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/audit/audit.h>
65 #include <security/mac/mac_framework.h>
68 #include <vm/vm_object.h>
70 static fo_rdwr_t vn_read;
71 static fo_rdwr_t vn_write;
72 static fo_truncate_t vn_truncate;
73 static fo_ioctl_t vn_ioctl;
74 static fo_poll_t vn_poll;
75 static fo_kqfilter_t vn_kqfilter;
76 static fo_stat_t vn_statfile;
77 static fo_close_t vn_closefile;
79 struct fileops vnops = {
82 .fo_truncate = vn_truncate,
85 .fo_kqfilter = vn_kqfilter,
86 .fo_stat = vn_statfile,
87 .fo_close = vn_closefile,
90 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
94 vn_open(ndp, flagp, cmode, fp)
95 struct nameidata *ndp;
99 struct thread *td = ndp->ni_cnd.cn_thread;
101 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
105 * Common code for vnode open operations.
106 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
108 * Note that this does NOT free nameidata for the successful case,
109 * due to the NDINIT being done elsewhere.
112 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
113 struct ucred *cred, struct file *fp)
117 struct thread *td = ndp->ni_cnd.cn_thread;
119 struct vattr *vap = &vat;
122 int vfslocked, mpsafe;
124 mpsafe = ndp->ni_cnd.cn_flags & MPSAFE;
128 if (fmode & O_CREAT) {
129 ndp->ni_cnd.cn_nameiop = CREATE;
130 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF |
132 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
133 ndp->ni_cnd.cn_flags |= FOLLOW;
134 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
135 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
137 if ((error = namei(ndp)) != 0)
139 vfslocked = NDHASGIANT(ndp);
141 ndp->ni_cnd.cn_flags &= ~MPSAFE;
142 if (ndp->ni_vp == NULL) {
145 vap->va_mode = cmode;
147 vap->va_vaflags |= VA_EXCLUSIVE;
148 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
149 NDFREE(ndp, NDF_ONLY_PNBUF);
151 VFS_UNLOCK_GIANT(vfslocked);
152 if ((error = vn_start_write(NULL, &mp,
153 V_XSLEEP | PCATCH)) != 0)
158 error = mac_vnode_check_create(cred, ndp->ni_dvp,
162 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
165 vn_finished_write(mp);
167 VFS_UNLOCK_GIANT(vfslocked);
168 NDFREE(ndp, NDF_ONLY_PNBUF);
174 if (ndp->ni_dvp == ndp->ni_vp)
180 if (fmode & O_EXCL) {
187 ndp->ni_cnd.cn_nameiop = LOOKUP;
188 ndp->ni_cnd.cn_flags = ISOPEN |
189 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) |
191 if (!(fmode & FWRITE))
192 ndp->ni_cnd.cn_flags |= LOCKSHARED;
193 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
194 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
195 if ((error = namei(ndp)) != 0)
198 ndp->ni_cnd.cn_flags &= ~MPSAFE;
199 vfslocked = NDHASGIANT(ndp);
202 if (vp->v_type == VLNK) {
206 if (vp->v_type == VSOCK) {
210 if (vp->v_type != VDIR && fmode & O_DIRECTORY) {
215 if (fmode & (FWRITE | O_TRUNC)) {
216 if (vp->v_type == VDIR) {
226 if ((fmode & O_APPEND) && (fmode & FWRITE))
229 error = mac_vnode_check_open(cred, vp, accmode);
233 if ((fmode & O_CREAT) == 0) {
234 if (accmode & VWRITE) {
235 error = vn_writechk(vp);
240 error = VOP_ACCESS(vp, accmode, cred, td);
245 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
251 ASSERT_VOP_LOCKED(vp, "vn_open_cred");
253 VFS_UNLOCK_GIANT(vfslocked);
256 NDFREE(ndp, NDF_ONLY_PNBUF);
258 VFS_UNLOCK_GIANT(vfslocked);
265 * Check for write permissions on the specified vnode.
266 * Prototype text segments cannot be written.
270 register struct vnode *vp;
273 ASSERT_VOP_LOCKED(vp, "vn_writechk");
275 * If there's shared text associated with
276 * the vnode, try to free it up once. If
277 * we fail, we can't allow writing.
279 if (vp->v_vflag & VV_TEXT)
289 vn_close(vp, flags, file_cred, td)
290 register struct vnode *vp;
292 struct ucred *file_cred;
296 int error, lock_flags;
298 if (!(flags & FWRITE) && vp->v_mount != NULL &&
299 vp->v_mount->mnt_kern_flag & MNTK_EXTENDED_SHARED)
300 lock_flags = LK_SHARED;
302 lock_flags = LK_EXCLUSIVE;
304 VFS_ASSERT_GIANT(vp->v_mount);
306 vn_start_write(vp, &mp, V_WAIT);
307 vn_lock(vp, lock_flags | LK_RETRY);
308 if (flags & FWRITE) {
309 VNASSERT(vp->v_writecount > 0, vp,
310 ("vn_close: negative writecount"));
313 error = VOP_CLOSE(vp, flags, file_cred, td);
315 vn_finished_write(mp);
320 * Heuristic to detect sequential operation.
323 sequential_heuristic(struct uio *uio, struct file *fp)
326 if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
327 return (fp->f_seqcount << IO_SEQSHIFT);
330 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
331 * that the first I/O is normally considered to be slightly
332 * sequential. Seeking to offset 0 doesn't change sequentiality
333 * unless previous seeks have reduced f_seqcount to 0, in which
334 * case offset 0 is not special.
336 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
337 uio->uio_offset == fp->f_nextoff) {
339 * f_seqcount is in units of fixed-size blocks so that it
340 * depends mainly on the amount of sequential I/O and not
341 * much on the number of sequential I/O's. The fixed size
342 * of 16384 is hard-coded here since it is (not quite) just
343 * a magic size that works well here. This size is more
344 * closely related to the best I/O size for real disks than
345 * to any block size used by software.
347 fp->f_seqcount += howmany(uio->uio_resid, 16384);
348 if (fp->f_seqcount > IO_SEQMAX)
349 fp->f_seqcount = IO_SEQMAX;
350 return (fp->f_seqcount << IO_SEQSHIFT);
353 /* Not sequential. Quickly draw-down sequentiality. */
354 if (fp->f_seqcount > 1)
362 * Package up an I/O request on a vnode into a uio and do it.
365 vn_rdwr(rw, vp, base, len, offset, segflg, ioflg, active_cred, file_cred,
374 struct ucred *active_cred;
375 struct ucred *file_cred;
383 int error, lock_flags;
385 VFS_ASSERT_GIANT(vp->v_mount);
387 if ((ioflg & IO_NODELOCKED) == 0) {
389 if (rw == UIO_WRITE) {
390 if (vp->v_type != VCHR &&
391 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
394 if (MNT_SHARED_WRITES(mp) ||
395 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) {
396 lock_flags = LK_SHARED;
398 lock_flags = LK_EXCLUSIVE;
400 vn_lock(vp, lock_flags | LK_RETRY);
402 vn_lock(vp, LK_SHARED | LK_RETRY);
405 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
406 auio.uio_iov = &aiov;
408 aiov.iov_base = base;
410 auio.uio_resid = len;
411 auio.uio_offset = offset;
412 auio.uio_segflg = segflg;
417 if ((ioflg & IO_NOMACCHECK) == 0) {
419 error = mac_vnode_check_read(active_cred, file_cred,
422 error = mac_vnode_check_write(active_cred, file_cred,
432 error = VOP_READ(vp, &auio, ioflg, cred);
434 error = VOP_WRITE(vp, &auio, ioflg, cred);
437 *aresid = auio.uio_resid;
439 if (auio.uio_resid && error == 0)
441 if ((ioflg & IO_NODELOCKED) == 0) {
442 if (rw == UIO_WRITE && vp->v_type != VCHR)
443 vn_finished_write(mp);
450 * Package up an I/O request on a vnode into a uio and do it. The I/O
451 * request is split up into smaller chunks and we try to avoid saturating
452 * the buffer cache while potentially holding a vnode locked, so we
453 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
454 * to give other processes a chance to lock the vnode (either other processes
455 * core'ing the same binary, or unrelated processes scanning the directory).
458 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
459 file_cred, aresid, td)
467 struct ucred *active_cred;
468 struct ucred *file_cred;
475 VFS_ASSERT_GIANT(vp->v_mount);
481 * Force `offset' to a multiple of MAXBSIZE except possibly
482 * for the first chunk, so that filesystems only need to
483 * write full blocks except possibly for the first and last
486 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
490 if (rw != UIO_READ && vp->v_type == VREG)
493 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
494 ioflg, active_cred, file_cred, &iaresid, td);
495 len -= chunk; /* aresid calc already includes length */
499 base = (char *)base + chunk;
500 kern_yield(PRI_USER);
503 *aresid = len + iaresid;
508 * File table vnode read routine.
511 vn_read(fp, uio, active_cred, flags, td)
514 struct ucred *active_cred;
521 int advice, vfslocked;
523 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
528 if (fp->f_flag & FNONBLOCK)
530 if (fp->f_flag & O_DIRECT)
532 advice = POSIX_FADV_NORMAL;
533 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
535 * According to McKusick the vn lock was protecting f_offset here.
536 * It is now protected by the FOFFSET_LOCKED flag.
538 if ((flags & FOF_OFFSET) == 0 || fp->f_advice != NULL) {
539 mtxp = mtx_pool_find(mtxpool_sleep, fp);
541 if ((flags & FOF_OFFSET) == 0) {
542 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
543 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
544 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
545 "vnread offlock", 0);
547 fp->f_vnread_flags |= FOFFSET_LOCKED;
548 uio->uio_offset = fp->f_offset;
550 if (fp->f_advice != NULL &&
551 uio->uio_offset >= fp->f_advice->fa_start &&
552 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
553 advice = fp->f_advice->fa_advice;
556 vn_lock(vp, LK_SHARED | LK_RETRY);
559 case POSIX_FADV_NORMAL:
560 case POSIX_FADV_SEQUENTIAL:
561 ioflag |= sequential_heuristic(uio, fp);
563 case POSIX_FADV_RANDOM:
564 /* Disable read-ahead for random I/O. */
566 case POSIX_FADV_NOREUSE:
568 * Request the underlying FS to discard the buffers
569 * and pages after the I/O is complete.
576 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
579 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
580 if ((flags & FOF_OFFSET) == 0) {
581 fp->f_offset = uio->uio_offset;
583 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
584 wakeup(&fp->f_vnread_flags);
585 fp->f_vnread_flags = 0;
588 fp->f_nextoff = uio->uio_offset;
590 VFS_UNLOCK_GIANT(vfslocked);
595 * File table vnode write routine.
598 vn_write(fp, uio, active_cred, flags, td)
601 struct ucred *active_cred;
607 int error, ioflag, lock_flags;
609 int advice, vfslocked;
611 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
614 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
615 if (vp->v_type == VREG)
618 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
620 if (fp->f_flag & FNONBLOCK)
622 if (fp->f_flag & O_DIRECT)
624 if ((fp->f_flag & O_FSYNC) ||
625 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
628 if (vp->v_type != VCHR &&
629 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
632 if ((MNT_SHARED_WRITES(mp) ||
633 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) &&
634 (flags & FOF_OFFSET) != 0) {
635 lock_flags = LK_SHARED;
637 lock_flags = LK_EXCLUSIVE;
640 vn_lock(vp, lock_flags | LK_RETRY);
641 if ((flags & FOF_OFFSET) == 0)
642 uio->uio_offset = fp->f_offset;
643 advice = POSIX_FADV_NORMAL;
644 if (fp->f_advice != NULL) {
645 mtxp = mtx_pool_find(mtxpool_sleep, fp);
647 if (fp->f_advice != NULL &&
648 uio->uio_offset >= fp->f_advice->fa_start &&
649 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
650 advice = fp->f_advice->fa_advice;
654 case POSIX_FADV_NORMAL:
655 case POSIX_FADV_SEQUENTIAL:
656 ioflag |= sequential_heuristic(uio, fp);
658 case POSIX_FADV_RANDOM:
659 /* XXX: Is this correct? */
661 case POSIX_FADV_NOREUSE:
663 * Request the underlying FS to discard the buffers
664 * and pages after the I/O is complete.
671 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
674 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
675 if ((flags & FOF_OFFSET) == 0)
676 fp->f_offset = uio->uio_offset;
677 fp->f_nextoff = uio->uio_offset;
679 if (vp->v_type != VCHR)
680 vn_finished_write(mp);
682 VFS_UNLOCK_GIANT(vfslocked);
687 * File table truncate routine.
690 vn_truncate(fp, length, active_cred, td)
693 struct ucred *active_cred;
703 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
704 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
706 VFS_UNLOCK_GIANT(vfslocked);
709 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
710 if (vp->v_type == VDIR) {
715 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
719 error = vn_writechk(vp);
722 vattr.va_size = length;
723 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
727 vn_finished_write(mp);
728 VFS_UNLOCK_GIANT(vfslocked);
733 * File table vnode stat routine.
736 vn_statfile(fp, sb, active_cred, td)
739 struct ucred *active_cred;
742 struct vnode *vp = fp->f_vnode;
746 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
747 vn_lock(vp, LK_SHARED | LK_RETRY);
748 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
750 VFS_UNLOCK_GIANT(vfslocked);
756 * Stat a vnode; implementation for the stat syscall
759 vn_stat(vp, sb, active_cred, file_cred, td)
761 register struct stat *sb;
762 struct ucred *active_cred;
763 struct ucred *file_cred;
767 register struct vattr *vap;
772 error = mac_vnode_check_stat(active_cred, file_cred, vp);
780 * Initialize defaults for new and unusual fields, so that file
781 * systems which don't support these fields don't need to know
784 vap->va_birthtime.tv_sec = -1;
785 vap->va_birthtime.tv_nsec = 0;
786 vap->va_fsid = VNOVAL;
787 vap->va_rdev = NODEV;
789 error = VOP_GETATTR(vp, vap, active_cred);
794 * Zero the spare stat fields
796 bzero(sb, sizeof *sb);
799 * Copy from vattr table
801 if (vap->va_fsid != VNOVAL)
802 sb->st_dev = vap->va_fsid;
804 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
805 sb->st_ino = vap->va_fileid;
807 switch (vap->va_type) {
833 sb->st_nlink = vap->va_nlink;
834 sb->st_uid = vap->va_uid;
835 sb->st_gid = vap->va_gid;
836 sb->st_rdev = vap->va_rdev;
837 if (vap->va_size > OFF_MAX)
839 sb->st_size = vap->va_size;
840 sb->st_atim = vap->va_atime;
841 sb->st_mtim = vap->va_mtime;
842 sb->st_ctim = vap->va_ctime;
843 sb->st_birthtim = vap->va_birthtime;
846 * According to www.opengroup.org, the meaning of st_blksize is
847 * "a filesystem-specific preferred I/O block size for this
848 * object. In some filesystem types, this may vary from file
850 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
853 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
855 sb->st_flags = vap->va_flags;
856 if (priv_check(td, PRIV_VFS_GENERATION))
859 sb->st_gen = vap->va_gen;
861 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
866 * File table vnode ioctl routine.
869 vn_ioctl(fp, com, data, active_cred, td)
873 struct ucred *active_cred;
876 struct vnode *vp = fp->f_vnode;
881 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
883 switch (vp->v_type) {
886 if (com == FIONREAD) {
887 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
888 error = VOP_GETATTR(vp, &vattr, active_cred);
891 *(int *)data = vattr.va_size - fp->f_offset;
893 if (com == FIONBIO || com == FIOASYNC) /* XXX */
896 error = VOP_IOCTL(vp, com, data, fp->f_flag,
903 VFS_UNLOCK_GIANT(vfslocked);
908 * File table vnode poll routine.
911 vn_poll(fp, events, active_cred, td)
914 struct ucred *active_cred;
922 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
924 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
925 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
930 error = VOP_POLL(vp, events, fp->f_cred, td);
931 VFS_UNLOCK_GIANT(vfslocked);
936 * Acquire the requested lock and then check for validity. LK_RETRY
937 * permits vn_lock to return doomed vnodes.
940 _vn_lock(struct vnode *vp, int flags, char *file, int line)
944 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
945 ("vn_lock called with no locktype."));
947 #ifdef DEBUG_VFS_LOCKS
948 KASSERT(vp->v_holdcnt != 0,
949 ("vn_lock %p: zero hold count", vp));
951 error = VOP_LOCK1(vp, flags, file, line);
952 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
953 KASSERT((flags & LK_RETRY) == 0 || error == 0,
954 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
957 * Callers specify LK_RETRY if they wish to get dead vnodes.
958 * If RETRY is not set, we return ENOENT instead.
960 if (error == 0 && vp->v_iflag & VI_DOOMED &&
961 (flags & LK_RETRY) == 0) {
966 } while (flags & LK_RETRY && error != 0);
971 * File table vnode close routine.
985 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
986 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
987 lf.l_whence = SEEK_SET;
991 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
994 fp->f_ops = &badfileops;
996 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
997 VFS_UNLOCK_GIANT(vfslocked);
1002 * Preparing to start a filesystem write operation. If the operation is
1003 * permitted, then we bump the count of operations in progress and
1004 * proceed. If a suspend request is in progress, we wait until the
1005 * suspension is over, and then proceed.
1008 vn_start_write(vp, mpp, flags)
1018 * If a vnode is provided, get and return the mount point that
1019 * to which it will write.
1022 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1024 if (error != EOPNOTSUPP)
1029 if ((mp = *mpp) == NULL)
1033 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1035 * As long as a vnode is not provided we need to acquire a
1036 * refcount for the provided mountpoint too, in order to
1037 * emulate a vfs_ref().
1044 * Check on status of suspension.
1046 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1047 mp->mnt_susp_owner != curthread) {
1048 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1049 if (flags & V_NOWAIT) {
1050 error = EWOULDBLOCK;
1053 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1054 (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1059 if (flags & V_XSLEEP)
1061 mp->mnt_writeopcount++;
1063 if (error != 0 || (flags & V_XSLEEP) != 0)
1070 * Secondary suspension. Used by operations such as vop_inactive
1071 * routines that are needed by the higher level functions. These
1072 * are allowed to proceed until all the higher level functions have
1073 * completed (indicated by mnt_writeopcount dropping to zero). At that
1074 * time, these operations are halted until the suspension is over.
1077 vn_start_secondary_write(vp, mpp, flags)
1087 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1089 if (error != EOPNOTSUPP)
1095 * If we are not suspended or have not yet reached suspended
1096 * mode, then let the operation proceed.
1098 if ((mp = *mpp) == NULL)
1102 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1104 * As long as a vnode is not provided we need to acquire a
1105 * refcount for the provided mountpoint too, in order to
1106 * emulate a vfs_ref().
1111 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1112 mp->mnt_secondary_writes++;
1113 mp->mnt_secondary_accwrites++;
1117 if (flags & V_NOWAIT) {
1120 return (EWOULDBLOCK);
1123 * Wait for the suspension to finish.
1125 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1126 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1134 * Filesystem write operation has completed. If we are suspending and this
1135 * operation is the last one, notify the suspender that the suspension is
1139 vn_finished_write(mp)
1146 mp->mnt_writeopcount--;
1147 if (mp->mnt_writeopcount < 0)
1148 panic("vn_finished_write: neg cnt");
1149 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1150 mp->mnt_writeopcount <= 0)
1151 wakeup(&mp->mnt_writeopcount);
1157 * Filesystem secondary write operation has completed. If we are
1158 * suspending and this operation is the last one, notify the suspender
1159 * that the suspension is now in effect.
1162 vn_finished_secondary_write(mp)
1169 mp->mnt_secondary_writes--;
1170 if (mp->mnt_secondary_writes < 0)
1171 panic("vn_finished_secondary_write: neg cnt");
1172 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1173 mp->mnt_secondary_writes <= 0)
1174 wakeup(&mp->mnt_secondary_writes);
1181 * Request a filesystem to suspend write operations.
1184 vfs_write_suspend(mp)
1190 if (mp->mnt_susp_owner == curthread) {
1194 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1195 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1196 mp->mnt_kern_flag |= MNTK_SUSPEND;
1197 mp->mnt_susp_owner = curthread;
1198 if (mp->mnt_writeopcount > 0)
1199 (void) msleep(&mp->mnt_writeopcount,
1200 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1203 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1204 vfs_write_resume(mp);
1209 * Request a filesystem to resume write operations.
1212 vfs_write_resume(mp)
1217 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1218 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1219 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1221 mp->mnt_susp_owner = NULL;
1222 wakeup(&mp->mnt_writeopcount);
1223 wakeup(&mp->mnt_flag);
1224 curthread->td_pflags &= ~TDP_IGNSUSP;
1232 * Implement kqueues for files by translating it to vnode operation.
1235 vn_kqfilter(struct file *fp, struct knote *kn)
1240 vfslocked = VFS_LOCK_GIANT(fp->f_vnode->v_mount);
1241 error = VOP_KQFILTER(fp->f_vnode, kn);
1242 VFS_UNLOCK_GIANT(vfslocked);
1248 * Simplified in-kernel wrapper calls for extended attribute access.
1249 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1250 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1253 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1254 const char *attrname, int *buflen, char *buf, struct thread *td)
1260 iov.iov_len = *buflen;
1263 auio.uio_iov = &iov;
1264 auio.uio_iovcnt = 1;
1265 auio.uio_rw = UIO_READ;
1266 auio.uio_segflg = UIO_SYSSPACE;
1268 auio.uio_offset = 0;
1269 auio.uio_resid = *buflen;
1271 if ((ioflg & IO_NODELOCKED) == 0)
1272 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1274 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1276 /* authorize attribute retrieval as kernel */
1277 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1280 if ((ioflg & IO_NODELOCKED) == 0)
1284 *buflen = *buflen - auio.uio_resid;
1291 * XXX failure mode if partially written?
1294 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1295 const char *attrname, int buflen, char *buf, struct thread *td)
1302 iov.iov_len = buflen;
1305 auio.uio_iov = &iov;
1306 auio.uio_iovcnt = 1;
1307 auio.uio_rw = UIO_WRITE;
1308 auio.uio_segflg = UIO_SYSSPACE;
1310 auio.uio_offset = 0;
1311 auio.uio_resid = buflen;
1313 if ((ioflg & IO_NODELOCKED) == 0) {
1314 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1316 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1319 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1321 /* authorize attribute setting as kernel */
1322 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1324 if ((ioflg & IO_NODELOCKED) == 0) {
1325 vn_finished_write(mp);
1333 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1334 const char *attrname, struct thread *td)
1339 if ((ioflg & IO_NODELOCKED) == 0) {
1340 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1342 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1345 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1347 /* authorize attribute removal as kernel */
1348 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1349 if (error == EOPNOTSUPP)
1350 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1353 if ((ioflg & IO_NODELOCKED) == 0) {
1354 vn_finished_write(mp);
1362 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1368 ltype = VOP_ISLOCKED(vp);
1369 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1370 ("vn_vget_ino: vp not locked"));
1371 error = vfs_busy(mp, MBF_NOWAIT);
1375 error = vfs_busy(mp, 0);
1376 vn_lock(vp, ltype | LK_RETRY);
1380 if (vp->v_iflag & VI_DOOMED) {
1386 error = VFS_VGET(mp, ino, lkflags, rvp);
1388 vn_lock(vp, ltype | LK_RETRY);
1389 if (vp->v_iflag & VI_DOOMED) {
1398 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
1399 const struct thread *td)
1402 if (vp->v_type != VREG || td == NULL)
1404 PROC_LOCK(td->td_proc);
1405 if ((uoff_t)uio->uio_offset + uio->uio_resid >
1406 lim_cur(td->td_proc, RLIMIT_FSIZE)) {
1407 kern_psignal(td->td_proc, SIGXFSZ);
1408 PROC_UNLOCK(td->td_proc);
1411 PROC_UNLOCK(td->td_proc);
1416 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
1420 int error, vfslocked;
1423 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1425 vn_lock(vp, LK_SHARED | LK_RETRY);
1426 AUDIT_ARG_VNODE1(vp);
1429 error = setfmode(td, active_cred, vp, mode);
1430 VFS_UNLOCK_GIANT(vfslocked);
1435 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
1439 int error, vfslocked;
1442 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1444 vn_lock(vp, LK_SHARED | LK_RETRY);
1445 AUDIT_ARG_VNODE1(vp);
1448 error = setfown(td, active_cred, vp, uid, gid);
1449 VFS_UNLOCK_GIANT(vfslocked);
1454 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
1458 if ((object = vp->v_object) == NULL)
1460 VM_OBJECT_LOCK(object);
1461 vm_object_page_remove(object, start, end, 0);
1462 VM_OBJECT_UNLOCK(object);
1466 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
1474 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
1475 ("Wrong command %lu", cmd));
1477 if (vn_lock(vp, LK_SHARED) != 0)
1479 if (vp->v_type != VREG) {
1483 error = VOP_GETATTR(vp, &va, cred);
1487 if (noff >= va.va_size) {
1491 bsize = vp->v_mount->mnt_stat.f_iosize;
1492 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
1493 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
1494 if (error == EOPNOTSUPP) {
1498 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
1499 (bnp != -1 && cmd == FIOSEEKDATA)) {
1506 if (noff > va.va_size)
1508 /* noff == va.va_size. There is an implicit hole at the end of file. */
1509 if (cmd == FIOSEEKDATA)