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 v_incr_usecount(struct vnode *);
107 static void v_decr_usecount(struct vnode *);
108 static void v_decr_useonly(struct vnode *);
109 static void v_upgrade_usecount(struct vnode *);
110 static void vfree(struct vnode *);
111 static void vnlru_free(int);
112 static void vgonel(struct vnode *);
113 static void vfs_knllock(void *arg);
114 static void vfs_knlunlock(void *arg);
115 static void vfs_knl_assert_locked(void *arg);
116 static void vfs_knl_assert_unlocked(void *arg);
117 static void destroy_vpollinfo(struct vpollinfo *vi);
120 * Number of vnodes in existence. Increased whenever getnewvnode()
121 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed
124 static unsigned long numvnodes;
126 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
127 "Number of vnodes in existence");
130 * Conversion tables for conversion from vnode types to inode formats
133 enum vtype iftovt_tab[16] = {
134 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
135 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
137 int vttoif_tab[10] = {
138 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
139 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
143 * List of vnodes that are ready for recycling.
145 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
148 * Free vnode target. Free vnodes may simply be files which have been stat'd
149 * but not read. This is somewhat common, and a small cache of such files
150 * should be kept to avoid recreation costs.
152 static u_long wantfreevnodes;
153 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
154 /* Number of vnodes in the free list. */
155 static u_long freevnodes;
156 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
157 "Number of vnodes in the free list");
159 static int vlru_allow_cache_src;
160 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
161 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
164 * Various variables used for debugging the new implementation of
166 * XXX these are probably of (very) limited utility now.
168 static int reassignbufcalls;
169 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
170 "Number of calls to reassignbuf");
173 * Cache for the mount type id assigned to NFS. This is used for
174 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
176 int nfs_mount_type = -1;
178 /* To keep more than one thread at a time from running vfs_getnewfsid */
179 static struct mtx mntid_mtx;
182 * Lock for any access to the following:
187 static struct mtx vnode_free_list_mtx;
189 /* Publicly exported FS */
190 struct nfs_public nfs_pub;
192 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
193 static uma_zone_t vnode_zone;
194 static uma_zone_t vnodepoll_zone;
197 * The workitem queue.
199 * It is useful to delay writes of file data and filesystem metadata
200 * for tens of seconds so that quickly created and deleted files need
201 * not waste disk bandwidth being created and removed. To realize this,
202 * we append vnodes to a "workitem" queue. When running with a soft
203 * updates implementation, most pending metadata dependencies should
204 * not wait for more than a few seconds. Thus, mounted on block devices
205 * are delayed only about a half the time that file data is delayed.
206 * Similarly, directory updates are more critical, so are only delayed
207 * about a third the time that file data is delayed. Thus, there are
208 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
209 * one each second (driven off the filesystem syncer process). The
210 * syncer_delayno variable indicates the next queue that is to be processed.
211 * Items that need to be processed soon are placed in this queue:
213 * syncer_workitem_pending[syncer_delayno]
215 * A delay of fifteen seconds is done by placing the request fifteen
216 * entries later in the queue:
218 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
221 static int syncer_delayno;
222 static long syncer_mask;
223 LIST_HEAD(synclist, bufobj);
224 static struct synclist *syncer_workitem_pending[2];
226 * The sync_mtx protects:
231 * syncer_workitem_pending
232 * syncer_worklist_len
235 static struct mtx sync_mtx;
236 static struct cv sync_wakeup;
238 #define SYNCER_MAXDELAY 32
239 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
240 static int syncdelay = 30; /* max time to delay syncing data */
241 static int filedelay = 30; /* time to delay syncing files */
242 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
243 "Time to delay syncing files (in seconds)");
244 static int dirdelay = 29; /* time to delay syncing directories */
245 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
246 "Time to delay syncing directories (in seconds)");
247 static int metadelay = 28; /* time to delay syncing metadata */
248 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
249 "Time to delay syncing metadata (in seconds)");
250 static int rushjob; /* number of slots to run ASAP */
251 static int stat_rush_requests; /* number of times I/O speeded up */
252 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
253 "Number of times I/O speeded up (rush requests)");
256 * When shutting down the syncer, run it at four times normal speed.
258 #define SYNCER_SHUTDOWN_SPEEDUP 4
259 static int sync_vnode_count;
260 static int syncer_worklist_len;
261 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
265 * Number of vnodes we want to exist at any one time. This is mostly used
266 * to size hash tables in vnode-related code. It is normally not used in
267 * getnewvnode(), as wantfreevnodes is normally nonzero.)
269 * XXX desiredvnodes is historical cruft and should not exist.
272 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
273 &desiredvnodes, 0, "Maximum number of vnodes");
274 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
275 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
276 static int vnlru_nowhere;
277 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
278 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
281 * Macros to control when a vnode is freed and recycled. All require
282 * the vnode interlock.
284 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
285 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
286 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
290 * Initialize the vnode management data structures.
292 * Reevaluate the following cap on the number of vnodes after the physical
293 * memory size exceeds 512GB. In the limit, as the physical memory size
294 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
296 #ifndef MAXVNODES_MAX
297 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
300 vntblinit(void *dummy __unused)
302 int physvnodes, virtvnodes;
305 * Desiredvnodes is a function of the physical memory size and the
306 * kernel's heap size. Generally speaking, it scales with the
307 * physical memory size. The ratio of desiredvnodes to physical pages
308 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
309 * marginal ratio of desiredvnodes to physical pages is one to
310 * sixteen. However, desiredvnodes is limited by the kernel's heap
311 * size. The memory required by desiredvnodes vnodes and vm objects
312 * may not exceed one seventh of the kernel's heap size.
314 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
315 cnt.v_page_count) / 16;
316 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
317 sizeof(struct vnode)));
318 desiredvnodes = min(physvnodes, virtvnodes);
319 if (desiredvnodes > MAXVNODES_MAX) {
321 printf("Reducing kern.maxvnodes %d -> %d\n",
322 desiredvnodes, MAXVNODES_MAX);
323 desiredvnodes = MAXVNODES_MAX;
325 wantfreevnodes = desiredvnodes / 4;
326 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
327 TAILQ_INIT(&vnode_free_list);
328 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
329 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
330 NULL, NULL, UMA_ALIGN_PTR, 0);
331 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
332 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
334 * Initialize the filesystem syncer.
336 syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
338 syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
340 syncer_maxdelay = syncer_mask + 1;
341 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
342 cv_init(&sync_wakeup, "syncer");
344 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
348 * Mark a mount point as busy. Used to synchronize access and to delay
349 * unmounting. Eventually, mountlist_mtx is not released on failure.
351 * vfs_busy() is a custom lock, it can block the caller.
352 * vfs_busy() only sleeps if the unmount is active on the mount point.
353 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
354 * vnode belonging to mp.
356 * Lookup uses vfs_busy() to traverse mount points.
358 * / vnode lock A / vnode lock (/var) D
359 * /var vnode lock B /log vnode lock(/var/log) E
360 * vfs_busy lock C vfs_busy lock F
362 * Within each file system, the lock order is C->A->B and F->D->E.
364 * When traversing across mounts, the system follows that lock order:
370 * The lookup() process for namei("/var") illustrates the process:
371 * VOP_LOOKUP() obtains B while A is held
372 * vfs_busy() obtains a shared lock on F while A and B are held
373 * vput() releases lock on B
374 * vput() releases lock on A
375 * VFS_ROOT() obtains lock on D while shared lock on F is held
376 * vfs_unbusy() releases shared lock on F
377 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
378 * Attempt to lock A (instead of vp_crossmp) while D is held would
379 * violate the global order, causing deadlocks.
381 * dounmount() locks B while F is drained.
384 vfs_busy(struct mount *mp, int flags)
387 MPASS((flags & ~MBF_MASK) == 0);
388 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
393 * If mount point is currenly being unmounted, sleep until the
394 * mount point fate is decided. If thread doing the unmounting fails,
395 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
396 * that this mount point has survived the unmount attempt and vfs_busy
397 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
398 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
399 * about to be really destroyed. vfs_busy needs to release its
400 * reference on the mount point in this case and return with ENOENT,
401 * telling the caller that mount mount it tried to busy is no longer
404 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
405 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
408 CTR1(KTR_VFS, "%s: failed busying before sleeping",
412 if (flags & MBF_MNTLSTLOCK)
413 mtx_unlock(&mountlist_mtx);
414 mp->mnt_kern_flag |= MNTK_MWAIT;
415 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
416 if (flags & MBF_MNTLSTLOCK)
417 mtx_lock(&mountlist_mtx);
420 if (flags & MBF_MNTLSTLOCK)
421 mtx_unlock(&mountlist_mtx);
428 * Free a busy filesystem.
431 vfs_unbusy(struct mount *mp)
434 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
437 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
439 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
440 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
441 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
442 mp->mnt_kern_flag &= ~MNTK_DRAINING;
443 wakeup(&mp->mnt_lockref);
449 * Lookup a mount point by filesystem identifier.
452 vfs_getvfs(fsid_t *fsid)
456 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
457 mtx_lock(&mountlist_mtx);
458 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
459 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
460 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
462 mtx_unlock(&mountlist_mtx);
466 mtx_unlock(&mountlist_mtx);
467 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
468 return ((struct mount *) 0);
472 * Lookup a mount point by filesystem identifier, busying it before
476 vfs_busyfs(fsid_t *fsid)
481 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
482 mtx_lock(&mountlist_mtx);
483 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
484 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
485 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
486 error = vfs_busy(mp, MBF_MNTLSTLOCK);
488 mtx_unlock(&mountlist_mtx);
494 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
495 mtx_unlock(&mountlist_mtx);
496 return ((struct mount *) 0);
500 * Check if a user can access privileged mount options.
503 vfs_suser(struct mount *mp, struct thread *td)
508 * If the thread is jailed, but this is not a jail-friendly file
509 * system, deny immediately.
511 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
515 * If the file system was mounted outside the jail of the calling
516 * thread, deny immediately.
518 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
522 * If file system supports delegated administration, we don't check
523 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
524 * by the file system itself.
525 * If this is not the user that did original mount, we check for
526 * the PRIV_VFS_MOUNT_OWNER privilege.
528 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
529 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
530 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
537 * Get a new unique fsid. Try to make its val[0] unique, since this value
538 * will be used to create fake device numbers for stat(). Also try (but
539 * not so hard) make its val[0] unique mod 2^16, since some emulators only
540 * support 16-bit device numbers. We end up with unique val[0]'s for the
541 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
543 * Keep in mind that several mounts may be running in parallel. Starting
544 * the search one past where the previous search terminated is both a
545 * micro-optimization and a defense against returning the same fsid to
549 vfs_getnewfsid(struct mount *mp)
551 static uint16_t mntid_base;
556 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
557 mtx_lock(&mntid_mtx);
558 mtype = mp->mnt_vfc->vfc_typenum;
559 tfsid.val[1] = mtype;
560 mtype = (mtype & 0xFF) << 24;
562 tfsid.val[0] = makedev(255,
563 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
565 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
569 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
570 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
571 mtx_unlock(&mntid_mtx);
575 * Knob to control the precision of file timestamps:
577 * 0 = seconds only; nanoseconds zeroed.
578 * 1 = seconds and nanoseconds, accurate within 1/HZ.
579 * 2 = seconds and nanoseconds, truncated to microseconds.
580 * >=3 = seconds and nanoseconds, maximum precision.
582 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
584 static int timestamp_precision = TSP_SEC;
585 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
586 ×tamp_precision, 0, "File timestamp precision (0: seconds, "
587 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
588 "3+: sec + ns (max. precision))");
591 * Get a current timestamp.
594 vfs_timestamp(struct timespec *tsp)
598 switch (timestamp_precision) {
600 tsp->tv_sec = time_second;
608 TIMEVAL_TO_TIMESPEC(&tv, tsp);
618 * Set vnode attributes to VNOVAL
621 vattr_null(struct vattr *vap)
625 vap->va_size = VNOVAL;
626 vap->va_bytes = VNOVAL;
627 vap->va_mode = VNOVAL;
628 vap->va_nlink = VNOVAL;
629 vap->va_uid = VNOVAL;
630 vap->va_gid = VNOVAL;
631 vap->va_fsid = VNOVAL;
632 vap->va_fileid = VNOVAL;
633 vap->va_blocksize = VNOVAL;
634 vap->va_rdev = VNOVAL;
635 vap->va_atime.tv_sec = VNOVAL;
636 vap->va_atime.tv_nsec = VNOVAL;
637 vap->va_mtime.tv_sec = VNOVAL;
638 vap->va_mtime.tv_nsec = VNOVAL;
639 vap->va_ctime.tv_sec = VNOVAL;
640 vap->va_ctime.tv_nsec = VNOVAL;
641 vap->va_birthtime.tv_sec = VNOVAL;
642 vap->va_birthtime.tv_nsec = VNOVAL;
643 vap->va_flags = VNOVAL;
644 vap->va_gen = VNOVAL;
649 * This routine is called when we have too many vnodes. It attempts
650 * to free <count> vnodes and will potentially free vnodes that still
651 * have VM backing store (VM backing store is typically the cause
652 * of a vnode blowout so we want to do this). Therefore, this operation
653 * is not considered cheap.
655 * A number of conditions may prevent a vnode from being reclaimed.
656 * the buffer cache may have references on the vnode, a directory
657 * vnode may still have references due to the namei cache representing
658 * underlying files, or the vnode may be in active use. It is not
659 * desireable to reuse such vnodes. These conditions may cause the
660 * number of vnodes to reach some minimum value regardless of what
661 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
664 vlrureclaim(struct mount *mp)
673 * Calculate the trigger point, don't allow user
674 * screwups to blow us up. This prevents us from
675 * recycling vnodes with lots of resident pages. We
676 * aren't trying to free memory, we are trying to
679 usevnodes = desiredvnodes;
682 trigger = cnt.v_page_count * 2 / usevnodes;
684 vn_start_write(NULL, &mp, V_WAIT);
686 count = mp->mnt_nvnodelistsize / 10 + 1;
688 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
689 while (vp != NULL && vp->v_type == VMARKER)
690 vp = TAILQ_NEXT(vp, v_nmntvnodes);
693 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
694 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
699 * If it's been deconstructed already, it's still
700 * referenced, or it exceeds the trigger, skip it.
702 if (vp->v_usecount ||
703 (!vlru_allow_cache_src &&
704 !LIST_EMPTY(&(vp)->v_cache_src)) ||
705 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
706 vp->v_object->resident_page_count > trigger)) {
712 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
714 goto next_iter_mntunlocked;
718 * v_usecount may have been bumped after VOP_LOCK() dropped
719 * the vnode interlock and before it was locked again.
721 * It is not necessary to recheck VI_DOOMED because it can
722 * only be set by another thread that holds both the vnode
723 * lock and vnode interlock. If another thread has the
724 * vnode lock before we get to VOP_LOCK() and obtains the
725 * vnode interlock after VOP_LOCK() drops the vnode
726 * interlock, the other thread will be unable to drop the
727 * vnode lock before our VOP_LOCK() call fails.
729 if (vp->v_usecount ||
730 (!vlru_allow_cache_src &&
731 !LIST_EMPTY(&(vp)->v_cache_src)) ||
732 (vp->v_object != NULL &&
733 vp->v_object->resident_page_count > trigger)) {
734 VOP_UNLOCK(vp, LK_INTERLOCK);
735 goto next_iter_mntunlocked;
737 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
738 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
743 next_iter_mntunlocked:
752 kern_yield(PRI_UNCHANGED);
757 vn_finished_write(mp);
762 * Attempt to keep the free list at wantfreevnodes length.
765 vnlru_free(int count)
770 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
771 for (; count > 0; count--) {
772 vp = TAILQ_FIRST(&vnode_free_list);
774 * The list can be modified while the free_list_mtx
775 * has been dropped and vp could be NULL here.
779 VNASSERT(vp->v_op != NULL, vp,
780 ("vnlru_free: vnode already reclaimed."));
781 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
783 * Don't recycle if we can't get the interlock.
785 if (!VI_TRYLOCK(vp)) {
786 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
789 VNASSERT(VCANRECYCLE(vp), vp,
790 ("vp inconsistent on freelist"));
792 vp->v_iflag &= ~VI_FREE;
794 mtx_unlock(&vnode_free_list_mtx);
796 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
798 VFS_UNLOCK_GIANT(vfslocked);
800 * If the recycled succeeded this vdrop will actually free
801 * the vnode. If not it will simply place it back on
805 mtx_lock(&vnode_free_list_mtx);
809 * Attempt to recycle vnodes in a context that is always safe to block.
810 * Calling vlrurecycle() from the bowels of filesystem code has some
811 * interesting deadlock problems.
813 static struct proc *vnlruproc;
814 static int vnlruproc_sig;
819 struct mount *mp, *nmp;
821 struct proc *p = vnlruproc;
823 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
827 kproc_suspend_check(p);
828 mtx_lock(&vnode_free_list_mtx);
829 if (freevnodes > wantfreevnodes)
830 vnlru_free(freevnodes - wantfreevnodes);
831 if (numvnodes <= desiredvnodes * 9 / 10) {
833 wakeup(&vnlruproc_sig);
834 msleep(vnlruproc, &vnode_free_list_mtx,
835 PVFS|PDROP, "vlruwt", hz);
838 mtx_unlock(&vnode_free_list_mtx);
840 mtx_lock(&mountlist_mtx);
841 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
842 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
843 nmp = TAILQ_NEXT(mp, mnt_list);
846 vfslocked = VFS_LOCK_GIANT(mp);
847 done += vlrureclaim(mp);
848 VFS_UNLOCK_GIANT(vfslocked);
849 mtx_lock(&mountlist_mtx);
850 nmp = TAILQ_NEXT(mp, mnt_list);
853 mtx_unlock(&mountlist_mtx);
856 /* These messages are temporary debugging aids */
857 if (vnlru_nowhere < 5)
858 printf("vnlru process getting nowhere..\n");
859 else if (vnlru_nowhere == 5)
860 printf("vnlru process messages stopped.\n");
863 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
865 kern_yield(PRI_UNCHANGED);
869 static struct kproc_desc vnlru_kp = {
874 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
878 * Routines having to do with the management of the vnode table.
882 vdestroy(struct vnode *vp)
886 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
887 mtx_lock(&vnode_free_list_mtx);
889 mtx_unlock(&vnode_free_list_mtx);
891 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
892 ("cleaned vnode still on the free list."));
893 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
894 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
895 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
896 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
897 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
898 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
899 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
900 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
901 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
902 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
903 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
904 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
907 mac_vnode_destroy(vp);
909 if (vp->v_pollinfo != NULL)
910 destroy_vpollinfo(vp->v_pollinfo);
912 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */
915 lockdestroy(vp->v_vnlock);
916 mtx_destroy(&vp->v_interlock);
917 mtx_destroy(BO_MTX(bo));
918 uma_zfree(vnode_zone, vp);
922 * Try to recycle a freed vnode. We abort if anyone picks up a reference
923 * before we actually vgone(). This function must be called with the vnode
924 * held to prevent the vnode from being returned to the free list midway
928 vtryrecycle(struct vnode *vp)
932 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
933 VNASSERT(vp->v_holdcnt, vp,
934 ("vtryrecycle: Recycling vp %p without a reference.", vp));
936 * This vnode may found and locked via some other list, if so we
937 * can't recycle it yet.
939 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
941 "%s: impossible to recycle, vp %p lock is already held",
943 return (EWOULDBLOCK);
946 * Don't recycle if its filesystem is being suspended.
948 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
951 "%s: impossible to recycle, cannot start the write for %p",
956 * If we got this far, we need to acquire the interlock and see if
957 * anyone picked up this vnode from another list. If not, we will
958 * mark it with DOOMED via vgonel() so that anyone who does find it
962 if (vp->v_usecount) {
963 VOP_UNLOCK(vp, LK_INTERLOCK);
964 vn_finished_write(vnmp);
966 "%s: impossible to recycle, %p is already referenced",
970 if ((vp->v_iflag & VI_DOOMED) == 0)
972 VOP_UNLOCK(vp, LK_INTERLOCK);
973 vn_finished_write(vnmp);
978 * Return the next vnode from the free list.
981 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
984 struct vnode *vp = NULL;
987 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
988 mtx_lock(&vnode_free_list_mtx);
990 * Lend our context to reclaim vnodes if they've exceeded the max.
992 if (freevnodes > wantfreevnodes)
995 * Wait for available vnodes.
997 if (numvnodes > desiredvnodes) {
998 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
1000 * File system is beeing suspended, we cannot risk a
1001 * deadlock here, so allocate new vnode anyway.
1003 if (freevnodes > wantfreevnodes)
1004 vnlru_free(freevnodes - wantfreevnodes);
1007 if (vnlruproc_sig == 0) {
1008 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
1011 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
1013 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1014 if (numvnodes > desiredvnodes) {
1015 mtx_unlock(&vnode_free_list_mtx);
1022 mtx_unlock(&vnode_free_list_mtx);
1023 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1027 vp->v_vnlock = &vp->v_lock;
1028 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1030 * By default, don't allow shared locks unless filesystems
1033 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
1035 * Initialize bufobj.
1038 bo->__bo_vnode = vp;
1039 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
1040 bo->bo_ops = &buf_ops_bio;
1041 bo->bo_private = vp;
1042 TAILQ_INIT(&bo->bo_clean.bv_hd);
1043 TAILQ_INIT(&bo->bo_dirty.bv_hd);
1045 * Initialize namecache.
1047 LIST_INIT(&vp->v_cache_src);
1048 TAILQ_INIT(&vp->v_cache_dst);
1050 * Finalize various vnode identity bits.
1055 v_incr_usecount(vp);
1059 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1060 mac_vnode_associate_singlelabel(mp, vp);
1061 else if (mp == NULL && vops != &dead_vnodeops)
1062 printf("NULL mp in getnewvnode()\n");
1065 bo->bo_bsize = mp->mnt_stat.f_iosize;
1066 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1067 vp->v_vflag |= VV_NOKNOTE;
1075 * Delete from old mount point vnode list, if on one.
1078 delmntque(struct vnode *vp)
1087 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1088 ("bad mount point vnode list size"));
1089 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1090 mp->mnt_nvnodelistsize--;
1096 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1100 vp->v_op = &dead_vnodeops;
1101 /* XXX non mp-safe fs may still call insmntque with vnode
1103 if (!VOP_ISLOCKED(vp))
1104 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1110 * Insert into list of vnodes for the new mount point, if available.
1113 insmntque1(struct vnode *vp, struct mount *mp,
1114 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1118 KASSERT(vp->v_mount == NULL,
1119 ("insmntque: vnode already on per mount vnode list"));
1120 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1121 #ifdef DEBUG_VFS_LOCKS
1122 if (!VFS_NEEDSGIANT(mp))
1123 ASSERT_VOP_ELOCKED(vp,
1124 "insmntque: mp-safe fs and non-locked vp");
1127 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1128 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1129 mp->mnt_nvnodelistsize == 0)) {
1130 locked = VOP_ISLOCKED(vp);
1131 if (!locked || (locked == LK_EXCLUSIVE &&
1132 (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1141 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1142 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1143 ("neg mount point vnode list size"));
1144 mp->mnt_nvnodelistsize++;
1150 insmntque(struct vnode *vp, struct mount *mp)
1153 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1157 * Flush out and invalidate all buffers associated with a bufobj
1158 * Called with the underlying object locked.
1161 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1166 if (flags & V_SAVE) {
1167 error = bufobj_wwait(bo, slpflag, slptimeo);
1172 if (bo->bo_dirty.bv_cnt > 0) {
1174 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1177 * XXX We could save a lock/unlock if this was only
1178 * enabled under INVARIANTS
1181 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1182 panic("vinvalbuf: dirty bufs");
1186 * If you alter this loop please notice that interlock is dropped and
1187 * reacquired in flushbuflist. Special care is needed to ensure that
1188 * no race conditions occur from this.
1191 error = flushbuflist(&bo->bo_clean,
1192 flags, bo, slpflag, slptimeo);
1193 if (error == 0 && !(flags & V_CLEANONLY))
1194 error = flushbuflist(&bo->bo_dirty,
1195 flags, bo, slpflag, slptimeo);
1196 if (error != 0 && error != EAGAIN) {
1200 } while (error != 0);
1203 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1204 * have write I/O in-progress but if there is a VM object then the
1205 * VM object can also have read-I/O in-progress.
1208 bufobj_wwait(bo, 0, 0);
1210 if (bo->bo_object != NULL) {
1211 VM_OBJECT_LOCK(bo->bo_object);
1212 vm_object_pip_wait(bo->bo_object, "bovlbx");
1213 VM_OBJECT_UNLOCK(bo->bo_object);
1216 } while (bo->bo_numoutput > 0);
1220 * Destroy the copy in the VM cache, too.
1222 if (bo->bo_object != NULL &&
1223 (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1224 VM_OBJECT_LOCK(bo->bo_object);
1225 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1226 OBJPR_CLEANONLY : 0);
1227 VM_OBJECT_UNLOCK(bo->bo_object);
1232 if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1233 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1234 panic("vinvalbuf: flush failed");
1241 * Flush out and invalidate all buffers associated with a vnode.
1242 * Called with the underlying object locked.
1245 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1248 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1249 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1250 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1254 * Flush out buffers on the specified list.
1258 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1261 struct buf *bp, *nbp;
1266 ASSERT_BO_LOCKED(bo);
1269 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1270 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1271 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1277 lblkno = nbp->b_lblkno;
1278 xflags = nbp->b_xflags &
1279 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1282 error = BUF_TIMELOCK(bp,
1283 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1284 "flushbuf", slpflag, slptimeo);
1287 return (error != ENOLCK ? error : EAGAIN);
1289 KASSERT(bp->b_bufobj == bo,
1290 ("bp %p wrong b_bufobj %p should be %p",
1291 bp, bp->b_bufobj, bo));
1292 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1298 * XXX Since there are no node locks for NFS, I
1299 * believe there is a slight chance that a delayed
1300 * write will occur while sleeping just above, so
1303 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1308 bp->b_flags |= B_ASYNC;
1311 return (EAGAIN); /* XXX: why not loop ? */
1316 bp->b_flags |= (B_INVAL | B_RELBUF);
1317 bp->b_flags &= ~B_ASYNC;
1321 (nbp->b_bufobj != bo ||
1322 nbp->b_lblkno != lblkno ||
1324 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1325 break; /* nbp invalid */
1331 * Truncate a file's buffer and pages to a specified length. This
1332 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1336 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1337 off_t length, int blksize)
1339 struct buf *bp, *nbp;
1344 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1345 vp, cred, blksize, (uintmax_t)length);
1348 * Round up to the *next* lbn.
1350 trunclbn = (length + blksize - 1) / blksize;
1352 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1359 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1360 if (bp->b_lblkno < trunclbn)
1363 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1364 BO_MTX(bo)) == ENOLCK)
1370 bp->b_flags |= (B_INVAL | B_RELBUF);
1371 bp->b_flags &= ~B_ASYNC;
1377 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1378 (nbp->b_vp != vp) ||
1379 (nbp->b_flags & B_DELWRI))) {
1385 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1386 if (bp->b_lblkno < trunclbn)
1389 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1390 BO_MTX(bo)) == ENOLCK)
1395 bp->b_flags |= (B_INVAL | B_RELBUF);
1396 bp->b_flags &= ~B_ASYNC;
1402 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1403 (nbp->b_vp != vp) ||
1404 (nbp->b_flags & B_DELWRI) == 0)) {
1413 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1414 if (bp->b_lblkno > 0)
1417 * Since we hold the vnode lock this should only
1418 * fail if we're racing with the buf daemon.
1421 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1422 BO_MTX(bo)) == ENOLCK) {
1425 VNASSERT((bp->b_flags & B_DELWRI), vp,
1426 ("buf(%p) on dirty queue without DELWRI", bp));
1437 bufobj_wwait(bo, 0, 0);
1439 vnode_pager_setsize(vp, length);
1445 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1448 * NOTE: We have to deal with the special case of a background bitmap
1449 * buffer, a situation where two buffers will have the same logical
1450 * block offset. We want (1) only the foreground buffer to be accessed
1451 * in a lookup and (2) must differentiate between the foreground and
1452 * background buffer in the splay tree algorithm because the splay
1453 * tree cannot normally handle multiple entities with the same 'index'.
1454 * We accomplish this by adding differentiating flags to the splay tree's
1459 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1462 struct buf *lefttreemax, *righttreemin, *y;
1466 lefttreemax = righttreemin = &dummy;
1468 if (lblkno < root->b_lblkno ||
1469 (lblkno == root->b_lblkno &&
1470 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1471 if ((y = root->b_left) == NULL)
1473 if (lblkno < y->b_lblkno) {
1475 root->b_left = y->b_right;
1478 if ((y = root->b_left) == NULL)
1481 /* Link into the new root's right tree. */
1482 righttreemin->b_left = root;
1483 righttreemin = root;
1484 } else if (lblkno > root->b_lblkno ||
1485 (lblkno == root->b_lblkno &&
1486 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1487 if ((y = root->b_right) == NULL)
1489 if (lblkno > y->b_lblkno) {
1491 root->b_right = y->b_left;
1494 if ((y = root->b_right) == NULL)
1497 /* Link into the new root's left tree. */
1498 lefttreemax->b_right = root;
1505 /* Assemble the new root. */
1506 lefttreemax->b_right = root->b_left;
1507 righttreemin->b_left = root->b_right;
1508 root->b_left = dummy.b_right;
1509 root->b_right = dummy.b_left;
1514 buf_vlist_remove(struct buf *bp)
1519 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1520 ASSERT_BO_LOCKED(bp->b_bufobj);
1521 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1522 (BX_VNDIRTY|BX_VNCLEAN),
1523 ("buf_vlist_remove: Buf %p is on two lists", bp));
1524 if (bp->b_xflags & BX_VNDIRTY)
1525 bv = &bp->b_bufobj->bo_dirty;
1527 bv = &bp->b_bufobj->bo_clean;
1528 if (bp != bv->bv_root) {
1529 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1530 KASSERT(root == bp, ("splay lookup failed in remove"));
1532 if (bp->b_left == NULL) {
1535 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1536 root->b_right = bp->b_right;
1539 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1541 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1545 * Add the buffer to the sorted clean or dirty block list using a
1546 * splay tree algorithm.
1548 * NOTE: xflags is passed as a constant, optimizing this inline function!
1551 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1556 ASSERT_BO_LOCKED(bo);
1557 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1558 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1559 bp->b_xflags |= xflags;
1560 if (xflags & BX_VNDIRTY)
1565 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1569 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1570 } else if (bp->b_lblkno < root->b_lblkno ||
1571 (bp->b_lblkno == root->b_lblkno &&
1572 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1573 bp->b_left = root->b_left;
1575 root->b_left = NULL;
1576 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1578 bp->b_right = root->b_right;
1580 root->b_right = NULL;
1581 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1588 * Lookup a buffer using the splay tree. Note that we specifically avoid
1589 * shadow buffers used in background bitmap writes.
1591 * This code isn't quite efficient as it could be because we are maintaining
1592 * two sorted lists and do not know which list the block resides in.
1594 * During a "make buildworld" the desired buffer is found at one of
1595 * the roots more than 60% of the time. Thus, checking both roots
1596 * before performing either splay eliminates unnecessary splays on the
1597 * first tree splayed.
1600 gbincore(struct bufobj *bo, daddr_t lblkno)
1604 ASSERT_BO_LOCKED(bo);
1605 if ((bp = bo->bo_clean.bv_root) != NULL &&
1606 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1608 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1609 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1611 if ((bp = bo->bo_clean.bv_root) != NULL) {
1612 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1613 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1616 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1617 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1618 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1625 * Associate a buffer with a vnode.
1628 bgetvp(struct vnode *vp, struct buf *bp)
1633 ASSERT_BO_LOCKED(bo);
1634 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1636 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1637 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1638 ("bgetvp: bp already attached! %p", bp));
1641 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1642 bp->b_flags |= B_NEEDSGIANT;
1646 * Insert onto list for new vnode.
1648 buf_vlist_add(bp, bo, BX_VNCLEAN);
1652 * Disassociate a buffer from a vnode.
1655 brelvp(struct buf *bp)
1660 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1661 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1664 * Delete from old vnode list, if on one.
1666 vp = bp->b_vp; /* XXX */
1669 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1670 buf_vlist_remove(bp);
1672 panic("brelvp: Buffer %p not on queue.", bp);
1673 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1674 bo->bo_flag &= ~BO_ONWORKLST;
1675 mtx_lock(&sync_mtx);
1676 LIST_REMOVE(bo, bo_synclist);
1677 syncer_worklist_len--;
1678 mtx_unlock(&sync_mtx);
1680 bp->b_flags &= ~B_NEEDSGIANT;
1682 bp->b_bufobj = NULL;
1688 * Add an item to the syncer work queue.
1691 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1695 ASSERT_BO_LOCKED(bo);
1697 mtx_lock(&sync_mtx);
1698 if (bo->bo_flag & BO_ONWORKLST)
1699 LIST_REMOVE(bo, bo_synclist);
1701 bo->bo_flag |= BO_ONWORKLST;
1702 syncer_worklist_len++;
1705 if (delay > syncer_maxdelay - 2)
1706 delay = syncer_maxdelay - 2;
1707 slot = (syncer_delayno + delay) & syncer_mask;
1709 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1711 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1713 mtx_unlock(&sync_mtx);
1717 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1721 mtx_lock(&sync_mtx);
1722 len = syncer_worklist_len - sync_vnode_count;
1723 mtx_unlock(&sync_mtx);
1724 error = SYSCTL_OUT(req, &len, sizeof(len));
1728 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1729 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1731 static struct proc *updateproc;
1732 static void sched_sync(void);
1733 static struct kproc_desc up_kp = {
1738 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1741 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1746 *bo = LIST_FIRST(slp);
1749 vp = (*bo)->__bo_vnode; /* XXX */
1750 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1753 * We use vhold in case the vnode does not
1754 * successfully sync. vhold prevents the vnode from
1755 * going away when we unlock the sync_mtx so that
1756 * we can acquire the vnode interlock.
1759 mtx_unlock(&sync_mtx);
1761 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1763 mtx_lock(&sync_mtx);
1764 return (*bo == LIST_FIRST(slp));
1766 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1767 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1769 vn_finished_write(mp);
1771 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1773 * Put us back on the worklist. The worklist
1774 * routine will remove us from our current
1775 * position and then add us back in at a later
1778 vn_syncer_add_to_worklist(*bo, syncdelay);
1782 mtx_lock(&sync_mtx);
1787 * System filesystem synchronizer daemon.
1792 struct synclist *gnext, *next;
1793 struct synclist *gslp, *slp;
1796 struct thread *td = curthread;
1798 int net_worklist_len;
1799 int syncer_final_iter;
1804 syncer_final_iter = 0;
1806 syncer_state = SYNCER_RUNNING;
1807 starttime = time_uptime;
1808 td->td_pflags |= TDP_NORUNNINGBUF;
1810 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1813 mtx_lock(&sync_mtx);
1815 if (syncer_state == SYNCER_FINAL_DELAY &&
1816 syncer_final_iter == 0) {
1817 mtx_unlock(&sync_mtx);
1818 kproc_suspend_check(td->td_proc);
1819 mtx_lock(&sync_mtx);
1821 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1822 if (syncer_state != SYNCER_RUNNING &&
1823 starttime != time_uptime) {
1825 printf("\nSyncing disks, vnodes remaining...");
1828 printf("%d ", net_worklist_len);
1830 starttime = time_uptime;
1833 * Push files whose dirty time has expired. Be careful
1834 * of interrupt race on slp queue.
1836 * Skip over empty worklist slots when shutting down.
1839 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1840 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1841 syncer_delayno += 1;
1842 if (syncer_delayno == syncer_maxdelay)
1844 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1845 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1847 * If the worklist has wrapped since the
1848 * it was emptied of all but syncer vnodes,
1849 * switch to the FINAL_DELAY state and run
1850 * for one more second.
1852 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1853 net_worklist_len == 0 &&
1854 last_work_seen == syncer_delayno) {
1855 syncer_state = SYNCER_FINAL_DELAY;
1856 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1858 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1859 LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1862 * Keep track of the last time there was anything
1863 * on the worklist other than syncer vnodes.
1864 * Return to the SHUTTING_DOWN state if any
1867 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1868 last_work_seen = syncer_delayno;
1869 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1870 syncer_state = SYNCER_SHUTTING_DOWN;
1871 while (!LIST_EMPTY(slp)) {
1872 error = sync_vnode(slp, &bo, td);
1874 LIST_REMOVE(bo, bo_synclist);
1875 LIST_INSERT_HEAD(next, bo, bo_synclist);
1879 if (first_printf == 0)
1880 wdog_kern_pat(WD_LASTVAL);
1883 if (!LIST_EMPTY(gslp)) {
1884 mtx_unlock(&sync_mtx);
1886 mtx_lock(&sync_mtx);
1887 while (!LIST_EMPTY(gslp)) {
1888 error = sync_vnode(gslp, &bo, td);
1890 LIST_REMOVE(bo, bo_synclist);
1891 LIST_INSERT_HEAD(gnext, bo,
1898 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1899 syncer_final_iter--;
1901 * The variable rushjob allows the kernel to speed up the
1902 * processing of the filesystem syncer process. A rushjob
1903 * value of N tells the filesystem syncer to process the next
1904 * N seconds worth of work on its queue ASAP. Currently rushjob
1905 * is used by the soft update code to speed up the filesystem
1906 * syncer process when the incore state is getting so far
1907 * ahead of the disk that the kernel memory pool is being
1908 * threatened with exhaustion.
1915 * Just sleep for a short period of time between
1916 * iterations when shutting down to allow some I/O
1919 * If it has taken us less than a second to process the
1920 * current work, then wait. Otherwise start right over
1921 * again. We can still lose time if any single round
1922 * takes more than two seconds, but it does not really
1923 * matter as we are just trying to generally pace the
1924 * filesystem activity.
1926 if (syncer_state != SYNCER_RUNNING ||
1927 time_uptime == starttime) {
1929 sched_prio(td, PPAUSE);
1932 if (syncer_state != SYNCER_RUNNING)
1933 cv_timedwait(&sync_wakeup, &sync_mtx,
1934 hz / SYNCER_SHUTDOWN_SPEEDUP);
1935 else if (time_uptime == starttime)
1936 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1941 * Request the syncer daemon to speed up its work.
1942 * We never push it to speed up more than half of its
1943 * normal turn time, otherwise it could take over the cpu.
1946 speedup_syncer(void)
1950 mtx_lock(&sync_mtx);
1951 if (rushjob < syncdelay / 2) {
1953 stat_rush_requests += 1;
1956 mtx_unlock(&sync_mtx);
1957 cv_broadcast(&sync_wakeup);
1962 * Tell the syncer to speed up its work and run though its work
1963 * list several times, then tell it to shut down.
1966 syncer_shutdown(void *arg, int howto)
1969 if (howto & RB_NOSYNC)
1971 mtx_lock(&sync_mtx);
1972 syncer_state = SYNCER_SHUTTING_DOWN;
1974 mtx_unlock(&sync_mtx);
1975 cv_broadcast(&sync_wakeup);
1976 kproc_shutdown(arg, howto);
1980 * Reassign a buffer from one vnode to another.
1981 * Used to assign file specific control information
1982 * (indirect blocks) to the vnode to which they belong.
1985 reassignbuf(struct buf *bp)
1998 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1999 bp, bp->b_vp, bp->b_flags);
2001 * B_PAGING flagged buffers cannot be reassigned because their vp
2002 * is not fully linked in.
2004 if (bp->b_flags & B_PAGING)
2005 panic("cannot reassign paging buffer");
2008 * Delete from old vnode list, if on one.
2011 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2012 buf_vlist_remove(bp);
2014 panic("reassignbuf: Buffer %p not on queue.", bp);
2016 * If dirty, put on list of dirty buffers; otherwise insert onto list
2019 if (bp->b_flags & B_DELWRI) {
2020 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2021 switch (vp->v_type) {
2031 vn_syncer_add_to_worklist(bo, delay);
2033 buf_vlist_add(bp, bo, BX_VNDIRTY);
2035 buf_vlist_add(bp, bo, BX_VNCLEAN);
2037 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2038 mtx_lock(&sync_mtx);
2039 LIST_REMOVE(bo, bo_synclist);
2040 syncer_worklist_len--;
2041 mtx_unlock(&sync_mtx);
2042 bo->bo_flag &= ~BO_ONWORKLST;
2047 bp = TAILQ_FIRST(&bv->bv_hd);
2048 KASSERT(bp == NULL || bp->b_bufobj == bo,
2049 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2050 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2051 KASSERT(bp == NULL || bp->b_bufobj == bo,
2052 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2054 bp = TAILQ_FIRST(&bv->bv_hd);
2055 KASSERT(bp == NULL || bp->b_bufobj == bo,
2056 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2057 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2058 KASSERT(bp == NULL || bp->b_bufobj == bo,
2059 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2065 * Increment the use and hold counts on the vnode, taking care to reference
2066 * the driver's usecount if this is a chardev. The vholdl() will remove
2067 * the vnode from the free list if it is presently free. Requires the
2068 * vnode interlock and returns with it held.
2071 v_incr_usecount(struct vnode *vp)
2074 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2076 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2078 vp->v_rdev->si_usecount++;
2085 * Turn a holdcnt into a use+holdcnt such that only one call to
2086 * v_decr_usecount is needed.
2089 v_upgrade_usecount(struct vnode *vp)
2092 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2094 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2096 vp->v_rdev->si_usecount++;
2102 * Decrement the vnode use and hold count along with the driver's usecount
2103 * if this is a chardev. The vdropl() below releases the vnode interlock
2104 * as it may free the vnode.
2107 v_decr_usecount(struct vnode *vp)
2110 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2111 VNASSERT(vp->v_usecount > 0, vp,
2112 ("v_decr_usecount: negative usecount"));
2113 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2115 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2117 vp->v_rdev->si_usecount--;
2124 * Decrement only the use count and driver use count. This is intended to
2125 * be paired with a follow on vdropl() to release the remaining hold count.
2126 * In this way we may vgone() a vnode with a 0 usecount without risk of
2127 * having it end up on a free list because the hold count is kept above 0.
2130 v_decr_useonly(struct vnode *vp)
2133 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2134 VNASSERT(vp->v_usecount > 0, vp,
2135 ("v_decr_useonly: negative usecount"));
2136 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2138 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2140 vp->v_rdev->si_usecount--;
2146 * Grab a particular vnode from the free list, increment its
2147 * reference count and lock it. VI_DOOMED is set if the vnode
2148 * is being destroyed. Only callers who specify LK_RETRY will
2149 * see doomed vnodes. If inactive processing was delayed in
2150 * vput try to do it here.
2153 vget(struct vnode *vp, int flags, struct thread *td)
2158 VFS_ASSERT_GIANT(vp->v_mount);
2159 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2160 ("vget: invalid lock operation"));
2161 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2163 if ((flags & LK_INTERLOCK) == 0)
2166 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2168 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2172 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2173 panic("vget: vn_lock failed to return ENOENT\n");
2175 /* Upgrade our holdcnt to a usecount. */
2176 v_upgrade_usecount(vp);
2178 * We don't guarantee that any particular close will
2179 * trigger inactive processing so just make a best effort
2180 * here at preventing a reference to a removed file. If
2181 * we don't succeed no harm is done.
2183 if (vp->v_iflag & VI_OWEINACT) {
2184 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2185 (flags & LK_NOWAIT) == 0)
2187 vp->v_iflag &= ~VI_OWEINACT;
2194 * Increase the reference count of a vnode.
2197 vref(struct vnode *vp)
2200 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2202 v_incr_usecount(vp);
2207 * Return reference count of a vnode.
2209 * The results of this call are only guaranteed when some mechanism other
2210 * than the VI lock is used to stop other processes from gaining references
2211 * to the vnode. This may be the case if the caller holds the only reference.
2212 * This is also useful when stale data is acceptable as race conditions may
2213 * be accounted for by some other means.
2216 vrefcnt(struct vnode *vp)
2221 usecnt = vp->v_usecount;
2227 #define VPUTX_VRELE 1
2228 #define VPUTX_VPUT 2
2229 #define VPUTX_VUNREF 3
2232 vputx(struct vnode *vp, int func)
2236 KASSERT(vp != NULL, ("vputx: null vp"));
2237 if (func == VPUTX_VUNREF)
2238 ASSERT_VOP_LOCKED(vp, "vunref");
2239 else if (func == VPUTX_VPUT)
2240 ASSERT_VOP_LOCKED(vp, "vput");
2242 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2243 VFS_ASSERT_GIANT(vp->v_mount);
2244 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2247 /* Skip this v_writecount check if we're going to panic below. */
2248 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2249 ("vputx: missed vn_close"));
2252 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2253 vp->v_usecount == 1)) {
2254 if (func == VPUTX_VPUT)
2256 v_decr_usecount(vp);
2260 if (vp->v_usecount != 1) {
2261 vprint("vputx: negative ref count", vp);
2262 panic("vputx: negative ref cnt");
2264 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2266 * We want to hold the vnode until the inactive finishes to
2267 * prevent vgone() races. We drop the use count here and the
2268 * hold count below when we're done.
2272 * We must call VOP_INACTIVE with the node locked. Mark
2273 * as VI_DOINGINACT to avoid recursion.
2275 vp->v_iflag |= VI_OWEINACT;
2278 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2282 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2283 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2289 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2293 if (vp->v_usecount > 0)
2294 vp->v_iflag &= ~VI_OWEINACT;
2296 if (vp->v_iflag & VI_OWEINACT)
2297 vinactive(vp, curthread);
2298 if (func != VPUTX_VUNREF)
2305 * Vnode put/release.
2306 * If count drops to zero, call inactive routine and return to freelist.
2309 vrele(struct vnode *vp)
2312 vputx(vp, VPUTX_VRELE);
2316 * Release an already locked vnode. This give the same effects as
2317 * unlock+vrele(), but takes less time and avoids releasing and
2318 * re-aquiring the lock (as vrele() acquires the lock internally.)
2321 vput(struct vnode *vp)
2324 vputx(vp, VPUTX_VPUT);
2328 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2331 vunref(struct vnode *vp)
2334 vputx(vp, VPUTX_VUNREF);
2338 * Somebody doesn't want the vnode recycled.
2341 vhold(struct vnode *vp)
2350 vholdl(struct vnode *vp)
2353 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2355 if (VSHOULDBUSY(vp))
2360 * Note that there is one less who cares about this vnode. vdrop() is the
2361 * opposite of vhold().
2364 vdrop(struct vnode *vp)
2372 * Drop the hold count of the vnode. If this is the last reference to
2373 * the vnode we will free it if it has been vgone'd otherwise it is
2374 * placed on the free list.
2377 vdropl(struct vnode *vp)
2380 ASSERT_VI_LOCKED(vp, "vdropl");
2381 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2382 if (vp->v_holdcnt <= 0)
2383 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2385 if (vp->v_holdcnt == 0) {
2386 if (vp->v_iflag & VI_DOOMED) {
2387 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__,
2398 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2399 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2400 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2401 * failed lock upgrade.
2404 vinactive(struct vnode *vp, struct thread *td)
2407 ASSERT_VOP_ELOCKED(vp, "vinactive");
2408 ASSERT_VI_LOCKED(vp, "vinactive");
2409 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2410 ("vinactive: recursed on VI_DOINGINACT"));
2411 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2412 vp->v_iflag |= VI_DOINGINACT;
2413 vp->v_iflag &= ~VI_OWEINACT;
2415 VOP_INACTIVE(vp, td);
2417 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2418 ("vinactive: lost VI_DOINGINACT"));
2419 vp->v_iflag &= ~VI_DOINGINACT;
2423 * Remove any vnodes in the vnode table belonging to mount point mp.
2425 * If FORCECLOSE is not specified, there should not be any active ones,
2426 * return error if any are found (nb: this is a user error, not a
2427 * system error). If FORCECLOSE is specified, detach any active vnodes
2430 * If WRITECLOSE is set, only flush out regular file vnodes open for
2433 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2435 * `rootrefs' specifies the base reference count for the root vnode
2436 * of this filesystem. The root vnode is considered busy if its
2437 * v_usecount exceeds this value. On a successful return, vflush(, td)
2438 * will call vrele() on the root vnode exactly rootrefs times.
2439 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2443 static int busyprt = 0; /* print out busy vnodes */
2444 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2448 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2450 struct vnode *vp, *mvp, *rootvp = NULL;
2452 int busy = 0, error;
2454 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2457 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2458 ("vflush: bad args"));
2460 * Get the filesystem root vnode. We can vput() it
2461 * immediately, since with rootrefs > 0, it won't go away.
2463 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2464 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2472 MNT_VNODE_FOREACH(vp, mp, mvp) {
2476 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2480 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
2484 * Skip over a vnodes marked VV_SYSTEM.
2486 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2493 * If WRITECLOSE is set, flush out unlinked but still open
2494 * files (even if open only for reading) and regular file
2495 * vnodes open for writing.
2497 if (flags & WRITECLOSE) {
2498 if (vp->v_object != NULL) {
2499 VM_OBJECT_LOCK(vp->v_object);
2500 vm_object_page_clean(vp->v_object, 0, 0, 0);
2501 VM_OBJECT_UNLOCK(vp->v_object);
2503 error = VOP_FSYNC(vp, MNT_WAIT, td);
2507 MNT_VNODE_FOREACH_ABORT(mp, mvp);
2510 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2513 if ((vp->v_type == VNON ||
2514 (error == 0 && vattr.va_nlink > 0)) &&
2515 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2524 * With v_usecount == 0, all we need to do is clear out the
2525 * vnode data structures and we are done.
2527 * If FORCECLOSE is set, forcibly close the vnode.
2529 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2530 VNASSERT(vp->v_usecount == 0 ||
2531 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2532 ("device VNODE %p is FORCECLOSED", vp));
2538 vprint("vflush: busy vnode", vp);
2546 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2548 * If just the root vnode is busy, and if its refcount
2549 * is equal to `rootrefs', then go ahead and kill it.
2552 KASSERT(busy > 0, ("vflush: not busy"));
2553 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2554 ("vflush: usecount %d < rootrefs %d",
2555 rootvp->v_usecount, rootrefs));
2556 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2557 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2559 VOP_UNLOCK(rootvp, 0);
2565 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2569 for (; rootrefs > 0; rootrefs--)
2575 * Recycle an unused vnode to the front of the free list.
2578 vrecycle(struct vnode *vp, struct thread *td)
2582 ASSERT_VOP_ELOCKED(vp, "vrecycle");
2583 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2586 if (vp->v_usecount == 0) {
2595 * Eliminate all activity associated with a vnode
2596 * in preparation for reuse.
2599 vgone(struct vnode *vp)
2607 * vgone, with the vp interlock held.
2610 vgonel(struct vnode *vp)
2617 ASSERT_VOP_ELOCKED(vp, "vgonel");
2618 ASSERT_VI_LOCKED(vp, "vgonel");
2619 VNASSERT(vp->v_holdcnt, vp,
2620 ("vgonel: vp %p has no reference.", vp));
2621 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2625 * Don't vgonel if we're already doomed.
2627 if (vp->v_iflag & VI_DOOMED)
2629 vp->v_iflag |= VI_DOOMED;
2631 * Check to see if the vnode is in use. If so, we have to call
2632 * VOP_CLOSE() and VOP_INACTIVE().
2634 active = vp->v_usecount;
2635 oweinact = (vp->v_iflag & VI_OWEINACT);
2638 * Clean out any buffers associated with the vnode.
2639 * If the flush fails, just toss the buffers.
2642 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2643 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2644 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2645 vinvalbuf(vp, 0, 0, 0);
2648 * If purging an active vnode, it must be closed and
2649 * deactivated before being reclaimed.
2652 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2653 if (oweinact || active) {
2655 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2659 if (vp->v_type == VSOCK)
2660 vfs_unp_reclaim(vp);
2662 * Reclaim the vnode.
2664 if (VOP_RECLAIM(vp, td))
2665 panic("vgone: cannot reclaim");
2667 vn_finished_secondary_write(mp);
2668 VNASSERT(vp->v_object == NULL, vp,
2669 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2671 * Clear the advisory locks and wake up waiting threads.
2673 (void)VOP_ADVLOCKPURGE(vp);
2675 * Delete from old mount point vnode list.
2680 * Done with purge, reset to the standard lock and invalidate
2684 vp->v_vnlock = &vp->v_lock;
2685 vp->v_op = &dead_vnodeops;
2691 * Calculate the total number of references to a special device.
2694 vcount(struct vnode *vp)
2699 count = vp->v_rdev->si_usecount;
2705 * Same as above, but using the struct cdev *as argument
2708 count_dev(struct cdev *dev)
2713 count = dev->si_usecount;
2719 * Print out a description of a vnode.
2721 static char *typename[] =
2722 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2726 vn_printf(struct vnode *vp, const char *fmt, ...)
2729 char buf[256], buf2[16];
2735 printf("%p: ", (void *)vp);
2736 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2737 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2738 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2741 if (vp->v_vflag & VV_ROOT)
2742 strlcat(buf, "|VV_ROOT", sizeof(buf));
2743 if (vp->v_vflag & VV_ISTTY)
2744 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2745 if (vp->v_vflag & VV_NOSYNC)
2746 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2747 if (vp->v_vflag & VV_CACHEDLABEL)
2748 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2749 if (vp->v_vflag & VV_TEXT)
2750 strlcat(buf, "|VV_TEXT", sizeof(buf));
2751 if (vp->v_vflag & VV_COPYONWRITE)
2752 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2753 if (vp->v_vflag & VV_SYSTEM)
2754 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2755 if (vp->v_vflag & VV_PROCDEP)
2756 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2757 if (vp->v_vflag & VV_NOKNOTE)
2758 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2759 if (vp->v_vflag & VV_DELETED)
2760 strlcat(buf, "|VV_DELETED", sizeof(buf));
2761 if (vp->v_vflag & VV_MD)
2762 strlcat(buf, "|VV_MD", sizeof(buf));
2763 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2764 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2765 VV_NOKNOTE | VV_DELETED | VV_MD);
2767 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2768 strlcat(buf, buf2, sizeof(buf));
2770 if (vp->v_iflag & VI_MOUNT)
2771 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2772 if (vp->v_iflag & VI_AGE)
2773 strlcat(buf, "|VI_AGE", sizeof(buf));
2774 if (vp->v_iflag & VI_DOOMED)
2775 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2776 if (vp->v_iflag & VI_FREE)
2777 strlcat(buf, "|VI_FREE", sizeof(buf));
2778 if (vp->v_iflag & VI_DOINGINACT)
2779 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2780 if (vp->v_iflag & VI_OWEINACT)
2781 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2782 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2783 VI_DOINGINACT | VI_OWEINACT);
2785 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2786 strlcat(buf, buf2, sizeof(buf));
2788 printf(" flags (%s)\n", buf + 1);
2789 if (mtx_owned(VI_MTX(vp)))
2790 printf(" VI_LOCKed");
2791 if (vp->v_object != NULL)
2792 printf(" v_object %p ref %d pages %d\n",
2793 vp->v_object, vp->v_object->ref_count,
2794 vp->v_object->resident_page_count);
2796 lockmgr_printinfo(vp->v_vnlock);
2797 if (vp->v_data != NULL)
2803 * List all of the locked vnodes in the system.
2804 * Called when debugging the kernel.
2806 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2808 struct mount *mp, *nmp;
2812 * Note: because this is DDB, we can't obey the locking semantics
2813 * for these structures, which means we could catch an inconsistent
2814 * state and dereference a nasty pointer. Not much to be done
2817 db_printf("Locked vnodes\n");
2818 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2819 nmp = TAILQ_NEXT(mp, mnt_list);
2820 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2821 if (vp->v_type != VMARKER &&
2825 nmp = TAILQ_NEXT(mp, mnt_list);
2830 * Show details about the given vnode.
2832 DB_SHOW_COMMAND(vnode, db_show_vnode)
2838 vp = (struct vnode *)addr;
2839 vn_printf(vp, "vnode ");
2843 * Show details about the given mount point.
2845 DB_SHOW_COMMAND(mount, db_show_mount)
2856 /* No address given, print short info about all mount points. */
2857 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2858 db_printf("%p %s on %s (%s)\n", mp,
2859 mp->mnt_stat.f_mntfromname,
2860 mp->mnt_stat.f_mntonname,
2861 mp->mnt_stat.f_fstypename);
2865 db_printf("\nMore info: show mount <addr>\n");
2869 mp = (struct mount *)addr;
2870 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2871 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2874 mflags = mp->mnt_flag;
2875 #define MNT_FLAG(flag) do { \
2876 if (mflags & (flag)) { \
2877 if (buf[0] != '\0') \
2878 strlcat(buf, ", ", sizeof(buf)); \
2879 strlcat(buf, (#flag) + 4, sizeof(buf)); \
2880 mflags &= ~(flag); \
2883 MNT_FLAG(MNT_RDONLY);
2884 MNT_FLAG(MNT_SYNCHRONOUS);
2885 MNT_FLAG(MNT_NOEXEC);
2886 MNT_FLAG(MNT_NOSUID);
2887 MNT_FLAG(MNT_UNION);
2888 MNT_FLAG(MNT_ASYNC);
2889 MNT_FLAG(MNT_SUIDDIR);
2890 MNT_FLAG(MNT_SOFTDEP);
2892 MNT_FLAG(MNT_NOSYMFOLLOW);
2893 MNT_FLAG(MNT_GJOURNAL);
2894 MNT_FLAG(MNT_MULTILABEL);
2896 MNT_FLAG(MNT_NOATIME);
2897 MNT_FLAG(MNT_NOCLUSTERR);
2898 MNT_FLAG(MNT_NOCLUSTERW);
2899 MNT_FLAG(MNT_NFS4ACLS);
2900 MNT_FLAG(MNT_EXRDONLY);
2901 MNT_FLAG(MNT_EXPORTED);
2902 MNT_FLAG(MNT_DEFEXPORTED);
2903 MNT_FLAG(MNT_EXPORTANON);
2904 MNT_FLAG(MNT_EXKERB);
2905 MNT_FLAG(MNT_EXPUBLIC);
2906 MNT_FLAG(MNT_LOCAL);
2907 MNT_FLAG(MNT_QUOTA);
2908 MNT_FLAG(MNT_ROOTFS);
2910 MNT_FLAG(MNT_IGNORE);
2911 MNT_FLAG(MNT_UPDATE);
2912 MNT_FLAG(MNT_DELEXPORT);
2913 MNT_FLAG(MNT_RELOAD);
2914 MNT_FLAG(MNT_FORCE);
2915 MNT_FLAG(MNT_SNAPSHOT);
2916 MNT_FLAG(MNT_BYFSID);
2920 strlcat(buf, ", ", sizeof(buf));
2921 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2922 "0x%016jx", mflags);
2924 db_printf(" mnt_flag = %s\n", buf);
2927 flags = mp->mnt_kern_flag;
2928 #define MNT_KERN_FLAG(flag) do { \
2929 if (flags & (flag)) { \
2930 if (buf[0] != '\0') \
2931 strlcat(buf, ", ", sizeof(buf)); \
2932 strlcat(buf, (#flag) + 5, sizeof(buf)); \
2936 MNT_KERN_FLAG(MNTK_UNMOUNTF);
2937 MNT_KERN_FLAG(MNTK_ASYNC);
2938 MNT_KERN_FLAG(MNTK_SOFTDEP);
2939 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
2940 MNT_KERN_FLAG(MNTK_DRAINING);
2941 MNT_KERN_FLAG(MNTK_REFEXPIRE);
2942 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
2943 MNT_KERN_FLAG(MNTK_SHARED_WRITES);
2944 MNT_KERN_FLAG(MNTK_NOASYNC);
2945 MNT_KERN_FLAG(MNTK_UNMOUNT);
2946 MNT_KERN_FLAG(MNTK_MWAIT);
2947 MNT_KERN_FLAG(MNTK_SUSPEND);
2948 MNT_KERN_FLAG(MNTK_SUSPEND2);
2949 MNT_KERN_FLAG(MNTK_SUSPENDED);
2950 MNT_KERN_FLAG(MNTK_MPSAFE);
2951 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
2952 MNT_KERN_FLAG(MNTK_NOKNOTE);
2953 #undef MNT_KERN_FLAG
2956 strlcat(buf, ", ", sizeof(buf));
2957 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
2960 db_printf(" mnt_kern_flag = %s\n", buf);
2962 db_printf(" mnt_opt = ");
2963 opt = TAILQ_FIRST(mp->mnt_opt);
2965 db_printf("%s", opt->name);
2966 opt = TAILQ_NEXT(opt, link);
2967 while (opt != NULL) {
2968 db_printf(", %s", opt->name);
2969 opt = TAILQ_NEXT(opt, link);
2975 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
2976 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
2977 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
2978 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
2979 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
2980 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
2981 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
2982 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
2983 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
2984 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
2985 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
2986 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
2988 db_printf(" mnt_cred = { uid=%u ruid=%u",
2989 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
2990 if (jailed(mp->mnt_cred))
2991 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
2993 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
2994 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
2995 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
2996 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
2997 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
2998 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
2999 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
3000 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3001 db_printf(" mnt_secondary_accwrites = %d\n",
3002 mp->mnt_secondary_accwrites);
3003 db_printf(" mnt_gjprovider = %s\n",
3004 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3007 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3008 if (vp->v_type != VMARKER) {
3009 vn_printf(vp, "vnode ");
3018 * Fill in a struct xvfsconf based on a struct vfsconf.
3021 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
3024 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
3025 xvfsp->vfc_typenum = vfsp->vfc_typenum;
3026 xvfsp->vfc_refcount = vfsp->vfc_refcount;
3027 xvfsp->vfc_flags = vfsp->vfc_flags;
3029 * These are unused in userland, we keep them
3030 * to not break binary compatibility.
3032 xvfsp->vfc_vfsops = NULL;
3033 xvfsp->vfc_next = NULL;
3037 * Top level filesystem related information gathering.
3040 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3042 struct vfsconf *vfsp;
3043 struct xvfsconf xvfsp;
3047 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3048 bzero(&xvfsp, sizeof(xvfsp));
3049 vfsconf2x(vfsp, &xvfsp);
3050 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
3057 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3058 NULL, 0, sysctl_vfs_conflist,
3059 "S,xvfsconf", "List of all configured filesystems");
3061 #ifndef BURN_BRIDGES
3062 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3065 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3067 int *name = (int *)arg1 - 1; /* XXX */
3068 u_int namelen = arg2 + 1; /* XXX */
3069 struct vfsconf *vfsp;
3070 struct xvfsconf xvfsp;
3072 printf("WARNING: userland calling deprecated sysctl, "
3073 "please rebuild world\n");
3075 #if 1 || defined(COMPAT_PRELITE2)
3076 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3078 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3082 case VFS_MAXTYPENUM:
3085 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3088 return (ENOTDIR); /* overloaded */
3089 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3090 if (vfsp->vfc_typenum == name[2])
3093 return (EOPNOTSUPP);
3094 bzero(&xvfsp, sizeof(xvfsp));
3095 vfsconf2x(vfsp, &xvfsp);
3096 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3098 return (EOPNOTSUPP);
3101 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3102 vfs_sysctl, "Generic filesystem");
3104 #if 1 || defined(COMPAT_PRELITE2)
3107 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3110 struct vfsconf *vfsp;
3111 struct ovfsconf ovfs;
3113 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3114 bzero(&ovfs, sizeof(ovfs));
3115 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3116 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3117 ovfs.vfc_index = vfsp->vfc_typenum;
3118 ovfs.vfc_refcount = vfsp->vfc_refcount;
3119 ovfs.vfc_flags = vfsp->vfc_flags;
3120 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3127 #endif /* 1 || COMPAT_PRELITE2 */
3128 #endif /* !BURN_BRIDGES */
3130 #define KINFO_VNODESLOP 10
3133 * Dump vnode list (via sysctl).
3137 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3145 * Stale numvnodes access is not fatal here.
3148 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3150 /* Make an estimate */
3151 return (SYSCTL_OUT(req, 0, len));
3153 error = sysctl_wire_old_buffer(req, 0);
3156 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3158 mtx_lock(&mountlist_mtx);
3159 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3160 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3163 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3167 xvn[n].xv_size = sizeof *xvn;
3168 xvn[n].xv_vnode = vp;
3169 xvn[n].xv_id = 0; /* XXX compat */
3170 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3172 XV_COPY(writecount);
3178 xvn[n].xv_flag = vp->v_vflag;
3180 switch (vp->v_type) {
3187 if (vp->v_rdev == NULL) {
3191 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3194 xvn[n].xv_socket = vp->v_socket;
3197 xvn[n].xv_fifo = vp->v_fifoinfo;
3202 /* shouldn't happen? */
3210 mtx_lock(&mountlist_mtx);
3215 mtx_unlock(&mountlist_mtx);
3217 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3222 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3223 0, 0, sysctl_vnode, "S,xvnode", "");
3227 * Unmount all filesystems. The list is traversed in reverse order
3228 * of mounting to avoid dependencies.
3231 vfs_unmountall(void)
3237 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3238 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3242 * Since this only runs when rebooting, it is not interlocked.
3244 while(!TAILQ_EMPTY(&mountlist)) {
3245 mp = TAILQ_LAST(&mountlist, mntlist);
3246 error = dounmount(mp, MNT_FORCE, td);
3248 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3250 * XXX: Due to the way in which we mount the root
3251 * file system off of devfs, devfs will generate a
3252 * "busy" warning when we try to unmount it before
3253 * the root. Don't print a warning as a result in
3254 * order to avoid false positive errors that may
3255 * cause needless upset.
3257 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3258 printf("unmount of %s failed (",
3259 mp->mnt_stat.f_mntonname);
3263 printf("%d)\n", error);
3266 /* The unmount has removed mp from the mountlist */
3272 * perform msync on all vnodes under a mount point
3273 * the mount point must be locked.
3276 vfs_msync(struct mount *mp, int flags)
3278 struct vnode *vp, *mvp;
3279 struct vm_object *obj;
3281 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3283 MNT_VNODE_FOREACH(vp, mp, mvp) {
3286 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3287 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3290 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3292 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3300 VM_OBJECT_LOCK(obj);
3301 vm_object_page_clean(obj, 0, 0,
3303 OBJPC_SYNC : OBJPC_NOSYNC);
3304 VM_OBJECT_UNLOCK(obj);
3316 * Mark a vnode as free, putting it up for recycling.
3319 vfree(struct vnode *vp)
3322 ASSERT_VI_LOCKED(vp, "vfree");
3323 mtx_lock(&vnode_free_list_mtx);
3324 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed."));
3325 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free"));
3326 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't"));
3327 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp,
3328 ("vfree: Freeing doomed vnode"));
3329 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3330 if (vp->v_iflag & VI_AGE) {
3331 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
3333 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
3336 vp->v_iflag &= ~VI_AGE;
3337 vp->v_iflag |= VI_FREE;
3338 mtx_unlock(&vnode_free_list_mtx);
3342 * Opposite of vfree() - mark a vnode as in use.
3345 vbusy(struct vnode *vp)
3347 ASSERT_VI_LOCKED(vp, "vbusy");
3348 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
3349 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed."));
3350 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3352 mtx_lock(&vnode_free_list_mtx);
3353 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
3355 vp->v_iflag &= ~(VI_FREE|VI_AGE);
3356 mtx_unlock(&vnode_free_list_mtx);
3360 destroy_vpollinfo(struct vpollinfo *vi)
3362 seldrain(&vi->vpi_selinfo);
3363 knlist_destroy(&vi->vpi_selinfo.si_note);
3364 mtx_destroy(&vi->vpi_lock);
3365 uma_zfree(vnodepoll_zone, vi);
3369 * Initalize per-vnode helper structure to hold poll-related state.
3372 v_addpollinfo(struct vnode *vp)
3374 struct vpollinfo *vi;
3376 if (vp->v_pollinfo != NULL)
3378 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3379 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3380 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3381 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3383 if (vp->v_pollinfo != NULL) {
3385 destroy_vpollinfo(vi);
3388 vp->v_pollinfo = vi;
3393 * Record a process's interest in events which might happen to
3394 * a vnode. Because poll uses the historic select-style interface
3395 * internally, this routine serves as both the ``check for any
3396 * pending events'' and the ``record my interest in future events''
3397 * functions. (These are done together, while the lock is held,
3398 * to avoid race conditions.)
3401 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3405 mtx_lock(&vp->v_pollinfo->vpi_lock);
3406 if (vp->v_pollinfo->vpi_revents & events) {
3408 * This leaves events we are not interested
3409 * in available for the other process which
3410 * which presumably had requested them
3411 * (otherwise they would never have been
3414 events &= vp->v_pollinfo->vpi_revents;
3415 vp->v_pollinfo->vpi_revents &= ~events;
3417 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3420 vp->v_pollinfo->vpi_events |= events;
3421 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3422 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3427 * Routine to create and manage a filesystem syncer vnode.
3429 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3430 static int sync_fsync(struct vop_fsync_args *);
3431 static int sync_inactive(struct vop_inactive_args *);
3432 static int sync_reclaim(struct vop_reclaim_args *);
3434 static struct vop_vector sync_vnodeops = {
3435 .vop_bypass = VOP_EOPNOTSUPP,
3436 .vop_close = sync_close, /* close */
3437 .vop_fsync = sync_fsync, /* fsync */
3438 .vop_inactive = sync_inactive, /* inactive */
3439 .vop_reclaim = sync_reclaim, /* reclaim */
3440 .vop_lock1 = vop_stdlock, /* lock */
3441 .vop_unlock = vop_stdunlock, /* unlock */
3442 .vop_islocked = vop_stdislocked, /* islocked */
3446 * Create a new filesystem syncer vnode for the specified mount point.
3449 vfs_allocate_syncvnode(struct mount *mp)
3453 static long start, incr, next;
3456 /* Allocate a new vnode */
3457 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3459 panic("vfs_allocate_syncvnode: getnewvnode() failed");
3461 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3462 vp->v_vflag |= VV_FORCEINSMQ;
3463 error = insmntque(vp, mp);
3465 panic("vfs_allocate_syncvnode: insmntque() failed");
3466 vp->v_vflag &= ~VV_FORCEINSMQ;
3469 * Place the vnode onto the syncer worklist. We attempt to
3470 * scatter them about on the list so that they will go off
3471 * at evenly distributed times even if all the filesystems
3472 * are mounted at once.
3475 if (next == 0 || next > syncer_maxdelay) {
3479 start = syncer_maxdelay / 2;
3480 incr = syncer_maxdelay;
3486 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3487 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3488 mtx_lock(&sync_mtx);
3490 if (mp->mnt_syncer == NULL) {
3491 mp->mnt_syncer = vp;
3494 mtx_unlock(&sync_mtx);
3497 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3504 vfs_deallocate_syncvnode(struct mount *mp)
3508 mtx_lock(&sync_mtx);
3509 vp = mp->mnt_syncer;
3511 mp->mnt_syncer = NULL;
3512 mtx_unlock(&sync_mtx);
3518 * Do a lazy sync of the filesystem.
3521 sync_fsync(struct vop_fsync_args *ap)
3523 struct vnode *syncvp = ap->a_vp;
3524 struct mount *mp = syncvp->v_mount;
3529 * We only need to do something if this is a lazy evaluation.
3531 if (ap->a_waitfor != MNT_LAZY)
3535 * Move ourselves to the back of the sync list.
3537 bo = &syncvp->v_bufobj;
3539 vn_syncer_add_to_worklist(bo, syncdelay);
3543 * Walk the list of vnodes pushing all that are dirty and
3544 * not already on the sync list.
3546 mtx_lock(&mountlist_mtx);
3547 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3548 mtx_unlock(&mountlist_mtx);
3551 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3555 save = curthread_pflags_set(TDP_SYNCIO);
3556 vfs_msync(mp, MNT_NOWAIT);
3557 error = VFS_SYNC(mp, MNT_LAZY);
3558 curthread_pflags_restore(save);
3559 vn_finished_write(mp);
3565 * The syncer vnode is no referenced.
3568 sync_inactive(struct vop_inactive_args *ap)
3576 * The syncer vnode is no longer needed and is being decommissioned.
3578 * Modifications to the worklist must be protected by sync_mtx.
3581 sync_reclaim(struct vop_reclaim_args *ap)
3583 struct vnode *vp = ap->a_vp;
3588 mtx_lock(&sync_mtx);
3589 if (vp->v_mount->mnt_syncer == vp)
3590 vp->v_mount->mnt_syncer = NULL;
3591 if (bo->bo_flag & BO_ONWORKLST) {
3592 LIST_REMOVE(bo, bo_synclist);
3593 syncer_worklist_len--;
3595 bo->bo_flag &= ~BO_ONWORKLST;
3597 mtx_unlock(&sync_mtx);
3604 * Check if vnode represents a disk device
3607 vn_isdisk(struct vnode *vp, int *errp)
3613 if (vp->v_type != VCHR)
3615 else if (vp->v_rdev == NULL)
3617 else if (vp->v_rdev->si_devsw == NULL)
3619 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3624 return (error == 0);
3628 * Common filesystem object access control check routine. Accepts a
3629 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3630 * and optional call-by-reference privused argument allowing vaccess()
3631 * to indicate to the caller whether privilege was used to satisfy the
3632 * request (obsoleted). Returns 0 on success, or an errno on failure.
3635 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3636 accmode_t accmode, struct ucred *cred, int *privused)
3638 accmode_t dac_granted;
3639 accmode_t priv_granted;
3641 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3642 ("invalid bit in accmode"));
3643 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3644 ("VAPPEND without VWRITE"));
3647 * Look for a normal, non-privileged way to access the file/directory
3648 * as requested. If it exists, go with that.
3651 if (privused != NULL)
3656 /* Check the owner. */
3657 if (cred->cr_uid == file_uid) {
3658 dac_granted |= VADMIN;
3659 if (file_mode & S_IXUSR)
3660 dac_granted |= VEXEC;
3661 if (file_mode & S_IRUSR)
3662 dac_granted |= VREAD;
3663 if (file_mode & S_IWUSR)
3664 dac_granted |= (VWRITE | VAPPEND);
3666 if ((accmode & dac_granted) == accmode)
3672 /* Otherwise, check the groups (first match) */
3673 if (groupmember(file_gid, cred)) {
3674 if (file_mode & S_IXGRP)
3675 dac_granted |= VEXEC;
3676 if (file_mode & S_IRGRP)
3677 dac_granted |= VREAD;
3678 if (file_mode & S_IWGRP)
3679 dac_granted |= (VWRITE | VAPPEND);
3681 if ((accmode & dac_granted) == accmode)
3687 /* Otherwise, check everyone else. */
3688 if (file_mode & S_IXOTH)
3689 dac_granted |= VEXEC;
3690 if (file_mode & S_IROTH)
3691 dac_granted |= VREAD;
3692 if (file_mode & S_IWOTH)
3693 dac_granted |= (VWRITE | VAPPEND);
3694 if ((accmode & dac_granted) == accmode)
3699 * Build a privilege mask to determine if the set of privileges
3700 * satisfies the requirements when combined with the granted mask
3701 * from above. For each privilege, if the privilege is required,
3702 * bitwise or the request type onto the priv_granted mask.
3708 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3709 * requests, instead of PRIV_VFS_EXEC.
3711 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3712 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3713 priv_granted |= VEXEC;
3716 * Ensure that at least one execute bit is on. Otherwise,
3717 * a privileged user will always succeed, and we don't want
3718 * this to happen unless the file really is executable.
3720 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3721 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3722 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3723 priv_granted |= VEXEC;
3726 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3727 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3728 priv_granted |= VREAD;
3730 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3731 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3732 priv_granted |= (VWRITE | VAPPEND);
3734 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3735 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3736 priv_granted |= VADMIN;
3738 if ((accmode & (priv_granted | dac_granted)) == accmode) {
3739 /* XXX audit: privilege used */
3740 if (privused != NULL)
3745 return ((accmode & VADMIN) ? EPERM : EACCES);
3749 * Credential check based on process requesting service, and per-attribute
3753 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3754 struct thread *td, accmode_t accmode)
3758 * Kernel-invoked always succeeds.
3764 * Do not allow privileged processes in jail to directly manipulate
3765 * system attributes.
3767 switch (attrnamespace) {
3768 case EXTATTR_NAMESPACE_SYSTEM:
3769 /* Potentially should be: return (EPERM); */
3770 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3771 case EXTATTR_NAMESPACE_USER:
3772 return (VOP_ACCESS(vp, accmode, cred, td));
3778 #ifdef DEBUG_VFS_LOCKS
3780 * This only exists to supress warnings from unlocked specfs accesses. It is
3781 * no longer ok to have an unlocked VFS.
3783 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
3784 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
3786 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3787 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3788 "Drop into debugger on lock violation");
3790 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3791 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3792 0, "Check for interlock across VOPs");
3794 int vfs_badlock_print = 1; /* Print lock violations. */
3795 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3796 0, "Print lock violations");
3799 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3800 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3801 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3805 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3809 if (vfs_badlock_backtrace)
3812 if (vfs_badlock_print)
3813 printf("%s: %p %s\n", str, (void *)vp, msg);
3814 if (vfs_badlock_ddb)
3815 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3819 assert_vi_locked(struct vnode *vp, const char *str)
3822 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3823 vfs_badlock("interlock is not locked but should be", str, vp);
3827 assert_vi_unlocked(struct vnode *vp, const char *str)
3830 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3831 vfs_badlock("interlock is locked but should not be", str, vp);
3835 assert_vop_locked(struct vnode *vp, const char *str)
3838 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3839 vfs_badlock("is not locked but should be", str, vp);
3843 assert_vop_unlocked(struct vnode *vp, const char *str)
3846 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3847 vfs_badlock("is locked but should not be", str, vp);
3851 assert_vop_elocked(struct vnode *vp, const char *str)
3854 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3855 vfs_badlock("is not exclusive locked but should be", str, vp);
3860 assert_vop_elocked_other(struct vnode *vp, const char *str)
3863 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3864 vfs_badlock("is not exclusive locked by another thread",
3869 assert_vop_slocked(struct vnode *vp, const char *str)
3872 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3873 vfs_badlock("is not locked shared but should be", str, vp);
3876 #endif /* DEBUG_VFS_LOCKS */
3879 vop_rename_fail(struct vop_rename_args *ap)
3882 if (ap->a_tvp != NULL)
3884 if (ap->a_tdvp == ap->a_tvp)
3893 vop_rename_pre(void *ap)
3895 struct vop_rename_args *a = ap;
3897 #ifdef DEBUG_VFS_LOCKS
3899 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3900 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3901 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3902 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3904 /* Check the source (from). */
3905 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
3906 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
3907 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3908 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
3909 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3911 /* Check the target. */
3913 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3914 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3916 if (a->a_tdvp != a->a_fdvp)
3918 if (a->a_tvp != a->a_fvp)
3926 vop_strategy_pre(void *ap)
3928 #ifdef DEBUG_VFS_LOCKS
3929 struct vop_strategy_args *a;
3936 * Cluster ops lock their component buffers but not the IO container.
3938 if ((bp->b_flags & B_CLUSTER) != 0)
3941 if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
3942 if (vfs_badlock_print)
3944 "VOP_STRATEGY: bp is not locked but should be\n");
3945 if (vfs_badlock_ddb)
3946 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3952 vop_lookup_pre(void *ap)
3954 #ifdef DEBUG_VFS_LOCKS
3955 struct vop_lookup_args *a;
3960 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3961 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3966 vop_lookup_post(void *ap, int rc)
3968 #ifdef DEBUG_VFS_LOCKS
3969 struct vop_lookup_args *a;
3977 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
3978 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
3981 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
3986 vop_lock_pre(void *ap)
3988 #ifdef DEBUG_VFS_LOCKS
3989 struct vop_lock1_args *a = ap;
3991 if ((a->a_flags & LK_INTERLOCK) == 0)
3992 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
3994 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
3999 vop_lock_post(void *ap, int rc)
4001 #ifdef DEBUG_VFS_LOCKS
4002 struct vop_lock1_args *a = ap;
4004 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4006 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4011 vop_unlock_pre(void *ap)
4013 #ifdef DEBUG_VFS_LOCKS
4014 struct vop_unlock_args *a = ap;
4016 if (a->a_flags & LK_INTERLOCK)
4017 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4018 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4023 vop_unlock_post(void *ap, int rc)
4025 #ifdef DEBUG_VFS_LOCKS
4026 struct vop_unlock_args *a = ap;
4028 if (a->a_flags & LK_INTERLOCK)
4029 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4034 vop_create_post(void *ap, int rc)
4036 struct vop_create_args *a = ap;
4039 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4043 vop_deleteextattr_post(void *ap, int rc)
4045 struct vop_deleteextattr_args *a = ap;
4048 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4052 vop_link_post(void *ap, int rc)
4054 struct vop_link_args *a = ap;
4057 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4058 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4063 vop_mkdir_post(void *ap, int rc)
4065 struct vop_mkdir_args *a = ap;
4068 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4072 vop_mknod_post(void *ap, int rc)
4074 struct vop_mknod_args *a = ap;
4077 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4081 vop_remove_post(void *ap, int rc)
4083 struct vop_remove_args *a = ap;
4086 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4087 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4092 vop_rename_post(void *ap, int rc)
4094 struct vop_rename_args *a = ap;
4097 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4098 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4099 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4101 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4103 if (a->a_tdvp != a->a_fdvp)
4105 if (a->a_tvp != a->a_fvp)
4113 vop_rmdir_post(void *ap, int rc)
4115 struct vop_rmdir_args *a = ap;
4118 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4119 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4124 vop_setattr_post(void *ap, int rc)
4126 struct vop_setattr_args *a = ap;
4129 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4133 vop_setextattr_post(void *ap, int rc)
4135 struct vop_setextattr_args *a = ap;
4138 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4142 vop_symlink_post(void *ap, int rc)
4144 struct vop_symlink_args *a = ap;
4147 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4150 static struct knlist fs_knlist;
4153 vfs_event_init(void *arg)
4155 knlist_init_mtx(&fs_knlist, NULL);
4157 /* XXX - correct order? */
4158 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4161 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4164 KNOTE_UNLOCKED(&fs_knlist, event);
4167 static int filt_fsattach(struct knote *kn);
4168 static void filt_fsdetach(struct knote *kn);
4169 static int filt_fsevent(struct knote *kn, long hint);
4171 struct filterops fs_filtops = {
4173 .f_attach = filt_fsattach,
4174 .f_detach = filt_fsdetach,
4175 .f_event = filt_fsevent
4179 filt_fsattach(struct knote *kn)
4182 kn->kn_flags |= EV_CLEAR;
4183 knlist_add(&fs_knlist, kn, 0);
4188 filt_fsdetach(struct knote *kn)
4191 knlist_remove(&fs_knlist, kn, 0);
4195 filt_fsevent(struct knote *kn, long hint)
4198 kn->kn_fflags |= hint;
4199 return (kn->kn_fflags != 0);
4203 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4209 error = SYSCTL_IN(req, &vc, sizeof(vc));
4212 if (vc.vc_vers != VFS_CTL_VERS1)
4214 mp = vfs_getvfs(&vc.vc_fsid);
4217 /* ensure that a specific sysctl goes to the right filesystem. */
4218 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4219 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4223 VCTLTOREQ(&vc, req);
4224 error = VFS_SYSCTL(mp, vc.vc_op, req);
4229 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4230 NULL, 0, sysctl_vfs_ctl, "",
4234 * Function to initialize a va_filerev field sensibly.
4235 * XXX: Wouldn't a random number make a lot more sense ??
4238 init_va_filerev(void)
4243 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4246 static int filt_vfsread(struct knote *kn, long hint);
4247 static int filt_vfswrite(struct knote *kn, long hint);
4248 static int filt_vfsvnode(struct knote *kn, long hint);
4249 static void filt_vfsdetach(struct knote *kn);
4250 static struct filterops vfsread_filtops = {
4252 .f_detach = filt_vfsdetach,
4253 .f_event = filt_vfsread
4255 static struct filterops vfswrite_filtops = {
4257 .f_detach = filt_vfsdetach,
4258 .f_event = filt_vfswrite
4260 static struct filterops vfsvnode_filtops = {
4262 .f_detach = filt_vfsdetach,
4263 .f_event = filt_vfsvnode
4267 vfs_knllock(void *arg)
4269 struct vnode *vp = arg;
4271 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4275 vfs_knlunlock(void *arg)
4277 struct vnode *vp = arg;
4283 vfs_knl_assert_locked(void *arg)
4285 #ifdef DEBUG_VFS_LOCKS
4286 struct vnode *vp = arg;
4288 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4293 vfs_knl_assert_unlocked(void *arg)
4295 #ifdef DEBUG_VFS_LOCKS
4296 struct vnode *vp = arg;
4298 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4303 vfs_kqfilter(struct vop_kqfilter_args *ap)
4305 struct vnode *vp = ap->a_vp;
4306 struct knote *kn = ap->a_kn;
4309 switch (kn->kn_filter) {
4311 kn->kn_fop = &vfsread_filtops;
4314 kn->kn_fop = &vfswrite_filtops;
4317 kn->kn_fop = &vfsvnode_filtops;
4323 kn->kn_hook = (caddr_t)vp;
4326 if (vp->v_pollinfo == NULL)
4328 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4329 knlist_add(knl, kn, 0);
4335 * Detach knote from vnode
4338 filt_vfsdetach(struct knote *kn)
4340 struct vnode *vp = (struct vnode *)kn->kn_hook;
4342 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4343 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4348 filt_vfsread(struct knote *kn, long hint)
4350 struct vnode *vp = (struct vnode *)kn->kn_hook;
4355 * filesystem is gone, so set the EOF flag and schedule
4356 * the knote for deletion.
4358 if (hint == NOTE_REVOKE) {
4360 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4365 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4369 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4370 res = (kn->kn_data != 0);
4377 filt_vfswrite(struct knote *kn, long hint)
4379 struct vnode *vp = (struct vnode *)kn->kn_hook;
4384 * filesystem is gone, so set the EOF flag and schedule
4385 * the knote for deletion.
4387 if (hint == NOTE_REVOKE)
4388 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4396 filt_vfsvnode(struct knote *kn, long hint)
4398 struct vnode *vp = (struct vnode *)kn->kn_hook;
4402 if (kn->kn_sfflags & hint)
4403 kn->kn_fflags |= hint;
4404 if (hint == NOTE_REVOKE) {
4405 kn->kn_flags |= EV_EOF;
4409 res = (kn->kn_fflags != 0);
4415 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4419 if (dp->d_reclen > ap->a_uio->uio_resid)
4420 return (ENAMETOOLONG);
4421 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4423 if (ap->a_ncookies != NULL) {
4424 if (ap->a_cookies != NULL)
4425 free(ap->a_cookies, M_TEMP);
4426 ap->a_cookies = NULL;
4427 *ap->a_ncookies = 0;
4431 if (ap->a_ncookies == NULL)
4434 KASSERT(ap->a_cookies,
4435 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4437 *ap->a_cookies = realloc(*ap->a_cookies,
4438 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4439 (*ap->a_cookies)[*ap->a_ncookies] = off;
4444 * Mark for update the access time of the file if the filesystem
4445 * supports VOP_MARKATIME. This functionality is used by execve and
4446 * mmap, so we want to avoid the I/O implied by directly setting
4447 * va_atime for the sake of efficiency.
4450 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4455 VFS_ASSERT_GIANT(mp);
4456 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4457 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4458 (void)VOP_MARKATIME(vp);
4462 * The purpose of this routine is to remove granularity from accmode_t,
4463 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4464 * VADMIN and VAPPEND.
4466 * If it returns 0, the caller is supposed to continue with the usual
4467 * access checks using 'accmode' as modified by this routine. If it
4468 * returns nonzero value, the caller is supposed to return that value
4471 * Note that after this routine runs, accmode may be zero.
4474 vfs_unixify_accmode(accmode_t *accmode)
4477 * There is no way to specify explicit "deny" rule using
4478 * file mode or POSIX.1e ACLs.
4480 if (*accmode & VEXPLICIT_DENY) {
4486 * None of these can be translated into usual access bits.
4487 * Also, the common case for NFSv4 ACLs is to not contain
4488 * either of these bits. Caller should check for VWRITE
4489 * on the containing directory instead.
4491 if (*accmode & (VDELETE_CHILD | VDELETE))
4494 if (*accmode & VADMIN_PERMS) {
4495 *accmode &= ~VADMIN_PERMS;
4500 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4501 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4503 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);