2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
38 * External virtual filesystem routines
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
45 #include "opt_watchdog.h"
47 #include <sys/param.h>
48 #include <sys/systm.h>
51 #include <sys/condvar.h>
53 #include <sys/dirent.h>
54 #include <sys/event.h>
55 #include <sys/eventhandler.h>
56 #include <sys/extattr.h>
58 #include <sys/fcntl.h>
61 #include <sys/kernel.h>
62 #include <sys/kthread.h>
63 #include <sys/lockf.h>
64 #include <sys/malloc.h>
65 #include <sys/mount.h>
66 #include <sys/namei.h>
68 #include <sys/reboot.h>
69 #include <sys/sched.h>
70 #include <sys/sleepqueue.h>
72 #include <sys/sysctl.h>
73 #include <sys/syslog.h>
74 #include <sys/vmmeter.h>
75 #include <sys/vnode.h>
77 #include <sys/watchdog.h>
80 #include <machine/stdarg.h>
82 #include <security/mac/mac_framework.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_extern.h>
88 #include <vm/vm_map.h>
89 #include <vm/vm_page.h>
90 #include <vm/vm_kern.h>
100 static void delmntque(struct vnode *vp);
101 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
102 int slpflag, int slptimeo);
103 static void syncer_shutdown(void *arg, int howto);
104 static int vtryrecycle(struct vnode *vp);
105 static void vbusy(struct vnode *vp);
106 static void vinactive(struct vnode *, struct thread *);
107 static void v_incr_usecount(struct vnode *);
108 static void v_decr_usecount(struct vnode *);
109 static void v_decr_useonly(struct vnode *);
110 static void v_upgrade_usecount(struct vnode *);
111 static void vfree(struct vnode *);
112 static void vnlru_free(int);
113 static void vgonel(struct vnode *);
114 static void vfs_knllock(void *arg);
115 static void vfs_knlunlock(void *arg);
116 static void vfs_knl_assert_locked(void *arg);
117 static void vfs_knl_assert_unlocked(void *arg);
118 static void destroy_vpollinfo(struct vpollinfo *vi);
121 * Number of vnodes in existence. Increased whenever getnewvnode()
122 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
125 static unsigned long numvnodes;
127 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
128 "Number of vnodes in existence");
131 * Conversion tables for conversion from vnode types to inode formats
134 enum vtype iftovt_tab[16] = {
135 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
136 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
138 int vttoif_tab[10] = {
139 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
140 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
144 * List of vnodes that are ready for recycling.
146 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
149 * Free vnode target. Free vnodes may simply be files which have been stat'd
150 * but not read. This is somewhat common, and a small cache of such files
151 * should be kept to avoid recreation costs.
153 static u_long wantfreevnodes;
154 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
155 /* Number of vnodes in the free list. */
156 static u_long freevnodes;
157 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
158 "Number of vnodes in the free list");
160 static int vlru_allow_cache_src;
161 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
162 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
165 * Various variables used for debugging the new implementation of
167 * XXX these are probably of (very) limited utility now.
169 static int reassignbufcalls;
170 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
171 "Number of calls to reassignbuf");
174 * Cache for the mount type id assigned to NFS. This is used for
175 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
177 int nfs_mount_type = -1;
179 /* To keep more than one thread at a time from running vfs_getnewfsid */
180 static struct mtx mntid_mtx;
183 * Lock for any access to the following:
188 static struct mtx vnode_free_list_mtx;
190 /* Publicly exported FS */
191 struct nfs_public nfs_pub;
193 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
194 static uma_zone_t vnode_zone;
195 static uma_zone_t vnodepoll_zone;
197 /* Set to 1 to print out reclaim of active vnodes */
201 * The workitem queue.
203 * It is useful to delay writes of file data and filesystem metadata
204 * for tens of seconds so that quickly created and deleted files need
205 * not waste disk bandwidth being created and removed. To realize this,
206 * we append vnodes to a "workitem" queue. When running with a soft
207 * updates implementation, most pending metadata dependencies should
208 * not wait for more than a few seconds. Thus, mounted on block devices
209 * are delayed only about a half the time that file data is delayed.
210 * Similarly, directory updates are more critical, so are only delayed
211 * about a third the time that file data is delayed. Thus, there are
212 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
213 * one each second (driven off the filesystem syncer process). The
214 * syncer_delayno variable indicates the next queue that is to be processed.
215 * Items that need to be processed soon are placed in this queue:
217 * syncer_workitem_pending[syncer_delayno]
219 * A delay of fifteen seconds is done by placing the request fifteen
220 * entries later in the queue:
222 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
225 static int syncer_delayno;
226 static long syncer_mask;
227 LIST_HEAD(synclist, bufobj);
228 static struct synclist *syncer_workitem_pending[2];
230 * The sync_mtx protects:
235 * syncer_workitem_pending
236 * syncer_worklist_len
239 static struct mtx sync_mtx;
240 static struct cv sync_wakeup;
242 #define SYNCER_MAXDELAY 32
243 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
244 static int syncdelay = 30; /* max time to delay syncing data */
245 static int filedelay = 30; /* time to delay syncing files */
246 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
247 "Time to delay syncing files (in seconds)");
248 static int dirdelay = 29; /* time to delay syncing directories */
249 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
250 "Time to delay syncing directories (in seconds)");
251 static int metadelay = 28; /* time to delay syncing metadata */
252 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
253 "Time to delay syncing metadata (in seconds)");
254 static int rushjob; /* number of slots to run ASAP */
255 static int stat_rush_requests; /* number of times I/O speeded up */
256 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
257 "Number of times I/O speeded up (rush requests)");
260 * When shutting down the syncer, run it at four times normal speed.
262 #define SYNCER_SHUTDOWN_SPEEDUP 4
263 static int sync_vnode_count;
264 static int syncer_worklist_len;
265 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
269 * Number of vnodes we want to exist at any one time. This is mostly used
270 * to size hash tables in vnode-related code. It is normally not used in
271 * getnewvnode(), as wantfreevnodes is normally nonzero.)
273 * XXX desiredvnodes is historical cruft and should not exist.
276 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
277 &desiredvnodes, 0, "Maximum number of vnodes");
278 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
279 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
280 static int vnlru_nowhere;
281 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
282 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
285 * Macros to control when a vnode is freed and recycled. All require
286 * the vnode interlock.
288 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
289 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
290 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
294 * Initialize the vnode management data structures.
296 * Reevaluate the following cap on the number of vnodes after the physical
297 * memory size exceeds 512GB. In the limit, as the physical memory size
298 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
300 #ifndef MAXVNODES_MAX
301 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
304 vntblinit(void *dummy __unused)
306 int physvnodes, virtvnodes;
309 * Desiredvnodes is a function of the physical memory size and the
310 * kernel's heap size. Generally speaking, it scales with the
311 * physical memory size. The ratio of desiredvnodes to physical pages
312 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
313 * marginal ratio of desiredvnodes to physical pages is one to
314 * sixteen. However, desiredvnodes is limited by the kernel's heap
315 * size. The memory required by desiredvnodes vnodes and vm objects
316 * may not exceed one seventh of the kernel's heap size.
318 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
319 cnt.v_page_count) / 16;
320 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
321 sizeof(struct vnode)));
322 desiredvnodes = min(physvnodes, virtvnodes);
323 if (desiredvnodes > MAXVNODES_MAX) {
325 printf("Reducing kern.maxvnodes %d -> %d\n",
326 desiredvnodes, MAXVNODES_MAX);
327 desiredvnodes = MAXVNODES_MAX;
329 wantfreevnodes = desiredvnodes / 4;
330 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
331 TAILQ_INIT(&vnode_free_list);
332 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
333 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
334 NULL, NULL, UMA_ALIGN_PTR, 0);
335 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
336 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
338 * Initialize the filesystem syncer.
340 syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
342 syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
344 syncer_maxdelay = syncer_mask + 1;
345 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
346 cv_init(&sync_wakeup, "syncer");
348 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
352 * Mark a mount point as busy. Used to synchronize access and to delay
353 * unmounting. Eventually, mountlist_mtx is not released on failure.
355 * vfs_busy() is a custom lock, it can block the caller.
356 * vfs_busy() only sleeps if the unmount is active on the mount point.
357 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
358 * vnode belonging to mp.
360 * Lookup uses vfs_busy() to traverse mount points.
362 * / vnode lock A / vnode lock (/var) D
363 * /var vnode lock B /log vnode lock(/var/log) E
364 * vfs_busy lock C vfs_busy lock F
366 * Within each file system, the lock order is C->A->B and F->D->E.
368 * When traversing across mounts, the system follows that lock order:
374 * The lookup() process for namei("/var") illustrates the process:
375 * VOP_LOOKUP() obtains B while A is held
376 * vfs_busy() obtains a shared lock on F while A and B are held
377 * vput() releases lock on B
378 * vput() releases lock on A
379 * VFS_ROOT() obtains lock on D while shared lock on F is held
380 * vfs_unbusy() releases shared lock on F
381 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
382 * Attempt to lock A (instead of vp_crossmp) while D is held would
383 * violate the global order, causing deadlocks.
385 * dounmount() locks B while F is drained.
388 vfs_busy(struct mount *mp, int flags)
391 MPASS((flags & ~MBF_MASK) == 0);
392 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
397 * If mount point is currenly being unmounted, sleep until the
398 * mount point fate is decided. If thread doing the unmounting fails,
399 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
400 * that this mount point has survived the unmount attempt and vfs_busy
401 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
402 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
403 * about to be really destroyed. vfs_busy needs to release its
404 * reference on the mount point in this case and return with ENOENT,
405 * telling the caller that mount mount it tried to busy is no longer
408 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
409 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
412 CTR1(KTR_VFS, "%s: failed busying before sleeping",
416 if (flags & MBF_MNTLSTLOCK)
417 mtx_unlock(&mountlist_mtx);
418 mp->mnt_kern_flag |= MNTK_MWAIT;
419 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
420 if (flags & MBF_MNTLSTLOCK)
421 mtx_lock(&mountlist_mtx);
424 if (flags & MBF_MNTLSTLOCK)
425 mtx_unlock(&mountlist_mtx);
432 * Free a busy filesystem.
435 vfs_unbusy(struct mount *mp)
438 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
441 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
443 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
444 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
445 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
446 mp->mnt_kern_flag &= ~MNTK_DRAINING;
447 wakeup(&mp->mnt_lockref);
453 * Lookup a mount point by filesystem identifier.
456 vfs_getvfs(fsid_t *fsid)
460 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
461 mtx_lock(&mountlist_mtx);
462 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
463 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
464 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
466 mtx_unlock(&mountlist_mtx);
470 mtx_unlock(&mountlist_mtx);
471 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
472 return ((struct mount *) 0);
476 * Lookup a mount point by filesystem identifier, busying it before
480 vfs_busyfs(fsid_t *fsid)
485 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
486 mtx_lock(&mountlist_mtx);
487 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
488 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
489 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
490 error = vfs_busy(mp, MBF_MNTLSTLOCK);
492 mtx_unlock(&mountlist_mtx);
498 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
499 mtx_unlock(&mountlist_mtx);
500 return ((struct mount *) 0);
504 * Check if a user can access privileged mount options.
507 vfs_suser(struct mount *mp, struct thread *td)
512 * If the thread is jailed, but this is not a jail-friendly file
513 * system, deny immediately.
515 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
519 * If the file system was mounted outside the jail of the calling
520 * thread, deny immediately.
522 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
526 * If file system supports delegated administration, we don't check
527 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
528 * by the file system itself.
529 * If this is not the user that did original mount, we check for
530 * the PRIV_VFS_MOUNT_OWNER privilege.
532 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
533 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
534 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
541 * Get a new unique fsid. Try to make its val[0] unique, since this value
542 * will be used to create fake device numbers for stat(). Also try (but
543 * not so hard) make its val[0] unique mod 2^16, since some emulators only
544 * support 16-bit device numbers. We end up with unique val[0]'s for the
545 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
547 * Keep in mind that several mounts may be running in parallel. Starting
548 * the search one past where the previous search terminated is both a
549 * micro-optimization and a defense against returning the same fsid to
553 vfs_getnewfsid(struct mount *mp)
555 static u_int16_t mntid_base;
560 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
561 mtx_lock(&mntid_mtx);
562 mtype = mp->mnt_vfc->vfc_typenum;
563 tfsid.val[1] = mtype;
564 mtype = (mtype & 0xFF) << 24;
566 tfsid.val[0] = makedev(255,
567 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
569 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
573 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
574 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
575 mtx_unlock(&mntid_mtx);
579 * Knob to control the precision of file timestamps:
581 * 0 = seconds only; nanoseconds zeroed.
582 * 1 = seconds and nanoseconds, accurate within 1/HZ.
583 * 2 = seconds and nanoseconds, truncated to microseconds.
584 * >=3 = seconds and nanoseconds, maximum precision.
586 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
588 static int timestamp_precision = TSP_SEC;
589 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
590 ×tamp_precision, 0, "File timestamp precision (0: seconds, "
591 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
592 "3+: sec + ns (max. precision))");
595 * Get a current timestamp.
598 vfs_timestamp(struct timespec *tsp)
602 switch (timestamp_precision) {
604 tsp->tv_sec = time_second;
612 TIMEVAL_TO_TIMESPEC(&tv, tsp);
622 * Set vnode attributes to VNOVAL
625 vattr_null(struct vattr *vap)
629 vap->va_size = VNOVAL;
630 vap->va_bytes = VNOVAL;
631 vap->va_mode = VNOVAL;
632 vap->va_nlink = VNOVAL;
633 vap->va_uid = VNOVAL;
634 vap->va_gid = VNOVAL;
635 vap->va_fsid = VNOVAL;
636 vap->va_fileid = VNOVAL;
637 vap->va_blocksize = VNOVAL;
638 vap->va_rdev = VNOVAL;
639 vap->va_atime.tv_sec = VNOVAL;
640 vap->va_atime.tv_nsec = VNOVAL;
641 vap->va_mtime.tv_sec = VNOVAL;
642 vap->va_mtime.tv_nsec = VNOVAL;
643 vap->va_ctime.tv_sec = VNOVAL;
644 vap->va_ctime.tv_nsec = VNOVAL;
645 vap->va_birthtime.tv_sec = VNOVAL;
646 vap->va_birthtime.tv_nsec = VNOVAL;
647 vap->va_flags = VNOVAL;
648 vap->va_gen = VNOVAL;
653 * This routine is called when we have too many vnodes. It attempts
654 * to free <count> vnodes and will potentially free vnodes that still
655 * have VM backing store (VM backing store is typically the cause
656 * of a vnode blowout so we want to do this). Therefore, this operation
657 * is not considered cheap.
659 * A number of conditions may prevent a vnode from being reclaimed.
660 * the buffer cache may have references on the vnode, a directory
661 * vnode may still have references due to the namei cache representing
662 * underlying files, or the vnode may be in active use. It is not
663 * desireable to reuse such vnodes. These conditions may cause the
664 * number of vnodes to reach some minimum value regardless of what
665 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
668 vlrureclaim(struct mount *mp)
677 * Calculate the trigger point, don't allow user
678 * screwups to blow us up. This prevents us from
679 * recycling vnodes with lots of resident pages. We
680 * aren't trying to free memory, we are trying to
683 usevnodes = desiredvnodes;
686 trigger = cnt.v_page_count * 2 / usevnodes;
688 vn_start_write(NULL, &mp, V_WAIT);
690 count = mp->mnt_nvnodelistsize / 10 + 1;
692 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
693 while (vp != NULL && vp->v_type == VMARKER)
694 vp = TAILQ_NEXT(vp, v_nmntvnodes);
697 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
698 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
703 * If it's been deconstructed already, it's still
704 * referenced, or it exceeds the trigger, skip it.
706 if (vp->v_usecount ||
707 (!vlru_allow_cache_src &&
708 !LIST_EMPTY(&(vp)->v_cache_src)) ||
709 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
710 vp->v_object->resident_page_count > trigger)) {
716 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
718 goto next_iter_mntunlocked;
722 * v_usecount may have been bumped after VOP_LOCK() dropped
723 * the vnode interlock and before it was locked again.
725 * It is not necessary to recheck VI_DOOMED because it can
726 * only be set by another thread that holds both the vnode
727 * lock and vnode interlock. If another thread has the
728 * vnode lock before we get to VOP_LOCK() and obtains the
729 * vnode interlock after VOP_LOCK() drops the vnode
730 * interlock, the other thread will be unable to drop the
731 * vnode lock before our VOP_LOCK() call fails.
733 if (vp->v_usecount ||
734 (!vlru_allow_cache_src &&
735 !LIST_EMPTY(&(vp)->v_cache_src)) ||
736 (vp->v_object != NULL &&
737 vp->v_object->resident_page_count > trigger)) {
738 VOP_UNLOCK(vp, LK_INTERLOCK);
739 goto next_iter_mntunlocked;
741 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
742 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
747 next_iter_mntunlocked:
748 if ((count % 256) != 0)
752 if ((count % 256) != 0)
761 vn_finished_write(mp);
766 * Attempt to keep the free list at wantfreevnodes length.
769 vnlru_free(int count)
774 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
775 for (; count > 0; count--) {
776 vp = TAILQ_FIRST(&vnode_free_list);
778 * The list can be modified while the free_list_mtx
779 * has been dropped and vp could be NULL here.
783 VNASSERT(vp->v_op != NULL, vp,
784 ("vnlru_free: vnode already reclaimed."));
785 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
787 * Don't recycle if we can't get the interlock.
789 if (!VI_TRYLOCK(vp)) {
790 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
793 VNASSERT(VCANRECYCLE(vp), vp,
794 ("vp inconsistent on freelist"));
796 vp->v_iflag &= ~VI_FREE;
798 mtx_unlock(&vnode_free_list_mtx);
800 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
802 VFS_UNLOCK_GIANT(vfslocked);
804 * If the recycled succeeded this vdrop will actually free
805 * the vnode. If not it will simply place it back on
809 mtx_lock(&vnode_free_list_mtx);
813 * Attempt to recycle vnodes in a context that is always safe to block.
814 * Calling vlrurecycle() from the bowels of filesystem code has some
815 * interesting deadlock problems.
817 static struct proc *vnlruproc;
818 static int vnlruproc_sig;
823 struct mount *mp, *nmp;
825 struct proc *p = vnlruproc;
827 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
831 kproc_suspend_check(p);
832 mtx_lock(&vnode_free_list_mtx);
833 if (freevnodes > wantfreevnodes)
834 vnlru_free(freevnodes - wantfreevnodes);
835 if (numvnodes <= desiredvnodes * 9 / 10) {
837 wakeup(&vnlruproc_sig);
838 msleep(vnlruproc, &vnode_free_list_mtx,
839 PVFS|PDROP, "vlruwt", hz);
842 mtx_unlock(&vnode_free_list_mtx);
844 mtx_lock(&mountlist_mtx);
845 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
846 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
847 nmp = TAILQ_NEXT(mp, mnt_list);
850 vfslocked = VFS_LOCK_GIANT(mp);
851 done += vlrureclaim(mp);
852 VFS_UNLOCK_GIANT(vfslocked);
853 mtx_lock(&mountlist_mtx);
854 nmp = TAILQ_NEXT(mp, mnt_list);
857 mtx_unlock(&mountlist_mtx);
860 /* These messages are temporary debugging aids */
861 if (vnlru_nowhere < 5)
862 printf("vnlru process getting nowhere..\n");
863 else if (vnlru_nowhere == 5)
864 printf("vnlru process messages stopped.\n");
867 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
873 static struct kproc_desc vnlru_kp = {
878 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
882 * Routines having to do with the management of the vnode table.
886 vdestroy(struct vnode *vp)
890 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
891 mtx_lock(&vnode_free_list_mtx);
893 mtx_unlock(&vnode_free_list_mtx);
895 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
896 ("cleaned vnode still on the free list."));
897 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
898 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
899 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
900 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
901 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
902 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
903 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
904 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
905 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
906 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
907 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
908 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
911 mac_vnode_destroy(vp);
913 if (vp->v_pollinfo != NULL)
914 destroy_vpollinfo(vp->v_pollinfo);
916 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
919 lockdestroy(vp->v_vnlock);
920 mtx_destroy(&vp->v_interlock);
921 mtx_destroy(BO_MTX(bo));
922 uma_zfree(vnode_zone, vp);
926 * Try to recycle a freed vnode. We abort if anyone picks up a reference
927 * before we actually vgone(). This function must be called with the vnode
928 * held to prevent the vnode from being returned to the free list midway
932 vtryrecycle(struct vnode *vp)
936 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
937 VNASSERT(vp->v_holdcnt, vp,
938 ("vtryrecycle: Recycling vp %p without a reference.", vp));
940 * This vnode may found and locked via some other list, if so we
941 * can't recycle it yet.
943 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
945 "%s: impossible to recycle, vp %p lock is already held",
947 return (EWOULDBLOCK);
950 * Don't recycle if its filesystem is being suspended.
952 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
955 "%s: impossible to recycle, cannot start the write for %p",
960 * If we got this far, we need to acquire the interlock and see if
961 * anyone picked up this vnode from another list. If not, we will
962 * mark it with DOOMED via vgonel() so that anyone who does find it
966 if (vp->v_usecount) {
967 VOP_UNLOCK(vp, LK_INTERLOCK);
968 vn_finished_write(vnmp);
970 "%s: impossible to recycle, %p is already referenced",
974 if ((vp->v_iflag & VI_DOOMED) == 0)
976 VOP_UNLOCK(vp, LK_INTERLOCK);
977 vn_finished_write(vnmp);
982 * Return the next vnode from the free list.
985 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
988 struct vnode *vp = NULL;
991 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
992 mtx_lock(&vnode_free_list_mtx);
994 * Lend our context to reclaim vnodes if they've exceeded the max.
996 if (freevnodes > wantfreevnodes)
999 * Wait for available vnodes.
1001 if (numvnodes > desiredvnodes) {
1002 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
1004 * File system is beeing suspended, we cannot risk a
1005 * deadlock here, so allocate new vnode anyway.
1007 if (freevnodes > wantfreevnodes)
1008 vnlru_free(freevnodes - wantfreevnodes);
1011 if (vnlruproc_sig == 0) {
1012 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
1015 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
1017 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1018 if (numvnodes > desiredvnodes) {
1019 mtx_unlock(&vnode_free_list_mtx);
1026 mtx_unlock(&vnode_free_list_mtx);
1027 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1031 vp->v_vnlock = &vp->v_lock;
1032 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1034 * By default, don't allow shared locks unless filesystems
1037 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
1039 * Initialize bufobj.
1042 bo->__bo_vnode = vp;
1043 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
1044 bo->bo_ops = &buf_ops_bio;
1045 bo->bo_private = vp;
1046 TAILQ_INIT(&bo->bo_clean.bv_hd);
1047 TAILQ_INIT(&bo->bo_dirty.bv_hd);
1049 * Initialize namecache.
1051 LIST_INIT(&vp->v_cache_src);
1052 TAILQ_INIT(&vp->v_cache_dst);
1054 * Finalize various vnode identity bits.
1059 v_incr_usecount(vp);
1063 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1064 mac_vnode_associate_singlelabel(mp, vp);
1065 else if (mp == NULL && vops != &dead_vnodeops)
1066 printf("NULL mp in getnewvnode()\n");
1069 bo->bo_bsize = mp->mnt_stat.f_iosize;
1070 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1071 vp->v_vflag |= VV_NOKNOTE;
1079 * Delete from old mount point vnode list, if on one.
1082 delmntque(struct vnode *vp)
1091 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1092 ("bad mount point vnode list size"));
1093 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1094 mp->mnt_nvnodelistsize--;
1100 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1104 vp->v_op = &dead_vnodeops;
1105 /* XXX non mp-safe fs may still call insmntque with vnode
1107 if (!VOP_ISLOCKED(vp))
1108 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1114 * Insert into list of vnodes for the new mount point, if available.
1117 insmntque1(struct vnode *vp, struct mount *mp,
1118 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1122 KASSERT(vp->v_mount == NULL,
1123 ("insmntque: vnode already on per mount vnode list"));
1124 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1125 #ifdef DEBUG_VFS_LOCKS
1126 if (!VFS_NEEDSGIANT(mp))
1127 ASSERT_VOP_ELOCKED(vp,
1128 "insmntque: mp-safe fs and non-locked vp");
1131 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1132 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1133 mp->mnt_nvnodelistsize == 0)) {
1134 locked = VOP_ISLOCKED(vp);
1135 if (!locked || (locked == LK_EXCLUSIVE &&
1136 (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1145 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1146 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1147 ("neg mount point vnode list size"));
1148 mp->mnt_nvnodelistsize++;
1154 insmntque(struct vnode *vp, struct mount *mp)
1157 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1161 * Flush out and invalidate all buffers associated with a bufobj
1162 * Called with the underlying object locked.
1165 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1170 if (flags & V_SAVE) {
1171 error = bufobj_wwait(bo, slpflag, slptimeo);
1176 if (bo->bo_dirty.bv_cnt > 0) {
1178 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1181 * XXX We could save a lock/unlock if this was only
1182 * enabled under INVARIANTS
1185 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1186 panic("vinvalbuf: dirty bufs");
1190 * If you alter this loop please notice that interlock is dropped and
1191 * reacquired in flushbuflist. Special care is needed to ensure that
1192 * no race conditions occur from this.
1195 error = flushbuflist(&bo->bo_clean,
1196 flags, bo, slpflag, slptimeo);
1197 if (error == 0 && !(flags & V_CLEANONLY))
1198 error = flushbuflist(&bo->bo_dirty,
1199 flags, bo, slpflag, slptimeo);
1200 if (error != 0 && error != EAGAIN) {
1204 } while (error != 0);
1207 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1208 * have write I/O in-progress but if there is a VM object then the
1209 * VM object can also have read-I/O in-progress.
1212 bufobj_wwait(bo, 0, 0);
1214 if (bo->bo_object != NULL) {
1215 VM_OBJECT_LOCK(bo->bo_object);
1216 vm_object_pip_wait(bo->bo_object, "bovlbx");
1217 VM_OBJECT_UNLOCK(bo->bo_object);
1220 } while (bo->bo_numoutput > 0);
1224 * Destroy the copy in the VM cache, too.
1226 if (bo->bo_object != NULL &&
1227 (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1228 VM_OBJECT_LOCK(bo->bo_object);
1229 vm_object_page_remove(bo->bo_object, 0, 0,
1230 (flags & V_SAVE) ? TRUE : FALSE);
1231 VM_OBJECT_UNLOCK(bo->bo_object);
1236 if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1237 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1238 panic("vinvalbuf: flush failed");
1245 * Flush out and invalidate all buffers associated with a vnode.
1246 * Called with the underlying object locked.
1249 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1252 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1253 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1254 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1258 * Flush out buffers on the specified list.
1262 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1265 struct buf *bp, *nbp;
1270 ASSERT_BO_LOCKED(bo);
1273 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1274 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1275 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1281 lblkno = nbp->b_lblkno;
1282 xflags = nbp->b_xflags &
1283 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1286 error = BUF_TIMELOCK(bp,
1287 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1288 "flushbuf", slpflag, slptimeo);
1291 return (error != ENOLCK ? error : EAGAIN);
1293 KASSERT(bp->b_bufobj == bo,
1294 ("bp %p wrong b_bufobj %p should be %p",
1295 bp, bp->b_bufobj, bo));
1296 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1302 * XXX Since there are no node locks for NFS, I
1303 * believe there is a slight chance that a delayed
1304 * write will occur while sleeping just above, so
1307 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1312 bp->b_flags |= B_ASYNC;
1315 return (EAGAIN); /* XXX: why not loop ? */
1320 bp->b_flags |= (B_INVAL | B_RELBUF);
1321 bp->b_flags &= ~B_ASYNC;
1325 (nbp->b_bufobj != bo ||
1326 nbp->b_lblkno != lblkno ||
1328 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1329 break; /* nbp invalid */
1335 * Truncate a file's buffer and pages to a specified length. This
1336 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1340 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1341 off_t length, int blksize)
1343 struct buf *bp, *nbp;
1348 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1349 vp, cred, blksize, (uintmax_t)length);
1352 * Round up to the *next* lbn.
1354 trunclbn = (length + blksize - 1) / blksize;
1356 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1363 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1364 if (bp->b_lblkno < trunclbn)
1367 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1368 BO_MTX(bo)) == ENOLCK)
1374 bp->b_flags |= (B_INVAL | B_RELBUF);
1375 bp->b_flags &= ~B_ASYNC;
1381 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1382 (nbp->b_vp != vp) ||
1383 (nbp->b_flags & B_DELWRI))) {
1389 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1390 if (bp->b_lblkno < trunclbn)
1393 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1394 BO_MTX(bo)) == ENOLCK)
1399 bp->b_flags |= (B_INVAL | B_RELBUF);
1400 bp->b_flags &= ~B_ASYNC;
1406 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1407 (nbp->b_vp != vp) ||
1408 (nbp->b_flags & B_DELWRI) == 0)) {
1417 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1418 if (bp->b_lblkno > 0)
1421 * Since we hold the vnode lock this should only
1422 * fail if we're racing with the buf daemon.
1425 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1426 BO_MTX(bo)) == ENOLCK) {
1429 VNASSERT((bp->b_flags & B_DELWRI), vp,
1430 ("buf(%p) on dirty queue without DELWRI", bp));
1441 bufobj_wwait(bo, 0, 0);
1443 vnode_pager_setsize(vp, length);
1449 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1452 * NOTE: We have to deal with the special case of a background bitmap
1453 * buffer, a situation where two buffers will have the same logical
1454 * block offset. We want (1) only the foreground buffer to be accessed
1455 * in a lookup and (2) must differentiate between the foreground and
1456 * background buffer in the splay tree algorithm because the splay
1457 * tree cannot normally handle multiple entities with the same 'index'.
1458 * We accomplish this by adding differentiating flags to the splay tree's
1463 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1466 struct buf *lefttreemax, *righttreemin, *y;
1470 lefttreemax = righttreemin = &dummy;
1472 if (lblkno < root->b_lblkno ||
1473 (lblkno == root->b_lblkno &&
1474 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1475 if ((y = root->b_left) == NULL)
1477 if (lblkno < y->b_lblkno) {
1479 root->b_left = y->b_right;
1482 if ((y = root->b_left) == NULL)
1485 /* Link into the new root's right tree. */
1486 righttreemin->b_left = root;
1487 righttreemin = root;
1488 } else if (lblkno > root->b_lblkno ||
1489 (lblkno == root->b_lblkno &&
1490 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1491 if ((y = root->b_right) == NULL)
1493 if (lblkno > y->b_lblkno) {
1495 root->b_right = y->b_left;
1498 if ((y = root->b_right) == NULL)
1501 /* Link into the new root's left tree. */
1502 lefttreemax->b_right = root;
1509 /* Assemble the new root. */
1510 lefttreemax->b_right = root->b_left;
1511 righttreemin->b_left = root->b_right;
1512 root->b_left = dummy.b_right;
1513 root->b_right = dummy.b_left;
1518 buf_vlist_remove(struct buf *bp)
1523 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1524 ASSERT_BO_LOCKED(bp->b_bufobj);
1525 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1526 (BX_VNDIRTY|BX_VNCLEAN),
1527 ("buf_vlist_remove: Buf %p is on two lists", bp));
1528 if (bp->b_xflags & BX_VNDIRTY)
1529 bv = &bp->b_bufobj->bo_dirty;
1531 bv = &bp->b_bufobj->bo_clean;
1532 if (bp != bv->bv_root) {
1533 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1534 KASSERT(root == bp, ("splay lookup failed in remove"));
1536 if (bp->b_left == NULL) {
1539 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1540 root->b_right = bp->b_right;
1543 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1545 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1549 * Add the buffer to the sorted clean or dirty block list using a
1550 * splay tree algorithm.
1552 * NOTE: xflags is passed as a constant, optimizing this inline function!
1555 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1560 ASSERT_BO_LOCKED(bo);
1561 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1562 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1563 bp->b_xflags |= xflags;
1564 if (xflags & BX_VNDIRTY)
1569 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1573 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1574 } else if (bp->b_lblkno < root->b_lblkno ||
1575 (bp->b_lblkno == root->b_lblkno &&
1576 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1577 bp->b_left = root->b_left;
1579 root->b_left = NULL;
1580 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1582 bp->b_right = root->b_right;
1584 root->b_right = NULL;
1585 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1592 * Lookup a buffer using the splay tree. Note that we specifically avoid
1593 * shadow buffers used in background bitmap writes.
1595 * This code isn't quite efficient as it could be because we are maintaining
1596 * two sorted lists and do not know which list the block resides in.
1598 * During a "make buildworld" the desired buffer is found at one of
1599 * the roots more than 60% of the time. Thus, checking both roots
1600 * before performing either splay eliminates unnecessary splays on the
1601 * first tree splayed.
1604 gbincore(struct bufobj *bo, daddr_t lblkno)
1608 ASSERT_BO_LOCKED(bo);
1609 if ((bp = bo->bo_clean.bv_root) != NULL &&
1610 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1612 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1613 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1615 if ((bp = bo->bo_clean.bv_root) != NULL) {
1616 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1617 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1620 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1621 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1622 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1629 * Associate a buffer with a vnode.
1632 bgetvp(struct vnode *vp, struct buf *bp)
1637 ASSERT_BO_LOCKED(bo);
1638 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1640 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1641 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1642 ("bgetvp: bp already attached! %p", bp));
1645 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1646 bp->b_flags |= B_NEEDSGIANT;
1650 * Insert onto list for new vnode.
1652 buf_vlist_add(bp, bo, BX_VNCLEAN);
1656 * Disassociate a buffer from a vnode.
1659 brelvp(struct buf *bp)
1664 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1665 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1668 * Delete from old vnode list, if on one.
1670 vp = bp->b_vp; /* XXX */
1673 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1674 buf_vlist_remove(bp);
1676 panic("brelvp: Buffer %p not on queue.", bp);
1677 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1678 bo->bo_flag &= ~BO_ONWORKLST;
1679 mtx_lock(&sync_mtx);
1680 LIST_REMOVE(bo, bo_synclist);
1681 syncer_worklist_len--;
1682 mtx_unlock(&sync_mtx);
1684 bp->b_flags &= ~B_NEEDSGIANT;
1686 bp->b_bufobj = NULL;
1692 * Add an item to the syncer work queue.
1695 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1699 ASSERT_BO_LOCKED(bo);
1701 mtx_lock(&sync_mtx);
1702 if (bo->bo_flag & BO_ONWORKLST)
1703 LIST_REMOVE(bo, bo_synclist);
1705 bo->bo_flag |= BO_ONWORKLST;
1706 syncer_worklist_len++;
1709 if (delay > syncer_maxdelay - 2)
1710 delay = syncer_maxdelay - 2;
1711 slot = (syncer_delayno + delay) & syncer_mask;
1713 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1715 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1717 mtx_unlock(&sync_mtx);
1721 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1725 mtx_lock(&sync_mtx);
1726 len = syncer_worklist_len - sync_vnode_count;
1727 mtx_unlock(&sync_mtx);
1728 error = SYSCTL_OUT(req, &len, sizeof(len));
1732 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1733 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1735 static struct proc *updateproc;
1736 static void sched_sync(void);
1737 static struct kproc_desc up_kp = {
1742 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1745 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1750 *bo = LIST_FIRST(slp);
1753 vp = (*bo)->__bo_vnode; /* XXX */
1754 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1757 * We use vhold in case the vnode does not
1758 * successfully sync. vhold prevents the vnode from
1759 * going away when we unlock the sync_mtx so that
1760 * we can acquire the vnode interlock.
1763 mtx_unlock(&sync_mtx);
1765 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1767 mtx_lock(&sync_mtx);
1768 return (*bo == LIST_FIRST(slp));
1770 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1771 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1773 vn_finished_write(mp);
1775 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1777 * Put us back on the worklist. The worklist
1778 * routine will remove us from our current
1779 * position and then add us back in at a later
1782 vn_syncer_add_to_worklist(*bo, syncdelay);
1786 mtx_lock(&sync_mtx);
1791 * System filesystem synchronizer daemon.
1796 struct synclist *gnext, *next;
1797 struct synclist *gslp, *slp;
1800 struct thread *td = curthread;
1802 int net_worklist_len;
1803 int syncer_final_iter;
1808 syncer_final_iter = 0;
1810 syncer_state = SYNCER_RUNNING;
1811 starttime = time_uptime;
1812 td->td_pflags |= TDP_NORUNNINGBUF;
1814 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1817 mtx_lock(&sync_mtx);
1819 if (syncer_state == SYNCER_FINAL_DELAY &&
1820 syncer_final_iter == 0) {
1821 mtx_unlock(&sync_mtx);
1822 kproc_suspend_check(td->td_proc);
1823 mtx_lock(&sync_mtx);
1825 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1826 if (syncer_state != SYNCER_RUNNING &&
1827 starttime != time_uptime) {
1829 printf("\nSyncing disks, vnodes remaining...");
1832 printf("%d ", net_worklist_len);
1834 starttime = time_uptime;
1837 * Push files whose dirty time has expired. Be careful
1838 * of interrupt race on slp queue.
1840 * Skip over empty worklist slots when shutting down.
1843 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1844 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1845 syncer_delayno += 1;
1846 if (syncer_delayno == syncer_maxdelay)
1848 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1849 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1851 * If the worklist has wrapped since the
1852 * it was emptied of all but syncer vnodes,
1853 * switch to the FINAL_DELAY state and run
1854 * for one more second.
1856 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1857 net_worklist_len == 0 &&
1858 last_work_seen == syncer_delayno) {
1859 syncer_state = SYNCER_FINAL_DELAY;
1860 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1862 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1863 LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1866 * Keep track of the last time there was anything
1867 * on the worklist other than syncer vnodes.
1868 * Return to the SHUTTING_DOWN state if any
1871 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1872 last_work_seen = syncer_delayno;
1873 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1874 syncer_state = SYNCER_SHUTTING_DOWN;
1875 while (!LIST_EMPTY(slp)) {
1876 error = sync_vnode(slp, &bo, td);
1878 LIST_REMOVE(bo, bo_synclist);
1879 LIST_INSERT_HEAD(next, bo, bo_synclist);
1883 if (first_printf == 0)
1884 wdog_kern_pat(WD_LASTVAL);
1887 if (!LIST_EMPTY(gslp)) {
1888 mtx_unlock(&sync_mtx);
1890 mtx_lock(&sync_mtx);
1891 while (!LIST_EMPTY(gslp)) {
1892 error = sync_vnode(gslp, &bo, td);
1894 LIST_REMOVE(bo, bo_synclist);
1895 LIST_INSERT_HEAD(gnext, bo,
1902 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1903 syncer_final_iter--;
1905 * The variable rushjob allows the kernel to speed up the
1906 * processing of the filesystem syncer process. A rushjob
1907 * value of N tells the filesystem syncer to process the next
1908 * N seconds worth of work on its queue ASAP. Currently rushjob
1909 * is used by the soft update code to speed up the filesystem
1910 * syncer process when the incore state is getting so far
1911 * ahead of the disk that the kernel memory pool is being
1912 * threatened with exhaustion.
1919 * Just sleep for a short period of time between
1920 * iterations when shutting down to allow some I/O
1923 * If it has taken us less than a second to process the
1924 * current work, then wait. Otherwise start right over
1925 * again. We can still lose time if any single round
1926 * takes more than two seconds, but it does not really
1927 * matter as we are just trying to generally pace the
1928 * filesystem activity.
1930 if (syncer_state != SYNCER_RUNNING ||
1931 time_uptime == starttime) {
1933 sched_prio(td, PPAUSE);
1936 if (syncer_state != SYNCER_RUNNING)
1937 cv_timedwait(&sync_wakeup, &sync_mtx,
1938 hz / SYNCER_SHUTDOWN_SPEEDUP);
1939 else if (time_uptime == starttime)
1940 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1945 * Request the syncer daemon to speed up its work.
1946 * We never push it to speed up more than half of its
1947 * normal turn time, otherwise it could take over the cpu.
1950 speedup_syncer(void)
1954 mtx_lock(&sync_mtx);
1955 if (rushjob < syncdelay / 2) {
1957 stat_rush_requests += 1;
1960 mtx_unlock(&sync_mtx);
1961 cv_broadcast(&sync_wakeup);
1966 * Tell the syncer to speed up its work and run though its work
1967 * list several times, then tell it to shut down.
1970 syncer_shutdown(void *arg, int howto)
1973 if (howto & RB_NOSYNC)
1975 mtx_lock(&sync_mtx);
1976 syncer_state = SYNCER_SHUTTING_DOWN;
1978 mtx_unlock(&sync_mtx);
1979 cv_broadcast(&sync_wakeup);
1980 kproc_shutdown(arg, howto);
1984 * Reassign a buffer from one vnode to another.
1985 * Used to assign file specific control information
1986 * (indirect blocks) to the vnode to which they belong.
1989 reassignbuf(struct buf *bp)
2002 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
2003 bp, bp->b_vp, bp->b_flags);
2005 * B_PAGING flagged buffers cannot be reassigned because their vp
2006 * is not fully linked in.
2008 if (bp->b_flags & B_PAGING)
2009 panic("cannot reassign paging buffer");
2012 * Delete from old vnode list, if on one.
2015 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2016 buf_vlist_remove(bp);
2018 panic("reassignbuf: Buffer %p not on queue.", bp);
2020 * If dirty, put on list of dirty buffers; otherwise insert onto list
2023 if (bp->b_flags & B_DELWRI) {
2024 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2025 switch (vp->v_type) {
2035 vn_syncer_add_to_worklist(bo, delay);
2037 buf_vlist_add(bp, bo, BX_VNDIRTY);
2039 buf_vlist_add(bp, bo, BX_VNCLEAN);
2041 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2042 mtx_lock(&sync_mtx);
2043 LIST_REMOVE(bo, bo_synclist);
2044 syncer_worklist_len--;
2045 mtx_unlock(&sync_mtx);
2046 bo->bo_flag &= ~BO_ONWORKLST;
2051 bp = TAILQ_FIRST(&bv->bv_hd);
2052 KASSERT(bp == NULL || bp->b_bufobj == bo,
2053 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2054 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2055 KASSERT(bp == NULL || bp->b_bufobj == bo,
2056 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2058 bp = TAILQ_FIRST(&bv->bv_hd);
2059 KASSERT(bp == NULL || bp->b_bufobj == bo,
2060 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2061 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2062 KASSERT(bp == NULL || bp->b_bufobj == bo,
2063 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2069 * Increment the use and hold counts on the vnode, taking care to reference
2070 * the driver's usecount if this is a chardev. The vholdl() will remove
2071 * the vnode from the free list if it is presently free. Requires the
2072 * vnode interlock and returns with it held.
2075 v_incr_usecount(struct vnode *vp)
2078 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2080 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2082 vp->v_rdev->si_usecount++;
2089 * Turn a holdcnt into a use+holdcnt such that only one call to
2090 * v_decr_usecount is needed.
2093 v_upgrade_usecount(struct vnode *vp)
2096 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2098 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2100 vp->v_rdev->si_usecount++;
2106 * Decrement the vnode use and hold count along with the driver's usecount
2107 * if this is a chardev. The vdropl() below releases the vnode interlock
2108 * as it may free the vnode.
2111 v_decr_usecount(struct vnode *vp)
2114 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2115 VNASSERT(vp->v_usecount > 0, vp,
2116 ("v_decr_usecount: negative usecount"));
2117 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2119 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2121 vp->v_rdev->si_usecount--;
2128 * Decrement only the use count and driver use count. This is intended to
2129 * be paired with a follow on vdropl() to release the remaining hold count.
2130 * In this way we may vgone() a vnode with a 0 usecount without risk of
2131 * having it end up on a free list because the hold count is kept above 0.
2134 v_decr_useonly(struct vnode *vp)
2137 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2138 VNASSERT(vp->v_usecount > 0, vp,
2139 ("v_decr_useonly: negative usecount"));
2140 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2142 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2144 vp->v_rdev->si_usecount--;
2150 * Grab a particular vnode from the free list, increment its
2151 * reference count and lock it. VI_DOOMED is set if the vnode
2152 * is being destroyed. Only callers who specify LK_RETRY will
2153 * see doomed vnodes. If inactive processing was delayed in
2154 * vput try to do it here.
2157 vget(struct vnode *vp, int flags, struct thread *td)
2162 VFS_ASSERT_GIANT(vp->v_mount);
2163 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2164 ("vget: invalid lock operation"));
2165 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2167 if ((flags & LK_INTERLOCK) == 0)
2170 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2172 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2176 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2177 panic("vget: vn_lock failed to return ENOENT\n");
2179 /* Upgrade our holdcnt to a usecount. */
2180 v_upgrade_usecount(vp);
2182 * We don't guarantee that any particular close will
2183 * trigger inactive processing so just make a best effort
2184 * here at preventing a reference to a removed file. If
2185 * we don't succeed no harm is done.
2187 if (vp->v_iflag & VI_OWEINACT) {
2188 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2189 (flags & LK_NOWAIT) == 0)
2191 vp->v_iflag &= ~VI_OWEINACT;
2198 * Increase the reference count of a vnode.
2201 vref(struct vnode *vp)
2204 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2206 v_incr_usecount(vp);
2211 * Return reference count of a vnode.
2213 * The results of this call are only guaranteed when some mechanism other
2214 * than the VI lock is used to stop other processes from gaining references
2215 * to the vnode. This may be the case if the caller holds the only reference.
2216 * This is also useful when stale data is acceptable as race conditions may
2217 * be accounted for by some other means.
2220 vrefcnt(struct vnode *vp)
2225 usecnt = vp->v_usecount;
2231 #define VPUTX_VRELE 1
2232 #define VPUTX_VPUT 2
2233 #define VPUTX_VUNREF 3
2236 vputx(struct vnode *vp, int func)
2240 KASSERT(vp != NULL, ("vputx: null vp"));
2241 if (func == VPUTX_VUNREF)
2242 ASSERT_VOP_LOCKED(vp, "vunref");
2243 else if (func == VPUTX_VPUT)
2244 ASSERT_VOP_LOCKED(vp, "vput");
2246 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2247 VFS_ASSERT_GIANT(vp->v_mount);
2248 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2251 /* Skip this v_writecount check if we're going to panic below. */
2252 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2253 ("vputx: missed vn_close"));
2256 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2257 vp->v_usecount == 1)) {
2258 if (func == VPUTX_VPUT)
2260 v_decr_usecount(vp);
2264 if (vp->v_usecount != 1) {
2265 vprint("vputx: negative ref count", vp);
2266 panic("vputx: negative ref cnt");
2268 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2270 * We want to hold the vnode until the inactive finishes to
2271 * prevent vgone() races. We drop the use count here and the
2272 * hold count below when we're done.
2276 * We must call VOP_INACTIVE with the node locked. Mark
2277 * as VI_DOINGINACT to avoid recursion.
2279 vp->v_iflag |= VI_OWEINACT;
2282 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2286 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2287 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2293 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2297 if (vp->v_usecount > 0)
2298 vp->v_iflag &= ~VI_OWEINACT;
2300 if (vp->v_iflag & VI_OWEINACT)
2301 vinactive(vp, curthread);
2302 if (func != VPUTX_VUNREF)
2309 * Vnode put/release.
2310 * If count drops to zero, call inactive routine and return to freelist.
2313 vrele(struct vnode *vp)
2316 vputx(vp, VPUTX_VRELE);
2320 * Release an already locked vnode. This give the same effects as
2321 * unlock+vrele(), but takes less time and avoids releasing and
2322 * re-aquiring the lock (as vrele() acquires the lock internally.)
2325 vput(struct vnode *vp)
2328 vputx(vp, VPUTX_VPUT);
2332 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2335 vunref(struct vnode *vp)
2338 vputx(vp, VPUTX_VUNREF);
2342 * Somebody doesn't want the vnode recycled.
2345 vhold(struct vnode *vp)
2354 vholdl(struct vnode *vp)
2357 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2359 if (VSHOULDBUSY(vp))
2364 * Note that there is one less who cares about this vnode. vdrop() is the
2365 * opposite of vhold().
2368 vdrop(struct vnode *vp)
2376 * Drop the hold count of the vnode. If this is the last reference to
2377 * the vnode we will free it if it has been vgone'd otherwise it is
2378 * placed on the free list.
2381 vdropl(struct vnode *vp)
2384 ASSERT_VI_LOCKED(vp, "vdropl");
2385 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2386 if (vp->v_holdcnt <= 0)
2387 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2389 if (vp->v_holdcnt == 0) {
2390 if (vp->v_iflag & VI_DOOMED) {
2391 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__,
2402 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2403 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2404 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2405 * failed lock upgrade.
2408 vinactive(struct vnode *vp, struct thread *td)
2411 ASSERT_VOP_ELOCKED(vp, "vinactive");
2412 ASSERT_VI_LOCKED(vp, "vinactive");
2413 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2414 ("vinactive: recursed on VI_DOINGINACT"));
2415 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2416 vp->v_iflag |= VI_DOINGINACT;
2417 vp->v_iflag &= ~VI_OWEINACT;
2419 VOP_INACTIVE(vp, td);
2421 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2422 ("vinactive: lost VI_DOINGINACT"));
2423 vp->v_iflag &= ~VI_DOINGINACT;
2427 * Remove any vnodes in the vnode table belonging to mount point mp.
2429 * If FORCECLOSE is not specified, there should not be any active ones,
2430 * return error if any are found (nb: this is a user error, not a
2431 * system error). If FORCECLOSE is specified, detach any active vnodes
2434 * If WRITECLOSE is set, only flush out regular file vnodes open for
2437 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2439 * `rootrefs' specifies the base reference count for the root vnode
2440 * of this filesystem. The root vnode is considered busy if its
2441 * v_usecount exceeds this value. On a successful return, vflush(, td)
2442 * will call vrele() on the root vnode exactly rootrefs times.
2443 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2447 static int busyprt = 0; /* print out busy vnodes */
2448 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2452 vflush( struct mount *mp, int rootrefs, int flags, struct thread *td)
2454 struct vnode *vp, *mvp, *rootvp = NULL;
2456 int busy = 0, error;
2458 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2461 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2462 ("vflush: bad args"));
2464 * Get the filesystem root vnode. We can vput() it
2465 * immediately, since with rootrefs > 0, it won't go away.
2467 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2468 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2477 MNT_VNODE_FOREACH(vp, mp, mvp) {
2482 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2486 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2490 * Skip over a vnodes marked VV_SYSTEM.
2492 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2499 * If WRITECLOSE is set, flush out unlinked but still open
2500 * files (even if open only for reading) and regular file
2501 * vnodes open for writing.
2503 if (flags & WRITECLOSE) {
2504 if (vp->v_object != NULL) {
2505 VM_OBJECT_LOCK(vp->v_object);
2506 vm_object_page_clean(vp->v_object, 0, 0, 0);
2507 VM_OBJECT_UNLOCK(vp->v_object);
2509 error = VOP_FSYNC(vp, MNT_WAIT, td);
2513 MNT_VNODE_FOREACH_ABORT(mp, mvp);
2516 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2519 if ((vp->v_type == VNON ||
2520 (error == 0 && vattr.va_nlink > 0)) &&
2521 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2530 * With v_usecount == 0, all we need to do is clear out the
2531 * vnode data structures and we are done.
2533 * If FORCECLOSE is set, forcibly close the vnode.
2535 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2536 VNASSERT(vp->v_usecount == 0 ||
2537 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2538 ("device VNODE %p is FORCECLOSED", vp));
2544 vprint("vflush: busy vnode", vp);
2552 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2554 * If just the root vnode is busy, and if its refcount
2555 * is equal to `rootrefs', then go ahead and kill it.
2558 KASSERT(busy > 0, ("vflush: not busy"));
2559 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2560 ("vflush: usecount %d < rootrefs %d",
2561 rootvp->v_usecount, rootrefs));
2562 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2563 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2565 VOP_UNLOCK(rootvp, 0);
2571 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2575 for (; rootrefs > 0; rootrefs--)
2581 * Recycle an unused vnode to the front of the free list.
2584 vrecycle(struct vnode *vp, struct thread *td)
2588 ASSERT_VOP_ELOCKED(vp, "vrecycle");
2589 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2592 if (vp->v_usecount == 0) {
2601 * Eliminate all activity associated with a vnode
2602 * in preparation for reuse.
2605 vgone(struct vnode *vp)
2613 * vgone, with the vp interlock held.
2616 vgonel(struct vnode *vp)
2623 ASSERT_VOP_ELOCKED(vp, "vgonel");
2624 ASSERT_VI_LOCKED(vp, "vgonel");
2625 VNASSERT(vp->v_holdcnt, vp,
2626 ("vgonel: vp %p has no reference.", vp));
2627 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2631 * Don't vgonel if we're already doomed.
2633 if (vp->v_iflag & VI_DOOMED)
2635 vp->v_iflag |= VI_DOOMED;
2637 * Check to see if the vnode is in use. If so, we have to call
2638 * VOP_CLOSE() and VOP_INACTIVE().
2640 active = vp->v_usecount;
2641 oweinact = (vp->v_iflag & VI_OWEINACT);
2644 * Clean out any buffers associated with the vnode.
2645 * If the flush fails, just toss the buffers.
2648 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2649 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2650 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2651 vinvalbuf(vp, 0, 0, 0);
2654 * If purging an active vnode, it must be closed and
2655 * deactivated before being reclaimed.
2658 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2659 if (oweinact || active) {
2661 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2666 * Reclaim the vnode.
2668 if (VOP_RECLAIM(vp, td))
2669 panic("vgone: cannot reclaim");
2671 vn_finished_secondary_write(mp);
2672 VNASSERT(vp->v_object == NULL, vp,
2673 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2675 * Clear the advisory locks and wake up waiting threads.
2677 lf_purgelocks(vp, &(vp->v_lockf));
2679 * Delete from old mount point vnode list.
2684 * Done with purge, reset to the standard lock and invalidate
2688 vp->v_vnlock = &vp->v_lock;
2689 vp->v_op = &dead_vnodeops;
2695 * Calculate the total number of references to a special device.
2698 vcount(struct vnode *vp)
2703 count = vp->v_rdev->si_usecount;
2709 * Same as above, but using the struct cdev *as argument
2712 count_dev(struct cdev *dev)
2717 count = dev->si_usecount;
2723 * Print out a description of a vnode.
2725 static char *typename[] =
2726 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2730 vn_printf(struct vnode *vp, const char *fmt, ...)
2733 char buf[256], buf2[16];
2739 printf("%p: ", (void *)vp);
2740 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2741 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2742 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2745 if (vp->v_vflag & VV_ROOT)
2746 strlcat(buf, "|VV_ROOT", sizeof(buf));
2747 if (vp->v_vflag & VV_ISTTY)
2748 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2749 if (vp->v_vflag & VV_NOSYNC)
2750 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2751 if (vp->v_vflag & VV_CACHEDLABEL)
2752 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2753 if (vp->v_vflag & VV_TEXT)
2754 strlcat(buf, "|VV_TEXT", sizeof(buf));
2755 if (vp->v_vflag & VV_COPYONWRITE)
2756 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2757 if (vp->v_vflag & VV_SYSTEM)
2758 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2759 if (vp->v_vflag & VV_PROCDEP)
2760 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2761 if (vp->v_vflag & VV_NOKNOTE)
2762 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2763 if (vp->v_vflag & VV_DELETED)
2764 strlcat(buf, "|VV_DELETED", sizeof(buf));
2765 if (vp->v_vflag & VV_MD)
2766 strlcat(buf, "|VV_MD", sizeof(buf));
2767 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2768 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2769 VV_NOKNOTE | VV_DELETED | VV_MD);
2771 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2772 strlcat(buf, buf2, sizeof(buf));
2774 if (vp->v_iflag & VI_MOUNT)
2775 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2776 if (vp->v_iflag & VI_AGE)
2777 strlcat(buf, "|VI_AGE", sizeof(buf));
2778 if (vp->v_iflag & VI_DOOMED)
2779 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2780 if (vp->v_iflag & VI_FREE)
2781 strlcat(buf, "|VI_FREE", sizeof(buf));
2782 if (vp->v_iflag & VI_DOINGINACT)
2783 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2784 if (vp->v_iflag & VI_OWEINACT)
2785 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2786 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2787 VI_DOINGINACT | VI_OWEINACT);
2789 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2790 strlcat(buf, buf2, sizeof(buf));
2792 printf(" flags (%s)\n", buf + 1);
2793 if (mtx_owned(VI_MTX(vp)))
2794 printf(" VI_LOCKed");
2795 if (vp->v_object != NULL)
2796 printf(" v_object %p ref %d pages %d\n",
2797 vp->v_object, vp->v_object->ref_count,
2798 vp->v_object->resident_page_count);
2800 lockmgr_printinfo(vp->v_vnlock);
2801 if (vp->v_data != NULL)
2807 * List all of the locked vnodes in the system.
2808 * Called when debugging the kernel.
2810 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2812 struct mount *mp, *nmp;
2816 * Note: because this is DDB, we can't obey the locking semantics
2817 * for these structures, which means we could catch an inconsistent
2818 * state and dereference a nasty pointer. Not much to be done
2821 db_printf("Locked vnodes\n");
2822 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2823 nmp = TAILQ_NEXT(mp, mnt_list);
2824 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2825 if (vp->v_type != VMARKER &&
2829 nmp = TAILQ_NEXT(mp, mnt_list);
2834 * Show details about the given vnode.
2836 DB_SHOW_COMMAND(vnode, db_show_vnode)
2842 vp = (struct vnode *)addr;
2843 vn_printf(vp, "vnode ");
2847 * Show details about the given mount point.
2849 DB_SHOW_COMMAND(mount, db_show_mount)
2859 /* No address given, print short info about all mount points. */
2860 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2861 db_printf("%p %s on %s (%s)\n", mp,
2862 mp->mnt_stat.f_mntfromname,
2863 mp->mnt_stat.f_mntonname,
2864 mp->mnt_stat.f_fstypename);
2868 db_printf("\nMore info: show mount <addr>\n");
2872 mp = (struct mount *)addr;
2873 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2874 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2877 flags = mp->mnt_flag;
2878 #define MNT_FLAG(flag) do { \
2879 if (flags & (flag)) { \
2880 if (buf[0] != '\0') \
2881 strlcat(buf, ", ", sizeof(buf)); \
2882 strlcat(buf, (#flag) + 4, sizeof(buf)); \
2886 MNT_FLAG(MNT_RDONLY);
2887 MNT_FLAG(MNT_SYNCHRONOUS);
2888 MNT_FLAG(MNT_NOEXEC);
2889 MNT_FLAG(MNT_NOSUID);
2890 MNT_FLAG(MNT_UNION);
2891 MNT_FLAG(MNT_ASYNC);
2892 MNT_FLAG(MNT_SUIDDIR);
2893 MNT_FLAG(MNT_SOFTDEP);
2894 MNT_FLAG(MNT_NOSYMFOLLOW);
2895 MNT_FLAG(MNT_GJOURNAL);
2896 MNT_FLAG(MNT_MULTILABEL);
2898 MNT_FLAG(MNT_NOATIME);
2899 MNT_FLAG(MNT_NOCLUSTERR);
2900 MNT_FLAG(MNT_NOCLUSTERW);
2901 MNT_FLAG(MNT_NFS4ACLS);
2902 MNT_FLAG(MNT_EXRDONLY);
2903 MNT_FLAG(MNT_EXPORTED);
2904 MNT_FLAG(MNT_DEFEXPORTED);
2905 MNT_FLAG(MNT_EXPORTANON);
2906 MNT_FLAG(MNT_EXKERB);
2907 MNT_FLAG(MNT_EXPUBLIC);
2908 MNT_FLAG(MNT_LOCAL);
2909 MNT_FLAG(MNT_QUOTA);
2910 MNT_FLAG(MNT_ROOTFS);
2912 MNT_FLAG(MNT_IGNORE);
2913 MNT_FLAG(MNT_UPDATE);
2914 MNT_FLAG(MNT_DELEXPORT);
2915 MNT_FLAG(MNT_RELOAD);
2916 MNT_FLAG(MNT_FORCE);
2917 MNT_FLAG(MNT_SNAPSHOT);
2918 MNT_FLAG(MNT_BYFSID);
2922 strlcat(buf, ", ", sizeof(buf));
2923 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2926 db_printf(" mnt_flag = %s\n", buf);
2929 flags = mp->mnt_kern_flag;
2930 #define MNT_KERN_FLAG(flag) do { \
2931 if (flags & (flag)) { \
2932 if (buf[0] != '\0') \
2933 strlcat(buf, ", ", sizeof(buf)); \
2934 strlcat(buf, (#flag) + 5, sizeof(buf)); \
2938 MNT_KERN_FLAG(MNTK_UNMOUNTF);
2939 MNT_KERN_FLAG(MNTK_ASYNC);
2940 MNT_KERN_FLAG(MNTK_SOFTDEP);
2941 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2942 MNT_KERN_FLAG(MNTK_UNMOUNT);
2943 MNT_KERN_FLAG(MNTK_MWAIT);
2944 MNT_KERN_FLAG(MNTK_SUSPEND);
2945 MNT_KERN_FLAG(MNTK_SUSPEND2);
2946 MNT_KERN_FLAG(MNTK_SUSPENDED);
2947 MNT_KERN_FLAG(MNTK_MPSAFE);
2948 MNT_KERN_FLAG(MNTK_NOKNOTE);
2949 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2950 #undef MNT_KERN_FLAG
2953 strlcat(buf, ", ", sizeof(buf));
2954 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2957 db_printf(" mnt_kern_flag = %s\n", buf);
2959 db_printf(" mnt_opt = ");
2960 opt = TAILQ_FIRST(mp->mnt_opt);
2962 db_printf("%s", opt->name);
2963 opt = TAILQ_NEXT(opt, link);
2964 while (opt != NULL) {
2965 db_printf(", %s", opt->name);
2966 opt = TAILQ_NEXT(opt, link);
2972 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
2973 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2974 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2975 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2976 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2977 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2978 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2979 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2980 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2981 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2982 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2983 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2985 db_printf(" mnt_cred = { uid=%u ruid=%u",
2986 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2987 if (jailed(mp->mnt_cred))
2988 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2990 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
2991 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
2992 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2993 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2994 db_printf(" mnt_noasync = %u\n", mp->mnt_noasync);
2995 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2996 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2997 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
2998 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
2999 db_printf(" mnt_secondary_accwrites = %d\n",
3000 mp->mnt_secondary_accwrites);
3001 db_printf(" mnt_gjprovider = %s\n",
3002 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3005 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3006 if (vp->v_type != VMARKER) {
3007 vn_printf(vp, "vnode ");
3016 * Fill in a struct xvfsconf based on a struct vfsconf.
3019 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
3022 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
3023 xvfsp->vfc_typenum = vfsp->vfc_typenum;
3024 xvfsp->vfc_refcount = vfsp->vfc_refcount;
3025 xvfsp->vfc_flags = vfsp->vfc_flags;
3027 * These are unused in userland, we keep them
3028 * to not break binary compatibility.
3030 xvfsp->vfc_vfsops = NULL;
3031 xvfsp->vfc_next = NULL;
3035 * Top level filesystem related information gathering.
3038 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3040 struct vfsconf *vfsp;
3041 struct xvfsconf xvfsp;
3045 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3046 bzero(&xvfsp, sizeof(xvfsp));
3047 vfsconf2x(vfsp, &xvfsp);
3048 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
3055 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
3056 "S,xvfsconf", "List of all configured filesystems");
3058 #ifndef BURN_BRIDGES
3059 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3062 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3064 int *name = (int *)arg1 - 1; /* XXX */
3065 u_int namelen = arg2 + 1; /* XXX */
3066 struct vfsconf *vfsp;
3067 struct xvfsconf xvfsp;
3069 printf("WARNING: userland calling deprecated sysctl, "
3070 "please rebuild world\n");
3072 #if 1 || defined(COMPAT_PRELITE2)
3073 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3075 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3079 case VFS_MAXTYPENUM:
3082 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3085 return (ENOTDIR); /* overloaded */
3086 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3087 if (vfsp->vfc_typenum == name[2])
3090 return (EOPNOTSUPP);
3091 bzero(&xvfsp, sizeof(xvfsp));
3092 vfsconf2x(vfsp, &xvfsp);
3093 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3095 return (EOPNOTSUPP);
3098 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3099 vfs_sysctl, "Generic filesystem");
3101 #if 1 || defined(COMPAT_PRELITE2)
3104 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3107 struct vfsconf *vfsp;
3108 struct ovfsconf ovfs;
3110 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3111 bzero(&ovfs, sizeof(ovfs));
3112 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3113 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3114 ovfs.vfc_index = vfsp->vfc_typenum;
3115 ovfs.vfc_refcount = vfsp->vfc_refcount;
3116 ovfs.vfc_flags = vfsp->vfc_flags;
3117 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3124 #endif /* 1 || COMPAT_PRELITE2 */
3125 #endif /* !BURN_BRIDGES */
3127 #define KINFO_VNODESLOP 10
3130 * Dump vnode list (via sysctl).
3134 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3142 * Stale numvnodes access is not fatal here.
3145 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3147 /* Make an estimate */
3148 return (SYSCTL_OUT(req, 0, len));
3150 error = sysctl_wire_old_buffer(req, 0);
3153 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3155 mtx_lock(&mountlist_mtx);
3156 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3157 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3160 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3164 xvn[n].xv_size = sizeof *xvn;
3165 xvn[n].xv_vnode = vp;
3166 xvn[n].xv_id = 0; /* XXX compat */
3167 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3169 XV_COPY(writecount);
3175 xvn[n].xv_flag = vp->v_vflag;
3177 switch (vp->v_type) {
3184 if (vp->v_rdev == NULL) {
3188 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3191 xvn[n].xv_socket = vp->v_socket;
3194 xvn[n].xv_fifo = vp->v_fifoinfo;
3199 /* shouldn't happen? */
3207 mtx_lock(&mountlist_mtx);
3212 mtx_unlock(&mountlist_mtx);
3214 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3219 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3220 0, 0, sysctl_vnode, "S,xvnode", "");
3224 * Unmount all filesystems. The list is traversed in reverse order
3225 * of mounting to avoid dependencies.
3228 vfs_unmountall(void)
3234 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3235 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3239 * Since this only runs when rebooting, it is not interlocked.
3241 while(!TAILQ_EMPTY(&mountlist)) {
3242 mp = TAILQ_LAST(&mountlist, mntlist);
3243 error = dounmount(mp, MNT_FORCE, td);
3245 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3247 * XXX: Due to the way in which we mount the root
3248 * file system off of devfs, devfs will generate a
3249 * "busy" warning when we try to unmount it before
3250 * the root. Don't print a warning as a result in
3251 * order to avoid false positive errors that may
3252 * cause needless upset.
3254 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3255 printf("unmount of %s failed (",
3256 mp->mnt_stat.f_mntonname);
3260 printf("%d)\n", error);
3263 /* The unmount has removed mp from the mountlist */
3269 * perform msync on all vnodes under a mount point
3270 * the mount point must be locked.
3273 vfs_msync(struct mount *mp, int flags)
3275 struct vnode *vp, *mvp;
3276 struct vm_object *obj;
3278 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3280 MNT_VNODE_FOREACH(vp, mp, mvp) {
3283 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3284 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3287 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3289 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3297 VM_OBJECT_LOCK(obj);
3298 vm_object_page_clean(obj, 0, 0,
3300 OBJPC_SYNC : OBJPC_NOSYNC);
3301 VM_OBJECT_UNLOCK(obj);
3313 * Mark a vnode as free, putting it up for recycling.
3316 vfree(struct vnode *vp)
3319 ASSERT_VI_LOCKED(vp, "vfree");
3320 mtx_lock(&vnode_free_list_mtx);
3321 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3322 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3323 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3324 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3325 ("vfree: Freeing doomed vnode"));
3326 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3327 if (vp->v_iflag & VI_AGE) {
3328 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3330 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3333 vp->v_iflag &= ~VI_AGE;
3334 vp->v_iflag |= VI_FREE;
3335 mtx_unlock(&vnode_free_list_mtx);
3339 * Opposite of vfree() - mark a vnode as in use.
3342 vbusy(struct vnode *vp)
3344 ASSERT_VI_LOCKED(vp, "vbusy");
3345 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3346 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3347 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3349 mtx_lock(&vnode_free_list_mtx);
3350 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3352 vp->v_iflag &= ~(VI_FREE|VI_AGE);
3353 mtx_unlock(&vnode_free_list_mtx);
3357 destroy_vpollinfo(struct vpollinfo *vi)
3359 seldrain(&vi->vpi_selinfo);
3360 knlist_destroy(&vi->vpi_selinfo.si_note);
3361 mtx_destroy(&vi->vpi_lock);
3362 uma_zfree(vnodepoll_zone, vi);
3366 * Initalize per-vnode helper structure to hold poll-related state.
3369 v_addpollinfo(struct vnode *vp)
3371 struct vpollinfo *vi;
3373 if (vp->v_pollinfo != NULL)
3375 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3376 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3377 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3378 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3380 if (vp->v_pollinfo != NULL) {
3382 destroy_vpollinfo(vi);
3385 vp->v_pollinfo = vi;
3390 * Record a process's interest in events which might happen to
3391 * a vnode. Because poll uses the historic select-style interface
3392 * internally, this routine serves as both the ``check for any
3393 * pending events'' and the ``record my interest in future events''
3394 * functions. (These are done together, while the lock is held,
3395 * to avoid race conditions.)
3398 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3402 mtx_lock(&vp->v_pollinfo->vpi_lock);
3403 if (vp->v_pollinfo->vpi_revents & events) {
3405 * This leaves events we are not interested
3406 * in available for the other process which
3407 * which presumably had requested them
3408 * (otherwise they would never have been
3411 events &= vp->v_pollinfo->vpi_revents;
3412 vp->v_pollinfo->vpi_revents &= ~events;
3414 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3417 vp->v_pollinfo->vpi_events |= events;
3418 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3419 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3424 * Routine to create and manage a filesystem syncer vnode.
3426 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3427 static int sync_fsync(struct vop_fsync_args *);
3428 static int sync_inactive(struct vop_inactive_args *);
3429 static int sync_reclaim(struct vop_reclaim_args *);
3431 static struct vop_vector sync_vnodeops = {
3432 .vop_bypass = VOP_EOPNOTSUPP,
3433 .vop_close = sync_close, /* close */
3434 .vop_fsync = sync_fsync, /* fsync */
3435 .vop_inactive = sync_inactive, /* inactive */
3436 .vop_reclaim = sync_reclaim, /* reclaim */
3437 .vop_lock1 = vop_stdlock, /* lock */
3438 .vop_unlock = vop_stdunlock, /* unlock */
3439 .vop_islocked = vop_stdislocked, /* islocked */
3443 * Create a new filesystem syncer vnode for the specified mount point.
3446 vfs_allocate_syncvnode(struct mount *mp)
3450 static long start, incr, next;
3453 /* Allocate a new vnode */
3454 if ((error = getnewvnode("syncer", mp, &sync_vnodeops, &vp)) != 0) {
3455 mp->mnt_syncer = NULL;
3459 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3460 vp->v_vflag |= VV_FORCEINSMQ;
3461 error = insmntque(vp, mp);
3463 panic("vfs_allocate_syncvnode: insmntque failed");
3464 vp->v_vflag &= ~VV_FORCEINSMQ;
3467 * Place the vnode onto the syncer worklist. We attempt to
3468 * scatter them about on the list so that they will go off
3469 * at evenly distributed times even if all the filesystems
3470 * are mounted at once.
3473 if (next == 0 || next > syncer_maxdelay) {
3477 start = syncer_maxdelay / 2;
3478 incr = syncer_maxdelay;
3484 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3485 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3486 mtx_lock(&sync_mtx);
3488 mtx_unlock(&sync_mtx);
3490 mp->mnt_syncer = vp;
3495 * Do a lazy sync of the filesystem.
3498 sync_fsync(struct vop_fsync_args *ap)
3500 struct vnode *syncvp = ap->a_vp;
3501 struct mount *mp = syncvp->v_mount;
3506 * We only need to do something if this is a lazy evaluation.
3508 if (ap->a_waitfor != MNT_LAZY)
3512 * Move ourselves to the back of the sync list.
3514 bo = &syncvp->v_bufobj;
3516 vn_syncer_add_to_worklist(bo, syncdelay);
3520 * Walk the list of vnodes pushing all that are dirty and
3521 * not already on the sync list.
3523 mtx_lock(&mountlist_mtx);
3524 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3525 mtx_unlock(&mountlist_mtx);
3528 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3532 save = curthread_pflags_set(TDP_SYNCIO);
3533 vfs_msync(mp, MNT_NOWAIT);
3534 error = VFS_SYNC(mp, MNT_LAZY);
3535 curthread_pflags_restore(save);
3536 vn_finished_write(mp);
3542 * The syncer vnode is no referenced.
3545 sync_inactive(struct vop_inactive_args *ap)
3553 * The syncer vnode is no longer needed and is being decommissioned.
3555 * Modifications to the worklist must be protected by sync_mtx.
3558 sync_reclaim(struct vop_reclaim_args *ap)
3560 struct vnode *vp = ap->a_vp;
3565 vp->v_mount->mnt_syncer = NULL;
3566 if (bo->bo_flag & BO_ONWORKLST) {
3567 mtx_lock(&sync_mtx);
3568 LIST_REMOVE(bo, bo_synclist);
3569 syncer_worklist_len--;
3571 mtx_unlock(&sync_mtx);
3572 bo->bo_flag &= ~BO_ONWORKLST;
3580 * Check if vnode represents a disk device
3583 vn_isdisk(struct vnode *vp, int *errp)
3589 if (vp->v_type != VCHR)
3591 else if (vp->v_rdev == NULL)
3593 else if (vp->v_rdev->si_devsw == NULL)
3595 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3600 return (error == 0);
3604 * Common filesystem object access control check routine. Accepts a
3605 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3606 * and optional call-by-reference privused argument allowing vaccess()
3607 * to indicate to the caller whether privilege was used to satisfy the
3608 * request (obsoleted). Returns 0 on success, or an errno on failure.
3611 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3612 accmode_t accmode, struct ucred *cred, int *privused)
3614 accmode_t dac_granted;
3615 accmode_t priv_granted;
3617 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3618 ("invalid bit in accmode"));
3621 * Look for a normal, non-privileged way to access the file/directory
3622 * as requested. If it exists, go with that.
3625 if (privused != NULL)
3630 /* Check the owner. */
3631 if (cred->cr_uid == file_uid) {
3632 dac_granted |= VADMIN;
3633 if (file_mode & S_IXUSR)
3634 dac_granted |= VEXEC;
3635 if (file_mode & S_IRUSR)
3636 dac_granted |= VREAD;
3637 if (file_mode & S_IWUSR)
3638 dac_granted |= (VWRITE | VAPPEND);
3640 if ((accmode & dac_granted) == accmode)
3646 /* Otherwise, check the groups (first match) */
3647 if (groupmember(file_gid, cred)) {
3648 if (file_mode & S_IXGRP)
3649 dac_granted |= VEXEC;
3650 if (file_mode & S_IRGRP)
3651 dac_granted |= VREAD;
3652 if (file_mode & S_IWGRP)
3653 dac_granted |= (VWRITE | VAPPEND);
3655 if ((accmode & dac_granted) == accmode)
3661 /* Otherwise, check everyone else. */
3662 if (file_mode & S_IXOTH)
3663 dac_granted |= VEXEC;
3664 if (file_mode & S_IROTH)
3665 dac_granted |= VREAD;
3666 if (file_mode & S_IWOTH)
3667 dac_granted |= (VWRITE | VAPPEND);
3668 if ((accmode & dac_granted) == accmode)
3673 * Build a privilege mask to determine if the set of privileges
3674 * satisfies the requirements when combined with the granted mask
3675 * from above. For each privilege, if the privilege is required,
3676 * bitwise or the request type onto the priv_granted mask.
3682 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3683 * requests, instead of PRIV_VFS_EXEC.
3685 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3686 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3687 priv_granted |= VEXEC;
3689 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3690 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3691 priv_granted |= VEXEC;
3694 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3695 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3696 priv_granted |= VREAD;
3698 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3699 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3700 priv_granted |= (VWRITE | VAPPEND);
3702 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3703 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3704 priv_granted |= VADMIN;
3706 if ((accmode & (priv_granted | dac_granted)) == accmode) {
3707 /* XXX audit: privilege used */
3708 if (privused != NULL)
3713 return ((accmode & VADMIN) ? EPERM : EACCES);
3717 * Credential check based on process requesting service, and per-attribute
3721 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3722 struct thread *td, accmode_t accmode)
3726 * Kernel-invoked always succeeds.
3732 * Do not allow privileged processes in jail to directly manipulate
3733 * system attributes.
3735 switch (attrnamespace) {
3736 case EXTATTR_NAMESPACE_SYSTEM:
3737 /* Potentially should be: return (EPERM); */
3738 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3739 case EXTATTR_NAMESPACE_USER:
3740 return (VOP_ACCESS(vp, accmode, cred, td));
3746 #ifdef DEBUG_VFS_LOCKS
3748 * This only exists to supress warnings from unlocked specfs accesses. It is
3749 * no longer ok to have an unlocked VFS.
3751 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
3752 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
3754 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3755 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3756 "Drop into debugger on lock violation");
3758 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3759 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3760 0, "Check for interlock across VOPs");
3762 int vfs_badlock_print = 1; /* Print lock violations. */
3763 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3764 0, "Print lock violations");
3767 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3768 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3769 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3773 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3777 if (vfs_badlock_backtrace)
3780 if (vfs_badlock_print)
3781 printf("%s: %p %s\n", str, (void *)vp, msg);
3782 if (vfs_badlock_ddb)
3783 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3787 assert_vi_locked(struct vnode *vp, const char *str)
3790 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3791 vfs_badlock("interlock is not locked but should be", str, vp);
3795 assert_vi_unlocked(struct vnode *vp, const char *str)
3798 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3799 vfs_badlock("interlock is locked but should not be", str, vp);
3803 assert_vop_locked(struct vnode *vp, const char *str)
3806 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3807 vfs_badlock("is not locked but should be", str, vp);
3811 assert_vop_unlocked(struct vnode *vp, const char *str)
3814 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3815 vfs_badlock("is locked but should not be", str, vp);
3819 assert_vop_elocked(struct vnode *vp, const char *str)
3822 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3823 vfs_badlock("is not exclusive locked but should be", str, vp);
3828 assert_vop_elocked_other(struct vnode *vp, const char *str)
3831 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3832 vfs_badlock("is not exclusive locked by another thread",
3837 assert_vop_slocked(struct vnode *vp, const char *str)
3840 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3841 vfs_badlock("is not locked shared but should be", str, vp);
3844 #endif /* DEBUG_VFS_LOCKS */
3847 vop_rename_fail(struct vop_rename_args *ap)
3850 if (ap->a_tvp != NULL)
3852 if (ap->a_tdvp == ap->a_tvp)
3861 vop_rename_pre(void *ap)
3863 struct vop_rename_args *a = ap;
3865 #ifdef DEBUG_VFS_LOCKS
3867 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3868 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3869 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3870 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3872 /* Check the source (from). */
3873 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
3874 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
3875 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3876 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
3877 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3879 /* Check the target. */
3881 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3882 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3884 if (a->a_tdvp != a->a_fdvp)
3886 if (a->a_tvp != a->a_fvp)
3894 vop_strategy_pre(void *ap)
3896 #ifdef DEBUG_VFS_LOCKS
3897 struct vop_strategy_args *a;
3904 * Cluster ops lock their component buffers but not the IO container.
3906 if ((bp->b_flags & B_CLUSTER) != 0)
3909 if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
3910 if (vfs_badlock_print)
3912 "VOP_STRATEGY: bp is not locked but should be\n");
3913 if (vfs_badlock_ddb)
3914 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3920 vop_lookup_pre(void *ap)
3922 #ifdef DEBUG_VFS_LOCKS
3923 struct vop_lookup_args *a;
3928 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3929 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3934 vop_lookup_post(void *ap, int rc)
3936 #ifdef DEBUG_VFS_LOCKS
3937 struct vop_lookup_args *a;
3945 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3946 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3949 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3954 vop_lock_pre(void *ap)
3956 #ifdef DEBUG_VFS_LOCKS
3957 struct vop_lock1_args *a = ap;
3959 if ((a->a_flags & LK_INTERLOCK) == 0)
3960 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3962 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3967 vop_lock_post(void *ap, int rc)
3969 #ifdef DEBUG_VFS_LOCKS
3970 struct vop_lock1_args *a = ap;
3972 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3974 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
3979 vop_unlock_pre(void *ap)
3981 #ifdef DEBUG_VFS_LOCKS
3982 struct vop_unlock_args *a = ap;
3984 if (a->a_flags & LK_INTERLOCK)
3985 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
3986 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
3991 vop_unlock_post(void *ap, int rc)
3993 #ifdef DEBUG_VFS_LOCKS
3994 struct vop_unlock_args *a = ap;
3996 if (a->a_flags & LK_INTERLOCK)
3997 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4002 vop_create_post(void *ap, int rc)
4004 struct vop_create_args *a = ap;
4007 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4011 vop_deleteextattr_post(void *ap, int rc)
4013 struct vop_deleteextattr_args *a = ap;
4016 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4020 vop_link_post(void *ap, int rc)
4022 struct vop_link_args *a = ap;
4025 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4026 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4031 vop_mkdir_post(void *ap, int rc)
4033 struct vop_mkdir_args *a = ap;
4036 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4040 vop_mknod_post(void *ap, int rc)
4042 struct vop_mknod_args *a = ap;
4045 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4049 vop_remove_post(void *ap, int rc)
4051 struct vop_remove_args *a = ap;
4054 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4055 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4060 vop_rename_post(void *ap, int rc)
4062 struct vop_rename_args *a = ap;
4065 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4066 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4067 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4069 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4071 if (a->a_tdvp != a->a_fdvp)
4073 if (a->a_tvp != a->a_fvp)
4081 vop_rmdir_post(void *ap, int rc)
4083 struct vop_rmdir_args *a = ap;
4086 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4087 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4092 vop_setattr_post(void *ap, int rc)
4094 struct vop_setattr_args *a = ap;
4097 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4101 vop_setextattr_post(void *ap, int rc)
4103 struct vop_setextattr_args *a = ap;
4106 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4110 vop_symlink_post(void *ap, int rc)
4112 struct vop_symlink_args *a = ap;
4115 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4118 static struct knlist fs_knlist;
4121 vfs_event_init(void *arg)
4123 knlist_init_mtx(&fs_knlist, NULL);
4125 /* XXX - correct order? */
4126 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4129 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
4132 KNOTE_UNLOCKED(&fs_knlist, event);
4135 static int filt_fsattach(struct knote *kn);
4136 static void filt_fsdetach(struct knote *kn);
4137 static int filt_fsevent(struct knote *kn, long hint);
4139 struct filterops fs_filtops =
4140 { 0, filt_fsattach, filt_fsdetach, filt_fsevent };
4143 filt_fsattach(struct knote *kn)
4146 kn->kn_flags |= EV_CLEAR;
4147 knlist_add(&fs_knlist, kn, 0);
4152 filt_fsdetach(struct knote *kn)
4155 knlist_remove(&fs_knlist, kn, 0);
4159 filt_fsevent(struct knote *kn, long hint)
4162 kn->kn_fflags |= hint;
4163 return (kn->kn_fflags != 0);
4167 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4173 error = SYSCTL_IN(req, &vc, sizeof(vc));
4176 if (vc.vc_vers != VFS_CTL_VERS1)
4178 mp = vfs_getvfs(&vc.vc_fsid);
4181 /* ensure that a specific sysctl goes to the right filesystem. */
4182 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4183 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4187 VCTLTOREQ(&vc, req);
4188 error = VFS_SYSCTL(mp, vc.vc_op, req);
4193 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "",
4197 * Function to initialize a va_filerev field sensibly.
4198 * XXX: Wouldn't a random number make a lot more sense ??
4201 init_va_filerev(void)
4206 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4209 static int filt_vfsread(struct knote *kn, long hint);
4210 static int filt_vfswrite(struct knote *kn, long hint);
4211 static int filt_vfsvnode(struct knote *kn, long hint);
4212 static void filt_vfsdetach(struct knote *kn);
4213 static struct filterops vfsread_filtops =
4214 { 1, NULL, filt_vfsdetach, filt_vfsread };
4215 static struct filterops vfswrite_filtops =
4216 { 1, NULL, filt_vfsdetach, filt_vfswrite };
4217 static struct filterops vfsvnode_filtops =
4218 { 1, NULL, filt_vfsdetach, filt_vfsvnode };
4221 vfs_knllock(void *arg)
4223 struct vnode *vp = arg;
4225 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4229 vfs_knlunlock(void *arg)
4231 struct vnode *vp = arg;
4237 vfs_knl_assert_locked(void *arg)
4239 #ifdef DEBUG_VFS_LOCKS
4240 struct vnode *vp = arg;
4242 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4247 vfs_knl_assert_unlocked(void *arg)
4249 #ifdef DEBUG_VFS_LOCKS
4250 struct vnode *vp = arg;
4252 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4257 vfs_kqfilter(struct vop_kqfilter_args *ap)
4259 struct vnode *vp = ap->a_vp;
4260 struct knote *kn = ap->a_kn;
4263 switch (kn->kn_filter) {
4265 kn->kn_fop = &vfsread_filtops;
4268 kn->kn_fop = &vfswrite_filtops;
4271 kn->kn_fop = &vfsvnode_filtops;
4277 kn->kn_hook = (caddr_t)vp;
4280 if (vp->v_pollinfo == NULL)
4282 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4283 knlist_add(knl, kn, 0);
4289 * Detach knote from vnode
4292 filt_vfsdetach(struct knote *kn)
4294 struct vnode *vp = (struct vnode *)kn->kn_hook;
4296 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4297 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4302 filt_vfsread(struct knote *kn, long hint)
4304 struct vnode *vp = (struct vnode *)kn->kn_hook;
4309 * filesystem is gone, so set the EOF flag and schedule
4310 * the knote for deletion.
4312 if (hint == NOTE_REVOKE) {
4314 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4319 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4323 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4324 res = (kn->kn_data != 0);
4331 filt_vfswrite(struct knote *kn, long hint)
4333 struct vnode *vp = (struct vnode *)kn->kn_hook;
4338 * filesystem is gone, so set the EOF flag and schedule
4339 * the knote for deletion.
4341 if (hint == NOTE_REVOKE)
4342 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4350 filt_vfsvnode(struct knote *kn, long hint)
4352 struct vnode *vp = (struct vnode *)kn->kn_hook;
4356 if (kn->kn_sfflags & hint)
4357 kn->kn_fflags |= hint;
4358 if (hint == NOTE_REVOKE) {
4359 kn->kn_flags |= EV_EOF;
4363 res = (kn->kn_fflags != 0);
4369 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4373 if (dp->d_reclen > ap->a_uio->uio_resid)
4374 return (ENAMETOOLONG);
4375 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4377 if (ap->a_ncookies != NULL) {
4378 if (ap->a_cookies != NULL)
4379 free(ap->a_cookies, M_TEMP);
4380 ap->a_cookies = NULL;
4381 *ap->a_ncookies = 0;
4385 if (ap->a_ncookies == NULL)
4388 KASSERT(ap->a_cookies,
4389 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4391 *ap->a_cookies = realloc(*ap->a_cookies,
4392 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4393 (*ap->a_cookies)[*ap->a_ncookies] = off;
4398 * Mark for update the access time of the file if the filesystem
4399 * supports VOP_MARKATIME. This functionality is used by execve and
4400 * mmap, so we want to avoid the I/O implied by directly setting
4401 * va_atime for the sake of efficiency.
4404 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4409 VFS_ASSERT_GIANT(mp);
4410 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4411 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4412 (void)VOP_MARKATIME(vp);
4416 * The purpose of this routine is to remove granularity from accmode_t,
4417 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4418 * VADMIN and VAPPEND.
4420 * If it returns 0, the caller is supposed to continue with the usual
4421 * access checks using 'accmode' as modified by this routine. If it
4422 * returns nonzero value, the caller is supposed to return that value
4425 * Note that after this routine runs, accmode may be zero.
4428 vfs_unixify_accmode(accmode_t *accmode)
4431 * There is no way to specify explicit "deny" rule using
4432 * file mode or POSIX.1e ACLs.
4434 if (*accmode & VEXPLICIT_DENY) {
4440 * None of these can be translated into usual access bits.
4441 * Also, the common case for NFSv4 ACLs is to not contain
4442 * either of these bits. Caller should check for VWRITE
4443 * on the containing directory instead.
4445 if (*accmode & (VDELETE_CHILD | VDELETE))
4448 if (*accmode & VADMIN_PERMS) {
4449 *accmode &= ~VADMIN_PERMS;
4454 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4455 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4457 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);