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. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the University of
21 * California, Berkeley and its contributors.
22 * 4. Neither the name of the University nor the names of its contributors
23 * may be used to endorse or promote products derived from this software
24 * without specific prior written permission.
26 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94
39 * $FreeBSD: src/sys/kern/vfs_vnops.c,v 1.87.2.13 2002/12/29 18:19:53 dillon Exp $
40 * $DragonFly: src/sys/kern/vfs_vnops.c,v 1.35 2006/03/27 16:18:34 dillon Exp $
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/fcntl.h>
49 #include <sys/mount.h>
50 #include <sys/nlookup.h>
51 #include <sys/vnode.h>
53 #include <sys/filio.h>
54 #include <sys/ttycom.h>
56 #include <sys/syslog.h>
58 static int vn_closefile (struct file *fp, struct thread *td);
59 static int vn_ioctl (struct file *fp, u_long com, caddr_t data,
61 static int vn_read (struct file *fp, struct uio *uio,
62 struct ucred *cred, int flags, struct thread *td);
63 static int svn_read (struct file *fp, struct uio *uio,
64 struct ucred *cred, int flags, struct thread *td);
65 static int vn_poll (struct file *fp, int events, struct ucred *cred,
67 static int vn_kqfilter (struct file *fp, struct knote *kn);
68 static int vn_statfile (struct file *fp, struct stat *sb, struct thread *td);
69 static int vn_write (struct file *fp, struct uio *uio,
70 struct ucred *cred, int flags, struct thread *td);
71 static int svn_write (struct file *fp, struct uio *uio,
72 struct ucred *cred, int flags, struct thread *td);
74 struct fileops vnode_fileops = {
77 vn_read, vn_write, vn_ioctl, vn_poll, vn_kqfilter,
78 vn_statfile, vn_closefile, nofo_shutdown
81 struct fileops specvnode_fileops = {
84 svn_read, svn_write, vn_ioctl, vn_poll, vn_kqfilter,
85 vn_statfile, vn_closefile, nofo_shutdown
89 * Shortcut the device read/write. This avoids a lot of vnode junk.
90 * Basically the specfs vnops for read and write take the locked vnode,
91 * unlock it (because we can't hold the vnode locked while reading or writing
92 * a device which may block indefinitely), issues the device operation, then
93 * relock the vnode before returning, plus other junk. This bypasses all
94 * of that and just does the device operation.
97 vn_setspecops(struct file *fp)
99 if (vfs_fastdev && fp->f_ops == &vnode_fileops) {
100 fp->f_ops = &specvnode_fileops;
105 * Common code for vnode open operations. Check permissions, and call
106 * the VOP_NOPEN or VOP_NCREATE routine.
108 * The caller is responsible for setting up nd with nlookup_init() and
109 * for cleaning it up with nlookup_done(), whether we return an error
112 * On success nd->nl_open_vp will hold a referenced and, if requested,
113 * locked vnode. A locked vnode is requested via NLC_LOCKVP. If fp
114 * is non-NULL the vnode will be installed in the file pointer.
116 * NOTE: The vnode is referenced just once on return whether or not it
117 * is also installed in the file pointer.
120 vn_open(struct nlookupdata *nd, struct file *fp, int fmode, int cmode)
123 struct thread *td = nd->nl_td;
124 struct ucred *cred = nd->nl_cred;
126 struct vattr *vap = &vat;
127 struct namecache *ncp;
131 * Lookup the path and create or obtain the vnode. After a
132 * successful lookup a locked nd->nl_ncp will be returned.
134 * The result of this section should be a locked vnode.
136 * XXX with only a little work we should be able to avoid locking
137 * the vnode if FWRITE, O_CREAT, and O_TRUNC are *not* set.
139 if (fmode & O_CREAT) {
141 * CONDITIONAL CREATE FILE CASE
143 * Setting NLC_CREATE causes a negative hit to store
144 * the negative hit ncp and not return an error. Then
145 * nc_error or nc_vp may be checked to see if the ncp
146 * represents a negative hit. NLC_CREATE also requires
147 * write permission on the governing directory or EPERM
150 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
151 nd->nl_flags |= NLC_FOLLOW;
152 nd->nl_flags |= NLC_CREATE;
157 * NORMAL OPEN FILE CASE
167 * split case to allow us to re-resolve and retry the ncp in case
171 if (fmode & O_CREAT) {
172 if (ncp->nc_vp == NULL) {
175 vap->va_mode = cmode;
177 vap->va_vaflags |= VA_EXCLUSIVE;
178 error = VOP_NCREATE(ncp, &vp, nd->nl_cred, vap);
182 ASSERT_VOP_LOCKED(vp, "create");
183 /* locked vnode is returned */
185 if (fmode & O_EXCL) {
188 error = cache_vget(ncp, cred,
196 error = cache_vget(ncp, cred, LK_EXCLUSIVE, &vp);
202 * We have a locked vnode and ncp now. Note that the ncp will
203 * be cleaned up by the caller if nd->nl_ncp is left intact.
205 if (vp->v_type == VLNK) {
209 if (vp->v_type == VSOCK) {
213 if ((fmode & O_CREAT) == 0) {
215 if (fmode & (FWRITE | O_TRUNC)) {
216 if (vp->v_type == VDIR) {
220 error = vn_writechk(vp);
223 * Special stale handling, re-resolve the
226 if (error == ESTALE) {
229 cache_setunresolved(ncp);
230 error = cache_resolve(ncp, cred);
241 error = VOP_ACCESS(vp, mode, cred, td);
244 * Special stale handling, re-resolve the
247 if (error == ESTALE) {
250 cache_setunresolved(ncp);
251 error = cache_resolve(ncp, cred);
259 if (fmode & O_TRUNC) {
260 VOP_UNLOCK(vp, 0, td); /* XXX */
261 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); /* XXX */
264 error = VOP_SETATTR(vp, vap, cred, td);
270 * Setup the fp so VOP_OPEN can override it. No descriptor has been
271 * associated with the fp yet so we own it clean. f_data will inherit
272 * our vp reference as long as we do not shift f_ops to &badfileops.
273 * f_ncp inherits nl_ncp .
276 fp->f_type = (vp->v_type == VFIFO ? DTYPE_FIFO : DTYPE_VNODE);
277 fp->f_flag = fmode & FMASK;
278 fp->f_ops = &vnode_fileops;
280 if (vp->v_type == VDIR) {
281 fp->f_ncp = nd->nl_ncp;
283 cache_unlock(fp->f_ncp);
288 * Get rid of nl_ncp. vn_open does not return it (it returns the
289 * vnode or the file pointer). Note: we can't leave nl_ncp locked
290 * through the VOP_OPEN anyway since the VOP_OPEN may block, e.g.
294 cache_put(nd->nl_ncp);
298 error = VOP_OPEN(vp, fmode, cred, fp, td);
301 * setting f_ops to &badfileops will prevent the descriptor
302 * code from trying to close and release the vnode, since
303 * the open failed we do not want to call close.
307 fp->f_ops = &badfileops;
315 * Make sure that a VM object is created for VMIO support. If this
316 * fails we have to be sure to match VOP_CLOSE's with VOP_OPEN's.
317 * Cleanup the fp so we can just vput() the vp in 'bad'.
319 if (vn_canvmio(vp) == TRUE) {
320 if ((error = vfs_object_create(vp, td)) != 0) {
323 fp->f_ops = &badfileops;
325 VOP_CLOSE(vp, fmode, td);
331 * Return the vnode. XXX needs some cleaning up. The vnode is
332 * only returned in the fp == NULL case, otherwise the vnode ref
333 * is inherited by the fp and we unconditionally unlock it.
337 nd->nl_vp_fmode = fmode;
338 if ((nd->nl_flags & NLC_LOCKVP) == 0)
339 VOP_UNLOCK(vp, 0, td);
341 VOP_UNLOCK(vp, 0, td);
351 * Check for write permissions on the specified vnode.
352 * Prototype text segments cannot be written.
360 * If there's shared text associated with
361 * the vnode, try to free it up once. If
362 * we fail, we can't allow writing.
364 if (vp->v_flag & VTEXT)
373 vn_close(struct vnode *vp, int flags, struct thread *td)
379 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td)) == 0) {
380 error = VOP_CLOSE(vp, flags, td);
381 VOP_UNLOCK(vp, 0, td);
389 sequential_heuristic(struct uio *uio, struct file *fp)
392 * Sequential heuristic - detect sequential operation
394 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
395 uio->uio_offset == fp->f_nextoff) {
396 int tmpseq = fp->f_seqcount;
398 * XXX we assume that the filesystem block size is
399 * the default. Not true, but still gives us a pretty
400 * good indicator of how sequential the read operations
403 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
404 if (tmpseq > IO_SEQMAX)
406 fp->f_seqcount = tmpseq;
407 return(fp->f_seqcount << IO_SEQSHIFT);
411 * Not sequential, quick draw-down of seqcount
413 if (fp->f_seqcount > 1)
421 * Package up an I/O request on a vnode into a uio and do it.
424 vn_rdwr(rw, vp, base, len, offset, segflg, ioflg, cred, aresid, td)
440 if ((ioflg & IO_NODELOCKED) == 0)
441 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
442 auio.uio_iov = &aiov;
444 aiov.iov_base = base;
446 auio.uio_resid = len;
447 auio.uio_offset = offset;
448 auio.uio_segflg = segflg;
451 if (rw == UIO_READ) {
452 error = VOP_READ(vp, &auio, ioflg, cred);
454 error = VOP_WRITE(vp, &auio, ioflg, cred);
457 *aresid = auio.uio_resid;
459 if (auio.uio_resid && error == 0)
461 if ((ioflg & IO_NODELOCKED) == 0)
462 VOP_UNLOCK(vp, 0, td);
467 * Package up an I/O request on a vnode into a uio and do it. The I/O
468 * request is split up into smaller chunks and we try to avoid saturating
469 * the buffer cache while potentially holding a vnode locked, so we
470 * check bwillwrite() before calling vn_rdwr(). We also call uio_yield()
471 * to give other processes a chance to lock the vnode (either other processes
472 * core'ing the same binary, or unrelated processes scanning the directory).
475 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, cred, aresid, td)
493 * Force `offset' to a multiple of MAXBSIZE except possibly
494 * for the first chunk, so that filesystems only need to
495 * write full blocks except possibly for the first and last
498 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
502 if (rw != UIO_READ && vp->v_type == VREG)
504 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
505 ioflg, cred, aresid, td);
506 len -= chunk; /* aresid calc already includes length */
519 * File table vnode read routine.
522 vn_read(fp, uio, cred, flags, td)
532 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
533 vp = (struct vnode *)fp->f_data;
535 if (fp->f_flag & FNONBLOCK)
537 if (fp->f_flag & O_DIRECT)
539 vn_lock(vp, LK_SHARED | LK_NOPAUSE | LK_RETRY, td);
540 if ((flags & FOF_OFFSET) == 0)
541 uio->uio_offset = fp->f_offset;
543 ioflag |= sequential_heuristic(uio, fp);
545 error = VOP_READ(vp, uio, ioflag, cred);
546 if ((flags & FOF_OFFSET) == 0)
547 fp->f_offset = uio->uio_offset;
548 fp->f_nextoff = uio->uio_offset;
549 VOP_UNLOCK(vp, 0, td);
554 * Device-optimized file table vnode read routine.
556 * This bypasses the VOP table and talks directly to the device. Most
557 * filesystems just route to specfs and can make this optimization.
560 svn_read(fp, uio, cred, flags, td)
572 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td));
574 vp = (struct vnode *)fp->f_data;
575 if (vp == NULL || vp->v_type == VBAD)
578 if ((dev = vp->v_rdev) == NULL)
582 if (uio->uio_resid == 0)
584 if ((flags & FOF_OFFSET) == 0)
585 uio->uio_offset = fp->f_offset;
588 if (fp->f_flag & FNONBLOCK)
590 if (fp->f_flag & O_DIRECT)
592 ioflag |= sequential_heuristic(uio, fp);
594 error = dev_dread(dev, uio, ioflag);
597 if ((flags & FOF_OFFSET) == 0)
598 fp->f_offset = uio->uio_offset;
599 fp->f_nextoff = uio->uio_offset;
604 * File table vnode write routine.
607 vn_write(fp, uio, cred, flags, td)
617 KASSERT(uio->uio_td == td, ("uio_procp %p is not p %p",
619 vp = (struct vnode *)fp->f_data;
620 if (vp->v_type == VREG)
622 vp = (struct vnode *)fp->f_data; /* XXX needed? */
624 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
626 if (fp->f_flag & FNONBLOCK)
628 if (fp->f_flag & O_DIRECT)
630 if ((fp->f_flag & O_FSYNC) ||
631 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
633 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
634 if ((flags & FOF_OFFSET) == 0)
635 uio->uio_offset = fp->f_offset;
636 ioflag |= sequential_heuristic(uio, fp);
637 error = VOP_WRITE(vp, uio, ioflag, cred);
638 if ((flags & FOF_OFFSET) == 0)
639 fp->f_offset = uio->uio_offset;
640 fp->f_nextoff = uio->uio_offset;
641 VOP_UNLOCK(vp, 0, td);
646 * Device-optimized file table vnode write routine.
648 * This bypasses the VOP table and talks directly to the device. Most
649 * filesystems just route to specfs and can make this optimization.
652 svn_write(fp, uio, cred, flags, td)
664 KASSERT(uio->uio_td == td, ("uio_procp %p is not p %p",
667 vp = (struct vnode *)fp->f_data;
668 if (vp == NULL || vp->v_type == VBAD)
670 if (vp->v_type == VREG)
672 vp = (struct vnode *)fp->f_data; /* XXX needed? */
674 if ((dev = vp->v_rdev) == NULL)
678 if ((flags & FOF_OFFSET) == 0)
679 uio->uio_offset = fp->f_offset;
682 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
684 if (fp->f_flag & FNONBLOCK)
686 if (fp->f_flag & O_DIRECT)
688 if ((fp->f_flag & O_FSYNC) ||
689 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
691 ioflag |= sequential_heuristic(uio, fp);
693 error = dev_dwrite(dev, uio, ioflag);
696 if ((flags & FOF_OFFSET) == 0)
697 fp->f_offset = uio->uio_offset;
698 fp->f_nextoff = uio->uio_offset;
704 * File table vnode stat routine.
707 vn_statfile(struct file *fp, struct stat *sb, struct thread *td)
709 struct vnode *vp = (struct vnode *)fp->f_data;
711 return vn_stat(vp, sb, td);
715 vn_stat(struct vnode *vp, struct stat *sb, struct thread *td)
724 error = VOP_GETATTR(vp, vap, td);
729 * Zero the spare stat fields
735 * Copy from vattr table
737 if (vap->va_fsid != VNOVAL)
738 sb->st_dev = vap->va_fsid;
740 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
741 sb->st_ino = vap->va_fileid;
743 switch (vap->va_type) {
758 /* This is a cosmetic change, symlinks do not have a mode. */
759 if (vp->v_mount->mnt_flag & MNT_NOSYMFOLLOW)
760 sb->st_mode &= ~ACCESSPERMS; /* 0000 */
762 sb->st_mode |= ACCESSPERMS; /* 0777 */
774 sb->st_nlink = vap->va_nlink;
775 sb->st_uid = vap->va_uid;
776 sb->st_gid = vap->va_gid;
777 sb->st_rdev = vap->va_rdev;
778 sb->st_size = vap->va_size;
779 sb->st_atimespec = vap->va_atime;
780 sb->st_mtimespec = vap->va_mtime;
781 sb->st_ctimespec = vap->va_ctime;
784 * A VCHR and VBLK device may track the last access and last modified
785 * time independantly of the filesystem. This is particularly true
786 * because device read and write calls may bypass the filesystem.
788 if (vp->v_type == VCHR || vp->v_type == VBLK) {
789 if ((dev = vp->v_rdev) != NULL) {
790 if (dev->si_lastread) {
791 sb->st_atimespec.tv_sec = dev->si_lastread;
792 sb->st_atimespec.tv_nsec = 0;
794 if (dev->si_lastwrite) {
795 sb->st_atimespec.tv_sec = dev->si_lastwrite;
796 sb->st_atimespec.tv_nsec = 0;
802 * According to www.opengroup.org, the meaning of st_blksize is
803 * "a filesystem-specific preferred I/O block size for this
804 * object. In some filesystem types, this may vary from file
806 * Default to PAGE_SIZE after much discussion.
809 if (vap->va_type == VREG) {
810 sb->st_blksize = vap->va_blocksize;
811 } else if (vn_isdisk(vp, NULL)) {
813 * XXX this is broken. If the device is not yet open (aka
814 * stat() call, aka v_rdev == NULL), how are we supposed
815 * to get a valid block size out of it?
819 if ((dev = vp->v_rdev) == NULL)
820 dev = udev2dev(vp->v_udev, vp->v_type == VBLK);
821 sb->st_blksize = dev->si_bsize_best;
822 if (sb->st_blksize < dev->si_bsize_phys)
823 sb->st_blksize = dev->si_bsize_phys;
824 if (sb->st_blksize < BLKDEV_IOSIZE)
825 sb->st_blksize = BLKDEV_IOSIZE;
827 sb->st_blksize = PAGE_SIZE;
830 sb->st_flags = vap->va_flags;
834 sb->st_gen = vap->va_gen;
836 #if (S_BLKSIZE == 512)
837 /* Optimize this case */
838 sb->st_blocks = vap->va_bytes >> 9;
840 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
842 sb->st_fsmid = vap->va_fsmid;
847 * File table vnode ioctl routine.
850 vn_ioctl(struct file *fp, u_long com, caddr_t data, struct thread *td)
852 struct vnode *vp = ((struct vnode *)fp->f_data);
858 KKASSERT(td->td_proc != NULL);
859 ucred = td->td_proc->p_ucred;
861 switch (vp->v_type) {
864 if (com == FIONREAD) {
865 error = VOP_GETATTR(vp, &vattr, td);
868 *(int *)data = vattr.va_size - fp->f_offset;
871 if (com == FIONBIO || com == FIOASYNC) /* XXX */
872 return (0); /* XXX */
881 if (com == FIODTYPE) {
882 if (vp->v_type != VCHR && vp->v_type != VBLK)
884 *(int *)data = dev_dflags(vp->v_rdev) & D_TYPEMASK;
887 error = VOP_IOCTL(vp, com, data, fp->f_flag, ucred, td);
888 if (error == 0 && com == TIOCSCTTY) {
889 struct session *sess = td->td_proc->p_session;
891 /* Do nothing if reassigning same control tty */
892 if (sess->s_ttyvp == vp)
895 /* Get rid of reference to old control tty */
907 * File table vnode poll routine.
910 vn_poll(struct file *fp, int events, struct ucred *cred, struct thread *td)
912 return (VOP_POLL(((struct vnode *)fp->f_data), events, cred, td));
916 * Check that the vnode is still valid, and if so
917 * acquire requested lock.
921 vn_lock(struct vnode *vp, int flags, struct thread *td)
923 debug_vn_lock(struct vnode *vp, int flags, struct thread *td,
924 const char *filename, int line)
931 vp->filename = filename;
934 error = VOP_LOCK(vp, flags | LK_NOPAUSE, td);
937 } while (flags & LK_RETRY);
940 * Because we (had better!) have a ref on the vnode, once it
941 * goes to VRECLAIMED state it will not be recycled until all
942 * refs go away. So we can just check the flag.
944 if (error == 0 && (vp->v_flag & VRECLAIMED)) {
945 VOP_UNLOCK(vp, 0, td);
952 * File table vnode close routine.
955 vn_closefile(struct file *fp, struct thread *td)
959 fp->f_ops = &badfileops;
960 err = vn_close(((struct vnode *)fp->f_data), fp->f_flag, td);
965 vn_kqfilter(struct file *fp, struct knote *kn)
968 return (VOP_KQFILTER(((struct vnode *)fp->f_data), kn));