2 * Copyright (c) 1982, 1986, 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_vnops.c 8.2 (Berkeley) 1/21/94
35 * $FreeBSD: src/sys/kern/vfs_vnops.c,v 1.87.2.13 2002/12/29 18:19:53 dillon Exp $
38 #include <sys/param.h>
39 #include <sys/systm.h>
41 #include <sys/fcntl.h>
46 #include <sys/mount.h>
47 #include <sys/nlookup.h>
48 #include <sys/vnode.h>
50 #include <sys/filio.h>
51 #include <sys/ttycom.h>
53 #include <sys/sysctl.h>
54 #include <sys/syslog.h>
56 #include <sys/mplock2.h>
58 static int vn_closefile (struct file
*fp
);
59 static int vn_ioctl (struct file
*fp
, u_long com
, caddr_t data
,
60 struct ucred
*cred
, struct sysmsg
*msg
);
61 static int vn_read (struct file
*fp
, struct uio
*uio
,
62 struct ucred
*cred
, int flags
);
63 static int vn_kqfilter (struct file
*fp
, struct knote
*kn
);
64 static int vn_statfile (struct file
*fp
, struct stat
*sb
, struct ucred
*cred
);
65 static int vn_write (struct file
*fp
, struct uio
*uio
,
66 struct ucred
*cred
, int flags
);
68 struct fileops vnode_fileops
= {
72 .fo_kqfilter
= vn_kqfilter
,
73 .fo_stat
= vn_statfile
,
74 .fo_close
= vn_closefile
,
75 .fo_shutdown
= nofo_shutdown
79 * Common code for vnode open operations. Check permissions, and call
80 * the VOP_NOPEN or VOP_NCREATE routine.
82 * The caller is responsible for setting up nd with nlookup_init() and
83 * for cleaning it up with nlookup_done(), whether we return an error
86 * On success nd->nl_open_vp will hold a referenced and, if requested,
87 * locked vnode. A locked vnode is requested via NLC_LOCKVP. If fp
88 * is non-NULL the vnode will be installed in the file pointer.
90 * NOTE: If the caller wishes the namecache entry to be operated with
91 * a shared lock it must use NLC_SHAREDLOCK. If NLC_LOCKVP is set
92 * then the vnode lock will also be shared.
94 * NOTE: The vnode is referenced just once on return whether or not it
95 * is also installed in the file pointer.
98 vn_open(struct nlookupdata
*nd
, struct file
**fpp
, int fmode
, int cmode
)
100 struct file
*fp
= fpp
? *fpp
: NULL
;
102 struct ucred
*cred
= nd
->nl_cred
;
104 struct vattr
*vap
= &vat
;
112 * Certain combinations are illegal
114 if ((fmode
& (FWRITE
| O_TRUNC
)) == O_TRUNC
)
118 * Lookup the path and create or obtain the vnode. After a
119 * successful lookup a locked nd->nl_nch will be returned.
121 * The result of this section should be a locked vnode.
123 * XXX with only a little work we should be able to avoid locking
124 * the vnode if FWRITE, O_CREAT, and O_TRUNC are *not* set.
126 nd
->nl_flags
|= NLC_OPEN
;
127 if (fmode
& O_APPEND
)
128 nd
->nl_flags
|= NLC_APPEND
;
130 nd
->nl_flags
|= NLC_TRUNCATE
;
132 nd
->nl_flags
|= NLC_READ
;
134 nd
->nl_flags
|= NLC_WRITE
;
135 if ((fmode
& O_EXCL
) == 0 && (fmode
& O_NOFOLLOW
) == 0)
136 nd
->nl_flags
|= NLC_FOLLOW
;
138 if (fmode
& O_CREAT
) {
140 * CONDITIONAL CREATE FILE CASE
142 * Setting NLC_CREATE causes a negative hit to store
143 * the negative hit ncp and not return an error. Then
144 * nc_error or nc_vp may be checked to see if the ncp
145 * represents a negative hit. NLC_CREATE also requires
146 * write permission on the governing directory or EPERM
149 * If the file exists but is missing write permission,
150 * nlookup() returns EACCES. This has to be handled specially
151 * when combined with O_EXCL.
153 nd
->nl_flags
|= NLC_CREATE
;
154 nd
->nl_flags
|= NLC_REFDVP
;
157 if (error
== EACCES
&& nd
->nl_nch
.ncp
->nc_vp
!= NULL
&&
163 * If no error and nd->nl_dvp is NULL, the nlookup represents
164 * a mount-point or cross-mount situation. e.g.
165 * open("/var/cache", O_CREAT), where /var/cache is a
166 * mount point or a null-mount point.
168 if (error
== 0 && nd
->nl_dvp
== NULL
)
172 * NORMAL OPEN FILE CASE
181 * split case to allow us to re-resolve and retry the ncp in case
184 * (error is 0 on entry / retry)
188 * Checks for (likely) filesystem-modifying cases and allows
189 * the filesystem to stall the front-end.
191 if ((fmode
& (FWRITE
| O_TRUNC
)) ||
192 ((fmode
& O_CREAT
) && nd
->nl_nch
.ncp
->nc_vp
== NULL
)) {
193 error
= ncp_writechk(&nd
->nl_nch
);
199 if (fmode
& O_CREAT
) {
200 if (nd
->nl_nch
.ncp
->nc_vp
== NULL
) {
203 vap
->va_mode
= cmode
;
204 vap
->va_fuseflags
= fmode
; /* FUSE */
206 vap
->va_vaflags
|= VA_EXCLUSIVE
;
207 error
= VOP_NCREATE(&nd
->nl_nch
, nd
->nl_dvp
, &vp
,
212 /* locked vnode is returned */
214 if (fmode
& O_EXCL
) {
217 error
= cache_vget(&nd
->nl_nch
, cred
,
226 * In most other cases a shared lock on the vnode is
227 * sufficient. However, the O_RDWR case needs an
228 * exclusive lock if the vnode is executable. The
229 * NLC_EXCLLOCK_IFEXEC and NCF_NOTX flags help resolve
232 * NOTE: If NCF_NOTX is not set, we do not know the
233 * the state of the 'x' bits and have to get
234 * an exclusive lock for the EXCLLOCK_IFEXEC case.
236 if ((nd
->nl_flags
& NLC_SHAREDLOCK
) &&
237 ((nd
->nl_flags
& NLC_EXCLLOCK_IFEXEC
) == 0 ||
238 nd
->nl_nch
.ncp
->nc_flag
& NCF_NOTX
)) {
239 error
= cache_vget(&nd
->nl_nch
, cred
, LK_SHARED
, &vp
);
242 error
= cache_vget(&nd
->nl_nch
, cred
,
250 * We have a locked vnode and ncp now. Note that the ncp will
251 * be cleaned up by the caller if nd->nl_nch is left intact.
253 if (vp
->v_type
== VLNK
) {
257 if (vp
->v_type
== VSOCK
) {
261 if (vp
->v_type
!= VDIR
&& (fmode
& O_DIRECTORY
)) {
265 if ((fmode
& O_CREAT
) == 0) {
266 if (fmode
& (FWRITE
| O_TRUNC
)) {
267 if (vp
->v_type
== VDIR
) {
273 * Additional checks on vnode (does not substitute
274 * for ncp_writechk()).
276 error
= vn_writechk(vp
);
279 * Special stale handling, re-resolve the
282 if (error
== ESTALE
) {
288 cache_unlock(&nd
->nl_nch
);
289 cache_lock(&nd
->nl_nch
);
291 cache_setunresolved(&nd
->nl_nch
);
292 error
= cache_resolve(&nd
->nl_nch
,
302 if (fmode
& O_TRUNC
) {
303 vn_unlock(vp
); /* XXX */
304 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
); /* XXX */
305 osize
= vp
->v_filesize
;
308 error
= VOP_SETATTR_FP(vp
, vap
, cred
, fp
);
311 error
= VOP_GETATTR(vp
, vap
);
315 VFS_ACCOUNT(mp
, vap
->va_uid
, vap
->va_gid
, -osize
);
319 * Set or clear VNSWAPCACHE on the vp based on nd->nl_nch.ncp->nc_flag.
320 * These particular bits a tracked all the way from the root.
322 * NOTE: Might not work properly on NFS servers due to the
323 * disconnected namecache.
325 flags
= nd
->nl_nch
.ncp
->nc_flag
;
326 if ((flags
& (NCF_UF_CACHE
| NCF_UF_PCACHE
)) &&
327 (flags
& (NCF_SF_NOCACHE
| NCF_SF_PNOCACHE
)) == 0) {
328 vsetflags(vp
, VSWAPCACHE
);
330 vclrflags(vp
, VSWAPCACHE
);
334 * Setup the fp so VOP_OPEN can override it. No descriptor has been
335 * associated with the fp yet so we own it clean.
337 * f_nchandle inherits nl_nch. This used to be necessary only for
338 * directories but now we do it unconditionally so f*() ops
339 * such as fchmod() can access the actual namespace that was
340 * used to open the file.
343 if (nd
->nl_flags
& NLC_APPENDONLY
)
344 fmode
|= FAPPENDONLY
;
345 fp
->f_nchandle
= nd
->nl_nch
;
346 cache_zero(&nd
->nl_nch
);
347 cache_unlock(&fp
->f_nchandle
);
351 * Get rid of nl_nch. vn_open does not return it (it returns the
352 * vnode or the file pointer).
354 * NOTE: We can't leave nl_nch locked through the VOP_OPEN anyway
355 * since the VOP_OPEN may block, e.g. on /dev/ttyd0
357 * NOTE: The VOP_OPEN() can replace the *fpp we supply with its own
358 * (it will fdrop/fhold), and can also set the *fpp up however
359 * it wants, not necessarily using DTYPE_VNODE.
362 cache_put(&nd
->nl_nch
);
364 error
= VOP_OPEN(vp
, fmode
, cred
, fpp
);
365 fp
= fpp
? *fpp
: NULL
;
369 * setting f_ops to &badfileops will prevent the descriptor
370 * code from trying to close and release the vnode, since
371 * the open failed we do not want to call close.
375 fp
->f_ops
= &badfileops
;
382 * Assert that VREG files have been setup for vmio.
384 KASSERT(vp
->v_type
!= VREG
|| vp
->v_object
!= NULL
,
385 ("vn_open: regular file was not VMIO enabled!"));
389 * Return the vnode. XXX needs some cleaning up. The vnode is
390 * only returned in the fp == NULL case.
392 * NOTE: vnode stored in fp may be different
396 nd
->nl_vp_fmode
= fmode
;
397 if ((nd
->nl_flags
& NLC_LOCKVP
) == 0)
410 vn_opendisk(const char *devname
, int fmode
, struct vnode
**vpp
)
415 if (strncmp(devname
, "/dev/", 5) == 0)
417 if ((vp
= getsynthvnode(devname
)) == NULL
) {
420 error
= VOP_OPEN(vp
, fmode
, proc0
.p_ucred
, NULL
);
432 * Checks for special conditions on the vnode which might prevent writing
433 * after the vnode has (likely) been locked. The vnode might or might not
434 * be locked as of this call, but will be at least referenced.
436 * Also re-checks the mount RDONLY flag that ncp_writechk() checked prior
437 * to the vnode being locked.
440 vn_writechk(struct vnode
*vp
)
443 * If there's shared text associated with
444 * the vnode, try to free it up once. If
445 * we fail, we can't allow writing.
447 if (vp
->v_flag
& VTEXT
)
449 if (vp
->v_mount
&& (vp
->v_mount
->mnt_flag
& MNT_RDONLY
))
455 * Check whether the underlying mount is read-only. The mount point
456 * referenced by the namecache may be different from the mount point
457 * used by the underlying vnode in the case of NULLFS, so a separate
460 * Must be called PRIOR to any vnodes being locked.
463 ncp_writechk(struct nchandle
*nch
)
467 if ((mp
= nch
->mount
) != NULL
) {
468 if (mp
->mnt_flag
& MNT_RDONLY
)
470 if (mp
->mnt_op
->vfs_modifying
!= vfs_stdmodifying
)
482 vn_close(struct vnode
*vp
, int flags
, struct file
*fp
)
486 error
= vn_lock(vp
, LK_SHARED
| LK_RETRY
| LK_FAILRECLAIM
);
488 error
= VOP_CLOSE(vp
, flags
, fp
);
496 * Sequential heuristic.
498 * MPSAFE (f_seqcount and f_nextoff are allowed to race)
502 sequential_heuristic(struct uio
*uio
, struct file
*fp
)
505 * Sequential heuristic - detect sequential operation
507 * NOTE: SMP: We allow f_seqcount updates to race.
509 if ((uio
->uio_offset
== 0 && fp
->f_seqcount
> 0) ||
510 uio
->uio_offset
== fp
->f_nextoff
) {
511 int tmpseq
= fp
->f_seqcount
;
513 tmpseq
+= howmany(uio
->uio_resid
, MAXBSIZE
);
514 if (tmpseq
> IO_SEQMAX
)
516 fp
->f_seqcount
= tmpseq
;
517 return(fp
->f_seqcount
<< IO_SEQSHIFT
);
521 * Not sequential, quick draw-down of seqcount
523 * NOTE: SMP: We allow f_seqcount updates to race.
525 if (fp
->f_seqcount
> 1)
533 * get - lock and return the f_offset field.
534 * set - set and unlock the f_offset field.
536 * These routines serve the dual purpose of serializing access to the
537 * f_offset field (at least on x86) and guaranteeing operational integrity
538 * when multiple read()ers and write()ers are present on the same fp.
542 static __inline off_t
543 vn_get_fpf_offset(struct file
*fp
)
549 * Shortcut critical path.
551 flags
= fp
->f_flag
& ~FOFFSETLOCK
;
552 if (atomic_cmpset_int(&fp
->f_flag
, flags
, flags
| FOFFSETLOCK
))
553 return(fp
->f_offset
);
560 if (flags
& FOFFSETLOCK
) {
561 nflags
= flags
| FOFFSETWAKE
;
562 tsleep_interlock(&fp
->f_flag
, 0);
563 if (atomic_cmpset_int(&fp
->f_flag
, flags
, nflags
))
564 tsleep(&fp
->f_flag
, PINTERLOCKED
, "fpoff", 0);
566 nflags
= flags
| FOFFSETLOCK
;
567 if (atomic_cmpset_int(&fp
->f_flag
, flags
, nflags
))
571 return(fp
->f_offset
);
578 vn_set_fpf_offset(struct file
*fp
, off_t offset
)
584 * We hold the lock so we can set the offset without interference.
586 fp
->f_offset
= offset
;
589 * Normal release is already a reasonably critical path.
593 nflags
= flags
& ~(FOFFSETLOCK
| FOFFSETWAKE
);
594 if (atomic_cmpset_int(&fp
->f_flag
, flags
, nflags
)) {
595 if (flags
& FOFFSETWAKE
)
605 static __inline off_t
606 vn_poll_fpf_offset(struct file
*fp
)
608 #if defined(__x86_64__)
609 return(fp
->f_offset
);
611 off_t off
= vn_get_fpf_offset(fp
);
612 vn_set_fpf_offset(fp
, off
);
618 * Package up an I/O request on a vnode into a uio and do it.
623 vn_rdwr(enum uio_rw rw
, struct vnode
*vp
, caddr_t base
, int len
,
624 off_t offset
, enum uio_seg segflg
, int ioflg
,
625 struct ucred
*cred
, int *aresid
)
631 if ((ioflg
& IO_NODELOCKED
) == 0)
632 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
);
633 auio
.uio_iov
= &aiov
;
635 aiov
.iov_base
= base
;
637 auio
.uio_resid
= len
;
638 auio
.uio_offset
= offset
;
639 auio
.uio_segflg
= segflg
;
641 auio
.uio_td
= curthread
;
642 if (rw
== UIO_READ
) {
643 error
= VOP_READ(vp
, &auio
, ioflg
, cred
);
645 error
= VOP_WRITE(vp
, &auio
, ioflg
, cred
);
648 *aresid
= auio
.uio_resid
;
650 if (auio
.uio_resid
&& error
== 0)
652 if ((ioflg
& IO_NODELOCKED
) == 0)
658 * Package up an I/O request on a vnode into a uio and do it. The I/O
659 * request is split up into smaller chunks and we try to avoid saturating
660 * the buffer cache while potentially holding a vnode locked, so we
661 * check bwillwrite() before calling vn_rdwr(). We also call lwkt_user_yield()
662 * to give other processes a chance to lock the vnode (either other processes
663 * core'ing the same binary, or unrelated processes scanning the directory).
668 vn_rdwr_inchunks(enum uio_rw rw
, struct vnode
*vp
, caddr_t base
, int len
,
669 off_t offset
, enum uio_seg segflg
, int ioflg
,
670 struct ucred
*cred
, int *aresid
)
678 * Force `offset' to a multiple of MAXBSIZE except possibly
679 * for the first chunk, so that filesystems only need to
680 * write full blocks except possibly for the first and last
683 chunk
= MAXBSIZE
- (uoff_t
)offset
% MAXBSIZE
;
687 if (vp
->v_type
== VREG
&& (ioflg
& IO_RECURSE
) == 0) {
697 error
= vn_rdwr(rw
, vp
, base
, chunk
, offset
, segflg
,
698 ioflg
, cred
, aresid
);
699 len
-= chunk
; /* aresid calc already includes length */
712 * File pointers can no longer get ripped up by revoke so
713 * we don't need to lock access to the vp.
715 * f_offset updates are not guaranteed against multiple readers
718 vn_read(struct file
*fp
, struct uio
*uio
, struct ucred
*cred
, int flags
)
723 KASSERT(uio
->uio_td
== curthread
,
724 ("uio_td %p is not td %p", uio
->uio_td
, curthread
));
725 vp
= (struct vnode
*)fp
->f_data
;
728 if (flags
& O_FBLOCKING
) {
729 /* ioflag &= ~IO_NDELAY; */
730 } else if (flags
& O_FNONBLOCKING
) {
732 } else if (fp
->f_flag
& FNONBLOCK
) {
735 if (fp
->f_flag
& O_DIRECT
) {
738 if ((flags
& O_FOFFSET
) == 0 && (vp
->v_flag
& VNOTSEEKABLE
) == 0)
739 uio
->uio_offset
= vn_get_fpf_offset(fp
);
740 vn_lock(vp
, LK_SHARED
| LK_RETRY
);
741 ioflag
|= sequential_heuristic(uio
, fp
);
743 error
= VOP_READ_FP(vp
, uio
, ioflag
, cred
, fp
);
744 fp
->f_nextoff
= uio
->uio_offset
;
746 if ((flags
& O_FOFFSET
) == 0 && (vp
->v_flag
& VNOTSEEKABLE
) == 0)
747 vn_set_fpf_offset(fp
, uio
->uio_offset
);
755 vn_write(struct file
*fp
, struct uio
*uio
, struct ucred
*cred
, int flags
)
760 KASSERT(uio
->uio_td
== curthread
,
761 ("uio_td %p is not p %p", uio
->uio_td
, curthread
));
762 vp
= (struct vnode
*)fp
->f_data
;
765 if (vp
->v_type
== VREG
&&
766 ((fp
->f_flag
& O_APPEND
) || (flags
& O_FAPPEND
))) {
770 if (flags
& O_FBLOCKING
) {
771 /* ioflag &= ~IO_NDELAY; */
772 } else if (flags
& O_FNONBLOCKING
) {
774 } else if (fp
->f_flag
& FNONBLOCK
) {
777 if (fp
->f_flag
& O_DIRECT
) {
780 if (flags
& O_FASYNCWRITE
) {
781 /* ioflag &= ~IO_SYNC; */
782 } else if (flags
& O_FSYNCWRITE
) {
784 } else if (fp
->f_flag
& O_FSYNC
) {
788 if (vp
->v_mount
&& (vp
->v_mount
->mnt_flag
& MNT_SYNCHRONOUS
))
790 if ((flags
& O_FOFFSET
) == 0)
791 uio
->uio_offset
= vn_get_fpf_offset(fp
);
793 VFS_MODIFYING(vp
->v_mount
);
794 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
);
795 ioflag
|= sequential_heuristic(uio
, fp
);
796 error
= VOP_WRITE_FP(vp
, uio
, ioflag
, cred
, fp
);
797 fp
->f_nextoff
= uio
->uio_offset
;
799 if ((flags
& O_FOFFSET
) == 0)
800 vn_set_fpf_offset(fp
, uio
->uio_offset
);
808 vn_statfile(struct file
*fp
, struct stat
*sb
, struct ucred
*cred
)
813 vp
= (struct vnode
*)fp
->f_data
;
814 error
= vn_stat(vp
, sb
, cred
);
822 vn_stat(struct vnode
*vp
, struct stat
*sb
, struct ucred
*cred
)
831 * vp already has a ref and is validated, can call unlocked.
834 error
= VOP_GETATTR(vp
, vap
);
839 * Zero the spare stat fields
845 * Copy from vattr table
847 if (vap
->va_fsid
!= VNOVAL
)
848 sb
->st_dev
= vap
->va_fsid
;
850 sb
->st_dev
= vp
->v_mount
->mnt_stat
.f_fsid
.val
[0];
851 sb
->st_ino
= vap
->va_fileid
;
853 switch (vap
->va_type
) {
871 /* This is a cosmetic change, symlinks do not have a mode. */
872 if (vp
->v_mount
->mnt_flag
& MNT_NOSYMFOLLOW
)
873 sb
->st_mode
&= ~ACCESSPERMS
; /* 0000 */
875 sb
->st_mode
|= ACCESSPERMS
; /* 0777 */
887 if (vap
->va_nlink
> (nlink_t
)-1)
888 sb
->st_nlink
= (nlink_t
)-1;
890 sb
->st_nlink
= vap
->va_nlink
;
891 sb
->st_uid
= vap
->va_uid
;
892 sb
->st_gid
= vap
->va_gid
;
893 sb
->st_rdev
= devid_from_dev(vp
->v_rdev
);
894 sb
->st_size
= vap
->va_size
;
895 sb
->st_atimespec
= vap
->va_atime
;
896 sb
->st_mtimespec
= vap
->va_mtime
;
897 sb
->st_ctimespec
= vap
->va_ctime
;
900 * A VCHR and VBLK device may track the last access and last modified
901 * time independantly of the filesystem. This is particularly true
902 * because device read and write calls may bypass the filesystem.
904 if (vp
->v_type
== VCHR
|| vp
->v_type
== VBLK
) {
907 if (dev
->si_lastread
) {
908 sb
->st_atimespec
.tv_sec
= time_second
+
911 sb
->st_atimespec
.tv_nsec
= 0;
913 if (dev
->si_lastwrite
) {
914 sb
->st_mtimespec
.tv_sec
= time_second
+
917 sb
->st_mtimespec
.tv_nsec
= 0;
923 * According to www.opengroup.org, the meaning of st_blksize is
924 * "a filesystem-specific preferred I/O block size for this
925 * object. In some filesystem types, this may vary from file
927 * Default to PAGE_SIZE after much discussion.
930 if (vap
->va_type
== VREG
) {
931 sb
->st_blksize
= vap
->va_blocksize
;
932 } else if (vn_isdisk(vp
, NULL
)) {
934 * XXX this is broken. If the device is not yet open (aka
935 * stat() call, aka v_rdev == NULL), how are we supposed
936 * to get a valid block size out of it?
940 sb
->st_blksize
= dev
->si_bsize_best
;
941 if (sb
->st_blksize
< dev
->si_bsize_phys
)
942 sb
->st_blksize
= dev
->si_bsize_phys
;
943 if (sb
->st_blksize
< BLKDEV_IOSIZE
)
944 sb
->st_blksize
= BLKDEV_IOSIZE
;
946 sb
->st_blksize
= PAGE_SIZE
;
949 sb
->st_flags
= vap
->va_flags
;
951 error
= priv_check_cred(cred
, PRIV_VFS_GENERATION
, 0);
955 sb
->st_gen
= (u_int32_t
)vap
->va_gen
;
957 sb
->st_blocks
= vap
->va_bytes
/ S_BLKSIZE
;
960 * This is for ABI compatibility <= 5.7 (for ABI change made in
963 sb
->__old_st_blksize
= sb
->st_blksize
;
969 * MPALMOSTSAFE - acquires mplock
972 vn_ioctl(struct file
*fp
, u_long com
, caddr_t data
, struct ucred
*ucred
,
975 struct vnode
*vp
= ((struct vnode
*)fp
->f_data
);
981 switch (vp
->v_type
) {
984 if (com
== FIONREAD
) {
985 error
= VOP_GETATTR(vp
, &vattr
);
988 size
= vattr
.va_size
;
989 if ((vp
->v_flag
& VNOTSEEKABLE
) == 0)
990 size
-= vn_poll_fpf_offset(fp
);
991 if (size
> 0x7FFFFFFF)
997 if (com
== FIOASYNC
) { /* XXX */
1009 if (com
== FIODTYPE
) {
1010 if (vp
->v_type
!= VCHR
&& vp
->v_type
!= VBLK
) {
1014 *(int *)data
= dev_dflags(vp
->v_rdev
) & D_TYPEMASK
;
1018 error
= VOP_IOCTL(vp
, com
, data
, fp
->f_flag
, ucred
, msg
);
1019 if (error
== 0 && com
== TIOCSCTTY
) {
1020 struct proc
*p
= curthread
->td_proc
;
1021 struct session
*sess
;
1029 sess
= p
->p_session
;
1030 /* Do nothing if reassigning same control tty */
1031 if (sess
->s_ttyvp
== vp
) {
1037 /* Get rid of reference to old control tty */
1038 ovp
= sess
->s_ttyvp
;
1051 * Obtain the requested vnode lock
1053 * LK_RETRY Automatically retry on timeout
1054 * LK_FAILRECLAIM Fail if the vnode is being reclaimed
1056 * Failures will occur if the vnode is undergoing recyclement, but not
1057 * all callers expect that the function will fail so the caller must pass
1058 * LK_FAILOK if it wants to process an error code.
1060 * Errors can occur for other reasons if you pass in other LK_ flags,
1061 * regardless of whether you pass in LK_FAILRECLAIM
1064 vn_lock(struct vnode
*vp
, int flags
)
1069 error
= lockmgr(&vp
->v_lock
, flags
);
1072 } while (flags
& LK_RETRY
);
1075 * Because we (had better!) have a ref on the vnode, once it
1076 * goes to VRECLAIMED state it will not be recycled until all
1077 * refs go away. So we can just check the flag.
1079 if (error
== 0 && (vp
->v_flag
& VRECLAIMED
)) {
1080 if (flags
& LK_FAILRECLAIM
) {
1081 lockmgr(&vp
->v_lock
, LK_RELEASE
);
1089 vn_relock(struct vnode
*vp
, int flags
)
1094 error
= lockmgr(&vp
->v_lock
, flags
);
1097 } while (flags
& LK_RETRY
);
1102 #ifdef DEBUG_VN_UNLOCK
1105 debug_vn_unlock(struct vnode
*vp
, const char *filename
, int line
)
1107 kprintf("vn_unlock from %s:%d\n", filename
, line
);
1108 lockmgr(&vp
->v_lock
, LK_RELEASE
);
1114 vn_unlock(struct vnode
*vp
)
1116 lockmgr(&vp
->v_lock
, LK_RELEASE
);
1125 vn_islocked(struct vnode
*vp
)
1127 return (lockstatus(&vp
->v_lock
, curthread
));
1131 * Return the lock status of a vnode and unlock the vnode
1132 * if we owned the lock. This is not a boolean, if the
1133 * caller cares what the lock status is the caller must
1134 * check the various possible values.
1136 * This only unlocks exclusive locks held by the caller,
1137 * it will NOT unlock shared locks (there is no way to
1138 * tell who the shared lock belongs to).
1143 vn_islocked_unlock(struct vnode
*vp
)
1147 vpls
= lockstatus(&vp
->v_lock
, curthread
);
1148 if (vpls
== LK_EXCLUSIVE
)
1149 lockmgr(&vp
->v_lock
, LK_RELEASE
);
1154 * Restore a vnode lock that we previously released via
1155 * vn_islocked_unlock(). This is a NOP if we did not
1156 * own the original lock.
1161 vn_islocked_relock(struct vnode
*vp
, int vpls
)
1165 if (vpls
== LK_EXCLUSIVE
)
1166 error
= lockmgr(&vp
->v_lock
, vpls
);
1173 vn_closefile(struct file
*fp
)
1177 fp
->f_ops
= &badfileops
;
1178 error
= vn_close(((struct vnode
*)fp
->f_data
), fp
->f_flag
, fp
);
1186 vn_kqfilter(struct file
*fp
, struct knote
*kn
)
1190 error
= VOP_KQFILTER(((struct vnode
*)fp
->f_data
), kn
);