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("$FreeBSD$");
44 #include "opt_compat.h"
46 #include "opt_watchdog.h"
48 #include <sys/param.h>
49 #include <sys/systm.h>
52 #include <sys/condvar.h>
54 #include <sys/dirent.h>
55 #include <sys/event.h>
56 #include <sys/eventhandler.h>
57 #include <sys/extattr.h>
59 #include <sys/fcntl.h>
62 #include <sys/kernel.h>
63 #include <sys/kthread.h>
64 #include <sys/lockf.h>
65 #include <sys/malloc.h>
66 #include <sys/mount.h>
67 #include <sys/namei.h>
69 #include <sys/reboot.h>
70 #include <sys/sched.h>
71 #include <sys/sleepqueue.h>
74 #include <sys/sysctl.h>
75 #include <sys/syslog.h>
76 #include <sys/vmmeter.h>
77 #include <sys/vnode.h>
78 #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 v_incr_usecount(struct vnode *);
106 static void v_decr_usecount(struct vnode *);
107 static void v_decr_useonly(struct vnode *);
108 static void v_upgrade_usecount(struct vnode *);
109 static void vnlru_free(int);
110 static void vgonel(struct vnode *);
111 static void vfs_knllock(void *arg);
112 static void vfs_knlunlock(void *arg);
113 static void vfs_knl_assert_locked(void *arg);
114 static void vfs_knl_assert_unlocked(void *arg);
115 static void destroy_vpollinfo(struct vpollinfo *vi);
118 * Number of vnodes in existence. Increased whenever getnewvnode()
119 * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
121 static unsigned long numvnodes;
123 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
124 "Number of vnodes in existence");
127 * Conversion tables for conversion from vnode types to inode formats
130 enum vtype iftovt_tab[16] = {
131 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
132 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
134 int vttoif_tab[10] = {
135 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
136 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
140 * List of vnodes that are ready for recycling.
142 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
145 * Free vnode target. Free vnodes may simply be files which have been stat'd
146 * but not read. This is somewhat common, and a small cache of such files
147 * should be kept to avoid recreation costs.
149 static u_long wantfreevnodes;
150 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
151 /* Number of vnodes in the free list. */
152 static u_long freevnodes;
153 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
154 "Number of vnodes in the free list");
156 static int vlru_allow_cache_src;
157 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
158 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
161 * Various variables used for debugging the new implementation of
163 * XXX these are probably of (very) limited utility now.
165 static int reassignbufcalls;
166 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
167 "Number of calls to reassignbuf");
170 * Cache for the mount type id assigned to NFS. This is used for
171 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
173 int nfs_mount_type = -1;
175 /* To keep more than one thread at a time from running vfs_getnewfsid */
176 static struct mtx mntid_mtx;
179 * Lock for any access to the following:
184 static struct mtx vnode_free_list_mtx;
186 /* Publicly exported FS */
187 struct nfs_public nfs_pub;
189 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
190 static uma_zone_t vnode_zone;
191 static uma_zone_t vnodepoll_zone;
194 * The workitem queue.
196 * It is useful to delay writes of file data and filesystem metadata
197 * for tens of seconds so that quickly created and deleted files need
198 * not waste disk bandwidth being created and removed. To realize this,
199 * we append vnodes to a "workitem" queue. When running with a soft
200 * updates implementation, most pending metadata dependencies should
201 * not wait for more than a few seconds. Thus, mounted on block devices
202 * are delayed only about a half the time that file data is delayed.
203 * Similarly, directory updates are more critical, so are only delayed
204 * about a third the time that file data is delayed. Thus, there are
205 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
206 * one each second (driven off the filesystem syncer process). The
207 * syncer_delayno variable indicates the next queue that is to be processed.
208 * Items that need to be processed soon are placed in this queue:
210 * syncer_workitem_pending[syncer_delayno]
212 * A delay of fifteen seconds is done by placing the request fifteen
213 * entries later in the queue:
215 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
218 static int syncer_delayno;
219 static long syncer_mask;
220 LIST_HEAD(synclist, bufobj);
221 static struct synclist *syncer_workitem_pending[2];
223 * The sync_mtx protects:
228 * syncer_workitem_pending
229 * syncer_worklist_len
232 static struct mtx sync_mtx;
233 static struct cv sync_wakeup;
235 #define SYNCER_MAXDELAY 32
236 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
237 static int syncdelay = 30; /* max time to delay syncing data */
238 static int filedelay = 30; /* time to delay syncing files */
239 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
240 "Time to delay syncing files (in seconds)");
241 static int dirdelay = 29; /* time to delay syncing directories */
242 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
243 "Time to delay syncing directories (in seconds)");
244 static int metadelay = 28; /* time to delay syncing metadata */
245 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
246 "Time to delay syncing metadata (in seconds)");
247 static int rushjob; /* number of slots to run ASAP */
248 static int stat_rush_requests; /* number of times I/O speeded up */
249 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
250 "Number of times I/O speeded up (rush requests)");
253 * When shutting down the syncer, run it at four times normal speed.
255 #define SYNCER_SHUTDOWN_SPEEDUP 4
256 static int sync_vnode_count;
257 static int syncer_worklist_len;
258 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
262 * Number of vnodes we want to exist at any one time. This is mostly used
263 * to size hash tables in vnode-related code. It is normally not used in
264 * getnewvnode(), as wantfreevnodes is normally nonzero.)
266 * XXX desiredvnodes is historical cruft and should not exist.
269 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
270 &desiredvnodes, 0, "Maximum number of vnodes");
271 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
272 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
273 static int vnlru_nowhere;
274 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
275 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
278 * Macros to control when a vnode is freed and recycled. All require
279 * the vnode interlock.
281 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
282 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
283 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
285 /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
289 * Initialize the vnode management data structures.
291 * Reevaluate the following cap on the number of vnodes after the physical
292 * memory size exceeds 512GB. In the limit, as the physical memory size
293 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
295 #ifndef MAXVNODES_MAX
296 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
299 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");
343 for (i = 1; i <= sizeof(struct vnode); i <<= 1)
347 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
351 * Mark a mount point as busy. Used to synchronize access and to delay
352 * unmounting. Eventually, mountlist_mtx is not released on failure.
354 * vfs_busy() is a custom lock, it can block the caller.
355 * vfs_busy() only sleeps if the unmount is active on the mount point.
356 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
357 * vnode belonging to mp.
359 * Lookup uses vfs_busy() to traverse mount points.
361 * / vnode lock A / vnode lock (/var) D
362 * /var vnode lock B /log vnode lock(/var/log) E
363 * vfs_busy lock C vfs_busy lock F
365 * Within each file system, the lock order is C->A->B and F->D->E.
367 * When traversing across mounts, the system follows that lock order:
373 * The lookup() process for namei("/var") illustrates the process:
374 * VOP_LOOKUP() obtains B while A is held
375 * vfs_busy() obtains a shared lock on F while A and B are held
376 * vput() releases lock on B
377 * vput() releases lock on A
378 * VFS_ROOT() obtains lock on D while shared lock on F is held
379 * vfs_unbusy() releases shared lock on F
380 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
381 * Attempt to lock A (instead of vp_crossmp) while D is held would
382 * violate the global order, causing deadlocks.
384 * dounmount() locks B while F is drained.
387 vfs_busy(struct mount *mp, int flags)
390 MPASS((flags & ~MBF_MASK) == 0);
391 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
396 * If mount point is currenly being unmounted, sleep until the
397 * mount point fate is decided. If thread doing the unmounting fails,
398 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
399 * that this mount point has survived the unmount attempt and vfs_busy
400 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
401 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
402 * about to be really destroyed. vfs_busy needs to release its
403 * reference on the mount point in this case and return with ENOENT,
404 * telling the caller that mount mount it tried to busy is no longer
407 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
408 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
411 CTR1(KTR_VFS, "%s: failed busying before sleeping",
415 if (flags & MBF_MNTLSTLOCK)
416 mtx_unlock(&mountlist_mtx);
417 mp->mnt_kern_flag |= MNTK_MWAIT;
418 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
419 if (flags & MBF_MNTLSTLOCK)
420 mtx_lock(&mountlist_mtx);
423 if (flags & MBF_MNTLSTLOCK)
424 mtx_unlock(&mountlist_mtx);
431 * Free a busy filesystem.
434 vfs_unbusy(struct mount *mp)
437 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
440 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
442 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
443 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
444 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
445 mp->mnt_kern_flag &= ~MNTK_DRAINING;
446 wakeup(&mp->mnt_lockref);
452 * Lookup a mount point by filesystem identifier.
455 vfs_getvfs(fsid_t *fsid)
459 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
460 mtx_lock(&mountlist_mtx);
461 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
462 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
463 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
465 mtx_unlock(&mountlist_mtx);
469 mtx_unlock(&mountlist_mtx);
470 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
471 return ((struct mount *) 0);
475 * Lookup a mount point by filesystem identifier, busying it before
478 * To avoid congestion on mountlist_mtx, implement simple direct-mapped
479 * cache for popular filesystem identifiers. The cache is lockess, using
480 * the fact that struct mount's are never freed. In worst case we may
481 * get pointer to unmounted or even different filesystem, so we have to
482 * check what we got, and go slow way if so.
485 vfs_busyfs(fsid_t *fsid)
487 #define FSID_CACHE_SIZE 256
488 typedef struct mount * volatile vmp_t;
489 static vmp_t cache[FSID_CACHE_SIZE];
494 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
495 hash = fsid->val[0] ^ fsid->val[1];
496 hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
499 mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
500 mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
502 if (vfs_busy(mp, 0) != 0) {
506 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
507 mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
513 mtx_lock(&mountlist_mtx);
514 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
515 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
516 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
517 error = vfs_busy(mp, MBF_MNTLSTLOCK);
520 mtx_unlock(&mountlist_mtx);
527 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
528 mtx_unlock(&mountlist_mtx);
529 return ((struct mount *) 0);
533 * Check if a user can access privileged mount options.
536 vfs_suser(struct mount *mp, struct thread *td)
541 * If the thread is jailed, but this is not a jail-friendly file
542 * system, deny immediately.
544 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
548 * If the file system was mounted outside the jail of the calling
549 * thread, deny immediately.
551 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
555 * If file system supports delegated administration, we don't check
556 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
557 * by the file system itself.
558 * If this is not the user that did original mount, we check for
559 * the PRIV_VFS_MOUNT_OWNER privilege.
561 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
562 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
563 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
570 * Get a new unique fsid. Try to make its val[0] unique, since this value
571 * will be used to create fake device numbers for stat(). Also try (but
572 * not so hard) make its val[0] unique mod 2^16, since some emulators only
573 * support 16-bit device numbers. We end up with unique val[0]'s for the
574 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
576 * Keep in mind that several mounts may be running in parallel. Starting
577 * the search one past where the previous search terminated is both a
578 * micro-optimization and a defense against returning the same fsid to
582 vfs_getnewfsid(struct mount *mp)
584 static uint16_t mntid_base;
589 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
590 mtx_lock(&mntid_mtx);
591 mtype = mp->mnt_vfc->vfc_typenum;
592 tfsid.val[1] = mtype;
593 mtype = (mtype & 0xFF) << 24;
595 tfsid.val[0] = makedev(255,
596 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
598 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
602 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
603 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
604 mtx_unlock(&mntid_mtx);
608 * Knob to control the precision of file timestamps:
610 * 0 = seconds only; nanoseconds zeroed.
611 * 1 = seconds and nanoseconds, accurate within 1/HZ.
612 * 2 = seconds and nanoseconds, truncated to microseconds.
613 * >=3 = seconds and nanoseconds, maximum precision.
615 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
617 static int timestamp_precision = TSP_SEC;
618 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
619 ×tamp_precision, 0, "File timestamp precision (0: seconds, "
620 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
621 "3+: sec + ns (max. precision))");
624 * Get a current timestamp.
627 vfs_timestamp(struct timespec *tsp)
631 switch (timestamp_precision) {
633 tsp->tv_sec = time_second;
641 TIMEVAL_TO_TIMESPEC(&tv, tsp);
651 * Set vnode attributes to VNOVAL
654 vattr_null(struct vattr *vap)
658 vap->va_size = VNOVAL;
659 vap->va_bytes = VNOVAL;
660 vap->va_mode = VNOVAL;
661 vap->va_nlink = VNOVAL;
662 vap->va_uid = VNOVAL;
663 vap->va_gid = VNOVAL;
664 vap->va_fsid = VNOVAL;
665 vap->va_fileid = VNOVAL;
666 vap->va_blocksize = VNOVAL;
667 vap->va_rdev = VNOVAL;
668 vap->va_atime.tv_sec = VNOVAL;
669 vap->va_atime.tv_nsec = VNOVAL;
670 vap->va_mtime.tv_sec = VNOVAL;
671 vap->va_mtime.tv_nsec = VNOVAL;
672 vap->va_ctime.tv_sec = VNOVAL;
673 vap->va_ctime.tv_nsec = VNOVAL;
674 vap->va_birthtime.tv_sec = VNOVAL;
675 vap->va_birthtime.tv_nsec = VNOVAL;
676 vap->va_flags = VNOVAL;
677 vap->va_gen = VNOVAL;
682 * This routine is called when we have too many vnodes. It attempts
683 * to free <count> vnodes and will potentially free vnodes that still
684 * have VM backing store (VM backing store is typically the cause
685 * of a vnode blowout so we want to do this). Therefore, this operation
686 * is not considered cheap.
688 * A number of conditions may prevent a vnode from being reclaimed.
689 * the buffer cache may have references on the vnode, a directory
690 * vnode may still have references due to the namei cache representing
691 * underlying files, or the vnode may be in active use. It is not
692 * desireable to reuse such vnodes. These conditions may cause the
693 * number of vnodes to reach some minimum value regardless of what
694 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
697 vlrureclaim(struct mount *mp)
706 * Calculate the trigger point, don't allow user
707 * screwups to blow us up. This prevents us from
708 * recycling vnodes with lots of resident pages. We
709 * aren't trying to free memory, we are trying to
712 usevnodes = desiredvnodes;
715 trigger = cnt.v_page_count * 2 / usevnodes;
717 vn_start_write(NULL, &mp, V_WAIT);
719 count = mp->mnt_nvnodelistsize / 10 + 1;
721 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
722 while (vp != NULL && vp->v_type == VMARKER)
723 vp = TAILQ_NEXT(vp, v_nmntvnodes);
726 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
727 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
732 * If it's been deconstructed already, it's still
733 * referenced, or it exceeds the trigger, skip it.
735 if (vp->v_usecount ||
736 (!vlru_allow_cache_src &&
737 !LIST_EMPTY(&(vp)->v_cache_src)) ||
738 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
739 vp->v_object->resident_page_count > trigger)) {
745 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
747 goto next_iter_mntunlocked;
751 * v_usecount may have been bumped after VOP_LOCK() dropped
752 * the vnode interlock and before it was locked again.
754 * It is not necessary to recheck VI_DOOMED because it can
755 * only be set by another thread that holds both the vnode
756 * lock and vnode interlock. If another thread has the
757 * vnode lock before we get to VOP_LOCK() and obtains the
758 * vnode interlock after VOP_LOCK() drops the vnode
759 * interlock, the other thread will be unable to drop the
760 * vnode lock before our VOP_LOCK() call fails.
762 if (vp->v_usecount ||
763 (!vlru_allow_cache_src &&
764 !LIST_EMPTY(&(vp)->v_cache_src)) ||
765 (vp->v_object != NULL &&
766 vp->v_object->resident_page_count > trigger)) {
767 VOP_UNLOCK(vp, LK_INTERLOCK);
768 goto next_iter_mntunlocked;
770 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
771 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
776 next_iter_mntunlocked:
785 kern_yield(PRI_UNCHANGED);
790 vn_finished_write(mp);
795 * Attempt to keep the free list at wantfreevnodes length.
798 vnlru_free(int count)
803 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
804 for (; count > 0; count--) {
805 vp = TAILQ_FIRST(&vnode_free_list);
807 * The list can be modified while the free_list_mtx
808 * has been dropped and vp could be NULL here.
812 VNASSERT(vp->v_op != NULL, vp,
813 ("vnlru_free: vnode already reclaimed."));
814 KASSERT((vp->v_iflag & VI_FREE) != 0,
815 ("Removing vnode not on freelist"));
816 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
817 ("Mangling active vnode"));
818 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
820 * Don't recycle if we can't get the interlock.
822 if (!VI_TRYLOCK(vp)) {
823 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
826 VNASSERT(VCANRECYCLE(vp), vp,
827 ("vp inconsistent on freelist"));
829 vp->v_iflag &= ~VI_FREE;
831 mtx_unlock(&vnode_free_list_mtx);
833 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
835 VFS_UNLOCK_GIANT(vfslocked);
837 * If the recycled succeeded this vdrop will actually free
838 * the vnode. If not it will simply place it back on
842 mtx_lock(&vnode_free_list_mtx);
846 * Attempt to recycle vnodes in a context that is always safe to block.
847 * Calling vlrurecycle() from the bowels of filesystem code has some
848 * interesting deadlock problems.
850 static struct proc *vnlruproc;
851 static int vnlruproc_sig;
856 struct mount *mp, *nmp;
858 struct proc *p = vnlruproc;
860 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
864 kproc_suspend_check(p);
865 mtx_lock(&vnode_free_list_mtx);
866 if (freevnodes > wantfreevnodes)
867 vnlru_free(freevnodes - wantfreevnodes);
868 if (numvnodes <= desiredvnodes * 9 / 10) {
870 wakeup(&vnlruproc_sig);
871 msleep(vnlruproc, &vnode_free_list_mtx,
872 PVFS|PDROP, "vlruwt", hz);
875 mtx_unlock(&vnode_free_list_mtx);
877 mtx_lock(&mountlist_mtx);
878 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
879 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
880 nmp = TAILQ_NEXT(mp, mnt_list);
883 vfslocked = VFS_LOCK_GIANT(mp);
884 done += vlrureclaim(mp);
885 VFS_UNLOCK_GIANT(vfslocked);
886 mtx_lock(&mountlist_mtx);
887 nmp = TAILQ_NEXT(mp, mnt_list);
890 mtx_unlock(&mountlist_mtx);
893 /* These messages are temporary debugging aids */
894 if (vnlru_nowhere < 5)
895 printf("vnlru process getting nowhere..\n");
896 else if (vnlru_nowhere == 5)
897 printf("vnlru process messages stopped.\n");
900 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
902 kern_yield(PRI_UNCHANGED);
906 static struct kproc_desc vnlru_kp = {
911 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
915 * Routines having to do with the management of the vnode table.
919 * Try to recycle a freed vnode. We abort if anyone picks up a reference
920 * before we actually vgone(). This function must be called with the vnode
921 * held to prevent the vnode from being returned to the free list midway
925 vtryrecycle(struct vnode *vp)
929 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
930 VNASSERT(vp->v_holdcnt, vp,
931 ("vtryrecycle: Recycling vp %p without a reference.", vp));
933 * This vnode may found and locked via some other list, if so we
934 * can't recycle it yet.
936 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
938 "%s: impossible to recycle, vp %p lock is already held",
940 return (EWOULDBLOCK);
943 * Don't recycle if its filesystem is being suspended.
945 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
948 "%s: impossible to recycle, cannot start the write for %p",
953 * If we got this far, we need to acquire the interlock and see if
954 * anyone picked up this vnode from another list. If not, we will
955 * mark it with DOOMED via vgonel() so that anyone who does find it
959 if (vp->v_usecount) {
960 VOP_UNLOCK(vp, LK_INTERLOCK);
961 vn_finished_write(vnmp);
963 "%s: impossible to recycle, %p is already referenced",
967 if ((vp->v_iflag & VI_DOOMED) == 0)
969 VOP_UNLOCK(vp, LK_INTERLOCK);
970 vn_finished_write(vnmp);
975 * Wait for available vnodes.
978 getnewvnode_wait(int suspended)
981 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
982 if (numvnodes > desiredvnodes) {
985 * File system is beeing suspended, we cannot risk a
986 * deadlock here, so allocate new vnode anyway.
988 if (freevnodes > wantfreevnodes)
989 vnlru_free(freevnodes - wantfreevnodes);
992 if (vnlruproc_sig == 0) {
993 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
996 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
999 return (numvnodes > desiredvnodes ? ENFILE : 0);
1003 getnewvnode_reserve(u_int count)
1008 /* First try to be quick and racy. */
1009 if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
1010 td->td_vp_reserv += count;
1013 atomic_subtract_long(&numvnodes, count);
1015 mtx_lock(&vnode_free_list_mtx);
1017 if (getnewvnode_wait(0) == 0) {
1020 atomic_add_long(&numvnodes, 1);
1023 mtx_unlock(&vnode_free_list_mtx);
1027 getnewvnode_drop_reserve(void)
1032 atomic_subtract_long(&numvnodes, td->td_vp_reserv);
1033 td->td_vp_reserv = 0;
1037 * Return the next vnode from the free list.
1040 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
1048 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
1051 if (td->td_vp_reserv > 0) {
1052 td->td_vp_reserv -= 1;
1055 mtx_lock(&vnode_free_list_mtx);
1057 * Lend our context to reclaim vnodes if they've exceeded the max.
1059 if (freevnodes > wantfreevnodes)
1061 error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1063 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1065 mtx_unlock(&vnode_free_list_mtx);
1069 atomic_add_long(&numvnodes, 1);
1070 mtx_unlock(&vnode_free_list_mtx);
1072 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1076 vp->v_vnlock = &vp->v_lock;
1077 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1079 * By default, don't allow shared locks unless filesystems
1082 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
1084 * Initialize bufobj.
1087 bo->__bo_vnode = vp;
1088 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
1089 bo->bo_ops = &buf_ops_bio;
1090 bo->bo_private = vp;
1091 TAILQ_INIT(&bo->bo_clean.bv_hd);
1092 TAILQ_INIT(&bo->bo_dirty.bv_hd);
1094 * Initialize namecache.
1096 LIST_INIT(&vp->v_cache_src);
1097 TAILQ_INIT(&vp->v_cache_dst);
1099 * Finalize various vnode identity bits.
1104 v_incr_usecount(vp);
1108 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1109 mac_vnode_associate_singlelabel(mp, vp);
1110 else if (mp == NULL && vops != &dead_vnodeops)
1111 printf("NULL mp in getnewvnode()\n");
1114 bo->bo_bsize = mp->mnt_stat.f_iosize;
1115 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1116 vp->v_vflag |= VV_NOKNOTE;
1118 rangelock_init(&vp->v_rl);
1121 * For the filesystems which do not use vfs_hash_insert(),
1122 * still initialize v_hash to have vfs_hash_index() useful.
1123 * E.g., nullfs uses vfs_hash_index() on the lower vnode for
1126 vp->v_hash = (uintptr_t)vp >> vnsz2log;
1133 * Delete from old mount point vnode list, if on one.
1136 delmntque(struct vnode *vp)
1146 KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1147 ("Active vnode list size %d > Vnode list size %d",
1148 mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1149 active = vp->v_iflag & VI_ACTIVE;
1150 vp->v_iflag &= ~VI_ACTIVE;
1152 mtx_lock(&vnode_free_list_mtx);
1153 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1154 mp->mnt_activevnodelistsize--;
1155 mtx_unlock(&vnode_free_list_mtx);
1159 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1160 ("bad mount point vnode list size"));
1161 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1162 mp->mnt_nvnodelistsize--;
1168 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1172 vp->v_op = &dead_vnodeops;
1173 /* XXX non mp-safe fs may still call insmntque with vnode
1175 if (!VOP_ISLOCKED(vp))
1176 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1182 * Insert into list of vnodes for the new mount point, if available.
1185 insmntque1(struct vnode *vp, struct mount *mp,
1186 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1190 KASSERT(vp->v_mount == NULL,
1191 ("insmntque: vnode already on per mount vnode list"));
1192 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1193 #ifdef DEBUG_VFS_LOCKS
1194 if (!VFS_NEEDSGIANT(mp))
1195 ASSERT_VOP_ELOCKED(vp,
1196 "insmntque: mp-safe fs and non-locked vp");
1199 * We acquire the vnode interlock early to ensure that the
1200 * vnode cannot be recycled by another process releasing a
1201 * holdcnt on it before we get it on both the vnode list
1202 * and the active vnode list. The mount mutex protects only
1203 * manipulation of the vnode list and the vnode freelist
1204 * mutex protects only manipulation of the active vnode list.
1205 * Hence the need to hold the vnode interlock throughout.
1209 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1210 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1211 mp->mnt_nvnodelistsize == 0)) {
1212 locked = VOP_ISLOCKED(vp);
1213 if (!locked || (locked == LK_EXCLUSIVE &&
1214 (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1224 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1225 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1226 ("neg mount point vnode list size"));
1227 mp->mnt_nvnodelistsize++;
1228 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1229 ("Activating already active vnode"));
1230 vp->v_iflag |= VI_ACTIVE;
1231 mtx_lock(&vnode_free_list_mtx);
1232 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1233 mp->mnt_activevnodelistsize++;
1234 mtx_unlock(&vnode_free_list_mtx);
1241 insmntque(struct vnode *vp, struct mount *mp)
1244 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1248 * Flush out and invalidate all buffers associated with a bufobj
1249 * Called with the underlying object locked.
1252 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1257 if (flags & V_SAVE) {
1258 error = bufobj_wwait(bo, slpflag, slptimeo);
1263 if (bo->bo_dirty.bv_cnt > 0) {
1265 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1268 * XXX We could save a lock/unlock if this was only
1269 * enabled under INVARIANTS
1272 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1273 panic("vinvalbuf: dirty bufs");
1277 * If you alter this loop please notice that interlock is dropped and
1278 * reacquired in flushbuflist. Special care is needed to ensure that
1279 * no race conditions occur from this.
1282 error = flushbuflist(&bo->bo_clean,
1283 flags, bo, slpflag, slptimeo);
1284 if (error == 0 && !(flags & V_CLEANONLY))
1285 error = flushbuflist(&bo->bo_dirty,
1286 flags, bo, slpflag, slptimeo);
1287 if (error != 0 && error != EAGAIN) {
1291 } while (error != 0);
1294 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1295 * have write I/O in-progress but if there is a VM object then the
1296 * VM object can also have read-I/O in-progress.
1299 bufobj_wwait(bo, 0, 0);
1301 if (bo->bo_object != NULL) {
1302 VM_OBJECT_LOCK(bo->bo_object);
1303 vm_object_pip_wait(bo->bo_object, "bovlbx");
1304 VM_OBJECT_UNLOCK(bo->bo_object);
1307 } while (bo->bo_numoutput > 0);
1311 * Destroy the copy in the VM cache, too.
1313 if (bo->bo_object != NULL &&
1314 (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1315 VM_OBJECT_LOCK(bo->bo_object);
1316 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1317 OBJPR_CLEANONLY : 0);
1318 VM_OBJECT_UNLOCK(bo->bo_object);
1323 if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1324 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1325 panic("vinvalbuf: flush failed");
1332 * Flush out and invalidate all buffers associated with a vnode.
1333 * Called with the underlying object locked.
1336 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1339 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1340 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1341 if (vp->v_object != NULL && vp->v_object->handle != vp)
1343 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1347 * Flush out buffers on the specified list.
1351 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1354 struct buf *bp, *nbp;
1359 ASSERT_BO_LOCKED(bo);
1362 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1363 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1364 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1370 lblkno = nbp->b_lblkno;
1371 xflags = nbp->b_xflags &
1372 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1375 error = BUF_TIMELOCK(bp,
1376 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1377 "flushbuf", slpflag, slptimeo);
1380 return (error != ENOLCK ? error : EAGAIN);
1382 KASSERT(bp->b_bufobj == bo,
1383 ("bp %p wrong b_bufobj %p should be %p",
1384 bp, bp->b_bufobj, bo));
1385 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1391 * XXX Since there are no node locks for NFS, I
1392 * believe there is a slight chance that a delayed
1393 * write will occur while sleeping just above, so
1396 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1401 bp->b_flags |= B_ASYNC;
1404 return (EAGAIN); /* XXX: why not loop ? */
1409 bp->b_flags |= (B_INVAL | B_RELBUF);
1410 bp->b_flags &= ~B_ASYNC;
1414 (nbp->b_bufobj != bo ||
1415 nbp->b_lblkno != lblkno ||
1417 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1418 break; /* nbp invalid */
1424 * Truncate a file's buffer and pages to a specified length. This
1425 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1429 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1430 off_t length, int blksize)
1432 struct buf *bp, *nbp;
1437 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1438 vp, cred, blksize, (uintmax_t)length);
1441 * Round up to the *next* lbn.
1443 trunclbn = (length + blksize - 1) / blksize;
1445 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1452 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1453 if (bp->b_lblkno < trunclbn)
1456 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1457 BO_MTX(bo)) == ENOLCK)
1463 bp->b_flags |= (B_INVAL | B_RELBUF);
1464 bp->b_flags &= ~B_ASYNC;
1470 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1471 (nbp->b_vp != vp) ||
1472 (nbp->b_flags & B_DELWRI))) {
1478 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1479 if (bp->b_lblkno < trunclbn)
1482 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1483 BO_MTX(bo)) == ENOLCK)
1488 bp->b_flags |= (B_INVAL | B_RELBUF);
1489 bp->b_flags &= ~B_ASYNC;
1495 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1496 (nbp->b_vp != vp) ||
1497 (nbp->b_flags & B_DELWRI) == 0)) {
1506 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1507 if (bp->b_lblkno > 0)
1510 * Since we hold the vnode lock this should only
1511 * fail if we're racing with the buf daemon.
1514 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1515 BO_MTX(bo)) == ENOLCK) {
1518 VNASSERT((bp->b_flags & B_DELWRI), vp,
1519 ("buf(%p) on dirty queue without DELWRI", bp));
1530 bufobj_wwait(bo, 0, 0);
1532 vnode_pager_setsize(vp, length);
1538 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1541 * NOTE: We have to deal with the special case of a background bitmap
1542 * buffer, a situation where two buffers will have the same logical
1543 * block offset. We want (1) only the foreground buffer to be accessed
1544 * in a lookup and (2) must differentiate between the foreground and
1545 * background buffer in the splay tree algorithm because the splay
1546 * tree cannot normally handle multiple entities with the same 'index'.
1547 * We accomplish this by adding differentiating flags to the splay tree's
1552 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1555 struct buf *lefttreemax, *righttreemin, *y;
1559 lefttreemax = righttreemin = &dummy;
1561 if (lblkno < root->b_lblkno ||
1562 (lblkno == root->b_lblkno &&
1563 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1564 if ((y = root->b_left) == NULL)
1566 if (lblkno < y->b_lblkno) {
1568 root->b_left = y->b_right;
1571 if ((y = root->b_left) == NULL)
1574 /* Link into the new root's right tree. */
1575 righttreemin->b_left = root;
1576 righttreemin = root;
1577 } else if (lblkno > root->b_lblkno ||
1578 (lblkno == root->b_lblkno &&
1579 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1580 if ((y = root->b_right) == NULL)
1582 if (lblkno > y->b_lblkno) {
1584 root->b_right = y->b_left;
1587 if ((y = root->b_right) == NULL)
1590 /* Link into the new root's left tree. */
1591 lefttreemax->b_right = root;
1598 /* Assemble the new root. */
1599 lefttreemax->b_right = root->b_left;
1600 righttreemin->b_left = root->b_right;
1601 root->b_left = dummy.b_right;
1602 root->b_right = dummy.b_left;
1607 buf_vlist_remove(struct buf *bp)
1612 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1613 ASSERT_BO_LOCKED(bp->b_bufobj);
1614 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1615 (BX_VNDIRTY|BX_VNCLEAN),
1616 ("buf_vlist_remove: Buf %p is on two lists", bp));
1617 if (bp->b_xflags & BX_VNDIRTY)
1618 bv = &bp->b_bufobj->bo_dirty;
1620 bv = &bp->b_bufobj->bo_clean;
1621 if (bp != bv->bv_root) {
1622 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1623 KASSERT(root == bp, ("splay lookup failed in remove"));
1625 if (bp->b_left == NULL) {
1628 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1629 root->b_right = bp->b_right;
1632 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1634 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1638 * Add the buffer to the sorted clean or dirty block list using a
1639 * splay tree algorithm.
1641 * NOTE: xflags is passed as a constant, optimizing this inline function!
1644 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1649 ASSERT_BO_LOCKED(bo);
1650 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1651 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1652 bp->b_xflags |= xflags;
1653 if (xflags & BX_VNDIRTY)
1658 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1662 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1663 } else if (bp->b_lblkno < root->b_lblkno ||
1664 (bp->b_lblkno == root->b_lblkno &&
1665 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1666 bp->b_left = root->b_left;
1668 root->b_left = NULL;
1669 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1671 bp->b_right = root->b_right;
1673 root->b_right = NULL;
1674 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1681 * Lookup a buffer using the splay tree. Note that we specifically avoid
1682 * shadow buffers used in background bitmap writes.
1684 * This code isn't quite efficient as it could be because we are maintaining
1685 * two sorted lists and do not know which list the block resides in.
1687 * During a "make buildworld" the desired buffer is found at one of
1688 * the roots more than 60% of the time. Thus, checking both roots
1689 * before performing either splay eliminates unnecessary splays on the
1690 * first tree splayed.
1693 gbincore(struct bufobj *bo, daddr_t lblkno)
1697 ASSERT_BO_LOCKED(bo);
1698 if ((bp = bo->bo_clean.bv_root) != NULL &&
1699 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1701 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1702 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1704 if ((bp = bo->bo_clean.bv_root) != NULL) {
1705 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1706 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1709 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1710 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1711 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1718 * Associate a buffer with a vnode.
1721 bgetvp(struct vnode *vp, struct buf *bp)
1726 ASSERT_BO_LOCKED(bo);
1727 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1729 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1730 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1731 ("bgetvp: bp already attached! %p", bp));
1734 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1735 bp->b_flags |= B_NEEDSGIANT;
1739 * Insert onto list for new vnode.
1741 buf_vlist_add(bp, bo, BX_VNCLEAN);
1745 * Disassociate a buffer from a vnode.
1748 brelvp(struct buf *bp)
1753 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1754 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1757 * Delete from old vnode list, if on one.
1759 vp = bp->b_vp; /* XXX */
1762 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1763 buf_vlist_remove(bp);
1765 panic("brelvp: Buffer %p not on queue.", bp);
1766 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1767 bo->bo_flag &= ~BO_ONWORKLST;
1768 mtx_lock(&sync_mtx);
1769 LIST_REMOVE(bo, bo_synclist);
1770 syncer_worklist_len--;
1771 mtx_unlock(&sync_mtx);
1773 bp->b_flags &= ~B_NEEDSGIANT;
1775 bp->b_bufobj = NULL;
1781 * Add an item to the syncer work queue.
1784 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1788 ASSERT_BO_LOCKED(bo);
1790 mtx_lock(&sync_mtx);
1791 if (bo->bo_flag & BO_ONWORKLST)
1792 LIST_REMOVE(bo, bo_synclist);
1794 bo->bo_flag |= BO_ONWORKLST;
1795 syncer_worklist_len++;
1798 if (delay > syncer_maxdelay - 2)
1799 delay = syncer_maxdelay - 2;
1800 slot = (syncer_delayno + delay) & syncer_mask;
1802 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1804 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1806 mtx_unlock(&sync_mtx);
1810 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1814 mtx_lock(&sync_mtx);
1815 len = syncer_worklist_len - sync_vnode_count;
1816 mtx_unlock(&sync_mtx);
1817 error = SYSCTL_OUT(req, &len, sizeof(len));
1821 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1822 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1824 static struct proc *updateproc;
1825 static void sched_sync(void);
1826 static struct kproc_desc up_kp = {
1831 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1834 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1839 *bo = LIST_FIRST(slp);
1842 vp = (*bo)->__bo_vnode; /* XXX */
1843 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1846 * We use vhold in case the vnode does not
1847 * successfully sync. vhold prevents the vnode from
1848 * going away when we unlock the sync_mtx so that
1849 * we can acquire the vnode interlock.
1852 mtx_unlock(&sync_mtx);
1854 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1856 mtx_lock(&sync_mtx);
1857 return (*bo == LIST_FIRST(slp));
1859 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1860 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1862 vn_finished_write(mp);
1864 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1866 * Put us back on the worklist. The worklist
1867 * routine will remove us from our current
1868 * position and then add us back in at a later
1871 vn_syncer_add_to_worklist(*bo, syncdelay);
1875 mtx_lock(&sync_mtx);
1880 * System filesystem synchronizer daemon.
1885 struct synclist *gnext, *next;
1886 struct synclist *gslp, *slp;
1889 struct thread *td = curthread;
1891 int net_worklist_len;
1892 int syncer_final_iter;
1897 syncer_final_iter = 0;
1899 syncer_state = SYNCER_RUNNING;
1900 starttime = time_uptime;
1901 td->td_pflags |= TDP_NORUNNINGBUF;
1903 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1906 mtx_lock(&sync_mtx);
1908 if (syncer_state == SYNCER_FINAL_DELAY &&
1909 syncer_final_iter == 0) {
1910 mtx_unlock(&sync_mtx);
1911 kproc_suspend_check(td->td_proc);
1912 mtx_lock(&sync_mtx);
1914 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1915 if (syncer_state != SYNCER_RUNNING &&
1916 starttime != time_uptime) {
1918 printf("\nSyncing disks, vnodes remaining...");
1921 printf("%d ", net_worklist_len);
1923 starttime = time_uptime;
1926 * Push files whose dirty time has expired. Be careful
1927 * of interrupt race on slp queue.
1929 * Skip over empty worklist slots when shutting down.
1932 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1933 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1934 syncer_delayno += 1;
1935 if (syncer_delayno == syncer_maxdelay)
1937 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1938 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1940 * If the worklist has wrapped since the
1941 * it was emptied of all but syncer vnodes,
1942 * switch to the FINAL_DELAY state and run
1943 * for one more second.
1945 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1946 net_worklist_len == 0 &&
1947 last_work_seen == syncer_delayno) {
1948 syncer_state = SYNCER_FINAL_DELAY;
1949 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1951 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1952 LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1955 * Keep track of the last time there was anything
1956 * on the worklist other than syncer vnodes.
1957 * Return to the SHUTTING_DOWN state if any
1960 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1961 last_work_seen = syncer_delayno;
1962 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1963 syncer_state = SYNCER_SHUTTING_DOWN;
1964 while (!LIST_EMPTY(slp)) {
1965 error = sync_vnode(slp, &bo, td);
1967 LIST_REMOVE(bo, bo_synclist);
1968 LIST_INSERT_HEAD(next, bo, bo_synclist);
1972 if (first_printf == 0)
1973 wdog_kern_pat(WD_LASTVAL);
1976 if (!LIST_EMPTY(gslp)) {
1977 mtx_unlock(&sync_mtx);
1979 mtx_lock(&sync_mtx);
1980 while (!LIST_EMPTY(gslp)) {
1981 error = sync_vnode(gslp, &bo, td);
1983 LIST_REMOVE(bo, bo_synclist);
1984 LIST_INSERT_HEAD(gnext, bo,
1991 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1992 syncer_final_iter--;
1994 * The variable rushjob allows the kernel to speed up the
1995 * processing of the filesystem syncer process. A rushjob
1996 * value of N tells the filesystem syncer to process the next
1997 * N seconds worth of work on its queue ASAP. Currently rushjob
1998 * is used by the soft update code to speed up the filesystem
1999 * syncer process when the incore state is getting so far
2000 * ahead of the disk that the kernel memory pool is being
2001 * threatened with exhaustion.
2008 * Just sleep for a short period of time between
2009 * iterations when shutting down to allow some I/O
2012 * If it has taken us less than a second to process the
2013 * current work, then wait. Otherwise start right over
2014 * again. We can still lose time if any single round
2015 * takes more than two seconds, but it does not really
2016 * matter as we are just trying to generally pace the
2017 * filesystem activity.
2019 if (syncer_state != SYNCER_RUNNING ||
2020 time_uptime == starttime) {
2022 sched_prio(td, PPAUSE);
2025 if (syncer_state != SYNCER_RUNNING)
2026 cv_timedwait(&sync_wakeup, &sync_mtx,
2027 hz / SYNCER_SHUTDOWN_SPEEDUP);
2028 else if (time_uptime == starttime)
2029 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
2034 * Request the syncer daemon to speed up its work.
2035 * We never push it to speed up more than half of its
2036 * normal turn time, otherwise it could take over the cpu.
2039 speedup_syncer(void)
2043 mtx_lock(&sync_mtx);
2044 if (rushjob < syncdelay / 2) {
2046 stat_rush_requests += 1;
2049 mtx_unlock(&sync_mtx);
2050 cv_broadcast(&sync_wakeup);
2055 * Tell the syncer to speed up its work and run though its work
2056 * list several times, then tell it to shut down.
2059 syncer_shutdown(void *arg, int howto)
2062 if (howto & RB_NOSYNC)
2064 mtx_lock(&sync_mtx);
2065 syncer_state = SYNCER_SHUTTING_DOWN;
2067 mtx_unlock(&sync_mtx);
2068 cv_broadcast(&sync_wakeup);
2069 kproc_shutdown(arg, howto);
2073 * Reassign a buffer from one vnode to another.
2074 * Used to assign file specific control information
2075 * (indirect blocks) to the vnode to which they belong.
2078 reassignbuf(struct buf *bp)
2091 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
2092 bp, bp->b_vp, bp->b_flags);
2094 * B_PAGING flagged buffers cannot be reassigned because their vp
2095 * is not fully linked in.
2097 if (bp->b_flags & B_PAGING)
2098 panic("cannot reassign paging buffer");
2101 * Delete from old vnode list, if on one.
2104 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2105 buf_vlist_remove(bp);
2107 panic("reassignbuf: Buffer %p not on queue.", bp);
2109 * If dirty, put on list of dirty buffers; otherwise insert onto list
2112 if (bp->b_flags & B_DELWRI) {
2113 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2114 switch (vp->v_type) {
2124 vn_syncer_add_to_worklist(bo, delay);
2126 buf_vlist_add(bp, bo, BX_VNDIRTY);
2128 buf_vlist_add(bp, bo, BX_VNCLEAN);
2130 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2131 mtx_lock(&sync_mtx);
2132 LIST_REMOVE(bo, bo_synclist);
2133 syncer_worklist_len--;
2134 mtx_unlock(&sync_mtx);
2135 bo->bo_flag &= ~BO_ONWORKLST;
2140 bp = TAILQ_FIRST(&bv->bv_hd);
2141 KASSERT(bp == NULL || bp->b_bufobj == bo,
2142 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2143 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2144 KASSERT(bp == NULL || bp->b_bufobj == bo,
2145 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2147 bp = TAILQ_FIRST(&bv->bv_hd);
2148 KASSERT(bp == NULL || bp->b_bufobj == bo,
2149 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2150 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2151 KASSERT(bp == NULL || bp->b_bufobj == bo,
2152 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2158 * Increment the use and hold counts on the vnode, taking care to reference
2159 * the driver's usecount if this is a chardev. The vholdl() will remove
2160 * the vnode from the free list if it is presently free. Requires the
2161 * vnode interlock and returns with it held.
2164 v_incr_usecount(struct vnode *vp)
2167 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2169 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2171 vp->v_rdev->si_usecount++;
2178 * Turn a holdcnt into a use+holdcnt such that only one call to
2179 * v_decr_usecount is needed.
2182 v_upgrade_usecount(struct vnode *vp)
2185 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2187 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2189 vp->v_rdev->si_usecount++;
2195 * Decrement the vnode use and hold count along with the driver's usecount
2196 * if this is a chardev. The vdropl() below releases the vnode interlock
2197 * as it may free the vnode.
2200 v_decr_usecount(struct vnode *vp)
2203 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2204 VNASSERT(vp->v_usecount > 0, vp,
2205 ("v_decr_usecount: negative usecount"));
2206 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2208 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2210 vp->v_rdev->si_usecount--;
2217 * Decrement only the use count and driver use count. This is intended to
2218 * be paired with a follow on vdropl() to release the remaining hold count.
2219 * In this way we may vgone() a vnode with a 0 usecount without risk of
2220 * having it end up on a free list because the hold count is kept above 0.
2223 v_decr_useonly(struct vnode *vp)
2226 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2227 VNASSERT(vp->v_usecount > 0, vp,
2228 ("v_decr_useonly: negative usecount"));
2229 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2231 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2233 vp->v_rdev->si_usecount--;
2239 * Grab a particular vnode from the free list, increment its
2240 * reference count and lock it. VI_DOOMED is set if the vnode
2241 * is being destroyed. Only callers who specify LK_RETRY will
2242 * see doomed vnodes. If inactive processing was delayed in
2243 * vput try to do it here.
2246 vget(struct vnode *vp, int flags, struct thread *td)
2251 VFS_ASSERT_GIANT(vp->v_mount);
2252 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2253 ("vget: invalid lock operation"));
2254 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2256 if ((flags & LK_INTERLOCK) == 0)
2259 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2261 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2265 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2266 panic("vget: vn_lock failed to return ENOENT\n");
2268 /* Upgrade our holdcnt to a usecount. */
2269 v_upgrade_usecount(vp);
2271 * We don't guarantee that any particular close will
2272 * trigger inactive processing so just make a best effort
2273 * here at preventing a reference to a removed file. If
2274 * we don't succeed no harm is done.
2276 if (vp->v_iflag & VI_OWEINACT) {
2277 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2278 (flags & LK_NOWAIT) == 0)
2280 vp->v_iflag &= ~VI_OWEINACT;
2287 * Increase the reference count of a vnode.
2290 vref(struct vnode *vp)
2293 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2295 v_incr_usecount(vp);
2300 * Return reference count of a vnode.
2302 * The results of this call are only guaranteed when some mechanism other
2303 * than the VI lock is used to stop other processes from gaining references
2304 * to the vnode. This may be the case if the caller holds the only reference.
2305 * This is also useful when stale data is acceptable as race conditions may
2306 * be accounted for by some other means.
2309 vrefcnt(struct vnode *vp)
2314 usecnt = vp->v_usecount;
2320 #define VPUTX_VRELE 1
2321 #define VPUTX_VPUT 2
2322 #define VPUTX_VUNREF 3
2325 vputx(struct vnode *vp, int func)
2329 KASSERT(vp != NULL, ("vputx: null vp"));
2330 if (func == VPUTX_VUNREF)
2331 ASSERT_VOP_LOCKED(vp, "vunref");
2332 else if (func == VPUTX_VPUT)
2333 ASSERT_VOP_LOCKED(vp, "vput");
2335 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2336 VFS_ASSERT_GIANT(vp->v_mount);
2337 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2340 /* Skip this v_writecount check if we're going to panic below. */
2341 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2342 ("vputx: missed vn_close"));
2345 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2346 vp->v_usecount == 1)) {
2347 if (func == VPUTX_VPUT)
2349 v_decr_usecount(vp);
2353 if (vp->v_usecount != 1) {
2354 vprint("vputx: negative ref count", vp);
2355 panic("vputx: negative ref cnt");
2357 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2359 * We want to hold the vnode until the inactive finishes to
2360 * prevent vgone() races. We drop the use count here and the
2361 * hold count below when we're done.
2365 * We must call VOP_INACTIVE with the node locked. Mark
2366 * as VI_DOINGINACT to avoid recursion.
2368 vp->v_iflag |= VI_OWEINACT;
2371 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2375 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2376 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2382 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2383 error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
2388 if (vp->v_usecount > 0)
2389 vp->v_iflag &= ~VI_OWEINACT;
2391 if (vp->v_iflag & VI_OWEINACT)
2392 vinactive(vp, curthread);
2393 if (func != VPUTX_VUNREF)
2400 * Vnode put/release.
2401 * If count drops to zero, call inactive routine and return to freelist.
2404 vrele(struct vnode *vp)
2407 vputx(vp, VPUTX_VRELE);
2411 * Release an already locked vnode. This give the same effects as
2412 * unlock+vrele(), but takes less time and avoids releasing and
2413 * re-aquiring the lock (as vrele() acquires the lock internally.)
2416 vput(struct vnode *vp)
2419 vputx(vp, VPUTX_VPUT);
2423 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2426 vunref(struct vnode *vp)
2429 vputx(vp, VPUTX_VUNREF);
2433 * Somebody doesn't want the vnode recycled.
2436 vhold(struct vnode *vp)
2445 * Increase the hold count and activate if this is the first reference.
2448 vholdl(struct vnode *vp)
2452 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2454 if (!VSHOULDBUSY(vp))
2456 ASSERT_VI_LOCKED(vp, "vholdl");
2457 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2458 VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2460 * Remove a vnode from the free list, mark it as in use,
2461 * and put it on the active list.
2463 mtx_lock(&vnode_free_list_mtx);
2464 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2466 vp->v_iflag &= ~(VI_FREE|VI_AGE);
2467 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2468 ("Activating already active vnode"));
2469 vp->v_iflag |= VI_ACTIVE;
2471 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2472 mp->mnt_activevnodelistsize++;
2473 mtx_unlock(&vnode_free_list_mtx);
2477 * Note that there is one less who cares about this vnode.
2478 * vdrop() is the opposite of vhold().
2481 vdrop(struct vnode *vp)
2489 * Drop the hold count of the vnode. If this is the last reference to
2490 * the vnode we place it on the free list unless it has been vgone'd
2491 * (marked VI_DOOMED) in which case we will free it.
2494 vdropl(struct vnode *vp)
2500 ASSERT_VI_LOCKED(vp, "vdropl");
2501 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2502 if (vp->v_holdcnt <= 0)
2503 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2505 if (vp->v_holdcnt > 0) {
2509 if ((vp->v_iflag & VI_DOOMED) == 0) {
2511 * Mark a vnode as free: remove it from its active list
2512 * and put it up for recycling on the freelist.
2514 VNASSERT(vp->v_op != NULL, vp,
2515 ("vdropl: vnode already reclaimed."));
2516 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2517 ("vnode already free"));
2518 VNASSERT(VSHOULDFREE(vp), vp,
2519 ("vdropl: freeing when we shouldn't"));
2520 active = vp->v_iflag & VI_ACTIVE;
2521 vp->v_iflag &= ~VI_ACTIVE;
2523 mtx_lock(&vnode_free_list_mtx);
2525 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2527 mp->mnt_activevnodelistsize--;
2529 if (vp->v_iflag & VI_AGE) {
2530 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
2532 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
2535 vp->v_iflag &= ~VI_AGE;
2536 vp->v_iflag |= VI_FREE;
2537 mtx_unlock(&vnode_free_list_mtx);
2542 * The vnode has been marked for destruction, so free it.
2544 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2545 atomic_subtract_long(&numvnodes, 1);
2547 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2548 ("cleaned vnode still on the free list."));
2549 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2550 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2551 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2552 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2553 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2554 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2555 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
2556 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2557 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
2558 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2559 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2560 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2563 mac_vnode_destroy(vp);
2565 if (vp->v_pollinfo != NULL)
2566 destroy_vpollinfo(vp->v_pollinfo);
2568 /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2571 rangelock_destroy(&vp->v_rl);
2572 lockdestroy(vp->v_vnlock);
2573 mtx_destroy(&vp->v_interlock);
2574 mtx_destroy(BO_MTX(bo));
2575 uma_zfree(vnode_zone, vp);
2579 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2580 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2581 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2582 * failed lock upgrade.
2585 vinactive(struct vnode *vp, struct thread *td)
2587 struct vm_object *obj;
2589 ASSERT_VOP_ELOCKED(vp, "vinactive");
2590 ASSERT_VI_LOCKED(vp, "vinactive");
2591 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2592 ("vinactive: recursed on VI_DOINGINACT"));
2593 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2594 vp->v_iflag |= VI_DOINGINACT;
2595 vp->v_iflag &= ~VI_OWEINACT;
2598 * Before moving off the active list, we must be sure that any
2599 * modified pages are on the vnode's dirty list since these will
2600 * no longer be checked once the vnode is on the inactive list.
2601 * Because the vnode vm object keeps a hold reference on the vnode
2602 * if there is at least one resident non-cached page, the vnode
2603 * cannot leave the active list without the page cleanup done.
2606 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2607 VM_OBJECT_LOCK(obj);
2608 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2609 VM_OBJECT_UNLOCK(obj);
2611 VOP_INACTIVE(vp, td);
2613 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2614 ("vinactive: lost VI_DOINGINACT"));
2615 vp->v_iflag &= ~VI_DOINGINACT;
2619 * Remove any vnodes in the vnode table belonging to mount point mp.
2621 * If FORCECLOSE is not specified, there should not be any active ones,
2622 * return error if any are found (nb: this is a user error, not a
2623 * system error). If FORCECLOSE is specified, detach any active vnodes
2626 * If WRITECLOSE is set, only flush out regular file vnodes open for
2629 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2631 * `rootrefs' specifies the base reference count for the root vnode
2632 * of this filesystem. The root vnode is considered busy if its
2633 * v_usecount exceeds this value. On a successful return, vflush(, td)
2634 * will call vrele() on the root vnode exactly rootrefs times.
2635 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2639 static int busyprt = 0; /* print out busy vnodes */
2640 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2644 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2646 struct vnode *vp, *mvp, *rootvp = NULL;
2648 int busy = 0, error;
2650 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2653 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2654 ("vflush: bad args"));
2656 * Get the filesystem root vnode. We can vput() it
2657 * immediately, since with rootrefs > 0, it won't go away.
2659 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2660 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2667 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2669 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2672 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2676 * Skip over a vnodes marked VV_SYSTEM.
2678 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2684 * If WRITECLOSE is set, flush out unlinked but still open
2685 * files (even if open only for reading) and regular file
2686 * vnodes open for writing.
2688 if (flags & WRITECLOSE) {
2689 if (vp->v_object != NULL) {
2690 VM_OBJECT_LOCK(vp->v_object);
2691 vm_object_page_clean(vp->v_object, 0, 0, 0);
2692 VM_OBJECT_UNLOCK(vp->v_object);
2694 error = VOP_FSYNC(vp, MNT_WAIT, td);
2698 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2701 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2704 if ((vp->v_type == VNON ||
2705 (error == 0 && vattr.va_nlink > 0)) &&
2706 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2714 * With v_usecount == 0, all we need to do is clear out the
2715 * vnode data structures and we are done.
2717 * If FORCECLOSE is set, forcibly close the vnode.
2719 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2720 VNASSERT(vp->v_usecount == 0 ||
2721 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2722 ("device VNODE %p is FORCECLOSED", vp));
2728 vprint("vflush: busy vnode", vp);
2734 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2736 * If just the root vnode is busy, and if its refcount
2737 * is equal to `rootrefs', then go ahead and kill it.
2740 KASSERT(busy > 0, ("vflush: not busy"));
2741 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2742 ("vflush: usecount %d < rootrefs %d",
2743 rootvp->v_usecount, rootrefs));
2744 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2745 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2747 VOP_UNLOCK(rootvp, 0);
2753 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2757 for (; rootrefs > 0; rootrefs--)
2763 * Recycle an unused vnode to the front of the free list.
2766 vrecycle(struct vnode *vp, struct thread *td)
2770 ASSERT_VOP_ELOCKED(vp, "vrecycle");
2771 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2774 if (vp->v_usecount == 0) {
2783 * Eliminate all activity associated with a vnode
2784 * in preparation for reuse.
2787 vgone(struct vnode *vp)
2795 notify_lowervp_vfs_dummy(struct mount *mp __unused,
2796 struct vnode *lowervp __unused)
2801 * Notify upper mounts about reclaimed or unlinked vnode.
2804 vfs_notify_upper(struct vnode *vp, int event)
2806 static struct vfsops vgonel_vfsops = {
2807 .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
2808 .vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
2810 struct mount *mp, *ump, *mmp;
2817 if (TAILQ_EMPTY(&mp->mnt_uppers))
2820 mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
2821 mmp->mnt_op = &vgonel_vfsops;
2822 mmp->mnt_kern_flag |= MNTK_MARKER;
2824 mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
2825 for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
2826 if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
2827 ump = TAILQ_NEXT(ump, mnt_upper_link);
2830 TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
2833 case VFS_NOTIFY_UPPER_RECLAIM:
2834 VFS_RECLAIM_LOWERVP(ump, vp);
2836 case VFS_NOTIFY_UPPER_UNLINK:
2837 VFS_UNLINK_LOWERVP(ump, vp);
2840 KASSERT(0, ("invalid event %d", event));
2844 ump = TAILQ_NEXT(mmp, mnt_upper_link);
2845 TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
2848 mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
2849 if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
2850 mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
2851 wakeup(&mp->mnt_uppers);
2858 * vgone, with the vp interlock held.
2861 vgonel(struct vnode *vp)
2868 ASSERT_VOP_ELOCKED(vp, "vgonel");
2869 ASSERT_VI_LOCKED(vp, "vgonel");
2870 VNASSERT(vp->v_holdcnt, vp,
2871 ("vgonel: vp %p has no reference.", vp));
2872 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2876 * Don't vgonel if we're already doomed.
2878 if (vp->v_iflag & VI_DOOMED)
2880 vp->v_iflag |= VI_DOOMED;
2883 * Check to see if the vnode is in use. If so, we have to call
2884 * VOP_CLOSE() and VOP_INACTIVE().
2886 active = vp->v_usecount;
2887 oweinact = (vp->v_iflag & VI_OWEINACT);
2889 vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
2892 * Clean out any buffers associated with the vnode.
2893 * If the flush fails, just toss the buffers.
2896 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2897 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2898 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2899 vinvalbuf(vp, 0, 0, 0);
2902 * If purging an active vnode, it must be closed and
2903 * deactivated before being reclaimed.
2906 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2907 if (oweinact || active) {
2909 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2913 if (vp->v_type == VSOCK)
2914 vfs_unp_reclaim(vp);
2916 * Reclaim the vnode.
2918 if (VOP_RECLAIM(vp, td))
2919 panic("vgone: cannot reclaim");
2921 vn_finished_secondary_write(mp);
2922 VNASSERT(vp->v_object == NULL, vp,
2923 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2925 * Clear the advisory locks and wake up waiting threads.
2927 (void)VOP_ADVLOCKPURGE(vp);
2929 * Delete from old mount point vnode list.
2934 * Done with purge, reset to the standard lock and invalidate
2938 vp->v_vnlock = &vp->v_lock;
2939 vp->v_op = &dead_vnodeops;
2945 * Calculate the total number of references to a special device.
2948 vcount(struct vnode *vp)
2953 count = vp->v_rdev->si_usecount;
2959 * Same as above, but using the struct cdev *as argument
2962 count_dev(struct cdev *dev)
2967 count = dev->si_usecount;
2973 * Print out a description of a vnode.
2975 static char *typename[] =
2976 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2980 vn_printf(struct vnode *vp, const char *fmt, ...)
2983 char buf[256], buf2[16];
2989 printf("%p: ", (void *)vp);
2990 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2991 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2992 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2995 if (vp->v_vflag & VV_ROOT)
2996 strlcat(buf, "|VV_ROOT", sizeof(buf));
2997 if (vp->v_vflag & VV_ISTTY)
2998 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2999 if (vp->v_vflag & VV_NOSYNC)
3000 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
3001 if (vp->v_vflag & VV_ETERNALDEV)
3002 strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
3003 if (vp->v_vflag & VV_CACHEDLABEL)
3004 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
3005 if (vp->v_vflag & VV_TEXT)
3006 strlcat(buf, "|VV_TEXT", sizeof(buf));
3007 if (vp->v_vflag & VV_COPYONWRITE)
3008 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
3009 if (vp->v_vflag & VV_SYSTEM)
3010 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
3011 if (vp->v_vflag & VV_PROCDEP)
3012 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
3013 if (vp->v_vflag & VV_NOKNOTE)
3014 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
3015 if (vp->v_vflag & VV_DELETED)
3016 strlcat(buf, "|VV_DELETED", sizeof(buf));
3017 if (vp->v_vflag & VV_MD)
3018 strlcat(buf, "|VV_MD", sizeof(buf));
3019 if (vp->v_vflag & VV_FORCEINSMQ)
3020 strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
3021 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
3022 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
3023 VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
3025 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
3026 strlcat(buf, buf2, sizeof(buf));
3028 if (vp->v_iflag & VI_MOUNT)
3029 strlcat(buf, "|VI_MOUNT", sizeof(buf));
3030 if (vp->v_iflag & VI_AGE)
3031 strlcat(buf, "|VI_AGE", sizeof(buf));
3032 if (vp->v_iflag & VI_DOOMED)
3033 strlcat(buf, "|VI_DOOMED", sizeof(buf));
3034 if (vp->v_iflag & VI_FREE)
3035 strlcat(buf, "|VI_FREE", sizeof(buf));
3036 if (vp->v_iflag & VI_ACTIVE)
3037 strlcat(buf, "|VI_ACTIVE", sizeof(buf));
3038 if (vp->v_iflag & VI_DOINGINACT)
3039 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
3040 if (vp->v_iflag & VI_OWEINACT)
3041 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
3042 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
3043 VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
3045 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
3046 strlcat(buf, buf2, sizeof(buf));
3048 printf(" flags (%s)\n", buf + 1);
3049 if (mtx_owned(VI_MTX(vp)))
3050 printf(" VI_LOCKed");
3051 if (vp->v_object != NULL)
3052 printf(" v_object %p ref %d pages %d "
3053 "cleanbuf %d dirtybuf %d\n",
3054 vp->v_object, vp->v_object->ref_count,
3055 vp->v_object->resident_page_count,
3056 vp->v_bufobj.bo_dirty.bv_cnt,
3057 vp->v_bufobj.bo_clean.bv_cnt);
3059 lockmgr_printinfo(vp->v_vnlock);
3060 if (vp->v_data != NULL)
3066 * List all of the locked vnodes in the system.
3067 * Called when debugging the kernel.
3069 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
3071 struct mount *mp, *nmp;
3075 * Note: because this is DDB, we can't obey the locking semantics
3076 * for these structures, which means we could catch an inconsistent
3077 * state and dereference a nasty pointer. Not much to be done
3080 db_printf("Locked vnodes\n");
3081 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
3082 nmp = TAILQ_NEXT(mp, mnt_list);
3083 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3084 if (vp->v_type != VMARKER &&
3088 nmp = TAILQ_NEXT(mp, mnt_list);
3093 * Show details about the given vnode.
3095 DB_SHOW_COMMAND(vnode, db_show_vnode)
3101 vp = (struct vnode *)addr;
3102 vn_printf(vp, "vnode ");
3106 * Show details about the given mount point.
3108 DB_SHOW_COMMAND(mount, db_show_mount)
3119 /* No address given, print short info about all mount points. */
3120 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3121 db_printf("%p %s on %s (%s)\n", mp,
3122 mp->mnt_stat.f_mntfromname,
3123 mp->mnt_stat.f_mntonname,
3124 mp->mnt_stat.f_fstypename);
3128 db_printf("\nMore info: show mount <addr>\n");
3132 mp = (struct mount *)addr;
3133 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
3134 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
3137 mflags = mp->mnt_flag;
3138 #define MNT_FLAG(flag) do { \
3139 if (mflags & (flag)) { \
3140 if (buf[0] != '\0') \
3141 strlcat(buf, ", ", sizeof(buf)); \
3142 strlcat(buf, (#flag) + 4, sizeof(buf)); \
3143 mflags &= ~(flag); \
3146 MNT_FLAG(MNT_RDONLY);
3147 MNT_FLAG(MNT_SYNCHRONOUS);
3148 MNT_FLAG(MNT_NOEXEC);
3149 MNT_FLAG(MNT_NOSUID);
3150 MNT_FLAG(MNT_NFS4ACLS);
3151 MNT_FLAG(MNT_UNION);
3152 MNT_FLAG(MNT_ASYNC);
3153 MNT_FLAG(MNT_SUIDDIR);
3154 MNT_FLAG(MNT_SOFTDEP);
3155 MNT_FLAG(MNT_NOSYMFOLLOW);
3156 MNT_FLAG(MNT_GJOURNAL);
3157 MNT_FLAG(MNT_MULTILABEL);
3159 MNT_FLAG(MNT_NOATIME);
3160 MNT_FLAG(MNT_NOCLUSTERR);
3161 MNT_FLAG(MNT_NOCLUSTERW);
3163 MNT_FLAG(MNT_EXRDONLY);
3164 MNT_FLAG(MNT_EXPORTED);
3165 MNT_FLAG(MNT_DEFEXPORTED);
3166 MNT_FLAG(MNT_EXPORTANON);
3167 MNT_FLAG(MNT_EXKERB);
3168 MNT_FLAG(MNT_EXPUBLIC);
3169 MNT_FLAG(MNT_LOCAL);
3170 MNT_FLAG(MNT_QUOTA);
3171 MNT_FLAG(MNT_ROOTFS);
3173 MNT_FLAG(MNT_IGNORE);
3174 MNT_FLAG(MNT_UPDATE);
3175 MNT_FLAG(MNT_DELEXPORT);
3176 MNT_FLAG(MNT_RELOAD);
3177 MNT_FLAG(MNT_FORCE);
3178 MNT_FLAG(MNT_SNAPSHOT);
3179 MNT_FLAG(MNT_BYFSID);
3183 strlcat(buf, ", ", sizeof(buf));
3184 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3185 "0x%016jx", mflags);
3187 db_printf(" mnt_flag = %s\n", buf);
3190 flags = mp->mnt_kern_flag;
3191 #define MNT_KERN_FLAG(flag) do { \
3192 if (flags & (flag)) { \
3193 if (buf[0] != '\0') \
3194 strlcat(buf, ", ", sizeof(buf)); \
3195 strlcat(buf, (#flag) + 5, sizeof(buf)); \
3199 MNT_KERN_FLAG(MNTK_UNMOUNTF);
3200 MNT_KERN_FLAG(MNTK_ASYNC);
3201 MNT_KERN_FLAG(MNTK_SOFTDEP);
3202 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3203 MNT_KERN_FLAG(MNTK_DRAINING);
3204 MNT_KERN_FLAG(MNTK_REFEXPIRE);
3205 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3206 MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3207 MNT_KERN_FLAG(MNTK_NO_IOPF);
3208 MNT_KERN_FLAG(MNTK_VGONE_UPPER);
3209 MNT_KERN_FLAG(MNTK_VGONE_WAITER);
3210 MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
3211 MNT_KERN_FLAG(MNTK_MARKER);
3212 MNT_KERN_FLAG(MNTK_NOASYNC);
3213 MNT_KERN_FLAG(MNTK_UNMOUNT);
3214 MNT_KERN_FLAG(MNTK_MWAIT);
3215 MNT_KERN_FLAG(MNTK_SUSPEND);
3216 MNT_KERN_FLAG(MNTK_SUSPEND2);
3217 MNT_KERN_FLAG(MNTK_SUSPENDED);
3218 MNT_KERN_FLAG(MNTK_MPSAFE);
3219 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3220 MNT_KERN_FLAG(MNTK_NOKNOTE);
3221 #undef MNT_KERN_FLAG
3224 strlcat(buf, ", ", sizeof(buf));
3225 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3228 db_printf(" mnt_kern_flag = %s\n", buf);
3230 db_printf(" mnt_opt = ");
3231 opt = TAILQ_FIRST(mp->mnt_opt);
3233 db_printf("%s", opt->name);
3234 opt = TAILQ_NEXT(opt, link);
3235 while (opt != NULL) {
3236 db_printf(", %s", opt->name);
3237 opt = TAILQ_NEXT(opt, link);
3243 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
3244 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3245 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3246 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3247 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3248 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3249 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3250 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3251 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3252 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3253 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3254 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3256 db_printf(" mnt_cred = { uid=%u ruid=%u",
3257 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3258 if (jailed(mp->mnt_cred))
3259 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3261 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
3262 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
3263 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3264 db_printf(" mnt_activevnodelistsize = %d\n",
3265 mp->mnt_activevnodelistsize);
3266 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3267 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3268 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3269 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
3270 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3271 db_printf(" mnt_secondary_accwrites = %d\n",
3272 mp->mnt_secondary_accwrites);
3273 db_printf(" mnt_gjprovider = %s\n",
3274 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3276 db_printf("\n\nList of active vnodes\n");
3277 TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3278 if (vp->v_type != VMARKER) {
3279 vn_printf(vp, "vnode ");
3284 db_printf("\n\nList of inactive vnodes\n");
3285 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3286 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3287 vn_printf(vp, "vnode ");
3296 * Fill in a struct xvfsconf based on a struct vfsconf.
3299 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3301 struct xvfsconf xvfsp;
3303 bzero(&xvfsp, sizeof(xvfsp));
3304 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3305 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3306 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3307 xvfsp.vfc_flags = vfsp->vfc_flags;
3309 * These are unused in userland, we keep them
3310 * to not break binary compatibility.
3312 xvfsp.vfc_vfsops = NULL;
3313 xvfsp.vfc_next = NULL;
3314 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3317 #ifdef COMPAT_FREEBSD32
3319 uint32_t vfc_vfsops;
3320 char vfc_name[MFSNAMELEN];
3321 int32_t vfc_typenum;
3322 int32_t vfc_refcount;
3328 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3330 struct xvfsconf32 xvfsp;
3332 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3333 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3334 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3335 xvfsp.vfc_flags = vfsp->vfc_flags;
3336 xvfsp.vfc_vfsops = 0;
3338 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3343 * Top level filesystem related information gathering.
3346 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3348 struct vfsconf *vfsp;
3352 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3353 #ifdef COMPAT_FREEBSD32
3354 if (req->flags & SCTL_MASK32)
3355 error = vfsconf2x32(req, vfsp);
3358 error = vfsconf2x(req, vfsp);
3365 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3366 NULL, 0, sysctl_vfs_conflist,
3367 "S,xvfsconf", "List of all configured filesystems");
3369 #ifndef BURN_BRIDGES
3370 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3373 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3375 int *name = (int *)arg1 - 1; /* XXX */
3376 u_int namelen = arg2 + 1; /* XXX */
3377 struct vfsconf *vfsp;
3379 printf("WARNING: userland calling deprecated sysctl, "
3380 "please rebuild world\n");
3382 #if 1 || defined(COMPAT_PRELITE2)
3383 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3385 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3389 case VFS_MAXTYPENUM:
3392 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3395 return (ENOTDIR); /* overloaded */
3396 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3397 if (vfsp->vfc_typenum == name[2])
3400 return (EOPNOTSUPP);
3401 #ifdef COMPAT_FREEBSD32
3402 if (req->flags & SCTL_MASK32)
3403 return (vfsconf2x32(req, vfsp));
3406 return (vfsconf2x(req, vfsp));
3408 return (EOPNOTSUPP);
3411 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3412 vfs_sysctl, "Generic filesystem");
3414 #if 1 || defined(COMPAT_PRELITE2)
3417 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3420 struct vfsconf *vfsp;
3421 struct ovfsconf ovfs;
3423 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3424 bzero(&ovfs, sizeof(ovfs));
3425 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3426 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3427 ovfs.vfc_index = vfsp->vfc_typenum;
3428 ovfs.vfc_refcount = vfsp->vfc_refcount;
3429 ovfs.vfc_flags = vfsp->vfc_flags;
3430 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3437 #endif /* 1 || COMPAT_PRELITE2 */
3438 #endif /* !BURN_BRIDGES */
3440 #define KINFO_VNODESLOP 10
3443 * Dump vnode list (via sysctl).
3447 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3455 * Stale numvnodes access is not fatal here.
3458 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3460 /* Make an estimate */
3461 return (SYSCTL_OUT(req, 0, len));
3463 error = sysctl_wire_old_buffer(req, 0);
3466 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3468 mtx_lock(&mountlist_mtx);
3469 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3470 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3473 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3477 xvn[n].xv_size = sizeof *xvn;
3478 xvn[n].xv_vnode = vp;
3479 xvn[n].xv_id = 0; /* XXX compat */
3480 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3482 XV_COPY(writecount);
3488 xvn[n].xv_flag = vp->v_vflag;
3490 switch (vp->v_type) {
3497 if (vp->v_rdev == NULL) {
3501 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3504 xvn[n].xv_socket = vp->v_socket;
3507 xvn[n].xv_fifo = vp->v_fifoinfo;
3512 /* shouldn't happen? */
3520 mtx_lock(&mountlist_mtx);
3525 mtx_unlock(&mountlist_mtx);
3527 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3532 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3533 0, 0, sysctl_vnode, "S,xvnode", "");
3537 * Unmount all filesystems. The list is traversed in reverse order
3538 * of mounting to avoid dependencies.
3541 vfs_unmountall(void)
3547 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3551 * Since this only runs when rebooting, it is not interlocked.
3553 while(!TAILQ_EMPTY(&mountlist)) {
3554 mp = TAILQ_LAST(&mountlist, mntlist);
3555 error = dounmount(mp, MNT_FORCE, td);
3557 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3559 * XXX: Due to the way in which we mount the root
3560 * file system off of devfs, devfs will generate a
3561 * "busy" warning when we try to unmount it before
3562 * the root. Don't print a warning as a result in
3563 * order to avoid false positive errors that may
3564 * cause needless upset.
3566 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3567 printf("unmount of %s failed (",
3568 mp->mnt_stat.f_mntonname);
3572 printf("%d)\n", error);
3575 /* The unmount has removed mp from the mountlist */
3581 * perform msync on all vnodes under a mount point
3582 * the mount point must be locked.
3585 vfs_msync(struct mount *mp, int flags)
3587 struct vnode *vp, *mvp;
3588 struct vm_object *obj;
3590 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3591 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3593 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3594 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3596 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3598 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3605 VM_OBJECT_LOCK(obj);
3606 vm_object_page_clean(obj, 0, 0,
3608 OBJPC_SYNC : OBJPC_NOSYNC);
3609 VM_OBJECT_UNLOCK(obj);
3619 destroy_vpollinfo_free(struct vpollinfo *vi)
3622 knlist_destroy(&vi->vpi_selinfo.si_note);
3623 mtx_destroy(&vi->vpi_lock);
3624 uma_zfree(vnodepoll_zone, vi);
3628 destroy_vpollinfo(struct vpollinfo *vi)
3631 knlist_clear(&vi->vpi_selinfo.si_note, 1);
3632 seldrain(&vi->vpi_selinfo);
3633 destroy_vpollinfo_free(vi);
3637 * Initalize per-vnode helper structure to hold poll-related state.
3640 v_addpollinfo(struct vnode *vp)
3642 struct vpollinfo *vi;
3644 if (vp->v_pollinfo != NULL)
3646 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3647 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3648 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3649 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3651 if (vp->v_pollinfo != NULL) {
3653 destroy_vpollinfo_free(vi);
3656 vp->v_pollinfo = vi;
3661 * Record a process's interest in events which might happen to
3662 * a vnode. Because poll uses the historic select-style interface
3663 * internally, this routine serves as both the ``check for any
3664 * pending events'' and the ``record my interest in future events''
3665 * functions. (These are done together, while the lock is held,
3666 * to avoid race conditions.)
3669 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3673 mtx_lock(&vp->v_pollinfo->vpi_lock);
3674 if (vp->v_pollinfo->vpi_revents & events) {
3676 * This leaves events we are not interested
3677 * in available for the other process which
3678 * which presumably had requested them
3679 * (otherwise they would never have been
3682 events &= vp->v_pollinfo->vpi_revents;
3683 vp->v_pollinfo->vpi_revents &= ~events;
3685 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3688 vp->v_pollinfo->vpi_events |= events;
3689 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3690 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3695 * Routine to create and manage a filesystem syncer vnode.
3697 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3698 static int sync_fsync(struct vop_fsync_args *);
3699 static int sync_inactive(struct vop_inactive_args *);
3700 static int sync_reclaim(struct vop_reclaim_args *);
3702 static struct vop_vector sync_vnodeops = {
3703 .vop_bypass = VOP_EOPNOTSUPP,
3704 .vop_close = sync_close, /* close */
3705 .vop_fsync = sync_fsync, /* fsync */
3706 .vop_inactive = sync_inactive, /* inactive */
3707 .vop_reclaim = sync_reclaim, /* reclaim */
3708 .vop_lock1 = vop_stdlock, /* lock */
3709 .vop_unlock = vop_stdunlock, /* unlock */
3710 .vop_islocked = vop_stdislocked, /* islocked */
3714 * Create a new filesystem syncer vnode for the specified mount point.
3717 vfs_allocate_syncvnode(struct mount *mp)
3721 static long start, incr, next;
3724 /* Allocate a new vnode */
3725 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3727 panic("vfs_allocate_syncvnode: getnewvnode() failed");
3729 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3730 vp->v_vflag |= VV_FORCEINSMQ;
3731 error = insmntque(vp, mp);
3733 panic("vfs_allocate_syncvnode: insmntque() failed");
3734 vp->v_vflag &= ~VV_FORCEINSMQ;
3737 * Place the vnode onto the syncer worklist. We attempt to
3738 * scatter them about on the list so that they will go off
3739 * at evenly distributed times even if all the filesystems
3740 * are mounted at once.
3743 if (next == 0 || next > syncer_maxdelay) {
3747 start = syncer_maxdelay / 2;
3748 incr = syncer_maxdelay;
3754 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3755 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3756 mtx_lock(&sync_mtx);
3758 if (mp->mnt_syncer == NULL) {
3759 mp->mnt_syncer = vp;
3762 mtx_unlock(&sync_mtx);
3765 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3772 vfs_deallocate_syncvnode(struct mount *mp)
3776 mtx_lock(&sync_mtx);
3777 vp = mp->mnt_syncer;
3779 mp->mnt_syncer = NULL;
3780 mtx_unlock(&sync_mtx);
3786 * Do a lazy sync of the filesystem.
3789 sync_fsync(struct vop_fsync_args *ap)
3791 struct vnode *syncvp = ap->a_vp;
3792 struct mount *mp = syncvp->v_mount;
3797 * We only need to do something if this is a lazy evaluation.
3799 if (ap->a_waitfor != MNT_LAZY)
3803 * Move ourselves to the back of the sync list.
3805 bo = &syncvp->v_bufobj;
3807 vn_syncer_add_to_worklist(bo, syncdelay);
3811 * Walk the list of vnodes pushing all that are dirty and
3812 * not already on the sync list.
3814 if (vfs_busy(mp, MBF_NOWAIT) != 0)
3816 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3820 save = curthread_pflags_set(TDP_SYNCIO);
3821 vfs_msync(mp, MNT_NOWAIT);
3822 error = VFS_SYNC(mp, MNT_LAZY);
3823 curthread_pflags_restore(save);
3824 vn_finished_write(mp);
3830 * The syncer vnode is no referenced.
3833 sync_inactive(struct vop_inactive_args *ap)
3841 * The syncer vnode is no longer needed and is being decommissioned.
3843 * Modifications to the worklist must be protected by sync_mtx.
3846 sync_reclaim(struct vop_reclaim_args *ap)
3848 struct vnode *vp = ap->a_vp;
3853 mtx_lock(&sync_mtx);
3854 if (vp->v_mount->mnt_syncer == vp)
3855 vp->v_mount->mnt_syncer = NULL;
3856 if (bo->bo_flag & BO_ONWORKLST) {
3857 LIST_REMOVE(bo, bo_synclist);
3858 syncer_worklist_len--;
3860 bo->bo_flag &= ~BO_ONWORKLST;
3862 mtx_unlock(&sync_mtx);
3869 * Check if vnode represents a disk device
3872 vn_isdisk(struct vnode *vp, int *errp)
3878 if (vp->v_type != VCHR)
3880 else if (vp->v_rdev == NULL)
3882 else if (vp->v_rdev->si_devsw == NULL)
3884 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3889 return (error == 0);
3893 * Common filesystem object access control check routine. Accepts a
3894 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3895 * and optional call-by-reference privused argument allowing vaccess()
3896 * to indicate to the caller whether privilege was used to satisfy the
3897 * request (obsoleted). Returns 0 on success, or an errno on failure.
3900 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3901 accmode_t accmode, struct ucred *cred, int *privused)
3903 accmode_t dac_granted;
3904 accmode_t priv_granted;
3906 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3907 ("invalid bit in accmode"));
3908 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3909 ("VAPPEND without VWRITE"));
3912 * Look for a normal, non-privileged way to access the file/directory
3913 * as requested. If it exists, go with that.
3916 if (privused != NULL)
3921 /* Check the owner. */
3922 if (cred->cr_uid == file_uid) {
3923 dac_granted |= VADMIN;
3924 if (file_mode & S_IXUSR)
3925 dac_granted |= VEXEC;
3926 if (file_mode & S_IRUSR)
3927 dac_granted |= VREAD;
3928 if (file_mode & S_IWUSR)
3929 dac_granted |= (VWRITE | VAPPEND);
3931 if ((accmode & dac_granted) == accmode)
3937 /* Otherwise, check the groups (first match) */
3938 if (groupmember(file_gid, cred)) {
3939 if (file_mode & S_IXGRP)
3940 dac_granted |= VEXEC;
3941 if (file_mode & S_IRGRP)
3942 dac_granted |= VREAD;
3943 if (file_mode & S_IWGRP)
3944 dac_granted |= (VWRITE | VAPPEND);
3946 if ((accmode & dac_granted) == accmode)
3952 /* Otherwise, check everyone else. */
3953 if (file_mode & S_IXOTH)
3954 dac_granted |= VEXEC;
3955 if (file_mode & S_IROTH)
3956 dac_granted |= VREAD;
3957 if (file_mode & S_IWOTH)
3958 dac_granted |= (VWRITE | VAPPEND);
3959 if ((accmode & dac_granted) == accmode)
3964 * Build a privilege mask to determine if the set of privileges
3965 * satisfies the requirements when combined with the granted mask
3966 * from above. For each privilege, if the privilege is required,
3967 * bitwise or the request type onto the priv_granted mask.
3973 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3974 * requests, instead of PRIV_VFS_EXEC.
3976 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3977 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3978 priv_granted |= VEXEC;
3981 * Ensure that at least one execute bit is on. Otherwise,
3982 * a privileged user will always succeed, and we don't want
3983 * this to happen unless the file really is executable.
3985 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3986 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3987 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3988 priv_granted |= VEXEC;
3991 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3992 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3993 priv_granted |= VREAD;
3995 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3996 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3997 priv_granted |= (VWRITE | VAPPEND);
3999 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
4000 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
4001 priv_granted |= VADMIN;
4003 if ((accmode & (priv_granted | dac_granted)) == accmode) {
4004 /* XXX audit: privilege used */
4005 if (privused != NULL)
4010 return ((accmode & VADMIN) ? EPERM : EACCES);
4014 * Credential check based on process requesting service, and per-attribute
4018 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
4019 struct thread *td, accmode_t accmode)
4023 * Kernel-invoked always succeeds.
4029 * Do not allow privileged processes in jail to directly manipulate
4030 * system attributes.
4032 switch (attrnamespace) {
4033 case EXTATTR_NAMESPACE_SYSTEM:
4034 /* Potentially should be: return (EPERM); */
4035 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
4036 case EXTATTR_NAMESPACE_USER:
4037 return (VOP_ACCESS(vp, accmode, cred, td));
4043 #ifdef DEBUG_VFS_LOCKS
4045 * This only exists to supress warnings from unlocked specfs accesses. It is
4046 * no longer ok to have an unlocked VFS.
4048 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
4049 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
4051 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
4052 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
4053 "Drop into debugger on lock violation");
4055 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
4056 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
4057 0, "Check for interlock across VOPs");
4059 int vfs_badlock_print = 1; /* Print lock violations. */
4060 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
4061 0, "Print lock violations");
4064 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
4065 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
4066 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
4070 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
4074 if (vfs_badlock_backtrace)
4077 if (vfs_badlock_print)
4078 printf("%s: %p %s\n", str, (void *)vp, msg);
4079 if (vfs_badlock_ddb)
4080 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4084 assert_vi_locked(struct vnode *vp, const char *str)
4087 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
4088 vfs_badlock("interlock is not locked but should be", str, vp);
4092 assert_vi_unlocked(struct vnode *vp, const char *str)
4095 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
4096 vfs_badlock("interlock is locked but should not be", str, vp);
4100 assert_vop_locked(struct vnode *vp, const char *str)
4104 if (!IGNORE_LOCK(vp)) {
4105 locked = VOP_ISLOCKED(vp);
4106 if (locked == 0 || locked == LK_EXCLOTHER)
4107 vfs_badlock("is not locked but should be", str, vp);
4112 assert_vop_unlocked(struct vnode *vp, const char *str)
4115 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
4116 vfs_badlock("is locked but should not be", str, vp);
4120 assert_vop_elocked(struct vnode *vp, const char *str)
4123 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
4124 vfs_badlock("is not exclusive locked but should be", str, vp);
4129 assert_vop_elocked_other(struct vnode *vp, const char *str)
4132 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
4133 vfs_badlock("is not exclusive locked by another thread",
4138 assert_vop_slocked(struct vnode *vp, const char *str)
4141 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
4142 vfs_badlock("is not locked shared but should be", str, vp);
4145 #endif /* DEBUG_VFS_LOCKS */
4148 vop_rename_fail(struct vop_rename_args *ap)
4151 if (ap->a_tvp != NULL)
4153 if (ap->a_tdvp == ap->a_tvp)
4162 vop_rename_pre(void *ap)
4164 struct vop_rename_args *a = ap;
4166 #ifdef DEBUG_VFS_LOCKS
4168 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4169 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4170 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4171 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4173 /* Check the source (from). */
4174 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4175 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4176 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4177 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4178 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4180 /* Check the target. */
4182 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4183 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4185 if (a->a_tdvp != a->a_fdvp)
4187 if (a->a_tvp != a->a_fvp)
4195 vop_strategy_pre(void *ap)
4197 #ifdef DEBUG_VFS_LOCKS
4198 struct vop_strategy_args *a;
4205 * Cluster ops lock their component buffers but not the IO container.
4207 if ((bp->b_flags & B_CLUSTER) != 0)
4210 if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4211 if (vfs_badlock_print)
4213 "VOP_STRATEGY: bp is not locked but should be\n");
4214 if (vfs_badlock_ddb)
4215 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4221 vop_lock_pre(void *ap)
4223 #ifdef DEBUG_VFS_LOCKS
4224 struct vop_lock1_args *a = ap;
4226 if ((a->a_flags & LK_INTERLOCK) == 0)
4227 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4229 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4234 vop_lock_post(void *ap, int rc)
4236 #ifdef DEBUG_VFS_LOCKS
4237 struct vop_lock1_args *a = ap;
4239 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4240 if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
4241 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4246 vop_unlock_pre(void *ap)
4248 #ifdef DEBUG_VFS_LOCKS
4249 struct vop_unlock_args *a = ap;
4251 if (a->a_flags & LK_INTERLOCK)
4252 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4253 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4258 vop_unlock_post(void *ap, int rc)
4260 #ifdef DEBUG_VFS_LOCKS
4261 struct vop_unlock_args *a = ap;
4263 if (a->a_flags & LK_INTERLOCK)
4264 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4269 vop_create_post(void *ap, int rc)
4271 struct vop_create_args *a = ap;
4274 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4278 vop_deleteextattr_post(void *ap, int rc)
4280 struct vop_deleteextattr_args *a = ap;
4283 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4287 vop_link_post(void *ap, int rc)
4289 struct vop_link_args *a = ap;
4292 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4293 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4298 vop_mkdir_post(void *ap, int rc)
4300 struct vop_mkdir_args *a = ap;
4303 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4307 vop_mknod_post(void *ap, int rc)
4309 struct vop_mknod_args *a = ap;
4312 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4316 vop_remove_post(void *ap, int rc)
4318 struct vop_remove_args *a = ap;
4321 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4322 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4327 vop_rename_post(void *ap, int rc)
4329 struct vop_rename_args *a = ap;
4332 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4333 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4334 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4336 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4338 if (a->a_tdvp != a->a_fdvp)
4340 if (a->a_tvp != a->a_fvp)
4348 vop_rmdir_post(void *ap, int rc)
4350 struct vop_rmdir_args *a = ap;
4353 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4354 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4359 vop_setattr_post(void *ap, int rc)
4361 struct vop_setattr_args *a = ap;
4364 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4368 vop_setextattr_post(void *ap, int rc)
4370 struct vop_setextattr_args *a = ap;
4373 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4377 vop_symlink_post(void *ap, int rc)
4379 struct vop_symlink_args *a = ap;
4382 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4385 static struct knlist fs_knlist;
4388 vfs_event_init(void *arg)
4390 knlist_init_mtx(&fs_knlist, NULL);
4392 /* XXX - correct order? */
4393 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4396 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4399 KNOTE_UNLOCKED(&fs_knlist, event);
4402 static int filt_fsattach(struct knote *kn);
4403 static void filt_fsdetach(struct knote *kn);
4404 static int filt_fsevent(struct knote *kn, long hint);
4406 struct filterops fs_filtops = {
4408 .f_attach = filt_fsattach,
4409 .f_detach = filt_fsdetach,
4410 .f_event = filt_fsevent
4414 filt_fsattach(struct knote *kn)
4417 kn->kn_flags |= EV_CLEAR;
4418 knlist_add(&fs_knlist, kn, 0);
4423 filt_fsdetach(struct knote *kn)
4426 knlist_remove(&fs_knlist, kn, 0);
4430 filt_fsevent(struct knote *kn, long hint)
4433 kn->kn_fflags |= hint;
4434 return (kn->kn_fflags != 0);
4438 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4444 error = SYSCTL_IN(req, &vc, sizeof(vc));
4447 if (vc.vc_vers != VFS_CTL_VERS1)
4449 mp = vfs_getvfs(&vc.vc_fsid);
4452 /* ensure that a specific sysctl goes to the right filesystem. */
4453 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4454 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4458 VCTLTOREQ(&vc, req);
4459 error = VFS_SYSCTL(mp, vc.vc_op, req);
4464 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4465 NULL, 0, sysctl_vfs_ctl, "",
4469 * Function to initialize a va_filerev field sensibly.
4470 * XXX: Wouldn't a random number make a lot more sense ??
4473 init_va_filerev(void)
4478 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4481 static int filt_vfsread(struct knote *kn, long hint);
4482 static int filt_vfswrite(struct knote *kn, long hint);
4483 static int filt_vfsvnode(struct knote *kn, long hint);
4484 static void filt_vfsdetach(struct knote *kn);
4485 static struct filterops vfsread_filtops = {
4487 .f_detach = filt_vfsdetach,
4488 .f_event = filt_vfsread
4490 static struct filterops vfswrite_filtops = {
4492 .f_detach = filt_vfsdetach,
4493 .f_event = filt_vfswrite
4495 static struct filterops vfsvnode_filtops = {
4497 .f_detach = filt_vfsdetach,
4498 .f_event = filt_vfsvnode
4502 vfs_knllock(void *arg)
4504 struct vnode *vp = arg;
4506 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4510 vfs_knlunlock(void *arg)
4512 struct vnode *vp = arg;
4518 vfs_knl_assert_locked(void *arg)
4520 #ifdef DEBUG_VFS_LOCKS
4521 struct vnode *vp = arg;
4523 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4528 vfs_knl_assert_unlocked(void *arg)
4530 #ifdef DEBUG_VFS_LOCKS
4531 struct vnode *vp = arg;
4533 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4538 vfs_kqfilter(struct vop_kqfilter_args *ap)
4540 struct vnode *vp = ap->a_vp;
4541 struct knote *kn = ap->a_kn;
4544 switch (kn->kn_filter) {
4546 kn->kn_fop = &vfsread_filtops;
4549 kn->kn_fop = &vfswrite_filtops;
4552 kn->kn_fop = &vfsvnode_filtops;
4558 kn->kn_hook = (caddr_t)vp;
4561 if (vp->v_pollinfo == NULL)
4563 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4565 knlist_add(knl, kn, 0);
4571 * Detach knote from vnode
4574 filt_vfsdetach(struct knote *kn)
4576 struct vnode *vp = (struct vnode *)kn->kn_hook;
4578 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4579 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4585 filt_vfsread(struct knote *kn, long hint)
4587 struct vnode *vp = (struct vnode *)kn->kn_hook;
4592 * filesystem is gone, so set the EOF flag and schedule
4593 * the knote for deletion.
4595 if (hint == NOTE_REVOKE) {
4597 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4602 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4606 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4607 res = (kn->kn_data != 0);
4614 filt_vfswrite(struct knote *kn, long hint)
4616 struct vnode *vp = (struct vnode *)kn->kn_hook;
4621 * filesystem is gone, so set the EOF flag and schedule
4622 * the knote for deletion.
4624 if (hint == NOTE_REVOKE)
4625 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4633 filt_vfsvnode(struct knote *kn, long hint)
4635 struct vnode *vp = (struct vnode *)kn->kn_hook;
4639 if (kn->kn_sfflags & hint)
4640 kn->kn_fflags |= hint;
4641 if (hint == NOTE_REVOKE) {
4642 kn->kn_flags |= EV_EOF;
4646 res = (kn->kn_fflags != 0);
4652 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4656 if (dp->d_reclen > ap->a_uio->uio_resid)
4657 return (ENAMETOOLONG);
4658 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4660 if (ap->a_ncookies != NULL) {
4661 if (ap->a_cookies != NULL)
4662 free(ap->a_cookies, M_TEMP);
4663 ap->a_cookies = NULL;
4664 *ap->a_ncookies = 0;
4668 if (ap->a_ncookies == NULL)
4671 KASSERT(ap->a_cookies,
4672 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4674 *ap->a_cookies = realloc(*ap->a_cookies,
4675 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4676 (*ap->a_cookies)[*ap->a_ncookies] = off;
4681 * Mark for update the access time of the file if the filesystem
4682 * supports VOP_MARKATIME. This functionality is used by execve and
4683 * mmap, so we want to avoid the I/O implied by directly setting
4684 * va_atime for the sake of efficiency.
4687 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4692 VFS_ASSERT_GIANT(mp);
4693 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4694 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4695 (void)VOP_MARKATIME(vp);
4699 * The purpose of this routine is to remove granularity from accmode_t,
4700 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4701 * VADMIN and VAPPEND.
4703 * If it returns 0, the caller is supposed to continue with the usual
4704 * access checks using 'accmode' as modified by this routine. If it
4705 * returns nonzero value, the caller is supposed to return that value
4708 * Note that after this routine runs, accmode may be zero.
4711 vfs_unixify_accmode(accmode_t *accmode)
4714 * There is no way to specify explicit "deny" rule using
4715 * file mode or POSIX.1e ACLs.
4717 if (*accmode & VEXPLICIT_DENY) {
4723 * None of these can be translated into usual access bits.
4724 * Also, the common case for NFSv4 ACLs is to not contain
4725 * either of these bits. Caller should check for VWRITE
4726 * on the containing directory instead.
4728 if (*accmode & (VDELETE_CHILD | VDELETE))
4731 if (*accmode & VADMIN_PERMS) {
4732 *accmode &= ~VADMIN_PERMS;
4737 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4738 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4740 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4746 * These are helper functions for filesystems to traverse all
4747 * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4749 * This interface replaces MNT_VNODE_FOREACH.
4752 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4755 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4760 kern_yield(PRI_UNCHANGED);
4762 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4763 vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4764 while (vp != NULL && (vp->v_type == VMARKER ||
4765 (vp->v_iflag & VI_DOOMED) != 0))
4766 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4768 /* Check if we are done */
4770 __mnt_vnode_markerfree_all(mvp, mp);
4771 /* MNT_IUNLOCK(mp); -- done in above function */
4772 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4775 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4776 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4783 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4787 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4790 (*mvp)->v_type = VMARKER;
4792 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4793 while (vp != NULL && (vp->v_type == VMARKER ||
4794 (vp->v_iflag & VI_DOOMED) != 0))
4795 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4797 /* Check if we are done */
4801 free(*mvp, M_VNODE_MARKER);
4805 (*mvp)->v_mount = mp;
4806 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4814 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4822 mtx_assert(MNT_MTX(mp), MA_OWNED);
4824 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4825 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4828 free(*mvp, M_VNODE_MARKER);
4833 * These are helper functions for filesystems to traverse their
4834 * active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4837 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4840 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4845 free(*mvp, M_VNODE_MARKER);
4850 #define ALWAYS_YIELD (mp_ncpus == 1)
4852 #define ALWAYS_YIELD 1
4855 static struct vnode *
4856 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4858 struct vnode *vp, *nvp;
4860 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
4861 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4863 vp = TAILQ_NEXT(*mvp, v_actfreelist);
4864 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4865 while (vp != NULL) {
4866 if (vp->v_type == VMARKER) {
4867 vp = TAILQ_NEXT(vp, v_actfreelist);
4870 if (!VI_TRYLOCK(vp)) {
4871 if (ALWAYS_YIELD || should_yield()) {
4872 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4873 mtx_unlock(&vnode_free_list_mtx);
4875 mtx_lock(&vnode_free_list_mtx);
4880 KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
4881 KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
4882 ("alien vnode on the active list %p %p", vp, mp));
4883 if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
4885 nvp = TAILQ_NEXT(vp, v_actfreelist);
4890 /* Check if we are done */
4892 mtx_unlock(&vnode_free_list_mtx);
4893 mnt_vnode_markerfree_active(mvp, mp);
4896 TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4897 mtx_unlock(&vnode_free_list_mtx);
4898 ASSERT_VI_LOCKED(vp, "active iter");
4899 KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
4905 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4909 kern_yield(PRI_UNCHANGED);
4910 mtx_lock(&vnode_free_list_mtx);
4911 return (mnt_vnode_next_active(mvp, mp));
4915 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4919 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4923 (*mvp)->v_type = VMARKER;
4924 (*mvp)->v_mount = mp;
4926 mtx_lock(&vnode_free_list_mtx);
4927 vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
4929 mtx_unlock(&vnode_free_list_mtx);
4930 mnt_vnode_markerfree_active(mvp, mp);
4933 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4934 return (mnt_vnode_next_active(mvp, mp));
4938 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4944 mtx_lock(&vnode_free_list_mtx);
4945 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4946 mtx_unlock(&vnode_free_list_mtx);
4947 mnt_vnode_markerfree_active(mvp, mp);