2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
38 * External virtual filesystem routines
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
45 #include "opt_watchdog.h"
47 #include <sys/param.h>
48 #include <sys/systm.h>
51 #include <sys/condvar.h>
53 #include <sys/dirent.h>
54 #include <sys/event.h>
55 #include <sys/eventhandler.h>
56 #include <sys/extattr.h>
58 #include <sys/fcntl.h>
61 #include <sys/kernel.h>
62 #include <sys/kthread.h>
63 #include <sys/lockf.h>
64 #include <sys/malloc.h>
65 #include <sys/mount.h>
66 #include <sys/namei.h>
68 #include <sys/reboot.h>
69 #include <sys/sched.h>
70 #include <sys/sleepqueue.h>
72 #include <sys/sysctl.h>
73 #include <sys/syslog.h>
74 #include <sys/vmmeter.h>
75 #include <sys/vnode.h>
76 #include <sys/watchdog.h>
78 #include <machine/stdarg.h>
80 #include <security/mac/mac_framework.h>
83 #include <vm/vm_object.h>
84 #include <vm/vm_extern.h>
86 #include <vm/vm_map.h>
87 #include <vm/vm_page.h>
88 #include <vm/vm_kern.h>
98 static void delmntque(struct vnode *vp);
99 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
100 int slpflag, int slptimeo);
101 static void syncer_shutdown(void *arg, int howto);
102 static int vtryrecycle(struct vnode *vp);
103 static void v_incr_usecount(struct vnode *);
104 static void v_decr_usecount(struct vnode *);
105 static void v_decr_useonly(struct vnode *);
106 static void v_upgrade_usecount(struct vnode *);
107 static void vnlru_free(int);
108 static void vgonel(struct vnode *);
109 static void vfs_knllock(void *arg);
110 static void vfs_knlunlock(void *arg);
111 static void vfs_knl_assert_locked(void *arg);
112 static void vfs_knl_assert_unlocked(void *arg);
113 static void destroy_vpollinfo(struct vpollinfo *vi);
116 * Number of vnodes in existence. Increased whenever getnewvnode()
117 * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
119 static unsigned long numvnodes;
121 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
122 "Number of vnodes in existence");
125 * Conversion tables for conversion from vnode types to inode formats
128 enum vtype iftovt_tab[16] = {
129 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
130 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
132 int vttoif_tab[10] = {
133 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
134 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
138 * List of vnodes that are ready for recycling.
140 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
143 * Free vnode target. Free vnodes may simply be files which have been stat'd
144 * but not read. This is somewhat common, and a small cache of such files
145 * should be kept to avoid recreation costs.
147 static u_long wantfreevnodes;
148 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
149 /* Number of vnodes in the free list. */
150 static u_long freevnodes;
151 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
152 "Number of vnodes in the free list");
154 static int vlru_allow_cache_src;
155 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
156 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
159 * Various variables used for debugging the new implementation of
161 * XXX these are probably of (very) limited utility now.
163 static int reassignbufcalls;
164 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
165 "Number of calls to reassignbuf");
168 * Cache for the mount type id assigned to NFS. This is used for
169 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
171 int nfs_mount_type = -1;
173 /* To keep more than one thread at a time from running vfs_getnewfsid */
174 static struct mtx mntid_mtx;
177 * Lock for any access to the following:
182 static struct mtx vnode_free_list_mtx;
184 /* Publicly exported FS */
185 struct nfs_public nfs_pub;
187 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
188 static uma_zone_t vnode_zone;
189 static uma_zone_t vnodepoll_zone;
192 * The workitem queue.
194 * It is useful to delay writes of file data and filesystem metadata
195 * for tens of seconds so that quickly created and deleted files need
196 * not waste disk bandwidth being created and removed. To realize this,
197 * we append vnodes to a "workitem" queue. When running with a soft
198 * updates implementation, most pending metadata dependencies should
199 * not wait for more than a few seconds. Thus, mounted on block devices
200 * are delayed only about a half the time that file data is delayed.
201 * Similarly, directory updates are more critical, so are only delayed
202 * about a third the time that file data is delayed. Thus, there are
203 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
204 * one each second (driven off the filesystem syncer process). The
205 * syncer_delayno variable indicates the next queue that is to be processed.
206 * Items that need to be processed soon are placed in this queue:
208 * syncer_workitem_pending[syncer_delayno]
210 * A delay of fifteen seconds is done by placing the request fifteen
211 * entries later in the queue:
213 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
216 static int syncer_delayno;
217 static long syncer_mask;
218 LIST_HEAD(synclist, bufobj);
219 static struct synclist *syncer_workitem_pending[2];
221 * The sync_mtx protects:
226 * syncer_workitem_pending
227 * syncer_worklist_len
230 static struct mtx sync_mtx;
231 static struct cv sync_wakeup;
233 #define SYNCER_MAXDELAY 32
234 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
235 static int syncdelay = 30; /* max time to delay syncing data */
236 static int filedelay = 30; /* time to delay syncing files */
237 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
238 "Time to delay syncing files (in seconds)");
239 static int dirdelay = 29; /* time to delay syncing directories */
240 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
241 "Time to delay syncing directories (in seconds)");
242 static int metadelay = 28; /* time to delay syncing metadata */
243 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
244 "Time to delay syncing metadata (in seconds)");
245 static int rushjob; /* number of slots to run ASAP */
246 static int stat_rush_requests; /* number of times I/O speeded up */
247 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
248 "Number of times I/O speeded up (rush requests)");
251 * When shutting down the syncer, run it at four times normal speed.
253 #define SYNCER_SHUTDOWN_SPEEDUP 4
254 static int sync_vnode_count;
255 static int syncer_worklist_len;
256 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
260 * Number of vnodes we want to exist at any one time. This is mostly used
261 * to size hash tables in vnode-related code. It is normally not used in
262 * getnewvnode(), as wantfreevnodes is normally nonzero.)
264 * XXX desiredvnodes is historical cruft and should not exist.
267 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
268 &desiredvnodes, 0, "Maximum number of vnodes");
269 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
270 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
271 static int vnlru_nowhere;
272 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
273 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
276 * Macros to control when a vnode is freed and recycled. All require
277 * the vnode interlock.
279 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
280 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
281 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
285 * Initialize the vnode management data structures.
287 * Reevaluate the following cap on the number of vnodes after the physical
288 * memory size exceeds 512GB. In the limit, as the physical memory size
289 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
291 #ifndef MAXVNODES_MAX
292 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
295 vntblinit(void *dummy __unused)
297 int physvnodes, virtvnodes;
300 * Desiredvnodes is a function of the physical memory size and the
301 * kernel's heap size. Generally speaking, it scales with the
302 * physical memory size. The ratio of desiredvnodes to physical pages
303 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
304 * marginal ratio of desiredvnodes to physical pages is one to
305 * sixteen. However, desiredvnodes is limited by the kernel's heap
306 * size. The memory required by desiredvnodes vnodes and vm objects
307 * may not exceed one seventh of the kernel's heap size.
309 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
310 cnt.v_page_count) / 16;
311 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
312 sizeof(struct vnode)));
313 desiredvnodes = min(physvnodes, virtvnodes);
314 if (desiredvnodes > MAXVNODES_MAX) {
316 printf("Reducing kern.maxvnodes %d -> %d\n",
317 desiredvnodes, MAXVNODES_MAX);
318 desiredvnodes = MAXVNODES_MAX;
320 wantfreevnodes = desiredvnodes / 4;
321 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
322 TAILQ_INIT(&vnode_free_list);
323 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
324 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
325 NULL, NULL, UMA_ALIGN_PTR, 0);
326 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
327 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
329 * Initialize the filesystem syncer.
331 syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
333 syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
335 syncer_maxdelay = syncer_mask + 1;
336 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
337 cv_init(&sync_wakeup, "syncer");
339 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
343 * Mark a mount point as busy. Used to synchronize access and to delay
344 * unmounting. Eventually, mountlist_mtx is not released on failure.
346 * vfs_busy() is a custom lock, it can block the caller.
347 * vfs_busy() only sleeps if the unmount is active on the mount point.
348 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
349 * vnode belonging to mp.
351 * Lookup uses vfs_busy() to traverse mount points.
353 * / vnode lock A / vnode lock (/var) D
354 * /var vnode lock B /log vnode lock(/var/log) E
355 * vfs_busy lock C vfs_busy lock F
357 * Within each file system, the lock order is C->A->B and F->D->E.
359 * When traversing across mounts, the system follows that lock order:
365 * The lookup() process for namei("/var") illustrates the process:
366 * VOP_LOOKUP() obtains B while A is held
367 * vfs_busy() obtains a shared lock on F while A and B are held
368 * vput() releases lock on B
369 * vput() releases lock on A
370 * VFS_ROOT() obtains lock on D while shared lock on F is held
371 * vfs_unbusy() releases shared lock on F
372 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
373 * Attempt to lock A (instead of vp_crossmp) while D is held would
374 * violate the global order, causing deadlocks.
376 * dounmount() locks B while F is drained.
379 vfs_busy(struct mount *mp, int flags)
382 MPASS((flags & ~MBF_MASK) == 0);
383 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
388 * If mount point is currenly being unmounted, sleep until the
389 * mount point fate is decided. If thread doing the unmounting fails,
390 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
391 * that this mount point has survived the unmount attempt and vfs_busy
392 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
393 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
394 * about to be really destroyed. vfs_busy needs to release its
395 * reference on the mount point in this case and return with ENOENT,
396 * telling the caller that mount mount it tried to busy is no longer
399 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
400 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
403 CTR1(KTR_VFS, "%s: failed busying before sleeping",
407 if (flags & MBF_MNTLSTLOCK)
408 mtx_unlock(&mountlist_mtx);
409 mp->mnt_kern_flag |= MNTK_MWAIT;
410 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
411 if (flags & MBF_MNTLSTLOCK)
412 mtx_lock(&mountlist_mtx);
415 if (flags & MBF_MNTLSTLOCK)
416 mtx_unlock(&mountlist_mtx);
423 * Free a busy filesystem.
426 vfs_unbusy(struct mount *mp)
429 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
432 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
434 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
435 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
436 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
437 mp->mnt_kern_flag &= ~MNTK_DRAINING;
438 wakeup(&mp->mnt_lockref);
444 * Lookup a mount point by filesystem identifier.
447 vfs_getvfs(fsid_t *fsid)
451 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
452 mtx_lock(&mountlist_mtx);
453 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
454 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
455 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
457 mtx_unlock(&mountlist_mtx);
461 mtx_unlock(&mountlist_mtx);
462 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
463 return ((struct mount *) 0);
467 * Lookup a mount point by filesystem identifier, busying it before
471 vfs_busyfs(fsid_t *fsid)
476 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
477 mtx_lock(&mountlist_mtx);
478 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
479 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
480 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
481 error = vfs_busy(mp, MBF_MNTLSTLOCK);
483 mtx_unlock(&mountlist_mtx);
489 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
490 mtx_unlock(&mountlist_mtx);
491 return ((struct mount *) 0);
495 * Check if a user can access privileged mount options.
498 vfs_suser(struct mount *mp, struct thread *td)
503 * If the thread is jailed, but this is not a jail-friendly file
504 * system, deny immediately.
506 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
510 * If the file system was mounted outside the jail of the calling
511 * thread, deny immediately.
513 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
517 * If file system supports delegated administration, we don't check
518 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
519 * by the file system itself.
520 * If this is not the user that did original mount, we check for
521 * the PRIV_VFS_MOUNT_OWNER privilege.
523 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
524 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
525 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
532 * Get a new unique fsid. Try to make its val[0] unique, since this value
533 * will be used to create fake device numbers for stat(). Also try (but
534 * not so hard) make its val[0] unique mod 2^16, since some emulators only
535 * support 16-bit device numbers. We end up with unique val[0]'s for the
536 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
538 * Keep in mind that several mounts may be running in parallel. Starting
539 * the search one past where the previous search terminated is both a
540 * micro-optimization and a defense against returning the same fsid to
544 vfs_getnewfsid(struct mount *mp)
546 static uint16_t mntid_base;
551 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
552 mtx_lock(&mntid_mtx);
553 mtype = mp->mnt_vfc->vfc_typenum;
554 tfsid.val[1] = mtype;
555 mtype = (mtype & 0xFF) << 24;
557 tfsid.val[0] = makedev(255,
558 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
560 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
564 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
565 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
566 mtx_unlock(&mntid_mtx);
570 * Knob to control the precision of file timestamps:
572 * 0 = seconds only; nanoseconds zeroed.
573 * 1 = seconds and nanoseconds, accurate within 1/HZ.
574 * 2 = seconds and nanoseconds, truncated to microseconds.
575 * >=3 = seconds and nanoseconds, maximum precision.
577 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
579 static int timestamp_precision = TSP_SEC;
580 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
581 ×tamp_precision, 0, "File timestamp precision (0: seconds, "
582 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
583 "3+: sec + ns (max. precision))");
586 * Get a current timestamp.
589 vfs_timestamp(struct timespec *tsp)
593 switch (timestamp_precision) {
595 tsp->tv_sec = time_second;
603 TIMEVAL_TO_TIMESPEC(&tv, tsp);
613 * Set vnode attributes to VNOVAL
616 vattr_null(struct vattr *vap)
620 vap->va_size = VNOVAL;
621 vap->va_bytes = VNOVAL;
622 vap->va_mode = VNOVAL;
623 vap->va_nlink = VNOVAL;
624 vap->va_uid = VNOVAL;
625 vap->va_gid = VNOVAL;
626 vap->va_fsid = VNOVAL;
627 vap->va_fileid = VNOVAL;
628 vap->va_blocksize = VNOVAL;
629 vap->va_rdev = VNOVAL;
630 vap->va_atime.tv_sec = VNOVAL;
631 vap->va_atime.tv_nsec = VNOVAL;
632 vap->va_mtime.tv_sec = VNOVAL;
633 vap->va_mtime.tv_nsec = VNOVAL;
634 vap->va_ctime.tv_sec = VNOVAL;
635 vap->va_ctime.tv_nsec = VNOVAL;
636 vap->va_birthtime.tv_sec = VNOVAL;
637 vap->va_birthtime.tv_nsec = VNOVAL;
638 vap->va_flags = VNOVAL;
639 vap->va_gen = VNOVAL;
644 * This routine is called when we have too many vnodes. It attempts
645 * to free <count> vnodes and will potentially free vnodes that still
646 * have VM backing store (VM backing store is typically the cause
647 * of a vnode blowout so we want to do this). Therefore, this operation
648 * is not considered cheap.
650 * A number of conditions may prevent a vnode from being reclaimed.
651 * the buffer cache may have references on the vnode, a directory
652 * vnode may still have references due to the namei cache representing
653 * underlying files, or the vnode may be in active use. It is not
654 * desireable to reuse such vnodes. These conditions may cause the
655 * number of vnodes to reach some minimum value regardless of what
656 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
659 vlrureclaim(struct mount *mp)
668 * Calculate the trigger point, don't allow user
669 * screwups to blow us up. This prevents us from
670 * recycling vnodes with lots of resident pages. We
671 * aren't trying to free memory, we are trying to
674 usevnodes = desiredvnodes;
677 trigger = cnt.v_page_count * 2 / usevnodes;
679 vn_start_write(NULL, &mp, V_WAIT);
681 count = mp->mnt_nvnodelistsize / 10 + 1;
683 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
684 while (vp != NULL && vp->v_type == VMARKER)
685 vp = TAILQ_NEXT(vp, v_nmntvnodes);
688 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
689 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
694 * If it's been deconstructed already, it's still
695 * referenced, or it exceeds the trigger, skip it.
697 if (vp->v_usecount ||
698 (!vlru_allow_cache_src &&
699 !LIST_EMPTY(&(vp)->v_cache_src)) ||
700 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
701 vp->v_object->resident_page_count > trigger)) {
707 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
709 goto next_iter_mntunlocked;
713 * v_usecount may have been bumped after VOP_LOCK() dropped
714 * the vnode interlock and before it was locked again.
716 * It is not necessary to recheck VI_DOOMED because it can
717 * only be set by another thread that holds both the vnode
718 * lock and vnode interlock. If another thread has the
719 * vnode lock before we get to VOP_LOCK() and obtains the
720 * vnode interlock after VOP_LOCK() drops the vnode
721 * interlock, the other thread will be unable to drop the
722 * vnode lock before our VOP_LOCK() call fails.
724 if (vp->v_usecount ||
725 (!vlru_allow_cache_src &&
726 !LIST_EMPTY(&(vp)->v_cache_src)) ||
727 (vp->v_object != NULL &&
728 vp->v_object->resident_page_count > trigger)) {
729 VOP_UNLOCK(vp, LK_INTERLOCK);
730 goto next_iter_mntunlocked;
732 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
733 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
738 next_iter_mntunlocked:
747 kern_yield(PRI_UNCHANGED);
752 vn_finished_write(mp);
757 * Attempt to keep the free list at wantfreevnodes length.
760 vnlru_free(int count)
765 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
766 for (; count > 0; count--) {
767 vp = TAILQ_FIRST(&vnode_free_list);
769 * The list can be modified while the free_list_mtx
770 * has been dropped and vp could be NULL here.
774 VNASSERT(vp->v_op != NULL, vp,
775 ("vnlru_free: vnode already reclaimed."));
776 KASSERT((vp->v_iflag & VI_FREE) != 0,
777 ("Removing vnode not on freelist"));
778 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
779 ("Mangling active vnode"));
780 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
782 * Don't recycle if we can't get the interlock.
784 if (!VI_TRYLOCK(vp)) {
785 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
788 VNASSERT(VCANRECYCLE(vp), vp,
789 ("vp inconsistent on freelist"));
791 vp->v_iflag &= ~VI_FREE;
793 mtx_unlock(&vnode_free_list_mtx);
795 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
797 VFS_UNLOCK_GIANT(vfslocked);
799 * If the recycled succeeded this vdrop will actually free
800 * the vnode. If not it will simply place it back on
804 mtx_lock(&vnode_free_list_mtx);
808 * Attempt to recycle vnodes in a context that is always safe to block.
809 * Calling vlrurecycle() from the bowels of filesystem code has some
810 * interesting deadlock problems.
812 static struct proc *vnlruproc;
813 static int vnlruproc_sig;
818 struct mount *mp, *nmp;
820 struct proc *p = vnlruproc;
822 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
826 kproc_suspend_check(p);
827 mtx_lock(&vnode_free_list_mtx);
828 if (freevnodes > wantfreevnodes)
829 vnlru_free(freevnodes - wantfreevnodes);
830 if (numvnodes <= desiredvnodes * 9 / 10) {
832 wakeup(&vnlruproc_sig);
833 msleep(vnlruproc, &vnode_free_list_mtx,
834 PVFS|PDROP, "vlruwt", hz);
837 mtx_unlock(&vnode_free_list_mtx);
839 mtx_lock(&mountlist_mtx);
840 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
841 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
842 nmp = TAILQ_NEXT(mp, mnt_list);
845 vfslocked = VFS_LOCK_GIANT(mp);
846 done += vlrureclaim(mp);
847 VFS_UNLOCK_GIANT(vfslocked);
848 mtx_lock(&mountlist_mtx);
849 nmp = TAILQ_NEXT(mp, mnt_list);
852 mtx_unlock(&mountlist_mtx);
855 /* These messages are temporary debugging aids */
856 if (vnlru_nowhere < 5)
857 printf("vnlru process getting nowhere..\n");
858 else if (vnlru_nowhere == 5)
859 printf("vnlru process messages stopped.\n");
862 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
864 kern_yield(PRI_UNCHANGED);
868 static struct kproc_desc vnlru_kp = {
873 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
877 * Routines having to do with the management of the vnode table.
881 * Try to recycle a freed vnode. We abort if anyone picks up a reference
882 * before we actually vgone(). This function must be called with the vnode
883 * held to prevent the vnode from being returned to the free list midway
887 vtryrecycle(struct vnode *vp)
891 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
892 VNASSERT(vp->v_holdcnt, vp,
893 ("vtryrecycle: Recycling vp %p without a reference.", vp));
895 * This vnode may found and locked via some other list, if so we
896 * can't recycle it yet.
898 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
900 "%s: impossible to recycle, vp %p lock is already held",
902 return (EWOULDBLOCK);
905 * Don't recycle if its filesystem is being suspended.
907 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
910 "%s: impossible to recycle, cannot start the write for %p",
915 * If we got this far, we need to acquire the interlock and see if
916 * anyone picked up this vnode from another list. If not, we will
917 * mark it with DOOMED via vgonel() so that anyone who does find it
921 if (vp->v_usecount) {
922 VOP_UNLOCK(vp, LK_INTERLOCK);
923 vn_finished_write(vnmp);
925 "%s: impossible to recycle, %p is already referenced",
929 if ((vp->v_iflag & VI_DOOMED) == 0)
931 VOP_UNLOCK(vp, LK_INTERLOCK);
932 vn_finished_write(vnmp);
937 * Return the next vnode from the free list.
940 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
943 struct vnode *vp = NULL;
946 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
947 mtx_lock(&vnode_free_list_mtx);
949 * Lend our context to reclaim vnodes if they've exceeded the max.
951 if (freevnodes > wantfreevnodes)
954 * Wait for available vnodes.
956 if (numvnodes > desiredvnodes) {
957 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) {
959 * File system is beeing suspended, we cannot risk a
960 * deadlock here, so allocate new vnode anyway.
962 if (freevnodes > wantfreevnodes)
963 vnlru_free(freevnodes - wantfreevnodes);
966 if (vnlruproc_sig == 0) {
967 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
970 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
972 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
973 if (numvnodes > desiredvnodes) {
974 mtx_unlock(&vnode_free_list_mtx);
981 mtx_unlock(&vnode_free_list_mtx);
982 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
986 vp->v_vnlock = &vp->v_lock;
987 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
989 * By default, don't allow shared locks unless filesystems
992 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
998 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
999 bo->bo_ops = &buf_ops_bio;
1000 bo->bo_private = vp;
1001 TAILQ_INIT(&bo->bo_clean.bv_hd);
1002 TAILQ_INIT(&bo->bo_dirty.bv_hd);
1004 * Initialize namecache.
1006 LIST_INIT(&vp->v_cache_src);
1007 TAILQ_INIT(&vp->v_cache_dst);
1009 * Finalize various vnode identity bits.
1014 v_incr_usecount(vp);
1018 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1019 mac_vnode_associate_singlelabel(mp, vp);
1020 else if (mp == NULL && vops != &dead_vnodeops)
1021 printf("NULL mp in getnewvnode()\n");
1024 bo->bo_bsize = mp->mnt_stat.f_iosize;
1025 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1026 vp->v_vflag |= VV_NOKNOTE;
1034 * Delete from old mount point vnode list, if on one.
1037 delmntque(struct vnode *vp)
1047 KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1048 ("Active vnode list size %d > Vnode list size %d",
1049 mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1050 active = vp->v_iflag & VI_ACTIVE;
1051 vp->v_iflag &= ~VI_ACTIVE;
1053 mtx_lock(&vnode_free_list_mtx);
1054 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1055 mp->mnt_activevnodelistsize--;
1056 mtx_unlock(&vnode_free_list_mtx);
1060 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1061 ("bad mount point vnode list size"));
1062 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1063 mp->mnt_nvnodelistsize--;
1069 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1073 vp->v_op = &dead_vnodeops;
1074 /* XXX non mp-safe fs may still call insmntque with vnode
1076 if (!VOP_ISLOCKED(vp))
1077 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1083 * Insert into list of vnodes for the new mount point, if available.
1086 insmntque1(struct vnode *vp, struct mount *mp,
1087 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1091 KASSERT(vp->v_mount == NULL,
1092 ("insmntque: vnode already on per mount vnode list"));
1093 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1094 #ifdef DEBUG_VFS_LOCKS
1095 if (!VFS_NEEDSGIANT(mp))
1096 ASSERT_VOP_ELOCKED(vp,
1097 "insmntque: mp-safe fs and non-locked vp");
1100 * We acquire the vnode interlock early to ensure that the
1101 * vnode cannot be recycled by another process releasing a
1102 * holdcnt on it before we get it on both the vnode list
1103 * and the active vnode list. The mount mutex protects only
1104 * manipulation of the vnode list and the vnode freelist
1105 * mutex protects only manipulation of the active vnode list.
1106 * Hence the need to hold the vnode interlock throughout.
1110 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1111 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1112 mp->mnt_nvnodelistsize == 0)) {
1113 locked = VOP_ISLOCKED(vp);
1114 if (!locked || (locked == LK_EXCLUSIVE &&
1115 (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
1125 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1126 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1127 ("neg mount point vnode list size"));
1128 mp->mnt_nvnodelistsize++;
1129 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1130 ("Activating already active vnode"));
1131 vp->v_iflag |= VI_ACTIVE;
1132 mtx_lock(&vnode_free_list_mtx);
1133 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1134 mp->mnt_activevnodelistsize++;
1135 mtx_unlock(&vnode_free_list_mtx);
1142 insmntque(struct vnode *vp, struct mount *mp)
1145 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1149 * Flush out and invalidate all buffers associated with a bufobj
1150 * Called with the underlying object locked.
1153 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1158 if (flags & V_SAVE) {
1159 error = bufobj_wwait(bo, slpflag, slptimeo);
1164 if (bo->bo_dirty.bv_cnt > 0) {
1166 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1169 * XXX We could save a lock/unlock if this was only
1170 * enabled under INVARIANTS
1173 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1174 panic("vinvalbuf: dirty bufs");
1178 * If you alter this loop please notice that interlock is dropped and
1179 * reacquired in flushbuflist. Special care is needed to ensure that
1180 * no race conditions occur from this.
1183 error = flushbuflist(&bo->bo_clean,
1184 flags, bo, slpflag, slptimeo);
1185 if (error == 0 && !(flags & V_CLEANONLY))
1186 error = flushbuflist(&bo->bo_dirty,
1187 flags, bo, slpflag, slptimeo);
1188 if (error != 0 && error != EAGAIN) {
1192 } while (error != 0);
1195 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1196 * have write I/O in-progress but if there is a VM object then the
1197 * VM object can also have read-I/O in-progress.
1200 bufobj_wwait(bo, 0, 0);
1202 if (bo->bo_object != NULL) {
1203 VM_OBJECT_LOCK(bo->bo_object);
1204 vm_object_pip_wait(bo->bo_object, "bovlbx");
1205 VM_OBJECT_UNLOCK(bo->bo_object);
1208 } while (bo->bo_numoutput > 0);
1212 * Destroy the copy in the VM cache, too.
1214 if (bo->bo_object != NULL &&
1215 (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1216 VM_OBJECT_LOCK(bo->bo_object);
1217 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1218 OBJPR_CLEANONLY : 0);
1219 VM_OBJECT_UNLOCK(bo->bo_object);
1224 if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1225 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1226 panic("vinvalbuf: flush failed");
1233 * Flush out and invalidate all buffers associated with a vnode.
1234 * Called with the underlying object locked.
1237 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1240 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1241 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1242 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1246 * Flush out buffers on the specified list.
1250 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1253 struct buf *bp, *nbp;
1258 ASSERT_BO_LOCKED(bo);
1261 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1262 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1263 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1269 lblkno = nbp->b_lblkno;
1270 xflags = nbp->b_xflags &
1271 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
1274 error = BUF_TIMELOCK(bp,
1275 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
1276 "flushbuf", slpflag, slptimeo);
1279 return (error != ENOLCK ? error : EAGAIN);
1281 KASSERT(bp->b_bufobj == bo,
1282 ("bp %p wrong b_bufobj %p should be %p",
1283 bp, bp->b_bufobj, bo));
1284 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1290 * XXX Since there are no node locks for NFS, I
1291 * believe there is a slight chance that a delayed
1292 * write will occur while sleeping just above, so
1295 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1300 bp->b_flags |= B_ASYNC;
1303 return (EAGAIN); /* XXX: why not loop ? */
1308 bp->b_flags |= (B_INVAL | B_RELBUF);
1309 bp->b_flags &= ~B_ASYNC;
1313 (nbp->b_bufobj != bo ||
1314 nbp->b_lblkno != lblkno ||
1316 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1317 break; /* nbp invalid */
1323 * Truncate a file's buffer and pages to a specified length. This
1324 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1328 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
1329 off_t length, int blksize)
1331 struct buf *bp, *nbp;
1336 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1337 vp, cred, blksize, (uintmax_t)length);
1340 * Round up to the *next* lbn.
1342 trunclbn = (length + blksize - 1) / blksize;
1344 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1351 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1352 if (bp->b_lblkno < trunclbn)
1355 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1356 BO_MTX(bo)) == ENOLCK)
1362 bp->b_flags |= (B_INVAL | B_RELBUF);
1363 bp->b_flags &= ~B_ASYNC;
1369 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1370 (nbp->b_vp != vp) ||
1371 (nbp->b_flags & B_DELWRI))) {
1377 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1378 if (bp->b_lblkno < trunclbn)
1381 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1382 BO_MTX(bo)) == ENOLCK)
1387 bp->b_flags |= (B_INVAL | B_RELBUF);
1388 bp->b_flags &= ~B_ASYNC;
1394 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1395 (nbp->b_vp != vp) ||
1396 (nbp->b_flags & B_DELWRI) == 0)) {
1405 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1406 if (bp->b_lblkno > 0)
1409 * Since we hold the vnode lock this should only
1410 * fail if we're racing with the buf daemon.
1413 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1414 BO_MTX(bo)) == ENOLCK) {
1417 VNASSERT((bp->b_flags & B_DELWRI), vp,
1418 ("buf(%p) on dirty queue without DELWRI", bp));
1429 bufobj_wwait(bo, 0, 0);
1431 vnode_pager_setsize(vp, length);
1437 * buf_splay() - splay tree core for the clean/dirty list of buffers in
1440 * NOTE: We have to deal with the special case of a background bitmap
1441 * buffer, a situation where two buffers will have the same logical
1442 * block offset. We want (1) only the foreground buffer to be accessed
1443 * in a lookup and (2) must differentiate between the foreground and
1444 * background buffer in the splay tree algorithm because the splay
1445 * tree cannot normally handle multiple entities with the same 'index'.
1446 * We accomplish this by adding differentiating flags to the splay tree's
1451 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
1454 struct buf *lefttreemax, *righttreemin, *y;
1458 lefttreemax = righttreemin = &dummy;
1460 if (lblkno < root->b_lblkno ||
1461 (lblkno == root->b_lblkno &&
1462 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1463 if ((y = root->b_left) == NULL)
1465 if (lblkno < y->b_lblkno) {
1467 root->b_left = y->b_right;
1470 if ((y = root->b_left) == NULL)
1473 /* Link into the new root's right tree. */
1474 righttreemin->b_left = root;
1475 righttreemin = root;
1476 } else if (lblkno > root->b_lblkno ||
1477 (lblkno == root->b_lblkno &&
1478 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
1479 if ((y = root->b_right) == NULL)
1481 if (lblkno > y->b_lblkno) {
1483 root->b_right = y->b_left;
1486 if ((y = root->b_right) == NULL)
1489 /* Link into the new root's left tree. */
1490 lefttreemax->b_right = root;
1497 /* Assemble the new root. */
1498 lefttreemax->b_right = root->b_left;
1499 righttreemin->b_left = root->b_right;
1500 root->b_left = dummy.b_right;
1501 root->b_right = dummy.b_left;
1506 buf_vlist_remove(struct buf *bp)
1511 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1512 ASSERT_BO_LOCKED(bp->b_bufobj);
1513 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1514 (BX_VNDIRTY|BX_VNCLEAN),
1515 ("buf_vlist_remove: Buf %p is on two lists", bp));
1516 if (bp->b_xflags & BX_VNDIRTY)
1517 bv = &bp->b_bufobj->bo_dirty;
1519 bv = &bp->b_bufobj->bo_clean;
1520 if (bp != bv->bv_root) {
1521 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1522 KASSERT(root == bp, ("splay lookup failed in remove"));
1524 if (bp->b_left == NULL) {
1527 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
1528 root->b_right = bp->b_right;
1531 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1533 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1537 * Add the buffer to the sorted clean or dirty block list using a
1538 * splay tree algorithm.
1540 * NOTE: xflags is passed as a constant, optimizing this inline function!
1543 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1548 ASSERT_BO_LOCKED(bo);
1549 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1550 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1551 bp->b_xflags |= xflags;
1552 if (xflags & BX_VNDIRTY)
1557 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
1561 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1562 } else if (bp->b_lblkno < root->b_lblkno ||
1563 (bp->b_lblkno == root->b_lblkno &&
1564 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
1565 bp->b_left = root->b_left;
1567 root->b_left = NULL;
1568 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
1570 bp->b_right = root->b_right;
1572 root->b_right = NULL;
1573 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
1580 * Lookup a buffer using the splay tree. Note that we specifically avoid
1581 * shadow buffers used in background bitmap writes.
1583 * This code isn't quite efficient as it could be because we are maintaining
1584 * two sorted lists and do not know which list the block resides in.
1586 * During a "make buildworld" the desired buffer is found at one of
1587 * the roots more than 60% of the time. Thus, checking both roots
1588 * before performing either splay eliminates unnecessary splays on the
1589 * first tree splayed.
1592 gbincore(struct bufobj *bo, daddr_t lblkno)
1596 ASSERT_BO_LOCKED(bo);
1597 if ((bp = bo->bo_clean.bv_root) != NULL &&
1598 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1600 if ((bp = bo->bo_dirty.bv_root) != NULL &&
1601 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1603 if ((bp = bo->bo_clean.bv_root) != NULL) {
1604 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
1605 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1608 if ((bp = bo->bo_dirty.bv_root) != NULL) {
1609 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
1610 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
1617 * Associate a buffer with a vnode.
1620 bgetvp(struct vnode *vp, struct buf *bp)
1625 ASSERT_BO_LOCKED(bo);
1626 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1628 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1629 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1630 ("bgetvp: bp already attached! %p", bp));
1633 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
1634 bp->b_flags |= B_NEEDSGIANT;
1638 * Insert onto list for new vnode.
1640 buf_vlist_add(bp, bo, BX_VNCLEAN);
1644 * Disassociate a buffer from a vnode.
1647 brelvp(struct buf *bp)
1652 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1653 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1656 * Delete from old vnode list, if on one.
1658 vp = bp->b_vp; /* XXX */
1661 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1662 buf_vlist_remove(bp);
1664 panic("brelvp: Buffer %p not on queue.", bp);
1665 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1666 bo->bo_flag &= ~BO_ONWORKLST;
1667 mtx_lock(&sync_mtx);
1668 LIST_REMOVE(bo, bo_synclist);
1669 syncer_worklist_len--;
1670 mtx_unlock(&sync_mtx);
1672 bp->b_flags &= ~B_NEEDSGIANT;
1674 bp->b_bufobj = NULL;
1680 * Add an item to the syncer work queue.
1683 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1687 ASSERT_BO_LOCKED(bo);
1689 mtx_lock(&sync_mtx);
1690 if (bo->bo_flag & BO_ONWORKLST)
1691 LIST_REMOVE(bo, bo_synclist);
1693 bo->bo_flag |= BO_ONWORKLST;
1694 syncer_worklist_len++;
1697 if (delay > syncer_maxdelay - 2)
1698 delay = syncer_maxdelay - 2;
1699 slot = (syncer_delayno + delay) & syncer_mask;
1701 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
1703 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
1705 mtx_unlock(&sync_mtx);
1709 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1713 mtx_lock(&sync_mtx);
1714 len = syncer_worklist_len - sync_vnode_count;
1715 mtx_unlock(&sync_mtx);
1716 error = SYSCTL_OUT(req, &len, sizeof(len));
1720 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1721 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1723 static struct proc *updateproc;
1724 static void sched_sync(void);
1725 static struct kproc_desc up_kp = {
1730 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1733 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1738 *bo = LIST_FIRST(slp);
1741 vp = (*bo)->__bo_vnode; /* XXX */
1742 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1745 * We use vhold in case the vnode does not
1746 * successfully sync. vhold prevents the vnode from
1747 * going away when we unlock the sync_mtx so that
1748 * we can acquire the vnode interlock.
1751 mtx_unlock(&sync_mtx);
1753 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1755 mtx_lock(&sync_mtx);
1756 return (*bo == LIST_FIRST(slp));
1758 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1759 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1761 vn_finished_write(mp);
1763 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1765 * Put us back on the worklist. The worklist
1766 * routine will remove us from our current
1767 * position and then add us back in at a later
1770 vn_syncer_add_to_worklist(*bo, syncdelay);
1774 mtx_lock(&sync_mtx);
1779 * System filesystem synchronizer daemon.
1784 struct synclist *gnext, *next;
1785 struct synclist *gslp, *slp;
1788 struct thread *td = curthread;
1790 int net_worklist_len;
1791 int syncer_final_iter;
1796 syncer_final_iter = 0;
1798 syncer_state = SYNCER_RUNNING;
1799 starttime = time_uptime;
1800 td->td_pflags |= TDP_NORUNNINGBUF;
1802 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1805 mtx_lock(&sync_mtx);
1807 if (syncer_state == SYNCER_FINAL_DELAY &&
1808 syncer_final_iter == 0) {
1809 mtx_unlock(&sync_mtx);
1810 kproc_suspend_check(td->td_proc);
1811 mtx_lock(&sync_mtx);
1813 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1814 if (syncer_state != SYNCER_RUNNING &&
1815 starttime != time_uptime) {
1817 printf("\nSyncing disks, vnodes remaining...");
1820 printf("%d ", net_worklist_len);
1822 starttime = time_uptime;
1825 * Push files whose dirty time has expired. Be careful
1826 * of interrupt race on slp queue.
1828 * Skip over empty worklist slots when shutting down.
1831 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1832 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1833 syncer_delayno += 1;
1834 if (syncer_delayno == syncer_maxdelay)
1836 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
1837 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
1839 * If the worklist has wrapped since the
1840 * it was emptied of all but syncer vnodes,
1841 * switch to the FINAL_DELAY state and run
1842 * for one more second.
1844 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1845 net_worklist_len == 0 &&
1846 last_work_seen == syncer_delayno) {
1847 syncer_state = SYNCER_FINAL_DELAY;
1848 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1850 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1851 LIST_EMPTY(gslp) && syncer_worklist_len > 0);
1854 * Keep track of the last time there was anything
1855 * on the worklist other than syncer vnodes.
1856 * Return to the SHUTTING_DOWN state if any
1859 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1860 last_work_seen = syncer_delayno;
1861 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1862 syncer_state = SYNCER_SHUTTING_DOWN;
1863 while (!LIST_EMPTY(slp)) {
1864 error = sync_vnode(slp, &bo, td);
1866 LIST_REMOVE(bo, bo_synclist);
1867 LIST_INSERT_HEAD(next, bo, bo_synclist);
1871 if (first_printf == 0)
1872 wdog_kern_pat(WD_LASTVAL);
1875 if (!LIST_EMPTY(gslp)) {
1876 mtx_unlock(&sync_mtx);
1878 mtx_lock(&sync_mtx);
1879 while (!LIST_EMPTY(gslp)) {
1880 error = sync_vnode(gslp, &bo, td);
1882 LIST_REMOVE(bo, bo_synclist);
1883 LIST_INSERT_HEAD(gnext, bo,
1890 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1891 syncer_final_iter--;
1893 * The variable rushjob allows the kernel to speed up the
1894 * processing of the filesystem syncer process. A rushjob
1895 * value of N tells the filesystem syncer to process the next
1896 * N seconds worth of work on its queue ASAP. Currently rushjob
1897 * is used by the soft update code to speed up the filesystem
1898 * syncer process when the incore state is getting so far
1899 * ahead of the disk that the kernel memory pool is being
1900 * threatened with exhaustion.
1907 * Just sleep for a short period of time between
1908 * iterations when shutting down to allow some I/O
1911 * If it has taken us less than a second to process the
1912 * current work, then wait. Otherwise start right over
1913 * again. We can still lose time if any single round
1914 * takes more than two seconds, but it does not really
1915 * matter as we are just trying to generally pace the
1916 * filesystem activity.
1918 if (syncer_state != SYNCER_RUNNING ||
1919 time_uptime == starttime) {
1921 sched_prio(td, PPAUSE);
1924 if (syncer_state != SYNCER_RUNNING)
1925 cv_timedwait(&sync_wakeup, &sync_mtx,
1926 hz / SYNCER_SHUTDOWN_SPEEDUP);
1927 else if (time_uptime == starttime)
1928 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1933 * Request the syncer daemon to speed up its work.
1934 * We never push it to speed up more than half of its
1935 * normal turn time, otherwise it could take over the cpu.
1938 speedup_syncer(void)
1942 mtx_lock(&sync_mtx);
1943 if (rushjob < syncdelay / 2) {
1945 stat_rush_requests += 1;
1948 mtx_unlock(&sync_mtx);
1949 cv_broadcast(&sync_wakeup);
1954 * Tell the syncer to speed up its work and run though its work
1955 * list several times, then tell it to shut down.
1958 syncer_shutdown(void *arg, int howto)
1961 if (howto & RB_NOSYNC)
1963 mtx_lock(&sync_mtx);
1964 syncer_state = SYNCER_SHUTTING_DOWN;
1966 mtx_unlock(&sync_mtx);
1967 cv_broadcast(&sync_wakeup);
1968 kproc_shutdown(arg, howto);
1972 * Reassign a buffer from one vnode to another.
1973 * Used to assign file specific control information
1974 * (indirect blocks) to the vnode to which they belong.
1977 reassignbuf(struct buf *bp)
1990 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1991 bp, bp->b_vp, bp->b_flags);
1993 * B_PAGING flagged buffers cannot be reassigned because their vp
1994 * is not fully linked in.
1996 if (bp->b_flags & B_PAGING)
1997 panic("cannot reassign paging buffer");
2000 * Delete from old vnode list, if on one.
2003 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2004 buf_vlist_remove(bp);
2006 panic("reassignbuf: Buffer %p not on queue.", bp);
2008 * If dirty, put on list of dirty buffers; otherwise insert onto list
2011 if (bp->b_flags & B_DELWRI) {
2012 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2013 switch (vp->v_type) {
2023 vn_syncer_add_to_worklist(bo, delay);
2025 buf_vlist_add(bp, bo, BX_VNDIRTY);
2027 buf_vlist_add(bp, bo, BX_VNCLEAN);
2029 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2030 mtx_lock(&sync_mtx);
2031 LIST_REMOVE(bo, bo_synclist);
2032 syncer_worklist_len--;
2033 mtx_unlock(&sync_mtx);
2034 bo->bo_flag &= ~BO_ONWORKLST;
2039 bp = TAILQ_FIRST(&bv->bv_hd);
2040 KASSERT(bp == NULL || bp->b_bufobj == bo,
2041 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2042 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2043 KASSERT(bp == NULL || bp->b_bufobj == bo,
2044 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2046 bp = TAILQ_FIRST(&bv->bv_hd);
2047 KASSERT(bp == NULL || bp->b_bufobj == bo,
2048 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2049 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2050 KASSERT(bp == NULL || bp->b_bufobj == bo,
2051 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2057 * Increment the use and hold counts on the vnode, taking care to reference
2058 * the driver's usecount if this is a chardev. The vholdl() will remove
2059 * the vnode from the free list if it is presently free. Requires the
2060 * vnode interlock and returns with it held.
2063 v_incr_usecount(struct vnode *vp)
2066 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2068 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2070 vp->v_rdev->si_usecount++;
2077 * Turn a holdcnt into a use+holdcnt such that only one call to
2078 * v_decr_usecount is needed.
2081 v_upgrade_usecount(struct vnode *vp)
2084 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2086 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2088 vp->v_rdev->si_usecount++;
2094 * Decrement the vnode use and hold count along with the driver's usecount
2095 * if this is a chardev. The vdropl() below releases the vnode interlock
2096 * as it may free the vnode.
2099 v_decr_usecount(struct vnode *vp)
2102 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2103 VNASSERT(vp->v_usecount > 0, vp,
2104 ("v_decr_usecount: negative usecount"));
2105 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2107 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2109 vp->v_rdev->si_usecount--;
2116 * Decrement only the use count and driver use count. This is intended to
2117 * be paired with a follow on vdropl() to release the remaining hold count.
2118 * In this way we may vgone() a vnode with a 0 usecount without risk of
2119 * having it end up on a free list because the hold count is kept above 0.
2122 v_decr_useonly(struct vnode *vp)
2125 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2126 VNASSERT(vp->v_usecount > 0, vp,
2127 ("v_decr_useonly: negative usecount"));
2128 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2130 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2132 vp->v_rdev->si_usecount--;
2138 * Grab a particular vnode from the free list, increment its
2139 * reference count and lock it. VI_DOOMED is set if the vnode
2140 * is being destroyed. Only callers who specify LK_RETRY will
2141 * see doomed vnodes. If inactive processing was delayed in
2142 * vput try to do it here.
2145 vget(struct vnode *vp, int flags, struct thread *td)
2150 VFS_ASSERT_GIANT(vp->v_mount);
2151 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2152 ("vget: invalid lock operation"));
2153 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2155 if ((flags & LK_INTERLOCK) == 0)
2158 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2160 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2164 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2165 panic("vget: vn_lock failed to return ENOENT\n");
2167 /* Upgrade our holdcnt to a usecount. */
2168 v_upgrade_usecount(vp);
2170 * We don't guarantee that any particular close will
2171 * trigger inactive processing so just make a best effort
2172 * here at preventing a reference to a removed file. If
2173 * we don't succeed no harm is done.
2175 if (vp->v_iflag & VI_OWEINACT) {
2176 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2177 (flags & LK_NOWAIT) == 0)
2179 vp->v_iflag &= ~VI_OWEINACT;
2186 * Increase the reference count of a vnode.
2189 vref(struct vnode *vp)
2192 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2194 v_incr_usecount(vp);
2199 * Return reference count of a vnode.
2201 * The results of this call are only guaranteed when some mechanism other
2202 * than the VI lock is used to stop other processes from gaining references
2203 * to the vnode. This may be the case if the caller holds the only reference.
2204 * This is also useful when stale data is acceptable as race conditions may
2205 * be accounted for by some other means.
2208 vrefcnt(struct vnode *vp)
2213 usecnt = vp->v_usecount;
2219 #define VPUTX_VRELE 1
2220 #define VPUTX_VPUT 2
2221 #define VPUTX_VUNREF 3
2224 vputx(struct vnode *vp, int func)
2228 KASSERT(vp != NULL, ("vputx: null vp"));
2229 if (func == VPUTX_VUNREF)
2230 ASSERT_VOP_LOCKED(vp, "vunref");
2231 else if (func == VPUTX_VPUT)
2232 ASSERT_VOP_LOCKED(vp, "vput");
2234 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2235 VFS_ASSERT_GIANT(vp->v_mount);
2236 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2239 /* Skip this v_writecount check if we're going to panic below. */
2240 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2241 ("vputx: missed vn_close"));
2244 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2245 vp->v_usecount == 1)) {
2246 if (func == VPUTX_VPUT)
2248 v_decr_usecount(vp);
2252 if (vp->v_usecount != 1) {
2253 vprint("vputx: negative ref count", vp);
2254 panic("vputx: negative ref cnt");
2256 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2258 * We want to hold the vnode until the inactive finishes to
2259 * prevent vgone() races. We drop the use count here and the
2260 * hold count below when we're done.
2264 * We must call VOP_INACTIVE with the node locked. Mark
2265 * as VI_DOINGINACT to avoid recursion.
2267 vp->v_iflag |= VI_OWEINACT;
2270 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2274 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2275 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2281 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
2285 if (vp->v_usecount > 0)
2286 vp->v_iflag &= ~VI_OWEINACT;
2288 if (vp->v_iflag & VI_OWEINACT)
2289 vinactive(vp, curthread);
2290 if (func != VPUTX_VUNREF)
2297 * Vnode put/release.
2298 * If count drops to zero, call inactive routine and return to freelist.
2301 vrele(struct vnode *vp)
2304 vputx(vp, VPUTX_VRELE);
2308 * Release an already locked vnode. This give the same effects as
2309 * unlock+vrele(), but takes less time and avoids releasing and
2310 * re-aquiring the lock (as vrele() acquires the lock internally.)
2313 vput(struct vnode *vp)
2316 vputx(vp, VPUTX_VPUT);
2320 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2323 vunref(struct vnode *vp)
2326 vputx(vp, VPUTX_VUNREF);
2330 * Somebody doesn't want the vnode recycled.
2333 vhold(struct vnode *vp)
2342 * Increase the hold count and activate if this is the first reference.
2345 vholdl(struct vnode *vp)
2349 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2351 if (!VSHOULDBUSY(vp))
2353 ASSERT_VI_LOCKED(vp, "vholdl");
2354 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2355 VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2357 * Remove a vnode from the free list, mark it as in use,
2358 * and put it on the active list.
2360 mtx_lock(&vnode_free_list_mtx);
2361 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2363 vp->v_iflag &= ~(VI_FREE|VI_AGE);
2364 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2365 ("Activating already active vnode"));
2366 vp->v_iflag |= VI_ACTIVE;
2368 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2369 mp->mnt_activevnodelistsize++;
2370 mtx_unlock(&vnode_free_list_mtx);
2374 * Note that there is one less who cares about this vnode.
2375 * vdrop() is the opposite of vhold().
2378 vdrop(struct vnode *vp)
2386 * Drop the hold count of the vnode. If this is the last reference to
2387 * the vnode we place it on the free list unless it has been vgone'd
2388 * (marked VI_DOOMED) in which case we will free it.
2391 vdropl(struct vnode *vp)
2397 ASSERT_VI_LOCKED(vp, "vdropl");
2398 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2399 if (vp->v_holdcnt <= 0)
2400 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2402 if (vp->v_holdcnt > 0) {
2406 if ((vp->v_iflag & VI_DOOMED) == 0) {
2408 * Mark a vnode as free: remove it from its active list
2409 * and put it up for recycling on the freelist.
2411 VNASSERT(vp->v_op != NULL, vp,
2412 ("vdropl: vnode already reclaimed."));
2413 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2414 ("vnode already free"));
2415 VNASSERT(VSHOULDFREE(vp), vp,
2416 ("vdropl: freeing when we shouldn't"));
2417 active = vp->v_iflag & VI_ACTIVE;
2418 vp->v_iflag &= ~VI_ACTIVE;
2420 mtx_lock(&vnode_free_list_mtx);
2422 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2424 mp->mnt_activevnodelistsize--;
2426 if (vp->v_iflag & VI_AGE) {
2427 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
2429 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
2432 vp->v_iflag &= ~VI_AGE;
2433 vp->v_iflag |= VI_FREE;
2434 mtx_unlock(&vnode_free_list_mtx);
2439 * The vnode has been marked for destruction, so free it.
2441 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2442 mtx_lock(&vnode_free_list_mtx);
2444 mtx_unlock(&vnode_free_list_mtx);
2446 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2447 ("cleaned vnode still on the free list."));
2448 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2449 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2450 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2451 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2452 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2453 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2454 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
2455 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2456 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
2457 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2458 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2459 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2462 mac_vnode_destroy(vp);
2464 if (vp->v_pollinfo != NULL)
2465 destroy_vpollinfo(vp->v_pollinfo);
2467 /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2470 lockdestroy(vp->v_vnlock);
2471 mtx_destroy(&vp->v_interlock);
2472 mtx_destroy(BO_MTX(bo));
2473 uma_zfree(vnode_zone, vp);
2477 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2478 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2479 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2480 * failed lock upgrade.
2483 vinactive(struct vnode *vp, struct thread *td)
2485 struct vm_object *obj;
2487 ASSERT_VOP_ELOCKED(vp, "vinactive");
2488 ASSERT_VI_LOCKED(vp, "vinactive");
2489 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2490 ("vinactive: recursed on VI_DOINGINACT"));
2491 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2492 vp->v_iflag |= VI_DOINGINACT;
2493 vp->v_iflag &= ~VI_OWEINACT;
2496 * Before moving off the active list, we must be sure that any
2497 * modified pages are on the vnode's dirty list since these will
2498 * no longer be checked once the vnode is on the inactive list.
2501 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2502 VM_OBJECT_LOCK(obj);
2503 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2504 VM_OBJECT_UNLOCK(obj);
2506 VOP_INACTIVE(vp, td);
2508 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2509 ("vinactive: lost VI_DOINGINACT"));
2510 vp->v_iflag &= ~VI_DOINGINACT;
2514 * Remove any vnodes in the vnode table belonging to mount point mp.
2516 * If FORCECLOSE is not specified, there should not be any active ones,
2517 * return error if any are found (nb: this is a user error, not a
2518 * system error). If FORCECLOSE is specified, detach any active vnodes
2521 * If WRITECLOSE is set, only flush out regular file vnodes open for
2524 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2526 * `rootrefs' specifies the base reference count for the root vnode
2527 * of this filesystem. The root vnode is considered busy if its
2528 * v_usecount exceeds this value. On a successful return, vflush(, td)
2529 * will call vrele() on the root vnode exactly rootrefs times.
2530 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2534 static int busyprt = 0; /* print out busy vnodes */
2535 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2539 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2541 struct vnode *vp, *mvp, *rootvp = NULL;
2543 int busy = 0, error;
2545 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2548 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2549 ("vflush: bad args"));
2551 * Get the filesystem root vnode. We can vput() it
2552 * immediately, since with rootrefs > 0, it won't go away.
2554 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2555 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2562 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2564 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2567 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2571 * Skip over a vnodes marked VV_SYSTEM.
2573 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2579 * If WRITECLOSE is set, flush out unlinked but still open
2580 * files (even if open only for reading) and regular file
2581 * vnodes open for writing.
2583 if (flags & WRITECLOSE) {
2584 if (vp->v_object != NULL) {
2585 VM_OBJECT_LOCK(vp->v_object);
2586 vm_object_page_clean(vp->v_object, 0, 0, 0);
2587 VM_OBJECT_UNLOCK(vp->v_object);
2589 error = VOP_FSYNC(vp, MNT_WAIT, td);
2593 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2596 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2599 if ((vp->v_type == VNON ||
2600 (error == 0 && vattr.va_nlink > 0)) &&
2601 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2609 * With v_usecount == 0, all we need to do is clear out the
2610 * vnode data structures and we are done.
2612 * If FORCECLOSE is set, forcibly close the vnode.
2614 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2615 VNASSERT(vp->v_usecount == 0 ||
2616 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2617 ("device VNODE %p is FORCECLOSED", vp));
2623 vprint("vflush: busy vnode", vp);
2629 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2631 * If just the root vnode is busy, and if its refcount
2632 * is equal to `rootrefs', then go ahead and kill it.
2635 KASSERT(busy > 0, ("vflush: not busy"));
2636 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2637 ("vflush: usecount %d < rootrefs %d",
2638 rootvp->v_usecount, rootrefs));
2639 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2640 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2642 VOP_UNLOCK(rootvp, 0);
2648 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2652 for (; rootrefs > 0; rootrefs--)
2658 * Recycle an unused vnode to the front of the free list.
2661 vrecycle(struct vnode *vp, struct thread *td)
2665 ASSERT_VOP_ELOCKED(vp, "vrecycle");
2666 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2669 if (vp->v_usecount == 0) {
2678 * Eliminate all activity associated with a vnode
2679 * in preparation for reuse.
2682 vgone(struct vnode *vp)
2690 * vgone, with the vp interlock held.
2693 vgonel(struct vnode *vp)
2700 ASSERT_VOP_ELOCKED(vp, "vgonel");
2701 ASSERT_VI_LOCKED(vp, "vgonel");
2702 VNASSERT(vp->v_holdcnt, vp,
2703 ("vgonel: vp %p has no reference.", vp));
2704 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2708 * Don't vgonel if we're already doomed.
2710 if (vp->v_iflag & VI_DOOMED)
2712 vp->v_iflag |= VI_DOOMED;
2714 * Check to see if the vnode is in use. If so, we have to call
2715 * VOP_CLOSE() and VOP_INACTIVE().
2717 active = vp->v_usecount;
2718 oweinact = (vp->v_iflag & VI_OWEINACT);
2721 * Clean out any buffers associated with the vnode.
2722 * If the flush fails, just toss the buffers.
2725 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2726 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2727 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2728 vinvalbuf(vp, 0, 0, 0);
2731 * If purging an active vnode, it must be closed and
2732 * deactivated before being reclaimed.
2735 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2736 if (oweinact || active) {
2738 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2742 if (vp->v_type == VSOCK)
2743 vfs_unp_reclaim(vp);
2745 * Reclaim the vnode.
2747 if (VOP_RECLAIM(vp, td))
2748 panic("vgone: cannot reclaim");
2750 vn_finished_secondary_write(mp);
2751 VNASSERT(vp->v_object == NULL, vp,
2752 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2754 * Clear the advisory locks and wake up waiting threads.
2756 (void)VOP_ADVLOCKPURGE(vp);
2758 * Delete from old mount point vnode list.
2763 * Done with purge, reset to the standard lock and invalidate
2767 vp->v_vnlock = &vp->v_lock;
2768 vp->v_op = &dead_vnodeops;
2774 * Calculate the total number of references to a special device.
2777 vcount(struct vnode *vp)
2782 count = vp->v_rdev->si_usecount;
2788 * Same as above, but using the struct cdev *as argument
2791 count_dev(struct cdev *dev)
2796 count = dev->si_usecount;
2802 * Print out a description of a vnode.
2804 static char *typename[] =
2805 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2809 vn_printf(struct vnode *vp, const char *fmt, ...)
2812 char buf[256], buf2[16];
2818 printf("%p: ", (void *)vp);
2819 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2820 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2821 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2824 if (vp->v_vflag & VV_ROOT)
2825 strlcat(buf, "|VV_ROOT", sizeof(buf));
2826 if (vp->v_vflag & VV_ISTTY)
2827 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2828 if (vp->v_vflag & VV_NOSYNC)
2829 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2830 if (vp->v_vflag & VV_CACHEDLABEL)
2831 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2832 if (vp->v_vflag & VV_TEXT)
2833 strlcat(buf, "|VV_TEXT", sizeof(buf));
2834 if (vp->v_vflag & VV_COPYONWRITE)
2835 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2836 if (vp->v_vflag & VV_SYSTEM)
2837 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2838 if (vp->v_vflag & VV_PROCDEP)
2839 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2840 if (vp->v_vflag & VV_NOKNOTE)
2841 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2842 if (vp->v_vflag & VV_DELETED)
2843 strlcat(buf, "|VV_DELETED", sizeof(buf));
2844 if (vp->v_vflag & VV_MD)
2845 strlcat(buf, "|VV_MD", sizeof(buf));
2846 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC |
2847 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2848 VV_NOKNOTE | VV_DELETED | VV_MD);
2850 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2851 strlcat(buf, buf2, sizeof(buf));
2853 if (vp->v_iflag & VI_MOUNT)
2854 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2855 if (vp->v_iflag & VI_AGE)
2856 strlcat(buf, "|VI_AGE", sizeof(buf));
2857 if (vp->v_iflag & VI_DOOMED)
2858 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2859 if (vp->v_iflag & VI_FREE)
2860 strlcat(buf, "|VI_FREE", sizeof(buf));
2861 if (vp->v_iflag & VI_DOINGINACT)
2862 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2863 if (vp->v_iflag & VI_OWEINACT)
2864 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2865 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2866 VI_DOINGINACT | VI_OWEINACT);
2868 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2869 strlcat(buf, buf2, sizeof(buf));
2871 printf(" flags (%s)\n", buf + 1);
2872 if (mtx_owned(VI_MTX(vp)))
2873 printf(" VI_LOCKed");
2874 if (vp->v_object != NULL)
2875 printf(" v_object %p ref %d pages %d\n",
2876 vp->v_object, vp->v_object->ref_count,
2877 vp->v_object->resident_page_count);
2879 lockmgr_printinfo(vp->v_vnlock);
2880 if (vp->v_data != NULL)
2886 * List all of the locked vnodes in the system.
2887 * Called when debugging the kernel.
2889 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2891 struct mount *mp, *nmp;
2895 * Note: because this is DDB, we can't obey the locking semantics
2896 * for these structures, which means we could catch an inconsistent
2897 * state and dereference a nasty pointer. Not much to be done
2900 db_printf("Locked vnodes\n");
2901 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2902 nmp = TAILQ_NEXT(mp, mnt_list);
2903 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2904 if (vp->v_type != VMARKER &&
2908 nmp = TAILQ_NEXT(mp, mnt_list);
2913 * Show details about the given vnode.
2915 DB_SHOW_COMMAND(vnode, db_show_vnode)
2921 vp = (struct vnode *)addr;
2922 vn_printf(vp, "vnode ");
2926 * Show details about the given mount point.
2928 DB_SHOW_COMMAND(mount, db_show_mount)
2939 /* No address given, print short info about all mount points. */
2940 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2941 db_printf("%p %s on %s (%s)\n", mp,
2942 mp->mnt_stat.f_mntfromname,
2943 mp->mnt_stat.f_mntonname,
2944 mp->mnt_stat.f_fstypename);
2948 db_printf("\nMore info: show mount <addr>\n");
2952 mp = (struct mount *)addr;
2953 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2954 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2957 mflags = mp->mnt_flag;
2958 #define MNT_FLAG(flag) do { \
2959 if (mflags & (flag)) { \
2960 if (buf[0] != '\0') \
2961 strlcat(buf, ", ", sizeof(buf)); \
2962 strlcat(buf, (#flag) + 4, sizeof(buf)); \
2963 mflags &= ~(flag); \
2966 MNT_FLAG(MNT_RDONLY);
2967 MNT_FLAG(MNT_SYNCHRONOUS);
2968 MNT_FLAG(MNT_NOEXEC);
2969 MNT_FLAG(MNT_NOSUID);
2970 MNT_FLAG(MNT_UNION);
2971 MNT_FLAG(MNT_ASYNC);
2972 MNT_FLAG(MNT_SUIDDIR);
2973 MNT_FLAG(MNT_SOFTDEP);
2975 MNT_FLAG(MNT_NOSYMFOLLOW);
2976 MNT_FLAG(MNT_GJOURNAL);
2977 MNT_FLAG(MNT_MULTILABEL);
2979 MNT_FLAG(MNT_NOATIME);
2980 MNT_FLAG(MNT_NOCLUSTERR);
2981 MNT_FLAG(MNT_NOCLUSTERW);
2982 MNT_FLAG(MNT_NFS4ACLS);
2983 MNT_FLAG(MNT_EXRDONLY);
2984 MNT_FLAG(MNT_EXPORTED);
2985 MNT_FLAG(MNT_DEFEXPORTED);
2986 MNT_FLAG(MNT_EXPORTANON);
2987 MNT_FLAG(MNT_EXKERB);
2988 MNT_FLAG(MNT_EXPUBLIC);
2989 MNT_FLAG(MNT_LOCAL);
2990 MNT_FLAG(MNT_QUOTA);
2991 MNT_FLAG(MNT_ROOTFS);
2993 MNT_FLAG(MNT_IGNORE);
2994 MNT_FLAG(MNT_UPDATE);
2995 MNT_FLAG(MNT_DELEXPORT);
2996 MNT_FLAG(MNT_RELOAD);
2997 MNT_FLAG(MNT_FORCE);
2998 MNT_FLAG(MNT_SNAPSHOT);
2999 MNT_FLAG(MNT_BYFSID);
3003 strlcat(buf, ", ", sizeof(buf));
3004 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3005 "0x%016jx", mflags);
3007 db_printf(" mnt_flag = %s\n", buf);
3010 flags = mp->mnt_kern_flag;
3011 #define MNT_KERN_FLAG(flag) do { \
3012 if (flags & (flag)) { \
3013 if (buf[0] != '\0') \
3014 strlcat(buf, ", ", sizeof(buf)); \
3015 strlcat(buf, (#flag) + 5, sizeof(buf)); \
3019 MNT_KERN_FLAG(MNTK_UNMOUNTF);
3020 MNT_KERN_FLAG(MNTK_ASYNC);
3021 MNT_KERN_FLAG(MNTK_SOFTDEP);
3022 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3023 MNT_KERN_FLAG(MNTK_DRAINING);
3024 MNT_KERN_FLAG(MNTK_REFEXPIRE);
3025 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3026 MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3027 MNT_KERN_FLAG(MNTK_NOASYNC);
3028 MNT_KERN_FLAG(MNTK_UNMOUNT);
3029 MNT_KERN_FLAG(MNTK_MWAIT);
3030 MNT_KERN_FLAG(MNTK_SUSPEND);
3031 MNT_KERN_FLAG(MNTK_SUSPEND2);
3032 MNT_KERN_FLAG(MNTK_SUSPENDED);
3033 MNT_KERN_FLAG(MNTK_MPSAFE);
3034 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3035 MNT_KERN_FLAG(MNTK_NOKNOTE);
3036 #undef MNT_KERN_FLAG
3039 strlcat(buf, ", ", sizeof(buf));
3040 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3043 db_printf(" mnt_kern_flag = %s\n", buf);
3045 db_printf(" mnt_opt = ");
3046 opt = TAILQ_FIRST(mp->mnt_opt);
3048 db_printf("%s", opt->name);
3049 opt = TAILQ_NEXT(opt, link);
3050 while (opt != NULL) {
3051 db_printf(", %s", opt->name);
3052 opt = TAILQ_NEXT(opt, link);
3058 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
3059 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3060 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3061 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3062 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3063 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3064 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3065 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3066 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3067 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3068 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3069 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3071 db_printf(" mnt_cred = { uid=%u ruid=%u",
3072 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3073 if (jailed(mp->mnt_cred))
3074 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3076 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
3077 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
3078 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3079 db_printf(" mnt_activevnodelistsize = %d\n",
3080 mp->mnt_activevnodelistsize);
3081 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3082 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3083 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3084 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
3085 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3086 db_printf(" mnt_secondary_accwrites = %d\n",
3087 mp->mnt_secondary_accwrites);
3088 db_printf(" mnt_gjprovider = %s\n",
3089 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3091 db_printf("\n\nList of active vnodes\n");
3092 TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3093 if (vp->v_type != VMARKER) {
3094 vn_printf(vp, "vnode ");
3099 db_printf("\n\nList of inactive vnodes\n");
3100 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3101 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3102 vn_printf(vp, "vnode ");
3111 * Fill in a struct xvfsconf based on a struct vfsconf.
3114 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
3117 strcpy(xvfsp->vfc_name, vfsp->vfc_name);
3118 xvfsp->vfc_typenum = vfsp->vfc_typenum;
3119 xvfsp->vfc_refcount = vfsp->vfc_refcount;
3120 xvfsp->vfc_flags = vfsp->vfc_flags;
3122 * These are unused in userland, we keep them
3123 * to not break binary compatibility.
3125 xvfsp->vfc_vfsops = NULL;
3126 xvfsp->vfc_next = NULL;
3130 * Top level filesystem related information gathering.
3133 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3135 struct vfsconf *vfsp;
3136 struct xvfsconf xvfsp;
3140 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3141 bzero(&xvfsp, sizeof(xvfsp));
3142 vfsconf2x(vfsp, &xvfsp);
3143 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
3150 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3151 NULL, 0, sysctl_vfs_conflist,
3152 "S,xvfsconf", "List of all configured filesystems");
3154 #ifndef BURN_BRIDGES
3155 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3158 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3160 int *name = (int *)arg1 - 1; /* XXX */
3161 u_int namelen = arg2 + 1; /* XXX */
3162 struct vfsconf *vfsp;
3163 struct xvfsconf xvfsp;
3165 printf("WARNING: userland calling deprecated sysctl, "
3166 "please rebuild world\n");
3168 #if 1 || defined(COMPAT_PRELITE2)
3169 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3171 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3175 case VFS_MAXTYPENUM:
3178 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3181 return (ENOTDIR); /* overloaded */
3182 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3183 if (vfsp->vfc_typenum == name[2])
3186 return (EOPNOTSUPP);
3187 bzero(&xvfsp, sizeof(xvfsp));
3188 vfsconf2x(vfsp, &xvfsp);
3189 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3191 return (EOPNOTSUPP);
3194 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3195 vfs_sysctl, "Generic filesystem");
3197 #if 1 || defined(COMPAT_PRELITE2)
3200 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3203 struct vfsconf *vfsp;
3204 struct ovfsconf ovfs;
3206 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3207 bzero(&ovfs, sizeof(ovfs));
3208 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3209 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3210 ovfs.vfc_index = vfsp->vfc_typenum;
3211 ovfs.vfc_refcount = vfsp->vfc_refcount;
3212 ovfs.vfc_flags = vfsp->vfc_flags;
3213 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3220 #endif /* 1 || COMPAT_PRELITE2 */
3221 #endif /* !BURN_BRIDGES */
3223 #define KINFO_VNODESLOP 10
3226 * Dump vnode list (via sysctl).
3230 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3238 * Stale numvnodes access is not fatal here.
3241 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3243 /* Make an estimate */
3244 return (SYSCTL_OUT(req, 0, len));
3246 error = sysctl_wire_old_buffer(req, 0);
3249 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3251 mtx_lock(&mountlist_mtx);
3252 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3253 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3256 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3260 xvn[n].xv_size = sizeof *xvn;
3261 xvn[n].xv_vnode = vp;
3262 xvn[n].xv_id = 0; /* XXX compat */
3263 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3265 XV_COPY(writecount);
3271 xvn[n].xv_flag = vp->v_vflag;
3273 switch (vp->v_type) {
3280 if (vp->v_rdev == NULL) {
3284 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3287 xvn[n].xv_socket = vp->v_socket;
3290 xvn[n].xv_fifo = vp->v_fifoinfo;
3295 /* shouldn't happen? */
3303 mtx_lock(&mountlist_mtx);
3308 mtx_unlock(&mountlist_mtx);
3310 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3315 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3316 0, 0, sysctl_vnode, "S,xvnode", "");
3320 * Unmount all filesystems. The list is traversed in reverse order
3321 * of mounting to avoid dependencies.
3324 vfs_unmountall(void)
3330 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread"));
3331 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3335 * Since this only runs when rebooting, it is not interlocked.
3337 while(!TAILQ_EMPTY(&mountlist)) {
3338 mp = TAILQ_LAST(&mountlist, mntlist);
3339 error = dounmount(mp, MNT_FORCE, td);
3341 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3343 * XXX: Due to the way in which we mount the root
3344 * file system off of devfs, devfs will generate a
3345 * "busy" warning when we try to unmount it before
3346 * the root. Don't print a warning as a result in
3347 * order to avoid false positive errors that may
3348 * cause needless upset.
3350 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3351 printf("unmount of %s failed (",
3352 mp->mnt_stat.f_mntonname);
3356 printf("%d)\n", error);
3359 /* The unmount has removed mp from the mountlist */
3365 * perform msync on all vnodes under a mount point
3366 * the mount point must be locked.
3369 vfs_msync(struct mount *mp, int flags)
3371 struct vnode *vp, *mvp;
3372 struct vm_object *obj;
3374 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3375 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3377 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3378 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3380 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3382 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3389 VM_OBJECT_LOCK(obj);
3390 vm_object_page_clean(obj, 0, 0,
3392 OBJPC_SYNC : OBJPC_NOSYNC);
3393 VM_OBJECT_UNLOCK(obj);
3403 destroy_vpollinfo(struct vpollinfo *vi)
3405 seldrain(&vi->vpi_selinfo);
3406 knlist_destroy(&vi->vpi_selinfo.si_note);
3407 mtx_destroy(&vi->vpi_lock);
3408 uma_zfree(vnodepoll_zone, vi);
3412 * Initalize per-vnode helper structure to hold poll-related state.
3415 v_addpollinfo(struct vnode *vp)
3417 struct vpollinfo *vi;
3419 if (vp->v_pollinfo != NULL)
3421 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3422 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3423 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3424 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3426 if (vp->v_pollinfo != NULL) {
3428 destroy_vpollinfo(vi);
3431 vp->v_pollinfo = vi;
3436 * Record a process's interest in events which might happen to
3437 * a vnode. Because poll uses the historic select-style interface
3438 * internally, this routine serves as both the ``check for any
3439 * pending events'' and the ``record my interest in future events''
3440 * functions. (These are done together, while the lock is held,
3441 * to avoid race conditions.)
3444 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3448 mtx_lock(&vp->v_pollinfo->vpi_lock);
3449 if (vp->v_pollinfo->vpi_revents & events) {
3451 * This leaves events we are not interested
3452 * in available for the other process which
3453 * which presumably had requested them
3454 * (otherwise they would never have been
3457 events &= vp->v_pollinfo->vpi_revents;
3458 vp->v_pollinfo->vpi_revents &= ~events;
3460 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3463 vp->v_pollinfo->vpi_events |= events;
3464 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3465 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3470 * Routine to create and manage a filesystem syncer vnode.
3472 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3473 static int sync_fsync(struct vop_fsync_args *);
3474 static int sync_inactive(struct vop_inactive_args *);
3475 static int sync_reclaim(struct vop_reclaim_args *);
3477 static struct vop_vector sync_vnodeops = {
3478 .vop_bypass = VOP_EOPNOTSUPP,
3479 .vop_close = sync_close, /* close */
3480 .vop_fsync = sync_fsync, /* fsync */
3481 .vop_inactive = sync_inactive, /* inactive */
3482 .vop_reclaim = sync_reclaim, /* reclaim */
3483 .vop_lock1 = vop_stdlock, /* lock */
3484 .vop_unlock = vop_stdunlock, /* unlock */
3485 .vop_islocked = vop_stdislocked, /* islocked */
3489 * Create a new filesystem syncer vnode for the specified mount point.
3492 vfs_allocate_syncvnode(struct mount *mp)
3496 static long start, incr, next;
3499 /* Allocate a new vnode */
3500 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3502 panic("vfs_allocate_syncvnode: getnewvnode() failed");
3504 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3505 vp->v_vflag |= VV_FORCEINSMQ;
3506 error = insmntque(vp, mp);
3508 panic("vfs_allocate_syncvnode: insmntque() failed");
3509 vp->v_vflag &= ~VV_FORCEINSMQ;
3512 * Place the vnode onto the syncer worklist. We attempt to
3513 * scatter them about on the list so that they will go off
3514 * at evenly distributed times even if all the filesystems
3515 * are mounted at once.
3518 if (next == 0 || next > syncer_maxdelay) {
3522 start = syncer_maxdelay / 2;
3523 incr = syncer_maxdelay;
3529 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3530 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3531 mtx_lock(&sync_mtx);
3533 if (mp->mnt_syncer == NULL) {
3534 mp->mnt_syncer = vp;
3537 mtx_unlock(&sync_mtx);
3540 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3547 vfs_deallocate_syncvnode(struct mount *mp)
3551 mtx_lock(&sync_mtx);
3552 vp = mp->mnt_syncer;
3554 mp->mnt_syncer = NULL;
3555 mtx_unlock(&sync_mtx);
3561 * Do a lazy sync of the filesystem.
3564 sync_fsync(struct vop_fsync_args *ap)
3566 struct vnode *syncvp = ap->a_vp;
3567 struct mount *mp = syncvp->v_mount;
3572 * We only need to do something if this is a lazy evaluation.
3574 if (ap->a_waitfor != MNT_LAZY)
3578 * Move ourselves to the back of the sync list.
3580 bo = &syncvp->v_bufobj;
3582 vn_syncer_add_to_worklist(bo, syncdelay);
3586 * Walk the list of vnodes pushing all that are dirty and
3587 * not already on the sync list.
3589 mtx_lock(&mountlist_mtx);
3590 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3591 mtx_unlock(&mountlist_mtx);
3594 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3598 save = curthread_pflags_set(TDP_SYNCIO);
3599 vfs_msync(mp, MNT_NOWAIT);
3600 error = VFS_SYNC(mp, MNT_LAZY);
3601 curthread_pflags_restore(save);
3602 vn_finished_write(mp);
3608 * The syncer vnode is no referenced.
3611 sync_inactive(struct vop_inactive_args *ap)
3619 * The syncer vnode is no longer needed and is being decommissioned.
3621 * Modifications to the worklist must be protected by sync_mtx.
3624 sync_reclaim(struct vop_reclaim_args *ap)
3626 struct vnode *vp = ap->a_vp;
3631 mtx_lock(&sync_mtx);
3632 if (vp->v_mount->mnt_syncer == vp)
3633 vp->v_mount->mnt_syncer = NULL;
3634 if (bo->bo_flag & BO_ONWORKLST) {
3635 LIST_REMOVE(bo, bo_synclist);
3636 syncer_worklist_len--;
3638 bo->bo_flag &= ~BO_ONWORKLST;
3640 mtx_unlock(&sync_mtx);
3647 * Check if vnode represents a disk device
3650 vn_isdisk(struct vnode *vp, int *errp)
3656 if (vp->v_type != VCHR)
3658 else if (vp->v_rdev == NULL)
3660 else if (vp->v_rdev->si_devsw == NULL)
3662 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3667 return (error == 0);
3671 * Common filesystem object access control check routine. Accepts a
3672 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3673 * and optional call-by-reference privused argument allowing vaccess()
3674 * to indicate to the caller whether privilege was used to satisfy the
3675 * request (obsoleted). Returns 0 on success, or an errno on failure.
3678 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3679 accmode_t accmode, struct ucred *cred, int *privused)
3681 accmode_t dac_granted;
3682 accmode_t priv_granted;
3684 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3685 ("invalid bit in accmode"));
3686 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3687 ("VAPPEND without VWRITE"));
3690 * Look for a normal, non-privileged way to access the file/directory
3691 * as requested. If it exists, go with that.
3694 if (privused != NULL)
3699 /* Check the owner. */
3700 if (cred->cr_uid == file_uid) {
3701 dac_granted |= VADMIN;
3702 if (file_mode & S_IXUSR)
3703 dac_granted |= VEXEC;
3704 if (file_mode & S_IRUSR)
3705 dac_granted |= VREAD;
3706 if (file_mode & S_IWUSR)
3707 dac_granted |= (VWRITE | VAPPEND);
3709 if ((accmode & dac_granted) == accmode)
3715 /* Otherwise, check the groups (first match) */
3716 if (groupmember(file_gid, cred)) {
3717 if (file_mode & S_IXGRP)
3718 dac_granted |= VEXEC;
3719 if (file_mode & S_IRGRP)
3720 dac_granted |= VREAD;
3721 if (file_mode & S_IWGRP)
3722 dac_granted |= (VWRITE | VAPPEND);
3724 if ((accmode & dac_granted) == accmode)
3730 /* Otherwise, check everyone else. */
3731 if (file_mode & S_IXOTH)
3732 dac_granted |= VEXEC;
3733 if (file_mode & S_IROTH)
3734 dac_granted |= VREAD;
3735 if (file_mode & S_IWOTH)
3736 dac_granted |= (VWRITE | VAPPEND);
3737 if ((accmode & dac_granted) == accmode)
3742 * Build a privilege mask to determine if the set of privileges
3743 * satisfies the requirements when combined with the granted mask
3744 * from above. For each privilege, if the privilege is required,
3745 * bitwise or the request type onto the priv_granted mask.
3751 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3752 * requests, instead of PRIV_VFS_EXEC.
3754 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3755 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3756 priv_granted |= VEXEC;
3759 * Ensure that at least one execute bit is on. Otherwise,
3760 * a privileged user will always succeed, and we don't want
3761 * this to happen unless the file really is executable.
3763 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3764 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3765 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3766 priv_granted |= VEXEC;
3769 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3770 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3771 priv_granted |= VREAD;
3773 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3774 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3775 priv_granted |= (VWRITE | VAPPEND);
3777 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3778 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3779 priv_granted |= VADMIN;
3781 if ((accmode & (priv_granted | dac_granted)) == accmode) {
3782 /* XXX audit: privilege used */
3783 if (privused != NULL)
3788 return ((accmode & VADMIN) ? EPERM : EACCES);
3792 * Credential check based on process requesting service, and per-attribute
3796 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3797 struct thread *td, accmode_t accmode)
3801 * Kernel-invoked always succeeds.
3807 * Do not allow privileged processes in jail to directly manipulate
3808 * system attributes.
3810 switch (attrnamespace) {
3811 case EXTATTR_NAMESPACE_SYSTEM:
3812 /* Potentially should be: return (EPERM); */
3813 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3814 case EXTATTR_NAMESPACE_USER:
3815 return (VOP_ACCESS(vp, accmode, cred, td));
3821 #ifdef DEBUG_VFS_LOCKS
3823 * This only exists to supress warnings from unlocked specfs accesses. It is
3824 * no longer ok to have an unlocked VFS.
3826 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
3827 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
3829 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3830 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3831 "Drop into debugger on lock violation");
3833 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3834 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3835 0, "Check for interlock across VOPs");
3837 int vfs_badlock_print = 1; /* Print lock violations. */
3838 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3839 0, "Print lock violations");
3842 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3843 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3844 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3848 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3852 if (vfs_badlock_backtrace)
3855 if (vfs_badlock_print)
3856 printf("%s: %p %s\n", str, (void *)vp, msg);
3857 if (vfs_badlock_ddb)
3858 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3862 assert_vi_locked(struct vnode *vp, const char *str)
3865 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3866 vfs_badlock("interlock is not locked but should be", str, vp);
3870 assert_vi_unlocked(struct vnode *vp, const char *str)
3873 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3874 vfs_badlock("interlock is locked but should not be", str, vp);
3878 assert_vop_locked(struct vnode *vp, const char *str)
3881 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0)
3882 vfs_badlock("is not locked but should be", str, vp);
3886 assert_vop_unlocked(struct vnode *vp, const char *str)
3889 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3890 vfs_badlock("is locked but should not be", str, vp);
3894 assert_vop_elocked(struct vnode *vp, const char *str)
3897 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3898 vfs_badlock("is not exclusive locked but should be", str, vp);
3903 assert_vop_elocked_other(struct vnode *vp, const char *str)
3906 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3907 vfs_badlock("is not exclusive locked by another thread",
3912 assert_vop_slocked(struct vnode *vp, const char *str)
3915 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3916 vfs_badlock("is not locked shared but should be", str, vp);
3919 #endif /* DEBUG_VFS_LOCKS */
3922 vop_rename_fail(struct vop_rename_args *ap)
3925 if (ap->a_tvp != NULL)
3927 if (ap->a_tdvp == ap->a_tvp)
3936 vop_rename_pre(void *ap)
3938 struct vop_rename_args *a = ap;
3940 #ifdef DEBUG_VFS_LOCKS
3942 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
3943 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
3944 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
3945 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
3947 /* Check the source (from). */
3948 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
3949 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
3950 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
3951 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
3952 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
3954 /* Check the target. */
3956 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
3957 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
3959 if (a->a_tdvp != a->a_fdvp)
3961 if (a->a_tvp != a->a_fvp)
3969 vop_strategy_pre(void *ap)
3971 #ifdef DEBUG_VFS_LOCKS
3972 struct vop_strategy_args *a;
3979 * Cluster ops lock their component buffers but not the IO container.
3981 if ((bp->b_flags & B_CLUSTER) != 0)
3984 if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
3985 if (vfs_badlock_print)
3987 "VOP_STRATEGY: bp is not locked but should be\n");
3988 if (vfs_badlock_ddb)
3989 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3995 vop_lookup_pre(void *ap)
3997 #ifdef DEBUG_VFS_LOCKS
3998 struct vop_lookup_args *a;
4003 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
4004 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
4009 vop_lookup_post(void *ap, int rc)
4011 #ifdef DEBUG_VFS_LOCKS
4012 struct vop_lookup_args *a;
4020 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
4021 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
4024 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)");
4029 vop_lock_pre(void *ap)
4031 #ifdef DEBUG_VFS_LOCKS
4032 struct vop_lock1_args *a = ap;
4034 if ((a->a_flags & LK_INTERLOCK) == 0)
4035 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4037 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4042 vop_lock_post(void *ap, int rc)
4044 #ifdef DEBUG_VFS_LOCKS
4045 struct vop_lock1_args *a = ap;
4047 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4049 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4054 vop_unlock_pre(void *ap)
4056 #ifdef DEBUG_VFS_LOCKS
4057 struct vop_unlock_args *a = ap;
4059 if (a->a_flags & LK_INTERLOCK)
4060 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4061 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4066 vop_unlock_post(void *ap, int rc)
4068 #ifdef DEBUG_VFS_LOCKS
4069 struct vop_unlock_args *a = ap;
4071 if (a->a_flags & LK_INTERLOCK)
4072 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4077 vop_create_post(void *ap, int rc)
4079 struct vop_create_args *a = ap;
4082 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4086 vop_deleteextattr_post(void *ap, int rc)
4088 struct vop_deleteextattr_args *a = ap;
4091 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4095 vop_link_post(void *ap, int rc)
4097 struct vop_link_args *a = ap;
4100 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4101 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4106 vop_mkdir_post(void *ap, int rc)
4108 struct vop_mkdir_args *a = ap;
4111 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4115 vop_mknod_post(void *ap, int rc)
4117 struct vop_mknod_args *a = ap;
4120 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4124 vop_remove_post(void *ap, int rc)
4126 struct vop_remove_args *a = ap;
4129 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4130 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4135 vop_rename_post(void *ap, int rc)
4137 struct vop_rename_args *a = ap;
4140 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4141 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4142 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4144 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4146 if (a->a_tdvp != a->a_fdvp)
4148 if (a->a_tvp != a->a_fvp)
4156 vop_rmdir_post(void *ap, int rc)
4158 struct vop_rmdir_args *a = ap;
4161 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4162 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4167 vop_setattr_post(void *ap, int rc)
4169 struct vop_setattr_args *a = ap;
4172 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4176 vop_setextattr_post(void *ap, int rc)
4178 struct vop_setextattr_args *a = ap;
4181 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4185 vop_symlink_post(void *ap, int rc)
4187 struct vop_symlink_args *a = ap;
4190 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4193 static struct knlist fs_knlist;
4196 vfs_event_init(void *arg)
4198 knlist_init_mtx(&fs_knlist, NULL);
4200 /* XXX - correct order? */
4201 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4204 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4207 KNOTE_UNLOCKED(&fs_knlist, event);
4210 static int filt_fsattach(struct knote *kn);
4211 static void filt_fsdetach(struct knote *kn);
4212 static int filt_fsevent(struct knote *kn, long hint);
4214 struct filterops fs_filtops = {
4216 .f_attach = filt_fsattach,
4217 .f_detach = filt_fsdetach,
4218 .f_event = filt_fsevent
4222 filt_fsattach(struct knote *kn)
4225 kn->kn_flags |= EV_CLEAR;
4226 knlist_add(&fs_knlist, kn, 0);
4231 filt_fsdetach(struct knote *kn)
4234 knlist_remove(&fs_knlist, kn, 0);
4238 filt_fsevent(struct knote *kn, long hint)
4241 kn->kn_fflags |= hint;
4242 return (kn->kn_fflags != 0);
4246 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4252 error = SYSCTL_IN(req, &vc, sizeof(vc));
4255 if (vc.vc_vers != VFS_CTL_VERS1)
4257 mp = vfs_getvfs(&vc.vc_fsid);
4260 /* ensure that a specific sysctl goes to the right filesystem. */
4261 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4262 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4266 VCTLTOREQ(&vc, req);
4267 error = VFS_SYSCTL(mp, vc.vc_op, req);
4272 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4273 NULL, 0, sysctl_vfs_ctl, "",
4277 * Function to initialize a va_filerev field sensibly.
4278 * XXX: Wouldn't a random number make a lot more sense ??
4281 init_va_filerev(void)
4286 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4289 static int filt_vfsread(struct knote *kn, long hint);
4290 static int filt_vfswrite(struct knote *kn, long hint);
4291 static int filt_vfsvnode(struct knote *kn, long hint);
4292 static void filt_vfsdetach(struct knote *kn);
4293 static struct filterops vfsread_filtops = {
4295 .f_detach = filt_vfsdetach,
4296 .f_event = filt_vfsread
4298 static struct filterops vfswrite_filtops = {
4300 .f_detach = filt_vfsdetach,
4301 .f_event = filt_vfswrite
4303 static struct filterops vfsvnode_filtops = {
4305 .f_detach = filt_vfsdetach,
4306 .f_event = filt_vfsvnode
4310 vfs_knllock(void *arg)
4312 struct vnode *vp = arg;
4314 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4318 vfs_knlunlock(void *arg)
4320 struct vnode *vp = arg;
4326 vfs_knl_assert_locked(void *arg)
4328 #ifdef DEBUG_VFS_LOCKS
4329 struct vnode *vp = arg;
4331 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4336 vfs_knl_assert_unlocked(void *arg)
4338 #ifdef DEBUG_VFS_LOCKS
4339 struct vnode *vp = arg;
4341 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4346 vfs_kqfilter(struct vop_kqfilter_args *ap)
4348 struct vnode *vp = ap->a_vp;
4349 struct knote *kn = ap->a_kn;
4352 switch (kn->kn_filter) {
4354 kn->kn_fop = &vfsread_filtops;
4357 kn->kn_fop = &vfswrite_filtops;
4360 kn->kn_fop = &vfsvnode_filtops;
4366 kn->kn_hook = (caddr_t)vp;
4369 if (vp->v_pollinfo == NULL)
4371 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4372 knlist_add(knl, kn, 0);
4378 * Detach knote from vnode
4381 filt_vfsdetach(struct knote *kn)
4383 struct vnode *vp = (struct vnode *)kn->kn_hook;
4385 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4386 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4391 filt_vfsread(struct knote *kn, long hint)
4393 struct vnode *vp = (struct vnode *)kn->kn_hook;
4398 * filesystem is gone, so set the EOF flag and schedule
4399 * the knote for deletion.
4401 if (hint == NOTE_REVOKE) {
4403 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4408 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4412 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4413 res = (kn->kn_data != 0);
4420 filt_vfswrite(struct knote *kn, long hint)
4422 struct vnode *vp = (struct vnode *)kn->kn_hook;
4427 * filesystem is gone, so set the EOF flag and schedule
4428 * the knote for deletion.
4430 if (hint == NOTE_REVOKE)
4431 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4439 filt_vfsvnode(struct knote *kn, long hint)
4441 struct vnode *vp = (struct vnode *)kn->kn_hook;
4445 if (kn->kn_sfflags & hint)
4446 kn->kn_fflags |= hint;
4447 if (hint == NOTE_REVOKE) {
4448 kn->kn_flags |= EV_EOF;
4452 res = (kn->kn_fflags != 0);
4458 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4462 if (dp->d_reclen > ap->a_uio->uio_resid)
4463 return (ENAMETOOLONG);
4464 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4466 if (ap->a_ncookies != NULL) {
4467 if (ap->a_cookies != NULL)
4468 free(ap->a_cookies, M_TEMP);
4469 ap->a_cookies = NULL;
4470 *ap->a_ncookies = 0;
4474 if (ap->a_ncookies == NULL)
4477 KASSERT(ap->a_cookies,
4478 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4480 *ap->a_cookies = realloc(*ap->a_cookies,
4481 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4482 (*ap->a_cookies)[*ap->a_ncookies] = off;
4487 * Mark for update the access time of the file if the filesystem
4488 * supports VOP_MARKATIME. This functionality is used by execve and
4489 * mmap, so we want to avoid the I/O implied by directly setting
4490 * va_atime for the sake of efficiency.
4493 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4498 VFS_ASSERT_GIANT(mp);
4499 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4500 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4501 (void)VOP_MARKATIME(vp);
4505 * The purpose of this routine is to remove granularity from accmode_t,
4506 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4507 * VADMIN and VAPPEND.
4509 * If it returns 0, the caller is supposed to continue with the usual
4510 * access checks using 'accmode' as modified by this routine. If it
4511 * returns nonzero value, the caller is supposed to return that value
4514 * Note that after this routine runs, accmode may be zero.
4517 vfs_unixify_accmode(accmode_t *accmode)
4520 * There is no way to specify explicit "deny" rule using
4521 * file mode or POSIX.1e ACLs.
4523 if (*accmode & VEXPLICIT_DENY) {
4529 * None of these can be translated into usual access bits.
4530 * Also, the common case for NFSv4 ACLs is to not contain
4531 * either of these bits. Caller should check for VWRITE
4532 * on the containing directory instead.
4534 if (*accmode & (VDELETE_CHILD | VDELETE))
4537 if (*accmode & VADMIN_PERMS) {
4538 *accmode &= ~VADMIN_PERMS;
4543 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4544 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4546 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4552 * These are helper functions for filesystems to traverse all
4553 * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4555 * This interface replaces MNT_VNODE_FOREACH.
4558 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4561 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4566 kern_yield(PRI_UNCHANGED);
4568 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4569 vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4570 while (vp != NULL && (vp->v_type == VMARKER ||
4571 (vp->v_iflag & VI_DOOMED) != 0))
4572 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4574 /* Check if we are done */
4576 __mnt_vnode_markerfree_all(mvp, mp);
4577 /* MNT_IUNLOCK(mp); -- done in above function */
4578 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4581 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4582 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4589 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4593 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4596 (*mvp)->v_type = VMARKER;
4598 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4599 while (vp != NULL && (vp->v_type == VMARKER ||
4600 (vp->v_iflag & VI_DOOMED) != 0))
4601 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4603 /* Check if we are done */
4607 free(*mvp, M_VNODE_MARKER);
4611 (*mvp)->v_mount = mp;
4612 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4620 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4628 mtx_assert(MNT_MTX(mp), MA_OWNED);
4630 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4631 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4634 free(*mvp, M_VNODE_MARKER);
4639 * These are helper functions for filesystems to traverse their
4640 * active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4643 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4645 struct vnode *vp, *nvp;
4648 kern_yield(PRI_UNCHANGED);
4650 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4651 vp = TAILQ_NEXT(*mvp, v_actfreelist);
4652 while (vp != NULL) {
4654 if (vp->v_mount == mp && vp->v_type != VMARKER &&
4655 (vp->v_iflag & VI_DOOMED) == 0)
4657 nvp = TAILQ_NEXT(vp, v_actfreelist);
4662 /* Check if we are done */
4664 __mnt_vnode_markerfree_active(mvp, mp);
4665 /* MNT_IUNLOCK(mp); -- done in above function */
4666 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4669 mtx_lock(&vnode_free_list_mtx);
4670 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4671 TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4672 mtx_unlock(&vnode_free_list_mtx);
4678 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4680 struct vnode *vp, *nvp;
4682 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4685 (*mvp)->v_type = VMARKER;
4687 vp = TAILQ_NEXT(*mvp, v_actfreelist);
4688 while (vp != NULL) {
4690 if (vp->v_mount == mp && vp->v_type != VMARKER &&
4691 (vp->v_iflag & VI_DOOMED) == 0)
4693 nvp = TAILQ_NEXT(vp, v_actfreelist);
4698 /* Check if we are done */
4702 free(*mvp, M_VNODE_MARKER);
4706 (*mvp)->v_mount = mp;
4707 mtx_lock(&vnode_free_list_mtx);
4708 TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4709 mtx_unlock(&vnode_free_list_mtx);
4715 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4723 mtx_assert(MNT_MTX(mp), MA_OWNED);
4725 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4726 mtx_lock(&vnode_free_list_mtx);
4727 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4728 mtx_unlock(&vnode_free_list_mtx);
4731 free(*mvp, M_VNODE_MARKER);