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/refcount.h>
72 #include <sys/rwlock.h>
73 #include <sys/sched.h>
74 #include <sys/sleepqueue.h>
77 #include <sys/sysctl.h>
78 #include <sys/syslog.h>
79 #include <sys/vmmeter.h>
80 #include <sys/vnode.h>
81 #include <sys/watchdog.h>
83 #include <machine/stdarg.h>
85 #include <security/mac/mac_framework.h>
88 #include <vm/vm_object.h>
89 #include <vm/vm_extern.h>
91 #include <vm/vm_map.h>
92 #include <vm/vm_page.h>
93 #include <vm/vm_kern.h>
100 static void delmntque(struct vnode *vp);
101 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
102 int slpflag, int slptimeo);
103 static void syncer_shutdown(void *arg, int howto);
104 static int vtryrecycle(struct vnode *vp);
105 static void v_init_counters(struct vnode *);
106 static void v_incr_usecount(struct vnode *);
107 static void v_incr_usecount_locked(struct vnode *);
108 static void v_incr_devcount(struct vnode *);
109 static void v_decr_devcount(struct vnode *);
110 static void vnlru_free(int);
111 static void vgonel(struct vnode *);
112 static void vfs_knllock(void *arg);
113 static void vfs_knlunlock(void *arg);
114 static void vfs_knl_assert_locked(void *arg);
115 static void vfs_knl_assert_unlocked(void *arg);
116 static void destroy_vpollinfo(struct vpollinfo *vi);
119 * Number of vnodes in existence. Increased whenever getnewvnode()
120 * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
122 static unsigned long numvnodes;
124 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
125 "Number of vnodes in existence");
127 static u_long vnodes_created;
128 SYSCTL_ULONG(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created,
129 0, "Number of vnodes created by getnewvnode");
132 * Conversion tables for conversion from vnode types to inode formats
135 enum vtype iftovt_tab[16] = {
136 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
137 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
139 int vttoif_tab[10] = {
140 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
141 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
145 * List of vnodes that are ready for recycling.
147 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
150 * "Free" vnode target. Free vnodes are rarely completely free, but are
151 * just ones that are cheap to recycle. Usually they are for files which
152 * have been stat'd but not read; these usually have inode and namecache
153 * data attached to them. This target is the preferred minimum size of a
154 * sub-cache consisting mostly of such files. The system balances the size
155 * of this sub-cache with its complement to try to prevent either from
156 * thrashing while the other is relatively inactive. The targets express
157 * a preference for the best balance.
159 * "Above" this target there are 2 further targets (watermarks) related
160 * to recyling of free vnodes. In the best-operating case, the cache is
161 * exactly full, the free list has size between vlowat and vhiwat above the
162 * free target, and recycling from it and normal use maintains this state.
163 * Sometimes the free list is below vlowat or even empty, but this state
164 * is even better for immediate use provided the cache is not full.
165 * Otherwise, vnlru_proc() runs to reclaim enough vnodes (usually non-free
166 * ones) to reach one of these states. The watermarks are currently hard-
167 * coded as 4% and 9% of the available space higher. These and the default
168 * of 25% for wantfreevnodes are too large if the memory size is large.
169 * E.g., 9% of 75% of MAXVNODES is more than 566000 vnodes to reclaim
170 * whenever vnlru_proc() becomes active.
172 static u_long wantfreevnodes;
173 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW,
174 &wantfreevnodes, 0, "Target for minimum number of \"free\" vnodes");
175 static u_long freevnodes;
176 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD,
177 &freevnodes, 0, "Number of \"free\" vnodes");
179 static u_long recycles_count;
180 SYSCTL_ULONG(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, 0,
181 "Number of vnodes recycled to meet vnode cache targets");
184 * Various variables used for debugging the new implementation of
186 * XXX these are probably of (very) limited utility now.
188 static int reassignbufcalls;
189 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
190 "Number of calls to reassignbuf");
192 static u_long free_owe_inact;
193 SYSCTL_ULONG(_vfs, OID_AUTO, free_owe_inact, CTLFLAG_RD, &free_owe_inact, 0,
194 "Number of times free vnodes kept on active list due to VFS "
195 "owing inactivation");
197 /* To keep more than one thread at a time from running vfs_getnewfsid */
198 static struct mtx mntid_mtx;
201 * Lock for any access to the following:
206 static struct mtx vnode_free_list_mtx;
208 /* Publicly exported FS */
209 struct nfs_public nfs_pub;
211 static uma_zone_t buf_trie_zone;
213 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
214 static uma_zone_t vnode_zone;
215 static uma_zone_t vnodepoll_zone;
218 * The workitem queue.
220 * It is useful to delay writes of file data and filesystem metadata
221 * for tens of seconds so that quickly created and deleted files need
222 * not waste disk bandwidth being created and removed. To realize this,
223 * we append vnodes to a "workitem" queue. When running with a soft
224 * updates implementation, most pending metadata dependencies should
225 * not wait for more than a few seconds. Thus, mounted on block devices
226 * are delayed only about a half the time that file data is delayed.
227 * Similarly, directory updates are more critical, so are only delayed
228 * about a third the time that file data is delayed. Thus, there are
229 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
230 * one each second (driven off the filesystem syncer process). The
231 * syncer_delayno variable indicates the next queue that is to be processed.
232 * Items that need to be processed soon are placed in this queue:
234 * syncer_workitem_pending[syncer_delayno]
236 * A delay of fifteen seconds is done by placing the request fifteen
237 * entries later in the queue:
239 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
242 static int syncer_delayno;
243 static long syncer_mask;
244 LIST_HEAD(synclist, bufobj);
245 static struct synclist *syncer_workitem_pending;
247 * The sync_mtx protects:
252 * syncer_workitem_pending
253 * syncer_worklist_len
256 static struct mtx sync_mtx;
257 static struct cv sync_wakeup;
259 #define SYNCER_MAXDELAY 32
260 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
261 static int syncdelay = 30; /* max time to delay syncing data */
262 static int filedelay = 30; /* time to delay syncing files */
263 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
264 "Time to delay syncing files (in seconds)");
265 static int dirdelay = 29; /* time to delay syncing directories */
266 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
267 "Time to delay syncing directories (in seconds)");
268 static int metadelay = 28; /* time to delay syncing metadata */
269 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
270 "Time to delay syncing metadata (in seconds)");
271 static int rushjob; /* number of slots to run ASAP */
272 static int stat_rush_requests; /* number of times I/O speeded up */
273 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
274 "Number of times I/O speeded up (rush requests)");
277 * When shutting down the syncer, run it at four times normal speed.
279 #define SYNCER_SHUTDOWN_SPEEDUP 4
280 static int sync_vnode_count;
281 static int syncer_worklist_len;
282 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
285 /* Target for maximum number of vnodes. */
287 static int gapvnodes; /* gap between wanted and desired */
288 static int vhiwat; /* enough extras after expansion */
289 static int vlowat; /* minimal extras before expansion */
290 static int vstir; /* nonzero to stir non-free vnodes */
291 static volatile int vsmalltrigger = 8; /* pref to keep if > this many pages */
294 sysctl_update_desiredvnodes(SYSCTL_HANDLER_ARGS)
296 int error, old_desiredvnodes;
298 old_desiredvnodes = desiredvnodes;
299 if ((error = sysctl_handle_int(oidp, arg1, arg2, req)) != 0)
301 if (old_desiredvnodes != desiredvnodes) {
302 wantfreevnodes = desiredvnodes / 4;
303 /* XXX locking seems to be incomplete. */
304 vfs_hash_changesize(desiredvnodes);
305 cache_changesize(desiredvnodes);
310 SYSCTL_PROC(_kern, KERN_MAXVNODES, maxvnodes,
311 CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, &desiredvnodes, 0,
312 sysctl_update_desiredvnodes, "I", "Target for maximum number of vnodes");
313 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
314 &wantfreevnodes, 0, "Old name for vfs.wantfreevnodes (legacy)");
315 static int vnlru_nowhere;
316 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
317 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
319 /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
323 * Support for the bufobj clean & dirty pctrie.
326 buf_trie_alloc(struct pctrie *ptree)
329 return uma_zalloc(buf_trie_zone, M_NOWAIT);
333 buf_trie_free(struct pctrie *ptree, void *node)
336 uma_zfree(buf_trie_zone, node);
338 PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
341 * Initialize the vnode management data structures.
343 * Reevaluate the following cap on the number of vnodes after the physical
344 * memory size exceeds 512GB. In the limit, as the physical memory size
345 * grows, the ratio of the memory size in KB to to vnodes approaches 64:1.
347 #ifndef MAXVNODES_MAX
348 #define MAXVNODES_MAX (512 * 1024 * 1024 / 64) /* 8M */
352 * Initialize a vnode as it first enters the zone.
355 vnode_init(void *mem, int size, int flags)
365 vp->v_vnlock = &vp->v_lock;
366 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
368 * By default, don't allow shared locks unless filesystems opt-in.
370 lockinit(vp->v_vnlock, PVFS, "vnode", VLKTIMEOUT,
371 LK_NOSHARE | LK_IS_VNODE);
377 rw_init(BO_LOCKPTR(bo), "bufobj interlock");
379 TAILQ_INIT(&bo->bo_clean.bv_hd);
380 TAILQ_INIT(&bo->bo_dirty.bv_hd);
382 * Initialize namecache.
384 LIST_INIT(&vp->v_cache_src);
385 TAILQ_INIT(&vp->v_cache_dst);
387 * Initialize rangelocks.
389 rangelock_init(&vp->v_rl);
394 * Free a vnode when it is cleared from the zone.
397 vnode_fini(void *mem, int size)
403 rangelock_destroy(&vp->v_rl);
404 lockdestroy(vp->v_vnlock);
405 mtx_destroy(&vp->v_interlock);
407 rw_destroy(BO_LOCKPTR(bo));
411 vntblinit(void *dummy __unused)
414 int physvnodes, virtvnodes;
417 * Desiredvnodes is a function of the physical memory size and the
418 * kernel's heap size. Generally speaking, it scales with the
419 * physical memory size. The ratio of desiredvnodes to the physical
420 * memory size is 1:16 until desiredvnodes exceeds 98,304.
422 * marginal ratio of desiredvnodes to the physical memory size is
423 * 1:64. However, desiredvnodes is limited by the kernel's heap
424 * size. The memory required by desiredvnodes vnodes and vm objects
425 * must not exceed 1/7th of the kernel's heap size.
427 physvnodes = maxproc + pgtok(vm_cnt.v_page_count) / 64 +
428 3 * min(98304 * 16, pgtok(vm_cnt.v_page_count)) / 64;
429 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
430 sizeof(struct vnode)));
431 desiredvnodes = min(physvnodes, virtvnodes);
432 if (desiredvnodes > MAXVNODES_MAX) {
434 printf("Reducing kern.maxvnodes %d -> %d\n",
435 desiredvnodes, MAXVNODES_MAX);
436 desiredvnodes = MAXVNODES_MAX;
438 wantfreevnodes = desiredvnodes / 4;
439 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
440 TAILQ_INIT(&vnode_free_list);
441 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
442 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
443 vnode_init, vnode_fini, UMA_ALIGN_PTR, 0);
444 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
445 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
447 * Preallocate enough nodes to support one-per buf so that
448 * we can not fail an insert. reassignbuf() callers can not
449 * tolerate the insertion failure.
451 buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
452 NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR,
453 UMA_ZONE_NOFREE | UMA_ZONE_VM);
454 uma_prealloc(buf_trie_zone, nbuf);
456 * Initialize the filesystem syncer.
458 syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
460 syncer_maxdelay = syncer_mask + 1;
461 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
462 cv_init(&sync_wakeup, "syncer");
463 for (i = 1; i <= sizeof(struct vnode); i <<= 1)
467 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
471 * Mark a mount point as busy. Used to synchronize access and to delay
472 * unmounting. Eventually, mountlist_mtx is not released on failure.
474 * vfs_busy() is a custom lock, it can block the caller.
475 * vfs_busy() only sleeps if the unmount is active on the mount point.
476 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
477 * vnode belonging to mp.
479 * Lookup uses vfs_busy() to traverse mount points.
481 * / vnode lock A / vnode lock (/var) D
482 * /var vnode lock B /log vnode lock(/var/log) E
483 * vfs_busy lock C vfs_busy lock F
485 * Within each file system, the lock order is C->A->B and F->D->E.
487 * When traversing across mounts, the system follows that lock order:
493 * The lookup() process for namei("/var") illustrates the process:
494 * VOP_LOOKUP() obtains B while A is held
495 * vfs_busy() obtains a shared lock on F while A and B are held
496 * vput() releases lock on B
497 * vput() releases lock on A
498 * VFS_ROOT() obtains lock on D while shared lock on F is held
499 * vfs_unbusy() releases shared lock on F
500 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
501 * Attempt to lock A (instead of vp_crossmp) while D is held would
502 * violate the global order, causing deadlocks.
504 * dounmount() locks B while F is drained.
507 vfs_busy(struct mount *mp, int flags)
510 MPASS((flags & ~MBF_MASK) == 0);
511 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
516 * If mount point is currenly being unmounted, sleep until the
517 * mount point fate is decided. If thread doing the unmounting fails,
518 * it will clear MNTK_UNMOUNT flag before waking us up, indicating
519 * that this mount point has survived the unmount attempt and vfs_busy
520 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
521 * flag in addition to MNTK_UNMOUNT, indicating that mount point is
522 * about to be really destroyed. vfs_busy needs to release its
523 * reference on the mount point in this case and return with ENOENT,
524 * telling the caller that mount mount it tried to busy is no longer
527 while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
528 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
531 CTR1(KTR_VFS, "%s: failed busying before sleeping",
535 if (flags & MBF_MNTLSTLOCK)
536 mtx_unlock(&mountlist_mtx);
537 mp->mnt_kern_flag |= MNTK_MWAIT;
538 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
539 if (flags & MBF_MNTLSTLOCK)
540 mtx_lock(&mountlist_mtx);
543 if (flags & MBF_MNTLSTLOCK)
544 mtx_unlock(&mountlist_mtx);
551 * Free a busy filesystem.
554 vfs_unbusy(struct mount *mp)
557 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
560 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
562 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
563 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
564 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
565 mp->mnt_kern_flag &= ~MNTK_DRAINING;
566 wakeup(&mp->mnt_lockref);
572 * Lookup a mount point by filesystem identifier.
575 vfs_getvfs(fsid_t *fsid)
579 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
580 mtx_lock(&mountlist_mtx);
581 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
582 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
583 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
585 mtx_unlock(&mountlist_mtx);
589 mtx_unlock(&mountlist_mtx);
590 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
591 return ((struct mount *) 0);
595 * Lookup a mount point by filesystem identifier, busying it before
598 * To avoid congestion on mountlist_mtx, implement simple direct-mapped
599 * cache for popular filesystem identifiers. The cache is lockess, using
600 * the fact that struct mount's are never freed. In worst case we may
601 * get pointer to unmounted or even different filesystem, so we have to
602 * check what we got, and go slow way if so.
605 vfs_busyfs(fsid_t *fsid)
607 #define FSID_CACHE_SIZE 256
608 typedef struct mount * volatile vmp_t;
609 static vmp_t cache[FSID_CACHE_SIZE];
614 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
615 hash = fsid->val[0] ^ fsid->val[1];
616 hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
619 mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
620 mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
622 if (vfs_busy(mp, 0) != 0) {
626 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
627 mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
633 mtx_lock(&mountlist_mtx);
634 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
635 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
636 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
637 error = vfs_busy(mp, MBF_MNTLSTLOCK);
640 mtx_unlock(&mountlist_mtx);
647 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
648 mtx_unlock(&mountlist_mtx);
649 return ((struct mount *) 0);
653 * Check if a user can access privileged mount options.
656 vfs_suser(struct mount *mp, struct thread *td)
661 * If the thread is jailed, but this is not a jail-friendly file
662 * system, deny immediately.
664 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
668 * If the file system was mounted outside the jail of the calling
669 * thread, deny immediately.
671 if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
675 * If file system supports delegated administration, we don't check
676 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
677 * by the file system itself.
678 * If this is not the user that did original mount, we check for
679 * the PRIV_VFS_MOUNT_OWNER privilege.
681 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
682 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
683 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
690 * Get a new unique fsid. Try to make its val[0] unique, since this value
691 * will be used to create fake device numbers for stat(). Also try (but
692 * not so hard) make its val[0] unique mod 2^16, since some emulators only
693 * support 16-bit device numbers. We end up with unique val[0]'s for the
694 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
696 * Keep in mind that several mounts may be running in parallel. Starting
697 * the search one past where the previous search terminated is both a
698 * micro-optimization and a defense against returning the same fsid to
702 vfs_getnewfsid(struct mount *mp)
704 static uint16_t mntid_base;
709 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
710 mtx_lock(&mntid_mtx);
711 mtype = mp->mnt_vfc->vfc_typenum;
712 tfsid.val[1] = mtype;
713 mtype = (mtype & 0xFF) << 24;
715 tfsid.val[0] = makedev(255,
716 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
718 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
722 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
723 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
724 mtx_unlock(&mntid_mtx);
728 * Knob to control the precision of file timestamps:
730 * 0 = seconds only; nanoseconds zeroed.
731 * 1 = seconds and nanoseconds, accurate within 1/HZ.
732 * 2 = seconds and nanoseconds, truncated to microseconds.
733 * >=3 = seconds and nanoseconds, maximum precision.
735 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
737 static int timestamp_precision = TSP_USEC;
738 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
739 ×tamp_precision, 0, "File timestamp precision (0: seconds, "
740 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
741 "3+: sec + ns (max. precision))");
744 * Get a current timestamp.
747 vfs_timestamp(struct timespec *tsp)
751 switch (timestamp_precision) {
753 tsp->tv_sec = time_second;
761 TIMEVAL_TO_TIMESPEC(&tv, tsp);
771 * Set vnode attributes to VNOVAL
774 vattr_null(struct vattr *vap)
778 vap->va_size = VNOVAL;
779 vap->va_bytes = VNOVAL;
780 vap->va_mode = VNOVAL;
781 vap->va_nlink = VNOVAL;
782 vap->va_uid = VNOVAL;
783 vap->va_gid = VNOVAL;
784 vap->va_fsid = VNOVAL;
785 vap->va_fileid = VNOVAL;
786 vap->va_blocksize = VNOVAL;
787 vap->va_rdev = VNOVAL;
788 vap->va_atime.tv_sec = VNOVAL;
789 vap->va_atime.tv_nsec = VNOVAL;
790 vap->va_mtime.tv_sec = VNOVAL;
791 vap->va_mtime.tv_nsec = VNOVAL;
792 vap->va_ctime.tv_sec = VNOVAL;
793 vap->va_ctime.tv_nsec = VNOVAL;
794 vap->va_birthtime.tv_sec = VNOVAL;
795 vap->va_birthtime.tv_nsec = VNOVAL;
796 vap->va_flags = VNOVAL;
797 vap->va_gen = VNOVAL;
802 * This routine is called when we have too many vnodes. It attempts
803 * to free <count> vnodes and will potentially free vnodes that still
804 * have VM backing store (VM backing store is typically the cause
805 * of a vnode blowout so we want to do this). Therefore, this operation
806 * is not considered cheap.
808 * A number of conditions may prevent a vnode from being reclaimed.
809 * the buffer cache may have references on the vnode, a directory
810 * vnode may still have references due to the namei cache representing
811 * underlying files, or the vnode may be in active use. It is not
812 * desireable to reuse such vnodes. These conditions may cause the
813 * number of vnodes to reach some minimum value regardless of what
814 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
817 vlrureclaim(struct mount *mp, int reclaim_nc_src, int trigger)
820 int count, done, target;
823 vn_start_write(NULL, &mp, V_WAIT);
825 count = mp->mnt_nvnodelistsize;
826 target = count * (int64_t)gapvnodes / imax(desiredvnodes, 1);
827 target = target / 10 + 1;
828 while (count != 0 && done < target) {
829 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
830 while (vp != NULL && vp->v_type == VMARKER)
831 vp = TAILQ_NEXT(vp, v_nmntvnodes);
835 * XXX LRU is completely broken for non-free vnodes. First
836 * by calling here in mountpoint order, then by moving
837 * unselected vnodes to the end here, and most grossly by
838 * removing the vlruvp() function that was supposed to
839 * maintain the order. (This function was born broken
840 * since syncer problems prevented it doing anything.) The
841 * order is closer to LRC (C = Created).
843 * LRU reclaiming of vnodes seems to have last worked in
844 * FreeBSD-3 where LRU wasn't mentioned under any spelling.
845 * Then there was no hold count, and inactive vnodes were
846 * simply put on the free list in LRU order. The separate
847 * lists also break LRU. We prefer to reclaim from the
848 * free list for technical reasons. This tends to thrash
849 * the free list to keep very unrecently used held vnodes.
850 * The problem is mitigated by keeping the free list large.
852 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
853 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
858 * If it's been deconstructed already, it's still
859 * referenced, or it exceeds the trigger, skip it.
860 * Also skip free vnodes. We are trying to make space
861 * to expand the free list, not reduce it.
863 if (vp->v_usecount ||
864 (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) ||
865 ((vp->v_iflag & VI_FREE) != 0) ||
866 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
867 vp->v_object->resident_page_count > trigger)) {
873 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
875 goto next_iter_mntunlocked;
879 * v_usecount may have been bumped after VOP_LOCK() dropped
880 * the vnode interlock and before it was locked again.
882 * It is not necessary to recheck VI_DOOMED because it can
883 * only be set by another thread that holds both the vnode
884 * lock and vnode interlock. If another thread has the
885 * vnode lock before we get to VOP_LOCK() and obtains the
886 * vnode interlock after VOP_LOCK() drops the vnode
887 * interlock, the other thread will be unable to drop the
888 * vnode lock before our VOP_LOCK() call fails.
890 if (vp->v_usecount ||
891 (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) ||
892 (vp->v_iflag & VI_FREE) != 0 ||
893 (vp->v_object != NULL &&
894 vp->v_object->resident_page_count > trigger)) {
895 VOP_UNLOCK(vp, LK_INTERLOCK);
897 goto next_iter_mntunlocked;
899 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
900 ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
901 atomic_add_long(&recycles_count, 1);
906 next_iter_mntunlocked:
915 kern_yield(PRI_USER);
920 vn_finished_write(mp);
925 * Attempt to reduce the free list by the requested amount.
928 vnlru_free(int count)
932 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
933 for (; count > 0; count--) {
934 vp = TAILQ_FIRST(&vnode_free_list);
936 * The list can be modified while the free_list_mtx
937 * has been dropped and vp could be NULL here.
941 VNASSERT(vp->v_op != NULL, vp,
942 ("vnlru_free: vnode already reclaimed."));
943 KASSERT((vp->v_iflag & VI_FREE) != 0,
944 ("Removing vnode not on freelist"));
945 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
946 ("Mangling active vnode"));
947 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
949 * Don't recycle if we can't get the interlock.
951 if (!VI_TRYLOCK(vp)) {
952 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
955 VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0,
956 vp, ("vp inconsistent on freelist"));
959 * The clear of VI_FREE prevents activation of the
960 * vnode. There is no sense in putting the vnode on
961 * the mount point active list, only to remove it
962 * later during recycling. Inline the relevant part
963 * of vholdl(), to avoid triggering assertions or
967 vp->v_iflag &= ~VI_FREE;
968 refcount_acquire(&vp->v_holdcnt);
970 mtx_unlock(&vnode_free_list_mtx);
974 * If the recycled succeeded this vdrop will actually free
975 * the vnode. If not it will simply place it back on
979 mtx_lock(&vnode_free_list_mtx);
983 /* XXX some names and initialization are bad for limits and watermarks. */
989 gapvnodes = imax(desiredvnodes - wantfreevnodes, 100);
990 vhiwat = gapvnodes / 11; /* 9% -- just under the 10% in vlrureclaim() */
992 if (numvnodes > desiredvnodes)
994 space = desiredvnodes - numvnodes;
995 if (freevnodes > wantfreevnodes)
996 space += freevnodes - wantfreevnodes;
1001 * Attempt to recycle vnodes in a context that is always safe to block.
1002 * Calling vlrurecycle() from the bowels of filesystem code has some
1003 * interesting deadlock problems.
1005 static struct proc *vnlruproc;
1006 static int vnlruproc_sig;
1011 struct mount *mp, *nmp;
1012 unsigned long ofreevnodes, onumvnodes;
1013 int done, force, reclaim_nc_src, trigger, usevnodes;
1015 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, vnlruproc,
1016 SHUTDOWN_PRI_FIRST);
1020 kproc_suspend_check(vnlruproc);
1021 mtx_lock(&vnode_free_list_mtx);
1023 * If numvnodes is too large (due to desiredvnodes being
1024 * adjusted using its sysctl, or emergency growth), first
1025 * try to reduce it by discarding from the free list.
1027 if (numvnodes > desiredvnodes && freevnodes > 0)
1028 vnlru_free(ulmin(numvnodes - desiredvnodes,
1031 * Sleep if the vnode cache is in a good state. This is
1032 * when it is not over-full and has space for about a 4%
1033 * or 9% expansion (by growing its size or inexcessively
1034 * reducing its free list). Otherwise, try to reclaim
1035 * space for a 10% expansion.
1037 if (vstir && force == 0) {
1041 if (vspace() >= vlowat && force == 0) {
1043 wakeup(&vnlruproc_sig);
1044 msleep(vnlruproc, &vnode_free_list_mtx,
1045 PVFS|PDROP, "vlruwt", hz);
1048 mtx_unlock(&vnode_free_list_mtx);
1050 ofreevnodes = freevnodes;
1051 onumvnodes = numvnodes;
1053 * Calculate parameters for recycling. These are the same
1054 * throughout the loop to give some semblance of fairness.
1055 * The trigger point is to avoid recycling vnodes with lots
1056 * of resident pages. We aren't trying to free memory; we
1057 * are trying to recycle or at least free vnodes.
1059 if (numvnodes <= desiredvnodes)
1060 usevnodes = numvnodes - freevnodes;
1062 usevnodes = numvnodes;
1066 * The trigger value is is chosen to give a conservatively
1067 * large value to ensure that it alone doesn't prevent
1068 * making progress. The value can easily be so large that
1069 * it is effectively infinite in some congested and
1070 * misconfigured cases, and this is necessary. Normally
1071 * it is about 8 to 100 (pages), which is quite large.
1073 trigger = vm_cnt.v_page_count * 2 / usevnodes;
1075 trigger = vsmalltrigger;
1076 reclaim_nc_src = force >= 3;
1077 mtx_lock(&mountlist_mtx);
1078 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
1079 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
1080 nmp = TAILQ_NEXT(mp, mnt_list);
1083 done += vlrureclaim(mp, reclaim_nc_src, trigger);
1084 mtx_lock(&mountlist_mtx);
1085 nmp = TAILQ_NEXT(mp, mnt_list);
1088 mtx_unlock(&mountlist_mtx);
1089 if (onumvnodes > desiredvnodes && numvnodes <= desiredvnodes)
1092 if (force == 0 || force == 1) {
1102 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
1104 kern_yield(PRI_USER);
1106 * After becoming active to expand above low water, keep
1107 * active until above high water.
1109 force = vspace() < vhiwat;
1113 static struct kproc_desc vnlru_kp = {
1118 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
1122 * Routines having to do with the management of the vnode table.
1126 * Try to recycle a freed vnode. We abort if anyone picks up a reference
1127 * before we actually vgone(). This function must be called with the vnode
1128 * held to prevent the vnode from being returned to the free list midway
1132 vtryrecycle(struct vnode *vp)
1136 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
1137 VNASSERT(vp->v_holdcnt, vp,
1138 ("vtryrecycle: Recycling vp %p without a reference.", vp));
1140 * This vnode may found and locked via some other list, if so we
1141 * can't recycle it yet.
1143 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
1145 "%s: impossible to recycle, vp %p lock is already held",
1147 return (EWOULDBLOCK);
1150 * Don't recycle if its filesystem is being suspended.
1152 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
1155 "%s: impossible to recycle, cannot start the write for %p",
1160 * If we got this far, we need to acquire the interlock and see if
1161 * anyone picked up this vnode from another list. If not, we will
1162 * mark it with DOOMED via vgonel() so that anyone who does find it
1163 * will skip over it.
1166 if (vp->v_usecount) {
1167 VOP_UNLOCK(vp, LK_INTERLOCK);
1168 vn_finished_write(vnmp);
1170 "%s: impossible to recycle, %p is already referenced",
1174 if ((vp->v_iflag & VI_DOOMED) == 0) {
1175 atomic_add_long(&recycles_count, 1);
1178 VOP_UNLOCK(vp, LK_INTERLOCK);
1179 vn_finished_write(vnmp);
1187 if (vspace() < vlowat && vnlruproc_sig == 0) {
1194 * Wait if necessary for space for a new vnode.
1197 getnewvnode_wait(int suspended)
1200 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
1201 if (numvnodes >= desiredvnodes) {
1204 * The file system is being suspended. We cannot
1205 * risk a deadlock here, so allow allocation of
1206 * another vnode even if this would give too many.
1210 if (vnlruproc_sig == 0) {
1211 vnlruproc_sig = 1; /* avoid unnecessary wakeups */
1214 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
1217 /* Post-adjust like the pre-adjust in getnewvnode(). */
1218 if (numvnodes + 1 > desiredvnodes && freevnodes > 1)
1220 return (numvnodes >= desiredvnodes ? ENFILE : 0);
1224 * This hack is fragile, and probably not needed any more now that the
1225 * watermark handling works.
1228 getnewvnode_reserve(u_int count)
1232 /* Pre-adjust like the pre-adjust in getnewvnode(), with any count. */
1233 /* XXX no longer so quick, but this part is not racy. */
1234 mtx_lock(&vnode_free_list_mtx);
1235 if (numvnodes + count > desiredvnodes && freevnodes > wantfreevnodes)
1236 vnlru_free(ulmin(numvnodes + count - desiredvnodes,
1237 freevnodes - wantfreevnodes));
1238 mtx_unlock(&vnode_free_list_mtx);
1241 /* First try to be quick and racy. */
1242 if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
1243 td->td_vp_reserv += count;
1244 vcheckspace(); /* XXX no longer so quick, but more racy */
1247 atomic_subtract_long(&numvnodes, count);
1249 mtx_lock(&vnode_free_list_mtx);
1251 if (getnewvnode_wait(0) == 0) {
1254 atomic_add_long(&numvnodes, 1);
1258 mtx_unlock(&vnode_free_list_mtx);
1262 * This hack is fragile, especially if desiredvnodes or wantvnodes are
1263 * misconfgured or changed significantly. Reducing desiredvnodes below
1264 * the reserved amount should cause bizarre behaviour like reducing it
1265 * below the number of active vnodes -- the system will try to reduce
1266 * numvnodes to match, but should fail, so the subtraction below should
1270 getnewvnode_drop_reserve(void)
1275 atomic_subtract_long(&numvnodes, td->td_vp_reserv);
1276 td->td_vp_reserv = 0;
1280 * Return the next vnode from the free list.
1283 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
1288 struct lock_object *lo;
1289 static int cyclecount;
1292 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
1295 if (td->td_vp_reserv > 0) {
1296 td->td_vp_reserv -= 1;
1299 mtx_lock(&vnode_free_list_mtx);
1300 if (numvnodes < desiredvnodes)
1302 else if (cyclecount++ >= freevnodes) {
1307 * Grow the vnode cache if it will not be above its target max
1308 * after growing. Otherwise, if the free list is nonempty, try
1309 * to reclaim 1 item from it before growing the cache (possibly
1310 * above its target max if the reclamation failed or is delayed).
1311 * Otherwise, wait for some space. In all cases, schedule
1312 * vnlru_proc() if we are getting short of space. The watermarks
1313 * should be chosen so that we never wait or even reclaim from
1314 * the free list to below its target minimum.
1316 if (numvnodes + 1 <= desiredvnodes)
1318 else if (freevnodes > 0)
1321 error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
1323 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
1325 mtx_unlock(&vnode_free_list_mtx);
1331 atomic_add_long(&numvnodes, 1);
1332 mtx_unlock(&vnode_free_list_mtx);
1334 atomic_add_long(&vnodes_created, 1);
1335 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK);
1337 * Locks are given the generic name "vnode" when created.
1338 * Follow the historic practice of using the filesystem
1339 * name when they allocated, e.g., "zfs", "ufs", "nfs, etc.
1341 * Locks live in a witness group keyed on their name. Thus,
1342 * when a lock is renamed, it must also move from the witness
1343 * group of its old name to the witness group of its new name.
1345 * The change only needs to be made when the vnode moves
1346 * from one filesystem type to another. We ensure that each
1347 * filesystem use a single static name pointer for its tag so
1348 * that we can compare pointers rather than doing a strcmp().
1350 lo = &vp->v_vnlock->lock_object;
1351 if (lo->lo_name != tag) {
1353 WITNESS_DESTROY(lo);
1354 WITNESS_INIT(lo, tag);
1357 * By default, don't allow shared locks unless filesystems opt-in.
1359 vp->v_vnlock->lock_object.lo_flags |= LK_NOSHARE;
1361 * Finalize various vnode identity bits.
1363 KASSERT(vp->v_object == NULL, ("stale v_object %p", vp));
1364 KASSERT(vp->v_lockf == NULL, ("stale v_lockf %p", vp));
1365 KASSERT(vp->v_pollinfo == NULL, ("stale v_pollinfo %p", vp));
1369 v_init_counters(vp);
1370 vp->v_bufobj.bo_ops = &buf_ops_bio;
1373 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
1374 mac_vnode_associate_singlelabel(mp, vp);
1375 else if (mp == NULL && vops != &dead_vnodeops)
1376 printf("NULL mp in getnewvnode()\n");
1379 vp->v_bufobj.bo_bsize = mp->mnt_stat.f_iosize;
1380 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
1381 vp->v_vflag |= VV_NOKNOTE;
1385 * For the filesystems which do not use vfs_hash_insert(),
1386 * still initialize v_hash to have vfs_hash_index() useful.
1387 * E.g., nullfs uses vfs_hash_index() on the lower vnode for
1390 vp->v_hash = (uintptr_t)vp >> vnsz2log;
1397 * Delete from old mount point vnode list, if on one.
1400 delmntque(struct vnode *vp)
1410 KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
1411 ("Active vnode list size %d > Vnode list size %d",
1412 mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
1413 active = vp->v_iflag & VI_ACTIVE;
1414 vp->v_iflag &= ~VI_ACTIVE;
1416 mtx_lock(&vnode_free_list_mtx);
1417 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
1418 mp->mnt_activevnodelistsize--;
1419 mtx_unlock(&vnode_free_list_mtx);
1423 VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
1424 ("bad mount point vnode list size"));
1425 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1426 mp->mnt_nvnodelistsize--;
1432 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
1436 vp->v_op = &dead_vnodeops;
1442 * Insert into list of vnodes for the new mount point, if available.
1445 insmntque1(struct vnode *vp, struct mount *mp,
1446 void (*dtr)(struct vnode *, void *), void *dtr_arg)
1449 KASSERT(vp->v_mount == NULL,
1450 ("insmntque: vnode already on per mount vnode list"));
1451 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
1452 ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
1455 * We acquire the vnode interlock early to ensure that the
1456 * vnode cannot be recycled by another process releasing a
1457 * holdcnt on it before we get it on both the vnode list
1458 * and the active vnode list. The mount mutex protects only
1459 * manipulation of the vnode list and the vnode freelist
1460 * mutex protects only manipulation of the active vnode list.
1461 * Hence the need to hold the vnode interlock throughout.
1465 if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
1466 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
1467 mp->mnt_nvnodelistsize == 0)) &&
1468 (vp->v_vflag & VV_FORCEINSMQ) == 0) {
1477 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
1478 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
1479 ("neg mount point vnode list size"));
1480 mp->mnt_nvnodelistsize++;
1481 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
1482 ("Activating already active vnode"));
1483 vp->v_iflag |= VI_ACTIVE;
1484 mtx_lock(&vnode_free_list_mtx);
1485 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
1486 mp->mnt_activevnodelistsize++;
1487 mtx_unlock(&vnode_free_list_mtx);
1494 insmntque(struct vnode *vp, struct mount *mp)
1497 return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1501 * Flush out and invalidate all buffers associated with a bufobj
1502 * Called with the underlying object locked.
1505 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1510 if (flags & V_SAVE) {
1511 error = bufobj_wwait(bo, slpflag, slptimeo);
1516 if (bo->bo_dirty.bv_cnt > 0) {
1518 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
1521 * XXX We could save a lock/unlock if this was only
1522 * enabled under INVARIANTS
1525 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
1526 panic("vinvalbuf: dirty bufs");
1530 * If you alter this loop please notice that interlock is dropped and
1531 * reacquired in flushbuflist. Special care is needed to ensure that
1532 * no race conditions occur from this.
1535 error = flushbuflist(&bo->bo_clean,
1536 flags, bo, slpflag, slptimeo);
1537 if (error == 0 && !(flags & V_CLEANONLY))
1538 error = flushbuflist(&bo->bo_dirty,
1539 flags, bo, slpflag, slptimeo);
1540 if (error != 0 && error != EAGAIN) {
1544 } while (error != 0);
1547 * Wait for I/O to complete. XXX needs cleaning up. The vnode can
1548 * have write I/O in-progress but if there is a VM object then the
1549 * VM object can also have read-I/O in-progress.
1552 bufobj_wwait(bo, 0, 0);
1554 if (bo->bo_object != NULL) {
1555 VM_OBJECT_WLOCK(bo->bo_object);
1556 vm_object_pip_wait(bo->bo_object, "bovlbx");
1557 VM_OBJECT_WUNLOCK(bo->bo_object);
1560 } while (bo->bo_numoutput > 0);
1564 * Destroy the copy in the VM cache, too.
1566 if (bo->bo_object != NULL &&
1567 (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
1568 VM_OBJECT_WLOCK(bo->bo_object);
1569 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
1570 OBJPR_CLEANONLY : 0);
1571 VM_OBJECT_WUNLOCK(bo->bo_object);
1576 if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
1577 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1578 panic("vinvalbuf: flush failed");
1585 * Flush out and invalidate all buffers associated with a vnode.
1586 * Called with the underlying object locked.
1589 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
1592 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
1593 ASSERT_VOP_LOCKED(vp, "vinvalbuf");
1594 if (vp->v_object != NULL && vp->v_object->handle != vp)
1596 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
1600 * Flush out buffers on the specified list.
1604 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
1607 struct buf *bp, *nbp;
1612 ASSERT_BO_WLOCKED(bo);
1615 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
1616 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
1617 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
1623 lblkno = nbp->b_lblkno;
1624 xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
1627 error = BUF_TIMELOCK(bp,
1628 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo),
1629 "flushbuf", slpflag, slptimeo);
1632 return (error != ENOLCK ? error : EAGAIN);
1634 KASSERT(bp->b_bufobj == bo,
1635 ("bp %p wrong b_bufobj %p should be %p",
1636 bp, bp->b_bufobj, bo));
1638 * XXX Since there are no node locks for NFS, I
1639 * believe there is a slight chance that a delayed
1640 * write will occur while sleeping just above, so
1643 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
1646 bp->b_flags |= B_ASYNC;
1649 return (EAGAIN); /* XXX: why not loop ? */
1652 bp->b_flags |= (B_INVAL | B_RELBUF);
1653 bp->b_flags &= ~B_ASYNC;
1656 nbp = gbincore(bo, lblkno);
1657 if (nbp == NULL || (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1659 break; /* nbp invalid */
1665 bnoreuselist(struct bufv *bufv, struct bufobj *bo, daddr_t startn, daddr_t endn)
1671 ASSERT_BO_LOCKED(bo);
1673 for (lblkno = startn;;) {
1675 bp = BUF_PCTRIE_LOOKUP_GE(&bufv->bv_root, lblkno);
1676 if (bp == NULL || bp->b_lblkno >= endn)
1678 error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1679 LK_INTERLOCK, BO_LOCKPTR(bo), "brlsfl", 0, 0);
1682 if (error == ENOLCK)
1686 KASSERT(bp->b_bufobj == bo,
1687 ("bp %p wrong b_bufobj %p should be %p",
1688 bp, bp->b_bufobj, bo));
1689 lblkno = bp->b_lblkno + 1;
1690 if ((bp->b_flags & B_MANAGED) == 0)
1692 bp->b_flags |= B_RELBUF;
1694 * In the VMIO case, use the B_NOREUSE flag to hint that the
1695 * pages backing each buffer in the range are unlikely to be
1696 * reused. Dirty buffers will have the hint applied once
1697 * they've been written.
1699 if (bp->b_vp->v_object != NULL)
1700 bp->b_flags |= B_NOREUSE;
1708 * Truncate a file's buffer and pages to a specified length. This
1709 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
1713 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
1715 struct buf *bp, *nbp;
1720 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
1721 vp, cred, blksize, (uintmax_t)length);
1724 * Round up to the *next* lbn.
1726 trunclbn = (length + blksize - 1) / blksize;
1728 ASSERT_VOP_LOCKED(vp, "vtruncbuf");
1735 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
1736 if (bp->b_lblkno < trunclbn)
1739 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1740 BO_LOCKPTR(bo)) == ENOLCK)
1744 bp->b_flags |= (B_INVAL | B_RELBUF);
1745 bp->b_flags &= ~B_ASYNC;
1751 (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
1752 (nbp->b_vp != vp) ||
1753 (nbp->b_flags & B_DELWRI))) {
1759 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1760 if (bp->b_lblkno < trunclbn)
1763 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1764 BO_LOCKPTR(bo)) == ENOLCK)
1767 bp->b_flags |= (B_INVAL | B_RELBUF);
1768 bp->b_flags &= ~B_ASYNC;
1774 (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
1775 (nbp->b_vp != vp) ||
1776 (nbp->b_flags & B_DELWRI) == 0)) {
1785 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
1786 if (bp->b_lblkno > 0)
1789 * Since we hold the vnode lock this should only
1790 * fail if we're racing with the buf daemon.
1793 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
1794 BO_LOCKPTR(bo)) == ENOLCK) {
1797 VNASSERT((bp->b_flags & B_DELWRI), vp,
1798 ("buf(%p) on dirty queue without DELWRI", bp));
1807 bufobj_wwait(bo, 0, 0);
1809 vnode_pager_setsize(vp, length);
1815 buf_vlist_remove(struct buf *bp)
1819 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
1820 ASSERT_BO_WLOCKED(bp->b_bufobj);
1821 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
1822 (BX_VNDIRTY|BX_VNCLEAN),
1823 ("buf_vlist_remove: Buf %p is on two lists", bp));
1824 if (bp->b_xflags & BX_VNDIRTY)
1825 bv = &bp->b_bufobj->bo_dirty;
1827 bv = &bp->b_bufobj->bo_clean;
1828 BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
1829 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
1831 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
1835 * Add the buffer to the sorted clean or dirty block list.
1837 * NOTE: xflags is passed as a constant, optimizing this inline function!
1840 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
1846 ASSERT_BO_WLOCKED(bo);
1847 KASSERT((xflags & BX_VNDIRTY) == 0 || (bo->bo_flag & BO_DEAD) == 0,
1848 ("dead bo %p", bo));
1849 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
1850 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
1851 bp->b_xflags |= xflags;
1852 if (xflags & BX_VNDIRTY)
1858 * Keep the list ordered. Optimize empty list insertion. Assume
1859 * we tend to grow at the tail so lookup_le should usually be cheaper
1862 if (bv->bv_cnt == 0 ||
1863 bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
1864 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
1865 else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
1866 TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
1868 TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
1869 error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
1871 panic("buf_vlist_add: Preallocated nodes insufficient.");
1876 * Look up a buffer using the buffer tries.
1879 gbincore(struct bufobj *bo, daddr_t lblkno)
1883 ASSERT_BO_LOCKED(bo);
1884 bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
1887 return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
1891 * Associate a buffer with a vnode.
1894 bgetvp(struct vnode *vp, struct buf *bp)
1899 ASSERT_BO_WLOCKED(bo);
1900 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1902 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
1903 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
1904 ("bgetvp: bp already attached! %p", bp));
1910 * Insert onto list for new vnode.
1912 buf_vlist_add(bp, bo, BX_VNCLEAN);
1916 * Disassociate a buffer from a vnode.
1919 brelvp(struct buf *bp)
1924 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1925 KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1928 * Delete from old vnode list, if on one.
1930 vp = bp->b_vp; /* XXX */
1933 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
1934 buf_vlist_remove(bp);
1936 panic("brelvp: Buffer %p not on queue.", bp);
1937 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
1938 bo->bo_flag &= ~BO_ONWORKLST;
1939 mtx_lock(&sync_mtx);
1940 LIST_REMOVE(bo, bo_synclist);
1941 syncer_worklist_len--;
1942 mtx_unlock(&sync_mtx);
1945 bp->b_bufobj = NULL;
1951 * Add an item to the syncer work queue.
1954 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
1958 ASSERT_BO_WLOCKED(bo);
1960 mtx_lock(&sync_mtx);
1961 if (bo->bo_flag & BO_ONWORKLST)
1962 LIST_REMOVE(bo, bo_synclist);
1964 bo->bo_flag |= BO_ONWORKLST;
1965 syncer_worklist_len++;
1968 if (delay > syncer_maxdelay - 2)
1969 delay = syncer_maxdelay - 2;
1970 slot = (syncer_delayno + delay) & syncer_mask;
1972 LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
1973 mtx_unlock(&sync_mtx);
1977 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
1981 mtx_lock(&sync_mtx);
1982 len = syncer_worklist_len - sync_vnode_count;
1983 mtx_unlock(&sync_mtx);
1984 error = SYSCTL_OUT(req, &len, sizeof(len));
1988 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
1989 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
1991 static struct proc *updateproc;
1992 static void sched_sync(void);
1993 static struct kproc_desc up_kp = {
1998 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
2001 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
2006 *bo = LIST_FIRST(slp);
2009 vp = (*bo)->__bo_vnode; /* XXX */
2010 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
2013 * We use vhold in case the vnode does not
2014 * successfully sync. vhold prevents the vnode from
2015 * going away when we unlock the sync_mtx so that
2016 * we can acquire the vnode interlock.
2019 mtx_unlock(&sync_mtx);
2021 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
2023 mtx_lock(&sync_mtx);
2024 return (*bo == LIST_FIRST(slp));
2026 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2027 (void) VOP_FSYNC(vp, MNT_LAZY, td);
2029 vn_finished_write(mp);
2031 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
2033 * Put us back on the worklist. The worklist
2034 * routine will remove us from our current
2035 * position and then add us back in at a later
2038 vn_syncer_add_to_worklist(*bo, syncdelay);
2042 mtx_lock(&sync_mtx);
2046 static int first_printf = 1;
2049 * System filesystem synchronizer daemon.
2054 struct synclist *next, *slp;
2057 struct thread *td = curthread;
2059 int net_worklist_len;
2060 int syncer_final_iter;
2064 syncer_final_iter = 0;
2065 syncer_state = SYNCER_RUNNING;
2066 starttime = time_uptime;
2067 td->td_pflags |= TDP_NORUNNINGBUF;
2069 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
2072 mtx_lock(&sync_mtx);
2074 if (syncer_state == SYNCER_FINAL_DELAY &&
2075 syncer_final_iter == 0) {
2076 mtx_unlock(&sync_mtx);
2077 kproc_suspend_check(td->td_proc);
2078 mtx_lock(&sync_mtx);
2080 net_worklist_len = syncer_worklist_len - sync_vnode_count;
2081 if (syncer_state != SYNCER_RUNNING &&
2082 starttime != time_uptime) {
2084 printf("\nSyncing disks, vnodes remaining...");
2087 printf("%d ", net_worklist_len);
2089 starttime = time_uptime;
2092 * Push files whose dirty time has expired. Be careful
2093 * of interrupt race on slp queue.
2095 * Skip over empty worklist slots when shutting down.
2098 slp = &syncer_workitem_pending[syncer_delayno];
2099 syncer_delayno += 1;
2100 if (syncer_delayno == syncer_maxdelay)
2102 next = &syncer_workitem_pending[syncer_delayno];
2104 * If the worklist has wrapped since the
2105 * it was emptied of all but syncer vnodes,
2106 * switch to the FINAL_DELAY state and run
2107 * for one more second.
2109 if (syncer_state == SYNCER_SHUTTING_DOWN &&
2110 net_worklist_len == 0 &&
2111 last_work_seen == syncer_delayno) {
2112 syncer_state = SYNCER_FINAL_DELAY;
2113 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
2115 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
2116 syncer_worklist_len > 0);
2119 * Keep track of the last time there was anything
2120 * on the worklist other than syncer vnodes.
2121 * Return to the SHUTTING_DOWN state if any
2124 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
2125 last_work_seen = syncer_delayno;
2126 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
2127 syncer_state = SYNCER_SHUTTING_DOWN;
2128 while (!LIST_EMPTY(slp)) {
2129 error = sync_vnode(slp, &bo, td);
2131 LIST_REMOVE(bo, bo_synclist);
2132 LIST_INSERT_HEAD(next, bo, bo_synclist);
2136 if (first_printf == 0) {
2138 * Drop the sync mutex, because some watchdog
2139 * drivers need to sleep while patting
2141 mtx_unlock(&sync_mtx);
2142 wdog_kern_pat(WD_LASTVAL);
2143 mtx_lock(&sync_mtx);
2147 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
2148 syncer_final_iter--;
2150 * The variable rushjob allows the kernel to speed up the
2151 * processing of the filesystem syncer process. A rushjob
2152 * value of N tells the filesystem syncer to process the next
2153 * N seconds worth of work on its queue ASAP. Currently rushjob
2154 * is used by the soft update code to speed up the filesystem
2155 * syncer process when the incore state is getting so far
2156 * ahead of the disk that the kernel memory pool is being
2157 * threatened with exhaustion.
2164 * Just sleep for a short period of time between
2165 * iterations when shutting down to allow some I/O
2168 * If it has taken us less than a second to process the
2169 * current work, then wait. Otherwise start right over
2170 * again. We can still lose time if any single round
2171 * takes more than two seconds, but it does not really
2172 * matter as we are just trying to generally pace the
2173 * filesystem activity.
2175 if (syncer_state != SYNCER_RUNNING ||
2176 time_uptime == starttime) {
2178 sched_prio(td, PPAUSE);
2181 if (syncer_state != SYNCER_RUNNING)
2182 cv_timedwait(&sync_wakeup, &sync_mtx,
2183 hz / SYNCER_SHUTDOWN_SPEEDUP);
2184 else if (time_uptime == starttime)
2185 cv_timedwait(&sync_wakeup, &sync_mtx, hz);
2190 * Request the syncer daemon to speed up its work.
2191 * We never push it to speed up more than half of its
2192 * normal turn time, otherwise it could take over the cpu.
2195 speedup_syncer(void)
2199 mtx_lock(&sync_mtx);
2200 if (rushjob < syncdelay / 2) {
2202 stat_rush_requests += 1;
2205 mtx_unlock(&sync_mtx);
2206 cv_broadcast(&sync_wakeup);
2211 * Tell the syncer to speed up its work and run though its work
2212 * list several times, then tell it to shut down.
2215 syncer_shutdown(void *arg, int howto)
2218 if (howto & RB_NOSYNC)
2220 mtx_lock(&sync_mtx);
2221 syncer_state = SYNCER_SHUTTING_DOWN;
2223 mtx_unlock(&sync_mtx);
2224 cv_broadcast(&sync_wakeup);
2225 kproc_shutdown(arg, howto);
2229 syncer_suspend(void)
2232 syncer_shutdown(updateproc, 0);
2239 mtx_lock(&sync_mtx);
2241 syncer_state = SYNCER_RUNNING;
2242 mtx_unlock(&sync_mtx);
2243 cv_broadcast(&sync_wakeup);
2244 kproc_resume(updateproc);
2248 * Reassign a buffer from one vnode to another.
2249 * Used to assign file specific control information
2250 * (indirect blocks) to the vnode to which they belong.
2253 reassignbuf(struct buf *bp)
2266 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
2267 bp, bp->b_vp, bp->b_flags);
2269 * B_PAGING flagged buffers cannot be reassigned because their vp
2270 * is not fully linked in.
2272 if (bp->b_flags & B_PAGING)
2273 panic("cannot reassign paging buffer");
2276 * Delete from old vnode list, if on one.
2279 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
2280 buf_vlist_remove(bp);
2282 panic("reassignbuf: Buffer %p not on queue.", bp);
2284 * If dirty, put on list of dirty buffers; otherwise insert onto list
2287 if (bp->b_flags & B_DELWRI) {
2288 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
2289 switch (vp->v_type) {
2299 vn_syncer_add_to_worklist(bo, delay);
2301 buf_vlist_add(bp, bo, BX_VNDIRTY);
2303 buf_vlist_add(bp, bo, BX_VNCLEAN);
2305 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
2306 mtx_lock(&sync_mtx);
2307 LIST_REMOVE(bo, bo_synclist);
2308 syncer_worklist_len--;
2309 mtx_unlock(&sync_mtx);
2310 bo->bo_flag &= ~BO_ONWORKLST;
2315 bp = TAILQ_FIRST(&bv->bv_hd);
2316 KASSERT(bp == NULL || bp->b_bufobj == bo,
2317 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2318 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2319 KASSERT(bp == NULL || bp->b_bufobj == bo,
2320 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2322 bp = TAILQ_FIRST(&bv->bv_hd);
2323 KASSERT(bp == NULL || bp->b_bufobj == bo,
2324 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2325 bp = TAILQ_LAST(&bv->bv_hd, buflists);
2326 KASSERT(bp == NULL || bp->b_bufobj == bo,
2327 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
2333 * A temporary hack until refcount_* APIs are sorted out.
2336 vfs_refcount_acquire_if_not_zero(volatile u_int *count)
2344 if (atomic_cmpset_int(count, old, old + 1))
2350 vfs_refcount_release_if_not_last(volatile u_int *count)
2358 if (atomic_cmpset_int(count, old, old - 1))
2364 v_init_counters(struct vnode *vp)
2367 VNASSERT(vp->v_type == VNON && vp->v_data == NULL && vp->v_iflag == 0,
2368 vp, ("%s called for an initialized vnode", __FUNCTION__));
2369 ASSERT_VI_UNLOCKED(vp, __FUNCTION__);
2371 refcount_init(&vp->v_holdcnt, 1);
2372 refcount_init(&vp->v_usecount, 1);
2376 v_incr_usecount_locked(struct vnode *vp)
2379 ASSERT_VI_LOCKED(vp, __func__);
2380 if ((vp->v_iflag & VI_OWEINACT) != 0) {
2381 VNASSERT(vp->v_usecount == 0, vp,
2382 ("vnode with usecount and VI_OWEINACT set"));
2383 vp->v_iflag &= ~VI_OWEINACT;
2385 refcount_acquire(&vp->v_usecount);
2386 v_incr_devcount(vp);
2390 * Increment the use and hold counts on the vnode, taking care to reference
2391 * the driver's usecount if this is a chardev. The _vhold() will remove
2392 * the vnode from the free list if it is presently free.
2395 v_incr_usecount(struct vnode *vp)
2398 ASSERT_VI_UNLOCKED(vp, __func__);
2399 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2401 if (vp->v_type != VCHR &&
2402 vfs_refcount_acquire_if_not_zero(&vp->v_usecount)) {
2403 VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp,
2404 ("vnode with usecount and VI_OWEINACT set"));
2407 v_incr_usecount_locked(vp);
2413 * Increment si_usecount of the associated device, if any.
2416 v_incr_devcount(struct vnode *vp)
2419 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2420 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2422 vp->v_rdev->si_usecount++;
2428 * Decrement si_usecount of the associated device, if any.
2431 v_decr_devcount(struct vnode *vp)
2434 ASSERT_VI_LOCKED(vp, __FUNCTION__);
2435 if (vp->v_type == VCHR && vp->v_rdev != NULL) {
2437 vp->v_rdev->si_usecount--;
2443 * Grab a particular vnode from the free list, increment its
2444 * reference count and lock it. VI_DOOMED is set if the vnode
2445 * is being destroyed. Only callers who specify LK_RETRY will
2446 * see doomed vnodes. If inactive processing was delayed in
2447 * vput try to do it here.
2449 * Notes on lockless counter manipulation:
2450 * _vhold, vputx and other routines make various decisions based
2451 * on either holdcnt or usecount being 0. As long as either contuner
2452 * is not transitioning 0->1 nor 1->0, the manipulation can be done
2453 * with atomic operations. Otherwise the interlock is taken.
2456 vget(struct vnode *vp, int flags, struct thread *td)
2458 int error, oweinact;
2460 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
2461 ("vget: invalid lock operation"));
2463 if ((flags & LK_INTERLOCK) != 0)
2464 ASSERT_VI_LOCKED(vp, __func__);
2466 ASSERT_VI_UNLOCKED(vp, __func__);
2467 if ((flags & LK_VNHELD) != 0)
2468 VNASSERT((vp->v_holdcnt > 0), vp,
2469 ("vget: LK_VNHELD passed but vnode not held"));
2471 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
2473 if ((flags & LK_VNHELD) == 0)
2474 _vhold(vp, (flags & LK_INTERLOCK) != 0);
2476 if ((error = vn_lock(vp, flags)) != 0) {
2478 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
2482 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
2483 panic("vget: vn_lock failed to return ENOENT\n");
2485 * We don't guarantee that any particular close will
2486 * trigger inactive processing so just make a best effort
2487 * here at preventing a reference to a removed file. If
2488 * we don't succeed no harm is done.
2490 * Upgrade our holdcnt to a usecount.
2492 if (vp->v_type != VCHR &&
2493 vfs_refcount_acquire_if_not_zero(&vp->v_usecount)) {
2494 VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp,
2495 ("vnode with usecount and VI_OWEINACT set"));
2498 if ((vp->v_iflag & VI_OWEINACT) == 0) {
2502 vp->v_iflag &= ~VI_OWEINACT;
2504 refcount_acquire(&vp->v_usecount);
2505 v_incr_devcount(vp);
2506 if (oweinact && VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
2507 (flags & LK_NOWAIT) == 0)
2515 * Increase the reference count of a vnode.
2518 vref(struct vnode *vp)
2521 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2523 v_incr_usecount(vp);
2527 vrefl(struct vnode *vp)
2530 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2532 v_incr_usecount_locked(vp);
2536 * Return reference count of a vnode.
2538 * The results of this call are only guaranteed when some mechanism is used to
2539 * stop other processes from gaining references to the vnode. This may be the
2540 * case if the caller holds the only reference. This is also useful when stale
2541 * data is acceptable as race conditions may be accounted for by some other
2545 vrefcnt(struct vnode *vp)
2548 return (vp->v_usecount);
2551 #define VPUTX_VRELE 1
2552 #define VPUTX_VPUT 2
2553 #define VPUTX_VUNREF 3
2556 * Decrement the use and hold counts for a vnode.
2558 * See an explanation near vget() as to why atomic operation is safe.
2561 vputx(struct vnode *vp, int func)
2565 KASSERT(vp != NULL, ("vputx: null vp"));
2566 if (func == VPUTX_VUNREF)
2567 ASSERT_VOP_LOCKED(vp, "vunref");
2568 else if (func == VPUTX_VPUT)
2569 ASSERT_VOP_LOCKED(vp, "vput");
2571 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
2572 ASSERT_VI_UNLOCKED(vp, __func__);
2573 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2575 if (vp->v_type != VCHR &&
2576 vfs_refcount_release_if_not_last(&vp->v_usecount)) {
2577 if (func == VPUTX_VPUT)
2586 * We want to hold the vnode until the inactive finishes to
2587 * prevent vgone() races. We drop the use count here and the
2588 * hold count below when we're done.
2590 if (!refcount_release(&vp->v_usecount) ||
2591 (vp->v_iflag & VI_DOINGINACT)) {
2592 if (func == VPUTX_VPUT)
2594 v_decr_devcount(vp);
2599 v_decr_devcount(vp);
2603 if (vp->v_usecount != 0) {
2604 vprint("vputx: usecount not zero", vp);
2605 panic("vputx: usecount not zero");
2608 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
2611 * We must call VOP_INACTIVE with the node locked. Mark
2612 * as VI_DOINGINACT to avoid recursion.
2614 vp->v_iflag |= VI_OWEINACT;
2617 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
2621 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2622 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
2628 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
2629 error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
2634 VNASSERT(vp->v_usecount == 0 || (vp->v_iflag & VI_OWEINACT) == 0, vp,
2635 ("vnode with usecount and VI_OWEINACT set"));
2637 if (vp->v_iflag & VI_OWEINACT)
2638 vinactive(vp, curthread);
2639 if (func != VPUTX_VUNREF)
2646 * Vnode put/release.
2647 * If count drops to zero, call inactive routine and return to freelist.
2650 vrele(struct vnode *vp)
2653 vputx(vp, VPUTX_VRELE);
2657 * Release an already locked vnode. This give the same effects as
2658 * unlock+vrele(), but takes less time and avoids releasing and
2659 * re-aquiring the lock (as vrele() acquires the lock internally.)
2662 vput(struct vnode *vp)
2665 vputx(vp, VPUTX_VPUT);
2669 * Release an exclusively locked vnode. Do not unlock the vnode lock.
2672 vunref(struct vnode *vp)
2675 vputx(vp, VPUTX_VUNREF);
2679 * Increase the hold count and activate if this is the first reference.
2682 _vhold(struct vnode *vp, bool locked)
2687 ASSERT_VI_LOCKED(vp, __func__);
2689 ASSERT_VI_UNLOCKED(vp, __func__);
2690 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2691 if (!locked && vfs_refcount_acquire_if_not_zero(&vp->v_holdcnt)) {
2692 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2693 ("_vhold: vnode with holdcnt is free"));
2699 if ((vp->v_iflag & VI_FREE) == 0) {
2700 refcount_acquire(&vp->v_holdcnt);
2705 VNASSERT(vp->v_holdcnt == 0, vp,
2706 ("%s: wrong hold count", __func__));
2707 VNASSERT(vp->v_op != NULL, vp,
2708 ("%s: vnode already reclaimed.", __func__));
2710 * Remove a vnode from the free list, mark it as in use,
2711 * and put it on the active list.
2713 mtx_lock(&vnode_free_list_mtx);
2714 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
2716 vp->v_iflag &= ~VI_FREE;
2717 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
2718 ("Activating already active vnode"));
2719 vp->v_iflag |= VI_ACTIVE;
2721 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
2722 mp->mnt_activevnodelistsize++;
2723 mtx_unlock(&vnode_free_list_mtx);
2724 refcount_acquire(&vp->v_holdcnt);
2730 * Drop the hold count of the vnode. If this is the last reference to
2731 * the vnode we place it on the free list unless it has been vgone'd
2732 * (marked VI_DOOMED) in which case we will free it.
2734 * Because the vnode vm object keeps a hold reference on the vnode if
2735 * there is at least one resident non-cached page, the vnode cannot
2736 * leave the active list without the page cleanup done.
2739 _vdrop(struct vnode *vp, bool locked)
2746 ASSERT_VI_LOCKED(vp, __func__);
2748 ASSERT_VI_UNLOCKED(vp, __func__);
2749 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2750 if ((int)vp->v_holdcnt <= 0)
2751 panic("vdrop: holdcnt %d", vp->v_holdcnt);
2752 if (vfs_refcount_release_if_not_last(&vp->v_holdcnt)) {
2760 if (refcount_release(&vp->v_holdcnt) == 0) {
2764 if ((vp->v_iflag & VI_DOOMED) == 0) {
2766 * Mark a vnode as free: remove it from its active list
2767 * and put it up for recycling on the freelist.
2769 VNASSERT(vp->v_op != NULL, vp,
2770 ("vdropl: vnode already reclaimed."));
2771 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2772 ("vnode already free"));
2773 VNASSERT(vp->v_holdcnt == 0, vp,
2774 ("vdropl: freeing when we shouldn't"));
2775 active = vp->v_iflag & VI_ACTIVE;
2776 if ((vp->v_iflag & VI_OWEINACT) == 0) {
2777 vp->v_iflag &= ~VI_ACTIVE;
2779 mtx_lock(&vnode_free_list_mtx);
2781 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
2783 mp->mnt_activevnodelistsize--;
2785 TAILQ_INSERT_TAIL(&vnode_free_list, vp,
2788 vp->v_iflag |= VI_FREE;
2789 mtx_unlock(&vnode_free_list_mtx);
2791 atomic_add_long(&free_owe_inact, 1);
2797 * The vnode has been marked for destruction, so free it.
2799 * The vnode will be returned to the zone where it will
2800 * normally remain until it is needed for another vnode. We
2801 * need to cleanup (or verify that the cleanup has already
2802 * been done) any residual data left from its current use
2803 * so as not to contaminate the freshly allocated vnode.
2805 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
2806 atomic_subtract_long(&numvnodes, 1);
2808 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
2809 ("cleaned vnode still on the free list."));
2810 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
2811 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
2812 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
2813 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
2814 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
2815 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
2816 VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
2817 ("clean blk trie not empty"));
2818 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
2819 VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
2820 ("dirty blk trie not empty"));
2821 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
2822 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
2823 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
2824 VNASSERT(TAILQ_EMPTY(&vp->v_rl.rl_waiters), vp,
2825 ("Dangling rangelock waiters"));
2828 mac_vnode_destroy(vp);
2830 if (vp->v_pollinfo != NULL) {
2831 destroy_vpollinfo(vp->v_pollinfo);
2832 vp->v_pollinfo = NULL;
2835 /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
2838 bzero(&vp->v_un, sizeof(vp->v_un));
2839 vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
2843 uma_zfree(vnode_zone, vp);
2847 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
2848 * flags. DOINGINACT prevents us from recursing in calls to vinactive.
2849 * OWEINACT tracks whether a vnode missed a call to inactive due to a
2850 * failed lock upgrade.
2853 vinactive(struct vnode *vp, struct thread *td)
2855 struct vm_object *obj;
2857 ASSERT_VOP_ELOCKED(vp, "vinactive");
2858 ASSERT_VI_LOCKED(vp, "vinactive");
2859 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
2860 ("vinactive: recursed on VI_DOINGINACT"));
2861 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
2862 vp->v_iflag |= VI_DOINGINACT;
2863 vp->v_iflag &= ~VI_OWEINACT;
2866 * Before moving off the active list, we must be sure that any
2867 * modified pages are converted into the vnode's dirty
2868 * buffers, since these will no longer be checked once the
2869 * vnode is on the inactive list.
2871 * The write-out of the dirty pages is asynchronous. At the
2872 * point that VOP_INACTIVE() is called, there could still be
2873 * pending I/O and dirty pages in the object.
2876 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
2877 VM_OBJECT_WLOCK(obj);
2878 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
2879 VM_OBJECT_WUNLOCK(obj);
2881 VOP_INACTIVE(vp, td);
2883 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
2884 ("vinactive: lost VI_DOINGINACT"));
2885 vp->v_iflag &= ~VI_DOINGINACT;
2889 * Remove any vnodes in the vnode table belonging to mount point mp.
2891 * If FORCECLOSE is not specified, there should not be any active ones,
2892 * return error if any are found (nb: this is a user error, not a
2893 * system error). If FORCECLOSE is specified, detach any active vnodes
2896 * If WRITECLOSE is set, only flush out regular file vnodes open for
2899 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
2901 * `rootrefs' specifies the base reference count for the root vnode
2902 * of this filesystem. The root vnode is considered busy if its
2903 * v_usecount exceeds this value. On a successful return, vflush(, td)
2904 * will call vrele() on the root vnode exactly rootrefs times.
2905 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
2909 static int busyprt = 0; /* print out busy vnodes */
2910 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
2914 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
2916 struct vnode *vp, *mvp, *rootvp = NULL;
2918 int busy = 0, error;
2920 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
2923 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
2924 ("vflush: bad args"));
2926 * Get the filesystem root vnode. We can vput() it
2927 * immediately, since with rootrefs > 0, it won't go away.
2929 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
2930 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
2937 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
2939 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
2942 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2946 * Skip over a vnodes marked VV_SYSTEM.
2948 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
2954 * If WRITECLOSE is set, flush out unlinked but still open
2955 * files (even if open only for reading) and regular file
2956 * vnodes open for writing.
2958 if (flags & WRITECLOSE) {
2959 if (vp->v_object != NULL) {
2960 VM_OBJECT_WLOCK(vp->v_object);
2961 vm_object_page_clean(vp->v_object, 0, 0, 0);
2962 VM_OBJECT_WUNLOCK(vp->v_object);
2964 error = VOP_FSYNC(vp, MNT_WAIT, td);
2968 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
2971 error = VOP_GETATTR(vp, &vattr, td->td_ucred);
2974 if ((vp->v_type == VNON ||
2975 (error == 0 && vattr.va_nlink > 0)) &&
2976 (vp->v_writecount == 0 || vp->v_type != VREG)) {
2984 * With v_usecount == 0, all we need to do is clear out the
2985 * vnode data structures and we are done.
2987 * If FORCECLOSE is set, forcibly close the vnode.
2989 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
2995 vprint("vflush: busy vnode", vp);
3001 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
3003 * If just the root vnode is busy, and if its refcount
3004 * is equal to `rootrefs', then go ahead and kill it.
3007 KASSERT(busy > 0, ("vflush: not busy"));
3008 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
3009 ("vflush: usecount %d < rootrefs %d",
3010 rootvp->v_usecount, rootrefs));
3011 if (busy == 1 && rootvp->v_usecount == rootrefs) {
3012 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
3014 VOP_UNLOCK(rootvp, 0);
3020 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
3024 for (; rootrefs > 0; rootrefs--)
3030 * Recycle an unused vnode to the front of the free list.
3033 vrecycle(struct vnode *vp)
3037 ASSERT_VOP_ELOCKED(vp, "vrecycle");
3038 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3041 if (vp->v_usecount == 0) {
3050 * Eliminate all activity associated with a vnode
3051 * in preparation for reuse.
3054 vgone(struct vnode *vp)
3062 notify_lowervp_vfs_dummy(struct mount *mp __unused,
3063 struct vnode *lowervp __unused)
3068 * Notify upper mounts about reclaimed or unlinked vnode.
3071 vfs_notify_upper(struct vnode *vp, int event)
3073 static struct vfsops vgonel_vfsops = {
3074 .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
3075 .vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
3077 struct mount *mp, *ump, *mmp;
3084 if (TAILQ_EMPTY(&mp->mnt_uppers))
3087 mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
3088 mmp->mnt_op = &vgonel_vfsops;
3089 mmp->mnt_kern_flag |= MNTK_MARKER;
3091 mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
3092 for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
3093 if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
3094 ump = TAILQ_NEXT(ump, mnt_upper_link);
3097 TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
3100 case VFS_NOTIFY_UPPER_RECLAIM:
3101 VFS_RECLAIM_LOWERVP(ump, vp);
3103 case VFS_NOTIFY_UPPER_UNLINK:
3104 VFS_UNLINK_LOWERVP(ump, vp);
3107 KASSERT(0, ("invalid event %d", event));
3111 ump = TAILQ_NEXT(mmp, mnt_upper_link);
3112 TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
3115 mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
3116 if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
3117 mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
3118 wakeup(&mp->mnt_uppers);
3125 * vgone, with the vp interlock held.
3128 vgonel(struct vnode *vp)
3135 ASSERT_VOP_ELOCKED(vp, "vgonel");
3136 ASSERT_VI_LOCKED(vp, "vgonel");
3137 VNASSERT(vp->v_holdcnt, vp,
3138 ("vgonel: vp %p has no reference.", vp));
3139 CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
3143 * Don't vgonel if we're already doomed.
3145 if (vp->v_iflag & VI_DOOMED)
3147 vp->v_iflag |= VI_DOOMED;
3150 * Check to see if the vnode is in use. If so, we have to call
3151 * VOP_CLOSE() and VOP_INACTIVE().
3153 active = vp->v_usecount;
3154 oweinact = (vp->v_iflag & VI_OWEINACT);
3156 vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
3159 * If purging an active vnode, it must be closed and
3160 * deactivated before being reclaimed.
3163 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
3164 if (oweinact || active) {
3166 if ((vp->v_iflag & VI_DOINGINACT) == 0)
3170 if (vp->v_type == VSOCK)
3171 vfs_unp_reclaim(vp);
3174 * Clean out any buffers associated with the vnode.
3175 * If the flush fails, just toss the buffers.
3178 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
3179 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
3180 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) {
3181 while (vinvalbuf(vp, 0, 0, 0) != 0)
3185 BO_LOCK(&vp->v_bufobj);
3186 KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) &&
3187 vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
3188 TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) &&
3189 vp->v_bufobj.bo_clean.bv_cnt == 0,
3190 ("vp %p bufobj not invalidated", vp));
3191 vp->v_bufobj.bo_flag |= BO_DEAD;
3192 BO_UNLOCK(&vp->v_bufobj);
3195 * Reclaim the vnode.
3197 if (VOP_RECLAIM(vp, td))
3198 panic("vgone: cannot reclaim");
3200 vn_finished_secondary_write(mp);
3201 VNASSERT(vp->v_object == NULL, vp,
3202 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
3204 * Clear the advisory locks and wake up waiting threads.
3206 (void)VOP_ADVLOCKPURGE(vp);
3209 * Delete from old mount point vnode list.
3214 * Done with purge, reset to the standard lock and invalidate
3218 vp->v_vnlock = &vp->v_lock;
3219 vp->v_op = &dead_vnodeops;
3225 * Calculate the total number of references to a special device.
3228 vcount(struct vnode *vp)
3233 count = vp->v_rdev->si_usecount;
3239 * Same as above, but using the struct cdev *as argument
3242 count_dev(struct cdev *dev)
3247 count = dev->si_usecount;
3253 * Print out a description of a vnode.
3255 static char *typename[] =
3256 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
3260 vn_printf(struct vnode *vp, const char *fmt, ...)
3263 char buf[256], buf2[16];
3269 printf("%p: ", (void *)vp);
3270 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
3271 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
3272 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
3275 if (vp->v_vflag & VV_ROOT)
3276 strlcat(buf, "|VV_ROOT", sizeof(buf));
3277 if (vp->v_vflag & VV_ISTTY)
3278 strlcat(buf, "|VV_ISTTY", sizeof(buf));
3279 if (vp->v_vflag & VV_NOSYNC)
3280 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
3281 if (vp->v_vflag & VV_ETERNALDEV)
3282 strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
3283 if (vp->v_vflag & VV_CACHEDLABEL)
3284 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
3285 if (vp->v_vflag & VV_TEXT)
3286 strlcat(buf, "|VV_TEXT", sizeof(buf));
3287 if (vp->v_vflag & VV_COPYONWRITE)
3288 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
3289 if (vp->v_vflag & VV_SYSTEM)
3290 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
3291 if (vp->v_vflag & VV_PROCDEP)
3292 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
3293 if (vp->v_vflag & VV_NOKNOTE)
3294 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
3295 if (vp->v_vflag & VV_DELETED)
3296 strlcat(buf, "|VV_DELETED", sizeof(buf));
3297 if (vp->v_vflag & VV_MD)
3298 strlcat(buf, "|VV_MD", sizeof(buf));
3299 if (vp->v_vflag & VV_FORCEINSMQ)
3300 strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
3301 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
3302 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
3303 VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
3305 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
3306 strlcat(buf, buf2, sizeof(buf));
3308 if (vp->v_iflag & VI_MOUNT)
3309 strlcat(buf, "|VI_MOUNT", sizeof(buf));
3310 if (vp->v_iflag & VI_DOOMED)
3311 strlcat(buf, "|VI_DOOMED", sizeof(buf));
3312 if (vp->v_iflag & VI_FREE)
3313 strlcat(buf, "|VI_FREE", sizeof(buf));
3314 if (vp->v_iflag & VI_ACTIVE)
3315 strlcat(buf, "|VI_ACTIVE", sizeof(buf));
3316 if (vp->v_iflag & VI_DOINGINACT)
3317 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
3318 if (vp->v_iflag & VI_OWEINACT)
3319 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
3320 flags = vp->v_iflag & ~(VI_MOUNT | VI_DOOMED | VI_FREE |
3321 VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
3323 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
3324 strlcat(buf, buf2, sizeof(buf));
3326 printf(" flags (%s)\n", buf + 1);
3327 if (mtx_owned(VI_MTX(vp)))
3328 printf(" VI_LOCKed");
3329 if (vp->v_object != NULL)
3330 printf(" v_object %p ref %d pages %d "
3331 "cleanbuf %d dirtybuf %d\n",
3332 vp->v_object, vp->v_object->ref_count,
3333 vp->v_object->resident_page_count,
3334 vp->v_bufobj.bo_clean.bv_cnt,
3335 vp->v_bufobj.bo_dirty.bv_cnt);
3337 lockmgr_printinfo(vp->v_vnlock);
3338 if (vp->v_data != NULL)
3344 * List all of the locked vnodes in the system.
3345 * Called when debugging the kernel.
3347 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
3353 * Note: because this is DDB, we can't obey the locking semantics
3354 * for these structures, which means we could catch an inconsistent
3355 * state and dereference a nasty pointer. Not much to be done
3358 db_printf("Locked vnodes\n");
3359 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3360 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3361 if (vp->v_type != VMARKER && VOP_ISLOCKED(vp))
3368 * Show details about the given vnode.
3370 DB_SHOW_COMMAND(vnode, db_show_vnode)
3376 vp = (struct vnode *)addr;
3377 vn_printf(vp, "vnode ");
3381 * Show details about the given mount point.
3383 DB_SHOW_COMMAND(mount, db_show_mount)
3394 /* No address given, print short info about all mount points. */
3395 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3396 db_printf("%p %s on %s (%s)\n", mp,
3397 mp->mnt_stat.f_mntfromname,
3398 mp->mnt_stat.f_mntonname,
3399 mp->mnt_stat.f_fstypename);
3403 db_printf("\nMore info: show mount <addr>\n");
3407 mp = (struct mount *)addr;
3408 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
3409 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
3412 mflags = mp->mnt_flag;
3413 #define MNT_FLAG(flag) do { \
3414 if (mflags & (flag)) { \
3415 if (buf[0] != '\0') \
3416 strlcat(buf, ", ", sizeof(buf)); \
3417 strlcat(buf, (#flag) + 4, sizeof(buf)); \
3418 mflags &= ~(flag); \
3421 MNT_FLAG(MNT_RDONLY);
3422 MNT_FLAG(MNT_SYNCHRONOUS);
3423 MNT_FLAG(MNT_NOEXEC);
3424 MNT_FLAG(MNT_NOSUID);
3425 MNT_FLAG(MNT_NFS4ACLS);
3426 MNT_FLAG(MNT_UNION);
3427 MNT_FLAG(MNT_ASYNC);
3428 MNT_FLAG(MNT_SUIDDIR);
3429 MNT_FLAG(MNT_SOFTDEP);
3430 MNT_FLAG(MNT_NOSYMFOLLOW);
3431 MNT_FLAG(MNT_GJOURNAL);
3432 MNT_FLAG(MNT_MULTILABEL);
3434 MNT_FLAG(MNT_NOATIME);
3435 MNT_FLAG(MNT_NOCLUSTERR);
3436 MNT_FLAG(MNT_NOCLUSTERW);
3438 MNT_FLAG(MNT_EXRDONLY);
3439 MNT_FLAG(MNT_EXPORTED);
3440 MNT_FLAG(MNT_DEFEXPORTED);
3441 MNT_FLAG(MNT_EXPORTANON);
3442 MNT_FLAG(MNT_EXKERB);
3443 MNT_FLAG(MNT_EXPUBLIC);
3444 MNT_FLAG(MNT_LOCAL);
3445 MNT_FLAG(MNT_QUOTA);
3446 MNT_FLAG(MNT_ROOTFS);
3448 MNT_FLAG(MNT_IGNORE);
3449 MNT_FLAG(MNT_UPDATE);
3450 MNT_FLAG(MNT_DELEXPORT);
3451 MNT_FLAG(MNT_RELOAD);
3452 MNT_FLAG(MNT_FORCE);
3453 MNT_FLAG(MNT_SNAPSHOT);
3454 MNT_FLAG(MNT_BYFSID);
3458 strlcat(buf, ", ", sizeof(buf));
3459 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3460 "0x%016jx", mflags);
3462 db_printf(" mnt_flag = %s\n", buf);
3465 flags = mp->mnt_kern_flag;
3466 #define MNT_KERN_FLAG(flag) do { \
3467 if (flags & (flag)) { \
3468 if (buf[0] != '\0') \
3469 strlcat(buf, ", ", sizeof(buf)); \
3470 strlcat(buf, (#flag) + 5, sizeof(buf)); \
3474 MNT_KERN_FLAG(MNTK_UNMOUNTF);
3475 MNT_KERN_FLAG(MNTK_ASYNC);
3476 MNT_KERN_FLAG(MNTK_SOFTDEP);
3477 MNT_KERN_FLAG(MNTK_NOINSMNTQ);
3478 MNT_KERN_FLAG(MNTK_DRAINING);
3479 MNT_KERN_FLAG(MNTK_REFEXPIRE);
3480 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
3481 MNT_KERN_FLAG(MNTK_SHARED_WRITES);
3482 MNT_KERN_FLAG(MNTK_NO_IOPF);
3483 MNT_KERN_FLAG(MNTK_VGONE_UPPER);
3484 MNT_KERN_FLAG(MNTK_VGONE_WAITER);
3485 MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
3486 MNT_KERN_FLAG(MNTK_MARKER);
3487 MNT_KERN_FLAG(MNTK_USES_BCACHE);
3488 MNT_KERN_FLAG(MNTK_NOASYNC);
3489 MNT_KERN_FLAG(MNTK_UNMOUNT);
3490 MNT_KERN_FLAG(MNTK_MWAIT);
3491 MNT_KERN_FLAG(MNTK_SUSPEND);
3492 MNT_KERN_FLAG(MNTK_SUSPEND2);
3493 MNT_KERN_FLAG(MNTK_SUSPENDED);
3494 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
3495 MNT_KERN_FLAG(MNTK_NOKNOTE);
3496 #undef MNT_KERN_FLAG
3499 strlcat(buf, ", ", sizeof(buf));
3500 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
3503 db_printf(" mnt_kern_flag = %s\n", buf);
3505 db_printf(" mnt_opt = ");
3506 opt = TAILQ_FIRST(mp->mnt_opt);
3508 db_printf("%s", opt->name);
3509 opt = TAILQ_NEXT(opt, link);
3510 while (opt != NULL) {
3511 db_printf(", %s", opt->name);
3512 opt = TAILQ_NEXT(opt, link);
3518 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
3519 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
3520 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
3521 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
3522 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
3523 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
3524 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
3525 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
3526 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
3527 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
3528 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
3529 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
3531 db_printf(" mnt_cred = { uid=%u ruid=%u",
3532 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
3533 if (jailed(mp->mnt_cred))
3534 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
3536 db_printf(" mnt_ref = %d\n", mp->mnt_ref);
3537 db_printf(" mnt_gen = %d\n", mp->mnt_gen);
3538 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
3539 db_printf(" mnt_activevnodelistsize = %d\n",
3540 mp->mnt_activevnodelistsize);
3541 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
3542 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
3543 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
3544 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
3545 db_printf(" mnt_lockref = %d\n", mp->mnt_lockref);
3546 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
3547 db_printf(" mnt_secondary_accwrites = %d\n",
3548 mp->mnt_secondary_accwrites);
3549 db_printf(" mnt_gjprovider = %s\n",
3550 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
3552 db_printf("\n\nList of active vnodes\n");
3553 TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
3554 if (vp->v_type != VMARKER) {
3555 vn_printf(vp, "vnode ");
3560 db_printf("\n\nList of inactive vnodes\n");
3561 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3562 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
3563 vn_printf(vp, "vnode ");
3572 * Fill in a struct xvfsconf based on a struct vfsconf.
3575 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
3577 struct xvfsconf xvfsp;
3579 bzero(&xvfsp, sizeof(xvfsp));
3580 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3581 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3582 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3583 xvfsp.vfc_flags = vfsp->vfc_flags;
3585 * These are unused in userland, we keep them
3586 * to not break binary compatibility.
3588 xvfsp.vfc_vfsops = NULL;
3589 xvfsp.vfc_next = NULL;
3590 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3593 #ifdef COMPAT_FREEBSD32
3595 uint32_t vfc_vfsops;
3596 char vfc_name[MFSNAMELEN];
3597 int32_t vfc_typenum;
3598 int32_t vfc_refcount;
3604 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
3606 struct xvfsconf32 xvfsp;
3608 strcpy(xvfsp.vfc_name, vfsp->vfc_name);
3609 xvfsp.vfc_typenum = vfsp->vfc_typenum;
3610 xvfsp.vfc_refcount = vfsp->vfc_refcount;
3611 xvfsp.vfc_flags = vfsp->vfc_flags;
3612 xvfsp.vfc_vfsops = 0;
3614 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
3619 * Top level filesystem related information gathering.
3622 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
3624 struct vfsconf *vfsp;
3629 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3630 #ifdef COMPAT_FREEBSD32
3631 if (req->flags & SCTL_MASK32)
3632 error = vfsconf2x32(req, vfsp);
3635 error = vfsconf2x(req, vfsp);
3643 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD |
3644 CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist,
3645 "S,xvfsconf", "List of all configured filesystems");
3647 #ifndef BURN_BRIDGES
3648 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
3651 vfs_sysctl(SYSCTL_HANDLER_ARGS)
3653 int *name = (int *)arg1 - 1; /* XXX */
3654 u_int namelen = arg2 + 1; /* XXX */
3655 struct vfsconf *vfsp;
3657 log(LOG_WARNING, "userland calling deprecated sysctl, "
3658 "please rebuild world\n");
3660 #if 1 || defined(COMPAT_PRELITE2)
3661 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
3663 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
3667 case VFS_MAXTYPENUM:
3670 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
3673 return (ENOTDIR); /* overloaded */
3675 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3676 if (vfsp->vfc_typenum == name[2])
3681 return (EOPNOTSUPP);
3682 #ifdef COMPAT_FREEBSD32
3683 if (req->flags & SCTL_MASK32)
3684 return (vfsconf2x32(req, vfsp));
3687 return (vfsconf2x(req, vfsp));
3689 return (EOPNOTSUPP);
3692 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP |
3693 CTLFLAG_MPSAFE, vfs_sysctl,
3694 "Generic filesystem");
3696 #if 1 || defined(COMPAT_PRELITE2)
3699 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
3702 struct vfsconf *vfsp;
3703 struct ovfsconf ovfs;
3706 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
3707 bzero(&ovfs, sizeof(ovfs));
3708 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
3709 strcpy(ovfs.vfc_name, vfsp->vfc_name);
3710 ovfs.vfc_index = vfsp->vfc_typenum;
3711 ovfs.vfc_refcount = vfsp->vfc_refcount;
3712 ovfs.vfc_flags = vfsp->vfc_flags;
3713 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
3723 #endif /* 1 || COMPAT_PRELITE2 */
3724 #endif /* !BURN_BRIDGES */
3726 #define KINFO_VNODESLOP 10
3729 * Dump vnode list (via sysctl).
3733 sysctl_vnode(SYSCTL_HANDLER_ARGS)
3741 * Stale numvnodes access is not fatal here.
3744 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
3746 /* Make an estimate */
3747 return (SYSCTL_OUT(req, 0, len));
3749 error = sysctl_wire_old_buffer(req, 0);
3752 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
3754 mtx_lock(&mountlist_mtx);
3755 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3756 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
3759 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
3763 xvn[n].xv_size = sizeof *xvn;
3764 xvn[n].xv_vnode = vp;
3765 xvn[n].xv_id = 0; /* XXX compat */
3766 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
3768 XV_COPY(writecount);
3774 xvn[n].xv_flag = vp->v_vflag;
3776 switch (vp->v_type) {
3783 if (vp->v_rdev == NULL) {
3787 xvn[n].xv_dev = dev2udev(vp->v_rdev);
3790 xvn[n].xv_socket = vp->v_socket;
3793 xvn[n].xv_fifo = vp->v_fifoinfo;
3798 /* shouldn't happen? */
3806 mtx_lock(&mountlist_mtx);
3811 mtx_unlock(&mountlist_mtx);
3813 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
3818 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD |
3819 CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode",
3824 unmount_or_warn(struct mount *mp)
3828 error = dounmount(mp, MNT_FORCE, curthread);
3830 printf("unmount of %s failed (", mp->mnt_stat.f_mntonname);
3834 printf("%d)\n", error);
3839 * Unmount all filesystems. The list is traversed in reverse order
3840 * of mounting to avoid dependencies.
3843 vfs_unmountall(void)
3845 struct mount *mp, *tmp;
3847 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
3850 * Since this only runs when rebooting, it is not interlocked.
3852 TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, tmp) {
3856 * Forcibly unmounting "/dev" before "/" would prevent clean
3857 * unmount of the latter.
3859 if (mp == rootdevmp)
3862 unmount_or_warn(mp);
3865 if (rootdevmp != NULL)
3866 unmount_or_warn(rootdevmp);
3870 * perform msync on all vnodes under a mount point
3871 * the mount point must be locked.
3874 vfs_msync(struct mount *mp, int flags)
3876 struct vnode *vp, *mvp;
3877 struct vm_object *obj;
3879 CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
3880 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
3882 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
3883 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
3885 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
3887 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
3894 VM_OBJECT_WLOCK(obj);
3895 vm_object_page_clean(obj, 0, 0,
3897 OBJPC_SYNC : OBJPC_NOSYNC);
3898 VM_OBJECT_WUNLOCK(obj);
3908 destroy_vpollinfo_free(struct vpollinfo *vi)
3911 knlist_destroy(&vi->vpi_selinfo.si_note);
3912 mtx_destroy(&vi->vpi_lock);
3913 uma_zfree(vnodepoll_zone, vi);
3917 destroy_vpollinfo(struct vpollinfo *vi)
3920 knlist_clear(&vi->vpi_selinfo.si_note, 1);
3921 seldrain(&vi->vpi_selinfo);
3922 destroy_vpollinfo_free(vi);
3926 * Initalize per-vnode helper structure to hold poll-related state.
3929 v_addpollinfo(struct vnode *vp)
3931 struct vpollinfo *vi;
3933 if (vp->v_pollinfo != NULL)
3935 vi = uma_zalloc(vnodepoll_zone, M_WAITOK | M_ZERO);
3936 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
3937 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
3938 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
3940 if (vp->v_pollinfo != NULL) {
3942 destroy_vpollinfo_free(vi);
3945 vp->v_pollinfo = vi;
3950 * Record a process's interest in events which might happen to
3951 * a vnode. Because poll uses the historic select-style interface
3952 * internally, this routine serves as both the ``check for any
3953 * pending events'' and the ``record my interest in future events''
3954 * functions. (These are done together, while the lock is held,
3955 * to avoid race conditions.)
3958 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
3962 mtx_lock(&vp->v_pollinfo->vpi_lock);
3963 if (vp->v_pollinfo->vpi_revents & events) {
3965 * This leaves events we are not interested
3966 * in available for the other process which
3967 * which presumably had requested them
3968 * (otherwise they would never have been
3971 events &= vp->v_pollinfo->vpi_revents;
3972 vp->v_pollinfo->vpi_revents &= ~events;
3974 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3977 vp->v_pollinfo->vpi_events |= events;
3978 selrecord(td, &vp->v_pollinfo->vpi_selinfo);
3979 mtx_unlock(&vp->v_pollinfo->vpi_lock);
3984 * Routine to create and manage a filesystem syncer vnode.
3986 #define sync_close ((int (*)(struct vop_close_args *))nullop)
3987 static int sync_fsync(struct vop_fsync_args *);
3988 static int sync_inactive(struct vop_inactive_args *);
3989 static int sync_reclaim(struct vop_reclaim_args *);
3991 static struct vop_vector sync_vnodeops = {
3992 .vop_bypass = VOP_EOPNOTSUPP,
3993 .vop_close = sync_close, /* close */
3994 .vop_fsync = sync_fsync, /* fsync */
3995 .vop_inactive = sync_inactive, /* inactive */
3996 .vop_reclaim = sync_reclaim, /* reclaim */
3997 .vop_lock1 = vop_stdlock, /* lock */
3998 .vop_unlock = vop_stdunlock, /* unlock */
3999 .vop_islocked = vop_stdislocked, /* islocked */
4003 * Create a new filesystem syncer vnode for the specified mount point.
4006 vfs_allocate_syncvnode(struct mount *mp)
4010 static long start, incr, next;
4013 /* Allocate a new vnode */
4014 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
4016 panic("vfs_allocate_syncvnode: getnewvnode() failed");
4018 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4019 vp->v_vflag |= VV_FORCEINSMQ;
4020 error = insmntque(vp, mp);
4022 panic("vfs_allocate_syncvnode: insmntque() failed");
4023 vp->v_vflag &= ~VV_FORCEINSMQ;
4026 * Place the vnode onto the syncer worklist. We attempt to
4027 * scatter them about on the list so that they will go off
4028 * at evenly distributed times even if all the filesystems
4029 * are mounted at once.
4032 if (next == 0 || next > syncer_maxdelay) {
4036 start = syncer_maxdelay / 2;
4037 incr = syncer_maxdelay;
4043 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
4044 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
4045 mtx_lock(&sync_mtx);
4047 if (mp->mnt_syncer == NULL) {
4048 mp->mnt_syncer = vp;
4051 mtx_unlock(&sync_mtx);
4054 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4061 vfs_deallocate_syncvnode(struct mount *mp)
4065 mtx_lock(&sync_mtx);
4066 vp = mp->mnt_syncer;
4068 mp->mnt_syncer = NULL;
4069 mtx_unlock(&sync_mtx);
4075 * Do a lazy sync of the filesystem.
4078 sync_fsync(struct vop_fsync_args *ap)
4080 struct vnode *syncvp = ap->a_vp;
4081 struct mount *mp = syncvp->v_mount;
4086 * We only need to do something if this is a lazy evaluation.
4088 if (ap->a_waitfor != MNT_LAZY)
4092 * Move ourselves to the back of the sync list.
4094 bo = &syncvp->v_bufobj;
4096 vn_syncer_add_to_worklist(bo, syncdelay);
4100 * Walk the list of vnodes pushing all that are dirty and
4101 * not already on the sync list.
4103 if (vfs_busy(mp, MBF_NOWAIT) != 0)
4105 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
4109 save = curthread_pflags_set(TDP_SYNCIO);
4110 vfs_msync(mp, MNT_NOWAIT);
4111 error = VFS_SYNC(mp, MNT_LAZY);
4112 curthread_pflags_restore(save);
4113 vn_finished_write(mp);
4119 * The syncer vnode is no referenced.
4122 sync_inactive(struct vop_inactive_args *ap)
4130 * The syncer vnode is no longer needed and is being decommissioned.
4132 * Modifications to the worklist must be protected by sync_mtx.
4135 sync_reclaim(struct vop_reclaim_args *ap)
4137 struct vnode *vp = ap->a_vp;
4142 mtx_lock(&sync_mtx);
4143 if (vp->v_mount->mnt_syncer == vp)
4144 vp->v_mount->mnt_syncer = NULL;
4145 if (bo->bo_flag & BO_ONWORKLST) {
4146 LIST_REMOVE(bo, bo_synclist);
4147 syncer_worklist_len--;
4149 bo->bo_flag &= ~BO_ONWORKLST;
4151 mtx_unlock(&sync_mtx);
4158 * Check if vnode represents a disk device
4161 vn_isdisk(struct vnode *vp, int *errp)
4165 if (vp->v_type != VCHR) {
4171 if (vp->v_rdev == NULL)
4173 else if (vp->v_rdev->si_devsw == NULL)
4175 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
4181 return (error == 0);
4185 * Common filesystem object access control check routine. Accepts a
4186 * vnode's type, "mode", uid and gid, requested access mode, credentials,
4187 * and optional call-by-reference privused argument allowing vaccess()
4188 * to indicate to the caller whether privilege was used to satisfy the
4189 * request (obsoleted). Returns 0 on success, or an errno on failure.
4192 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
4193 accmode_t accmode, struct ucred *cred, int *privused)
4195 accmode_t dac_granted;
4196 accmode_t priv_granted;
4198 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
4199 ("invalid bit in accmode"));
4200 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
4201 ("VAPPEND without VWRITE"));
4204 * Look for a normal, non-privileged way to access the file/directory
4205 * as requested. If it exists, go with that.
4208 if (privused != NULL)
4213 /* Check the owner. */
4214 if (cred->cr_uid == file_uid) {
4215 dac_granted |= VADMIN;
4216 if (file_mode & S_IXUSR)
4217 dac_granted |= VEXEC;
4218 if (file_mode & S_IRUSR)
4219 dac_granted |= VREAD;
4220 if (file_mode & S_IWUSR)
4221 dac_granted |= (VWRITE | VAPPEND);
4223 if ((accmode & dac_granted) == accmode)
4229 /* Otherwise, check the groups (first match) */
4230 if (groupmember(file_gid, cred)) {
4231 if (file_mode & S_IXGRP)
4232 dac_granted |= VEXEC;
4233 if (file_mode & S_IRGRP)
4234 dac_granted |= VREAD;
4235 if (file_mode & S_IWGRP)
4236 dac_granted |= (VWRITE | VAPPEND);
4238 if ((accmode & dac_granted) == accmode)
4244 /* Otherwise, check everyone else. */
4245 if (file_mode & S_IXOTH)
4246 dac_granted |= VEXEC;
4247 if (file_mode & S_IROTH)
4248 dac_granted |= VREAD;
4249 if (file_mode & S_IWOTH)
4250 dac_granted |= (VWRITE | VAPPEND);
4251 if ((accmode & dac_granted) == accmode)
4256 * Build a privilege mask to determine if the set of privileges
4257 * satisfies the requirements when combined with the granted mask
4258 * from above. For each privilege, if the privilege is required,
4259 * bitwise or the request type onto the priv_granted mask.
4265 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
4266 * requests, instead of PRIV_VFS_EXEC.
4268 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
4269 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
4270 priv_granted |= VEXEC;
4273 * Ensure that at least one execute bit is on. Otherwise,
4274 * a privileged user will always succeed, and we don't want
4275 * this to happen unless the file really is executable.
4277 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
4278 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
4279 !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
4280 priv_granted |= VEXEC;
4283 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
4284 !priv_check_cred(cred, PRIV_VFS_READ, 0))
4285 priv_granted |= VREAD;
4287 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
4288 !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
4289 priv_granted |= (VWRITE | VAPPEND);
4291 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
4292 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
4293 priv_granted |= VADMIN;
4295 if ((accmode & (priv_granted | dac_granted)) == accmode) {
4296 /* XXX audit: privilege used */
4297 if (privused != NULL)
4302 return ((accmode & VADMIN) ? EPERM : EACCES);
4306 * Credential check based on process requesting service, and per-attribute
4310 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
4311 struct thread *td, accmode_t accmode)
4315 * Kernel-invoked always succeeds.
4321 * Do not allow privileged processes in jail to directly manipulate
4322 * system attributes.
4324 switch (attrnamespace) {
4325 case EXTATTR_NAMESPACE_SYSTEM:
4326 /* Potentially should be: return (EPERM); */
4327 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
4328 case EXTATTR_NAMESPACE_USER:
4329 return (VOP_ACCESS(vp, accmode, cred, td));
4335 #ifdef DEBUG_VFS_LOCKS
4337 * This only exists to supress warnings from unlocked specfs accesses. It is
4338 * no longer ok to have an unlocked VFS.
4340 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
4341 (vp)->v_type == VCHR || (vp)->v_type == VBAD)
4343 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
4344 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
4345 "Drop into debugger on lock violation");
4347 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
4348 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
4349 0, "Check for interlock across VOPs");
4351 int vfs_badlock_print = 1; /* Print lock violations. */
4352 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
4353 0, "Print lock violations");
4356 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
4357 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
4358 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
4362 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
4366 if (vfs_badlock_backtrace)
4369 if (vfs_badlock_print)
4370 printf("%s: %p %s\n", str, (void *)vp, msg);
4371 if (vfs_badlock_ddb)
4372 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4376 assert_vi_locked(struct vnode *vp, const char *str)
4379 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
4380 vfs_badlock("interlock is not locked but should be", str, vp);
4384 assert_vi_unlocked(struct vnode *vp, const char *str)
4387 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
4388 vfs_badlock("interlock is locked but should not be", str, vp);
4392 assert_vop_locked(struct vnode *vp, const char *str)
4396 if (!IGNORE_LOCK(vp)) {
4397 locked = VOP_ISLOCKED(vp);
4398 if (locked == 0 || locked == LK_EXCLOTHER)
4399 vfs_badlock("is not locked but should be", str, vp);
4404 assert_vop_unlocked(struct vnode *vp, const char *str)
4407 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
4408 vfs_badlock("is locked but should not be", str, vp);
4412 assert_vop_elocked(struct vnode *vp, const char *str)
4415 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
4416 vfs_badlock("is not exclusive locked but should be", str, vp);
4421 assert_vop_elocked_other(struct vnode *vp, const char *str)
4424 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
4425 vfs_badlock("is not exclusive locked by another thread",
4430 assert_vop_slocked(struct vnode *vp, const char *str)
4433 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
4434 vfs_badlock("is not locked shared but should be", str, vp);
4437 #endif /* DEBUG_VFS_LOCKS */
4440 vop_rename_fail(struct vop_rename_args *ap)
4443 if (ap->a_tvp != NULL)
4445 if (ap->a_tdvp == ap->a_tvp)
4454 vop_rename_pre(void *ap)
4456 struct vop_rename_args *a = ap;
4458 #ifdef DEBUG_VFS_LOCKS
4460 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
4461 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
4462 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
4463 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
4465 /* Check the source (from). */
4466 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
4467 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
4468 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
4469 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
4470 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
4472 /* Check the target. */
4474 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
4475 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
4477 if (a->a_tdvp != a->a_fdvp)
4479 if (a->a_tvp != a->a_fvp)
4487 vop_strategy_pre(void *ap)
4489 #ifdef DEBUG_VFS_LOCKS
4490 struct vop_strategy_args *a;
4497 * Cluster ops lock their component buffers but not the IO container.
4499 if ((bp->b_flags & B_CLUSTER) != 0)
4502 if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
4503 if (vfs_badlock_print)
4505 "VOP_STRATEGY: bp is not locked but should be\n");
4506 if (vfs_badlock_ddb)
4507 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
4513 vop_lock_pre(void *ap)
4515 #ifdef DEBUG_VFS_LOCKS
4516 struct vop_lock1_args *a = ap;
4518 if ((a->a_flags & LK_INTERLOCK) == 0)
4519 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4521 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
4526 vop_lock_post(void *ap, int rc)
4528 #ifdef DEBUG_VFS_LOCKS
4529 struct vop_lock1_args *a = ap;
4531 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
4532 if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
4533 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
4538 vop_unlock_pre(void *ap)
4540 #ifdef DEBUG_VFS_LOCKS
4541 struct vop_unlock_args *a = ap;
4543 if (a->a_flags & LK_INTERLOCK)
4544 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
4545 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
4550 vop_unlock_post(void *ap, int rc)
4552 #ifdef DEBUG_VFS_LOCKS
4553 struct vop_unlock_args *a = ap;
4555 if (a->a_flags & LK_INTERLOCK)
4556 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
4561 vop_create_post(void *ap, int rc)
4563 struct vop_create_args *a = ap;
4566 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4570 vop_deleteextattr_post(void *ap, int rc)
4572 struct vop_deleteextattr_args *a = ap;
4575 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4579 vop_link_post(void *ap, int rc)
4581 struct vop_link_args *a = ap;
4584 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
4585 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
4590 vop_mkdir_post(void *ap, int rc)
4592 struct vop_mkdir_args *a = ap;
4595 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4599 vop_mknod_post(void *ap, int rc)
4601 struct vop_mknod_args *a = ap;
4604 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4608 vop_reclaim_post(void *ap, int rc)
4610 struct vop_reclaim_args *a = ap;
4613 VFS_KNOTE_LOCKED(a->a_vp, NOTE_REVOKE);
4617 vop_remove_post(void *ap, int rc)
4619 struct vop_remove_args *a = ap;
4622 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4623 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4628 vop_rename_post(void *ap, int rc)
4630 struct vop_rename_args *a = ap;
4633 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
4634 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
4635 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
4637 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
4639 if (a->a_tdvp != a->a_fdvp)
4641 if (a->a_tvp != a->a_fvp)
4649 vop_rmdir_post(void *ap, int rc)
4651 struct vop_rmdir_args *a = ap;
4654 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
4655 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
4660 vop_setattr_post(void *ap, int rc)
4662 struct vop_setattr_args *a = ap;
4665 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4669 vop_setextattr_post(void *ap, int rc)
4671 struct vop_setextattr_args *a = ap;
4674 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
4678 vop_symlink_post(void *ap, int rc)
4680 struct vop_symlink_args *a = ap;
4683 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
4686 static struct knlist fs_knlist;
4689 vfs_event_init(void *arg)
4691 knlist_init_mtx(&fs_knlist, NULL);
4693 /* XXX - correct order? */
4694 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
4697 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
4700 KNOTE_UNLOCKED(&fs_knlist, event);
4703 static int filt_fsattach(struct knote *kn);
4704 static void filt_fsdetach(struct knote *kn);
4705 static int filt_fsevent(struct knote *kn, long hint);
4707 struct filterops fs_filtops = {
4709 .f_attach = filt_fsattach,
4710 .f_detach = filt_fsdetach,
4711 .f_event = filt_fsevent
4715 filt_fsattach(struct knote *kn)
4718 kn->kn_flags |= EV_CLEAR;
4719 knlist_add(&fs_knlist, kn, 0);
4724 filt_fsdetach(struct knote *kn)
4727 knlist_remove(&fs_knlist, kn, 0);
4731 filt_fsevent(struct knote *kn, long hint)
4734 kn->kn_fflags |= hint;
4735 return (kn->kn_fflags != 0);
4739 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
4745 error = SYSCTL_IN(req, &vc, sizeof(vc));
4748 if (vc.vc_vers != VFS_CTL_VERS1)
4750 mp = vfs_getvfs(&vc.vc_fsid);
4753 /* ensure that a specific sysctl goes to the right filesystem. */
4754 if (strcmp(vc.vc_fstypename, "*") != 0 &&
4755 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
4759 VCTLTOREQ(&vc, req);
4760 error = VFS_SYSCTL(mp, vc.vc_op, req);
4765 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
4766 NULL, 0, sysctl_vfs_ctl, "",
4770 * Function to initialize a va_filerev field sensibly.
4771 * XXX: Wouldn't a random number make a lot more sense ??
4774 init_va_filerev(void)
4779 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
4782 static int filt_vfsread(struct knote *kn, long hint);
4783 static int filt_vfswrite(struct knote *kn, long hint);
4784 static int filt_vfsvnode(struct knote *kn, long hint);
4785 static void filt_vfsdetach(struct knote *kn);
4786 static struct filterops vfsread_filtops = {
4788 .f_detach = filt_vfsdetach,
4789 .f_event = filt_vfsread
4791 static struct filterops vfswrite_filtops = {
4793 .f_detach = filt_vfsdetach,
4794 .f_event = filt_vfswrite
4796 static struct filterops vfsvnode_filtops = {
4798 .f_detach = filt_vfsdetach,
4799 .f_event = filt_vfsvnode
4803 vfs_knllock(void *arg)
4805 struct vnode *vp = arg;
4807 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
4811 vfs_knlunlock(void *arg)
4813 struct vnode *vp = arg;
4819 vfs_knl_assert_locked(void *arg)
4821 #ifdef DEBUG_VFS_LOCKS
4822 struct vnode *vp = arg;
4824 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
4829 vfs_knl_assert_unlocked(void *arg)
4831 #ifdef DEBUG_VFS_LOCKS
4832 struct vnode *vp = arg;
4834 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
4839 vfs_kqfilter(struct vop_kqfilter_args *ap)
4841 struct vnode *vp = ap->a_vp;
4842 struct knote *kn = ap->a_kn;
4845 switch (kn->kn_filter) {
4847 kn->kn_fop = &vfsread_filtops;
4850 kn->kn_fop = &vfswrite_filtops;
4853 kn->kn_fop = &vfsvnode_filtops;
4859 kn->kn_hook = (caddr_t)vp;
4862 if (vp->v_pollinfo == NULL)
4864 knl = &vp->v_pollinfo->vpi_selinfo.si_note;
4866 knlist_add(knl, kn, 0);
4872 * Detach knote from vnode
4875 filt_vfsdetach(struct knote *kn)
4877 struct vnode *vp = (struct vnode *)kn->kn_hook;
4879 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
4880 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
4886 filt_vfsread(struct knote *kn, long hint)
4888 struct vnode *vp = (struct vnode *)kn->kn_hook;
4893 * filesystem is gone, so set the EOF flag and schedule
4894 * the knote for deletion.
4896 if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) {
4898 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4903 if (VOP_GETATTR(vp, &va, curthread->td_ucred))
4907 kn->kn_data = va.va_size - kn->kn_fp->f_offset;
4908 res = (kn->kn_sfflags & NOTE_FILE_POLL) != 0 || kn->kn_data != 0;
4915 filt_vfswrite(struct knote *kn, long hint)
4917 struct vnode *vp = (struct vnode *)kn->kn_hook;
4922 * filesystem is gone, so set the EOF flag and schedule
4923 * the knote for deletion.
4925 if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD))
4926 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
4934 filt_vfsvnode(struct knote *kn, long hint)
4936 struct vnode *vp = (struct vnode *)kn->kn_hook;
4940 if (kn->kn_sfflags & hint)
4941 kn->kn_fflags |= hint;
4942 if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) {
4943 kn->kn_flags |= EV_EOF;
4947 res = (kn->kn_fflags != 0);
4953 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
4957 if (dp->d_reclen > ap->a_uio->uio_resid)
4958 return (ENAMETOOLONG);
4959 error = uiomove(dp, dp->d_reclen, ap->a_uio);
4961 if (ap->a_ncookies != NULL) {
4962 if (ap->a_cookies != NULL)
4963 free(ap->a_cookies, M_TEMP);
4964 ap->a_cookies = NULL;
4965 *ap->a_ncookies = 0;
4969 if (ap->a_ncookies == NULL)
4972 KASSERT(ap->a_cookies,
4973 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
4975 *ap->a_cookies = realloc(*ap->a_cookies,
4976 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
4977 (*ap->a_cookies)[*ap->a_ncookies] = off;
4982 * Mark for update the access time of the file if the filesystem
4983 * supports VOP_MARKATIME. This functionality is used by execve and
4984 * mmap, so we want to avoid the I/O implied by directly setting
4985 * va_atime for the sake of efficiency.
4988 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
4993 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
4994 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
4995 (void)VOP_MARKATIME(vp);
4999 * The purpose of this routine is to remove granularity from accmode_t,
5000 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
5001 * VADMIN and VAPPEND.
5003 * If it returns 0, the caller is supposed to continue with the usual
5004 * access checks using 'accmode' as modified by this routine. If it
5005 * returns nonzero value, the caller is supposed to return that value
5008 * Note that after this routine runs, accmode may be zero.
5011 vfs_unixify_accmode(accmode_t *accmode)
5014 * There is no way to specify explicit "deny" rule using
5015 * file mode or POSIX.1e ACLs.
5017 if (*accmode & VEXPLICIT_DENY) {
5023 * None of these can be translated into usual access bits.
5024 * Also, the common case for NFSv4 ACLs is to not contain
5025 * either of these bits. Caller should check for VWRITE
5026 * on the containing directory instead.
5028 if (*accmode & (VDELETE_CHILD | VDELETE))
5031 if (*accmode & VADMIN_PERMS) {
5032 *accmode &= ~VADMIN_PERMS;
5037 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
5038 * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
5040 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
5046 * These are helper functions for filesystems to traverse all
5047 * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
5049 * This interface replaces MNT_VNODE_FOREACH.
5052 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
5055 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
5060 kern_yield(PRI_USER);
5062 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
5063 vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
5064 while (vp != NULL && (vp->v_type == VMARKER ||
5065 (vp->v_iflag & VI_DOOMED) != 0))
5066 vp = TAILQ_NEXT(vp, v_nmntvnodes);
5068 /* Check if we are done */
5070 __mnt_vnode_markerfree_all(mvp, mp);
5071 /* MNT_IUNLOCK(mp); -- done in above function */
5072 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
5075 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
5076 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
5083 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
5087 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
5090 (*mvp)->v_type = VMARKER;
5092 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
5093 while (vp != NULL && (vp->v_type == VMARKER ||
5094 (vp->v_iflag & VI_DOOMED) != 0))
5095 vp = TAILQ_NEXT(vp, v_nmntvnodes);
5097 /* Check if we are done */
5101 free(*mvp, M_VNODE_MARKER);
5105 (*mvp)->v_mount = mp;
5106 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
5114 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
5122 mtx_assert(MNT_MTX(mp), MA_OWNED);
5124 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
5125 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
5128 free(*mvp, M_VNODE_MARKER);
5133 * These are helper functions for filesystems to traverse their
5134 * active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
5137 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
5140 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
5145 free(*mvp, M_VNODE_MARKER);
5149 static struct vnode *
5150 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
5152 struct vnode *vp, *nvp;
5154 mtx_assert(&vnode_free_list_mtx, MA_OWNED);
5155 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
5157 vp = TAILQ_NEXT(*mvp, v_actfreelist);
5158 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
5159 while (vp != NULL) {
5160 if (vp->v_type == VMARKER) {
5161 vp = TAILQ_NEXT(vp, v_actfreelist);
5164 if (!VI_TRYLOCK(vp)) {
5165 if (mp_ncpus == 1 || should_yield()) {
5166 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
5167 mtx_unlock(&vnode_free_list_mtx);
5169 mtx_lock(&vnode_free_list_mtx);
5174 KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
5175 KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
5176 ("alien vnode on the active list %p %p", vp, mp));
5177 if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
5179 nvp = TAILQ_NEXT(vp, v_actfreelist);
5184 /* Check if we are done */
5186 mtx_unlock(&vnode_free_list_mtx);
5187 mnt_vnode_markerfree_active(mvp, mp);
5190 TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
5191 mtx_unlock(&vnode_free_list_mtx);
5192 ASSERT_VI_LOCKED(vp, "active iter");
5193 KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
5198 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
5202 kern_yield(PRI_USER);
5203 mtx_lock(&vnode_free_list_mtx);
5204 return (mnt_vnode_next_active(mvp, mp));
5208 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
5212 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
5216 (*mvp)->v_type = VMARKER;
5217 (*mvp)->v_mount = mp;
5219 mtx_lock(&vnode_free_list_mtx);
5220 vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
5222 mtx_unlock(&vnode_free_list_mtx);
5223 mnt_vnode_markerfree_active(mvp, mp);
5226 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
5227 return (mnt_vnode_next_active(mvp, mp));
5231 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
5237 mtx_lock(&vnode_free_list_mtx);
5238 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
5239 mtx_unlock(&vnode_free_list_mtx);
5240 mnt_vnode_markerfree_active(mvp, mp);