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 * 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_vnops.c 8.2 (Berkeley) 1/21/94
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/fcntl.h>
48 #include <sys/limits.h>
50 #include <sys/mount.h>
51 #include <sys/mutex.h>
52 #include <sys/namei.h>
53 #include <sys/vnode.h>
56 #include <sys/filio.h>
57 #include <sys/resourcevar.h>
59 #include <sys/ttycom.h>
61 #include <sys/syslog.h>
62 #include <sys/unistd.h>
64 #include <security/mac/mac_framework.h>
67 #include <vm/vm_object.h>
69 static fo_rdwr_t vn_read;
70 static fo_rdwr_t vn_write;
71 static fo_truncate_t vn_truncate;
72 static fo_ioctl_t vn_ioctl;
73 static fo_poll_t vn_poll;
74 static fo_kqfilter_t vn_kqfilter;
75 static fo_stat_t vn_statfile;
76 static fo_close_t vn_closefile;
78 struct fileops vnops = {
81 .fo_truncate = vn_truncate,
84 .fo_kqfilter = vn_kqfilter,
85 .fo_stat = vn_statfile,
86 .fo_close = vn_closefile,
87 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
91 vn_open(ndp, flagp, cmode, fp)
92 struct nameidata *ndp;
96 struct thread *td = ndp->ni_cnd.cn_thread;
98 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
102 * Common code for vnode open operations.
103 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
105 * Note that this does NOT free nameidata for the successful case,
106 * due to the NDINIT being done elsewhere.
109 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
110 struct ucred *cred, struct file *fp)
114 struct thread *td = ndp->ni_cnd.cn_thread;
116 struct vattr *vap = &vat;
119 int vfslocked, mpsafe;
121 mpsafe = ndp->ni_cnd.cn_flags & MPSAFE;
125 if (fmode & O_CREAT) {
126 ndp->ni_cnd.cn_nameiop = CREATE;
127 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF |
129 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
130 ndp->ni_cnd.cn_flags |= FOLLOW;
131 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
132 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
134 if ((error = namei(ndp)) != 0)
136 vfslocked = NDHASGIANT(ndp);
138 ndp->ni_cnd.cn_flags &= ~MPSAFE;
139 if (ndp->ni_vp == NULL) {
142 vap->va_mode = cmode;
144 vap->va_vaflags |= VA_EXCLUSIVE;
145 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
146 NDFREE(ndp, NDF_ONLY_PNBUF);
148 VFS_UNLOCK_GIANT(vfslocked);
149 if ((error = vn_start_write(NULL, &mp,
150 V_XSLEEP | PCATCH)) != 0)
155 error = mac_vnode_check_create(cred, ndp->ni_dvp,
159 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
162 vn_finished_write(mp);
164 VFS_UNLOCK_GIANT(vfslocked);
165 NDFREE(ndp, NDF_ONLY_PNBUF);
171 if (ndp->ni_dvp == ndp->ni_vp)
177 if (fmode & O_EXCL) {
184 ndp->ni_cnd.cn_nameiop = LOOKUP;
185 ndp->ni_cnd.cn_flags = ISOPEN |
186 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) |
188 if (!(fmode & FWRITE))
189 ndp->ni_cnd.cn_flags |= LOCKSHARED;
190 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
191 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
192 if ((error = namei(ndp)) != 0)
195 ndp->ni_cnd.cn_flags &= ~MPSAFE;
196 vfslocked = NDHASGIANT(ndp);
199 if (vp->v_type == VLNK) {
203 if (vp->v_type == VSOCK) {
208 if (fmode & (FWRITE | O_TRUNC)) {
209 if (vp->v_type == VDIR) {
219 if ((fmode & O_APPEND) && (fmode & FWRITE))
222 error = mac_vnode_check_open(cred, vp, accmode);
226 if ((fmode & O_CREAT) == 0) {
227 if (accmode & VWRITE) {
228 error = vn_writechk(vp);
233 error = VOP_ACCESS(vp, accmode, cred, td);
238 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
244 ASSERT_VOP_LOCKED(vp, "vn_open_cred");
246 VFS_UNLOCK_GIANT(vfslocked);
249 NDFREE(ndp, NDF_ONLY_PNBUF);
251 VFS_UNLOCK_GIANT(vfslocked);
258 * Check for write permissions on the specified vnode.
259 * Prototype text segments cannot be written.
263 register struct vnode *vp;
266 ASSERT_VOP_LOCKED(vp, "vn_writechk");
268 * If there's shared text associated with
269 * the vnode, try to free it up once. If
270 * we fail, we can't allow writing.
272 if (vp->v_vflag & VV_TEXT)
282 vn_close(vp, flags, file_cred, td)
283 register struct vnode *vp;
285 struct ucred *file_cred;
289 int error, lock_flags;
291 if (!(flags & FWRITE) && vp->v_mount != NULL &&
292 vp->v_mount->mnt_kern_flag & MNTK_EXTENDED_SHARED)
293 lock_flags = LK_SHARED;
295 lock_flags = LK_EXCLUSIVE;
297 VFS_ASSERT_GIANT(vp->v_mount);
299 vn_start_write(vp, &mp, V_WAIT);
300 vn_lock(vp, lock_flags | LK_RETRY);
301 if (flags & FWRITE) {
302 VNASSERT(vp->v_writecount > 0, vp,
303 ("vn_close: negative writecount"));
306 error = VOP_CLOSE(vp, flags, file_cred, td);
308 vn_finished_write(mp);
313 * Heuristic to detect sequential operation.
316 sequential_heuristic(struct uio *uio, struct file *fp)
319 if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
320 return (fp->f_seqcount << IO_SEQSHIFT);
323 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
324 * that the first I/O is normally considered to be slightly
325 * sequential. Seeking to offset 0 doesn't change sequentiality
326 * unless previous seeks have reduced f_seqcount to 0, in which
327 * case offset 0 is not special.
329 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
330 uio->uio_offset == fp->f_nextoff) {
332 * f_seqcount is in units of fixed-size blocks so that it
333 * depends mainly on the amount of sequential I/O and not
334 * much on the number of sequential I/O's. The fixed size
335 * of 16384 is hard-coded here since it is (not quite) just
336 * a magic size that works well here. This size is more
337 * closely related to the best I/O size for real disks than
338 * to any block size used by software.
340 fp->f_seqcount += howmany(uio->uio_resid, 16384);
341 if (fp->f_seqcount > IO_SEQMAX)
342 fp->f_seqcount = IO_SEQMAX;
343 return (fp->f_seqcount << IO_SEQSHIFT);
346 /* Not sequential. Quickly draw-down sequentiality. */
347 if (fp->f_seqcount > 1)
355 * Package up an I/O request on a vnode into a uio and do it.
358 vn_rdwr(rw, vp, base, len, offset, segflg, ioflg, active_cred, file_cred,
367 struct ucred *active_cred;
368 struct ucred *file_cred;
376 int error, lock_flags;
378 VFS_ASSERT_GIANT(vp->v_mount);
380 if ((ioflg & IO_NODELOCKED) == 0) {
382 if (rw == UIO_WRITE) {
383 if (vp->v_type != VCHR &&
384 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
387 if (MNT_SHARED_WRITES(mp) ||
388 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) {
389 lock_flags = LK_SHARED;
391 lock_flags = LK_EXCLUSIVE;
393 vn_lock(vp, lock_flags | LK_RETRY);
395 vn_lock(vp, LK_SHARED | LK_RETRY);
398 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
399 auio.uio_iov = &aiov;
401 aiov.iov_base = base;
403 auio.uio_resid = len;
404 auio.uio_offset = offset;
405 auio.uio_segflg = segflg;
410 if ((ioflg & IO_NOMACCHECK) == 0) {
412 error = mac_vnode_check_read(active_cred, file_cred,
415 error = mac_vnode_check_write(active_cred, file_cred,
425 error = VOP_READ(vp, &auio, ioflg, cred);
427 error = VOP_WRITE(vp, &auio, ioflg, cred);
430 *aresid = auio.uio_resid;
432 if (auio.uio_resid && error == 0)
434 if ((ioflg & IO_NODELOCKED) == 0) {
435 if (rw == UIO_WRITE && vp->v_type != VCHR)
436 vn_finished_write(mp);
443 * Package up an I/O request on a vnode into a uio and do it. The I/O
444 * request is split up into smaller chunks and we try to avoid saturating
445 * the buffer cache while potentially holding a vnode locked, so we
446 * check bwillwrite() before calling vn_rdwr(). We also call uio_yield()
447 * to give other processes a chance to lock the vnode (either other processes
448 * core'ing the same binary, or unrelated processes scanning the directory).
451 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
452 file_cred, aresid, td)
460 struct ucred *active_cred;
461 struct ucred *file_cred;
468 VFS_ASSERT_GIANT(vp->v_mount);
474 * Force `offset' to a multiple of MAXBSIZE except possibly
475 * for the first chunk, so that filesystems only need to
476 * write full blocks except possibly for the first and last
479 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
483 if (rw != UIO_READ && vp->v_type == VREG)
486 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
487 ioflg, active_cred, file_cred, &iaresid, td);
488 len -= chunk; /* aresid calc already includes length */
492 base = (char *)base + chunk;
496 *aresid = len + iaresid;
501 * File table vnode read routine.
504 vn_read(fp, uio, active_cred, flags, td)
507 struct ucred *active_cred;
514 int advice, vfslocked;
515 off_t offset, start, end;
517 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
522 if (fp->f_flag & FNONBLOCK)
524 if (fp->f_flag & O_DIRECT)
526 advice = POSIX_FADV_NORMAL;
527 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
529 * According to McKusick the vn lock was protecting f_offset here.
530 * It is now protected by the FOFFSET_LOCKED flag.
532 if ((flags & FOF_OFFSET) == 0 || fp->f_advice != NULL) {
533 mtxp = mtx_pool_find(mtxpool_sleep, fp);
535 if ((flags & FOF_OFFSET) == 0) {
536 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
537 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
538 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
539 "vnread offlock", 0);
541 fp->f_vnread_flags |= FOFFSET_LOCKED;
542 uio->uio_offset = fp->f_offset;
544 if (fp->f_advice != NULL &&
545 uio->uio_offset >= fp->f_advice->fa_start &&
546 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
547 advice = fp->f_advice->fa_advice;
550 vn_lock(vp, LK_SHARED | LK_RETRY);
553 case POSIX_FADV_NORMAL:
554 case POSIX_FADV_SEQUENTIAL:
555 case POSIX_FADV_NOREUSE:
556 ioflag |= sequential_heuristic(uio, fp);
558 case POSIX_FADV_RANDOM:
559 /* Disable read-ahead for random I/O. */
562 offset = uio->uio_offset;
565 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
568 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
569 if ((flags & FOF_OFFSET) == 0) {
570 fp->f_offset = uio->uio_offset;
572 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
573 wakeup(&fp->f_vnread_flags);
574 fp->f_vnread_flags = 0;
577 fp->f_nextoff = uio->uio_offset;
579 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
580 offset != uio->uio_offset) {
582 * Use POSIX_FADV_DONTNEED to flush clean pages and
583 * buffers for the backing file after a
584 * POSIX_FADV_NOREUSE read(2). To optimize the common
585 * case of using POSIX_FADV_NOREUSE with sequential
586 * access, track the previous implicit DONTNEED
587 * request and grow this request to include the
588 * current read(2) in addition to the previous
589 * DONTNEED. With purely sequential access this will
590 * cause the DONTNEED requests to continously grow to
591 * cover all of the previously read regions of the
592 * file. This allows filesystem blocks that are
593 * accessed by multiple calls to read(2) to be flushed
594 * once the last read(2) finishes.
597 end = uio->uio_offset - 1;
599 if (fp->f_advice != NULL &&
600 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
601 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
602 start = fp->f_advice->fa_prevstart;
603 else if (fp->f_advice->fa_prevstart != 0 &&
604 fp->f_advice->fa_prevstart == end + 1)
605 end = fp->f_advice->fa_prevend;
606 fp->f_advice->fa_prevstart = start;
607 fp->f_advice->fa_prevend = end;
610 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
612 VFS_UNLOCK_GIANT(vfslocked);
617 * File table vnode write routine.
620 vn_write(fp, uio, active_cred, flags, td)
623 struct ucred *active_cred;
629 int error, ioflag, lock_flags;
631 int advice, vfslocked;
632 off_t offset, start, end;
634 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
637 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
638 if (vp->v_type == VREG)
641 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
643 if (fp->f_flag & FNONBLOCK)
645 if (fp->f_flag & O_DIRECT)
647 if ((fp->f_flag & O_FSYNC) ||
648 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
651 if (vp->v_type != VCHR &&
652 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
655 if ((MNT_SHARED_WRITES(mp) ||
656 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) &&
657 (flags & FOF_OFFSET) != 0) {
658 lock_flags = LK_SHARED;
660 lock_flags = LK_EXCLUSIVE;
663 vn_lock(vp, lock_flags | LK_RETRY);
664 if ((flags & FOF_OFFSET) == 0)
665 uio->uio_offset = fp->f_offset;
666 advice = POSIX_FADV_NORMAL;
668 if (fp->f_advice != NULL) {
669 mtxp = mtx_pool_find(mtxpool_sleep, fp);
671 if (fp->f_advice != NULL &&
672 uio->uio_offset >= fp->f_advice->fa_start &&
673 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
674 advice = fp->f_advice->fa_advice;
678 case POSIX_FADV_NORMAL:
679 case POSIX_FADV_SEQUENTIAL:
680 case POSIX_FADV_NOREUSE:
681 ioflag |= sequential_heuristic(uio, fp);
683 case POSIX_FADV_RANDOM:
684 /* XXX: Is this correct? */
687 offset = uio->uio_offset;
690 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
693 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
694 if ((flags & FOF_OFFSET) == 0)
695 fp->f_offset = uio->uio_offset;
696 fp->f_nextoff = uio->uio_offset;
698 if (vp->v_type != VCHR)
699 vn_finished_write(mp);
700 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
701 offset != uio->uio_offset) {
703 * Use POSIX_FADV_DONTNEED to flush clean pages and
704 * buffers for the backing file after a
705 * POSIX_FADV_NOREUSE write(2). To optimize the
706 * common case of using POSIX_FADV_NOREUSE with
707 * sequential access, track the previous implicit
708 * DONTNEED request and grow this request to include
709 * the current write(2) in addition to the previous
710 * DONTNEED. With purely sequential access this will
711 * cause the DONTNEED requests to continously grow to
712 * cover all of the previously written regions of the
715 * Note that the blocks just written are almost
716 * certainly still dirty, so this only works when
717 * VOP_ADVISE() calls from subsequent writes push out
718 * the data written by this write(2) once the backing
719 * buffers are clean. However, as compared to forcing
720 * IO_DIRECT, this gives much saner behavior. Write
721 * clustering is still allowed, and clean pages are
722 * merely moved to the cache page queue rather than
723 * outright thrown away. This means a subsequent
724 * read(2) can still avoid hitting the disk if the
725 * pages have not been reclaimed.
727 * This does make POSIX_FADV_NOREUSE largely useless
728 * with non-sequential access. However, sequential
729 * access is the more common use case and the flag is
733 end = uio->uio_offset - 1;
735 if (fp->f_advice != NULL &&
736 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
737 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
738 start = fp->f_advice->fa_prevstart;
739 else if (fp->f_advice->fa_prevstart != 0 &&
740 fp->f_advice->fa_prevstart == end + 1)
741 end = fp->f_advice->fa_prevend;
742 fp->f_advice->fa_prevstart = start;
743 fp->f_advice->fa_prevend = end;
746 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
750 VFS_UNLOCK_GIANT(vfslocked);
755 * File table truncate routine.
758 vn_truncate(fp, length, active_cred, td)
761 struct ucred *active_cred;
771 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
772 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
774 VFS_UNLOCK_GIANT(vfslocked);
777 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
778 if (vp->v_type == VDIR) {
783 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
787 error = vn_writechk(vp);
790 vattr.va_size = length;
791 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
795 vn_finished_write(mp);
796 VFS_UNLOCK_GIANT(vfslocked);
801 * File table vnode stat routine.
804 vn_statfile(fp, sb, active_cred, td)
807 struct ucred *active_cred;
810 struct vnode *vp = fp->f_vnode;
814 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
815 vn_lock(vp, LK_SHARED | LK_RETRY);
816 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
818 VFS_UNLOCK_GIANT(vfslocked);
824 * Stat a vnode; implementation for the stat syscall
827 vn_stat(vp, sb, active_cred, file_cred, td)
829 register struct stat *sb;
830 struct ucred *active_cred;
831 struct ucred *file_cred;
835 register struct vattr *vap;
840 error = mac_vnode_check_stat(active_cred, file_cred, vp);
848 * Initialize defaults for new and unusual fields, so that file
849 * systems which don't support these fields don't need to know
852 vap->va_birthtime.tv_sec = -1;
853 vap->va_birthtime.tv_nsec = 0;
854 vap->va_fsid = VNOVAL;
855 vap->va_rdev = NODEV;
857 error = VOP_GETATTR(vp, vap, active_cred);
862 * Zero the spare stat fields
864 bzero(sb, sizeof *sb);
867 * Copy from vattr table
869 if (vap->va_fsid != VNOVAL)
870 sb->st_dev = vap->va_fsid;
872 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
873 sb->st_ino = vap->va_fileid;
875 switch (vap->va_type) {
901 sb->st_nlink = vap->va_nlink;
902 sb->st_uid = vap->va_uid;
903 sb->st_gid = vap->va_gid;
904 sb->st_rdev = vap->va_rdev;
905 if (vap->va_size > OFF_MAX)
907 sb->st_size = vap->va_size;
908 sb->st_atimespec = vap->va_atime;
909 sb->st_mtimespec = vap->va_mtime;
910 sb->st_ctimespec = vap->va_ctime;
911 sb->st_birthtimespec = vap->va_birthtime;
914 * According to www.opengroup.org, the meaning of st_blksize is
915 * "a filesystem-specific preferred I/O block size for this
916 * object. In some filesystem types, this may vary from file
918 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
921 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
923 sb->st_flags = vap->va_flags;
924 if (priv_check(td, PRIV_VFS_GENERATION))
927 sb->st_gen = vap->va_gen;
929 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
934 * File table vnode ioctl routine.
937 vn_ioctl(fp, com, data, active_cred, td)
941 struct ucred *active_cred;
944 struct vnode *vp = fp->f_vnode;
949 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
951 switch (vp->v_type) {
954 if (com == FIONREAD) {
955 vn_lock(vp, LK_SHARED | LK_RETRY);
956 error = VOP_GETATTR(vp, &vattr, active_cred);
959 *(int *)data = vattr.va_size - fp->f_offset;
960 } else if (com == FIONBIO || com == FIOASYNC) /* XXX */
963 error = VOP_IOCTL(vp, com, data, fp->f_flag,
970 VFS_UNLOCK_GIANT(vfslocked);
975 * File table vnode poll routine.
978 vn_poll(fp, events, active_cred, td)
981 struct ucred *active_cred;
989 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
991 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
992 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
997 error = VOP_POLL(vp, events, fp->f_cred, td);
998 VFS_UNLOCK_GIANT(vfslocked);
1003 * Acquire the requested lock and then check for validity. LK_RETRY
1004 * permits vn_lock to return doomed vnodes.
1007 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1011 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1012 ("vn_lock called with no locktype."));
1014 #ifdef DEBUG_VFS_LOCKS
1015 KASSERT(vp->v_holdcnt != 0,
1016 ("vn_lock %p: zero hold count", vp));
1018 error = VOP_LOCK1(vp, flags, file, line);
1019 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1020 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1021 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1024 * Callers specify LK_RETRY if they wish to get dead vnodes.
1025 * If RETRY is not set, we return ENOENT instead.
1027 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1028 (flags & LK_RETRY) == 0) {
1033 } while (flags & LK_RETRY && error != 0);
1038 * File table vnode close routine.
1041 vn_closefile(fp, td)
1052 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1053 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1054 lf.l_whence = SEEK_SET;
1057 lf.l_type = F_UNLCK;
1058 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1061 fp->f_ops = &badfileops;
1063 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1064 VFS_UNLOCK_GIANT(vfslocked);
1069 * Preparing to start a filesystem write operation. If the operation is
1070 * permitted, then we bump the count of operations in progress and
1071 * proceed. If a suspend request is in progress, we wait until the
1072 * suspension is over, and then proceed.
1075 vn_start_write(vp, mpp, flags)
1085 * If a vnode is provided, get and return the mount point that
1086 * to which it will write.
1089 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1091 if (error != EOPNOTSUPP)
1096 if ((mp = *mpp) == NULL)
1100 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1102 * As long as a vnode is not provided we need to acquire a
1103 * refcount for the provided mountpoint too, in order to
1104 * emulate a vfs_ref().
1111 * Check on status of suspension.
1113 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1114 mp->mnt_susp_owner != curthread) {
1115 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1116 if (flags & V_NOWAIT) {
1117 error = EWOULDBLOCK;
1120 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1121 (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1126 if (flags & V_XSLEEP)
1128 mp->mnt_writeopcount++;
1130 if (error != 0 || (flags & V_XSLEEP) != 0)
1137 * Secondary suspension. Used by operations such as vop_inactive
1138 * routines that are needed by the higher level functions. These
1139 * are allowed to proceed until all the higher level functions have
1140 * completed (indicated by mnt_writeopcount dropping to zero). At that
1141 * time, these operations are halted until the suspension is over.
1144 vn_start_secondary_write(vp, mpp, flags)
1154 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1156 if (error != EOPNOTSUPP)
1162 * If we are not suspended or have not yet reached suspended
1163 * mode, then let the operation proceed.
1165 if ((mp = *mpp) == NULL)
1169 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1171 * As long as a vnode is not provided we need to acquire a
1172 * refcount for the provided mountpoint too, in order to
1173 * emulate a vfs_ref().
1178 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1179 mp->mnt_secondary_writes++;
1180 mp->mnt_secondary_accwrites++;
1184 if (flags & V_NOWAIT) {
1187 return (EWOULDBLOCK);
1190 * Wait for the suspension to finish.
1192 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1193 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1201 * Filesystem write operation has completed. If we are suspending and this
1202 * operation is the last one, notify the suspender that the suspension is
1206 vn_finished_write(mp)
1213 mp->mnt_writeopcount--;
1214 if (mp->mnt_writeopcount < 0)
1215 panic("vn_finished_write: neg cnt");
1216 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1217 mp->mnt_writeopcount <= 0)
1218 wakeup(&mp->mnt_writeopcount);
1224 * Filesystem secondary write operation has completed. If we are
1225 * suspending and this operation is the last one, notify the suspender
1226 * that the suspension is now in effect.
1229 vn_finished_secondary_write(mp)
1236 mp->mnt_secondary_writes--;
1237 if (mp->mnt_secondary_writes < 0)
1238 panic("vn_finished_secondary_write: neg cnt");
1239 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1240 mp->mnt_secondary_writes <= 0)
1241 wakeup(&mp->mnt_secondary_writes);
1248 * Request a filesystem to suspend write operations.
1251 vfs_write_suspend(mp)
1257 if (mp->mnt_susp_owner == curthread) {
1261 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1262 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1263 mp->mnt_kern_flag |= MNTK_SUSPEND;
1264 mp->mnt_susp_owner = curthread;
1265 if (mp->mnt_writeopcount > 0)
1266 (void) msleep(&mp->mnt_writeopcount,
1267 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1270 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1271 vfs_write_resume(mp);
1276 * Request a filesystem to resume write operations.
1279 vfs_write_resume(mp)
1284 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1285 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1286 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1288 mp->mnt_susp_owner = NULL;
1289 wakeup(&mp->mnt_writeopcount);
1290 wakeup(&mp->mnt_flag);
1291 curthread->td_pflags &= ~TDP_IGNSUSP;
1299 * Implement kqueues for files by translating it to vnode operation.
1302 vn_kqfilter(struct file *fp, struct knote *kn)
1307 vfslocked = VFS_LOCK_GIANT(fp->f_vnode->v_mount);
1308 error = VOP_KQFILTER(fp->f_vnode, kn);
1309 VFS_UNLOCK_GIANT(vfslocked);
1315 * Simplified in-kernel wrapper calls for extended attribute access.
1316 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1317 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1320 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1321 const char *attrname, int *buflen, char *buf, struct thread *td)
1327 iov.iov_len = *buflen;
1330 auio.uio_iov = &iov;
1331 auio.uio_iovcnt = 1;
1332 auio.uio_rw = UIO_READ;
1333 auio.uio_segflg = UIO_SYSSPACE;
1335 auio.uio_offset = 0;
1336 auio.uio_resid = *buflen;
1338 if ((ioflg & IO_NODELOCKED) == 0)
1339 vn_lock(vp, LK_SHARED | LK_RETRY);
1341 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1343 /* authorize attribute retrieval as kernel */
1344 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1347 if ((ioflg & IO_NODELOCKED) == 0)
1351 *buflen = *buflen - auio.uio_resid;
1358 * XXX failure mode if partially written?
1361 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1362 const char *attrname, int buflen, char *buf, struct thread *td)
1369 iov.iov_len = buflen;
1372 auio.uio_iov = &iov;
1373 auio.uio_iovcnt = 1;
1374 auio.uio_rw = UIO_WRITE;
1375 auio.uio_segflg = UIO_SYSSPACE;
1377 auio.uio_offset = 0;
1378 auio.uio_resid = buflen;
1380 if ((ioflg & IO_NODELOCKED) == 0) {
1381 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1383 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1386 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1388 /* authorize attribute setting as kernel */
1389 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1391 if ((ioflg & IO_NODELOCKED) == 0) {
1392 vn_finished_write(mp);
1400 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1401 const char *attrname, struct thread *td)
1406 if ((ioflg & IO_NODELOCKED) == 0) {
1407 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1409 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1412 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1414 /* authorize attribute removal as kernel */
1415 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1416 if (error == EOPNOTSUPP)
1417 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1420 if ((ioflg & IO_NODELOCKED) == 0) {
1421 vn_finished_write(mp);
1429 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1435 ltype = VOP_ISLOCKED(vp);
1436 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1437 ("vn_vget_ino: vp not locked"));
1438 error = vfs_busy(mp, MBF_NOWAIT);
1442 error = vfs_busy(mp, 0);
1443 vn_lock(vp, ltype | LK_RETRY);
1447 if (vp->v_iflag & VI_DOOMED) {
1453 error = VFS_VGET(mp, ino, lkflags, rvp);
1455 vn_lock(vp, ltype | LK_RETRY);
1456 if (vp->v_iflag & VI_DOOMED) {
1465 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, const struct thread *td)
1467 if (vp->v_type != VREG || td == NULL)
1470 PROC_LOCK(td->td_proc);
1471 if (uio->uio_offset + uio->uio_resid >
1472 lim_cur(td->td_proc, RLIMIT_FSIZE)) {
1473 psignal(td->td_proc, SIGXFSZ);
1474 PROC_UNLOCK(td->td_proc);
1477 PROC_UNLOCK(td->td_proc);
1483 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
1487 if ((object = vp->v_object) == NULL)
1489 VM_OBJECT_LOCK(object);
1490 vm_object_page_remove(object, start, end, 0);
1491 VM_OBJECT_UNLOCK(object);
1495 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
1503 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
1504 ("Wrong command %lu", cmd));
1506 if (vn_lock(vp, LK_SHARED) != 0)
1508 if (vp->v_type != VREG) {
1512 error = VOP_GETATTR(vp, &va, cred);
1516 if (noff >= va.va_size) {
1520 bsize = vp->v_mount->mnt_stat.f_iosize;
1521 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
1522 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
1523 if (error == EOPNOTSUPP) {
1527 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
1528 (bnp != -1 && cmd == FIOSEEKDATA)) {
1535 if (noff > va.va_size)
1537 /* noff == va.va_size. There is an implicit hole at the end of file. */
1538 if (cmd == FIOSEEKDATA)