2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
38 * External virtual filesystem routines
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
46 #include <sys/param.h>
47 #include <sys/systm.h>
50 #include <sys/condvar.h>
52 #include <sys/dirent.h>
53 #include <sys/event.h>
54 #include <sys/eventhandler.h>
55 #include <sys/extattr.h>
57 #include <sys/fcntl.h>
60 #include <sys/kernel.h>
61 #include <sys/kthread.h>
62 #include <sys/lockf.h>
63 #include <sys/malloc.h>
64 #include <sys/mount.h>
65 #include <sys/namei.h>
67 #include <sys/reboot.h>
68 #include <sys/sleepqueue.h>
70 #include <sys/sysctl.h>
71 #include <sys/syslog.h>
72 #include <sys/vmmeter.h>
73 #include <sys/vnode.h>
75 #include <machine/stdarg.h>
77 #include <security/mac/mac_framework.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_extern.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_kern.h>
95 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure");
97 static void delmntque(struct vnode *vp);
98 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
99 int slpflag, int slptimeo);
100 static void syncer_shutdown(void *arg, int howto);
101 static int vtryrecycle(struct vnode *vp);
102 static void vbusy(struct vnode *vp);
103 static void vinactive(struct vnode *, struct thread *);
104 static void v_incr_usecount(struct vnode *);
105 static void v_decr_usecount(struct vnode *);
106 static void v_decr_useonly(struct vnode *);
107 static void v_upgrade_usecount(struct vnode *);
108 static void vfree(struct vnode *);
109 static void vnlru_free(int);
110 static void vgonel(struct vnode *);
111 static void vfs_knllock(void *arg);
112 static void vfs_knlunlock(void *arg);
113 static void vfs_knl_assert_locked(void *arg);
114 static void vfs_knl_assert_unlocked(void *arg);
115 static void destroy_vpollinfo(struct vpollinfo *vi);
118 * Number of vnodes in existence. Increased whenever getnewvnode()
119 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
122 static unsigned long numvnodes;
124 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
127 * Conversion tables for conversion from vnode types to inode formats
130 enum vtype iftovt_tab[16] = {
131 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
132 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
134 int vttoif_tab[10] = {
135 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
136 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
140 * List of vnodes that are ready for recycling.
142 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
145 * Free vnode target. Free vnodes may simply be files which have been stat'd
146 * but not read. This is somewhat common, and a small cache of such files
147 * should be kept to avoid recreation costs.
149 static u_long wantfreevnodes;
150 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
151 /* Number of vnodes in the free list. */
152 static u_long freevnodes;
153 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
155 static int vlru_allow_cache_src;
156 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
157 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
160 * Various variables used for debugging the new implementation of
162 * XXX these are probably of (very) limited utility now.
164 static int reassignbufcalls;
165 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
168 * Cache for the mount type id assigned to NFS. This is used for
169 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
171 int nfs_mount_type = -1;
173 /* To keep more than one thread at a time from running vfs_getnewfsid */
174 static struct mtx mntid_mtx;
177 * Lock for any access to the following:
182 static struct mtx vnode_free_list_mtx;
184 /* Publicly exported FS */
185 struct nfs_public nfs_pub;
187 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
188 static uma_zone_t vnode_zone;
189 static uma_zone_t vnodepoll_zone;
191 /* Set to 1 to print out reclaim of active vnodes */
195 * The workitem queue.
197 * It is useful to delay writes of file data and filesystem metadata
198 * for tens of seconds so that quickly created and deleted files need
199 * not waste disk bandwidth being created and removed. To realize this,
200 * we append vnodes to a "workitem" queue. When running with a soft
201 * updates implementation, most pending metadata dependencies should
202 * not wait for more than a few seconds. Thus, mounted on block devices
203 * are delayed only about a half the time that file data is delayed.
204 * Similarly, directory updates are more critical, so are only delayed
205 * about a third the time that file data is delayed. Thus, there are
206 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
207 * one each second (driven off the filesystem syncer process). The
208 * syncer_delayno variable indicates the next queue that is to be processed.
209 * Items that need to be processed soon are placed in this queue:
211 * syncer_workitem_pending[syncer_delayno]
213 * A delay of fifteen seconds is done by placing the request fifteen
214 * entries later in the queue:
216 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
219 static int syncer_delayno;
220 static long syncer_mask;
221 LIST_HEAD(synclist, bufobj);
222 static struct synclist *syncer_workitem_pending[2];
224 * The sync_mtx protects:
229 * syncer_workitem_pending
230 * syncer_worklist_len
233 static struct mtx sync_mtx;
234 static struct cv sync_wakeup;
236 #define SYNCER_MAXDELAY 32
237 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
238 static int syncdelay = 30; /* max time to delay syncing data */
239 static int filedelay = 30; /* time to delay syncing files */
240 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
241 static int dirdelay = 29; /* time to delay syncing directories */
242 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
243 static int metadelay = 28; /* time to delay syncing metadata */
244 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
245 static int rushjob; /* number of slots to run ASAP */
246 static int stat_rush_requests; /* number of times I/O speeded up */
247 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
250 * When shutting down the syncer, run it at four times normal speed.
252 #define SYNCER_SHUTDOWN_SPEEDUP 4
253 static int sync_vnode_count;
254 static int syncer_worklist_len;
255 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
259 * Number of vnodes we want to exist at any one time. This is mostly used
260 * to size hash tables in vnode-related code. It is normally not used in
261 * getnewvnode(), as wantfreevnodes is normally nonzero.)
263 * XXX desiredvnodes is historical cruft and should not exist.
266 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
267 &desiredvnodes, 0, "Maximum number of vnodes");
268 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
269 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
270 static int vnlru_nowhere;
271 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
272 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
275 * Macros to control when a vnode is freed and recycled. All require
276 * the vnode interlock.
278 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
279 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
280 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
284 * Initialize the vnode management data structures.
286 * Reevaluate the following cap on the number of vnodes after the physical
287 * memory size exceeds 512GB. In the limit, as the physical memory size
288 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
290 #ifndef MAXVNODES_MAX
291 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
294 vntblinit(void *dummy __unused)
296 int physvnodes, virtvnodes;
299 * Desiredvnodes is a function of the physical memory size and the
300 * kernel's heap size. Generally speaking, it scales with the
301 * physical memory size. The ratio of desiredvnodes to physical pages
302 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
303 * marginal ratio of desiredvnodes to physical pages is one to
304 * sixteen. However, desiredvnodes is limited by the kernel's heap
305 * size. The memory required by desiredvnodes vnodes and vm objects
306 * may not exceed one seventh of the kernel's heap size.
308 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
309 cnt.v_page_count) / 16;
310 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
311 sizeof(struct vnode)));
312 desiredvnodes = min(physvnodes, virtvnodes);
313 if (desiredvnodes > MAXVNODES_MAX) {
315 printf("Reducing kern.maxvnodes %d -> %d\n",
316 desiredvnodes, MAXVNODES_MAX);
317 desiredvnodes = MAXVNODES_MAX;
319 wantfreevnodes = desiredvnodes / 4;
320 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
321 TAILQ_INIT(&vnode_free_list);
322 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
323 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
324 NULL, NULL, UMA_ALIGN_PTR, 0);
325 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
326 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
328 * Initialize the filesystem syncer.
330 syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
332 syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
334 syncer_maxdelay = syncer_mask + 1;
335 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
336 cv_init(&sync_wakeup, "syncer");
338 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
342 * Mark a mount point as busy. Used to synchronize access and to delay
343 * unmounting. Eventually, mountlist_mtx is not released on failure.
346 vfs_busy(struct mount *mp, int flags)
349 MPASS((flags & ~MBF_MASK) == 0);
350 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
355 * If mount point is currenly being unmounted, sleep until the
356 * mount point fate is decided. If thread doing the unmounting fails,
357 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
358 * that this mount point has survived the unmount attempt and vfs_busy
359 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
360 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
361 * about to be really destroyed. vfs_busy needs to release its
362 * reference on the mount point in this case and return with ENOENT,
363 * telling the caller that mount mount it tried to busy is no longer
366 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
367 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
370 CTR1(KTR_VFS, "%s: failed busying before sleeping",
374 if (flags & MBF_MNTLSTLOCK)
375 mtx_unlock(&mountlist_mtx);
376 mp->mnt_kern_flag |= MNTK_MWAIT;
377 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0);
378 if (flags & MBF_MNTLSTLOCK)
379 mtx_lock(&mountlist_mtx);
381 if (flags & MBF_MNTLSTLOCK)
382 mtx_unlock(&mountlist_mtx);
389 * Free a busy filesystem.
392 vfs_unbusy(struct mount *mp)
395 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
398 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
400 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
401 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
402 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
403 mp->mnt_kern_flag &= ~MNTK_DRAINING;
404 wakeup(&mp->mnt_lockref);
410 * Lookup a mount point by filesystem identifier.
413 vfs_getvfs(fsid_t *fsid)
417 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
418 mtx_lock(&mountlist_mtx);
419 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
420 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
421 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
423 mtx_unlock(&mountlist_mtx);
427 mtx_unlock(&mountlist_mtx);
428 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
429 return ((struct mount *) 0);
433 * Lookup a mount point by filesystem identifier, busying it before
437 vfs_busyfs(fsid_t *fsid)
442 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
443 mtx_lock(&mountlist_mtx);
444 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
445 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
446 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
447 error = vfs_busy(mp, MBF_MNTLSTLOCK);
449 mtx_unlock(&mountlist_mtx);
455 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
456 mtx_unlock(&mountlist_mtx);
457 return ((struct mount *) 0);
461 * Check if a user can access privileged mount options.
464 vfs_suser(struct mount *mp, struct thread *td)
469 * If the thread is jailed, but this is not a jail-friendly file
470 * system, deny immediately.
472 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
476 * If the file system was mounted outside the jail of the calling
477 * thread, deny immediately.
479 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
483 * If file system supports delegated administration, we don't check
484 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
485 * by the file system itself.
486 * If this is not the user that did original mount, we check for
487 * the PRIV_VFS_MOUNT_OWNER privilege.
489 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
490 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
491 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
498 * Get a new unique fsid. Try to make its val[0] unique, since this value
499 * will be used to create fake device numbers for stat(). Also try (but
500 * not so hard) make its val[0] unique mod 2^16, since some emulators only
501 * support 16-bit device numbers. We end up with unique val[0]'s for the
502 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
504 * Keep in mind that several mounts may be running in parallel. Starting
505 * the search one past where the previous search terminated is both a
506 * micro-optimization and a defense against returning the same fsid to
510 vfs_getnewfsid(struct mount *mp)
512 static u_int16_t mntid_base;
517 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
518 mtx_lock(&mntid_mtx);
519 mtype = mp->mnt_vfc->vfc_typenum;
520 tfsid.val[1] = mtype;
521 mtype = (mtype & 0xFF) << 24;
523 tfsid.val[0] = makedev(255,
524 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
526 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
530 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
531 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
532 mtx_unlock(&mntid_mtx);
536 * Knob to control the precision of file timestamps:
538 * 0 = seconds only; nanoseconds zeroed.
539 * 1 = seconds and nanoseconds, accurate within 1/HZ.
540 * 2 = seconds and nanoseconds, truncated to microseconds.
541 * >=3 = seconds and nanoseconds, maximum precision.
543 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
545 static int timestamp_precision = TSP_SEC;
546 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
547 ×tamp_precision, 0, "");
550 * Get a current timestamp.
553 vfs_timestamp(struct timespec *tsp)
557 switch (timestamp_precision) {
559 tsp->tv_sec = time_second;
567 TIMEVAL_TO_TIMESPEC(&tv, tsp);
577 * Set vnode attributes to VNOVAL
580 vattr_null(struct vattr *vap)
584 vap->va_size = VNOVAL;
585 vap->va_bytes = VNOVAL;
586 vap->va_mode = VNOVAL;
587 vap->va_nlink = VNOVAL;
588 vap->va_uid = VNOVAL;
589 vap->va_gid = VNOVAL;
590 vap->va_fsid = VNOVAL;
591 vap->va_fileid = VNOVAL;
592 vap->va_blocksize = VNOVAL;
593 vap->va_rdev = VNOVAL;
594 vap->va_atime.tv_sec = VNOVAL;
595 vap->va_atime.tv_nsec = VNOVAL;
596 vap->va_mtime.tv_sec = VNOVAL;
597 vap->va_mtime.tv_nsec = VNOVAL;
598 vap->va_ctime.tv_sec = VNOVAL;
599 vap->va_ctime.tv_nsec = VNOVAL;
600 vap->va_birthtime.tv_sec = VNOVAL;
601 vap->va_birthtime.tv_nsec = VNOVAL;
602 vap->va_flags = VNOVAL;
603 vap->va_gen = VNOVAL;
608 * This routine is called when we have too many vnodes. It attempts
609 * to free <count> vnodes and will potentially free vnodes that still
610 * have VM backing store (VM backing store is typically the cause
611 * of a vnode blowout so we want to do this). Therefore, this operation
612 * is not considered cheap.
614 * A number of conditions may prevent a vnode from being reclaimed.
615 * the buffer cache may have references on the vnode, a directory
616 * vnode may still have references due to the namei cache representing
617 * underlying files, or the vnode may be in active use. It is not
618 * desireable to reuse such vnodes. These conditions may cause the
619 * number of vnodes to reach some minimum value regardless of what
620 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
623 vlrureclaim(struct mount *mp)
632 * Calculate the trigger point, don't allow user
633 * screwups to blow us up. This prevents us from
634 * recycling vnodes with lots of resident pages. We
635 * aren't trying to free memory, we are trying to
638 usevnodes = desiredvnodes;
641 trigger = cnt.v_page_count * 2 / usevnodes;
643 vn_start_write(NULL, &mp, V_WAIT);
645 count = mp->mnt_nvnodelistsize / 10 + 1;
647 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
648 while (vp != NULL && vp->v_type == VMARKER)
649 vp = TAILQ_NEXT(vp, v_nmntvnodes);
652 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
653 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
658 * If it's been deconstructed already, it's still
659 * referenced, or it exceeds the trigger, skip it.
661 if (vp->v_usecount ||
662 (!vlru_allow_cache_src &&
663 !LIST_EMPTY(&(vp)->v_cache_src)) ||
664 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
665 vp->v_object->resident_page_count > trigger)) {
671 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
673 goto next_iter_mntunlocked;
677 * v_usecount may have been bumped after VOP_LOCK() dropped
678 * the vnode interlock and before it was locked again.
680 * It is not necessary to recheck VI_DOOMED because it can
681 * only be set by another thread that holds both the vnode
682 * lock and vnode interlock. If another thread has the
683 * vnode lock before we get to VOP_LOCK() and obtains the
684 * vnode interlock after VOP_LOCK() drops the vnode
685 * interlock, the other thread will be unable to drop the
686 * vnode lock before our VOP_LOCK() call fails.
688 if (vp->v_usecount ||
689 (!vlru_allow_cache_src &&
690 !LIST_EMPTY(&(vp)->v_cache_src)) ||
691 (vp->v_object != NULL &&
692 vp->v_object->resident_page_count > trigger)) {
693 VOP_UNLOCK(vp, LK_INTERLOCK);
694 goto next_iter_mntunlocked;
696 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
697 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
702 next_iter_mntunlocked:
703 if ((count % 256) != 0)
707 if ((count % 256) != 0)
716 vn_finished_write(mp);
721 * Attempt to keep the free list at wantfreevnodes length.
724 vnlru_free(int count)
729 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
730 for (; count > 0; count--) {
731 vp = TAILQ_FIRST(&vnode_free_list);
733 * The list can be modified while the free_list_mtx
734 * has been dropped and vp could be NULL here.
738 VNASSERT(vp->v_op != NULL, vp,
739 ("vnlru_free: vnode already reclaimed."));
740 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
742 * Don't recycle if we can't get the interlock.
744 if (!VI_TRYLOCK(vp)) {
745 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
748 VNASSERT(VCANRECYCLE(vp), vp,
749 ("vp inconsistent on freelist"));
751 vp->v_iflag &= ~VI_FREE;
753 mtx_unlock(&vnode_free_list_mtx);
755 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
757 VFS_UNLOCK_GIANT(vfslocked);
759 * If the recycled succeeded this vdrop will actually free
760 * the vnode. If not it will simply place it back on
764 mtx_lock(&vnode_free_list_mtx);
768 * Attempt to recycle vnodes in a context that is always safe to block.
769 * Calling vlrurecycle() from the bowels of filesystem code has some
770 * interesting deadlock problems.
772 static struct proc *vnlruproc;
773 static int vnlruproc_sig;
778 struct mount *mp, *nmp;
780 struct proc *p = vnlruproc;
782 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
786 kproc_suspend_check(p);
787 mtx_lock(&vnode_free_list_mtx);
788 if (freevnodes > wantfreevnodes)
789 vnlru_free(freevnodes - wantfreevnodes);
790 if (numvnodes <= desiredvnodes * 9 / 10) {
792 wakeup(&vnlruproc_sig);
793 msleep(vnlruproc, &vnode_free_list_mtx,
794 PVFS|PDROP, "vlruwt", hz);
797 mtx_unlock(&vnode_free_list_mtx);
799 mtx_lock(&mountlist_mtx);
800 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
801 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
802 nmp = TAILQ_NEXT(mp, mnt_list);
805 vfslocked = VFS_LOCK_GIANT(mp);
806 done += vlrureclaim(mp);
807 VFS_UNLOCK_GIANT(vfslocked);
808 mtx_lock(&mountlist_mtx);
809 nmp = TAILQ_NEXT(mp, mnt_list);
812 mtx_unlock(&mountlist_mtx);
815 /* These messages are temporary debugging aids */
816 if (vnlru_nowhere < 5)
817 printf("vnlru process getting nowhere..\n");
818 else if (vnlru_nowhere == 5)
819 printf("vnlru process messages stopped.\n");
822 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
828 static struct kproc_desc vnlru_kp = {
833 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
837 * Routines having to do with the management of the vnode table.
841 vdestroy(struct vnode *vp)
845 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
846 mtx_lock(&vnode_free_list_mtx);
848 mtx_unlock(&vnode_free_list_mtx);
850 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
851 ("cleaned vnode still on the free list."));
852 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
853 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
854 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
855 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
856 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
857 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
858 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
859 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
860 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
861 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
862 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
863 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
866 mac_vnode_destroy(vp);
868 if (vp->v_pollinfo != NULL)
869 destroy_vpollinfo(vp->v_pollinfo);
871 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
874 lockdestroy(vp->v_vnlock);
875 mtx_destroy(&vp->v_interlock);
876 mtx_destroy(BO_MTX(bo));
877 uma_zfree(vnode_zone, vp);
881 * Try to recycle a freed vnode. We abort if anyone picks up a reference
882 * before we actually vgone(). This function must be called with the vnode
883 * held to prevent the vnode from being returned to the free list midway
887 vtryrecycle(struct vnode *vp)
891 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
892 VNASSERT(vp->v_holdcnt, vp,
893 ("vtryrecycle: Recycling vp %p without a reference.", vp));
895 * This vnode may found and locked via some other list, if so we
896 * can't recycle it yet.
898 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
900 "%s: impossible to recycle, vp %p lock is already held",
902 return (EWOULDBLOCK);
905 * Don't recycle if its filesystem is being suspended.
907 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
910 "%s: impossible to recycle, cannot start the write for %p",
915 * If we got this far, we need to acquire the interlock and see if
916 * anyone picked up this vnode from another list. If not, we will
917 * mark it with DOOMED via vgonel() so that anyone who does find it
921 if (vp->v_usecount) {
922 VOP_UNLOCK(vp, LK_INTERLOCK);
923 vn_finished_write(vnmp);
925 "%s: impossible to recycle, %p is already referenced",
929 if ((vp->v_iflag & VI_DOOMED) == 0)
931 VOP_UNLOCK(vp, LK_INTERLOCK);
932 vn_finished_write(vnmp);
937 * Return the next vnode from the free list.
940 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
943 struct vnode *vp = NULL;
946 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
947 mtx_lock(&vnode_free_list_mtx);
949 * Lend our context to reclaim vnodes if they've exceeded the max.
951 if (freevnodes > wantfreevnodes)
954 * Wait for available vnodes.
956 if (numvnodes > desiredvnodes) {
957 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
959 * File system is beeing suspended, we cannot risk a
960 * deadlock here, so allocate new vnode anyway.
962 if (freevnodes > wantfreevnodes)
963 vnlru_free(freevnodes - wantfreevnodes);
966 if (vnlruproc_sig == 0) {
967 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
970 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
972 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
973 if (numvnodes > desiredvnodes) {
974 mtx_unlock(&vnode_free_list_mtx);
981 mtx_unlock(&vnode_free_list_mtx);
982 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
986 vp->v_vnlock = &vp->v_lock;
987 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
989 * By default, don't allow shared locks unless filesystems
992 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
998 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
999 bo->bo_ops = &buf_ops_bio;
1000 bo->bo_private = vp;
1001 TAILQ_INIT(&bo->bo_clean.bv_hd);
1002 TAILQ_INIT(&bo->bo_dirty.bv_hd);
1004 * Initialize namecache.
1006 LIST_INIT(&vp->v_cache_src);
1007 TAILQ_INIT(&vp->v_cache_dst);
1009 * Finalize various vnode identity bits.
1014 v_incr_usecount(vp);
1018 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1019 mac_vnode_associate_singlelabel(mp, vp);
1020 else if (mp == NULL && vops != &dead_vnodeops)
1021 printf("NULL mp in getnewvnode()\n");
1024 bo->bo_bsize = mp->mnt_stat.f_iosize;
1025 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1026 vp->v_vflag |= VV_NOKNOTE;
1034 * Delete from old mount point vnode list, if on one.
1037 delmntque(struct vnode *vp)
1046 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1047 ("bad mount point vnode list size"));
1048 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1049 mp->mnt_nvnodelistsize--;
1055 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1059 vp->v_op = &dead_vnodeops;
1060 /* XXX non mp-safe fs may still call insmntque with vnode
1062 if (!VOP_ISLOCKED(vp))
1063 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1069 * Insert into list of vnodes for the new mount point, if available.
1072 insmntque1(struct vnode *vp, struct mount *mp,
1073 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1077 KASSERT(vp->v_mount == NULL,
1078 ("insmntque: vnode already on per mount vnode list"));
1079 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1080 #ifdef DEBUG_VFS_LOCKS
1081 if (!VFS_NEEDSGIANT(mp))
1082 ASSERT_VOP_ELOCKED(vp,
1083 "insmntque: mp-safe fs and non-locked vp");
1086 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1087 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1088 mp->mnt_nvnodelistsize == 0)) {
1089 locked = VOP_ISLOCKED(vp);
1090 if (!locked || (locked == LK_EXCLUSIVE &&
1091 (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1100 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1101 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1102 ("neg mount point vnode list size"));
1103 mp->mnt_nvnodelistsize++;
1109 insmntque(struct vnode *vp, struct mount *mp)
1112 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1116 * Flush out and invalidate all buffers associated with a bufobj
1117 * Called with the underlying object locked.
1120 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1125 if (flags & V_SAVE) {
1126 error = bufobj_wwait(bo, slpflag, slptimeo);
1131 if (bo->bo_dirty.bv_cnt > 0) {
1133 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1136 * XXX We could save a lock/unlock if this was only
1137 * enabled under INVARIANTS
1140 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1141 panic("vinvalbuf: dirty bufs");
1145 * If you alter this loop please notice that interlock is dropped and
1146 * reacquired in flushbuflist. Special care is needed to ensure that
1147 * no race conditions occur from this.
1150 error = flushbuflist(&bo->bo_clean,
1151 flags, bo, slpflag, slptimeo);
1153 error = flushbuflist(&bo->bo_dirty,
1154 flags, bo, slpflag, slptimeo);
1155 if (error != 0 && error != EAGAIN) {
1159 } while (error != 0);
1162 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1163 * have write I/O in-progress but if there is a VM object then the
1164 * VM object can also have read-I/O in-progress.
1167 bufobj_wwait(bo, 0, 0);
1169 if (bo->bo_object != NULL) {
1170 VM_OBJECT_LOCK(bo->bo_object);
1171 vm_object_pip_wait(bo->bo_object, "bovlbx");
1172 VM_OBJECT_UNLOCK(bo->bo_object);
1175 } while (bo->bo_numoutput > 0);
1179 * Destroy the copy in the VM cache, too.
1181 if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) {
1182 VM_OBJECT_LOCK(bo->bo_object);
1183 vm_object_page_remove(bo->bo_object, 0, 0,
1184 (flags & V_SAVE) ? TRUE : FALSE);
1185 VM_OBJECT_UNLOCK(bo->bo_object);
1190 if ((flags & (V_ALT | V_NORMAL)) == 0 &&
1191 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1192 panic("vinvalbuf: flush failed");
1199 * Flush out and invalidate all buffers associated with a vnode.
1200 * Called with the underlying object locked.
1203 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1206 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1207 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1208 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1212 * Flush out buffers on the specified list.
1216 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1219 struct buf *bp, *nbp;
1224 ASSERT_BO_LOCKED(bo);
1227 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1228 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1229 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1235 lblkno = nbp->b_lblkno;
1236 xflags = nbp->b_xflags &
1237 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1240 error = BUF_TIMELOCK(bp,
1241 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1242 "flushbuf", slpflag, slptimeo);
1245 return (error != ENOLCK ? error : EAGAIN);
1247 KASSERT(bp->b_bufobj == bo,
1248 ("bp %p wrong b_bufobj %p should be %p",
1249 bp, bp->b_bufobj, bo));
1250 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1256 * XXX Since there are no node locks for NFS, I
1257 * believe there is a slight chance that a delayed
1258 * write will occur while sleeping just above, so
1261 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1266 bp->b_flags |= B_ASYNC;
1269 return (EAGAIN); /* XXX: why not loop ? */
1274 bp->b_flags |= (B_INVAL | B_RELBUF);
1275 bp->b_flags &= ~B_ASYNC;
1279 (nbp->b_bufobj != bo ||
1280 nbp->b_lblkno != lblkno ||
1282 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1283 break; /* nbp invalid */
1289 * Truncate a file's buffer and pages to a specified length. This
1290 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1294 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1295 off_t length, int blksize)
1297 struct buf *bp, *nbp;
1302 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1303 vp, cred, blksize, (uintmax_t)length);
1306 * Round up to the *next* lbn.
1308 trunclbn = (length + blksize - 1) / blksize;
1310 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1317 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1318 if (bp->b_lblkno < trunclbn)
1321 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1322 BO_MTX(bo)) == ENOLCK)
1328 bp->b_flags |= (B_INVAL | B_RELBUF);
1329 bp->b_flags &= ~B_ASYNC;
1334 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1335 (nbp->b_vp != vp) ||
1336 (nbp->b_flags & B_DELWRI))) {
1342 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1343 if (bp->b_lblkno < trunclbn)
1346 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1347 BO_MTX(bo)) == ENOLCK)
1352 bp->b_flags |= (B_INVAL | B_RELBUF);
1353 bp->b_flags &= ~B_ASYNC;
1357 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1358 (nbp->b_vp != vp) ||
1359 (nbp->b_flags & B_DELWRI) == 0)) {
1368 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1369 if (bp->b_lblkno > 0)
1372 * Since we hold the vnode lock this should only
1373 * fail if we're racing with the buf daemon.
1376 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1377 BO_MTX(bo)) == ENOLCK) {
1380 VNASSERT((bp->b_flags & B_DELWRI), vp,
1381 ("buf(%p) on dirty queue without DELWRI", bp));
1392 bufobj_wwait(bo, 0, 0);
1394 vnode_pager_setsize(vp, length);
1400 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1403 * NOTE: We have to deal with the special case of a background bitmap
1404 * buffer, a situation where two buffers will have the same logical
1405 * block offset. We want (1) only the foreground buffer to be accessed
1406 * in a lookup and (2) must differentiate between the foreground and
1407 * background buffer in the splay tree algorithm because the splay
1408 * tree cannot normally handle multiple entities with the same 'index'.
1409 * We accomplish this by adding differentiating flags to the splay tree's
1414 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1417 struct buf *lefttreemax, *righttreemin, *y;
1421 lefttreemax = righttreemin = &dummy;
1423 if (lblkno < root->b_lblkno ||
1424 (lblkno == root->b_lblkno &&
1425 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1426 if ((y = root->b_left) == NULL)
1428 if (lblkno < y->b_lblkno) {
1430 root->b_left = y->b_right;
1433 if ((y = root->b_left) == NULL)
1436 /* Link into the new root's right tree. */
1437 righttreemin->b_left = root;
1438 righttreemin = root;
1439 } else if (lblkno > root->b_lblkno ||
1440 (lblkno == root->b_lblkno &&
1441 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1442 if ((y = root->b_right) == NULL)
1444 if (lblkno > y->b_lblkno) {
1446 root->b_right = y->b_left;
1449 if ((y = root->b_right) == NULL)
1452 /* Link into the new root's left tree. */
1453 lefttreemax->b_right = root;
1460 /* Assemble the new root. */
1461 lefttreemax->b_right = root->b_left;
1462 righttreemin->b_left = root->b_right;
1463 root->b_left = dummy.b_right;
1464 root->b_right = dummy.b_left;
1469 buf_vlist_remove(struct buf *bp)
1474 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1475 ASSERT_BO_LOCKED(bp->b_bufobj);
1476 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1477 (BX_VNDIRTY|BX_VNCLEAN),
1478 ("buf_vlist_remove: Buf %p is on two lists", bp));
1479 if (bp->b_xflags & BX_VNDIRTY)
1480 bv = &bp->b_bufobj->bo_dirty;
1482 bv = &bp->b_bufobj->bo_clean;
1483 if (bp != bv->bv_root) {
1484 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1485 KASSERT(root == bp, ("splay lookup failed in remove"));
1487 if (bp->b_left == NULL) {
1490 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1491 root->b_right = bp->b_right;
1494 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1496 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1500 * Add the buffer to the sorted clean or dirty block list using a
1501 * splay tree algorithm.
1503 * NOTE: xflags is passed as a constant, optimizing this inline function!
1506 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1511 ASSERT_BO_LOCKED(bo);
1512 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1513 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1514 bp->b_xflags |= xflags;
1515 if (xflags & BX_VNDIRTY)
1520 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1524 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1525 } else if (bp->b_lblkno < root->b_lblkno ||
1526 (bp->b_lblkno == root->b_lblkno &&
1527 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1528 bp->b_left = root->b_left;
1530 root->b_left = NULL;
1531 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1533 bp->b_right = root->b_right;
1535 root->b_right = NULL;
1536 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1543 * Lookup a buffer using the splay tree. Note that we specifically avoid
1544 * shadow buffers used in background bitmap writes.
1546 * This code isn't quite efficient as it could be because we are maintaining
1547 * two sorted lists and do not know which list the block resides in.
1549 * During a "make buildworld" the desired buffer is found at one of
1550 * the roots more than 60% of the time. Thus, checking both roots
1551 * before performing either splay eliminates unnecessary splays on the
1552 * first tree splayed.
1555 gbincore(struct bufobj *bo, daddr_t lblkno)
1559 ASSERT_BO_LOCKED(bo);
1560 if ((bp = bo->bo_clean.bv_root) != NULL &&
1561 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1563 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1564 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1566 if ((bp = bo->bo_clean.bv_root) != NULL) {
1567 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1568 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1571 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1572 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1573 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1580 * Associate a buffer with a vnode.
1583 bgetvp(struct vnode *vp, struct buf *bp)
1588 ASSERT_BO_LOCKED(bo);
1589 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1591 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1592 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1593 ("bgetvp: bp already attached! %p", bp));
1596 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1597 bp->b_flags |= B_NEEDSGIANT;
1601 * Insert onto list for new vnode.
1603 buf_vlist_add(bp, bo, BX_VNCLEAN);
1607 * Disassociate a buffer from a vnode.
1610 brelvp(struct buf *bp)
1615 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1616 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1619 * Delete from old vnode list, if on one.
1621 vp = bp->b_vp; /* XXX */
1624 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1625 buf_vlist_remove(bp);
1627 panic("brelvp: Buffer %p not on queue.", bp);
1628 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1629 bo->bo_flag &= ~BO_ONWORKLST;
1630 mtx_lock(&sync_mtx);
1631 LIST_REMOVE(bo, bo_synclist);
1632 syncer_worklist_len--;
1633 mtx_unlock(&sync_mtx);
1635 bp->b_flags &= ~B_NEEDSGIANT;
1637 bp->b_bufobj = NULL;
1643 * Add an item to the syncer work queue.
1646 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1650 ASSERT_BO_LOCKED(bo);
1652 mtx_lock(&sync_mtx);
1653 if (bo->bo_flag & BO_ONWORKLST)
1654 LIST_REMOVE(bo, bo_synclist);
1656 bo->bo_flag |= BO_ONWORKLST;
1657 syncer_worklist_len++;
1660 if (delay > syncer_maxdelay - 2)
1661 delay = syncer_maxdelay - 2;
1662 slot = (syncer_delayno + delay) & syncer_mask;
1664 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1666 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1668 mtx_unlock(&sync_mtx);
1672 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1676 mtx_lock(&sync_mtx);
1677 len = syncer_worklist_len - sync_vnode_count;
1678 mtx_unlock(&sync_mtx);
1679 error = SYSCTL_OUT(req, &len, sizeof(len));
1683 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1684 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1686 static struct proc *updateproc;
1687 static void sched_sync(void);
1688 static struct kproc_desc up_kp = {
1693 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1696 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1701 *bo = LIST_FIRST(slp);
1704 vp = (*bo)->__bo_vnode; /* XXX */
1705 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1708 * We use vhold in case the vnode does not
1709 * successfully sync. vhold prevents the vnode from
1710 * going away when we unlock the sync_mtx so that
1711 * we can acquire the vnode interlock.
1714 mtx_unlock(&sync_mtx);
1716 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1718 mtx_lock(&sync_mtx);
1719 return (*bo == LIST_FIRST(slp));
1721 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1722 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1724 vn_finished_write(mp);
1726 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1728 * Put us back on the worklist. The worklist
1729 * routine will remove us from our current
1730 * position and then add us back in at a later
1733 vn_syncer_add_to_worklist(*bo, syncdelay);
1737 mtx_lock(&sync_mtx);
1742 * System filesystem synchronizer daemon.
1747 struct synclist *gnext, *next;
1748 struct synclist *gslp, *slp;
1751 struct thread *td = curthread;
1753 int net_worklist_len;
1754 int syncer_final_iter;
1759 syncer_final_iter = 0;
1761 syncer_state = SYNCER_RUNNING;
1762 starttime = time_uptime;
1763 td->td_pflags |= TDP_NORUNNINGBUF;
1765 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1768 mtx_lock(&sync_mtx);
1770 if (syncer_state == SYNCER_FINAL_DELAY &&
1771 syncer_final_iter == 0) {
1772 mtx_unlock(&sync_mtx);
1773 kproc_suspend_check(td->td_proc);
1774 mtx_lock(&sync_mtx);
1776 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1777 if (syncer_state != SYNCER_RUNNING &&
1778 starttime != time_uptime) {
1780 printf("\nSyncing disks, vnodes remaining...");
1783 printf("%d ", net_worklist_len);
1785 starttime = time_uptime;
1788 * Push files whose dirty time has expired. Be careful
1789 * of interrupt race on slp queue.
1791 * Skip over empty worklist slots when shutting down.
1794 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1795 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1796 syncer_delayno += 1;
1797 if (syncer_delayno == syncer_maxdelay)
1799 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1800 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1802 * If the worklist has wrapped since the
1803 * it was emptied of all but syncer vnodes,
1804 * switch to the FINAL_DELAY state and run
1805 * for one more second.
1807 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1808 net_worklist_len == 0 &&
1809 last_work_seen == syncer_delayno) {
1810 syncer_state = SYNCER_FINAL_DELAY;
1811 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1813 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1814 LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1817 * Keep track of the last time there was anything
1818 * on the worklist other than syncer vnodes.
1819 * Return to the SHUTTING_DOWN state if any
1822 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1823 last_work_seen = syncer_delayno;
1824 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1825 syncer_state = SYNCER_SHUTTING_DOWN;
1826 while (!LIST_EMPTY(slp)) {
1827 error = sync_vnode(slp, &bo, td);
1829 LIST_REMOVE(bo, bo_synclist);
1830 LIST_INSERT_HEAD(next, bo, bo_synclist);
1834 if (!LIST_EMPTY(gslp)) {
1835 mtx_unlock(&sync_mtx);
1837 mtx_lock(&sync_mtx);
1838 while (!LIST_EMPTY(gslp)) {
1839 error = sync_vnode(gslp, &bo, td);
1841 LIST_REMOVE(bo, bo_synclist);
1842 LIST_INSERT_HEAD(gnext, bo,
1849 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1850 syncer_final_iter--;
1852 * The variable rushjob allows the kernel to speed up the
1853 * processing of the filesystem syncer process. A rushjob
1854 * value of N tells the filesystem syncer to process the next
1855 * N seconds worth of work on its queue ASAP. Currently rushjob
1856 * is used by the soft update code to speed up the filesystem
1857 * syncer process when the incore state is getting so far
1858 * ahead of the disk that the kernel memory pool is being
1859 * threatened with exhaustion.
1866 * Just sleep for a short period of time between
1867 * iterations when shutting down to allow some I/O
1870 * If it has taken us less than a second to process the
1871 * current work, then wait. Otherwise start right over
1872 * again. We can still lose time if any single round
1873 * takes more than two seconds, but it does not really
1874 * matter as we are just trying to generally pace the
1875 * filesystem activity.
1877 if (syncer_state != SYNCER_RUNNING)
1878 cv_timedwait(&sync_wakeup, &sync_mtx,
1879 hz / SYNCER_SHUTDOWN_SPEEDUP);
1880 else if (time_uptime == starttime)
1881 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1886 * Request the syncer daemon to speed up its work.
1887 * We never push it to speed up more than half of its
1888 * normal turn time, otherwise it could take over the cpu.
1891 speedup_syncer(void)
1895 mtx_lock(&sync_mtx);
1896 if (rushjob < syncdelay / 2) {
1898 stat_rush_requests += 1;
1901 mtx_unlock(&sync_mtx);
1902 cv_broadcast(&sync_wakeup);
1907 * Tell the syncer to speed up its work and run though its work
1908 * list several times, then tell it to shut down.
1911 syncer_shutdown(void *arg, int howto)
1914 if (howto & RB_NOSYNC)
1916 mtx_lock(&sync_mtx);
1917 syncer_state = SYNCER_SHUTTING_DOWN;
1919 mtx_unlock(&sync_mtx);
1920 cv_broadcast(&sync_wakeup);
1921 kproc_shutdown(arg, howto);
1925 * Reassign a buffer from one vnode to another.
1926 * Used to assign file specific control information
1927 * (indirect blocks) to the vnode to which they belong.
1930 reassignbuf(struct buf *bp)
1943 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1944 bp, bp->b_vp, bp->b_flags);
1946 * B_PAGING flagged buffers cannot be reassigned because their vp
1947 * is not fully linked in.
1949 if (bp->b_flags & B_PAGING)
1950 panic("cannot reassign paging buffer");
1953 * Delete from old vnode list, if on one.
1956 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1957 buf_vlist_remove(bp);
1959 panic("reassignbuf: Buffer %p not on queue.", bp);
1961 * If dirty, put on list of dirty buffers; otherwise insert onto list
1964 if (bp->b_flags & B_DELWRI) {
1965 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1966 switch (vp->v_type) {
1976 vn_syncer_add_to_worklist(bo, delay);
1978 buf_vlist_add(bp, bo, BX_VNDIRTY);
1980 buf_vlist_add(bp, bo, BX_VNCLEAN);
1982 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1983 mtx_lock(&sync_mtx);
1984 LIST_REMOVE(bo, bo_synclist);
1985 syncer_worklist_len--;
1986 mtx_unlock(&sync_mtx);
1987 bo->bo_flag &= ~BO_ONWORKLST;
1992 bp = TAILQ_FIRST(&bv->bv_hd);
1993 KASSERT(bp == NULL || bp->b_bufobj == bo,
1994 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1995 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1996 KASSERT(bp == NULL || bp->b_bufobj == bo,
1997 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1999 bp = TAILQ_FIRST(&bv->bv_hd);
2000 KASSERT(bp == NULL || bp->b_bufobj == bo,
2001 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2002 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2003 KASSERT(bp == NULL || bp->b_bufobj == bo,
2004 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2010 * Increment the use and hold counts on the vnode, taking care to reference
2011 * the driver's usecount if this is a chardev. The vholdl() will remove
2012 * the vnode from the free list if it is presently free. Requires the
2013 * vnode interlock and returns with it held.
2016 v_incr_usecount(struct vnode *vp)
2019 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2021 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2023 vp->v_rdev->si_usecount++;
2030 * Turn a holdcnt into a use+holdcnt such that only one call to
2031 * v_decr_usecount is needed.
2034 v_upgrade_usecount(struct vnode *vp)
2037 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2039 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2041 vp->v_rdev->si_usecount++;
2047 * Decrement the vnode use and hold count along with the driver's usecount
2048 * if this is a chardev. The vdropl() below releases the vnode interlock
2049 * as it may free the vnode.
2052 v_decr_usecount(struct vnode *vp)
2055 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2056 VNASSERT(vp->v_usecount > 0, vp,
2057 ("v_decr_usecount: negative usecount"));
2058 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2060 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2062 vp->v_rdev->si_usecount--;
2069 * Decrement only the use count and driver use count. This is intended to
2070 * be paired with a follow on vdropl() to release the remaining hold count.
2071 * In this way we may vgone() a vnode with a 0 usecount without risk of
2072 * having it end up on a free list because the hold count is kept above 0.
2075 v_decr_useonly(struct vnode *vp)
2078 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2079 VNASSERT(vp->v_usecount > 0, vp,
2080 ("v_decr_useonly: negative usecount"));
2081 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2083 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2085 vp->v_rdev->si_usecount--;
2091 * Grab a particular vnode from the free list, increment its
2092 * reference count and lock it. VI_DOOMED is set if the vnode
2093 * is being destroyed. Only callers who specify LK_RETRY will
2094 * see doomed vnodes. If inactive processing was delayed in
2095 * vput try to do it here.
2098 vget(struct vnode *vp, int flags, struct thread *td)
2103 VFS_ASSERT_GIANT(vp->v_mount);
2104 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2105 ("vget: invalid lock operation"));
2106 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2108 if ((flags & LK_INTERLOCK) == 0)
2111 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2113 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2117 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2118 panic("vget: vn_lock failed to return ENOENT\n");
2120 /* Upgrade our holdcnt to a usecount. */
2121 v_upgrade_usecount(vp);
2123 * We don't guarantee that any particular close will
2124 * trigger inactive processing so just make a best effort
2125 * here at preventing a reference to a removed file. If
2126 * we don't succeed no harm is done.
2128 if (vp->v_iflag & VI_OWEINACT) {
2129 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2130 (flags & LK_NOWAIT) == 0)
2132 vp->v_iflag &= ~VI_OWEINACT;
2139 * Increase the reference count of a vnode.
2142 vref(struct vnode *vp)
2145 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2147 v_incr_usecount(vp);
2152 * Return reference count of a vnode.
2154 * The results of this call are only guaranteed when some mechanism other
2155 * than the VI lock is used to stop other processes from gaining references
2156 * to the vnode. This may be the case if the caller holds the only reference.
2157 * This is also useful when stale data is acceptable as race conditions may
2158 * be accounted for by some other means.
2161 vrefcnt(struct vnode *vp)
2166 usecnt = vp->v_usecount;
2172 #define VPUTX_VRELE 1
2173 #define VPUTX_VPUT 2
2174 #define VPUTX_VUNREF 3
2177 vputx(struct vnode *vp, int func)
2181 KASSERT(vp != NULL, ("vputx: null vp"));
2182 if (func == VPUTX_VUNREF)
2183 ASSERT_VOP_ELOCKED(vp, "vunref");
2184 else if (func == VPUTX_VPUT)
2185 ASSERT_VOP_LOCKED(vp, "vput");
2187 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2188 VFS_ASSERT_GIANT(vp->v_mount);
2189 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2192 /* Skip this v_writecount check if we're going to panic below. */
2193 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2194 ("vputx: missed vn_close"));
2197 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2198 vp->v_usecount == 1)) {
2199 if (func == VPUTX_VPUT)
2201 v_decr_usecount(vp);
2205 if (vp->v_usecount != 1) {
2207 vprint("vputx: negative ref count", vp);
2209 panic("vputx: negative ref cnt");
2211 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2213 * We want to hold the vnode until the inactive finishes to
2214 * prevent vgone() races. We drop the use count here and the
2215 * hold count below when we're done.
2219 * We must call VOP_INACTIVE with the node locked. Mark
2220 * as VI_DOINGINACT to avoid recursion.
2222 vp->v_iflag |= VI_OWEINACT;
2223 if (func == VPUTX_VRELE) {
2224 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2226 } else if (func == VPUTX_VPUT && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2227 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | LK_NOWAIT);
2230 if (vp->v_usecount > 0)
2231 vp->v_iflag &= ~VI_OWEINACT;
2233 if (vp->v_iflag & VI_OWEINACT)
2234 vinactive(vp, curthread);
2235 if (func != VPUTX_VUNREF)
2242 * Vnode put/release.
2243 * If count drops to zero, call inactive routine and return to freelist.
2246 vrele(struct vnode *vp)
2249 vputx(vp, VPUTX_VRELE);
2253 * Release an already locked vnode. This give the same effects as
2254 * unlock+vrele(), but takes less time and avoids releasing and
2255 * re-aquiring the lock (as vrele() acquires the lock internally.)
2258 vput(struct vnode *vp)
2261 vputx(vp, VPUTX_VPUT);
2265 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2268 vunref(struct vnode *vp)
2271 vputx(vp, VPUTX_VUNREF);
2275 * Somebody doesn't want the vnode recycled.
2278 vhold(struct vnode *vp)
2287 vholdl(struct vnode *vp)
2290 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2292 if (VSHOULDBUSY(vp))
2297 * Note that there is one less who cares about this vnode. vdrop() is the
2298 * opposite of vhold().
2301 vdrop(struct vnode *vp)
2309 * Drop the hold count of the vnode. If this is the last reference to
2310 * the vnode we will free it if it has been vgone'd otherwise it is
2311 * placed on the free list.
2314 vdropl(struct vnode *vp)
2317 ASSERT_VI_LOCKED(vp, "vdropl");
2318 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2319 if (vp->v_holdcnt <= 0)
2320 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2322 if (vp->v_holdcnt == 0) {
2323 if (vp->v_iflag & VI_DOOMED) {
2324 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__,
2335 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2336 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2337 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2338 * failed lock upgrade.
2341 vinactive(struct vnode *vp, struct thread *td)
2344 ASSERT_VOP_ELOCKED(vp, "vinactive");
2345 ASSERT_VI_LOCKED(vp, "vinactive");
2346 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2347 ("vinactive: recursed on VI_DOINGINACT"));
2348 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2349 vp->v_iflag |= VI_DOINGINACT;
2350 vp->v_iflag &= ~VI_OWEINACT;
2352 VOP_INACTIVE(vp, td);
2354 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2355 ("vinactive: lost VI_DOINGINACT"));
2356 vp->v_iflag &= ~VI_DOINGINACT;
2360 * Remove any vnodes in the vnode table belonging to mount point mp.
2362 * If FORCECLOSE is not specified, there should not be any active ones,
2363 * return error if any are found (nb: this is a user error, not a
2364 * system error). If FORCECLOSE is specified, detach any active vnodes
2367 * If WRITECLOSE is set, only flush out regular file vnodes open for
2370 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2372 * `rootrefs' specifies the base reference count for the root vnode
2373 * of this filesystem. The root vnode is considered busy if its
2374 * v_usecount exceeds this value. On a successful return, vflush(, td)
2375 * will call vrele() on the root vnode exactly rootrefs times.
2376 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2380 static int busyprt = 0; /* print out busy vnodes */
2381 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
2385 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2387 struct vnode *vp, *mvp, *rootvp = NULL;
2389 int busy = 0, error;
2391 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2394 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2395 ("vflush: bad args"));
2397 * Get the filesystem root vnode. We can vput() it
2398 * immediately, since with rootrefs > 0, it won't go away.
2400 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2401 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2410 MNT_VNODE_FOREACH(vp, mp, mvp) {
2415 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2419 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2423 * Skip over a vnodes marked VV_SYSTEM.
2425 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2432 * If WRITECLOSE is set, flush out unlinked but still open
2433 * files (even if open only for reading) and regular file
2434 * vnodes open for writing.
2436 if (flags & WRITECLOSE) {
2437 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2440 if ((vp->v_type == VNON ||
2441 (error == 0 && vattr.va_nlink > 0)) &&
2442 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2451 * With v_usecount == 0, all we need to do is clear out the
2452 * vnode data structures and we are done.
2454 * If FORCECLOSE is set, forcibly close the vnode.
2456 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2457 VNASSERT(vp->v_usecount == 0 ||
2458 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2459 ("device VNODE %p is FORCECLOSED", vp));
2465 vprint("vflush: busy vnode", vp);
2473 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2475 * If just the root vnode is busy, and if its refcount
2476 * is equal to `rootrefs', then go ahead and kill it.
2479 KASSERT(busy > 0, ("vflush: not busy"));
2480 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2481 ("vflush: usecount %d < rootrefs %d",
2482 rootvp->v_usecount, rootrefs));
2483 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2484 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2486 VOP_UNLOCK(rootvp, 0);
2492 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2496 for (; rootrefs > 0; rootrefs--)
2502 * Recycle an unused vnode to the front of the free list.
2505 vrecycle(struct vnode *vp, struct thread *td)
2509 ASSERT_VOP_ELOCKED(vp, "vrecycle");
2510 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2513 if (vp->v_usecount == 0) {
2522 * Eliminate all activity associated with a vnode
2523 * in preparation for reuse.
2526 vgone(struct vnode *vp)
2534 * vgone, with the vp interlock held.
2537 vgonel(struct vnode *vp)
2544 ASSERT_VOP_ELOCKED(vp, "vgonel");
2545 ASSERT_VI_LOCKED(vp, "vgonel");
2546 VNASSERT(vp->v_holdcnt, vp,
2547 ("vgonel: vp %p has no reference.", vp));
2548 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2552 * Don't vgonel if we're already doomed.
2554 if (vp->v_iflag & VI_DOOMED)
2556 vp->v_iflag |= VI_DOOMED;
2558 * Check to see if the vnode is in use. If so, we have to call
2559 * VOP_CLOSE() and VOP_INACTIVE().
2561 active = vp->v_usecount;
2562 oweinact = (vp->v_iflag & VI_OWEINACT);
2565 * Clean out any buffers associated with the vnode.
2566 * If the flush fails, just toss the buffers.
2569 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2570 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2571 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2572 vinvalbuf(vp, 0, 0, 0);
2575 * If purging an active vnode, it must be closed and
2576 * deactivated before being reclaimed.
2579 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2580 if (oweinact || active) {
2582 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2587 * Reclaim the vnode.
2589 if (VOP_RECLAIM(vp, td))
2590 panic("vgone: cannot reclaim");
2592 vn_finished_secondary_write(mp);
2593 VNASSERT(vp->v_object == NULL, vp,
2594 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2596 * Clear the advisory locks and wake up waiting threads.
2598 lf_purgelocks(vp, &(vp->v_lockf));
2600 * Delete from old mount point vnode list.
2605 * Done with purge, reset to the standard lock and invalidate
2609 vp->v_vnlock = &vp->v_lock;
2610 vp->v_op = &dead_vnodeops;
2616 * Calculate the total number of references to a special device.
2619 vcount(struct vnode *vp)
2624 count = vp->v_rdev->si_usecount;
2630 * Same as above, but using the struct cdev *as argument
2633 count_dev(struct cdev *dev)
2638 count = dev->si_usecount;
2644 * Print out a description of a vnode.
2646 static char *typename[] =
2647 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2651 vn_printf(struct vnode *vp, const char *fmt, ...)
2654 char buf[256], buf2[16];
2660 printf("%p: ", (void *)vp);
2661 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2662 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2663 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2666 if (vp->v_vflag & VV_ROOT)
2667 strlcat(buf, "|VV_ROOT", sizeof(buf));
2668 if (vp->v_vflag & VV_ISTTY)
2669 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2670 if (vp->v_vflag & VV_NOSYNC)
2671 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2672 if (vp->v_vflag & VV_CACHEDLABEL)
2673 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2674 if (vp->v_vflag & VV_TEXT)
2675 strlcat(buf, "|VV_TEXT", sizeof(buf));
2676 if (vp->v_vflag & VV_COPYONWRITE)
2677 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2678 if (vp->v_vflag & VV_SYSTEM)
2679 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2680 if (vp->v_vflag & VV_PROCDEP)
2681 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2682 if (vp->v_vflag & VV_NOKNOTE)
2683 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2684 if (vp->v_vflag & VV_DELETED)
2685 strlcat(buf, "|VV_DELETED", sizeof(buf));
2686 if (vp->v_vflag & VV_MD)
2687 strlcat(buf, "|VV_MD", sizeof(buf));
2688 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2689 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2690 VV_NOKNOTE | VV_DELETED | VV_MD);
2692 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2693 strlcat(buf, buf2, sizeof(buf));
2695 if (vp->v_iflag & VI_MOUNT)
2696 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2697 if (vp->v_iflag & VI_AGE)
2698 strlcat(buf, "|VI_AGE", sizeof(buf));
2699 if (vp->v_iflag & VI_DOOMED)
2700 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2701 if (vp->v_iflag & VI_FREE)
2702 strlcat(buf, "|VI_FREE", sizeof(buf));
2703 if (vp->v_iflag & VI_DOINGINACT)
2704 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2705 if (vp->v_iflag & VI_OWEINACT)
2706 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2707 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2708 VI_DOINGINACT | VI_OWEINACT);
2710 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2711 strlcat(buf, buf2, sizeof(buf));
2713 printf(" flags (%s)\n", buf + 1);
2714 if (mtx_owned(VI_MTX(vp)))
2715 printf(" VI_LOCKed");
2716 if (vp->v_object != NULL)
2717 printf(" v_object %p ref %d pages %d\n",
2718 vp->v_object, vp->v_object->ref_count,
2719 vp->v_object->resident_page_count);
2721 lockmgr_printinfo(vp->v_vnlock);
2722 if (vp->v_data != NULL)
2728 * List all of the locked vnodes in the system.
2729 * Called when debugging the kernel.
2731 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2733 struct mount *mp, *nmp;
2737 * Note: because this is DDB, we can't obey the locking semantics
2738 * for these structures, which means we could catch an inconsistent
2739 * state and dereference a nasty pointer. Not much to be done
2742 db_printf("Locked vnodes\n");
2743 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2744 nmp = TAILQ_NEXT(mp, mnt_list);
2745 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2746 if (vp->v_type != VMARKER &&
2750 nmp = TAILQ_NEXT(mp, mnt_list);
2755 * Show details about the given vnode.
2757 DB_SHOW_COMMAND(vnode, db_show_vnode)
2763 vp = (struct vnode *)addr;
2764 vn_printf(vp, "vnode ");
2768 * Show details about the given mount point.
2770 DB_SHOW_COMMAND(mount, db_show_mount)
2780 /* No address given, print short info about all mount points. */
2781 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2782 db_printf("%p %s on %s (%s)\n", mp,
2783 mp->mnt_stat.f_mntfromname,
2784 mp->mnt_stat.f_mntonname,
2785 mp->mnt_stat.f_fstypename);
2789 db_printf("\nMore info: show mount <addr>\n");
2793 mp = (struct mount *)addr;
2794 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2795 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2798 flags = mp->mnt_flag;
2799 #define MNT_FLAG(flag) do { \
2800 if (flags & (flag)) { \
2801 if (buf[0] != '\0') \
2802 strlcat(buf, ", ", sizeof(buf)); \
2803 strlcat(buf, (#flag) + 4, sizeof(buf)); \
2807 MNT_FLAG(MNT_RDONLY);
2808 MNT_FLAG(MNT_SYNCHRONOUS);
2809 MNT_FLAG(MNT_NOEXEC);
2810 MNT_FLAG(MNT_NOSUID);
2811 MNT_FLAG(MNT_UNION);
2812 MNT_FLAG(MNT_ASYNC);
2813 MNT_FLAG(MNT_SUIDDIR);
2814 MNT_FLAG(MNT_SOFTDEP);
2815 MNT_FLAG(MNT_NOSYMFOLLOW);
2816 MNT_FLAG(MNT_GJOURNAL);
2817 MNT_FLAG(MNT_MULTILABEL);
2819 MNT_FLAG(MNT_NOATIME);
2820 MNT_FLAG(MNT_NOCLUSTERR);
2821 MNT_FLAG(MNT_NOCLUSTERW);
2822 MNT_FLAG(MNT_NFS4ACLS);
2823 MNT_FLAG(MNT_EXRDONLY);
2824 MNT_FLAG(MNT_EXPORTED);
2825 MNT_FLAG(MNT_DEFEXPORTED);
2826 MNT_FLAG(MNT_EXPORTANON);
2827 MNT_FLAG(MNT_EXKERB);
2828 MNT_FLAG(MNT_EXPUBLIC);
2829 MNT_FLAG(MNT_LOCAL);
2830 MNT_FLAG(MNT_QUOTA);
2831 MNT_FLAG(MNT_ROOTFS);
2833 MNT_FLAG(MNT_IGNORE);
2834 MNT_FLAG(MNT_UPDATE);
2835 MNT_FLAG(MNT_DELEXPORT);
2836 MNT_FLAG(MNT_RELOAD);
2837 MNT_FLAG(MNT_FORCE);
2838 MNT_FLAG(MNT_SNAPSHOT);
2839 MNT_FLAG(MNT_BYFSID);
2843 strlcat(buf, ", ", sizeof(buf));
2844 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2847 db_printf(" mnt_flag = %s\n", buf);
2850 flags = mp->mnt_kern_flag;
2851 #define MNT_KERN_FLAG(flag) do { \
2852 if (flags & (flag)) { \
2853 if (buf[0] != '\0') \
2854 strlcat(buf, ", ", sizeof(buf)); \
2855 strlcat(buf, (#flag) + 5, sizeof(buf)); \
2859 MNT_KERN_FLAG(MNTK_UNMOUNTF);
2860 MNT_KERN_FLAG(MNTK_ASYNC);
2861 MNT_KERN_FLAG(MNTK_SOFTDEP);
2862 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2863 MNT_KERN_FLAG(MNTK_UNMOUNT);
2864 MNT_KERN_FLAG(MNTK_MWAIT);
2865 MNT_KERN_FLAG(MNTK_SUSPEND);
2866 MNT_KERN_FLAG(MNTK_SUSPEND2);
2867 MNT_KERN_FLAG(MNTK_SUSPENDED);
2868 MNT_KERN_FLAG(MNTK_MPSAFE);
2869 MNT_KERN_FLAG(MNTK_NOKNOTE);
2870 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2871 #undef MNT_KERN_FLAG
2874 strlcat(buf, ", ", sizeof(buf));
2875 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2878 db_printf(" mnt_kern_flag = %s\n", buf);
2880 db_printf(" mnt_opt = ");
2881 opt = TAILQ_FIRST(mp->mnt_opt);
2883 db_printf("%s", opt->name);
2884 opt = TAILQ_NEXT(opt, link);
2885 while (opt != NULL) {
2886 db_printf(", %s", opt->name);
2887 opt = TAILQ_NEXT(opt, link);
2893 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
2894 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2895 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2896 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2897 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2898 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2899 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2900 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2901 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2902 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2903 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2904 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2906 db_printf(" mnt_cred = { uid=%u ruid=%u",
2907 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2908 if (jailed(mp->mnt_cred))
2909 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2911 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
2912 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
2913 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2914 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2915 db_printf(" mnt_noasync = %u\n", mp->mnt_noasync);
2916 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2917 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2918 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
2919 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2920 db_printf(" mnt_secondary_accwrites = %d\n",
2921 mp->mnt_secondary_accwrites);
2922 db_printf(" mnt_gjprovider = %s\n",
2923 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
2926 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2927 if (vp->v_type != VMARKER) {
2928 vn_printf(vp, "vnode ");
2937 * Fill in a struct xvfsconf based on a struct vfsconf.
2940 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
2943 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
2944 xvfsp->vfc_typenum = vfsp->vfc_typenum;
2945 xvfsp->vfc_refcount = vfsp->vfc_refcount;
2946 xvfsp->vfc_flags = vfsp->vfc_flags;
2948 * These are unused in userland, we keep them
2949 * to not break binary compatibility.
2951 xvfsp->vfc_vfsops = NULL;
2952 xvfsp->vfc_next = NULL;
2956 * Top level filesystem related information gathering.
2959 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
2961 struct vfsconf *vfsp;
2962 struct xvfsconf xvfsp;
2966 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
2967 bzero(&xvfsp, sizeof(xvfsp));
2968 vfsconf2x(vfsp, &xvfsp);
2969 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
2976 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
2977 "S,xvfsconf", "List of all configured filesystems");
2979 #ifndef BURN_BRIDGES
2980 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
2983 vfs_sysctl(SYSCTL_HANDLER_ARGS)
2985 int *name = (int *)arg1 - 1; /* XXX */
2986 u_int namelen = arg2 + 1; /* XXX */
2987 struct vfsconf *vfsp;
2988 struct xvfsconf xvfsp;
2990 printf("WARNING: userland calling deprecated sysctl, "
2991 "please rebuild world\n");
2993 #if 1 || defined(COMPAT_PRELITE2)
2994 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
2996 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3000 case VFS_MAXTYPENUM:
3003 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3006 return (ENOTDIR); /* overloaded */
3007 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3008 if (vfsp->vfc_typenum == name[2])
3011 return (EOPNOTSUPP);
3012 bzero(&xvfsp, sizeof(xvfsp));
3013 vfsconf2x(vfsp, &xvfsp);
3014 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3016 return (EOPNOTSUPP);
3019 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3020 vfs_sysctl, "Generic filesystem");
3022 #if 1 || defined(COMPAT_PRELITE2)
3025 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3028 struct vfsconf *vfsp;
3029 struct ovfsconf ovfs;
3031 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3032 bzero(&ovfs, sizeof(ovfs));
3033 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3034 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3035 ovfs.vfc_index = vfsp->vfc_typenum;
3036 ovfs.vfc_refcount = vfsp->vfc_refcount;
3037 ovfs.vfc_flags = vfsp->vfc_flags;
3038 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3045 #endif /* 1 || COMPAT_PRELITE2 */
3046 #endif /* !BURN_BRIDGES */
3048 #define KINFO_VNODESLOP 10
3051 * Dump vnode list (via sysctl).
3055 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3063 * Stale numvnodes access is not fatal here.
3066 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3068 /* Make an estimate */
3069 return (SYSCTL_OUT(req, 0, len));
3071 error = sysctl_wire_old_buffer(req, 0);
3074 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3076 mtx_lock(&mountlist_mtx);
3077 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3078 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3081 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3085 xvn[n].xv_size = sizeof *xvn;
3086 xvn[n].xv_vnode = vp;
3087 xvn[n].xv_id = 0; /* XXX compat */
3088 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3090 XV_COPY(writecount);
3096 xvn[n].xv_flag = vp->v_vflag;
3098 switch (vp->v_type) {
3105 if (vp->v_rdev == NULL) {
3109 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3112 xvn[n].xv_socket = vp->v_socket;
3115 xvn[n].xv_fifo = vp->v_fifoinfo;
3120 /* shouldn't happen? */
3128 mtx_lock(&mountlist_mtx);
3133 mtx_unlock(&mountlist_mtx);
3135 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3140 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3141 0, 0, sysctl_vnode, "S,xvnode", "");
3145 * Unmount all filesystems. The list is traversed in reverse order
3146 * of mounting to avoid dependencies.
3149 vfs_unmountall(void)
3155 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3156 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3160 * Since this only runs when rebooting, it is not interlocked.
3162 while(!TAILQ_EMPTY(&mountlist)) {
3163 mp = TAILQ_LAST(&mountlist, mntlist);
3164 error = dounmount(mp, MNT_FORCE, td);
3166 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3168 * XXX: Due to the way in which we mount the root
3169 * file system off of devfs, devfs will generate a
3170 * "busy" warning when we try to unmount it before
3171 * the root. Don't print a warning as a result in
3172 * order to avoid false positive errors that may
3173 * cause needless upset.
3175 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3176 printf("unmount of %s failed (",
3177 mp->mnt_stat.f_mntonname);
3181 printf("%d)\n", error);
3184 /* The unmount has removed mp from the mountlist */
3190 * perform msync on all vnodes under a mount point
3191 * the mount point must be locked.
3194 vfs_msync(struct mount *mp, int flags)
3196 struct vnode *vp, *mvp;
3197 struct vm_object *obj;
3199 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3201 MNT_VNODE_FOREACH(vp, mp, mvp) {
3204 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3205 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3208 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3210 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3218 VM_OBJECT_LOCK(obj);
3219 vm_object_page_clean(obj, 0, 0,
3221 OBJPC_SYNC : OBJPC_NOSYNC);
3222 VM_OBJECT_UNLOCK(obj);
3234 * Mark a vnode as free, putting it up for recycling.
3237 vfree(struct vnode *vp)
3240 ASSERT_VI_LOCKED(vp, "vfree");
3241 mtx_lock(&vnode_free_list_mtx);
3242 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3243 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3244 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3245 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3246 ("vfree: Freeing doomed vnode"));
3247 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3248 if (vp->v_iflag & VI_AGE) {
3249 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3251 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3254 vp->v_iflag &= ~VI_AGE;
3255 vp->v_iflag |= VI_FREE;
3256 mtx_unlock(&vnode_free_list_mtx);
3260 * Opposite of vfree() - mark a vnode as in use.
3263 vbusy(struct vnode *vp)
3265 ASSERT_VI_LOCKED(vp, "vbusy");
3266 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3267 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3268 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3270 mtx_lock(&vnode_free_list_mtx);
3271 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3273 vp->v_iflag &= ~(VI_FREE|VI_AGE);
3274 mtx_unlock(&vnode_free_list_mtx);
3278 destroy_vpollinfo(struct vpollinfo *vi)
3280 knlist_destroy(&vi->vpi_selinfo.si_note);
3281 mtx_destroy(&vi->vpi_lock);
3282 uma_zfree(vnodepoll_zone, vi);
3286 * Initalize per-vnode helper structure to hold poll-related state.
3289 v_addpollinfo(struct vnode *vp)
3291 struct vpollinfo *vi;
3293 if (vp->v_pollinfo != NULL)
3295 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3296 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3297 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3298 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3300 if (vp->v_pollinfo != NULL) {
3302 destroy_vpollinfo(vi);
3305 vp->v_pollinfo = vi;
3310 * Record a process's interest in events which might happen to
3311 * a vnode. Because poll uses the historic select-style interface
3312 * internally, this routine serves as both the ``check for any
3313 * pending events'' and the ``record my interest in future events''
3314 * functions. (These are done together, while the lock is held,
3315 * to avoid race conditions.)
3318 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3322 mtx_lock(&vp->v_pollinfo->vpi_lock);
3323 if (vp->v_pollinfo->vpi_revents & events) {
3325 * This leaves events we are not interested
3326 * in available for the other process which
3327 * which presumably had requested them
3328 * (otherwise they would never have been
3331 events &= vp->v_pollinfo->vpi_revents;
3332 vp->v_pollinfo->vpi_revents &= ~events;
3334 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3337 vp->v_pollinfo->vpi_events |= events;
3338 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3339 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3344 * Routine to create and manage a filesystem syncer vnode.
3346 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3347 static int sync_fsync(struct vop_fsync_args *);
3348 static int sync_inactive(struct vop_inactive_args *);
3349 static int sync_reclaim(struct vop_reclaim_args *);
3351 static struct vop_vector sync_vnodeops = {
3352 .vop_bypass = VOP_EOPNOTSUPP,
3353 .vop_close = sync_close, /* close */
3354 .vop_fsync = sync_fsync, /* fsync */
3355 .vop_inactive = sync_inactive, /* inactive */
3356 .vop_reclaim = sync_reclaim, /* reclaim */
3357 .vop_lock1 = vop_stdlock, /* lock */
3358 .vop_unlock = vop_stdunlock, /* unlock */
3359 .vop_islocked = vop_stdislocked, /* islocked */
3363 * Create a new filesystem syncer vnode for the specified mount point.
3366 vfs_allocate_syncvnode(struct mount *mp)
3370 static long start, incr, next;
3373 /* Allocate a new vnode */
3374 if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3375 mp->mnt_syncer = NULL;
3379 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3380 vp->v_vflag |= VV_FORCEINSMQ;
3381 error = insmntque(vp, mp);
3383 panic("vfs_allocate_syncvnode: insmntque failed");
3384 vp->v_vflag &= ~VV_FORCEINSMQ;
3387 * Place the vnode onto the syncer worklist. We attempt to
3388 * scatter them about on the list so that they will go off
3389 * at evenly distributed times even if all the filesystems
3390 * are mounted at once.
3393 if (next == 0 || next > syncer_maxdelay) {
3397 start = syncer_maxdelay / 2;
3398 incr = syncer_maxdelay;
3404 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3405 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3406 mtx_lock(&sync_mtx);
3408 mtx_unlock(&sync_mtx);
3410 mp->mnt_syncer = vp;
3415 * Do a lazy sync of the filesystem.
3418 sync_fsync(struct vop_fsync_args *ap)
3420 struct vnode *syncvp = ap->a_vp;
3421 struct mount *mp = syncvp->v_mount;
3426 * We only need to do something if this is a lazy evaluation.
3428 if (ap->a_waitfor != MNT_LAZY)
3432 * Move ourselves to the back of the sync list.
3434 bo = &syncvp->v_bufobj;
3436 vn_syncer_add_to_worklist(bo, syncdelay);
3440 * Walk the list of vnodes pushing all that are dirty and
3441 * not already on the sync list.
3443 mtx_lock(&mountlist_mtx);
3444 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3445 mtx_unlock(&mountlist_mtx);
3448 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3454 mp->mnt_kern_flag &= ~MNTK_ASYNC;
3456 vfs_msync(mp, MNT_NOWAIT);
3457 error = VFS_SYNC(mp, MNT_LAZY);
3460 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0)
3461 mp->mnt_kern_flag |= MNTK_ASYNC;
3463 vn_finished_write(mp);
3469 * The syncer vnode is no referenced.
3472 sync_inactive(struct vop_inactive_args *ap)
3480 * The syncer vnode is no longer needed and is being decommissioned.
3482 * Modifications to the worklist must be protected by sync_mtx.
3485 sync_reclaim(struct vop_reclaim_args *ap)
3487 struct vnode *vp = ap->a_vp;
3492 vp->v_mount->mnt_syncer = NULL;
3493 if (bo->bo_flag & BO_ONWORKLST) {
3494 mtx_lock(&sync_mtx);
3495 LIST_REMOVE(bo, bo_synclist);
3496 syncer_worklist_len--;
3498 mtx_unlock(&sync_mtx);
3499 bo->bo_flag &= ~BO_ONWORKLST;
3507 * Check if vnode represents a disk device
3510 vn_isdisk(struct vnode *vp, int *errp)
3516 if (vp->v_type != VCHR)
3518 else if (vp->v_rdev == NULL)
3520 else if (vp->v_rdev->si_devsw == NULL)
3522 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3527 return (error == 0);
3531 * Common filesystem object access control check routine. Accepts a
3532 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3533 * and optional call-by-reference privused argument allowing vaccess()
3534 * to indicate to the caller whether privilege was used to satisfy the
3535 * request (obsoleted). Returns 0 on success, or an errno on failure.
3537 * The ifdef'd CAPABILITIES version is here for reference, but is not
3541 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3542 accmode_t accmode, struct ucred *cred, int *privused)
3544 accmode_t dac_granted;
3545 accmode_t priv_granted;
3547 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3548 ("invalid bit in accmode"));
3551 * Look for a normal, non-privileged way to access the file/directory
3552 * as requested. If it exists, go with that.
3555 if (privused != NULL)
3560 /* Check the owner. */
3561 if (cred->cr_uid == file_uid) {
3562 dac_granted |= VADMIN;
3563 if (file_mode & S_IXUSR)
3564 dac_granted |= VEXEC;
3565 if (file_mode & S_IRUSR)
3566 dac_granted |= VREAD;
3567 if (file_mode & S_IWUSR)
3568 dac_granted |= (VWRITE | VAPPEND);
3570 if ((accmode & dac_granted) == accmode)
3576 /* Otherwise, check the groups (first match) */
3577 if (groupmember(file_gid, cred)) {
3578 if (file_mode & S_IXGRP)
3579 dac_granted |= VEXEC;
3580 if (file_mode & S_IRGRP)
3581 dac_granted |= VREAD;
3582 if (file_mode & S_IWGRP)
3583 dac_granted |= (VWRITE | VAPPEND);
3585 if ((accmode & dac_granted) == accmode)
3591 /* Otherwise, check everyone else. */
3592 if (file_mode & S_IXOTH)
3593 dac_granted |= VEXEC;
3594 if (file_mode & S_IROTH)
3595 dac_granted |= VREAD;
3596 if (file_mode & S_IWOTH)
3597 dac_granted |= (VWRITE | VAPPEND);
3598 if ((accmode & dac_granted) == accmode)
3603 * Build a privilege mask to determine if the set of privileges
3604 * satisfies the requirements when combined with the granted mask
3605 * from above. For each privilege, if the privilege is required,
3606 * bitwise or the request type onto the priv_granted mask.
3612 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3613 * requests, instead of PRIV_VFS_EXEC.
3615 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3616 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3617 priv_granted |= VEXEC;
3619 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3620 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3621 priv_granted |= VEXEC;
3624 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3625 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3626 priv_granted |= VREAD;
3628 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3629 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3630 priv_granted |= (VWRITE | VAPPEND);
3632 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3633 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3634 priv_granted |= VADMIN;
3636 if ((accmode & (priv_granted | dac_granted)) == accmode) {
3637 /* XXX audit: privilege used */
3638 if (privused != NULL)
3643 return ((accmode & VADMIN) ? EPERM : EACCES);
3647 * Credential check based on process requesting service, and per-attribute
3651 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3652 struct thread *td, accmode_t accmode)
3656 * Kernel-invoked always succeeds.
3662 * Do not allow privileged processes in jail to directly manipulate
3663 * system attributes.
3665 switch (attrnamespace) {
3666 case EXTATTR_NAMESPACE_SYSTEM:
3667 /* Potentially should be: return (EPERM); */
3668 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3669 case EXTATTR_NAMESPACE_USER:
3670 return (VOP_ACCESS(vp, accmode, cred, td));
3676 #ifdef DEBUG_VFS_LOCKS
3678 * This only exists to supress warnings from unlocked specfs accesses. It is
3679 * no longer ok to have an unlocked VFS.
3681 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
3682 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
3684 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3685 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
3687 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3688 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
3690 int vfs_badlock_print = 1; /* Print lock violations. */
3691 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
3694 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3695 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
3699 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3703 if (vfs_badlock_backtrace)
3706 if (vfs_badlock_print)
3707 printf("%s: %p %s\n", str, (void *)vp, msg);
3708 if (vfs_badlock_ddb)
3709 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3713 assert_vi_locked(struct vnode *vp, const char *str)
3716 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3717 vfs_badlock("interlock is not locked but should be", str, vp);
3721 assert_vi_unlocked(struct vnode *vp, const char *str)
3724 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3725 vfs_badlock("interlock is locked but should not be", str, vp);
3729 assert_vop_locked(struct vnode *vp, const char *str)
3732 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3733 vfs_badlock("is not locked but should be", str, vp);
3737 assert_vop_unlocked(struct vnode *vp, const char *str)
3740 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3741 vfs_badlock("is locked but should not be", str, vp);
3745 assert_vop_elocked(struct vnode *vp, const char *str)
3748 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3749 vfs_badlock("is not exclusive locked but should be", str, vp);
3754 assert_vop_elocked_other(struct vnode *vp, const char *str)
3757 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3758 vfs_badlock("is not exclusive locked by another thread",
3763 assert_vop_slocked(struct vnode *vp, const char *str)
3766 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3767 vfs_badlock("is not locked shared but should be", str, vp);
3770 #endif /* DEBUG_VFS_LOCKS */
3773 vop_rename_fail(struct vop_rename_args *ap)
3776 if (ap->a_tvp != NULL)
3778 if (ap->a_tdvp == ap->a_tvp)
3787 vop_rename_pre(void *ap)
3789 struct vop_rename_args *a = ap;
3791 #ifdef DEBUG_VFS_LOCKS
3793 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3794 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3795 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3796 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3798 /* Check the source (from). */
3799 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
3800 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
3801 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3802 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
3803 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3805 /* Check the target. */
3807 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3808 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3810 if (a->a_tdvp != a->a_fdvp)
3812 if (a->a_tvp != a->a_fvp)
3820 vop_strategy_pre(void *ap)
3822 #ifdef DEBUG_VFS_LOCKS
3823 struct vop_strategy_args *a;
3830 * Cluster ops lock their component buffers but not the IO container.
3832 if ((bp->b_flags & B_CLUSTER) != 0)
3835 if (!BUF_ISLOCKED(bp)) {
3836 if (vfs_badlock_print)
3838 "VOP_STRATEGY: bp is not locked but should be\n");
3839 if (vfs_badlock_ddb)
3840 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3846 vop_lookup_pre(void *ap)
3848 #ifdef DEBUG_VFS_LOCKS
3849 struct vop_lookup_args *a;
3854 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3855 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3860 vop_lookup_post(void *ap, int rc)
3862 #ifdef DEBUG_VFS_LOCKS
3863 struct vop_lookup_args *a;
3871 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3872 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3875 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3880 vop_lock_pre(void *ap)
3882 #ifdef DEBUG_VFS_LOCKS
3883 struct vop_lock1_args *a = ap;
3885 if ((a->a_flags & LK_INTERLOCK) == 0)
3886 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3888 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3893 vop_lock_post(void *ap, int rc)
3895 #ifdef DEBUG_VFS_LOCKS
3896 struct vop_lock1_args *a = ap;
3898 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3900 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3905 vop_unlock_pre(void *ap)
3907 #ifdef DEBUG_VFS_LOCKS
3908 struct vop_unlock_args *a = ap;
3910 if (a->a_flags & LK_INTERLOCK)
3911 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3912 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3917 vop_unlock_post(void *ap, int rc)
3919 #ifdef DEBUG_VFS_LOCKS
3920 struct vop_unlock_args *a = ap;
3922 if (a->a_flags & LK_INTERLOCK)
3923 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
3928 vop_create_post(void *ap, int rc)
3930 struct vop_create_args *a = ap;
3933 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3937 vop_link_post(void *ap, int rc)
3939 struct vop_link_args *a = ap;
3942 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
3943 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
3948 vop_mkdir_post(void *ap, int rc)
3950 struct vop_mkdir_args *a = ap;
3953 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
3957 vop_mknod_post(void *ap, int rc)
3959 struct vop_mknod_args *a = ap;
3962 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3966 vop_remove_post(void *ap, int rc)
3968 struct vop_remove_args *a = ap;
3971 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
3972 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
3977 vop_rename_post(void *ap, int rc)
3979 struct vop_rename_args *a = ap;
3982 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
3983 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
3984 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
3986 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
3988 if (a->a_tdvp != a->a_fdvp)
3990 if (a->a_tvp != a->a_fvp)
3998 vop_rmdir_post(void *ap, int rc)
4000 struct vop_rmdir_args *a = ap;
4003 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4004 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4009 vop_setattr_post(void *ap, int rc)
4011 struct vop_setattr_args *a = ap;
4014 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4018 vop_symlink_post(void *ap, int rc)
4020 struct vop_symlink_args *a = ap;
4023 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4026 static struct knlist fs_knlist;
4029 vfs_event_init(void *arg)
4031 knlist_init_mtx(&fs_knlist, NULL);
4033 /* XXX - correct order? */
4034 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4037 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
4040 KNOTE_UNLOCKED(&fs_knlist, event);
4043 static int filt_fsattach(struct knote *kn);
4044 static void filt_fsdetach(struct knote *kn);
4045 static int filt_fsevent(struct knote *kn, long hint);
4047 struct filterops fs_filtops =
4048 { 0, filt_fsattach, filt_fsdetach, filt_fsevent };
4051 filt_fsattach(struct knote *kn)
4054 kn->kn_flags |= EV_CLEAR;
4055 knlist_add(&fs_knlist, kn, 0);
4060 filt_fsdetach(struct knote *kn)
4063 knlist_remove(&fs_knlist, kn, 0);
4067 filt_fsevent(struct knote *kn, long hint)
4070 kn->kn_fflags |= hint;
4071 return (kn->kn_fflags != 0);
4075 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4081 error = SYSCTL_IN(req, &vc, sizeof(vc));
4084 if (vc.vc_vers != VFS_CTL_VERS1)
4086 mp = vfs_getvfs(&vc.vc_fsid);
4089 /* ensure that a specific sysctl goes to the right filesystem. */
4090 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4091 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4095 VCTLTOREQ(&vc, req);
4096 error = VFS_SYSCTL(mp, vc.vc_op, req);
4101 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
4105 * Function to initialize a va_filerev field sensibly.
4106 * XXX: Wouldn't a random number make a lot more sense ??
4109 init_va_filerev(void)
4114 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4117 static int filt_vfsread(struct knote *kn, long hint);
4118 static int filt_vfswrite(struct knote *kn, long hint);
4119 static int filt_vfsvnode(struct knote *kn, long hint);
4120 static void filt_vfsdetach(struct knote *kn);
4121 static struct filterops vfsread_filtops =
4122 { 1, NULL, filt_vfsdetach, filt_vfsread };
4123 static struct filterops vfswrite_filtops =
4124 { 1, NULL, filt_vfsdetach, filt_vfswrite };
4125 static struct filterops vfsvnode_filtops =
4126 { 1, NULL, filt_vfsdetach, filt_vfsvnode };
4129 vfs_knllock(void *arg)
4131 struct vnode *vp = arg;
4133 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4137 vfs_knlunlock(void *arg)
4139 struct vnode *vp = arg;
4145 vfs_knl_assert_locked(void *arg)
4147 #ifdef DEBUG_VFS_LOCKS
4148 struct vnode *vp = arg;
4150 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4155 vfs_knl_assert_unlocked(void *arg)
4157 #ifdef DEBUG_VFS_LOCKS
4158 struct vnode *vp = arg;
4160 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4165 vfs_kqfilter(struct vop_kqfilter_args *ap)
4167 struct vnode *vp = ap->a_vp;
4168 struct knote *kn = ap->a_kn;
4171 switch (kn->kn_filter) {
4173 kn->kn_fop = &vfsread_filtops;
4176 kn->kn_fop = &vfswrite_filtops;
4179 kn->kn_fop = &vfsvnode_filtops;
4185 kn->kn_hook = (caddr_t)vp;
4188 if (vp->v_pollinfo == NULL)
4190 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4191 knlist_add(knl, kn, 0);
4197 * Detach knote from vnode
4200 filt_vfsdetach(struct knote *kn)
4202 struct vnode *vp = (struct vnode *)kn->kn_hook;
4204 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4205 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4210 filt_vfsread(struct knote *kn, long hint)
4212 struct vnode *vp = (struct vnode *)kn->kn_hook;
4217 * filesystem is gone, so set the EOF flag and schedule
4218 * the knote for deletion.
4220 if (hint == NOTE_REVOKE) {
4222 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4227 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4231 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4232 res = (kn->kn_data != 0);
4239 filt_vfswrite(struct knote *kn, long hint)
4241 struct vnode *vp = (struct vnode *)kn->kn_hook;
4246 * filesystem is gone, so set the EOF flag and schedule
4247 * the knote for deletion.
4249 if (hint == NOTE_REVOKE)
4250 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4258 filt_vfsvnode(struct knote *kn, long hint)
4260 struct vnode *vp = (struct vnode *)kn->kn_hook;
4264 if (kn->kn_sfflags & hint)
4265 kn->kn_fflags |= hint;
4266 if (hint == NOTE_REVOKE) {
4267 kn->kn_flags |= EV_EOF;
4271 res = (kn->kn_fflags != 0);
4277 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4281 if (dp->d_reclen > ap->a_uio->uio_resid)
4282 return (ENAMETOOLONG);
4283 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4285 if (ap->a_ncookies != NULL) {
4286 if (ap->a_cookies != NULL)
4287 free(ap->a_cookies, M_TEMP);
4288 ap->a_cookies = NULL;
4289 *ap->a_ncookies = 0;
4293 if (ap->a_ncookies == NULL)
4296 KASSERT(ap->a_cookies,
4297 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4299 *ap->a_cookies = realloc(*ap->a_cookies,
4300 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4301 (*ap->a_cookies)[*ap->a_ncookies] = off;
4306 * Mark for update the access time of the file if the filesystem
4307 * supports VOP_MARKATIME. This functionality is used by execve and
4308 * mmap, so we want to avoid the I/O implied by directly setting
4309 * va_atime for the sake of efficiency.
4312 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4317 VFS_ASSERT_GIANT(mp);
4318 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4319 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4320 (void)VOP_MARKATIME(vp);
4324 * The purpose of this routine is to remove granularity from accmode_t,
4325 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4326 * VADMIN and VAPPEND.
4328 * If it returns 0, the caller is supposed to continue with the usual
4329 * access checks using 'accmode' as modified by this routine. If it
4330 * returns nonzero value, the caller is supposed to return that value
4333 * Note that after this routine runs, accmode may be zero.
4336 vfs_unixify_accmode(accmode_t *accmode)
4339 * There is no way to specify explicit "deny" rule using
4340 * file mode or POSIX.1e ACLs.
4342 if (*accmode & VEXPLICIT_DENY) {
4348 * None of these can be translated into usual access bits.
4349 * Also, the common case for NFSv4 ACLs is to not contain
4350 * either of these bits. Caller should check for VWRITE
4351 * on the containing directory instead.
4353 if (*accmode & (VDELETE_CHILD | VDELETE))
4356 if (*accmode & VADMIN_PERMS) {
4357 *accmode &= ~VADMIN_PERMS;
4362 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4363 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4365 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);