2381 ASSERT_VI_LOCKED(vp, "vdropl");
2382 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2383 if (vp->v_holdcnt <= 0)
2384 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2386 if (vp->v_holdcnt == 0) {
2387 if (vp->v_iflag & VI_DOOMED) {
2388 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__,
2399 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2400 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2401 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2402 * failed lock upgrade.
2405 vinactive(struct vnode *vp, struct thread *td)
2408 ASSERT_VOP_ELOCKED(vp, "vinactive");
2409 ASSERT_VI_LOCKED(vp, "vinactive");
2410 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2411 ("vinactive: recursed on VI_DOINGINACT"));
2412 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2413 vp->v_iflag |= VI_DOINGINACT;
2414 vp->v_iflag &= ~VI_OWEINACT;
2416 VOP_INACTIVE(vp, td);
2418 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2419 ("vinactive: lost VI_DOINGINACT"));
2420 vp->v_iflag &= ~VI_DOINGINACT;
2424 * Remove any vnodes in the vnode table belonging to mount point mp.
2426 * If FORCECLOSE is not specified, there should not be any active ones,
2427 * return error if any are found (nb: this is a user error, not a
2428 * system error). If FORCECLOSE is specified, detach any active vnodes
2431 * If WRITECLOSE is set, only flush out regular file vnodes open for
2434 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2436 * `rootrefs' specifies the base reference count for the root vnode
2437 * of this filesystem. The root vnode is considered busy if its
2438 * v_usecount exceeds this value. On a successful return, vflush(, td)
2439 * will call vrele() on the root vnode exactly rootrefs times.
2440 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2444 static int busyprt = 0; /* print out busy vnodes */
2445 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2449 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2451 struct vnode *vp, *mvp, *rootvp = NULL;
2453 int busy = 0, error;
2455 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2458 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2459 ("vflush: bad args"));
2461 * Get the filesystem root vnode. We can vput() it
2462 * immediately, since with rootrefs > 0, it won't go away.
2464 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2465 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2473 MNT_VNODE_FOREACH(vp, mp, mvp) {
2477 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2481 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2485 * Skip over a vnodes marked VV_SYSTEM.
2487 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2494 * If WRITECLOSE is set, flush out unlinked but still open
2495 * files (even if open only for reading) and regular file
2496 * vnodes open for writing.
2498 if (flags & WRITECLOSE) {
2499 if (vp->v_object != NULL) {
2500 VM_OBJECT_LOCK(vp->v_object);
2501 vm_object_page_clean(vp->v_object, 0, 0, 0);
2502 VM_OBJECT_UNLOCK(vp->v_object);
2504 error = VOP_FSYNC(vp, MNT_WAIT, td);
2508 MNT_VNODE_FOREACH_ABORT(mp, mvp);
2511 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2514 if ((vp->v_type == VNON ||
2515 (error == 0 && vattr.va_nlink > 0)) &&
2516 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2525 * With v_usecount == 0, all we need to do is clear out the
2526 * vnode data structures and we are done.
2528 * If FORCECLOSE is set, forcibly close the vnode.
2530 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2531 VNASSERT(vp->v_usecount == 0 ||
2532 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2533 ("device VNODE %p is FORCECLOSED", vp));
2539 vprint("vflush: busy vnode", vp);
2547 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2549 * If just the root vnode is busy, and if its refcount
2550 * is equal to `rootrefs', then go ahead and kill it.
2553 KASSERT(busy > 0, ("vflush: not busy"));
2554 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2555 ("vflush: usecount %d < rootrefs %d",
2556 rootvp->v_usecount, rootrefs));
2557 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2558 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2560 VOP_UNLOCK(rootvp, 0);
2566 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2570 for (; rootrefs > 0; rootrefs--)
2576 * Recycle an unused vnode to the front of the free list.
2579 vrecycle(struct vnode *vp, struct thread *td)
2583 ASSERT_VOP_ELOCKED(vp, "vrecycle");
2584 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2587 if (vp->v_usecount == 0) {
2596 * Eliminate all activity associated with a vnode
2597 * in preparation for reuse.
2600 vgone(struct vnode *vp)
2608 * vgone, with the vp interlock held.
2611 vgonel(struct vnode *vp)
2618 ASSERT_VOP_ELOCKED(vp, "vgonel");
2619 ASSERT_VI_LOCKED(vp, "vgonel");
2620 VNASSERT(vp->v_holdcnt, vp,
2621 ("vgonel: vp %p has no reference.", vp));
2622 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2626 * Don't vgonel if we're already doomed.
2628 if (vp->v_iflag & VI_DOOMED)
2630 vp->v_iflag |= VI_DOOMED;
2632 * Check to see if the vnode is in use. If so, we have to call
2633 * VOP_CLOSE() and VOP_INACTIVE().
2635 active = vp->v_usecount;
2636 oweinact = (vp->v_iflag & VI_OWEINACT);
2639 * Clean out any buffers associated with the vnode.
2640 * If the flush fails, just toss the buffers.
2643 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2644 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2645 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2646 vinvalbuf(vp, 0, 0, 0);
2649 * If purging an active vnode, it must be closed and
2650 * deactivated before being reclaimed.
2653 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2654 if (oweinact || active) {
2656 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2661 * Reclaim the vnode.
2663 if (VOP_RECLAIM(vp, td))
2664 panic("vgone: cannot reclaim");
2666 vn_finished_secondary_write(mp);
2667 VNASSERT(vp->v_object == NULL, vp,
2668 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2670 * Clear the advisory locks and wake up waiting threads.
2672 (void)VOP_ADVLOCKPURGE(vp);
2674 * Delete from old mount point vnode list.
2679 * Done with purge, reset to the standard lock and invalidate
2683 vp->v_vnlock = &vp->v_lock;
2684 vp->v_op = &dead_vnodeops;
2690 * Calculate the total number of references to a special device.
2693 vcount(struct vnode *vp)
2698 count = vp->v_rdev->si_usecount;
2704 * Same as above, but using the struct cdev *as argument
2707 count_dev(struct cdev *dev)
2712 count = dev->si_usecount;
2718 * Print out a description of a vnode.
2720 static char *typename[] =
2721 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2725 vn_printf(struct vnode *vp, const char *fmt, ...)
2728 char buf[256], buf2[16];
2734 printf("%p: ", (void *)vp);
2735 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2736 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2737 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2740 if (vp->v_vflag & VV_ROOT)
2741 strlcat(buf, "|VV_ROOT", sizeof(buf));
2742 if (vp->v_vflag & VV_ISTTY)
2743 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2744 if (vp->v_vflag & VV_NOSYNC)
2745 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2746 if (vp->v_vflag & VV_CACHEDLABEL)
2747 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2748 if (vp->v_vflag & VV_TEXT)
2749 strlcat(buf, "|VV_TEXT", sizeof(buf));
2750 if (vp->v_vflag & VV_COPYONWRITE)
2751 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2752 if (vp->v_vflag & VV_SYSTEM)
2753 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2754 if (vp->v_vflag & VV_PROCDEP)
2755 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2756 if (vp->v_vflag & VV_NOKNOTE)
2757 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2758 if (vp->v_vflag & VV_DELETED)
2759 strlcat(buf, "|VV_DELETED", sizeof(buf));
2760 if (vp->v_vflag & VV_MD)
2761 strlcat(buf, "|VV_MD", sizeof(buf));
2762 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2763 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2764 VV_NOKNOTE | VV_DELETED | VV_MD);
2766 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2767 strlcat(buf, buf2, sizeof(buf));
2769 if (vp->v_iflag & VI_MOUNT)
2770 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2771 if (vp->v_iflag & VI_AGE)
2772 strlcat(buf, "|VI_AGE", sizeof(buf));
2773 if (vp->v_iflag & VI_DOOMED)
2774 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2775 if (vp->v_iflag & VI_FREE)
2776 strlcat(buf, "|VI_FREE", sizeof(buf));
2777 if (vp->v_iflag & VI_DOINGINACT)
2778 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2779 if (vp->v_iflag & VI_OWEINACT)
2780 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2781 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2782 VI_DOINGINACT | VI_OWEINACT);
2784 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2785 strlcat(buf, buf2, sizeof(buf));
2787 printf(" flags (%s)\n", buf + 1);
2788 if (mtx_owned(VI_MTX(vp)))
2789 printf(" VI_LOCKed");
2790 if (vp->v_object != NULL)
2791 printf(" v_object %p ref %d pages %d\n",
2792 vp->v_object, vp->v_object->ref_count,
2793 vp->v_object->resident_page_count);
2795 lockmgr_printinfo(vp->v_vnlock);
2796 if (vp->v_data != NULL)
2802 * List all of the locked vnodes in the system.
2803 * Called when debugging the kernel.
2805 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2807 struct mount *mp, *nmp;
2811 * Note: because this is DDB, we can't obey the locking semantics
2812 * for these structures, which means we could catch an inconsistent
2813 * state and dereference a nasty pointer. Not much to be done
2816 db_printf("Locked vnodes\n");
2817 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2818 nmp = TAILQ_NEXT(mp, mnt_list);
2819 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2820 if (vp->v_type != VMARKER &&
2824 nmp = TAILQ_NEXT(mp, mnt_list);
2829 * Show details about the given vnode.
2831 DB_SHOW_COMMAND(vnode, db_show_vnode)
2837 vp = (struct vnode *)addr;
2838 vn_printf(vp, "vnode ");
2842 * Show details about the given mount point.
2844 DB_SHOW_COMMAND(mount, db_show_mount)
2855 /* No address given, print short info about all mount points. */
2856 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2857 db_printf("%p %s on %s (%s)\n", mp,
2858 mp->mnt_stat.f_mntfromname,
2859 mp->mnt_stat.f_mntonname,
2860 mp->mnt_stat.f_fstypename);
2864 db_printf("\nMore info: show mount <addr>\n");
2868 mp = (struct mount *)addr;
2869 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2870 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2873 mflags = mp->mnt_flag;
2874 #define MNT_FLAG(flag) do { \
2875 if (mflags & (flag)) { \
2876 if (buf[0] != '\0') \
2877 strlcat(buf, ", ", sizeof(buf)); \
2878 strlcat(buf, (#flag) + 4, sizeof(buf)); \
2879 mflags &= ~(flag); \
2882 MNT_FLAG(MNT_RDONLY);
2883 MNT_FLAG(MNT_SYNCHRONOUS);
2884 MNT_FLAG(MNT_NOEXEC);
2885 MNT_FLAG(MNT_NOSUID);
2886 MNT_FLAG(MNT_UNION);
2887 MNT_FLAG(MNT_ASYNC);
2888 MNT_FLAG(MNT_SUIDDIR);
2889 MNT_FLAG(MNT_SOFTDEP);
2891 MNT_FLAG(MNT_NOSYMFOLLOW);
2892 MNT_FLAG(MNT_GJOURNAL);
2893 MNT_FLAG(MNT_MULTILABEL);
2895 MNT_FLAG(MNT_NOATIME);
2896 MNT_FLAG(MNT_NOCLUSTERR);
2897 MNT_FLAG(MNT_NOCLUSTERW);
2898 MNT_FLAG(MNT_NFS4ACLS);
2899 MNT_FLAG(MNT_EXRDONLY);
2900 MNT_FLAG(MNT_EXPORTED);
2901 MNT_FLAG(MNT_DEFEXPORTED);
2902 MNT_FLAG(MNT_EXPORTANON);
2903 MNT_FLAG(MNT_EXKERB);
2904 MNT_FLAG(MNT_EXPUBLIC);
2905 MNT_FLAG(MNT_LOCAL);
2906 MNT_FLAG(MNT_QUOTA);
2907 MNT_FLAG(MNT_ROOTFS);
2909 MNT_FLAG(MNT_IGNORE);
2910 MNT_FLAG(MNT_UPDATE);
2911 MNT_FLAG(MNT_DELEXPORT);
2912 MNT_FLAG(MNT_RELOAD);
2913 MNT_FLAG(MNT_FORCE);
2914 MNT_FLAG(MNT_SNAPSHOT);
2915 MNT_FLAG(MNT_BYFSID);
2919 strlcat(buf, ", ", sizeof(buf));
2920 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2921 "0x%016jx", mflags);
2923 db_printf(" mnt_flag = %s\n", buf);
2926 flags = mp->mnt_kern_flag;
2927 #define MNT_KERN_FLAG(flag) do { \
2928 if (flags & (flag)) { \
2929 if (buf[0] != '\0') \
2930 strlcat(buf, ", ", sizeof(buf)); \
2931 strlcat(buf, (#flag) + 5, sizeof(buf)); \
2935 MNT_KERN_FLAG(MNTK_UNMOUNTF);
2936 MNT_KERN_FLAG(MNTK_ASYNC);
2937 MNT_KERN_FLAG(MNTK_SOFTDEP);
2938 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2939 MNT_KERN_FLAG(MNTK_DRAINING);
2940 MNT_KERN_FLAG(MNTK_REFEXPIRE);
2941 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
2942 MNT_KERN_FLAG(MNTK_SHARED_WRITES);
2943 MNT_KERN_FLAG(MNTK_NOASYNC);
2944 MNT_KERN_FLAG(MNTK_UNMOUNT);
2945 MNT_KERN_FLAG(MNTK_MWAIT);
2946 MNT_KERN_FLAG(MNTK_SUSPEND);
2947 MNT_KERN_FLAG(MNTK_SUSPEND2);
2948 MNT_KERN_FLAG(MNTK_SUSPENDED);
2949 MNT_KERN_FLAG(MNTK_MPSAFE);
2950 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2951 MNT_KERN_FLAG(MNTK_NOKNOTE);
2952 #undef MNT_KERN_FLAG
2955 strlcat(buf, ", ", sizeof(buf));
2956 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2959 db_printf(" mnt_kern_flag = %s\n", buf);
2961 db_printf(" mnt_opt = ");
2962 opt = TAILQ_FIRST(mp->mnt_opt);
2964 db_printf("%s", opt->name);
2965 opt = TAILQ_NEXT(opt, link);
2966 while (opt != NULL) {
2967 db_printf(", %s", opt->name);
2968 opt = TAILQ_NEXT(opt, link);
2974 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
2975 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2976 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2977 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2978 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2979 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2980 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2981 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2982 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2983 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2984 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2985 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2987 db_printf(" mnt_cred = { uid=%u ruid=%u",
2988 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2989 if (jailed(mp->mnt_cred))
2990 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2992 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
2993 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
2994 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2995 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2996 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2997 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2998 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
2999 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3000 db_printf(" mnt_secondary_accwrites = %d\n",
3001 mp->mnt_secondary_accwrites);
3002 db_printf(" mnt_gjprovider = %s\n",
3003 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3006 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3007 if (vp->v_type != VMARKER) {
3008 vn_printf(vp, "vnode ");
3017 * Fill in a struct xvfsconf based on a struct vfsconf.
3020 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
3023 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
3024 xvfsp->vfc_typenum = vfsp->vfc_typenum;
3025 xvfsp->vfc_refcount = vfsp->vfc_refcount;
3026 xvfsp->vfc_flags = vfsp->vfc_flags;
3028 * These are unused in userland, we keep them
3029 * to not break binary compatibility.
3031 xvfsp->vfc_vfsops = NULL;
3032 xvfsp->vfc_next = NULL;
3036 * Top level filesystem related information gathering.
3039 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3041 struct vfsconf *vfsp;
3042 struct xvfsconf xvfsp;
3046 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3047 bzero(&xvfsp, sizeof(xvfsp));
3048 vfsconf2x(vfsp, &xvfsp);
3049 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
3056 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3057 NULL, 0, sysctl_vfs_conflist,
3058 "S,xvfsconf", "List of all configured filesystems");
3060 #ifndef BURN_BRIDGES
3061 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3064 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3066 int *name = (int *)arg1 - 1; /* XXX */
3067 u_int namelen = arg2 + 1; /* XXX */
3068 struct vfsconf *vfsp;
3069 struct xvfsconf xvfsp;
3071 printf("WARNING: userland calling deprecated sysctl, "
3072 "please rebuild world\n");
3074 #if 1 || defined(COMPAT_PRELITE2)
3075 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3077 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3081 case VFS_MAXTYPENUM:
3084 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3087 return (ENOTDIR); /* overloaded */
3088 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3089 if (vfsp->vfc_typenum == name[2])
3092 return (EOPNOTSUPP);
3093 bzero(&xvfsp, sizeof(xvfsp));
3094 vfsconf2x(vfsp, &xvfsp);
3095 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3097 return (EOPNOTSUPP);
3100 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3101 vfs_sysctl, "Generic filesystem");
3103 #if 1 || defined(COMPAT_PRELITE2)
3106 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3109 struct vfsconf *vfsp;
3110 struct ovfsconf ovfs;
3112 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3113 bzero(&ovfs, sizeof(ovfs));
3114 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3115 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3116 ovfs.vfc_index = vfsp->vfc_typenum;
3117 ovfs.vfc_refcount = vfsp->vfc_refcount;
3118 ovfs.vfc_flags = vfsp->vfc_flags;
3119 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3126 #endif /* 1 || COMPAT_PRELITE2 */
3127 #endif /* !BURN_BRIDGES */
3129 #define KINFO_VNODESLOP 10
3132 * Dump vnode list (via sysctl).
3136 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3144 * Stale numvnodes access is not fatal here.
3147 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3149 /* Make an estimate */
3150 return (SYSCTL_OUT(req, 0, len));
3152 error = sysctl_wire_old_buffer(req, 0);
3155 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3157 mtx_lock(&mountlist_mtx);
3158 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3159 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3162 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3166 xvn[n].xv_size = sizeof *xvn;
3167 xvn[n].xv_vnode = vp;
3168 xvn[n].xv_id = 0; /* XXX compat */
3169 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3171 XV_COPY(writecount);
3177 xvn[n].xv_flag = vp->v_vflag;
3179 switch (vp->v_type) {
3186 if (vp->v_rdev == NULL) {
3190 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3193 xvn[n].xv_socket = vp->v_socket;
3196 xvn[n].xv_fifo = vp->v_fifoinfo;
3201 /* shouldn't happen? */
3209 mtx_lock(&mountlist_mtx);
3214 mtx_unlock(&mountlist_mtx);
3216 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3221 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3222 0, 0, sysctl_vnode, "S,xvnode", "");
3226 * Unmount all filesystems. The list is traversed in reverse order
3227 * of mounting to avoid dependencies.
3230 vfs_unmountall(void)
3236 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3237 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3241 * Since this only runs when rebooting, it is not interlocked.
3243 while(!TAILQ_EMPTY(&mountlist)) {
3244 mp = TAILQ_LAST(&mountlist, mntlist);
3245 error = dounmount(mp, MNT_FORCE, td);
3247 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3249 * XXX: Due to the way in which we mount the root
3250 * file system off of devfs, devfs will generate a
3251 * "busy" warning when we try to unmount it before
3252 * the root. Don't print a warning as a result in
3253 * order to avoid false positive errors that may
3254 * cause needless upset.
3256 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3257 printf("unmount of %s failed (",
3258 mp->mnt_stat.f_mntonname);
3262 printf("%d)\n", error);
3265 /* The unmount has removed mp from the mountlist */
3271 * perform msync on all vnodes under a mount point
3272 * the mount point must be locked.
3275 vfs_msync(struct mount *mp, int flags)
3277 struct vnode *vp, *mvp;
3278 struct vm_object *obj;
3280 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3282 MNT_VNODE_FOREACH(vp, mp, mvp) {
3285 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3286 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3289 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3291 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3299 VM_OBJECT_LOCK(obj);
3300 vm_object_page_clean(obj, 0, 0,
3302 OBJPC_SYNC : OBJPC_NOSYNC);
3303 VM_OBJECT_UNLOCK(obj);
3315 * Mark a vnode as free, putting it up for recycling.
3318 vfree(struct vnode *vp)
3321 ASSERT_VI_LOCKED(vp, "vfree");
3322 mtx_lock(&vnode_free_list_mtx);
3323 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3324 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3325 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3326 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3327 ("vfree: Freeing doomed vnode"));
3328 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3329 if (vp->v_iflag & VI_AGE) {
3330 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3332 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3335 vp->v_iflag &= ~VI_AGE;
3336 vp->v_iflag |= VI_FREE;
3337 mtx_unlock(&vnode_free_list_mtx);
3341 * Opposite of vfree() - mark a vnode as in use.
3344 vbusy(struct vnode *vp)
3346 ASSERT_VI_LOCKED(vp, "vbusy");
3347 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3348 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3349 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3351 mtx_lock(&vnode_free_list_mtx);
3352 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3354 vp->v_iflag &= ~(VI_FREE|VI_AGE);
3355 mtx_unlock(&vnode_free_list_mtx);
3359 destroy_vpollinfo(struct vpollinfo *vi)
3361 seldrain(&vi->vpi_selinfo);
3362 knlist_destroy(&vi->vpi_selinfo.si_note);
3363 mtx_destroy(&vi->vpi_lock);
3364 uma_zfree(vnodepoll_zone, vi);
3368 * Initalize per-vnode helper structure to hold poll-related state.
3371 v_addpollinfo(struct vnode *vp)
3373 struct vpollinfo *vi;
3375 if (vp->v_pollinfo != NULL)
3377 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3378 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3379 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3380 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3382 if (vp->v_pollinfo != NULL) {
3384 destroy_vpollinfo(vi);
3387 vp->v_pollinfo = vi;
3392 * Record a process's interest in events which might happen to
3393 * a vnode. Because poll uses the historic select-style interface
3394 * internally, this routine serves as both the ``check for any
3395 * pending events'' and the ``record my interest in future events''
3396 * functions. (These are done together, while the lock is held,
3397 * to avoid race conditions.)
3400 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3404 mtx_lock(&vp->v_pollinfo->vpi_lock);
3405 if (vp->v_pollinfo->vpi_revents & events) {
3407 * This leaves events we are not interested
3408 * in available for the other process which
3409 * which presumably had requested them
3410 * (otherwise they would never have been
3413 events &= vp->v_pollinfo->vpi_revents;
3414 vp->v_pollinfo->vpi_revents &= ~events;
3416 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3419 vp->v_pollinfo->vpi_events |= events;
3420 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3421 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3426 * Routine to create and manage a filesystem syncer vnode.
3428 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3429 static int sync_fsync(struct vop_fsync_args *);
3430 static int sync_inactive(struct vop_inactive_args *);
3431 static int sync_reclaim(struct vop_reclaim_args *);
3433 static struct vop_vector sync_vnodeops = {
3434 .vop_bypass = VOP_EOPNOTSUPP,
3435 .vop_close = sync_close, /* close */
3436 .vop_fsync = sync_fsync, /* fsync */
3437 .vop_inactive = sync_inactive, /* inactive */
3438 .vop_reclaim = sync_reclaim, /* reclaim */
3439 .vop_lock1 = vop_stdlock, /* lock */
3440 .vop_unlock = vop_stdunlock, /* unlock */
3441 .vop_islocked = vop_stdislocked, /* islocked */
3445 * Create a new filesystem syncer vnode for the specified mount point.
3448 vfs_allocate_syncvnode(struct mount *mp)
3452 static long start, incr, next;
3455 /* Allocate a new vnode */
3456 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3458 panic("vfs_allocate_syncvnode: getnewvnode() failed");
3460 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3461 vp->v_vflag |= VV_FORCEINSMQ;
3462 error = insmntque(vp, mp);
3464 panic("vfs_allocate_syncvnode: insmntque() failed");
3465 vp->v_vflag &= ~VV_FORCEINSMQ;
3468 * Place the vnode onto the syncer worklist. We attempt to
3469 * scatter them about on the list so that they will go off
3470 * at evenly distributed times even if all the filesystems
3471 * are mounted at once.
3474 if (next == 0 || next > syncer_maxdelay) {
3478 start = syncer_maxdelay / 2;
3479 incr = syncer_maxdelay;
3485 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3486 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3487 mtx_lock(&sync_mtx);
3489 if (mp->mnt_syncer == NULL) {
3490 mp->mnt_syncer = vp;
3493 mtx_unlock(&sync_mtx);
3496 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3503 vfs_deallocate_syncvnode(struct mount *mp)
3507 mtx_lock(&sync_mtx);
3508 vp = mp->mnt_syncer;
3510 mp->mnt_syncer = NULL;
3511 mtx_unlock(&sync_mtx);
3517 * Do a lazy sync of the filesystem.
3520 sync_fsync(struct vop_fsync_args *ap)
3522 struct vnode *syncvp = ap->a_vp;
3523 struct mount *mp = syncvp->v_mount;
3528 * We only need to do something if this is a lazy evaluation.
3530 if (ap->a_waitfor != MNT_LAZY)
3534 * Move ourselves to the back of the sync list.
3536 bo = &syncvp->v_bufobj;
3538 vn_syncer_add_to_worklist(bo, syncdelay);
3542 * Walk the list of vnodes pushing all that are dirty and
3543 * not already on the sync list.
3545 mtx_lock(&mountlist_mtx);
3546 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3547 mtx_unlock(&mountlist_mtx);
3550 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3554 save = curthread_pflags_set(TDP_SYNCIO);
3555 vfs_msync(mp, MNT_NOWAIT);
3556 error = VFS_SYNC(mp, MNT_LAZY);
3557 curthread_pflags_restore(save);
3558 vn_finished_write(mp);
3564 * The syncer vnode is no referenced.
3567 sync_inactive(struct vop_inactive_args *ap)
3575 * The syncer vnode is no longer needed and is being decommissioned.
3577 * Modifications to the worklist must be protected by sync_mtx.
3580 sync_reclaim(struct vop_reclaim_args *ap)
3582 struct vnode *vp = ap->a_vp;
3587 mtx_lock(&sync_mtx);
3588 if (vp->v_mount->mnt_syncer == vp)
3589 vp->v_mount->mnt_syncer = NULL;
3590 if (bo->bo_flag & BO_ONWORKLST) {
3591 LIST_REMOVE(bo, bo_synclist);
3592 syncer_worklist_len--;
3594 bo->bo_flag &= ~BO_ONWORKLST;
3596 mtx_unlock(&sync_mtx);
3603 * Check if vnode represents a disk device
3606 vn_isdisk(struct vnode *vp, int *errp)
3612 if (vp->v_type != VCHR)
3614 else if (vp->v_rdev == NULL)
3616 else if (vp->v_rdev->si_devsw == NULL)
3618 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3623 return (error == 0);
3627 * Common filesystem object access control check routine. Accepts a
3628 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3629 * and optional call-by-reference privused argument allowing vaccess()
3630 * to indicate to the caller whether privilege was used to satisfy the
3631 * request (obsoleted). Returns 0 on success, or an errno on failure.
3634 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3635 accmode_t accmode, struct ucred *cred, int *privused)
3637 accmode_t dac_granted;
3638 accmode_t priv_granted;
3640 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3641 ("invalid bit in accmode"));
3642 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3643 ("VAPPEND without VWRITE"));
3646 * Look for a normal, non-privileged way to access the file/directory
3647 * as requested. If it exists, go with that.
3650 if (privused != NULL)
3655 /* Check the owner. */
3656 if (cred->cr_uid == file_uid) {
3657 dac_granted |= VADMIN;
3658 if (file_mode & S_IXUSR)
3659 dac_granted |= VEXEC;
3660 if (file_mode & S_IRUSR)
3661 dac_granted |= VREAD;
3662 if (file_mode & S_IWUSR)
3663 dac_granted |= (VWRITE | VAPPEND);
3665 if ((accmode & dac_granted) == accmode)
3671 /* Otherwise, check the groups (first match) */
3672 if (groupmember(file_gid, cred)) {
3673 if (file_mode & S_IXGRP)
3674 dac_granted |= VEXEC;
3675 if (file_mode & S_IRGRP)
3676 dac_granted |= VREAD;
3677 if (file_mode & S_IWGRP)
3678 dac_granted |= (VWRITE | VAPPEND);
3680 if ((accmode & dac_granted) == accmode)
3686 /* Otherwise, check everyone else. */
3687 if (file_mode & S_IXOTH)
3688 dac_granted |= VEXEC;
3689 if (file_mode & S_IROTH)
3690 dac_granted |= VREAD;
3691 if (file_mode & S_IWOTH)
3692 dac_granted |= (VWRITE | VAPPEND);
3693 if ((accmode & dac_granted) == accmode)
3698 * Build a privilege mask to determine if the set of privileges
3699 * satisfies the requirements when combined with the granted mask
3700 * from above. For each privilege, if the privilege is required,
3701 * bitwise or the request type onto the priv_granted mask.
3707 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3708 * requests, instead of PRIV_VFS_EXEC.
3710 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3711 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3712 priv_granted |= VEXEC;
3715 * Ensure that at least one execute bit is on. Otherwise,
3716 * a privileged user will always succeed, and we don't want
3717 * this to happen unless the file really is executable.
3719 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3720 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3721 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3722 priv_granted |= VEXEC;
3725 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3726 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3727 priv_granted |= VREAD;
3729 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3730 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3731 priv_granted |= (VWRITE | VAPPEND);
3733 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3734 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3735 priv_granted |= VADMIN;
3737 if ((accmode & (priv_granted | dac_granted)) == accmode) {
3738 /* XXX audit: privilege used */
3739 if (privused != NULL)
3744 return ((accmode & VADMIN) ? EPERM : EACCES);
3748 * Credential check based on process requesting service, and per-attribute
3752 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3753 struct thread *td, accmode_t accmode)
3757 * Kernel-invoked always succeeds.
3763 * Do not allow privileged processes in jail to directly manipulate
3764 * system attributes.
3766 switch (attrnamespace) {
3767 case EXTATTR_NAMESPACE_SYSTEM:
3768 /* Potentially should be: return (EPERM); */
3769 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3770 case EXTATTR_NAMESPACE_USER:
3771 return (VOP_ACCESS(vp, accmode, cred, td));
3777 #ifdef DEBUG_VFS_LOCKS
3779 * This only exists to supress warnings from unlocked specfs accesses. It is
3780 * no longer ok to have an unlocked VFS.
3782 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
3783 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
3785 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3786 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3787 "Drop into debugger on lock violation");
3789 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3790 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3791 0, "Check for interlock across VOPs");
3793 int vfs_badlock_print = 1; /* Print lock violations. */
3794 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3795 0, "Print lock violations");
3798 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3799 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3800 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3804 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3808 if (vfs_badlock_backtrace)
3811 if (vfs_badlock_print)
3812 printf("%s: %p %s\n", str, (void *)vp, msg);
3813 if (vfs_badlock_ddb)
3814 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3818 assert_vi_locked(struct vnode *vp, const char *str)
3821 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3822 vfs_badlock("interlock is not locked but should be", str, vp);
3826 assert_vi_unlocked(struct vnode *vp, const char *str)
3829 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3830 vfs_badlock("interlock is locked but should not be", str, vp);
3834 assert_vop_locked(struct vnode *vp, const char *str)
3837 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3838 vfs_badlock("is not locked but should be", str, vp);
3842 assert_vop_unlocked(struct vnode *vp, const char *str)
3845 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3846 vfs_badlock("is locked but should not be", str, vp);
3850 assert_vop_elocked(struct vnode *vp, const char *str)
3853 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3854 vfs_badlock("is not exclusive locked but should be", str, vp);
3859 assert_vop_elocked_other(struct vnode *vp, const char *str)
3862 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3863 vfs_badlock("is not exclusive locked by another thread",
3868 assert_vop_slocked(struct vnode *vp, const char *str)
3871 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3872 vfs_badlock("is not locked shared but should be", str, vp);
3875 #endif /* DEBUG_VFS_LOCKS */
3878 vop_rename_fail(struct vop_rename_args *ap)
3881 if (ap->a_tvp != NULL)
3883 if (ap->a_tdvp == ap->a_tvp)
3892 vop_rename_pre(void *ap)
3894 struct vop_rename_args *a = ap;
3896 #ifdef DEBUG_VFS_LOCKS
3898 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3899 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3900 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3901 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3903 /* Check the source (from). */
3904 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
3905 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
3906 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3907 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
3908 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3910 /* Check the target. */
3912 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3913 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3915 if (a->a_tdvp != a->a_fdvp)
3917 if (a->a_tvp != a->a_fvp)
3925 vop_strategy_pre(void *ap)
3927 #ifdef DEBUG_VFS_LOCKS
3928 struct vop_strategy_args *a;
3935 * Cluster ops lock their component buffers but not the IO container.
3937 if ((bp->b_flags & B_CLUSTER) != 0)
3940 if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
3941 if (vfs_badlock_print)
3943 "VOP_STRATEGY: bp is not locked but should be\n");
3944 if (vfs_badlock_ddb)
3945 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3951 vop_lookup_pre(void *ap)
3953 #ifdef DEBUG_VFS_LOCKS
3954 struct vop_lookup_args *a;
3959 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3960 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3965 vop_lookup_post(void *ap, int rc)
3967 #ifdef DEBUG_VFS_LOCKS
3968 struct vop_lookup_args *a;
3976 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3977 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3980 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3985 vop_lock_pre(void *ap)
3987 #ifdef DEBUG_VFS_LOCKS
3988 struct vop_lock1_args *a = ap;
3990 if ((a->a_flags & LK_INTERLOCK) == 0)
3991 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3993 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3998 vop_lock_post(void *ap, int rc)
4000 #ifdef DEBUG_VFS_LOCKS
4001 struct vop_lock1_args *a = ap;
4003 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4005 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4010 vop_unlock_pre(void *ap)
4012 #ifdef DEBUG_VFS_LOCKS
4013 struct vop_unlock_args *a = ap;
4015 if (a->a_flags & LK_INTERLOCK)
4016 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4017 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4022 vop_unlock_post(void *ap, int rc)
4024 #ifdef DEBUG_VFS_LOCKS
4025 struct vop_unlock_args *a = ap;
4027 if (a->a_flags & LK_INTERLOCK)
4028 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4033 vop_create_post(void *ap, int rc)
4035 struct vop_create_args *a = ap;
4038 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4042 vop_deleteextattr_post(void *ap, int rc)
4044 struct vop_deleteextattr_args *a = ap;
4047 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4051 vop_link_post(void *ap, int rc)
4053 struct vop_link_args *a = ap;
4056 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4057 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4062 vop_mkdir_post(void *ap, int rc)
4064 struct vop_mkdir_args *a = ap;
4067 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4071 vop_mknod_post(void *ap, int rc)
4073 struct vop_mknod_args *a = ap;
4076 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4080 vop_remove_post(void *ap, int rc)
4082 struct vop_remove_args *a = ap;
4085 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4086 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4091 vop_rename_post(void *ap, int rc)
4093 struct vop_rename_args *a = ap;
4096 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4097 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4098 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4100 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4102 if (a->a_tdvp != a->a_fdvp)
4104 if (a->a_tvp != a->a_fvp)
4112 vop_rmdir_post(void *ap, int rc)
4114 struct vop_rmdir_args *a = ap;
4117 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4118 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4123 vop_setattr_post(void *ap, int rc)
4125 struct vop_setattr_args *a = ap;
4128 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4132 vop_setextattr_post(void *ap, int rc)
4134 struct vop_setextattr_args *a = ap;
4137 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4141 vop_symlink_post(void *ap, int rc)
4143 struct vop_symlink_args *a = ap;
4146 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4149 static struct knlist fs_knlist;
4152 vfs_event_init(void *arg)
4154 knlist_init_mtx(&fs_knlist, NULL);
4156 /* XXX - correct order? */
4157 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4160 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4163 KNOTE_UNLOCKED(&fs_knlist, event);
4166 static int filt_fsattach(struct knote *kn);
4167 static void filt_fsdetach(struct knote *kn);
4168 static int filt_fsevent(struct knote *kn, long hint);
4170 struct filterops fs_filtops = {
4172 .f_attach = filt_fsattach,
4173 .f_detach = filt_fsdetach,
4174 .f_event = filt_fsevent
4178 filt_fsattach(struct knote *kn)
4181 kn->kn_flags |= EV_CLEAR;
4182 knlist_add(&fs_knlist, kn, 0);
4187 filt_fsdetach(struct knote *kn)
4190 knlist_remove(&fs_knlist, kn, 0);
4194 filt_fsevent(struct knote *kn, long hint)
4197 kn->kn_fflags |= hint;
4198 return (kn->kn_fflags != 0);
4202 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4208 error = SYSCTL_IN(req, &vc, sizeof(vc));
4211 if (vc.vc_vers != VFS_CTL_VERS1)
4213 mp = vfs_getvfs(&vc.vc_fsid);
4216 /* ensure that a specific sysctl goes to the right filesystem. */
4217 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4218 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4222 VCTLTOREQ(&vc, req);
4223 error = VFS_SYSCTL(mp, vc.vc_op, req);
4228 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4229 NULL, 0, sysctl_vfs_ctl, "",
4233 * Function to initialize a va_filerev field sensibly.
4234 * XXX: Wouldn't a random number make a lot more sense ??
4237 init_va_filerev(void)
4242 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4245 static int filt_vfsread(struct knote *kn, long hint);
4246 static int filt_vfswrite(struct knote *kn, long hint);
4247 static int filt_vfsvnode(struct knote *kn, long hint);
4248 static void filt_vfsdetach(struct knote *kn);
4249 static struct filterops vfsread_filtops = {
4251 .f_detach = filt_vfsdetach,
4252 .f_event = filt_vfsread
4254 static struct filterops vfswrite_filtops = {
4256 .f_detach = filt_vfsdetach,
4257 .f_event = filt_vfswrite
4259 static struct filterops vfsvnode_filtops = {
4261 .f_detach = filt_vfsdetach,
4262 .f_event = filt_vfsvnode
4266 vfs_knllock(void *arg)
4268 struct vnode *vp = arg;
4270 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4274 vfs_knlunlock(void *arg)
4276 struct vnode *vp = arg;
4282 vfs_knl_assert_locked(void *arg)
4284 #ifdef DEBUG_VFS_LOCKS
4285 struct vnode *vp = arg;
4287 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4292 vfs_knl_assert_unlocked(void *arg)
4294 #ifdef DEBUG_VFS_LOCKS
4295 struct vnode *vp = arg;
4297 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4302 vfs_kqfilter(struct vop_kqfilter_args *ap)
4304 struct vnode *vp = ap->a_vp;
4305 struct knote *kn = ap->a_kn;
4308 switch (kn->kn_filter) {
4310 kn->kn_fop = &vfsread_filtops;
4313 kn->kn_fop = &vfswrite_filtops;
4316 kn->kn_fop = &vfsvnode_filtops;
4322 kn->kn_hook = (caddr_t)vp;
4325 if (vp->v_pollinfo == NULL)
4327 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4328 knlist_add(knl, kn, 0);
4334 * Detach knote from vnode
4337 filt_vfsdetach(struct knote *kn)
4339 struct vnode *vp = (struct vnode *)kn->kn_hook;
4341 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4342 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4347 filt_vfsread(struct knote *kn, long hint)
4349 struct vnode *vp = (struct vnode *)kn->kn_hook;
4354 * filesystem is gone, so set the EOF flag and schedule
4355 * the knote for deletion.
4357 if (hint == NOTE_REVOKE) {
4359 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4364 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4368 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4369 res = (kn->kn_data != 0);
4376 filt_vfswrite(struct knote *kn, long hint)
4378 struct vnode *vp = (struct vnode *)kn->kn_hook;
4383 * filesystem is gone, so set the EOF flag and schedule
4384 * the knote for deletion.
4386 if (hint == NOTE_REVOKE)
4387 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4395 filt_vfsvnode(struct knote *kn, long hint)
4397 struct vnode *vp = (struct vnode *)kn->kn_hook;
4401 if (kn->kn_sfflags & hint)
4402 kn->kn_fflags |= hint;
4403 if (hint == NOTE_REVOKE) {
4404 kn->kn_flags |= EV_EOF;
4408 res = (kn->kn_fflags != 0);
4414 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4418 if (dp->d_reclen > ap->a_uio->uio_resid)
4419 return (ENAMETOOLONG);
4420 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4422 if (ap->a_ncookies != NULL) {
4423 if (ap->a_cookies != NULL)
4424 free(ap->a_cookies, M_TEMP);
4425 ap->a_cookies = NULL;
4426 *ap->a_ncookies = 0;
4430 if (ap->a_ncookies == NULL)
4433 KASSERT(ap->a_cookies,
4434 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4436 *ap->a_cookies = realloc(*ap->a_cookies,
4437 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4438 (*ap->a_cookies)[*ap->a_ncookies] = off;
4443 * Mark for update the access time of the file if the filesystem
4444 * supports VOP_MARKATIME. This functionality is used by execve and
4445 * mmap, so we want to avoid the I/O implied by directly setting
4446 * va_atime for the sake of efficiency.
4449 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4454 VFS_ASSERT_GIANT(mp);
4455 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4456 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4457 (void)VOP_MARKATIME(vp);
4461 * The purpose of this routine is to remove granularity from accmode_t,
4462 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4463 * VADMIN and VAPPEND.
4465 * If it returns 0, the caller is supposed to continue with the usual
4466 * access checks using 'accmode' as modified by this routine. If it
4467 * returns nonzero value, the caller is supposed to return that value
4470 * Note that after this routine runs, accmode may be zero.
4473 vfs_unixify_accmode(accmode_t *accmode)
4476 * There is no way to specify explicit "deny" rule using
4477 * file mode or POSIX.1e ACLs.
4479 if (*accmode & VEXPLICIT_DENY) {
4485 * None of these can be translated into usual access bits.
4486 * Also, the common case for NFSv4 ACLs is to not contain
4487 * either of these bits. Caller should check for VWRITE
4488 * on the containing directory instead.
4490 if (*accmode & (VDELETE_CHILD | VDELETE))
4493 if (*accmode & VADMIN_PERMS) {
4494 *accmode &= ~VADMIN_PERMS;
4499 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4500 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4502 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);