2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
38 * External virtual filesystem routines
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
44 #include "opt_compat.h"
46 #include "opt_watchdog.h"
48 #include <sys/param.h>
49 #include <sys/systm.h>
52 #include <sys/condvar.h>
54 #include <sys/dirent.h>
55 #include <sys/event.h>
56 #include <sys/eventhandler.h>
57 #include <sys/extattr.h>
59 #include <sys/fcntl.h>
62 #include <sys/kernel.h>
63 #include <sys/kthread.h>
64 #include <sys/lockf.h>
65 #include <sys/malloc.h>
66 #include <sys/mount.h>
67 #include <sys/namei.h>
68 #include <sys/pctrie.h>
70 #include <sys/reboot.h>
71 #include <sys/rwlock.h>
72 #include <sys/sched.h>
73 #include <sys/sleepqueue.h>
76 #include <sys/sysctl.h>
77 #include <sys/syslog.h>
78 #include <sys/vmmeter.h>
79 #include <sys/vnode.h>
80 #include <sys/watchdog.h>
82 #include <machine/stdarg.h>
84 #include <security/mac/mac_framework.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_extern.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_page.h>
92 #include <vm/vm_kern.h>
99 static void delmntque(struct vnode *vp);
100 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
101 int slpflag, int slptimeo);
102 static void syncer_shutdown(void *arg, int howto);
103 static int vtryrecycle(struct vnode *vp);
104 static void v_incr_usecount(struct vnode *);
105 static void v_decr_usecount(struct vnode *);
106 static void v_decr_useonly(struct vnode *);
107 static void v_upgrade_usecount(struct vnode *);
108 static void vnlru_free(int);
109 static void vgonel(struct vnode *);
110 static void vfs_knllock(void *arg);
111 static void vfs_knlunlock(void *arg);
112 static void vfs_knl_assert_locked(void *arg);
113 static void vfs_knl_assert_unlocked(void *arg);
114 static void destroy_vpollinfo(struct vpollinfo *vi);
117 * Number of vnodes in existence. Increased whenever getnewvnode()
118 * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
120 static unsigned long numvnodes;
122 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
123 "Number of vnodes in existence");
126 * Conversion tables for conversion from vnode types to inode formats
129 enum vtype iftovt_tab[16] = {
130 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
131 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
133 int vttoif_tab[10] = {
134 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
135 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
139 * List of vnodes that are ready for recycling.
141 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
144 * Free vnode target. Free vnodes may simply be files which have been stat'd
145 * but not read. This is somewhat common, and a small cache of such files
146 * should be kept to avoid recreation costs.
148 static u_long wantfreevnodes;
149 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
150 /* Number of vnodes in the free list. */
151 static u_long freevnodes;
152 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
153 "Number of vnodes in the free list");
155 static int vlru_allow_cache_src;
156 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
157 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
160 * Various variables used for debugging the new implementation of
162 * XXX these are probably of (very) limited utility now.
164 static int reassignbufcalls;
165 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
166 "Number of calls to reassignbuf");
169 * Cache for the mount type id assigned to NFS. This is used for
170 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
172 int nfs_mount_type = -1;
174 /* To keep more than one thread at a time from running vfs_getnewfsid */
175 static struct mtx mntid_mtx;
178 * Lock for any access to the following:
183 static struct mtx vnode_free_list_mtx;
185 /* Publicly exported FS */
186 struct nfs_public nfs_pub;
188 static uma_zone_t buf_trie_zone;
190 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
191 static uma_zone_t vnode_zone;
192 static uma_zone_t vnodepoll_zone;
195 * The workitem queue.
197 * It is useful to delay writes of file data and filesystem metadata
198 * for tens of seconds so that quickly created and deleted files need
199 * not waste disk bandwidth being created and removed. To realize this,
200 * we append vnodes to a "workitem" queue. When running with a soft
201 * updates implementation, most pending metadata dependencies should
202 * not wait for more than a few seconds. Thus, mounted on block devices
203 * are delayed only about a half the time that file data is delayed.
204 * Similarly, directory updates are more critical, so are only delayed
205 * about a third the time that file data is delayed. Thus, there are
206 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
207 * one each second (driven off the filesystem syncer process). The
208 * syncer_delayno variable indicates the next queue that is to be processed.
209 * Items that need to be processed soon are placed in this queue:
211 * syncer_workitem_pending[syncer_delayno]
213 * A delay of fifteen seconds is done by placing the request fifteen
214 * entries later in the queue:
216 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
219 static int syncer_delayno;
220 static long syncer_mask;
221 LIST_HEAD(synclist, bufobj);
222 static struct synclist *syncer_workitem_pending;
224 * The sync_mtx protects:
229 * syncer_workitem_pending
230 * syncer_worklist_len
233 static struct mtx sync_mtx;
234 static struct cv sync_wakeup;
236 #define SYNCER_MAXDELAY 32
237 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
238 static int syncdelay = 30; /* max time to delay syncing data */
239 static int filedelay = 30; /* time to delay syncing files */
240 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
241 "Time to delay syncing files (in seconds)");
242 static int dirdelay = 29; /* time to delay syncing directories */
243 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
244 "Time to delay syncing directories (in seconds)");
245 static int metadelay = 28; /* time to delay syncing metadata */
246 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
247 "Time to delay syncing metadata (in seconds)");
248 static int rushjob; /* number of slots to run ASAP */
249 static int stat_rush_requests; /* number of times I/O speeded up */
250 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
251 "Number of times I/O speeded up (rush requests)");
254 * When shutting down the syncer, run it at four times normal speed.
256 #define SYNCER_SHUTDOWN_SPEEDUP 4
257 static int sync_vnode_count;
258 static int syncer_worklist_len;
259 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
263 * Number of vnodes we want to exist at any one time. This is mostly used
264 * to size hash tables in vnode-related code. It is normally not used in
265 * getnewvnode(), as wantfreevnodes is normally nonzero.)
267 * XXX desiredvnodes is historical cruft and should not exist.
270 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
271 &desiredvnodes, 0, "Maximum number of vnodes");
272 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
273 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
274 static int vnlru_nowhere;
275 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
276 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
279 * Macros to control when a vnode is freed and recycled. All require
280 * the vnode interlock.
282 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
283 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
284 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
286 /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
290 * Support for the bufobj clean & dirty pctrie.
293 buf_trie_alloc(struct pctrie *ptree)
296 return uma_zalloc(buf_trie_zone, M_NOWAIT);
300 buf_trie_free(struct pctrie *ptree, void *node)
303 uma_zfree(buf_trie_zone, node);
305 PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
308 * Initialize the vnode management data structures.
310 * Reevaluate the following cap on the number of vnodes after the physical
311 * memory size exceeds 512GB. In the limit, as the physical memory size
312 * grows, the ratio of physical pages to vnodes approaches sixteen to one.
314 #ifndef MAXVNODES_MAX
315 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
318 vntblinit(void *dummy __unused)
321 int physvnodes, virtvnodes;
324 * Desiredvnodes is a function of the physical memory size and the
325 * kernel's heap size. Generally speaking, it scales with the
326 * physical memory size. The ratio of desiredvnodes to physical pages
327 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the
328 * marginal ratio of desiredvnodes to physical pages is one to
329 * sixteen. However, desiredvnodes is limited by the kernel's heap
330 * size. The memory required by desiredvnodes vnodes and vm objects
331 * may not exceed one seventh of the kernel's heap size.
333 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
334 cnt.v_page_count) / 16;
335 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
336 sizeof(struct vnode)));
337 desiredvnodes = min(physvnodes, virtvnodes);
338 if (desiredvnodes > MAXVNODES_MAX) {
340 printf("Reducing kern.maxvnodes %d -> %d\n",
341 desiredvnodes, MAXVNODES_MAX);
342 desiredvnodes = MAXVNODES_MAX;
344 wantfreevnodes = desiredvnodes / 4;
345 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
346 TAILQ_INIT(&vnode_free_list);
347 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
348 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
349 NULL, NULL, UMA_ALIGN_PTR, 0);
350 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
351 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
353 * Preallocate enough nodes to support one-per buf so that
354 * we can not fail an insert. reassignbuf() callers can not
355 * tolerate the insertion failure.
357 buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
358 NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR,
359 UMA_ZONE_NOFREE | UMA_ZONE_VM);
360 uma_prealloc(buf_trie_zone, nbuf);
362 * Initialize the filesystem syncer.
364 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
366 syncer_maxdelay = syncer_mask + 1;
367 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
368 cv_init(&sync_wakeup, "syncer");
369 for (i = 1; i <= sizeof(struct vnode); i <<= 1)
373 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
377 * Mark a mount point as busy. Used to synchronize access and to delay
378 * unmounting. Eventually, mountlist_mtx is not released on failure.
380 * vfs_busy() is a custom lock, it can block the caller.
381 * vfs_busy() only sleeps if the unmount is active on the mount point.
382 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
383 * vnode belonging to mp.
385 * Lookup uses vfs_busy() to traverse mount points.
387 * / vnode lock A / vnode lock (/var) D
388 * /var vnode lock B /log vnode lock(/var/log) E
389 * vfs_busy lock C vfs_busy lock F
391 * Within each file system, the lock order is C->A->B and F->D->E.
393 * When traversing across mounts, the system follows that lock order:
399 * The lookup() process for namei("/var") illustrates the process:
400 * VOP_LOOKUP() obtains B while A is held
401 * vfs_busy() obtains a shared lock on F while A and B are held
402 * vput() releases lock on B
403 * vput() releases lock on A
404 * VFS_ROOT() obtains lock on D while shared lock on F is held
405 * vfs_unbusy() releases shared lock on F
406 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
407 * Attempt to lock A (instead of vp_crossmp) while D is held would
408 * violate the global order, causing deadlocks.
410 * dounmount() locks B while F is drained.
413 vfs_busy(struct mount *mp, int flags)
416 MPASS((flags & ~MBF_MASK) == 0);
417 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
422 * If mount point is currenly being unmounted, sleep until the
423 * mount point fate is decided. If thread doing the unmounting fails,
424 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
425 * that this mount point has survived the unmount attempt and vfs_busy
426 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
427 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
428 * about to be really destroyed. vfs_busy needs to release its
429 * reference on the mount point in this case and return with ENOENT,
430 * telling the caller that mount mount it tried to busy is no longer
433 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
434 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
437 CTR1(KTR_VFS, "%s: failed busying before sleeping",
441 if (flags & MBF_MNTLSTLOCK)
442 mtx_unlock(&mountlist_mtx);
443 mp->mnt_kern_flag |= MNTK_MWAIT;
444 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
445 if (flags & MBF_MNTLSTLOCK)
446 mtx_lock(&mountlist_mtx);
449 if (flags & MBF_MNTLSTLOCK)
450 mtx_unlock(&mountlist_mtx);
457 * Free a busy filesystem.
460 vfs_unbusy(struct mount *mp)
463 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
466 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
468 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
469 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
470 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
471 mp->mnt_kern_flag &= ~MNTK_DRAINING;
472 wakeup(&mp->mnt_lockref);
478 * Lookup a mount point by filesystem identifier.
481 vfs_getvfs(fsid_t *fsid)
485 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
486 mtx_lock(&mountlist_mtx);
487 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
488 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
489 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
491 mtx_unlock(&mountlist_mtx);
495 mtx_unlock(&mountlist_mtx);
496 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
497 return ((struct mount *) 0);
501 * Lookup a mount point by filesystem identifier, busying it before
505 vfs_busyfs(fsid_t *fsid)
510 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
511 mtx_lock(&mountlist_mtx);
512 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
513 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
514 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
515 error = vfs_busy(mp, MBF_MNTLSTLOCK);
517 mtx_unlock(&mountlist_mtx);
523 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
524 mtx_unlock(&mountlist_mtx);
525 return ((struct mount *) 0);
529 * Check if a user can access privileged mount options.
532 vfs_suser(struct mount *mp, struct thread *td)
537 * If the thread is jailed, but this is not a jail-friendly file
538 * system, deny immediately.
540 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
544 * If the file system was mounted outside the jail of the calling
545 * thread, deny immediately.
547 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
551 * If file system supports delegated administration, we don't check
552 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
553 * by the file system itself.
554 * If this is not the user that did original mount, we check for
555 * the PRIV_VFS_MOUNT_OWNER privilege.
557 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
558 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
559 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
566 * Get a new unique fsid. Try to make its val[0] unique, since this value
567 * will be used to create fake device numbers for stat(). Also try (but
568 * not so hard) make its val[0] unique mod 2^16, since some emulators only
569 * support 16-bit device numbers. We end up with unique val[0]'s for the
570 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
572 * Keep in mind that several mounts may be running in parallel. Starting
573 * the search one past where the previous search terminated is both a
574 * micro-optimization and a defense against returning the same fsid to
578 vfs_getnewfsid(struct mount *mp)
580 static uint16_t mntid_base;
585 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
586 mtx_lock(&mntid_mtx);
587 mtype = mp->mnt_vfc->vfc_typenum;
588 tfsid.val[1] = mtype;
589 mtype = (mtype & 0xFF) << 24;
591 tfsid.val[0] = makedev(255,
592 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
594 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
598 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
599 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
600 mtx_unlock(&mntid_mtx);
604 * Knob to control the precision of file timestamps:
606 * 0 = seconds only; nanoseconds zeroed.
607 * 1 = seconds and nanoseconds, accurate within 1/HZ.
608 * 2 = seconds and nanoseconds, truncated to microseconds.
609 * >=3 = seconds and nanoseconds, maximum precision.
611 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
613 static int timestamp_precision = TSP_SEC;
614 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
615 ×tamp_precision, 0, "File timestamp precision (0: seconds, "
616 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
617 "3+: sec + ns (max. precision))");
620 * Get a current timestamp.
623 vfs_timestamp(struct timespec *tsp)
627 switch (timestamp_precision) {
629 tsp->tv_sec = time_second;
637 TIMEVAL_TO_TIMESPEC(&tv, tsp);
647 * Set vnode attributes to VNOVAL
650 vattr_null(struct vattr *vap)
654 vap->va_size = VNOVAL;
655 vap->va_bytes = VNOVAL;
656 vap->va_mode = VNOVAL;
657 vap->va_nlink = VNOVAL;
658 vap->va_uid = VNOVAL;
659 vap->va_gid = VNOVAL;
660 vap->va_fsid = VNOVAL;
661 vap->va_fileid = VNOVAL;
662 vap->va_blocksize = VNOVAL;
663 vap->va_rdev = VNOVAL;
664 vap->va_atime.tv_sec = VNOVAL;
665 vap->va_atime.tv_nsec = VNOVAL;
666 vap->va_mtime.tv_sec = VNOVAL;
667 vap->va_mtime.tv_nsec = VNOVAL;
668 vap->va_ctime.tv_sec = VNOVAL;
669 vap->va_ctime.tv_nsec = VNOVAL;
670 vap->va_birthtime.tv_sec = VNOVAL;
671 vap->va_birthtime.tv_nsec = VNOVAL;
672 vap->va_flags = VNOVAL;
673 vap->va_gen = VNOVAL;
678 * This routine is called when we have too many vnodes. It attempts
679 * to free <count> vnodes and will potentially free vnodes that still
680 * have VM backing store (VM backing store is typically the cause
681 * of a vnode blowout so we want to do this). Therefore, this operation
682 * is not considered cheap.
684 * A number of conditions may prevent a vnode from being reclaimed.
685 * the buffer cache may have references on the vnode, a directory
686 * vnode may still have references due to the namei cache representing
687 * underlying files, or the vnode may be in active use. It is not
688 * desireable to reuse such vnodes. These conditions may cause the
689 * number of vnodes to reach some minimum value regardless of what
690 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
693 vlrureclaim(struct mount *mp)
702 * Calculate the trigger point, don't allow user
703 * screwups to blow us up. This prevents us from
704 * recycling vnodes with lots of resident pages. We
705 * aren't trying to free memory, we are trying to
708 usevnodes = desiredvnodes;
711 trigger = cnt.v_page_count * 2 / usevnodes;
713 vn_start_write(NULL, &mp, V_WAIT);
715 count = mp->mnt_nvnodelistsize / 10 + 1;
717 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
718 while (vp != NULL && vp->v_type == VMARKER)
719 vp = TAILQ_NEXT(vp, v_nmntvnodes);
722 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
723 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
728 * If it's been deconstructed already, it's still
729 * referenced, or it exceeds the trigger, skip it.
731 if (vp->v_usecount ||
732 (!vlru_allow_cache_src &&
733 !LIST_EMPTY(&(vp)->v_cache_src)) ||
734 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
735 vp->v_object->resident_page_count > trigger)) {
741 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
743 goto next_iter_mntunlocked;
747 * v_usecount may have been bumped after VOP_LOCK() dropped
748 * the vnode interlock and before it was locked again.
750 * It is not necessary to recheck VI_DOOMED because it can
751 * only be set by another thread that holds both the vnode
752 * lock and vnode interlock. If another thread has the
753 * vnode lock before we get to VOP_LOCK() and obtains the
754 * vnode interlock after VOP_LOCK() drops the vnode
755 * interlock, the other thread will be unable to drop the
756 * vnode lock before our VOP_LOCK() call fails.
758 if (vp->v_usecount ||
759 (!vlru_allow_cache_src &&
760 !LIST_EMPTY(&(vp)->v_cache_src)) ||
761 (vp->v_object != NULL &&
762 vp->v_object->resident_page_count > trigger)) {
763 VOP_UNLOCK(vp, LK_INTERLOCK);
765 goto next_iter_mntunlocked;
767 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
768 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
773 next_iter_mntunlocked:
782 kern_yield(PRI_USER);
787 vn_finished_write(mp);
792 * Attempt to keep the free list at wantfreevnodes length.
795 vnlru_free(int count)
799 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
800 for (; count > 0; count--) {
801 vp = TAILQ_FIRST(&vnode_free_list);
803 * The list can be modified while the free_list_mtx
804 * has been dropped and vp could be NULL here.
808 VNASSERT(vp->v_op != NULL, vp,
809 ("vnlru_free: vnode already reclaimed."));
810 KASSERT((vp->v_iflag & VI_FREE) != 0,
811 ("Removing vnode not on freelist"));
812 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
813 ("Mangling active vnode"));
814 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
816 * Don't recycle if we can't get the interlock.
818 if (!VI_TRYLOCK(vp)) {
819 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
822 VNASSERT(VCANRECYCLE(vp), vp,
823 ("vp inconsistent on freelist"));
825 vp->v_iflag &= ~VI_FREE;
827 mtx_unlock(&vnode_free_list_mtx);
831 * If the recycled succeeded this vdrop will actually free
832 * the vnode. If not it will simply place it back on
836 mtx_lock(&vnode_free_list_mtx);
840 * Attempt to recycle vnodes in a context that is always safe to block.
841 * Calling vlrurecycle() from the bowels of filesystem code has some
842 * interesting deadlock problems.
844 static struct proc *vnlruproc;
845 static int vnlruproc_sig;
850 struct mount *mp, *nmp;
852 struct proc *p = vnlruproc;
854 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
858 kproc_suspend_check(p);
859 mtx_lock(&vnode_free_list_mtx);
860 if (freevnodes > wantfreevnodes)
861 vnlru_free(freevnodes - wantfreevnodes);
862 if (numvnodes <= desiredvnodes * 9 / 10) {
864 wakeup(&vnlruproc_sig);
865 msleep(vnlruproc, &vnode_free_list_mtx,
866 PVFS|PDROP, "vlruwt", hz);
869 mtx_unlock(&vnode_free_list_mtx);
871 mtx_lock(&mountlist_mtx);
872 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
873 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
874 nmp = TAILQ_NEXT(mp, mnt_list);
877 done += vlrureclaim(mp);
878 mtx_lock(&mountlist_mtx);
879 nmp = TAILQ_NEXT(mp, mnt_list);
882 mtx_unlock(&mountlist_mtx);
885 /* These messages are temporary debugging aids */
886 if (vnlru_nowhere < 5)
887 printf("vnlru process getting nowhere..\n");
888 else if (vnlru_nowhere == 5)
889 printf("vnlru process messages stopped.\n");
892 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
894 kern_yield(PRI_USER);
898 static struct kproc_desc vnlru_kp = {
903 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
907 * Routines having to do with the management of the vnode table.
911 * Try to recycle a freed vnode. We abort if anyone picks up a reference
912 * before we actually vgone(). This function must be called with the vnode
913 * held to prevent the vnode from being returned to the free list midway
917 vtryrecycle(struct vnode *vp)
921 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
922 VNASSERT(vp->v_holdcnt, vp,
923 ("vtryrecycle: Recycling vp %p without a reference.", vp));
925 * This vnode may found and locked via some other list, if so we
926 * can't recycle it yet.
928 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
930 "%s: impossible to recycle, vp %p lock is already held",
932 return (EWOULDBLOCK);
935 * Don't recycle if its filesystem is being suspended.
937 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
940 "%s: impossible to recycle, cannot start the write for %p",
945 * If we got this far, we need to acquire the interlock and see if
946 * anyone picked up this vnode from another list. If not, we will
947 * mark it with DOOMED via vgonel() so that anyone who does find it
951 if (vp->v_usecount) {
952 VOP_UNLOCK(vp, LK_INTERLOCK);
953 vn_finished_write(vnmp);
955 "%s: impossible to recycle, %p is already referenced",
959 if ((vp->v_iflag & VI_DOOMED) == 0)
961 VOP_UNLOCK(vp, LK_INTERLOCK);
962 vn_finished_write(vnmp);
967 * Wait for available vnodes.
970 getnewvnode_wait(int suspended)
973 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
974 if (numvnodes > desiredvnodes) {
977 * File system is beeing suspended, we cannot risk a
978 * deadlock here, so allocate new vnode anyway.
980 if (freevnodes > wantfreevnodes)
981 vnlru_free(freevnodes - wantfreevnodes);
984 if (vnlruproc_sig == 0) {
985 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
988 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
991 return (numvnodes > desiredvnodes ? ENFILE : 0);
995 getnewvnode_reserve(u_int count)
1000 /* First try to be quick and racy. */
1001 if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
1002 td->td_vp_reserv += count;
1005 atomic_subtract_long(&numvnodes, count);
1007 mtx_lock(&vnode_free_list_mtx);
1009 if (getnewvnode_wait(0) == 0) {
1012 atomic_add_long(&numvnodes, 1);
1015 mtx_unlock(&vnode_free_list_mtx);
1019 getnewvnode_drop_reserve(void)
1024 atomic_subtract_long(&numvnodes, td->td_vp_reserv);
1025 td->td_vp_reserv = 0;
1029 * Return the next vnode from the free list.
1032 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
1040 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
1043 if (td->td_vp_reserv > 0) {
1044 td->td_vp_reserv -= 1;
1047 mtx_lock(&vnode_free_list_mtx);
1049 * Lend our context to reclaim vnodes if they've exceeded the max.
1051 if (freevnodes > wantfreevnodes)
1053 error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1055 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1057 mtx_unlock(&vnode_free_list_mtx);
1061 atomic_add_long(&numvnodes, 1);
1062 mtx_unlock(&vnode_free_list_mtx);
1064 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
1068 vp->v_vnlock = &vp->v_lock;
1069 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
1071 * By default, don't allow shared locks unless filesystems
1074 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE);
1076 * Initialize bufobj.
1079 bo->__bo_vnode = vp;
1080 rw_init(BO_LOCKPTR(bo), "bufobj interlock");
1081 bo->bo_ops = &buf_ops_bio;
1082 bo->bo_private = vp;
1083 TAILQ_INIT(&bo->bo_clean.bv_hd);
1084 TAILQ_INIT(&bo->bo_dirty.bv_hd);
1086 * Initialize namecache.
1088 LIST_INIT(&vp->v_cache_src);
1089 TAILQ_INIT(&vp->v_cache_dst);
1091 * Finalize various vnode identity bits.
1096 v_incr_usecount(vp);
1100 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1101 mac_vnode_associate_singlelabel(mp, vp);
1102 else if (mp == NULL && vops != &dead_vnodeops)
1103 printf("NULL mp in getnewvnode()\n");
1106 bo->bo_bsize = mp->mnt_stat.f_iosize;
1107 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1108 vp->v_vflag |= VV_NOKNOTE;
1110 rangelock_init(&vp->v_rl);
1113 * For the filesystems which do not use vfs_hash_insert(),
1114 * still initialize v_hash to have vfs_hash_index() useful.
1115 * E.g., nullfs uses vfs_hash_index() on the lower vnode for
1118 vp->v_hash = (uintptr_t)vp >> vnsz2log;
1125 * Delete from old mount point vnode list, if on one.
1128 delmntque(struct vnode *vp)
1138 KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1139 ("Active vnode list size %d > Vnode list size %d",
1140 mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1141 active = vp->v_iflag & VI_ACTIVE;
1142 vp->v_iflag &= ~VI_ACTIVE;
1144 mtx_lock(&vnode_free_list_mtx);
1145 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1146 mp->mnt_activevnodelistsize--;
1147 mtx_unlock(&vnode_free_list_mtx);
1151 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1152 ("bad mount point vnode list size"));
1153 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1154 mp->mnt_nvnodelistsize--;
1160 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1164 vp->v_op = &dead_vnodeops;
1170 * Insert into list of vnodes for the new mount point, if available.
1173 insmntque1(struct vnode *vp, struct mount *mp,
1174 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1177 KASSERT(vp->v_mount == NULL,
1178 ("insmntque: vnode already on per mount vnode list"));
1179 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1180 ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
1183 * We acquire the vnode interlock early to ensure that the
1184 * vnode cannot be recycled by another process releasing a
1185 * holdcnt on it before we get it on both the vnode list
1186 * and the active vnode list. The mount mutex protects only
1187 * manipulation of the vnode list and the vnode freelist
1188 * mutex protects only manipulation of the active vnode list.
1189 * Hence the need to hold the vnode interlock throughout.
1193 if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1194 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1195 mp->mnt_nvnodelistsize == 0)) &&
1196 (vp->v_vflag & VV_FORCEINSMQ) == 0) {
1205 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1206 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1207 ("neg mount point vnode list size"));
1208 mp->mnt_nvnodelistsize++;
1209 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1210 ("Activating already active vnode"));
1211 vp->v_iflag |= VI_ACTIVE;
1212 mtx_lock(&vnode_free_list_mtx);
1213 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1214 mp->mnt_activevnodelistsize++;
1215 mtx_unlock(&vnode_free_list_mtx);
1222 insmntque(struct vnode *vp, struct mount *mp)
1225 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1229 * Flush out and invalidate all buffers associated with a bufobj
1230 * Called with the underlying object locked.
1233 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1238 if (flags & V_SAVE) {
1239 error = bufobj_wwait(bo, slpflag, slptimeo);
1244 if (bo->bo_dirty.bv_cnt > 0) {
1246 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1249 * XXX We could save a lock/unlock if this was only
1250 * enabled under INVARIANTS
1253 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1254 panic("vinvalbuf: dirty bufs");
1258 * If you alter this loop please notice that interlock is dropped and
1259 * reacquired in flushbuflist. Special care is needed to ensure that
1260 * no race conditions occur from this.
1263 error = flushbuflist(&bo->bo_clean,
1264 flags, bo, slpflag, slptimeo);
1265 if (error == 0 && !(flags & V_CLEANONLY))
1266 error = flushbuflist(&bo->bo_dirty,
1267 flags, bo, slpflag, slptimeo);
1268 if (error != 0 && error != EAGAIN) {
1272 } while (error != 0);
1275 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1276 * have write I/O in-progress but if there is a VM object then the
1277 * VM object can also have read-I/O in-progress.
1280 bufobj_wwait(bo, 0, 0);
1282 if (bo->bo_object != NULL) {
1283 VM_OBJECT_WLOCK(bo->bo_object);
1284 vm_object_pip_wait(bo->bo_object, "bovlbx");
1285 VM_OBJECT_WUNLOCK(bo->bo_object);
1288 } while (bo->bo_numoutput > 0);
1292 * Destroy the copy in the VM cache, too.
1294 if (bo->bo_object != NULL &&
1295 (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1296 VM_OBJECT_WLOCK(bo->bo_object);
1297 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1298 OBJPR_CLEANONLY : 0);
1299 VM_OBJECT_WUNLOCK(bo->bo_object);
1304 if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1305 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1306 panic("vinvalbuf: flush failed");
1313 * Flush out and invalidate all buffers associated with a vnode.
1314 * Called with the underlying object locked.
1317 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1320 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1321 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1322 if (vp->v_object != NULL && vp->v_object->handle != vp)
1324 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1328 * Flush out buffers on the specified list.
1332 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1335 struct buf *bp, *nbp;
1340 ASSERT_BO_WLOCKED(bo);
1343 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1344 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1345 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1351 lblkno = nbp->b_lblkno;
1352 xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
1355 error = BUF_TIMELOCK(bp,
1356 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo),
1357 "flushbuf", slpflag, slptimeo);
1360 return (error != ENOLCK ? error : EAGAIN);
1362 KASSERT(bp->b_bufobj == bo,
1363 ("bp %p wrong b_bufobj %p should be %p",
1364 bp, bp->b_bufobj, bo));
1365 if (bp->b_bufobj != bo) { /* XXX: necessary ? */
1371 * XXX Since there are no node locks for NFS, I
1372 * believe there is a slight chance that a delayed
1373 * write will occur while sleeping just above, so
1376 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1379 bp->b_flags |= B_ASYNC;
1382 return (EAGAIN); /* XXX: why not loop ? */
1385 bp->b_flags |= (B_INVAL | B_RELBUF);
1386 bp->b_flags &= ~B_ASYNC;
1390 (nbp->b_bufobj != bo ||
1391 nbp->b_lblkno != lblkno ||
1392 (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags))
1393 break; /* nbp invalid */
1399 * Truncate a file's buffer and pages to a specified length. This
1400 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1404 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1406 struct buf *bp, *nbp;
1411 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1412 vp, cred, blksize, (uintmax_t)length);
1415 * Round up to the *next* lbn.
1417 trunclbn = (length + blksize - 1) / blksize;
1419 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1426 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1427 if (bp->b_lblkno < trunclbn)
1430 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1431 BO_LOCKPTR(bo)) == ENOLCK)
1435 bp->b_flags |= (B_INVAL | B_RELBUF);
1436 bp->b_flags &= ~B_ASYNC;
1442 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1443 (nbp->b_vp != vp) ||
1444 (nbp->b_flags & B_DELWRI))) {
1450 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1451 if (bp->b_lblkno < trunclbn)
1454 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1455 BO_LOCKPTR(bo)) == ENOLCK)
1458 bp->b_flags |= (B_INVAL | B_RELBUF);
1459 bp->b_flags &= ~B_ASYNC;
1465 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1466 (nbp->b_vp != vp) ||
1467 (nbp->b_flags & B_DELWRI) == 0)) {
1476 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1477 if (bp->b_lblkno > 0)
1480 * Since we hold the vnode lock this should only
1481 * fail if we're racing with the buf daemon.
1484 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1485 BO_LOCKPTR(bo)) == ENOLCK) {
1488 VNASSERT((bp->b_flags & B_DELWRI), vp,
1489 ("buf(%p) on dirty queue without DELWRI", bp));
1498 bufobj_wwait(bo, 0, 0);
1500 vnode_pager_setsize(vp, length);
1506 buf_vlist_remove(struct buf *bp)
1510 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1511 ASSERT_BO_WLOCKED(bp->b_bufobj);
1512 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1513 (BX_VNDIRTY|BX_VNCLEAN),
1514 ("buf_vlist_remove: Buf %p is on two lists", bp));
1515 if (bp->b_xflags & BX_VNDIRTY)
1516 bv = &bp->b_bufobj->bo_dirty;
1518 bv = &bp->b_bufobj->bo_clean;
1519 BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
1520 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1522 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1526 * Add the buffer to the sorted clean or dirty block list.
1528 * NOTE: xflags is passed as a constant, optimizing this inline function!
1531 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1537 ASSERT_BO_WLOCKED(bo);
1538 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1539 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1540 bp->b_xflags |= xflags;
1541 if (xflags & BX_VNDIRTY)
1547 * Keep the list ordered. Optimize empty list insertion. Assume
1548 * we tend to grow at the tail so lookup_le should usually be cheaper
1551 if (bv->bv_cnt == 0 ||
1552 bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
1553 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1554 else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
1555 TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
1557 TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
1558 error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
1560 panic("buf_vlist_add: Preallocated nodes insufficient.");
1565 * Lookup a buffer using the splay tree. Note that we specifically avoid
1566 * shadow buffers used in background bitmap writes.
1568 * This code isn't quite efficient as it could be because we are maintaining
1569 * two sorted lists and do not know which list the block resides in.
1571 * During a "make buildworld" the desired buffer is found at one of
1572 * the roots more than 60% of the time. Thus, checking both roots
1573 * before performing either splay eliminates unnecessary splays on the
1574 * first tree splayed.
1577 gbincore(struct bufobj *bo, daddr_t lblkno)
1581 ASSERT_BO_LOCKED(bo);
1582 bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
1585 return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
1589 * Associate a buffer with a vnode.
1592 bgetvp(struct vnode *vp, struct buf *bp)
1597 ASSERT_BO_WLOCKED(bo);
1598 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1600 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1601 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1602 ("bgetvp: bp already attached! %p", bp));
1608 * Insert onto list for new vnode.
1610 buf_vlist_add(bp, bo, BX_VNCLEAN);
1614 * Disassociate a buffer from a vnode.
1617 brelvp(struct buf *bp)
1622 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1623 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1626 * Delete from old vnode list, if on one.
1628 vp = bp->b_vp; /* XXX */
1631 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1632 buf_vlist_remove(bp);
1634 panic("brelvp: Buffer %p not on queue.", bp);
1635 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1636 bo->bo_flag &= ~BO_ONWORKLST;
1637 mtx_lock(&sync_mtx);
1638 LIST_REMOVE(bo, bo_synclist);
1639 syncer_worklist_len--;
1640 mtx_unlock(&sync_mtx);
1643 bp->b_bufobj = NULL;
1649 * Add an item to the syncer work queue.
1652 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1656 ASSERT_BO_WLOCKED(bo);
1658 mtx_lock(&sync_mtx);
1659 if (bo->bo_flag & BO_ONWORKLST)
1660 LIST_REMOVE(bo, bo_synclist);
1662 bo->bo_flag |= BO_ONWORKLST;
1663 syncer_worklist_len++;
1666 if (delay > syncer_maxdelay - 2)
1667 delay = syncer_maxdelay - 2;
1668 slot = (syncer_delayno + delay) & syncer_mask;
1670 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1671 mtx_unlock(&sync_mtx);
1675 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1679 mtx_lock(&sync_mtx);
1680 len = syncer_worklist_len - sync_vnode_count;
1681 mtx_unlock(&sync_mtx);
1682 error = SYSCTL_OUT(req, &len, sizeof(len));
1686 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1687 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1689 static struct proc *updateproc;
1690 static void sched_sync(void);
1691 static struct kproc_desc up_kp = {
1696 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
1699 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
1704 *bo = LIST_FIRST(slp);
1707 vp = (*bo)->__bo_vnode; /* XXX */
1708 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
1711 * We use vhold in case the vnode does not
1712 * successfully sync. vhold prevents the vnode from
1713 * going away when we unlock the sync_mtx so that
1714 * we can acquire the vnode interlock.
1717 mtx_unlock(&sync_mtx);
1719 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
1721 mtx_lock(&sync_mtx);
1722 return (*bo == LIST_FIRST(slp));
1724 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1725 (void) VOP_FSYNC(vp, MNT_LAZY, td);
1727 vn_finished_write(mp);
1729 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
1731 * Put us back on the worklist. The worklist
1732 * routine will remove us from our current
1733 * position and then add us back in at a later
1736 vn_syncer_add_to_worklist(*bo, syncdelay);
1740 mtx_lock(&sync_mtx);
1745 * System filesystem synchronizer daemon.
1750 struct synclist *next, *slp;
1753 struct thread *td = curthread;
1755 int net_worklist_len;
1756 int syncer_final_iter;
1761 syncer_final_iter = 0;
1763 syncer_state = SYNCER_RUNNING;
1764 starttime = time_uptime;
1765 td->td_pflags |= TDP_NORUNNINGBUF;
1767 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
1770 mtx_lock(&sync_mtx);
1772 if (syncer_state == SYNCER_FINAL_DELAY &&
1773 syncer_final_iter == 0) {
1774 mtx_unlock(&sync_mtx);
1775 kproc_suspend_check(td->td_proc);
1776 mtx_lock(&sync_mtx);
1778 net_worklist_len = syncer_worklist_len - sync_vnode_count;
1779 if (syncer_state != SYNCER_RUNNING &&
1780 starttime != time_uptime) {
1782 printf("\nSyncing disks, vnodes remaining...");
1785 printf("%d ", net_worklist_len);
1787 starttime = time_uptime;
1790 * Push files whose dirty time has expired. Be careful
1791 * of interrupt race on slp queue.
1793 * Skip over empty worklist slots when shutting down.
1796 slp = &syncer_workitem_pending[syncer_delayno];
1797 syncer_delayno += 1;
1798 if (syncer_delayno == syncer_maxdelay)
1800 next = &syncer_workitem_pending[syncer_delayno];
1802 * If the worklist has wrapped since the
1803 * it was emptied of all but syncer vnodes,
1804 * switch to the FINAL_DELAY state and run
1805 * for one more second.
1807 if (syncer_state == SYNCER_SHUTTING_DOWN &&
1808 net_worklist_len == 0 &&
1809 last_work_seen == syncer_delayno) {
1810 syncer_state = SYNCER_FINAL_DELAY;
1811 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
1813 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
1814 syncer_worklist_len > 0);
1817 * Keep track of the last time there was anything
1818 * on the worklist other than syncer vnodes.
1819 * Return to the SHUTTING_DOWN state if any
1822 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
1823 last_work_seen = syncer_delayno;
1824 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
1825 syncer_state = SYNCER_SHUTTING_DOWN;
1826 while (!LIST_EMPTY(slp)) {
1827 error = sync_vnode(slp, &bo, td);
1829 LIST_REMOVE(bo, bo_synclist);
1830 LIST_INSERT_HEAD(next, bo, bo_synclist);
1834 if (first_printf == 0)
1835 wdog_kern_pat(WD_LASTVAL);
1838 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
1839 syncer_final_iter--;
1841 * The variable rushjob allows the kernel to speed up the
1842 * processing of the filesystem syncer process. A rushjob
1843 * value of N tells the filesystem syncer to process the next
1844 * N seconds worth of work on its queue ASAP. Currently rushjob
1845 * is used by the soft update code to speed up the filesystem
1846 * syncer process when the incore state is getting so far
1847 * ahead of the disk that the kernel memory pool is being
1848 * threatened with exhaustion.
1855 * Just sleep for a short period of time between
1856 * iterations when shutting down to allow some I/O
1859 * If it has taken us less than a second to process the
1860 * current work, then wait. Otherwise start right over
1861 * again. We can still lose time if any single round
1862 * takes more than two seconds, but it does not really
1863 * matter as we are just trying to generally pace the
1864 * filesystem activity.
1866 if (syncer_state != SYNCER_RUNNING ||
1867 time_uptime == starttime) {
1869 sched_prio(td, PPAUSE);
1872 if (syncer_state != SYNCER_RUNNING)
1873 cv_timedwait(&sync_wakeup, &sync_mtx,
1874 hz / SYNCER_SHUTDOWN_SPEEDUP);
1875 else if (time_uptime == starttime)
1876 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
1881 * Request the syncer daemon to speed up its work.
1882 * We never push it to speed up more than half of its
1883 * normal turn time, otherwise it could take over the cpu.
1886 speedup_syncer(void)
1890 mtx_lock(&sync_mtx);
1891 if (rushjob < syncdelay / 2) {
1893 stat_rush_requests += 1;
1896 mtx_unlock(&sync_mtx);
1897 cv_broadcast(&sync_wakeup);
1902 * Tell the syncer to speed up its work and run though its work
1903 * list several times, then tell it to shut down.
1906 syncer_shutdown(void *arg, int howto)
1909 if (howto & RB_NOSYNC)
1911 mtx_lock(&sync_mtx);
1912 syncer_state = SYNCER_SHUTTING_DOWN;
1914 mtx_unlock(&sync_mtx);
1915 cv_broadcast(&sync_wakeup);
1916 kproc_shutdown(arg, howto);
1920 * Reassign a buffer from one vnode to another.
1921 * Used to assign file specific control information
1922 * (indirect blocks) to the vnode to which they belong.
1925 reassignbuf(struct buf *bp)
1938 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
1939 bp, bp->b_vp, bp->b_flags);
1941 * B_PAGING flagged buffers cannot be reassigned because their vp
1942 * is not fully linked in.
1944 if (bp->b_flags & B_PAGING)
1945 panic("cannot reassign paging buffer");
1948 * Delete from old vnode list, if on one.
1951 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1952 buf_vlist_remove(bp);
1954 panic("reassignbuf: Buffer %p not on queue.", bp);
1956 * If dirty, put on list of dirty buffers; otherwise insert onto list
1959 if (bp->b_flags & B_DELWRI) {
1960 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
1961 switch (vp->v_type) {
1971 vn_syncer_add_to_worklist(bo, delay);
1973 buf_vlist_add(bp, bo, BX_VNDIRTY);
1975 buf_vlist_add(bp, bo, BX_VNCLEAN);
1977 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1978 mtx_lock(&sync_mtx);
1979 LIST_REMOVE(bo, bo_synclist);
1980 syncer_worklist_len--;
1981 mtx_unlock(&sync_mtx);
1982 bo->bo_flag &= ~BO_ONWORKLST;
1987 bp = TAILQ_FIRST(&bv->bv_hd);
1988 KASSERT(bp == NULL || bp->b_bufobj == bo,
1989 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1990 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1991 KASSERT(bp == NULL || bp->b_bufobj == bo,
1992 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1994 bp = TAILQ_FIRST(&bv->bv_hd);
1995 KASSERT(bp == NULL || bp->b_bufobj == bo,
1996 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
1997 bp = TAILQ_LAST(&bv->bv_hd, buflists);
1998 KASSERT(bp == NULL || bp->b_bufobj == bo,
1999 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2005 * Increment the use and hold counts on the vnode, taking care to reference
2006 * the driver's usecount if this is a chardev. The vholdl() will remove
2007 * the vnode from the free list if it is presently free. Requires the
2008 * vnode interlock and returns with it held.
2011 v_incr_usecount(struct vnode *vp)
2014 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2016 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2018 vp->v_rdev->si_usecount++;
2025 * Turn a holdcnt into a use+holdcnt such that only one call to
2026 * v_decr_usecount is needed.
2029 v_upgrade_usecount(struct vnode *vp)
2032 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2034 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2036 vp->v_rdev->si_usecount++;
2042 * Decrement the vnode use and hold count along with the driver's usecount
2043 * if this is a chardev. The vdropl() below releases the vnode interlock
2044 * as it may free the vnode.
2047 v_decr_usecount(struct vnode *vp)
2050 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2051 VNASSERT(vp->v_usecount > 0, vp,
2052 ("v_decr_usecount: negative usecount"));
2053 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2055 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2057 vp->v_rdev->si_usecount--;
2064 * Decrement only the use count and driver use count. This is intended to
2065 * be paired with a follow on vdropl() to release the remaining hold count.
2066 * In this way we may vgone() a vnode with a 0 usecount without risk of
2067 * having it end up on a free list because the hold count is kept above 0.
2070 v_decr_useonly(struct vnode *vp)
2073 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2074 VNASSERT(vp->v_usecount > 0, vp,
2075 ("v_decr_useonly: negative usecount"));
2076 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2078 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2080 vp->v_rdev->si_usecount--;
2086 * Grab a particular vnode from the free list, increment its
2087 * reference count and lock it. VI_DOOMED is set if the vnode
2088 * is being destroyed. Only callers who specify LK_RETRY will
2089 * see doomed vnodes. If inactive processing was delayed in
2090 * vput try to do it here.
2093 vget(struct vnode *vp, int flags, struct thread *td)
2098 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2099 ("vget: invalid lock operation"));
2100 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2102 if ((flags & LK_INTERLOCK) == 0)
2105 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
2107 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2111 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2112 panic("vget: vn_lock failed to return ENOENT\n");
2114 /* Upgrade our holdcnt to a usecount. */
2115 v_upgrade_usecount(vp);
2117 * We don't guarantee that any particular close will
2118 * trigger inactive processing so just make a best effort
2119 * here at preventing a reference to a removed file. If
2120 * we don't succeed no harm is done.
2122 if (vp->v_iflag & VI_OWEINACT) {
2123 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2124 (flags & LK_NOWAIT) == 0)
2126 vp->v_iflag &= ~VI_OWEINACT;
2133 * Increase the reference count of a vnode.
2136 vref(struct vnode *vp)
2139 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2141 v_incr_usecount(vp);
2146 * Return reference count of a vnode.
2148 * The results of this call are only guaranteed when some mechanism other
2149 * than the VI lock is used to stop other processes from gaining references
2150 * to the vnode. This may be the case if the caller holds the only reference.
2151 * This is also useful when stale data is acceptable as race conditions may
2152 * be accounted for by some other means.
2155 vrefcnt(struct vnode *vp)
2160 usecnt = vp->v_usecount;
2166 #define VPUTX_VRELE 1
2167 #define VPUTX_VPUT 2
2168 #define VPUTX_VUNREF 3
2171 vputx(struct vnode *vp, int func)
2175 KASSERT(vp != NULL, ("vputx: null vp"));
2176 if (func == VPUTX_VUNREF)
2177 ASSERT_VOP_LOCKED(vp, "vunref");
2178 else if (func == VPUTX_VPUT)
2179 ASSERT_VOP_LOCKED(vp, "vput");
2181 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2182 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2185 /* Skip this v_writecount check if we're going to panic below. */
2186 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
2187 ("vputx: missed vn_close"));
2190 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
2191 vp->v_usecount == 1)) {
2192 if (func == VPUTX_VPUT)
2194 v_decr_usecount(vp);
2198 if (vp->v_usecount != 1) {
2199 vprint("vputx: negative ref count", vp);
2200 panic("vputx: negative ref cnt");
2202 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2204 * We want to hold the vnode until the inactive finishes to
2205 * prevent vgone() races. We drop the use count here and the
2206 * hold count below when we're done.
2210 * We must call VOP_INACTIVE with the node locked. Mark
2211 * as VI_DOINGINACT to avoid recursion.
2213 vp->v_iflag |= VI_OWEINACT;
2216 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2220 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2221 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2227 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2228 error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
2233 if (vp->v_usecount > 0)
2234 vp->v_iflag &= ~VI_OWEINACT;
2236 if (vp->v_iflag & VI_OWEINACT)
2237 vinactive(vp, curthread);
2238 if (func != VPUTX_VUNREF)
2245 * Vnode put/release.
2246 * If count drops to zero, call inactive routine and return to freelist.
2249 vrele(struct vnode *vp)
2252 vputx(vp, VPUTX_VRELE);
2256 * Release an already locked vnode. This give the same effects as
2257 * unlock+vrele(), but takes less time and avoids releasing and
2258 * re-aquiring the lock (as vrele() acquires the lock internally.)
2261 vput(struct vnode *vp)
2264 vputx(vp, VPUTX_VPUT);
2268 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2271 vunref(struct vnode *vp)
2274 vputx(vp, VPUTX_VUNREF);
2278 * Somebody doesn't want the vnode recycled.
2281 vhold(struct vnode *vp)
2290 * Increase the hold count and activate if this is the first reference.
2293 vholdl(struct vnode *vp)
2297 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2299 if (!VSHOULDBUSY(vp))
2301 ASSERT_VI_LOCKED(vp, "vholdl");
2302 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
2303 VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
2305 * Remove a vnode from the free list, mark it as in use,
2306 * and put it on the active list.
2308 mtx_lock(&vnode_free_list_mtx);
2309 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2311 vp->v_iflag &= ~(VI_FREE|VI_AGE);
2312 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2313 ("Activating already active vnode"));
2314 vp->v_iflag |= VI_ACTIVE;
2316 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2317 mp->mnt_activevnodelistsize++;
2318 mtx_unlock(&vnode_free_list_mtx);
2322 * Note that there is one less who cares about this vnode.
2323 * vdrop() is the opposite of vhold().
2326 vdrop(struct vnode *vp)
2334 * Drop the hold count of the vnode. If this is the last reference to
2335 * the vnode we place it on the free list unless it has been vgone'd
2336 * (marked VI_DOOMED) in which case we will free it.
2339 vdropl(struct vnode *vp)
2345 ASSERT_VI_LOCKED(vp, "vdropl");
2346 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2347 if (vp->v_holdcnt <= 0)
2348 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2350 if (vp->v_holdcnt > 0) {
2354 if ((vp->v_iflag & VI_DOOMED) == 0) {
2356 * Mark a vnode as free: remove it from its active list
2357 * and put it up for recycling on the freelist.
2359 VNASSERT(vp->v_op != NULL, vp,
2360 ("vdropl: vnode already reclaimed."));
2361 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2362 ("vnode already free"));
2363 VNASSERT(VSHOULDFREE(vp), vp,
2364 ("vdropl: freeing when we shouldn't"));
2365 active = vp->v_iflag & VI_ACTIVE;
2366 vp->v_iflag &= ~VI_ACTIVE;
2368 mtx_lock(&vnode_free_list_mtx);
2370 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2372 mp->mnt_activevnodelistsize--;
2374 if (vp->v_iflag & VI_AGE) {
2375 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
2377 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
2380 vp->v_iflag &= ~VI_AGE;
2381 vp->v_iflag |= VI_FREE;
2382 mtx_unlock(&vnode_free_list_mtx);
2387 * The vnode has been marked for destruction, so free it.
2389 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2390 atomic_subtract_long(&numvnodes, 1);
2392 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2393 ("cleaned vnode still on the free list."));
2394 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2395 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2396 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2397 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2398 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2399 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2400 VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
2401 ("clean blk trie not empty"));
2402 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2403 VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
2404 ("dirty blk trie not empty"));
2405 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2406 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2407 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2410 mac_vnode_destroy(vp);
2412 if (vp->v_pollinfo != NULL)
2413 destroy_vpollinfo(vp->v_pollinfo);
2415 /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2418 rangelock_destroy(&vp->v_rl);
2419 lockdestroy(vp->v_vnlock);
2420 mtx_destroy(&vp->v_interlock);
2421 rw_destroy(BO_LOCKPTR(bo));
2422 uma_zfree(vnode_zone, vp);
2426 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2427 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2428 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2429 * failed lock upgrade.
2432 vinactive(struct vnode *vp, struct thread *td)
2434 struct vm_object *obj;
2436 ASSERT_VOP_ELOCKED(vp, "vinactive");
2437 ASSERT_VI_LOCKED(vp, "vinactive");
2438 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2439 ("vinactive: recursed on VI_DOINGINACT"));
2440 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2441 vp->v_iflag |= VI_DOINGINACT;
2442 vp->v_iflag &= ~VI_OWEINACT;
2445 * Before moving off the active list, we must be sure that any
2446 * modified pages are on the vnode's dirty list since these will
2447 * no longer be checked once the vnode is on the inactive list.
2448 * Because the vnode vm object keeps a hold reference on the vnode
2449 * if there is at least one resident non-cached page, the vnode
2450 * cannot leave the active list without the page cleanup done.
2453 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2454 VM_OBJECT_WLOCK(obj);
2455 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2456 VM_OBJECT_WUNLOCK(obj);
2458 VOP_INACTIVE(vp, td);
2460 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2461 ("vinactive: lost VI_DOINGINACT"));
2462 vp->v_iflag &= ~VI_DOINGINACT;
2466 * Remove any vnodes in the vnode table belonging to mount point mp.
2468 * If FORCECLOSE is not specified, there should not be any active ones,
2469 * return error if any are found (nb: this is a user error, not a
2470 * system error). If FORCECLOSE is specified, detach any active vnodes
2473 * If WRITECLOSE is set, only flush out regular file vnodes open for
2476 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2478 * `rootrefs' specifies the base reference count for the root vnode
2479 * of this filesystem. The root vnode is considered busy if its
2480 * v_usecount exceeds this value. On a successful return, vflush(, td)
2481 * will call vrele() on the root vnode exactly rootrefs times.
2482 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2486 static int busyprt = 0; /* print out busy vnodes */
2487 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2491 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2493 struct vnode *vp, *mvp, *rootvp = NULL;
2495 int busy = 0, error;
2497 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2500 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2501 ("vflush: bad args"));
2503 * Get the filesystem root vnode. We can vput() it
2504 * immediately, since with rootrefs > 0, it won't go away.
2506 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2507 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2514 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2516 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2519 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2523 * Skip over a vnodes marked VV_SYSTEM.
2525 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2531 * If WRITECLOSE is set, flush out unlinked but still open
2532 * files (even if open only for reading) and regular file
2533 * vnodes open for writing.
2535 if (flags & WRITECLOSE) {
2536 if (vp->v_object != NULL) {
2537 VM_OBJECT_WLOCK(vp->v_object);
2538 vm_object_page_clean(vp->v_object, 0, 0, 0);
2539 VM_OBJECT_WUNLOCK(vp->v_object);
2541 error = VOP_FSYNC(vp, MNT_WAIT, td);
2545 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2548 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2551 if ((vp->v_type == VNON ||
2552 (error == 0 && vattr.va_nlink > 0)) &&
2553 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2561 * With v_usecount == 0, all we need to do is clear out the
2562 * vnode data structures and we are done.
2564 * If FORCECLOSE is set, forcibly close the vnode.
2566 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2567 VNASSERT(vp->v_usecount == 0 ||
2568 (vp->v_type != VCHR && vp->v_type != VBLK), vp,
2569 ("device VNODE %p is FORCECLOSED", vp));
2575 vprint("vflush: busy vnode", vp);
2581 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
2583 * If just the root vnode is busy, and if its refcount
2584 * is equal to `rootrefs', then go ahead and kill it.
2587 KASSERT(busy > 0, ("vflush: not busy"));
2588 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
2589 ("vflush: usecount %d < rootrefs %d",
2590 rootvp->v_usecount, rootrefs));
2591 if (busy == 1 && rootvp->v_usecount == rootrefs) {
2592 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
2594 VOP_UNLOCK(rootvp, 0);
2600 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
2604 for (; rootrefs > 0; rootrefs--)
2610 * Recycle an unused vnode to the front of the free list.
2613 vrecycle(struct vnode *vp)
2617 ASSERT_VOP_ELOCKED(vp, "vrecycle");
2618 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2621 if (vp->v_usecount == 0) {
2630 * Eliminate all activity associated with a vnode
2631 * in preparation for reuse.
2634 vgone(struct vnode *vp)
2642 notify_lowervp_vfs_dummy(struct mount *mp __unused,
2643 struct vnode *lowervp __unused)
2648 * Notify upper mounts about reclaimed or unlinked vnode.
2651 vfs_notify_upper(struct vnode *vp, int event)
2653 static struct vfsops vgonel_vfsops = {
2654 .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
2655 .vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
2657 struct mount *mp, *ump, *mmp;
2664 if (TAILQ_EMPTY(&mp->mnt_uppers))
2667 mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
2668 mmp->mnt_op = &vgonel_vfsops;
2669 mmp->mnt_kern_flag |= MNTK_MARKER;
2671 mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
2672 for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
2673 if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
2674 ump = TAILQ_NEXT(ump, mnt_upper_link);
2677 TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
2680 case VFS_NOTIFY_UPPER_RECLAIM:
2681 VFS_RECLAIM_LOWERVP(ump, vp);
2683 case VFS_NOTIFY_UPPER_UNLINK:
2684 VFS_UNLINK_LOWERVP(ump, vp);
2687 KASSERT(0, ("invalid event %d", event));
2691 ump = TAILQ_NEXT(mmp, mnt_upper_link);
2692 TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
2695 mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
2696 if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
2697 mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
2698 wakeup(&mp->mnt_uppers);
2705 * vgone, with the vp interlock held.
2708 vgonel(struct vnode *vp)
2715 ASSERT_VOP_ELOCKED(vp, "vgonel");
2716 ASSERT_VI_LOCKED(vp, "vgonel");
2717 VNASSERT(vp->v_holdcnt, vp,
2718 ("vgonel: vp %p has no reference.", vp));
2719 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2723 * Don't vgonel if we're already doomed.
2725 if (vp->v_iflag & VI_DOOMED)
2727 vp->v_iflag |= VI_DOOMED;
2730 * Check to see if the vnode is in use. If so, we have to call
2731 * VOP_CLOSE() and VOP_INACTIVE().
2733 active = vp->v_usecount;
2734 oweinact = (vp->v_iflag & VI_OWEINACT);
2736 vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
2739 * Clean out any buffers associated with the vnode.
2740 * If the flush fails, just toss the buffers.
2743 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
2744 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
2745 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
2746 vinvalbuf(vp, 0, 0, 0);
2749 * If purging an active vnode, it must be closed and
2750 * deactivated before being reclaimed.
2753 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
2754 if (oweinact || active) {
2756 if ((vp->v_iflag & VI_DOINGINACT) == 0)
2760 if (vp->v_type == VSOCK)
2761 vfs_unp_reclaim(vp);
2763 * Reclaim the vnode.
2765 if (VOP_RECLAIM(vp, td))
2766 panic("vgone: cannot reclaim");
2768 vn_finished_secondary_write(mp);
2769 VNASSERT(vp->v_object == NULL, vp,
2770 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
2772 * Clear the advisory locks and wake up waiting threads.
2774 (void)VOP_ADVLOCKPURGE(vp);
2776 * Delete from old mount point vnode list.
2781 * Done with purge, reset to the standard lock and invalidate
2785 vp->v_vnlock = &vp->v_lock;
2786 vp->v_op = &dead_vnodeops;
2792 * Calculate the total number of references to a special device.
2795 vcount(struct vnode *vp)
2800 count = vp->v_rdev->si_usecount;
2806 * Same as above, but using the struct cdev *as argument
2809 count_dev(struct cdev *dev)
2814 count = dev->si_usecount;
2820 * Print out a description of a vnode.
2822 static char *typename[] =
2823 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
2827 vn_printf(struct vnode *vp, const char *fmt, ...)
2830 char buf[256], buf2[16];
2836 printf("%p: ", (void *)vp);
2837 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
2838 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
2839 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
2842 if (vp->v_vflag & VV_ROOT)
2843 strlcat(buf, "|VV_ROOT", sizeof(buf));
2844 if (vp->v_vflag & VV_ISTTY)
2845 strlcat(buf, "|VV_ISTTY", sizeof(buf));
2846 if (vp->v_vflag & VV_NOSYNC)
2847 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
2848 if (vp->v_vflag & VV_ETERNALDEV)
2849 strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
2850 if (vp->v_vflag & VV_CACHEDLABEL)
2851 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
2852 if (vp->v_vflag & VV_TEXT)
2853 strlcat(buf, "|VV_TEXT", sizeof(buf));
2854 if (vp->v_vflag & VV_COPYONWRITE)
2855 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
2856 if (vp->v_vflag & VV_SYSTEM)
2857 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
2858 if (vp->v_vflag & VV_PROCDEP)
2859 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
2860 if (vp->v_vflag & VV_NOKNOTE)
2861 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
2862 if (vp->v_vflag & VV_DELETED)
2863 strlcat(buf, "|VV_DELETED", sizeof(buf));
2864 if (vp->v_vflag & VV_MD)
2865 strlcat(buf, "|VV_MD", sizeof(buf));
2866 if (vp->v_vflag & VV_FORCEINSMQ)
2867 strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
2868 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
2869 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
2870 VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
2872 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
2873 strlcat(buf, buf2, sizeof(buf));
2875 if (vp->v_iflag & VI_MOUNT)
2876 strlcat(buf, "|VI_MOUNT", sizeof(buf));
2877 if (vp->v_iflag & VI_AGE)
2878 strlcat(buf, "|VI_AGE", sizeof(buf));
2879 if (vp->v_iflag & VI_DOOMED)
2880 strlcat(buf, "|VI_DOOMED", sizeof(buf));
2881 if (vp->v_iflag & VI_FREE)
2882 strlcat(buf, "|VI_FREE", sizeof(buf));
2883 if (vp->v_iflag & VI_ACTIVE)
2884 strlcat(buf, "|VI_ACTIVE", sizeof(buf));
2885 if (vp->v_iflag & VI_DOINGINACT)
2886 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
2887 if (vp->v_iflag & VI_OWEINACT)
2888 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
2889 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
2890 VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
2892 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
2893 strlcat(buf, buf2, sizeof(buf));
2895 printf(" flags (%s)\n", buf + 1);
2896 if (mtx_owned(VI_MTX(vp)))
2897 printf(" VI_LOCKed");
2898 if (vp->v_object != NULL)
2899 printf(" v_object %p ref %d pages %d "
2900 "cleanbuf %d dirtybuf %d\n",
2901 vp->v_object, vp->v_object->ref_count,
2902 vp->v_object->resident_page_count,
2903 vp->v_bufobj.bo_dirty.bv_cnt,
2904 vp->v_bufobj.bo_clean.bv_cnt);
2906 lockmgr_printinfo(vp->v_vnlock);
2907 if (vp->v_data != NULL)
2913 * List all of the locked vnodes in the system.
2914 * Called when debugging the kernel.
2916 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
2922 * Note: because this is DDB, we can't obey the locking semantics
2923 * for these structures, which means we could catch an inconsistent
2924 * state and dereference a nasty pointer. Not much to be done
2927 db_printf("Locked vnodes\n");
2928 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2929 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
2930 if (vp->v_type != VMARKER && VOP_ISLOCKED(vp))
2937 * Show details about the given vnode.
2939 DB_SHOW_COMMAND(vnode, db_show_vnode)
2945 vp = (struct vnode *)addr;
2946 vn_printf(vp, "vnode ");
2950 * Show details about the given mount point.
2952 DB_SHOW_COMMAND(mount, db_show_mount)
2963 /* No address given, print short info about all mount points. */
2964 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2965 db_printf("%p %s on %s (%s)\n", mp,
2966 mp->mnt_stat.f_mntfromname,
2967 mp->mnt_stat.f_mntonname,
2968 mp->mnt_stat.f_fstypename);
2972 db_printf("\nMore info: show mount <addr>\n");
2976 mp = (struct mount *)addr;
2977 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
2978 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
2981 mflags = mp->mnt_flag;
2982 #define MNT_FLAG(flag) do { \
2983 if (mflags & (flag)) { \
2984 if (buf[0] != '\0') \
2985 strlcat(buf, ", ", sizeof(buf)); \
2986 strlcat(buf, (#flag) + 4, sizeof(buf)); \
2987 mflags &= ~(flag); \
2990 MNT_FLAG(MNT_RDONLY);
2991 MNT_FLAG(MNT_SYNCHRONOUS);
2992 MNT_FLAG(MNT_NOEXEC);
2993 MNT_FLAG(MNT_NOSUID);
2994 MNT_FLAG(MNT_NFS4ACLS);
2995 MNT_FLAG(MNT_UNION);
2996 MNT_FLAG(MNT_ASYNC);
2997 MNT_FLAG(MNT_SUIDDIR);
2998 MNT_FLAG(MNT_SOFTDEP);
2999 MNT_FLAG(MNT_NOSYMFOLLOW);
3000 MNT_FLAG(MNT_GJOURNAL);
3001 MNT_FLAG(MNT_MULTILABEL);
3003 MNT_FLAG(MNT_NOATIME);
3004 MNT_FLAG(MNT_NOCLUSTERR);
3005 MNT_FLAG(MNT_NOCLUSTERW);
3007 MNT_FLAG(MNT_EXRDONLY);
3008 MNT_FLAG(MNT_EXPORTED);
3009 MNT_FLAG(MNT_DEFEXPORTED);
3010 MNT_FLAG(MNT_EXPORTANON);
3011 MNT_FLAG(MNT_EXKERB);
3012 MNT_FLAG(MNT_EXPUBLIC);
3013 MNT_FLAG(MNT_LOCAL);
3014 MNT_FLAG(MNT_QUOTA);
3015 MNT_FLAG(MNT_ROOTFS);
3017 MNT_FLAG(MNT_IGNORE);
3018 MNT_FLAG(MNT_UPDATE);
3019 MNT_FLAG(MNT_DELEXPORT);
3020 MNT_FLAG(MNT_RELOAD);
3021 MNT_FLAG(MNT_FORCE);
3022 MNT_FLAG(MNT_SNAPSHOT);
3023 MNT_FLAG(MNT_BYFSID);
3027 strlcat(buf, ", ", sizeof(buf));
3028 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3029 "0x%016jx", mflags);
3031 db_printf(" mnt_flag = %s\n", buf);
3034 flags = mp->mnt_kern_flag;
3035 #define MNT_KERN_FLAG(flag) do { \
3036 if (flags & (flag)) { \
3037 if (buf[0] != '\0') \
3038 strlcat(buf, ", ", sizeof(buf)); \
3039 strlcat(buf, (#flag) + 5, sizeof(buf)); \
3043 MNT_KERN_FLAG(MNTK_UNMOUNTF);
3044 MNT_KERN_FLAG(MNTK_ASYNC);
3045 MNT_KERN_FLAG(MNTK_SOFTDEP);
3046 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3047 MNT_KERN_FLAG(MNTK_DRAINING);
3048 MNT_KERN_FLAG(MNTK_REFEXPIRE);
3049 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3050 MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3051 MNT_KERN_FLAG(MNTK_NO_IOPF);
3052 MNT_KERN_FLAG(MNTK_VGONE_UPPER);
3053 MNT_KERN_FLAG(MNTK_VGONE_WAITER);
3054 MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
3055 MNT_KERN_FLAG(MNTK_MARKER);
3056 MNT_KERN_FLAG(MNTK_NOASYNC);
3057 MNT_KERN_FLAG(MNTK_UNMOUNT);
3058 MNT_KERN_FLAG(MNTK_MWAIT);
3059 MNT_KERN_FLAG(MNTK_SUSPEND);
3060 MNT_KERN_FLAG(MNTK_SUSPEND2);
3061 MNT_KERN_FLAG(MNTK_SUSPENDED);
3062 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3063 MNT_KERN_FLAG(MNTK_NOKNOTE);
3064 #undef MNT_KERN_FLAG
3067 strlcat(buf, ", ", sizeof(buf));
3068 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3071 db_printf(" mnt_kern_flag = %s\n", buf);
3073 db_printf(" mnt_opt = ");
3074 opt = TAILQ_FIRST(mp->mnt_opt);
3076 db_printf("%s", opt->name);
3077 opt = TAILQ_NEXT(opt, link);
3078 while (opt != NULL) {
3079 db_printf(", %s", opt->name);
3080 opt = TAILQ_NEXT(opt, link);
3086 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
3087 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3088 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3089 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3090 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3091 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3092 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3093 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3094 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3095 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3096 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3097 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3099 db_printf(" mnt_cred = { uid=%u ruid=%u",
3100 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3101 if (jailed(mp->mnt_cred))
3102 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3104 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
3105 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
3106 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3107 db_printf(" mnt_activevnodelistsize = %d\n",
3108 mp->mnt_activevnodelistsize);
3109 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3110 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3111 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3112 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
3113 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3114 db_printf(" mnt_secondary_accwrites = %d\n",
3115 mp->mnt_secondary_accwrites);
3116 db_printf(" mnt_gjprovider = %s\n",
3117 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3119 db_printf("\n\nList of active vnodes\n");
3120 TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3121 if (vp->v_type != VMARKER) {
3122 vn_printf(vp, "vnode ");
3127 db_printf("\n\nList of inactive vnodes\n");
3128 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3129 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3130 vn_printf(vp, "vnode ");
3139 * Fill in a struct xvfsconf based on a struct vfsconf.
3142 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3144 struct xvfsconf xvfsp;
3146 bzero(&xvfsp, sizeof(xvfsp));
3147 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3148 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3149 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3150 xvfsp.vfc_flags = vfsp->vfc_flags;
3152 * These are unused in userland, we keep them
3153 * to not break binary compatibility.
3155 xvfsp.vfc_vfsops = NULL;
3156 xvfsp.vfc_next = NULL;
3157 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3160 #ifdef COMPAT_FREEBSD32
3162 uint32_t vfc_vfsops;
3163 char vfc_name[MFSNAMELEN];
3164 int32_t vfc_typenum;
3165 int32_t vfc_refcount;
3171 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3173 struct xvfsconf32 xvfsp;
3175 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3176 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3177 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3178 xvfsp.vfc_flags = vfsp->vfc_flags;
3179 xvfsp.vfc_vfsops = 0;
3181 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3186 * Top level filesystem related information gathering.
3189 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3191 struct vfsconf *vfsp;
3195 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3196 #ifdef COMPAT_FREEBSD32
3197 if (req->flags & SCTL_MASK32)
3198 error = vfsconf2x32(req, vfsp);
3201 error = vfsconf2x(req, vfsp);
3208 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
3209 NULL, 0, sysctl_vfs_conflist,
3210 "S,xvfsconf", "List of all configured filesystems");
3212 #ifndef BURN_BRIDGES
3213 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3216 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3218 int *name = (int *)arg1 - 1; /* XXX */
3219 u_int namelen = arg2 + 1; /* XXX */
3220 struct vfsconf *vfsp;
3222 log(LOG_WARNING, "userland calling deprecated sysctl, "
3223 "please rebuild world\n");
3225 #if 1 || defined(COMPAT_PRELITE2)
3226 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3228 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3232 case VFS_MAXTYPENUM:
3235 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3238 return (ENOTDIR); /* overloaded */
3239 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
3240 if (vfsp->vfc_typenum == name[2])
3243 return (EOPNOTSUPP);
3244 #ifdef COMPAT_FREEBSD32
3245 if (req->flags & SCTL_MASK32)
3246 return (vfsconf2x32(req, vfsp));
3249 return (vfsconf2x(req, vfsp));
3251 return (EOPNOTSUPP);
3254 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
3255 vfs_sysctl, "Generic filesystem");
3257 #if 1 || defined(COMPAT_PRELITE2)
3260 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3263 struct vfsconf *vfsp;
3264 struct ovfsconf ovfs;
3266 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3267 bzero(&ovfs, sizeof(ovfs));
3268 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3269 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3270 ovfs.vfc_index = vfsp->vfc_typenum;
3271 ovfs.vfc_refcount = vfsp->vfc_refcount;
3272 ovfs.vfc_flags = vfsp->vfc_flags;
3273 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3280 #endif /* 1 || COMPAT_PRELITE2 */
3281 #endif /* !BURN_BRIDGES */
3283 #define KINFO_VNODESLOP 10
3286 * Dump vnode list (via sysctl).
3290 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3298 * Stale numvnodes access is not fatal here.
3301 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3303 /* Make an estimate */
3304 return (SYSCTL_OUT(req, 0, len));
3306 error = sysctl_wire_old_buffer(req, 0);
3309 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3311 mtx_lock(&mountlist_mtx);
3312 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3313 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3316 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3320 xvn[n].xv_size = sizeof *xvn;
3321 xvn[n].xv_vnode = vp;
3322 xvn[n].xv_id = 0; /* XXX compat */
3323 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3325 XV_COPY(writecount);
3331 xvn[n].xv_flag = vp->v_vflag;
3333 switch (vp->v_type) {
3340 if (vp->v_rdev == NULL) {
3344 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3347 xvn[n].xv_socket = vp->v_socket;
3350 xvn[n].xv_fifo = vp->v_fifoinfo;
3355 /* shouldn't happen? */
3363 mtx_lock(&mountlist_mtx);
3368 mtx_unlock(&mountlist_mtx);
3370 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3375 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
3376 0, 0, sysctl_vnode, "S,xvnode", "");
3380 * Unmount all filesystems. The list is traversed in reverse order
3381 * of mounting to avoid dependencies.
3384 vfs_unmountall(void)
3390 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3394 * Since this only runs when rebooting, it is not interlocked.
3396 while(!TAILQ_EMPTY(&mountlist)) {
3397 mp = TAILQ_LAST(&mountlist, mntlist);
3398 error = dounmount(mp, MNT_FORCE, td);
3400 TAILQ_REMOVE(&mountlist, mp, mnt_list);
3402 * XXX: Due to the way in which we mount the root
3403 * file system off of devfs, devfs will generate a
3404 * "busy" warning when we try to unmount it before
3405 * the root. Don't print a warning as a result in
3406 * order to avoid false positive errors that may
3407 * cause needless upset.
3409 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
3410 printf("unmount of %s failed (",
3411 mp->mnt_stat.f_mntonname);
3415 printf("%d)\n", error);
3418 /* The unmount has removed mp from the mountlist */
3424 * perform msync on all vnodes under a mount point
3425 * the mount point must be locked.
3428 vfs_msync(struct mount *mp, int flags)
3430 struct vnode *vp, *mvp;
3431 struct vm_object *obj;
3433 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3434 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3436 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3437 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3439 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3441 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3448 VM_OBJECT_WLOCK(obj);
3449 vm_object_page_clean(obj, 0, 0,
3451 OBJPC_SYNC : OBJPC_NOSYNC);
3452 VM_OBJECT_WUNLOCK(obj);
3462 destroy_vpollinfo_free(struct vpollinfo *vi)
3465 knlist_destroy(&vi->vpi_selinfo.si_note);
3466 mtx_destroy(&vi->vpi_lock);
3467 uma_zfree(vnodepoll_zone, vi);
3471 destroy_vpollinfo(struct vpollinfo *vi)
3474 knlist_clear(&vi->vpi_selinfo.si_note, 1);
3475 seldrain(&vi->vpi_selinfo);
3476 destroy_vpollinfo_free(vi);
3480 * Initalize per-vnode helper structure to hold poll-related state.
3483 v_addpollinfo(struct vnode *vp)
3485 struct vpollinfo *vi;
3487 if (vp->v_pollinfo != NULL)
3489 vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
3490 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3491 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3492 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3494 if (vp->v_pollinfo != NULL) {
3496 destroy_vpollinfo_free(vi);
3499 vp->v_pollinfo = vi;
3504 * Record a process's interest in events which might happen to
3505 * a vnode. Because poll uses the historic select-style interface
3506 * internally, this routine serves as both the ``check for any
3507 * pending events'' and the ``record my interest in future events''
3508 * functions. (These are done together, while the lock is held,
3509 * to avoid race conditions.)
3512 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3516 mtx_lock(&vp->v_pollinfo->vpi_lock);
3517 if (vp->v_pollinfo->vpi_revents & events) {
3519 * This leaves events we are not interested
3520 * in available for the other process which
3521 * which presumably had requested them
3522 * (otherwise they would never have been
3525 events &= vp->v_pollinfo->vpi_revents;
3526 vp->v_pollinfo->vpi_revents &= ~events;
3528 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3531 vp->v_pollinfo->vpi_events |= events;
3532 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3533 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3538 * Routine to create and manage a filesystem syncer vnode.
3540 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3541 static int sync_fsync(struct vop_fsync_args *);
3542 static int sync_inactive(struct vop_inactive_args *);
3543 static int sync_reclaim(struct vop_reclaim_args *);
3545 static struct vop_vector sync_vnodeops = {
3546 .vop_bypass = VOP_EOPNOTSUPP,
3547 .vop_close = sync_close, /* close */
3548 .vop_fsync = sync_fsync, /* fsync */
3549 .vop_inactive = sync_inactive, /* inactive */
3550 .vop_reclaim = sync_reclaim, /* reclaim */
3551 .vop_lock1 = vop_stdlock, /* lock */
3552 .vop_unlock = vop_stdunlock, /* unlock */
3553 .vop_islocked = vop_stdislocked, /* islocked */
3557 * Create a new filesystem syncer vnode for the specified mount point.
3560 vfs_allocate_syncvnode(struct mount *mp)
3564 static long start, incr, next;
3567 /* Allocate a new vnode */
3568 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
3570 panic("vfs_allocate_syncvnode: getnewvnode() failed");
3572 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3573 vp->v_vflag |= VV_FORCEINSMQ;
3574 error = insmntque(vp, mp);
3576 panic("vfs_allocate_syncvnode: insmntque() failed");
3577 vp->v_vflag &= ~VV_FORCEINSMQ;
3580 * Place the vnode onto the syncer worklist. We attempt to
3581 * scatter them about on the list so that they will go off
3582 * at evenly distributed times even if all the filesystems
3583 * are mounted at once.
3586 if (next == 0 || next > syncer_maxdelay) {
3590 start = syncer_maxdelay / 2;
3591 incr = syncer_maxdelay;
3597 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
3598 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
3599 mtx_lock(&sync_mtx);
3601 if (mp->mnt_syncer == NULL) {
3602 mp->mnt_syncer = vp;
3605 mtx_unlock(&sync_mtx);
3608 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
3615 vfs_deallocate_syncvnode(struct mount *mp)
3619 mtx_lock(&sync_mtx);
3620 vp = mp->mnt_syncer;
3622 mp->mnt_syncer = NULL;
3623 mtx_unlock(&sync_mtx);
3629 * Do a lazy sync of the filesystem.
3632 sync_fsync(struct vop_fsync_args *ap)
3634 struct vnode *syncvp = ap->a_vp;
3635 struct mount *mp = syncvp->v_mount;
3640 * We only need to do something if this is a lazy evaluation.
3642 if (ap->a_waitfor != MNT_LAZY)
3646 * Move ourselves to the back of the sync list.
3648 bo = &syncvp->v_bufobj;
3650 vn_syncer_add_to_worklist(bo, syncdelay);
3654 * Walk the list of vnodes pushing all that are dirty and
3655 * not already on the sync list.
3657 mtx_lock(&mountlist_mtx);
3658 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
3659 mtx_unlock(&mountlist_mtx);
3662 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
3666 save = curthread_pflags_set(TDP_SYNCIO);
3667 vfs_msync(mp, MNT_NOWAIT);
3668 error = VFS_SYNC(mp, MNT_LAZY);
3669 curthread_pflags_restore(save);
3670 vn_finished_write(mp);
3676 * The syncer vnode is no referenced.
3679 sync_inactive(struct vop_inactive_args *ap)
3687 * The syncer vnode is no longer needed and is being decommissioned.
3689 * Modifications to the worklist must be protected by sync_mtx.
3692 sync_reclaim(struct vop_reclaim_args *ap)
3694 struct vnode *vp = ap->a_vp;
3699 mtx_lock(&sync_mtx);
3700 if (vp->v_mount->mnt_syncer == vp)
3701 vp->v_mount->mnt_syncer = NULL;
3702 if (bo->bo_flag & BO_ONWORKLST) {
3703 LIST_REMOVE(bo, bo_synclist);
3704 syncer_worklist_len--;
3706 bo->bo_flag &= ~BO_ONWORKLST;
3708 mtx_unlock(&sync_mtx);
3715 * Check if vnode represents a disk device
3718 vn_isdisk(struct vnode *vp, int *errp)
3724 if (vp->v_type != VCHR)
3726 else if (vp->v_rdev == NULL)
3728 else if (vp->v_rdev->si_devsw == NULL)
3730 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
3735 return (error == 0);
3739 * Common filesystem object access control check routine. Accepts a
3740 * vnode's type, "mode", uid and gid, requested access mode, credentials,
3741 * and optional call-by-reference privused argument allowing vaccess()
3742 * to indicate to the caller whether privilege was used to satisfy the
3743 * request (obsoleted). Returns 0 on success, or an errno on failure.
3746 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
3747 accmode_t accmode, struct ucred *cred, int *privused)
3749 accmode_t dac_granted;
3750 accmode_t priv_granted;
3752 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
3753 ("invalid bit in accmode"));
3754 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
3755 ("VAPPEND without VWRITE"));
3758 * Look for a normal, non-privileged way to access the file/directory
3759 * as requested. If it exists, go with that.
3762 if (privused != NULL)
3767 /* Check the owner. */
3768 if (cred->cr_uid == file_uid) {
3769 dac_granted |= VADMIN;
3770 if (file_mode & S_IXUSR)
3771 dac_granted |= VEXEC;
3772 if (file_mode & S_IRUSR)
3773 dac_granted |= VREAD;
3774 if (file_mode & S_IWUSR)
3775 dac_granted |= (VWRITE | VAPPEND);
3777 if ((accmode & dac_granted) == accmode)
3783 /* Otherwise, check the groups (first match) */
3784 if (groupmember(file_gid, cred)) {
3785 if (file_mode & S_IXGRP)
3786 dac_granted |= VEXEC;
3787 if (file_mode & S_IRGRP)
3788 dac_granted |= VREAD;
3789 if (file_mode & S_IWGRP)
3790 dac_granted |= (VWRITE | VAPPEND);
3792 if ((accmode & dac_granted) == accmode)
3798 /* Otherwise, check everyone else. */
3799 if (file_mode & S_IXOTH)
3800 dac_granted |= VEXEC;
3801 if (file_mode & S_IROTH)
3802 dac_granted |= VREAD;
3803 if (file_mode & S_IWOTH)
3804 dac_granted |= (VWRITE | VAPPEND);
3805 if ((accmode & dac_granted) == accmode)
3810 * Build a privilege mask to determine if the set of privileges
3811 * satisfies the requirements when combined with the granted mask
3812 * from above. For each privilege, if the privilege is required,
3813 * bitwise or the request type onto the priv_granted mask.
3819 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
3820 * requests, instead of PRIV_VFS_EXEC.
3822 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3823 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
3824 priv_granted |= VEXEC;
3827 * Ensure that at least one execute bit is on. Otherwise,
3828 * a privileged user will always succeed, and we don't want
3829 * this to happen unless the file really is executable.
3831 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
3832 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
3833 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
3834 priv_granted |= VEXEC;
3837 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
3838 !priv_check_cred(cred, PRIV_VFS_READ, 0))
3839 priv_granted |= VREAD;
3841 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
3842 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
3843 priv_granted |= (VWRITE | VAPPEND);
3845 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
3846 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
3847 priv_granted |= VADMIN;
3849 if ((accmode & (priv_granted | dac_granted)) == accmode) {
3850 /* XXX audit: privilege used */
3851 if (privused != NULL)
3856 return ((accmode & VADMIN) ? EPERM : EACCES);
3860 * Credential check based on process requesting service, and per-attribute
3864 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
3865 struct thread *td, accmode_t accmode)
3869 * Kernel-invoked always succeeds.
3875 * Do not allow privileged processes in jail to directly manipulate
3876 * system attributes.
3878 switch (attrnamespace) {
3879 case EXTATTR_NAMESPACE_SYSTEM:
3880 /* Potentially should be: return (EPERM); */
3881 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
3882 case EXTATTR_NAMESPACE_USER:
3883 return (VOP_ACCESS(vp, accmode, cred, td));
3889 #ifdef DEBUG_VFS_LOCKS
3891 * This only exists to supress warnings from unlocked specfs accesses. It is
3892 * no longer ok to have an unlocked VFS.
3894 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
3895 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
3897 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
3898 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
3899 "Drop into debugger on lock violation");
3901 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
3902 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
3903 0, "Check for interlock across VOPs");
3905 int vfs_badlock_print = 1; /* Print lock violations. */
3906 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
3907 0, "Print lock violations");
3910 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
3911 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
3912 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
3916 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
3920 if (vfs_badlock_backtrace)
3923 if (vfs_badlock_print)
3924 printf("%s: %p %s\n", str, (void *)vp, msg);
3925 if (vfs_badlock_ddb)
3926 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
3930 assert_vi_locked(struct vnode *vp, const char *str)
3933 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
3934 vfs_badlock("interlock is not locked but should be", str, vp);
3938 assert_vi_unlocked(struct vnode *vp, const char *str)
3941 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
3942 vfs_badlock("interlock is locked but should not be", str, vp);
3946 assert_vop_locked(struct vnode *vp, const char *str)
3950 if (!IGNORE_LOCK(vp)) {
3951 locked = VOP_ISLOCKED(vp);
3952 if (locked == 0 || locked == LK_EXCLOTHER)
3953 vfs_badlock("is not locked but should be", str, vp);
3958 assert_vop_unlocked(struct vnode *vp, const char *str)
3961 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
3962 vfs_badlock("is locked but should not be", str, vp);
3966 assert_vop_elocked(struct vnode *vp, const char *str)
3969 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
3970 vfs_badlock("is not exclusive locked but should be", str, vp);
3975 assert_vop_elocked_other(struct vnode *vp, const char *str)
3978 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
3979 vfs_badlock("is not exclusive locked by another thread",
3984 assert_vop_slocked(struct vnode *vp, const char *str)
3987 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
3988 vfs_badlock("is not locked shared but should be", str, vp);
3991 #endif /* DEBUG_VFS_LOCKS */
3994 vop_rename_fail(struct vop_rename_args *ap)
3997 if (ap->a_tvp != NULL)
3999 if (ap->a_tdvp == ap->a_tvp)
4008 vop_rename_pre(void *ap)
4010 struct vop_rename_args *a = ap;
4012 #ifdef DEBUG_VFS_LOCKS
4014 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4015 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4016 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4017 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4019 /* Check the source (from). */
4020 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4021 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4022 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4023 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4024 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4026 /* Check the target. */
4028 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4029 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4031 if (a->a_tdvp != a->a_fdvp)
4033 if (a->a_tvp != a->a_fvp)
4041 vop_strategy_pre(void *ap)
4043 #ifdef DEBUG_VFS_LOCKS
4044 struct vop_strategy_args *a;
4051 * Cluster ops lock their component buffers but not the IO container.
4053 if ((bp->b_flags & B_CLUSTER) != 0)
4056 if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4057 if (vfs_badlock_print)
4059 "VOP_STRATEGY: bp is not locked but should be\n");
4060 if (vfs_badlock_ddb)
4061 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4067 vop_lock_pre(void *ap)
4069 #ifdef DEBUG_VFS_LOCKS
4070 struct vop_lock1_args *a = ap;
4072 if ((a->a_flags & LK_INTERLOCK) == 0)
4073 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4075 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4080 vop_lock_post(void *ap, int rc)
4082 #ifdef DEBUG_VFS_LOCKS
4083 struct vop_lock1_args *a = ap;
4085 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4086 if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
4087 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4092 vop_unlock_pre(void *ap)
4094 #ifdef DEBUG_VFS_LOCKS
4095 struct vop_unlock_args *a = ap;
4097 if (a->a_flags & LK_INTERLOCK)
4098 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4099 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4104 vop_unlock_post(void *ap, int rc)
4106 #ifdef DEBUG_VFS_LOCKS
4107 struct vop_unlock_args *a = ap;
4109 if (a->a_flags & LK_INTERLOCK)
4110 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4115 vop_create_post(void *ap, int rc)
4117 struct vop_create_args *a = ap;
4120 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4124 vop_deleteextattr_post(void *ap, int rc)
4126 struct vop_deleteextattr_args *a = ap;
4129 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4133 vop_link_post(void *ap, int rc)
4135 struct vop_link_args *a = ap;
4138 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4139 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4144 vop_mkdir_post(void *ap, int rc)
4146 struct vop_mkdir_args *a = ap;
4149 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4153 vop_mknod_post(void *ap, int rc)
4155 struct vop_mknod_args *a = ap;
4158 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4162 vop_remove_post(void *ap, int rc)
4164 struct vop_remove_args *a = ap;
4167 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4168 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4173 vop_rename_post(void *ap, int rc)
4175 struct vop_rename_args *a = ap;
4178 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4179 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4180 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4182 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4184 if (a->a_tdvp != a->a_fdvp)
4186 if (a->a_tvp != a->a_fvp)
4194 vop_rmdir_post(void *ap, int rc)
4196 struct vop_rmdir_args *a = ap;
4199 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4200 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4205 vop_setattr_post(void *ap, int rc)
4207 struct vop_setattr_args *a = ap;
4210 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4214 vop_setextattr_post(void *ap, int rc)
4216 struct vop_setextattr_args *a = ap;
4219 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4223 vop_symlink_post(void *ap, int rc)
4225 struct vop_symlink_args *a = ap;
4228 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4231 static struct knlist fs_knlist;
4234 vfs_event_init(void *arg)
4236 knlist_init_mtx(&fs_knlist, NULL);
4238 /* XXX - correct order? */
4239 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4242 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4245 KNOTE_UNLOCKED(&fs_knlist, event);
4248 static int filt_fsattach(struct knote *kn);
4249 static void filt_fsdetach(struct knote *kn);
4250 static int filt_fsevent(struct knote *kn, long hint);
4252 struct filterops fs_filtops = {
4254 .f_attach = filt_fsattach,
4255 .f_detach = filt_fsdetach,
4256 .f_event = filt_fsevent
4260 filt_fsattach(struct knote *kn)
4263 kn->kn_flags |= EV_CLEAR;
4264 knlist_add(&fs_knlist, kn, 0);
4269 filt_fsdetach(struct knote *kn)
4272 knlist_remove(&fs_knlist, kn, 0);
4276 filt_fsevent(struct knote *kn, long hint)
4279 kn->kn_fflags |= hint;
4280 return (kn->kn_fflags != 0);
4284 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4290 error = SYSCTL_IN(req, &vc, sizeof(vc));
4293 if (vc.vc_vers != VFS_CTL_VERS1)
4295 mp = vfs_getvfs(&vc.vc_fsid);
4298 /* ensure that a specific sysctl goes to the right filesystem. */
4299 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4300 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4304 VCTLTOREQ(&vc, req);
4305 error = VFS_SYSCTL(mp, vc.vc_op, req);
4310 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4311 NULL, 0, sysctl_vfs_ctl, "",
4315 * Function to initialize a va_filerev field sensibly.
4316 * XXX: Wouldn't a random number make a lot more sense ??
4319 init_va_filerev(void)
4324 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4327 static int filt_vfsread(struct knote *kn, long hint);
4328 static int filt_vfswrite(struct knote *kn, long hint);
4329 static int filt_vfsvnode(struct knote *kn, long hint);
4330 static void filt_vfsdetach(struct knote *kn);
4331 static struct filterops vfsread_filtops = {
4333 .f_detach = filt_vfsdetach,
4334 .f_event = filt_vfsread
4336 static struct filterops vfswrite_filtops = {
4338 .f_detach = filt_vfsdetach,
4339 .f_event = filt_vfswrite
4341 static struct filterops vfsvnode_filtops = {
4343 .f_detach = filt_vfsdetach,
4344 .f_event = filt_vfsvnode
4348 vfs_knllock(void *arg)
4350 struct vnode *vp = arg;
4352 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4356 vfs_knlunlock(void *arg)
4358 struct vnode *vp = arg;
4364 vfs_knl_assert_locked(void *arg)
4366 #ifdef DEBUG_VFS_LOCKS
4367 struct vnode *vp = arg;
4369 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4374 vfs_knl_assert_unlocked(void *arg)
4376 #ifdef DEBUG_VFS_LOCKS
4377 struct vnode *vp = arg;
4379 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4384 vfs_kqfilter(struct vop_kqfilter_args *ap)
4386 struct vnode *vp = ap->a_vp;
4387 struct knote *kn = ap->a_kn;
4390 switch (kn->kn_filter) {
4392 kn->kn_fop = &vfsread_filtops;
4395 kn->kn_fop = &vfswrite_filtops;
4398 kn->kn_fop = &vfsvnode_filtops;
4404 kn->kn_hook = (caddr_t)vp;
4407 if (vp->v_pollinfo == NULL)
4409 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4411 knlist_add(knl, kn, 0);
4417 * Detach knote from vnode
4420 filt_vfsdetach(struct knote *kn)
4422 struct vnode *vp = (struct vnode *)kn->kn_hook;
4424 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4425 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4431 filt_vfsread(struct knote *kn, long hint)
4433 struct vnode *vp = (struct vnode *)kn->kn_hook;
4438 * filesystem is gone, so set the EOF flag and schedule
4439 * the knote for deletion.
4441 if (hint == NOTE_REVOKE) {
4443 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4448 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4452 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4453 res = (kn->kn_data != 0);
4460 filt_vfswrite(struct knote *kn, long hint)
4462 struct vnode *vp = (struct vnode *)kn->kn_hook;
4467 * filesystem is gone, so set the EOF flag and schedule
4468 * the knote for deletion.
4470 if (hint == NOTE_REVOKE)
4471 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4479 filt_vfsvnode(struct knote *kn, long hint)
4481 struct vnode *vp = (struct vnode *)kn->kn_hook;
4485 if (kn->kn_sfflags & hint)
4486 kn->kn_fflags |= hint;
4487 if (hint == NOTE_REVOKE) {
4488 kn->kn_flags |= EV_EOF;
4492 res = (kn->kn_fflags != 0);
4498 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4502 if (dp->d_reclen > ap->a_uio->uio_resid)
4503 return (ENAMETOOLONG);
4504 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4506 if (ap->a_ncookies != NULL) {
4507 if (ap->a_cookies != NULL)
4508 free(ap->a_cookies, M_TEMP);
4509 ap->a_cookies = NULL;
4510 *ap->a_ncookies = 0;
4514 if (ap->a_ncookies == NULL)
4517 KASSERT(ap->a_cookies,
4518 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4520 *ap->a_cookies = realloc(*ap->a_cookies,
4521 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4522 (*ap->a_cookies)[*ap->a_ncookies] = off;
4527 * Mark for update the access time of the file if the filesystem
4528 * supports VOP_MARKATIME. This functionality is used by execve and
4529 * mmap, so we want to avoid the I/O implied by directly setting
4530 * va_atime for the sake of efficiency.
4533 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4538 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4539 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4540 (void)VOP_MARKATIME(vp);
4544 * The purpose of this routine is to remove granularity from accmode_t,
4545 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
4546 * VADMIN and VAPPEND.
4548 * If it returns 0, the caller is supposed to continue with the usual
4549 * access checks using 'accmode' as modified by this routine. If it
4550 * returns nonzero value, the caller is supposed to return that value
4553 * Note that after this routine runs, accmode may be zero.
4556 vfs_unixify_accmode(accmode_t *accmode)
4559 * There is no way to specify explicit "deny" rule using
4560 * file mode or POSIX.1e ACLs.
4562 if (*accmode & VEXPLICIT_DENY) {
4568 * None of these can be translated into usual access bits.
4569 * Also, the common case for NFSv4 ACLs is to not contain
4570 * either of these bits. Caller should check for VWRITE
4571 * on the containing directory instead.
4573 if (*accmode & (VDELETE_CHILD | VDELETE))
4576 if (*accmode & VADMIN_PERMS) {
4577 *accmode &= ~VADMIN_PERMS;
4582 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
4583 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
4585 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
4591 * These are helper functions for filesystems to traverse all
4592 * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
4594 * This interface replaces MNT_VNODE_FOREACH.
4597 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
4600 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
4605 kern_yield(PRI_USER);
4607 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4608 vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
4609 while (vp != NULL && (vp->v_type == VMARKER ||
4610 (vp->v_iflag & VI_DOOMED) != 0))
4611 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4613 /* Check if we are done */
4615 __mnt_vnode_markerfree_all(mvp, mp);
4616 /* MNT_IUNLOCK(mp); -- done in above function */
4617 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
4620 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4621 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4628 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
4632 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4635 (*mvp)->v_type = VMARKER;
4637 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
4638 while (vp != NULL && (vp->v_type == VMARKER ||
4639 (vp->v_iflag & VI_DOOMED) != 0))
4640 vp = TAILQ_NEXT(vp, v_nmntvnodes);
4642 /* Check if we are done */
4646 free(*mvp, M_VNODE_MARKER);
4650 (*mvp)->v_mount = mp;
4651 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
4659 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
4667 mtx_assert(MNT_MTX(mp), MA_OWNED);
4669 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4670 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
4673 free(*mvp, M_VNODE_MARKER);
4678 * These are helper functions for filesystems to traverse their
4679 * active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
4682 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4685 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4690 free(*mvp, M_VNODE_MARKER);
4694 static struct vnode *
4695 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4697 struct vnode *vp, *nvp;
4699 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
4700 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
4702 vp = TAILQ_NEXT(*mvp, v_actfreelist);
4703 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4704 while (vp != NULL) {
4705 if (vp->v_type == VMARKER) {
4706 vp = TAILQ_NEXT(vp, v_actfreelist);
4709 if (!VI_TRYLOCK(vp)) {
4710 if (mp_ncpus == 1 || should_yield()) {
4711 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4712 mtx_unlock(&vnode_free_list_mtx);
4714 mtx_lock(&vnode_free_list_mtx);
4719 KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
4720 KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
4721 ("alien vnode on the active list %p %p", vp, mp));
4722 if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
4724 nvp = TAILQ_NEXT(vp, v_actfreelist);
4729 /* Check if we are done */
4731 mtx_unlock(&vnode_free_list_mtx);
4732 mnt_vnode_markerfree_active(mvp, mp);
4735 TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
4736 mtx_unlock(&vnode_free_list_mtx);
4737 ASSERT_VI_LOCKED(vp, "active iter");
4738 KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
4743 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
4747 kern_yield(PRI_USER);
4748 mtx_lock(&vnode_free_list_mtx);
4749 return (mnt_vnode_next_active(mvp, mp));
4753 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
4757 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
4761 (*mvp)->v_type = VMARKER;
4762 (*mvp)->v_mount = mp;
4764 mtx_lock(&vnode_free_list_mtx);
4765 vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
4767 mtx_unlock(&vnode_free_list_mtx);
4768 mnt_vnode_markerfree_active(mvp, mp);
4771 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
4772 return (mnt_vnode_next_active(mvp, mp));
4776 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
4782 mtx_lock(&vnode_free_list_mtx);
4783 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
4784 mtx_unlock(&vnode_free_list_mtx);
4785 mnt_vnode_markerfree_active(mvp, mp);