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 * 3. 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
35 * $FreeBSD: src/sys/kern/vfs_subr.c,v 1.249.2.30 2003/04/04 20:35:57 tegge Exp $
39 * External virtual filesystem routines
43 #include "opt_inet6.h"
45 #include <sys/param.h>
46 #include <sys/systm.h>
49 #include <sys/dirent.h>
50 #include <sys/eventhandler.h>
51 #include <sys/fcntl.h>
53 #include <sys/kernel.h>
54 #include <sys/kthread.h>
55 #include <sys/malloc.h>
57 #include <sys/mount.h>
60 #include <sys/reboot.h>
61 #include <sys/socket.h>
63 #include <sys/sysctl.h>
64 #include <sys/syslog.h>
65 #include <sys/unistd.h>
66 #include <sys/vmmeter.h>
67 #include <sys/vnode.h>
69 #include <machine/limits.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_extern.h>
74 #include <vm/vm_kern.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_pager.h>
79 #include <vm/vnode_pager.h>
80 #include <vm/vm_zone.h>
83 #include <sys/thread2.h>
84 #include <sys/sysref2.h>
85 #include <sys/mplock2.h>
87 #include <netinet/in.h>
89 static MALLOC_DEFINE(M_NETCRED
, "Export Host", "Export host address structure");
92 SYSCTL_INT(_debug
, OID_AUTO
, numvnodes
, CTLFLAG_RD
, &numvnodes
, 0,
93 "Number of vnodes allocated");
95 SYSCTL_INT(_debug
, OID_AUTO
, verbose_reclaims
, CTLFLAG_RD
, &verbose_reclaims
, 0,
96 "Output filename of reclaimed vnode(s)");
98 enum vtype iftovt_tab
[16] = {
99 VNON
, VFIFO
, VCHR
, VNON
, VDIR
, VNON
, VBLK
, VNON
,
100 VREG
, VNON
, VLNK
, VNON
, VSOCK
, VNON
, VNON
, VBAD
,
102 int vttoif_tab
[9] = {
103 0, S_IFREG
, S_IFDIR
, S_IFBLK
, S_IFCHR
, S_IFLNK
,
104 S_IFSOCK
, S_IFIFO
, S_IFMT
,
107 static int reassignbufcalls
;
108 SYSCTL_INT(_vfs
, OID_AUTO
, reassignbufcalls
, CTLFLAG_RW
, &reassignbufcalls
,
109 0, "Number of times buffers have been reassigned to the proper list");
111 static int check_buf_overlap
= 2; /* invasive check */
112 SYSCTL_INT(_vfs
, OID_AUTO
, check_buf_overlap
, CTLFLAG_RW
, &check_buf_overlap
,
113 0, "Enable overlapping buffer checks");
115 int nfs_mount_type
= -1;
116 static struct lwkt_token spechash_token
;
117 struct nfs_public nfs_pub
; /* publicly exported FS */
120 SYSCTL_INT(_kern
, KERN_MAXVNODES
, maxvnodes
, CTLFLAG_RW
,
121 &maxvnodes
, 0, "Maximum number of vnodes");
123 static struct radix_node_head
*vfs_create_addrlist_af(int af
,
124 struct netexport
*nep
);
125 static void vfs_free_addrlist (struct netexport
*nep
);
126 static int vfs_free_netcred (struct radix_node
*rn
, void *w
);
127 static void vfs_free_addrlist_af (struct radix_node_head
**prnh
);
128 static int vfs_hang_addrlist (struct mount
*mp
, struct netexport
*nep
,
129 const struct export_args
*argp
);
131 int prtactive
= 0; /* 1 => print out reclaim of active vnodes */
134 * Red black tree functions
136 static int rb_buf_compare(struct buf
*b1
, struct buf
*b2
);
137 RB_GENERATE2(buf_rb_tree
, buf
, b_rbnode
, rb_buf_compare
, off_t
, b_loffset
);
138 RB_GENERATE2(buf_rb_hash
, buf
, b_rbhash
, rb_buf_compare
, off_t
, b_loffset
);
141 rb_buf_compare(struct buf
*b1
, struct buf
*b2
)
143 if (b1
->b_loffset
< b2
->b_loffset
)
145 if (b1
->b_loffset
> b2
->b_loffset
)
151 * Initialize the vnode management data structures.
153 * Called from vfsinit()
162 * Desiredvnodes is kern.maxvnodes. We want to scale it
163 * according to available system memory but we may also have
164 * to limit it based on available KVM, which is capped on 32 bit
165 * systems, to ~80K vnodes or so.
167 * WARNING! For machines with 64-256M of ram we have to be sure
168 * that the default limit scales down well due to HAMMER
169 * taking up significantly more memory per-vnode vs UFS.
170 * We want around ~5800 on a 128M machine.
172 factor1
= 25 * (sizeof(struct vm_object
) + sizeof(struct vnode
));
173 factor2
= 30 * (sizeof(struct vm_object
) + sizeof(struct vnode
));
174 maxvnodes
= imin((int64_t)vmstats
.v_page_count
* PAGE_SIZE
/ factor1
,
176 maxvnodes
= imax(maxvnodes
, maxproc
* 8);
178 lwkt_token_init(&spechash_token
, "spechash");
182 * Knob to control the precision of file timestamps:
184 * 0 = seconds only; nanoseconds zeroed.
185 * 1 = seconds and nanoseconds, accurate within 1/HZ.
186 * 2 = seconds and nanoseconds, truncated to microseconds.
187 * >=3 = seconds and nanoseconds, maximum precision.
189 enum { TSP_SEC
, TSP_HZ
, TSP_USEC
, TSP_NSEC
};
191 static int timestamp_precision
= TSP_SEC
;
192 SYSCTL_INT(_vfs
, OID_AUTO
, timestamp_precision
, CTLFLAG_RW
,
193 ×tamp_precision
, 0, "Precision of file timestamps");
196 * Get a current timestamp.
201 vfs_timestamp(struct timespec
*tsp
)
205 switch (timestamp_precision
) {
207 tsp
->tv_sec
= time_second
;
215 TIMEVAL_TO_TIMESPEC(&tv
, tsp
);
225 * Set vnode attributes to VNOVAL
228 vattr_null(struct vattr
*vap
)
231 vap
->va_size
= VNOVAL
;
232 vap
->va_bytes
= VNOVAL
;
233 vap
->va_mode
= VNOVAL
;
234 vap
->va_nlink
= VNOVAL
;
235 vap
->va_uid
= VNOVAL
;
236 vap
->va_gid
= VNOVAL
;
237 vap
->va_fsid
= VNOVAL
;
238 vap
->va_fileid
= VNOVAL
;
239 vap
->va_blocksize
= VNOVAL
;
240 vap
->va_rmajor
= VNOVAL
;
241 vap
->va_rminor
= VNOVAL
;
242 vap
->va_atime
.tv_sec
= VNOVAL
;
243 vap
->va_atime
.tv_nsec
= VNOVAL
;
244 vap
->va_mtime
.tv_sec
= VNOVAL
;
245 vap
->va_mtime
.tv_nsec
= VNOVAL
;
246 vap
->va_ctime
.tv_sec
= VNOVAL
;
247 vap
->va_ctime
.tv_nsec
= VNOVAL
;
248 vap
->va_flags
= VNOVAL
;
249 vap
->va_gen
= VNOVAL
;
251 /* va_*_uuid fields are only valid if related flags are set */
255 * Flush out and invalidate all buffers associated with a vnode.
259 static int vinvalbuf_bp(struct buf
*bp
, void *data
);
261 struct vinvalbuf_bp_info
{
270 vinvalbuf(struct vnode
*vp
, int flags
, int slpflag
, int slptimeo
)
272 struct vinvalbuf_bp_info info
;
276 lwkt_gettoken(&vp
->v_token
);
279 * If we are being asked to save, call fsync to ensure that the inode
282 if (flags
& V_SAVE
) {
283 error
= bio_track_wait(&vp
->v_track_write
, slpflag
, slptimeo
);
286 if (!RB_EMPTY(&vp
->v_rbdirty_tree
)) {
287 if ((error
= VOP_FSYNC(vp
, MNT_WAIT
, 0)) != 0)
291 * Dirty bufs may be left or generated via races
292 * in circumstances where vinvalbuf() is called on
293 * a vnode not undergoing reclamation. Only
294 * panic if we are trying to reclaim the vnode.
296 if ((vp
->v_flag
& VRECLAIMED
) &&
297 (bio_track_active(&vp
->v_track_write
) ||
298 !RB_EMPTY(&vp
->v_rbdirty_tree
))) {
299 panic("vinvalbuf: dirty bufs");
304 info
.slptimeo
= slptimeo
;
305 info
.lkflags
= LK_EXCLUSIVE
| LK_SLEEPFAIL
;
306 if (slpflag
& PCATCH
)
307 info
.lkflags
|= LK_PCATCH
;
312 * Flush the buffer cache until nothing is left, wait for all I/O
313 * to complete. At least one pass is required. We might block
314 * in the pip code so we have to re-check. Order is important.
320 if (!RB_EMPTY(&vp
->v_rbclean_tree
)) {
322 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
,
323 NULL
, vinvalbuf_bp
, &info
);
325 if (!RB_EMPTY(&vp
->v_rbdirty_tree
)) {
327 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
328 NULL
, vinvalbuf_bp
, &info
);
332 * Wait for I/O completion.
334 bio_track_wait(&vp
->v_track_write
, 0, 0);
335 if ((object
= vp
->v_object
) != NULL
)
336 refcount_wait(&object
->paging_in_progress
, "vnvlbx");
337 } while (bio_track_active(&vp
->v_track_write
) ||
338 !RB_EMPTY(&vp
->v_rbclean_tree
) ||
339 !RB_EMPTY(&vp
->v_rbdirty_tree
));
342 * Destroy the copy in the VM cache, too.
344 if ((object
= vp
->v_object
) != NULL
) {
345 vm_object_page_remove(object
, 0, 0,
346 (flags
& V_SAVE
) ? TRUE
: FALSE
);
349 if (!RB_EMPTY(&vp
->v_rbdirty_tree
) || !RB_EMPTY(&vp
->v_rbclean_tree
))
350 panic("vinvalbuf: flush failed");
351 if (!RB_EMPTY(&vp
->v_rbhash_tree
))
352 panic("vinvalbuf: flush failed, buffers still present");
355 lwkt_reltoken(&vp
->v_token
);
360 vinvalbuf_bp(struct buf
*bp
, void *data
)
362 struct vinvalbuf_bp_info
*info
= data
;
365 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
366 atomic_add_int(&bp
->b_refs
, 1);
367 error
= BUF_TIMELOCK(bp
, info
->lkflags
,
368 "vinvalbuf", info
->slptimeo
);
369 atomic_subtract_int(&bp
->b_refs
, 1);
378 KKASSERT(bp
->b_vp
== info
->vp
);
381 * Must check clean/dirty status after successfully locking as
384 if ((info
->clean
&& (bp
->b_flags
& B_DELWRI
)) ||
385 (info
->clean
== 0 && (bp
->b_flags
& B_DELWRI
) == 0)) {
391 * NOTE: NO B_LOCKED CHECK. Also no buf_checkwrite()
392 * check. This code will write out the buffer, period.
395 if (((bp
->b_flags
& (B_DELWRI
| B_INVAL
)) == B_DELWRI
) &&
396 (info
->flags
& V_SAVE
)) {
398 } else if (info
->flags
& V_SAVE
) {
400 * Cannot set B_NOCACHE on a clean buffer as this will
401 * destroy the VM backing store which might actually
402 * be dirty (and unsynchronized).
404 bp
->b_flags
|= (B_INVAL
| B_RELBUF
);
407 bp
->b_flags
|= (B_INVAL
| B_NOCACHE
| B_RELBUF
);
414 * Truncate a file's buffer and pages to a specified length. This
415 * is in lieu of the old vinvalbuf mechanism, which performed unneeded
418 * The vnode must be locked.
420 static int vtruncbuf_bp_trunc_cmp(struct buf
*bp
, void *data
);
421 static int vtruncbuf_bp_trunc(struct buf
*bp
, void *data
);
422 static int vtruncbuf_bp_metasync_cmp(struct buf
*bp
, void *data
);
423 static int vtruncbuf_bp_metasync(struct buf
*bp
, void *data
);
425 struct vtruncbuf_info
{
432 vtruncbuf(struct vnode
*vp
, off_t length
, int blksize
)
434 struct vtruncbuf_info info
;
435 const char *filename
;
439 * Round up to the *next* block, then destroy the buffers in question.
440 * Since we are only removing some of the buffers we must rely on the
441 * scan count to determine whether a loop is necessary.
443 if ((count
= (int)(length
% blksize
)) != 0)
444 info
.truncloffset
= length
+ (blksize
- count
);
446 info
.truncloffset
= length
;
449 lwkt_gettoken(&vp
->v_token
);
452 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
,
453 vtruncbuf_bp_trunc_cmp
,
454 vtruncbuf_bp_trunc
, &info
);
456 count
+= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
457 vtruncbuf_bp_trunc_cmp
,
458 vtruncbuf_bp_trunc
, &info
);
462 * For safety, fsync any remaining metadata if the file is not being
463 * truncated to 0. Since the metadata does not represent the entire
464 * dirty list we have to rely on the hit count to ensure that we get
469 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
470 vtruncbuf_bp_metasync_cmp
,
471 vtruncbuf_bp_metasync
, &info
);
476 * Clean out any left over VM backing store.
478 * It is possible to have in-progress I/O from buffers that were
479 * not part of the truncation. This should not happen if we
480 * are truncating to 0-length.
482 vnode_pager_setsize(vp
, length
);
483 bio_track_wait(&vp
->v_track_write
, 0, 0);
488 spin_lock(&vp
->v_spin
);
489 filename
= TAILQ_FIRST(&vp
->v_namecache
) ?
490 TAILQ_FIRST(&vp
->v_namecache
)->nc_name
: "?";
491 spin_unlock(&vp
->v_spin
);
494 * Make sure no buffers were instantiated while we were trying
495 * to clean out the remaining VM pages. This could occur due
496 * to busy dirty VM pages being flushed out to disk.
500 count
= RB_SCAN(buf_rb_tree
, &vp
->v_rbclean_tree
,
501 vtruncbuf_bp_trunc_cmp
,
502 vtruncbuf_bp_trunc
, &info
);
504 count
+= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
505 vtruncbuf_bp_trunc_cmp
,
506 vtruncbuf_bp_trunc
, &info
);
508 kprintf("Warning: vtruncbuf(): Had to re-clean %d "
509 "left over buffers in %s\n", count
, filename
);
513 lwkt_reltoken(&vp
->v_token
);
519 * The callback buffer is beyond the new file EOF and must be destroyed.
520 * Note that the compare function must conform to the RB_SCAN's requirements.
524 vtruncbuf_bp_trunc_cmp(struct buf
*bp
, void *data
)
526 struct vtruncbuf_info
*info
= data
;
528 if (bp
->b_loffset
>= info
->truncloffset
)
535 vtruncbuf_bp_trunc(struct buf
*bp
, void *data
)
537 struct vtruncbuf_info
*info
= data
;
540 * Do not try to use a buffer we cannot immediately lock, but sleep
541 * anyway to prevent a livelock. The code will loop until all buffers
544 * We must always revalidate the buffer after locking it to deal
547 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
548 atomic_add_int(&bp
->b_refs
, 1);
549 if (BUF_LOCK(bp
, LK_EXCLUSIVE
|LK_SLEEPFAIL
) == 0)
551 atomic_subtract_int(&bp
->b_refs
, 1);
552 } else if ((info
->clean
&& (bp
->b_flags
& B_DELWRI
)) ||
553 (info
->clean
== 0 && (bp
->b_flags
& B_DELWRI
) == 0) ||
554 bp
->b_vp
!= info
->vp
||
555 vtruncbuf_bp_trunc_cmp(bp
, data
)) {
559 bp
->b_flags
|= (B_INVAL
| B_RELBUF
| B_NOCACHE
);
566 * Fsync all meta-data after truncating a file to be non-zero. Only metadata
567 * blocks (with a negative loffset) are scanned.
568 * Note that the compare function must conform to the RB_SCAN's requirements.
571 vtruncbuf_bp_metasync_cmp(struct buf
*bp
, void *data __unused
)
573 if (bp
->b_loffset
< 0)
579 vtruncbuf_bp_metasync(struct buf
*bp
, void *data
)
581 struct vtruncbuf_info
*info
= data
;
583 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
584 atomic_add_int(&bp
->b_refs
, 1);
585 if (BUF_LOCK(bp
, LK_EXCLUSIVE
|LK_SLEEPFAIL
) == 0)
587 atomic_subtract_int(&bp
->b_refs
, 1);
588 } else if ((bp
->b_flags
& B_DELWRI
) == 0 ||
589 bp
->b_vp
!= info
->vp
||
590 vtruncbuf_bp_metasync_cmp(bp
, data
)) {
594 if (bp
->b_vp
== info
->vp
)
603 * vfsync - implements a multipass fsync on a file which understands
604 * dependancies and meta-data. The passed vnode must be locked. The
605 * waitfor argument may be MNT_WAIT or MNT_NOWAIT, or MNT_LAZY.
607 * When fsyncing data asynchronously just do one consolidated pass starting
608 * with the most negative block number. This may not get all the data due
611 * When fsyncing data synchronously do a data pass, then a metadata pass,
612 * then do additional data+metadata passes to try to get all the data out.
614 * Caller must ref the vnode but does not have to lock it.
616 static int vfsync_wait_output(struct vnode
*vp
,
617 int (*waitoutput
)(struct vnode
*, struct thread
*));
618 static int vfsync_dummy_cmp(struct buf
*bp __unused
, void *data __unused
);
619 static int vfsync_data_only_cmp(struct buf
*bp
, void *data
);
620 static int vfsync_meta_only_cmp(struct buf
*bp
, void *data
);
621 static int vfsync_lazy_range_cmp(struct buf
*bp
, void *data
);
622 static int vfsync_bp(struct buf
*bp
, void *data
);
632 int (*checkdef
)(struct buf
*);
633 int (*cmpfunc
)(struct buf
*, void *);
637 vfsync(struct vnode
*vp
, int waitfor
, int passes
,
638 int (*checkdef
)(struct buf
*),
639 int (*waitoutput
)(struct vnode
*, struct thread
*))
641 struct vfsync_info info
;
644 bzero(&info
, sizeof(info
));
646 if ((info
.checkdef
= checkdef
) == NULL
)
649 lwkt_gettoken(&vp
->v_token
);
652 case MNT_LAZY
| MNT_NOWAIT
:
655 * Lazy (filesystem syncer typ) Asynchronous plus limit the
656 * number of data (not meta) pages we try to flush to 1MB.
657 * A non-zero return means that lazy limit was reached.
659 info
.lazylimit
= 1024 * 1024;
661 info
.cmpfunc
= vfsync_lazy_range_cmp
;
662 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
663 vfsync_lazy_range_cmp
, vfsync_bp
, &info
);
664 info
.cmpfunc
= vfsync_meta_only_cmp
;
665 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
,
666 vfsync_meta_only_cmp
, vfsync_bp
, &info
);
669 else if (!RB_EMPTY(&vp
->v_rbdirty_tree
))
670 vn_syncer_add(vp
, 1);
675 * Asynchronous. Do a data-only pass and a meta-only pass.
678 info
.cmpfunc
= vfsync_data_only_cmp
;
679 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_data_only_cmp
,
681 info
.cmpfunc
= vfsync_meta_only_cmp
;
682 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_meta_only_cmp
,
688 * Synchronous. Do a data-only pass, then a meta-data+data
689 * pass, then additional integrated passes to try to get
690 * all the dependancies flushed.
692 info
.cmpfunc
= vfsync_data_only_cmp
;
694 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, vfsync_data_only_cmp
,
697 error
= vfsync_wait_output(vp
, waitoutput
);
699 info
.skippedbufs
= 0;
700 info
.cmpfunc
= vfsync_dummy_cmp
;
701 RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
703 error
= vfsync_wait_output(vp
, waitoutput
);
704 if (info
.skippedbufs
) {
705 kprintf("Warning: vfsync skipped %d dirty "
708 ((info
.skippedbufs
> 1) ? "s" : ""));
711 while (error
== 0 && passes
> 0 &&
712 !RB_EMPTY(&vp
->v_rbdirty_tree
)
714 info
.skippedbufs
= 0;
716 info
.synchronous
= 1;
719 info
.cmpfunc
= vfsync_dummy_cmp
;
720 error
= RB_SCAN(buf_rb_tree
, &vp
->v_rbdirty_tree
, NULL
,
726 error
= vfsync_wait_output(vp
, waitoutput
);
727 if (info
.skippedbufs
&& passes
== 0) {
728 kprintf("Warning: vfsync skipped %d dirty "
729 "buf%s in final pass!\n",
731 ((info
.skippedbufs
> 1) ? "s" : ""));
736 * This case can occur normally because vnode lock might
739 if (!RB_EMPTY(&vp
->v_rbdirty_tree
))
740 kprintf("dirty bufs left after final pass\n");
744 lwkt_reltoken(&vp
->v_token
);
750 vfsync_wait_output(struct vnode
*vp
,
751 int (*waitoutput
)(struct vnode
*, struct thread
*))
755 error
= bio_track_wait(&vp
->v_track_write
, 0, 0);
757 error
= waitoutput(vp
, curthread
);
762 vfsync_dummy_cmp(struct buf
*bp __unused
, void *data __unused
)
768 vfsync_data_only_cmp(struct buf
*bp
, void *data
)
770 if (bp
->b_loffset
< 0)
776 vfsync_meta_only_cmp(struct buf
*bp
, void *data
)
778 if (bp
->b_loffset
< 0)
784 vfsync_lazy_range_cmp(struct buf
*bp
, void *data
)
786 struct vfsync_info
*info
= data
;
788 if (bp
->b_loffset
< info
->vp
->v_lazyw
)
794 vfsync_bp(struct buf
*bp
, void *data
)
796 struct vfsync_info
*info
= data
;
797 struct vnode
*vp
= info
->vp
;
800 if (info
->fastpass
) {
802 * Ignore buffers that we cannot immediately lock.
804 if (BUF_LOCK(bp
, LK_EXCLUSIVE
| LK_NOWAIT
)) {
805 if (BUF_TIMELOCK(bp
, LK_EXCLUSIVE
, "bflst1", 1)) {
810 } else if (info
->synchronous
== 0) {
812 * Normal pass, give the buffer a little time to become
815 if (BUF_TIMELOCK(bp
, LK_EXCLUSIVE
, "bflst2", hz
/ 10)) {
821 * Synchronous pass, give the buffer a lot of time before
824 if (BUF_TIMELOCK(bp
, LK_EXCLUSIVE
, "bflst3", hz
* 10)) {
831 * We must revalidate the buffer after locking.
833 if ((bp
->b_flags
& B_DELWRI
) == 0 ||
834 bp
->b_vp
!= info
->vp
||
835 info
->cmpfunc(bp
, data
)) {
841 * If syncdeps is not set we do not try to write buffers which have
844 if (!info
->synchronous
&& info
->syncdeps
== 0 && info
->checkdef(bp
)) {
850 * B_NEEDCOMMIT (primarily used by NFS) is a state where the buffer
851 * has been written but an additional handshake with the device
852 * is required before we can dispose of the buffer. We have no idea
853 * how to do this so we have to skip these buffers.
855 if (bp
->b_flags
& B_NEEDCOMMIT
) {
861 * Ask bioops if it is ok to sync. If not the VFS may have
862 * set B_LOCKED so we have to cycle the buffer.
864 if (LIST_FIRST(&bp
->b_dep
) != NULL
&& buf_checkwrite(bp
)) {
870 if (info
->synchronous
) {
872 * Synchronous flushing. An error may be returned.
878 * Asynchronous flushing. A negative return value simply
879 * stops the scan and is not considered an error. We use
880 * this to support limited MNT_LAZY flushes.
882 vp
->v_lazyw
= bp
->b_loffset
;
884 info
->lazycount
+= cluster_awrite(bp
);
885 waitrunningbufspace();
887 if (info
->lazylimit
&& info
->lazycount
>= info
->lazylimit
)
896 * Associate a buffer with a vnode.
901 bgetvp(struct vnode
*vp
, struct buf
*bp
, int testsize
)
903 KASSERT(bp
->b_vp
== NULL
, ("bgetvp: not free"));
904 KKASSERT((bp
->b_flags
& (B_HASHED
|B_DELWRI
|B_VNCLEAN
|B_VNDIRTY
)) == 0);
907 * Insert onto list for new vnode.
909 lwkt_gettoken(&vp
->v_token
);
911 if (buf_rb_hash_RB_INSERT(&vp
->v_rbhash_tree
, bp
)) {
912 lwkt_reltoken(&vp
->v_token
);
917 * Diagnostics (mainly for HAMMER debugging). Check for
918 * overlapping buffers.
920 if (check_buf_overlap
) {
922 bx
= buf_rb_hash_RB_PREV(bp
);
924 if (bx
->b_loffset
+ bx
->b_bufsize
> bp
->b_loffset
) {
925 kprintf("bgetvp: overlapl %016jx/%d %016jx "
927 (intmax_t)bx
->b_loffset
,
929 (intmax_t)bp
->b_loffset
,
931 if (check_buf_overlap
> 1)
932 panic("bgetvp - overlapping buffer");
935 bx
= buf_rb_hash_RB_NEXT(bp
);
937 if (bp
->b_loffset
+ testsize
> bx
->b_loffset
) {
938 kprintf("bgetvp: overlapr %016jx/%d %016jx "
940 (intmax_t)bp
->b_loffset
,
942 (intmax_t)bx
->b_loffset
,
944 if (check_buf_overlap
> 1)
945 panic("bgetvp - overlapping buffer");
950 bp
->b_flags
|= B_HASHED
;
951 bp
->b_flags
|= B_VNCLEAN
;
952 if (buf_rb_tree_RB_INSERT(&vp
->v_rbclean_tree
, bp
))
953 panic("reassignbuf: dup lblk/clean vp %p bp %p", vp
, bp
);
955 lwkt_reltoken(&vp
->v_token
);
960 * Disassociate a buffer from a vnode.
965 brelvp(struct buf
*bp
)
969 KASSERT(bp
->b_vp
!= NULL
, ("brelvp: NULL"));
972 * Delete from old vnode list, if on one.
975 lwkt_gettoken(&vp
->v_token
);
976 if (bp
->b_flags
& (B_VNDIRTY
| B_VNCLEAN
)) {
977 if (bp
->b_flags
& B_VNDIRTY
)
978 buf_rb_tree_RB_REMOVE(&vp
->v_rbdirty_tree
, bp
);
980 buf_rb_tree_RB_REMOVE(&vp
->v_rbclean_tree
, bp
);
981 bp
->b_flags
&= ~(B_VNDIRTY
| B_VNCLEAN
);
983 if (bp
->b_flags
& B_HASHED
) {
984 buf_rb_hash_RB_REMOVE(&vp
->v_rbhash_tree
, bp
);
985 bp
->b_flags
&= ~B_HASHED
;
989 * Only remove from synclist when no dirty buffers are left AND
990 * the VFS has not flagged the vnode's inode as being dirty.
992 if ((vp
->v_flag
& (VONWORKLST
| VISDIRTY
| VOBJDIRTY
)) == VONWORKLST
&&
993 RB_EMPTY(&vp
->v_rbdirty_tree
)) {
994 vn_syncer_remove(vp
, 0);
998 lwkt_reltoken(&vp
->v_token
);
1004 * Reassign the buffer to the proper clean/dirty list based on B_DELWRI.
1005 * This routine is called when the state of the B_DELWRI bit is changed.
1007 * Must be called with vp->v_token held.
1011 reassignbuf(struct buf
*bp
)
1013 struct vnode
*vp
= bp
->b_vp
;
1016 ASSERT_LWKT_TOKEN_HELD(&vp
->v_token
);
1020 * B_PAGING flagged buffers cannot be reassigned because their vp
1021 * is not fully linked in.
1023 if (bp
->b_flags
& B_PAGING
)
1024 panic("cannot reassign paging buffer");
1026 if (bp
->b_flags
& B_DELWRI
) {
1028 * Move to the dirty list, add the vnode to the worklist
1030 if (bp
->b_flags
& B_VNCLEAN
) {
1031 buf_rb_tree_RB_REMOVE(&vp
->v_rbclean_tree
, bp
);
1032 bp
->b_flags
&= ~B_VNCLEAN
;
1034 if ((bp
->b_flags
& B_VNDIRTY
) == 0) {
1035 if (buf_rb_tree_RB_INSERT(&vp
->v_rbdirty_tree
, bp
)) {
1036 panic("reassignbuf: dup lblk vp %p bp %p",
1039 bp
->b_flags
|= B_VNDIRTY
;
1041 if ((vp
->v_flag
& VONWORKLST
) == 0) {
1042 switch (vp
->v_type
) {
1049 vp
->v_rdev
->si_mountpoint
!= NULL
) {
1057 vn_syncer_add(vp
, delay
);
1061 * Move to the clean list, remove the vnode from the worklist
1062 * if no dirty blocks remain.
1064 if (bp
->b_flags
& B_VNDIRTY
) {
1065 buf_rb_tree_RB_REMOVE(&vp
->v_rbdirty_tree
, bp
);
1066 bp
->b_flags
&= ~B_VNDIRTY
;
1068 if ((bp
->b_flags
& B_VNCLEAN
) == 0) {
1069 if (buf_rb_tree_RB_INSERT(&vp
->v_rbclean_tree
, bp
)) {
1070 panic("reassignbuf: dup lblk vp %p bp %p",
1073 bp
->b_flags
|= B_VNCLEAN
;
1077 * Only remove from synclist when no dirty buffers are left
1078 * AND the VFS has not flagged the vnode's inode as being
1081 if ((vp
->v_flag
& (VONWORKLST
| VISDIRTY
| VOBJDIRTY
)) ==
1083 RB_EMPTY(&vp
->v_rbdirty_tree
)) {
1084 vn_syncer_remove(vp
, 0);
1090 * Create a vnode for a block device. Used for mounting the root file
1093 * A vref()'d vnode is returned.
1095 extern struct vop_ops
*devfs_vnode_dev_vops_p
;
1097 bdevvp(cdev_t dev
, struct vnode
**vpp
)
1107 error
= getspecialvnode(VT_NON
, NULL
, &devfs_vnode_dev_vops_p
,
1118 v_associate_rdev(vp
, dev
);
1119 vp
->v_umajor
= dev
->si_umajor
;
1120 vp
->v_uminor
= dev
->si_uminor
;
1127 v_associate_rdev(struct vnode
*vp
, cdev_t dev
)
1131 if (dev_is_good(dev
) == 0)
1133 KKASSERT(vp
->v_rdev
== NULL
);
1134 vp
->v_rdev
= reference_dev(dev
);
1135 lwkt_gettoken(&spechash_token
);
1136 SLIST_INSERT_HEAD(&dev
->si_hlist
, vp
, v_cdevnext
);
1137 lwkt_reltoken(&spechash_token
);
1142 v_release_rdev(struct vnode
*vp
)
1146 if ((dev
= vp
->v_rdev
) != NULL
) {
1147 lwkt_gettoken(&spechash_token
);
1148 SLIST_REMOVE(&dev
->si_hlist
, vp
, vnode
, v_cdevnext
);
1151 lwkt_reltoken(&spechash_token
);
1156 * Add a vnode to the alias list hung off the cdev_t. We only associate
1157 * the device number with the vnode. The actual device is not associated
1158 * until the vnode is opened (usually in spec_open()), and will be
1159 * disassociated on last close.
1162 addaliasu(struct vnode
*nvp
, int x
, int y
)
1164 if (nvp
->v_type
!= VBLK
&& nvp
->v_type
!= VCHR
)
1165 panic("addaliasu on non-special vnode");
1171 * Simple call that a filesystem can make to try to get rid of a
1172 * vnode. It will fail if anyone is referencing the vnode (including
1175 * The filesystem can check whether its in-memory inode structure still
1176 * references the vp on return.
1178 * May only be called if the vnode is in a known state (i.e. being prevented
1179 * from being deallocated by some other condition such as a vfs inode hold).
1182 vclean_unlocked(struct vnode
*vp
)
1185 if (VREFCNT(vp
) <= 1)
1191 * Disassociate a vnode from its underlying filesystem.
1193 * The vnode must be VX locked and referenced. In all normal situations
1194 * there are no active references. If vclean_vxlocked() is called while
1195 * there are active references, the vnode is being ripped out and we have
1196 * to call VOP_CLOSE() as appropriate before we can reclaim it.
1199 vclean_vxlocked(struct vnode
*vp
, int flags
)
1204 struct namecache
*ncp
;
1207 * If the vnode has already been reclaimed we have nothing to do.
1209 if (vp
->v_flag
& VRECLAIMED
)
1213 * Set flag to interlock operation, flag finalization to ensure
1214 * that the vnode winds up on the inactive list, and set v_act to 0.
1216 vsetflags(vp
, VRECLAIMED
);
1217 atomic_set_int(&vp
->v_refcnt
, VREF_FINALIZE
);
1220 if (verbose_reclaims
) {
1221 if ((ncp
= TAILQ_FIRST(&vp
->v_namecache
)) != NULL
)
1222 kprintf("Debug: reclaim %p %s\n", vp
, ncp
->nc_name
);
1226 * Scrap the vfs cache
1228 while (cache_inval_vp(vp
, 0) != 0) {
1229 kprintf("Warning: vnode %p clean/cache_resolution "
1230 "race detected\n", vp
);
1231 tsleep(vp
, 0, "vclninv", 2);
1235 * Check to see if the vnode is in use. If so we have to reference it
1236 * before we clean it out so that its count cannot fall to zero and
1237 * generate a race against ourselves to recycle it.
1239 active
= (VREFCNT(vp
) > 0);
1242 * Clean out any buffers associated with the vnode and destroy its
1243 * object, if it has one.
1245 vinvalbuf(vp
, V_SAVE
, 0, 0);
1246 KKASSERT(lockcountnb(&vp
->v_lock
) == 1);
1249 * If purging an active vnode (typically during a forced unmount
1250 * or reboot), it must be closed and deactivated before being
1251 * reclaimed. This isn't really all that safe, but what can
1254 * Note that neither of these routines unlocks the vnode.
1256 if (active
&& (flags
& DOCLOSE
)) {
1257 while ((n
= vp
->v_opencount
) != 0) {
1258 if (vp
->v_writecount
)
1259 VOP_CLOSE(vp
, FWRITE
|FNONBLOCK
, NULL
);
1261 VOP_CLOSE(vp
, FNONBLOCK
, NULL
);
1262 if (vp
->v_opencount
== n
) {
1263 kprintf("Warning: unable to force-close"
1271 * If the vnode has not been deactivated, deactivated it. Deactivation
1272 * can create new buffers and VM pages so we have to call vinvalbuf()
1273 * again to make sure they all get flushed.
1275 * This can occur if a file with a link count of 0 needs to be
1278 * If the vnode is already dead don't try to deactivate it.
1280 if ((vp
->v_flag
& VINACTIVE
) == 0) {
1281 vsetflags(vp
, VINACTIVE
);
1284 vinvalbuf(vp
, V_SAVE
, 0, 0);
1286 KKASSERT(lockcountnb(&vp
->v_lock
) == 1);
1289 * If the vnode has an object, destroy it.
1291 while ((object
= vp
->v_object
) != NULL
) {
1292 vm_object_hold(object
);
1293 if (object
== vp
->v_object
)
1295 vm_object_drop(object
);
1298 if (object
!= NULL
) {
1299 if (object
->ref_count
== 0) {
1300 if ((object
->flags
& OBJ_DEAD
) == 0)
1301 vm_object_terminate(object
);
1302 vm_object_drop(object
);
1303 vclrflags(vp
, VOBJBUF
);
1305 vm_pager_deallocate(object
);
1306 vclrflags(vp
, VOBJBUF
);
1307 vm_object_drop(object
);
1310 KKASSERT((vp
->v_flag
& VOBJBUF
) == 0);
1313 * Reclaim the vnode if not already dead.
1315 if (vp
->v_mount
&& VOP_RECLAIM(vp
))
1316 panic("vclean: cannot reclaim");
1319 * Done with purge, notify sleepers of the grim news.
1321 vp
->v_ops
= &dead_vnode_vops_p
;
1326 * If we are destroying an active vnode, reactivate it now that
1327 * we have reassociated it with deadfs. This prevents the system
1328 * from crashing on the vnode due to it being unexpectedly marked
1329 * as inactive or reclaimed.
1331 if (active
&& (flags
& DOCLOSE
)) {
1332 vclrflags(vp
, VINACTIVE
| VRECLAIMED
);
1337 * Eliminate all activity associated with the requested vnode
1338 * and with all vnodes aliased to the requested vnode.
1340 * The vnode must be referenced but should not be locked.
1343 vrevoke(struct vnode
*vp
, struct ucred
*cred
)
1351 * If the vnode has a device association, scrap all vnodes associated
1352 * with the device. Don't let the device disappear on us while we
1353 * are scrapping the vnodes.
1355 * The passed vp will probably show up in the list, do not VX lock
1358 * Releasing the vnode's rdev here can mess up specfs's call to
1359 * device close, so don't do it. The vnode has been disassociated
1360 * and the device will be closed after the last ref on the related
1361 * fp goes away (if not still open by e.g. the kernel).
1363 if (vp
->v_type
!= VCHR
) {
1364 error
= fdrevoke(vp
, DTYPE_VNODE
, cred
);
1367 if ((dev
= vp
->v_rdev
) == NULL
) {
1371 lwkt_gettoken(&spechash_token
);
1374 vqn
= SLIST_FIRST(&dev
->si_hlist
);
1377 while ((vq
= vqn
) != NULL
) {
1378 if (VREFCNT(vq
) > 0) {
1380 fdrevoke(vq
, DTYPE_VNODE
, cred
);
1381 /*v_release_rdev(vq);*/
1383 if (vq
->v_rdev
!= dev
) {
1388 vqn
= SLIST_NEXT(vq
, v_cdevnext
);
1393 lwkt_reltoken(&spechash_token
);
1400 * This is called when the object underlying a vnode is being destroyed,
1401 * such as in a remove(). Try to recycle the vnode immediately if the
1402 * only active reference is our reference.
1404 * Directory vnodes in the namecache with children cannot be immediately
1405 * recycled because numerous VOP_N*() ops require them to be stable.
1407 * To avoid recursive recycling from VOP_INACTIVE implemenetations this
1408 * function is a NOP if VRECLAIMED is already set.
1411 vrecycle(struct vnode
*vp
)
1413 if (VREFCNT(vp
) <= 1 && (vp
->v_flag
& VRECLAIMED
) == 0) {
1414 if (cache_inval_vp_nonblock(vp
))
1423 * Return the maximum I/O size allowed for strategy calls on VP.
1425 * If vp is VCHR or VBLK we dive the device, otherwise we use
1426 * the vp's mount info.
1428 * The returned value is clamped at MAXPHYS as most callers cannot use
1429 * buffers larger than that size.
1432 vmaxiosize(struct vnode
*vp
)
1436 if (vp
->v_type
== VBLK
|| vp
->v_type
== VCHR
)
1437 maxiosize
= vp
->v_rdev
->si_iosize_max
;
1439 maxiosize
= vp
->v_mount
->mnt_iosize_max
;
1441 if (maxiosize
> MAXPHYS
)
1442 maxiosize
= MAXPHYS
;
1447 * Eliminate all activity associated with a vnode in preparation for
1450 * The vnode must be VX locked and refd and will remain VX locked and refd
1451 * on return. This routine may be called with the vnode in any state, as
1452 * long as it is VX locked. The vnode will be cleaned out and marked
1453 * VRECLAIMED but will not actually be reused until all existing refs and
1456 * NOTE: This routine may be called on a vnode which has not yet been
1457 * already been deactivated (VOP_INACTIVE), or on a vnode which has
1458 * already been reclaimed.
1460 * This routine is not responsible for placing us back on the freelist.
1461 * Instead, it happens automatically when the caller releases the VX lock
1462 * (assuming there aren't any other references).
1465 vgone_vxlocked(struct vnode
*vp
)
1468 * assert that the VX lock is held. This is an absolute requirement
1469 * now for vgone_vxlocked() to be called.
1471 KKASSERT(lockcountnb(&vp
->v_lock
) == 1);
1474 * Clean out the filesystem specific data and set the VRECLAIMED
1475 * bit. Also deactivate the vnode if necessary.
1477 * The vnode should have automatically been removed from the syncer
1478 * list as syncer/dirty flags cleared during the cleaning.
1480 vclean_vxlocked(vp
, DOCLOSE
);
1483 * Normally panic if the vnode is still dirty, unless we are doing
1484 * a forced unmount (tmpfs typically).
1486 if (vp
->v_flag
& VONWORKLST
) {
1487 if (vp
->v_mount
->mnt_kern_flag
& MNTK_UNMOUNTF
) {
1489 vn_syncer_remove(vp
, 1);
1491 panic("vp %p still dirty in vgone after flush", vp
);
1496 * Delete from old mount point vnode list, if on one.
1498 if (vp
->v_mount
!= NULL
) {
1499 KKASSERT(vp
->v_data
== NULL
);
1500 insmntque(vp
, NULL
);
1504 * If special device, remove it from special device alias list
1505 * if it is on one. This should normally only occur if a vnode is
1506 * being revoked as the device should otherwise have been released
1509 if ((vp
->v_type
== VBLK
|| vp
->v_type
== VCHR
) && vp
->v_rdev
!= NULL
) {
1520 * Lookup a vnode by device number.
1522 * Returns non-zero and *vpp set to a vref'd vnode on success.
1523 * Returns zero on failure.
1526 vfinddev(cdev_t dev
, enum vtype type
, struct vnode
**vpp
)
1530 lwkt_gettoken(&spechash_token
);
1531 SLIST_FOREACH(vp
, &dev
->si_hlist
, v_cdevnext
) {
1532 if (type
== vp
->v_type
) {
1535 lwkt_reltoken(&spechash_token
);
1539 lwkt_reltoken(&spechash_token
);
1544 * Calculate the total number of references to a special device. This
1545 * routine may only be called for VBLK and VCHR vnodes since v_rdev is
1546 * an overloaded field. Since udev2dev can now return NULL, we have
1547 * to check for a NULL v_rdev.
1550 count_dev(cdev_t dev
)
1555 if (SLIST_FIRST(&dev
->si_hlist
)) {
1556 lwkt_gettoken(&spechash_token
);
1557 SLIST_FOREACH(vp
, &dev
->si_hlist
, v_cdevnext
) {
1558 count
+= vp
->v_opencount
;
1560 lwkt_reltoken(&spechash_token
);
1566 vcount(struct vnode
*vp
)
1568 if (vp
->v_rdev
== NULL
)
1570 return(count_dev(vp
->v_rdev
));
1574 * Initialize VMIO for a vnode. This routine MUST be called before a
1575 * VFS can issue buffer cache ops on a vnode. It is typically called
1576 * when a vnode is initialized from its inode.
1579 vinitvmio(struct vnode
*vp
, off_t filesize
, int blksize
, int boff
)
1584 object
= vp
->v_object
;
1586 vm_object_hold(object
);
1587 KKASSERT(vp
->v_object
== object
);
1590 if (object
== NULL
) {
1591 object
= vnode_pager_alloc(vp
, filesize
, 0, 0, blksize
, boff
);
1594 * Dereference the reference we just created. This assumes
1595 * that the object is associated with the vp. Allow it to
1596 * have zero refs. It cannot be destroyed as long as it
1597 * is associated with the vnode.
1599 vm_object_hold(object
);
1600 atomic_add_int(&object
->ref_count
, -1);
1603 KKASSERT((object
->flags
& OBJ_DEAD
) == 0);
1605 KASSERT(vp
->v_object
!= NULL
, ("vinitvmio: NULL object"));
1606 vsetflags(vp
, VOBJBUF
);
1607 vm_object_drop(object
);
1614 * Print out a description of a vnode.
1616 static char *typename
[] =
1617 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
1620 vprint(char *label
, struct vnode
*vp
)
1625 kprintf("%s: %p: ", label
, (void *)vp
);
1627 kprintf("%p: ", (void *)vp
);
1628 kprintf("type %s, refcnt %08x, writecount %d, holdcnt %d,",
1629 typename
[vp
->v_type
],
1630 vp
->v_refcnt
, vp
->v_writecount
, vp
->v_auxrefs
);
1632 if (vp
->v_flag
& VROOT
)
1633 strcat(buf
, "|VROOT");
1634 if (vp
->v_flag
& VPFSROOT
)
1635 strcat(buf
, "|VPFSROOT");
1636 if (vp
->v_flag
& VTEXT
)
1637 strcat(buf
, "|VTEXT");
1638 if (vp
->v_flag
& VSYSTEM
)
1639 strcat(buf
, "|VSYSTEM");
1640 if (vp
->v_flag
& VOBJBUF
)
1641 strcat(buf
, "|VOBJBUF");
1643 kprintf(" flags (%s)", &buf
[1]);
1644 if (vp
->v_data
== NULL
) {
1653 * Do the usual access checking.
1654 * file_mode, uid and gid are from the vnode in question,
1655 * while acc_mode and cred are from the VOP_ACCESS parameter list
1658 vaccess(enum vtype type
, mode_t file_mode
, uid_t uid
, gid_t gid
,
1659 mode_t acc_mode
, struct ucred
*cred
)
1665 * Super-user always gets read/write access, but execute access depends
1666 * on at least one execute bit being set.
1668 if (priv_check_cred(cred
, PRIV_ROOT
, 0) == 0) {
1669 if ((acc_mode
& VEXEC
) && type
!= VDIR
&&
1670 (file_mode
& (S_IXUSR
|S_IXGRP
|S_IXOTH
)) == 0)
1677 /* Otherwise, check the owner. */
1678 if (cred
->cr_uid
== uid
) {
1679 if (acc_mode
& VEXEC
)
1681 if (acc_mode
& VREAD
)
1683 if (acc_mode
& VWRITE
)
1685 return ((file_mode
& mask
) == mask
? 0 : EACCES
);
1688 /* Otherwise, check the groups. */
1689 ismember
= groupmember(gid
, cred
);
1690 if (cred
->cr_svgid
== gid
|| ismember
) {
1691 if (acc_mode
& VEXEC
)
1693 if (acc_mode
& VREAD
)
1695 if (acc_mode
& VWRITE
)
1697 return ((file_mode
& mask
) == mask
? 0 : EACCES
);
1700 /* Otherwise, check everyone else. */
1701 if (acc_mode
& VEXEC
)
1703 if (acc_mode
& VREAD
)
1705 if (acc_mode
& VWRITE
)
1707 return ((file_mode
& mask
) == mask
? 0 : EACCES
);
1711 #include <ddb/ddb.h>
1713 static int db_show_locked_vnodes(struct mount
*mp
, void *data
);
1716 * List all of the locked vnodes in the system.
1717 * Called when debugging the kernel.
1719 DB_SHOW_COMMAND(lockedvnodes
, lockedvnodes
)
1721 kprintf("Locked vnodes\n");
1722 mountlist_scan(db_show_locked_vnodes
, NULL
,
1723 MNTSCAN_FORWARD
|MNTSCAN_NOBUSY
);
1727 db_show_locked_vnodes(struct mount
*mp
, void *data __unused
)
1731 TAILQ_FOREACH(vp
, &mp
->mnt_nvnodelist
, v_nmntvnodes
) {
1732 if (vn_islocked(vp
))
1740 * Top level filesystem related information gathering.
1742 static int sysctl_ovfs_conf (SYSCTL_HANDLER_ARGS
);
1745 vfs_sysctl(SYSCTL_HANDLER_ARGS
)
1747 int *name
= (int *)arg1
- 1; /* XXX */
1748 u_int namelen
= arg2
+ 1; /* XXX */
1749 struct vfsconf
*vfsp
;
1752 #if 1 || defined(COMPAT_PRELITE2)
1753 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
1755 return (sysctl_ovfs_conf(oidp
, arg1
, arg2
, req
));
1759 /* all sysctl names at this level are at least name and field */
1761 return (ENOTDIR
); /* overloaded */
1762 if (name
[0] != VFS_GENERIC
) {
1763 vfsp
= vfsconf_find_by_typenum(name
[0]);
1765 return (EOPNOTSUPP
);
1766 return ((*vfsp
->vfc_vfsops
->vfs_sysctl
)(&name
[1], namelen
- 1,
1767 oldp
, oldlenp
, newp
, newlen
, p
));
1771 case VFS_MAXTYPENUM
:
1774 maxtypenum
= vfsconf_get_maxtypenum();
1775 return (SYSCTL_OUT(req
, &maxtypenum
, sizeof(maxtypenum
)));
1778 return (ENOTDIR
); /* overloaded */
1779 vfsp
= vfsconf_find_by_typenum(name
[2]);
1781 return (EOPNOTSUPP
);
1782 return (SYSCTL_OUT(req
, vfsp
, sizeof *vfsp
));
1784 return (EOPNOTSUPP
);
1787 SYSCTL_NODE(_vfs
, VFS_GENERIC
, generic
, CTLFLAG_RD
, vfs_sysctl
,
1788 "Generic filesystem");
1790 #if 1 || defined(COMPAT_PRELITE2)
1793 sysctl_ovfs_conf_iter(struct vfsconf
*vfsp
, void *data
)
1796 struct ovfsconf ovfs
;
1797 struct sysctl_req
*req
= (struct sysctl_req
*) data
;
1799 bzero(&ovfs
, sizeof(ovfs
));
1800 ovfs
.vfc_vfsops
= vfsp
->vfc_vfsops
; /* XXX used as flag */
1801 strcpy(ovfs
.vfc_name
, vfsp
->vfc_name
);
1802 ovfs
.vfc_index
= vfsp
->vfc_typenum
;
1803 ovfs
.vfc_refcount
= vfsp
->vfc_refcount
;
1804 ovfs
.vfc_flags
= vfsp
->vfc_flags
;
1805 error
= SYSCTL_OUT(req
, &ovfs
, sizeof ovfs
);
1807 return error
; /* abort iteration with error code */
1809 return 0; /* continue iterating with next element */
1813 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS
)
1815 return vfsconf_each(sysctl_ovfs_conf_iter
, (void*)req
);
1818 #endif /* 1 || COMPAT_PRELITE2 */
1821 * Check to see if a filesystem is mounted on a block device.
1824 vfs_mountedon(struct vnode
*vp
)
1828 if ((dev
= vp
->v_rdev
) == NULL
) {
1829 /* if (vp->v_type != VBLK)
1830 dev = get_dev(vp->v_uminor, vp->v_umajor); */
1832 if (dev
!= NULL
&& dev
->si_mountpoint
)
1838 * Unmount all filesystems. The list is traversed in reverse order
1839 * of mounting to avoid dependencies.
1841 * We want the umountall to be able to break out of its loop if a
1842 * failure occurs, after scanning all possible mounts, so the callback
1843 * returns 0 on error.
1845 * NOTE: Do not call mountlist_remove(mp) on error any more, this will
1846 * confuse mountlist_scan()'s unbusy check.
1848 static int vfs_umountall_callback(struct mount
*mp
, void *data
);
1851 vfs_unmountall(void)
1856 count
= mountlist_scan(vfs_umountall_callback
,
1857 NULL
, MNTSCAN_REVERSE
|MNTSCAN_NOBUSY
);
1863 vfs_umountall_callback(struct mount
*mp
, void *data
)
1867 error
= dounmount(mp
, MNT_FORCE
);
1869 kprintf("unmount of filesystem mounted from %s failed (",
1870 mp
->mnt_stat
.f_mntfromname
);
1874 kprintf("%d)\n", error
);
1882 * Checks the mount flags for parameter mp and put the names comma-separated
1883 * into a string buffer buf with a size limit specified by len.
1885 * It returns the number of bytes written into buf, and (*errorp) will be
1886 * set to 0, EINVAL (if passed length is 0), or ENOSPC (supplied buffer was
1887 * not large enough). The buffer will be 0-terminated if len was not 0.
1890 vfs_flagstostr(int flags
, const struct mountctl_opt
*optp
,
1891 char *buf
, size_t len
, int *errorp
)
1893 static const struct mountctl_opt optnames
[] = {
1894 { MNT_RDONLY
, "read-only" },
1895 { MNT_SYNCHRONOUS
, "synchronous" },
1896 { MNT_NOEXEC
, "noexec" },
1897 { MNT_NOSUID
, "nosuid" },
1898 { MNT_NODEV
, "nodev" },
1899 { MNT_AUTOMOUNTED
, "automounted" },
1900 { MNT_ASYNC
, "asynchronous" },
1901 { MNT_SUIDDIR
, "suiddir" },
1902 { MNT_SOFTDEP
, "soft-updates" },
1903 { MNT_NOSYMFOLLOW
, "nosymfollow" },
1904 { MNT_TRIM
, "trim" },
1905 { MNT_NOATIME
, "noatime" },
1906 { MNT_NOCLUSTERR
, "noclusterr" },
1907 { MNT_NOCLUSTERW
, "noclusterw" },
1908 { MNT_EXRDONLY
, "NFS read-only" },
1909 { MNT_EXPORTED
, "NFS exported" },
1910 /* Remaining NFS flags could come here */
1911 { MNT_LOCAL
, "local" },
1912 { MNT_QUOTA
, "with-quotas" },
1913 /* { MNT_ROOTFS, "rootfs" }, */
1914 /* { MNT_IGNORE, "ignore" }, */
1924 bleft
= len
- 1; /* leave room for trailing \0 */
1927 * Checks the size of the string. If it contains
1928 * any data, then we will append the new flags to
1931 actsize
= strlen(buf
);
1935 /* Default flags if no flags passed */
1939 if (bleft
< 0) { /* degenerate case, 0-length buffer */
1944 for (; flags
&& optp
->o_opt
; ++optp
) {
1945 if ((flags
& optp
->o_opt
) == 0)
1947 optlen
= strlen(optp
->o_name
);
1948 if (bwritten
|| actsize
> 0) {
1953 buf
[bwritten
++] = ',';
1954 buf
[bwritten
++] = ' ';
1957 if (bleft
< optlen
) {
1961 bcopy(optp
->o_name
, buf
+ bwritten
, optlen
);
1964 flags
&= ~optp
->o_opt
;
1968 * Space already reserved for trailing \0
1975 * Build hash lists of net addresses and hang them off the mount point.
1976 * Called by ufs_mount() to set up the lists of export addresses.
1979 vfs_hang_addrlist(struct mount
*mp
, struct netexport
*nep
,
1980 const struct export_args
*argp
)
1983 struct radix_node_head
*rnh
;
1985 struct radix_node
*rn
;
1986 struct sockaddr
*saddr
, *smask
= NULL
;
1989 if (argp
->ex_addrlen
== 0) {
1990 if (mp
->mnt_flag
& MNT_DEFEXPORTED
)
1992 np
= &nep
->ne_defexported
;
1993 np
->netc_exflags
= argp
->ex_flags
;
1994 np
->netc_anon
= argp
->ex_anon
;
1995 np
->netc_anon
.cr_ref
= 1;
1996 mp
->mnt_flag
|= MNT_DEFEXPORTED
;
2000 if (argp
->ex_addrlen
< 0 || argp
->ex_addrlen
> MLEN
)
2002 if (argp
->ex_masklen
< 0 || argp
->ex_masklen
> MLEN
)
2005 i
= sizeof(struct netcred
) + argp
->ex_addrlen
+ argp
->ex_masklen
;
2006 np
= (struct netcred
*)kmalloc(i
, M_NETCRED
, M_WAITOK
| M_ZERO
);
2007 saddr
= (struct sockaddr
*) (np
+ 1);
2008 if ((error
= copyin(argp
->ex_addr
, (caddr_t
) saddr
, argp
->ex_addrlen
)))
2010 if (saddr
->sa_len
> argp
->ex_addrlen
)
2011 saddr
->sa_len
= argp
->ex_addrlen
;
2012 if (argp
->ex_masklen
) {
2013 smask
= (struct sockaddr
*)((caddr_t
)saddr
+ argp
->ex_addrlen
);
2014 error
= copyin(argp
->ex_mask
, (caddr_t
)smask
, argp
->ex_masklen
);
2017 if (smask
->sa_len
> argp
->ex_masklen
)
2018 smask
->sa_len
= argp
->ex_masklen
;
2021 if (nep
->ne_maskhead
== NULL
) {
2022 if (!rn_inithead((void **)&nep
->ne_maskhead
, NULL
, 0)) {
2027 if ((rnh
= vfs_create_addrlist_af(saddr
->sa_family
, nep
)) == NULL
) {
2031 rn
= (*rnh
->rnh_addaddr
)((char *)saddr
, (char *)smask
, rnh
,
2034 if (rn
== NULL
|| np
!= (struct netcred
*)rn
) { /* already exists */
2038 np
->netc_exflags
= argp
->ex_flags
;
2039 np
->netc_anon
= argp
->ex_anon
;
2040 np
->netc_anon
.cr_ref
= 1;
2044 kfree(np
, M_NETCRED
);
2049 * Free netcred structures installed in the netexport
2052 vfs_free_netcred(struct radix_node
*rn
, void *w
)
2054 struct radix_node_head
*rnh
= (struct radix_node_head
*)w
;
2056 (*rnh
->rnh_deladdr
) (rn
->rn_key
, rn
->rn_mask
, rnh
);
2057 kfree(rn
, M_NETCRED
);
2063 * callback to free an element of the mask table installed in the
2064 * netexport. These may be created indirectly and are not netcred
2068 vfs_free_netcred_mask(struct radix_node
*rn
, void *w
)
2070 struct radix_node_head
*rnh
= (struct radix_node_head
*)w
;
2072 (*rnh
->rnh_deladdr
) (rn
->rn_key
, rn
->rn_mask
, rnh
);
2073 kfree(rn
, M_RTABLE
);
2078 static struct radix_node_head
*
2079 vfs_create_addrlist_af(int af
, struct netexport
*nep
)
2081 struct radix_node_head
*rnh
= NULL
;
2082 #if defined(INET) || defined(INET6)
2083 struct radix_node_head
*maskhead
= nep
->ne_maskhead
;
2087 NE_ASSERT_LOCKED(nep
);
2088 KKASSERT(maskhead
!= NULL
);
2092 if ((rnh
= nep
->ne_inethead
) == NULL
) {
2093 off
= offsetof(struct sockaddr_in
, sin_addr
) << 3;
2094 if (!rn_inithead((void **)&rnh
, maskhead
, off
))
2096 nep
->ne_inethead
= rnh
;
2102 if ((rnh
= nep
->ne_inet6head
) == NULL
) {
2103 off
= offsetof(struct sockaddr_in6
, sin6_addr
) << 3;
2104 if (!rn_inithead((void **)&rnh
, maskhead
, off
))
2106 nep
->ne_inet6head
= rnh
;
2115 * helper function for freeing netcred elements
2118 vfs_free_addrlist_af(struct radix_node_head
**prnh
)
2120 struct radix_node_head
*rnh
= *prnh
;
2122 (*rnh
->rnh_walktree
) (rnh
, vfs_free_netcred
, rnh
);
2123 kfree(rnh
, M_RTABLE
);
2128 * helper function for freeing mask elements
2131 vfs_free_addrlist_masks(struct radix_node_head
**prnh
)
2133 struct radix_node_head
*rnh
= *prnh
;
2135 (*rnh
->rnh_walktree
) (rnh
, vfs_free_netcred_mask
, rnh
);
2136 kfree(rnh
, M_RTABLE
);
2141 * Free the net address hash lists that are hanging off the mount points.
2144 vfs_free_addrlist(struct netexport
*nep
)
2147 if (nep
->ne_inethead
!= NULL
)
2148 vfs_free_addrlist_af(&nep
->ne_inethead
);
2149 if (nep
->ne_inet6head
!= NULL
)
2150 vfs_free_addrlist_af(&nep
->ne_inet6head
);
2151 if (nep
->ne_maskhead
)
2152 vfs_free_addrlist_masks(&nep
->ne_maskhead
);
2157 vfs_export(struct mount
*mp
, struct netexport
*nep
,
2158 const struct export_args
*argp
)
2162 if (argp
->ex_flags
& MNT_DELEXPORT
) {
2163 if (mp
->mnt_flag
& MNT_EXPUBLIC
) {
2164 vfs_setpublicfs(NULL
, NULL
, NULL
);
2165 mp
->mnt_flag
&= ~MNT_EXPUBLIC
;
2167 vfs_free_addrlist(nep
);
2168 mp
->mnt_flag
&= ~(MNT_EXPORTED
| MNT_DEFEXPORTED
);
2170 if (argp
->ex_flags
& MNT_EXPORTED
) {
2171 if (argp
->ex_flags
& MNT_EXPUBLIC
) {
2172 if ((error
= vfs_setpublicfs(mp
, nep
, argp
)) != 0)
2174 mp
->mnt_flag
|= MNT_EXPUBLIC
;
2176 if ((error
= vfs_hang_addrlist(mp
, nep
, argp
)))
2178 mp
->mnt_flag
|= MNT_EXPORTED
;
2185 * Set the publicly exported filesystem (WebNFS). Currently, only
2186 * one public filesystem is possible in the spec (RFC 2054 and 2055)
2189 vfs_setpublicfs(struct mount
*mp
, struct netexport
*nep
,
2190 const struct export_args
*argp
)
2197 * mp == NULL -> invalidate the current info, the FS is
2198 * no longer exported. May be called from either vfs_export
2199 * or unmount, so check if it hasn't already been done.
2202 if (nfs_pub
.np_valid
) {
2203 nfs_pub
.np_valid
= 0;
2204 if (nfs_pub
.np_index
!= NULL
) {
2205 kfree(nfs_pub
.np_index
, M_TEMP
);
2206 nfs_pub
.np_index
= NULL
;
2213 * Only one allowed at a time.
2215 if (nfs_pub
.np_valid
!= 0 && mp
!= nfs_pub
.np_mount
)
2219 * Get real filehandle for root of exported FS.
2221 bzero((caddr_t
)&nfs_pub
.np_handle
, sizeof(nfs_pub
.np_handle
));
2222 nfs_pub
.np_handle
.fh_fsid
= mp
->mnt_stat
.f_fsid
;
2224 if ((error
= VFS_ROOT(mp
, &rvp
)))
2227 if ((error
= VFS_VPTOFH(rvp
, &nfs_pub
.np_handle
.fh_fid
)))
2233 * If an indexfile was specified, pull it in.
2235 if (argp
->ex_indexfile
!= NULL
) {
2238 error
= vn_get_namelen(rvp
, &namelen
);
2241 nfs_pub
.np_index
= kmalloc(namelen
, M_TEMP
, M_WAITOK
);
2242 error
= copyinstr(argp
->ex_indexfile
, nfs_pub
.np_index
,
2246 * Check for illegal filenames.
2248 for (cp
= nfs_pub
.np_index
; *cp
; cp
++) {
2256 kfree(nfs_pub
.np_index
, M_TEMP
);
2261 nfs_pub
.np_mount
= mp
;
2262 nfs_pub
.np_valid
= 1;
2267 vfs_export_lookup(struct mount
*mp
, struct netexport
*nep
,
2268 struct sockaddr
*nam
)
2271 struct radix_node_head
*rnh
;
2272 struct sockaddr
*saddr
;
2275 if (mp
->mnt_flag
& MNT_EXPORTED
) {
2277 * Lookup in the export list first.
2282 switch (saddr
->sa_family
) {
2285 rnh
= nep
->ne_inethead
;
2290 rnh
= nep
->ne_inet6head
;
2297 np
= (struct netcred
*)
2298 (*rnh
->rnh_matchaddr
)((char *)saddr
,
2300 if (np
&& np
->netc_rnodes
->rn_flags
& RNF_ROOT
)
2306 * If no address match, use the default if it exists.
2308 if (np
== NULL
&& mp
->mnt_flag
& MNT_DEFEXPORTED
)
2309 np
= &nep
->ne_defexported
;
2315 * perform msync on all vnodes under a mount point. The mount point must
2316 * be locked. This code is also responsible for lazy-freeing unreferenced
2317 * vnodes whos VM objects no longer contain pages.
2319 * NOTE: MNT_WAIT still skips vnodes in the VXLOCK state.
2321 * NOTE: XXX VOP_PUTPAGES and friends requires that the vnode be locked,
2322 * but vnode_pager_putpages() doesn't lock the vnode. We have to do it
2323 * way up in this high level function.
2325 static int vfs_msync_scan1(struct mount
*mp
, struct vnode
*vp
, void *data
);
2326 static int vfs_msync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
);
2329 vfs_msync(struct mount
*mp
, int flags
)
2334 * tmpfs sets this flag to prevent msync(), sync, and the
2335 * filesystem periodic syncer from trying to flush VM pages
2336 * to swap. Only pure memory pressure flushes tmpfs VM pages
2339 if (mp
->mnt_kern_flag
& MNTK_NOMSYNC
)
2343 * Ok, scan the vnodes for work. If the filesystem is using the
2344 * syncer thread feature we can use vsyncscan() instead of
2345 * vmntvnodescan(), which is much faster.
2347 vmsc_flags
= VMSC_GETVP
;
2348 if (flags
!= MNT_WAIT
)
2349 vmsc_flags
|= VMSC_NOWAIT
;
2351 if (mp
->mnt_kern_flag
& MNTK_THR_SYNC
) {
2352 vsyncscan(mp
, vmsc_flags
, vfs_msync_scan2
,
2353 (void *)(intptr_t)flags
);
2355 vmntvnodescan(mp
, vmsc_flags
,
2356 vfs_msync_scan1
, vfs_msync_scan2
,
2357 (void *)(intptr_t)flags
);
2362 * scan1 is a fast pre-check. There could be hundreds of thousands of
2363 * vnodes, we cannot afford to do anything heavy weight until we have a
2364 * fairly good indication that there is work to do.
2368 vfs_msync_scan1(struct mount
*mp
, struct vnode
*vp
, void *data
)
2370 int flags
= (int)(intptr_t)data
;
2372 if ((vp
->v_flag
& VRECLAIMED
) == 0) {
2373 if (vp
->v_auxrefs
== 0 && VREFCNT(vp
) <= 0 &&
2375 return(0); /* call scan2 */
2377 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0 &&
2378 (vp
->v_flag
& VOBJDIRTY
) &&
2379 (flags
== MNT_WAIT
|| vn_islocked(vp
) == 0)) {
2380 return(0); /* call scan2 */
2385 * do not call scan2, continue the loop
2391 * This callback is handed a locked vnode.
2395 vfs_msync_scan2(struct mount
*mp
, struct vnode
*vp
, void *data
)
2398 int flags
= (int)(intptr_t)data
;
2400 if (vp
->v_flag
& VRECLAIMED
)
2403 if ((mp
->mnt_flag
& MNT_RDONLY
) == 0 && (vp
->v_flag
& VOBJDIRTY
)) {
2404 if ((obj
= vp
->v_object
) != NULL
) {
2405 vm_object_page_clean(obj
, 0, 0,
2406 flags
== MNT_WAIT
? OBJPC_SYNC
: OBJPC_NOSYNC
);
2413 * Wake up anyone interested in vp because it is being revoked.
2416 vn_gone(struct vnode
*vp
)
2418 lwkt_gettoken(&vp
->v_token
);
2419 KNOTE(&vp
->v_pollinfo
.vpi_kqinfo
.ki_note
, NOTE_REVOKE
);
2420 lwkt_reltoken(&vp
->v_token
);
2424 * extract the cdev_t from a VBLK or VCHR. The vnode must have been opened
2425 * (or v_rdev might be NULL).
2428 vn_todev(struct vnode
*vp
)
2430 if (vp
->v_type
!= VBLK
&& vp
->v_type
!= VCHR
)
2432 KKASSERT(vp
->v_rdev
!= NULL
);
2433 return (vp
->v_rdev
);
2437 * Check if vnode represents a disk device. The vnode does not need to be
2443 vn_isdisk(struct vnode
*vp
, int *errp
)
2447 if (vp
->v_type
!= VCHR
) {
2460 if (dev_is_good(dev
) == 0) {
2465 if ((dev_dflags(dev
) & D_DISK
) == 0) {
2476 vn_get_namelen(struct vnode
*vp
, int *namelen
)
2479 register_t retval
[2];
2481 error
= VOP_PATHCONF(vp
, _PC_NAME_MAX
, retval
);
2484 *namelen
= (int)retval
[0];
2489 vop_write_dirent(int *error
, struct uio
*uio
, ino_t d_ino
, uint8_t d_type
,
2490 uint16_t d_namlen
, const char *d_name
)
2495 len
= _DIRENT_RECLEN(d_namlen
);
2496 if (len
> uio
->uio_resid
)
2499 dp
= kmalloc(len
, M_TEMP
, M_WAITOK
| M_ZERO
);
2502 dp
->d_namlen
= d_namlen
;
2503 dp
->d_type
= d_type
;
2504 bcopy(d_name
, dp
->d_name
, d_namlen
);
2506 *error
= uiomove((caddr_t
)dp
, len
, uio
);
2514 vn_mark_atime(struct vnode
*vp
, struct thread
*td
)
2516 struct proc
*p
= td
->td_proc
;
2517 struct ucred
*cred
= p
? p
->p_ucred
: proc0
.p_ucred
;
2519 if ((vp
->v_mount
->mnt_flag
& (MNT_NOATIME
| MNT_RDONLY
)) == 0) {
2520 VOP_MARKATIME(vp
, cred
);
2525 * Calculate the number of entries in an inode-related chained hash table.
2526 * With today's memory sizes, maxvnodes can wind up being a very large
2527 * number. There is no reason to waste memory, so tolerate some stacking.
2530 vfs_inodehashsize(void)
2535 while (hsize
< maxvnodes
)
2537 while (hsize
> maxvnodes
* 2)
2538 hsize
>>= 1; /* nominal 2x stacking */
2540 if (maxvnodes
> 1024 * 1024)
2541 hsize
>>= 1; /* nominal 8x stacking */
2543 if (maxvnodes
> 128 * 1024)
2544 hsize
>>= 1; /* nominal 4x stacking */