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 * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org>
11 * Copyright (c) 2013 The FreeBSD Foundation
13 * Portions of this software were developed by Konstantin Belousov
14 * under sponsorship from the FreeBSD Foundation.
16 * Redistribution and use in source and binary forms, with or without
17 * modification, are permitted provided that the following conditions
19 * 1. Redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer.
21 * 2. Redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
46 #include <sys/param.h>
47 #include <sys/systm.h>
49 #include <sys/fcntl.h>
55 #include <sys/limits.h>
57 #include <sys/mount.h>
58 #include <sys/mutex.h>
59 #include <sys/namei.h>
60 #include <sys/vnode.h>
63 #include <sys/filio.h>
64 #include <sys/resourcevar.h>
65 #include <sys/rwlock.h>
67 #include <sys/sysctl.h>
68 #include <sys/ttycom.h>
70 #include <sys/syslog.h>
71 #include <sys/unistd.h>
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
77 #include <vm/vm_extern.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_page.h>
83 static fo_rdwr_t vn_read;
84 static fo_rdwr_t vn_write;
85 static fo_rdwr_t vn_io_fault;
86 static fo_truncate_t vn_truncate;
87 static fo_ioctl_t vn_ioctl;
88 static fo_poll_t vn_poll;
89 static fo_kqfilter_t vn_kqfilter;
90 static fo_stat_t vn_statfile;
91 static fo_close_t vn_closefile;
93 struct fileops vnops = {
94 .fo_read = vn_io_fault,
95 .fo_write = vn_io_fault,
96 .fo_truncate = vn_truncate,
99 .fo_kqfilter = vn_kqfilter,
100 .fo_stat = vn_statfile,
101 .fo_close = vn_closefile,
102 .fo_chmod = vn_chmod,
103 .fo_chown = vn_chown,
104 .fo_sendfile = vn_sendfile,
106 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
110 vn_open(ndp, flagp, cmode, fp)
111 struct nameidata *ndp;
115 struct thread *td = ndp->ni_cnd.cn_thread;
117 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
121 * Common code for vnode open operations via a name lookup.
122 * Lookup the vnode and invoke VOP_CREATE if needed.
123 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
125 * Note that this does NOT free nameidata for the successful case,
126 * due to the NDINIT being done elsewhere.
129 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
130 struct ucred *cred, struct file *fp)
134 struct thread *td = ndp->ni_cnd.cn_thread;
136 struct vattr *vap = &vat;
141 if (fmode & O_CREAT) {
142 ndp->ni_cnd.cn_nameiop = CREATE;
143 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF;
144 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
145 ndp->ni_cnd.cn_flags |= FOLLOW;
146 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
147 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
148 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
149 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
151 if ((error = namei(ndp)) != 0)
153 if (ndp->ni_vp == NULL) {
156 vap->va_mode = cmode;
158 vap->va_vaflags |= VA_EXCLUSIVE;
159 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
160 NDFREE(ndp, NDF_ONLY_PNBUF);
162 if ((error = vn_start_write(NULL, &mp,
163 V_XSLEEP | PCATCH)) != 0)
168 error = mac_vnode_check_create(cred, ndp->ni_dvp,
172 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
175 vn_finished_write(mp);
177 NDFREE(ndp, NDF_ONLY_PNBUF);
183 if (ndp->ni_dvp == ndp->ni_vp)
189 if (fmode & O_EXCL) {
196 ndp->ni_cnd.cn_nameiop = LOOKUP;
197 ndp->ni_cnd.cn_flags = ISOPEN |
198 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
199 if (!(fmode & FWRITE))
200 ndp->ni_cnd.cn_flags |= LOCKSHARED;
201 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
202 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
203 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
204 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
205 if ((error = namei(ndp)) != 0)
209 error = vn_open_vnode(vp, fmode, cred, td, fp);
215 NDFREE(ndp, NDF_ONLY_PNBUF);
223 * Common code for vnode open operations once a vnode is located.
224 * Check permissions, and call the VOP_OPEN routine.
227 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
228 struct thread *td, struct file *fp)
233 int error, have_flock, lock_flags, type;
235 if (vp->v_type == VLNK)
237 if (vp->v_type == VSOCK)
239 if (vp->v_type != VDIR && fmode & O_DIRECTORY)
242 if (fmode & (FWRITE | O_TRUNC)) {
243 if (vp->v_type == VDIR)
251 if ((fmode & O_APPEND) && (fmode & FWRITE))
254 error = mac_vnode_check_open(cred, vp, accmode);
258 if ((fmode & O_CREAT) == 0) {
259 if (accmode & VWRITE) {
260 error = vn_writechk(vp);
265 error = VOP_ACCESS(vp, accmode, cred, td);
270 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
271 vn_lock(vp, LK_UPGRADE | LK_RETRY);
272 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
275 if (fmode & (O_EXLOCK | O_SHLOCK)) {
276 KASSERT(fp != NULL, ("open with flock requires fp"));
277 lock_flags = VOP_ISLOCKED(vp);
279 lf.l_whence = SEEK_SET;
282 if (fmode & O_EXLOCK)
287 if ((fmode & FNONBLOCK) == 0)
289 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
290 have_flock = (error == 0);
291 vn_lock(vp, lock_flags | LK_RETRY);
292 if (error == 0 && vp->v_iflag & VI_DOOMED)
295 * Another thread might have used this vnode as an
296 * executable while the vnode lock was dropped.
297 * Ensure the vnode is still able to be opened for
298 * writing after the lock has been obtained.
300 if (error == 0 && accmode & VWRITE)
301 error = vn_writechk(vp);
305 lf.l_whence = SEEK_SET;
309 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
312 vn_start_write(vp, &mp, V_WAIT);
313 vn_lock(vp, lock_flags | LK_RETRY);
314 (void)VOP_CLOSE(vp, fmode, cred, td);
315 vn_finished_write(mp);
316 /* Prevent second close from fdrop()->vn_close(). */
318 fp->f_ops= &badfileops;
321 fp->f_flag |= FHASLOCK;
323 if (fmode & FWRITE) {
324 VOP_ADD_WRITECOUNT(vp, 1);
325 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
326 __func__, vp, vp->v_writecount);
328 ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
333 * Check for write permissions on the specified vnode.
334 * Prototype text segments cannot be written.
338 register struct vnode *vp;
341 ASSERT_VOP_LOCKED(vp, "vn_writechk");
343 * If there's shared text associated with
344 * the vnode, try to free it up once. If
345 * we fail, we can't allow writing.
357 vn_close(vp, flags, file_cred, td)
358 register struct vnode *vp;
360 struct ucred *file_cred;
364 int error, lock_flags;
366 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
367 MNT_EXTENDED_SHARED(vp->v_mount))
368 lock_flags = LK_SHARED;
370 lock_flags = LK_EXCLUSIVE;
372 vn_start_write(vp, &mp, V_WAIT);
373 vn_lock(vp, lock_flags | LK_RETRY);
374 if (flags & FWRITE) {
375 VNASSERT(vp->v_writecount > 0, vp,
376 ("vn_close: negative writecount"));
377 VOP_ADD_WRITECOUNT(vp, -1);
378 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
379 __func__, vp, vp->v_writecount);
381 error = VOP_CLOSE(vp, flags, file_cred, td);
383 vn_finished_write(mp);
388 * Heuristic to detect sequential operation.
391 sequential_heuristic(struct uio *uio, struct file *fp)
394 if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
395 return (fp->f_seqcount << IO_SEQSHIFT);
398 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
399 * that the first I/O is normally considered to be slightly
400 * sequential. Seeking to offset 0 doesn't change sequentiality
401 * unless previous seeks have reduced f_seqcount to 0, in which
402 * case offset 0 is not special.
404 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
405 uio->uio_offset == fp->f_nextoff) {
407 * f_seqcount is in units of fixed-size blocks so that it
408 * depends mainly on the amount of sequential I/O and not
409 * much on the number of sequential I/O's. The fixed size
410 * of 16384 is hard-coded here since it is (not quite) just
411 * a magic size that works well here. This size is more
412 * closely related to the best I/O size for real disks than
413 * to any block size used by software.
415 fp->f_seqcount += howmany(uio->uio_resid, 16384);
416 if (fp->f_seqcount > IO_SEQMAX)
417 fp->f_seqcount = IO_SEQMAX;
418 return (fp->f_seqcount << IO_SEQSHIFT);
421 /* Not sequential. Quickly draw-down sequentiality. */
422 if (fp->f_seqcount > 1)
430 * Package up an I/O request on a vnode into a uio and do it.
433 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
434 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
435 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
442 int error, lock_flags;
444 auio.uio_iov = &aiov;
446 aiov.iov_base = base;
448 auio.uio_resid = len;
449 auio.uio_offset = offset;
450 auio.uio_segflg = segflg;
455 if ((ioflg & IO_NODELOCKED) == 0) {
456 if (rw == UIO_READ) {
457 rl_cookie = vn_rangelock_rlock(vp, offset,
460 rl_cookie = vn_rangelock_wlock(vp, offset,
464 if (rw == UIO_WRITE) {
465 if (vp->v_type != VCHR &&
466 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
469 if (MNT_SHARED_WRITES(mp) ||
470 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
471 lock_flags = LK_SHARED;
473 lock_flags = LK_EXCLUSIVE;
475 lock_flags = LK_SHARED;
476 vn_lock(vp, lock_flags | LK_RETRY);
480 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
482 if ((ioflg & IO_NOMACCHECK) == 0) {
484 error = mac_vnode_check_read(active_cred, file_cred,
487 error = mac_vnode_check_write(active_cred, file_cred,
492 if (file_cred != NULL)
497 error = VOP_READ(vp, &auio, ioflg, cred);
499 error = VOP_WRITE(vp, &auio, ioflg, cred);
502 *aresid = auio.uio_resid;
504 if (auio.uio_resid && error == 0)
506 if ((ioflg & IO_NODELOCKED) == 0) {
509 vn_finished_write(mp);
512 if (rl_cookie != NULL)
513 vn_rangelock_unlock(vp, rl_cookie);
518 * Package up an I/O request on a vnode into a uio and do it. The I/O
519 * request is split up into smaller chunks and we try to avoid saturating
520 * the buffer cache while potentially holding a vnode locked, so we
521 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
522 * to give other processes a chance to lock the vnode (either other processes
523 * core'ing the same binary, or unrelated processes scanning the directory).
526 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
527 file_cred, aresid, td)
535 struct ucred *active_cred;
536 struct ucred *file_cred;
547 * Force `offset' to a multiple of MAXBSIZE except possibly
548 * for the first chunk, so that filesystems only need to
549 * write full blocks except possibly for the first and last
552 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
556 if (rw != UIO_READ && vp->v_type == VREG)
559 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
560 ioflg, active_cred, file_cred, &iaresid, td);
561 len -= chunk; /* aresid calc already includes length */
565 base = (char *)base + chunk;
566 kern_yield(PRI_USER);
569 *aresid = len + iaresid;
574 foffset_lock(struct file *fp, int flags)
579 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
581 #if OFF_MAX <= LONG_MAX
583 * Caller only wants the current f_offset value. Assume that
584 * the long and shorter integer types reads are atomic.
586 if ((flags & FOF_NOLOCK) != 0)
587 return (fp->f_offset);
591 * According to McKusick the vn lock was protecting f_offset here.
592 * It is now protected by the FOFFSET_LOCKED flag.
594 mtxp = mtx_pool_find(mtxpool_sleep, fp);
596 if ((flags & FOF_NOLOCK) == 0) {
597 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
598 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
599 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
602 fp->f_vnread_flags |= FOFFSET_LOCKED;
610 foffset_unlock(struct file *fp, off_t val, int flags)
614 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
616 #if OFF_MAX <= LONG_MAX
617 if ((flags & FOF_NOLOCK) != 0) {
618 if ((flags & FOF_NOUPDATE) == 0)
620 if ((flags & FOF_NEXTOFF) != 0)
626 mtxp = mtx_pool_find(mtxpool_sleep, fp);
628 if ((flags & FOF_NOUPDATE) == 0)
630 if ((flags & FOF_NEXTOFF) != 0)
632 if ((flags & FOF_NOLOCK) == 0) {
633 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
634 ("Lost FOFFSET_LOCKED"));
635 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
636 wakeup(&fp->f_vnread_flags);
637 fp->f_vnread_flags = 0;
643 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
646 if ((flags & FOF_OFFSET) == 0)
647 uio->uio_offset = foffset_lock(fp, flags);
651 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
654 if ((flags & FOF_OFFSET) == 0)
655 foffset_unlock(fp, uio->uio_offset, flags);
659 get_advice(struct file *fp, struct uio *uio)
664 ret = POSIX_FADV_NORMAL;
665 if (fp->f_advice == NULL)
668 mtxp = mtx_pool_find(mtxpool_sleep, fp);
670 if (uio->uio_offset >= fp->f_advice->fa_start &&
671 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
672 ret = fp->f_advice->fa_advice;
678 * File table vnode read routine.
681 vn_read(fp, uio, active_cred, flags, td)
684 struct ucred *active_cred;
692 off_t offset, start, end;
694 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
696 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
699 if (fp->f_flag & FNONBLOCK)
701 if (fp->f_flag & O_DIRECT)
703 advice = get_advice(fp, uio);
704 vn_lock(vp, LK_SHARED | LK_RETRY);
707 case POSIX_FADV_NORMAL:
708 case POSIX_FADV_SEQUENTIAL:
709 case POSIX_FADV_NOREUSE:
710 ioflag |= sequential_heuristic(uio, fp);
712 case POSIX_FADV_RANDOM:
713 /* Disable read-ahead for random I/O. */
716 offset = uio->uio_offset;
719 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
722 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
723 fp->f_nextoff = uio->uio_offset;
725 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
726 offset != uio->uio_offset) {
728 * Use POSIX_FADV_DONTNEED to flush clean pages and
729 * buffers for the backing file after a
730 * POSIX_FADV_NOREUSE read(2). To optimize the common
731 * case of using POSIX_FADV_NOREUSE with sequential
732 * access, track the previous implicit DONTNEED
733 * request and grow this request to include the
734 * current read(2) in addition to the previous
735 * DONTNEED. With purely sequential access this will
736 * cause the DONTNEED requests to continously grow to
737 * cover all of the previously read regions of the
738 * file. This allows filesystem blocks that are
739 * accessed by multiple calls to read(2) to be flushed
740 * once the last read(2) finishes.
743 end = uio->uio_offset - 1;
744 mtxp = mtx_pool_find(mtxpool_sleep, fp);
746 if (fp->f_advice != NULL &&
747 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
748 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
749 start = fp->f_advice->fa_prevstart;
750 else if (fp->f_advice->fa_prevstart != 0 &&
751 fp->f_advice->fa_prevstart == end + 1)
752 end = fp->f_advice->fa_prevend;
753 fp->f_advice->fa_prevstart = start;
754 fp->f_advice->fa_prevend = end;
757 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
763 * File table vnode write routine.
766 vn_write(fp, uio, active_cred, flags, td)
769 struct ucred *active_cred;
776 int error, ioflag, lock_flags;
778 off_t offset, start, end;
780 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
782 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
784 if (vp->v_type == VREG)
787 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
789 if (fp->f_flag & FNONBLOCK)
791 if (fp->f_flag & O_DIRECT)
793 if ((fp->f_flag & O_FSYNC) ||
794 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
797 if (vp->v_type != VCHR &&
798 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
801 advice = get_advice(fp, uio);
803 if (MNT_SHARED_WRITES(mp) ||
804 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
805 lock_flags = LK_SHARED;
807 lock_flags = LK_EXCLUSIVE;
810 vn_lock(vp, lock_flags | LK_RETRY);
812 case POSIX_FADV_NORMAL:
813 case POSIX_FADV_SEQUENTIAL:
814 case POSIX_FADV_NOREUSE:
815 ioflag |= sequential_heuristic(uio, fp);
817 case POSIX_FADV_RANDOM:
818 /* XXX: Is this correct? */
821 offset = uio->uio_offset;
824 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
827 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
828 fp->f_nextoff = uio->uio_offset;
830 if (vp->v_type != VCHR)
831 vn_finished_write(mp);
832 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
833 offset != uio->uio_offset) {
835 * Use POSIX_FADV_DONTNEED to flush clean pages and
836 * buffers for the backing file after a
837 * POSIX_FADV_NOREUSE write(2). To optimize the
838 * common case of using POSIX_FADV_NOREUSE with
839 * sequential access, track the previous implicit
840 * DONTNEED request and grow this request to include
841 * the current write(2) in addition to the previous
842 * DONTNEED. With purely sequential access this will
843 * cause the DONTNEED requests to continously grow to
844 * cover all of the previously written regions of the
847 * Note that the blocks just written are almost
848 * certainly still dirty, so this only works when
849 * VOP_ADVISE() calls from subsequent writes push out
850 * the data written by this write(2) once the backing
851 * buffers are clean. However, as compared to forcing
852 * IO_DIRECT, this gives much saner behavior. Write
853 * clustering is still allowed, and clean pages are
854 * merely moved to the cache page queue rather than
855 * outright thrown away. This means a subsequent
856 * read(2) can still avoid hitting the disk if the
857 * pages have not been reclaimed.
859 * This does make POSIX_FADV_NOREUSE largely useless
860 * with non-sequential access. However, sequential
861 * access is the more common use case and the flag is
865 end = uio->uio_offset - 1;
866 mtxp = mtx_pool_find(mtxpool_sleep, fp);
868 if (fp->f_advice != NULL &&
869 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
870 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
871 start = fp->f_advice->fa_prevstart;
872 else if (fp->f_advice->fa_prevstart != 0 &&
873 fp->f_advice->fa_prevstart == end + 1)
874 end = fp->f_advice->fa_prevend;
875 fp->f_advice->fa_prevstart = start;
876 fp->f_advice->fa_prevend = end;
879 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
886 static const int io_hold_cnt = 16;
887 static int vn_io_fault_enable = 1;
888 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
889 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
890 static u_long vn_io_faults_cnt;
891 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
892 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
895 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
896 * prevent the following deadlock:
898 * Assume that the thread A reads from the vnode vp1 into userspace
899 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
900 * currently not resident, then system ends up with the call chain
901 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
902 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
903 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
904 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
905 * backed by the pages of vnode vp1, and some page in buf2 is not
906 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
908 * To prevent the lock order reversal and deadlock, vn_io_fault() does
909 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
910 * Instead, it first tries to do the whole range i/o with pagefaults
911 * disabled. If all pages in the i/o buffer are resident and mapped,
912 * VOP will succeed (ignoring the genuine filesystem errors).
913 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
914 * i/o in chunks, with all pages in the chunk prefaulted and held
915 * using vm_fault_quick_hold_pages().
917 * Filesystems using this deadlock avoidance scheme should use the
918 * array of the held pages from uio, saved in the curthread->td_ma,
919 * instead of doing uiomove(). A helper function
920 * vn_io_fault_uiomove() converts uiomove request into
921 * uiomove_fromphys() over td_ma array.
923 * Since vnode locks do not cover the whole i/o anymore, rangelocks
924 * make the current i/o request atomic with respect to other i/os and
928 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
929 int flags, struct thread *td)
931 vm_page_t ma[io_hold_cnt + 2];
932 struct uio *uio_clone, short_uio;
933 struct iovec short_iovec[1];
938 vm_page_t *prev_td_ma;
939 int error, cnt, save, saveheld, prev_td_ma_cnt;
940 vm_offset_t addr, end;
945 if (uio->uio_rw == UIO_READ)
950 foffset_lock_uio(fp, uio, flags);
952 if (uio->uio_segflg != UIO_USERSPACE || vp->v_type != VREG ||
953 ((mp = vp->v_mount) != NULL &&
954 (mp->mnt_kern_flag & MNTK_NO_IOPF) == 0) ||
955 !vn_io_fault_enable) {
956 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
961 * The UFS follows IO_UNIT directive and replays back both
962 * uio_offset and uio_resid if an error is encountered during the
963 * operation. But, since the iovec may be already advanced,
964 * uio is still in an inconsistent state.
966 * Cache a copy of the original uio, which is advanced to the redo
967 * point using UIO_NOCOPY below.
969 uio_clone = cloneuio(uio);
970 resid = uio->uio_resid;
972 short_uio.uio_segflg = UIO_USERSPACE;
973 short_uio.uio_rw = uio->uio_rw;
974 short_uio.uio_td = uio->uio_td;
976 if (uio->uio_rw == UIO_READ) {
977 prot = VM_PROT_WRITE;
978 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
979 uio->uio_offset + uio->uio_resid);
982 if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0)
983 /* For appenders, punt and lock the whole range. */
984 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
986 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
987 uio->uio_offset + uio->uio_resid);
990 save = vm_fault_disable_pagefaults();
991 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
995 atomic_add_long(&vn_io_faults_cnt, 1);
996 uio_clone->uio_segflg = UIO_NOCOPY;
997 uiomove(NULL, resid - uio->uio_resid, uio_clone);
998 uio_clone->uio_segflg = uio->uio_segflg;
1000 saveheld = curthread_pflags_set(TDP_UIOHELD);
1001 prev_td_ma = td->td_ma;
1002 prev_td_ma_cnt = td->td_ma_cnt;
1004 while (uio_clone->uio_resid != 0) {
1005 len = uio_clone->uio_iov->iov_len;
1007 KASSERT(uio_clone->uio_iovcnt >= 1,
1008 ("iovcnt underflow"));
1009 uio_clone->uio_iov++;
1010 uio_clone->uio_iovcnt--;
1013 if (len > io_hold_cnt * PAGE_SIZE)
1014 len = io_hold_cnt * PAGE_SIZE;
1015 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1016 end = round_page(addr + len);
1021 cnt = atop(end - trunc_page(addr));
1023 * A perfectly misaligned address and length could cause
1024 * both the start and the end of the chunk to use partial
1025 * page. +2 accounts for such a situation.
1027 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1028 addr, len, prot, ma, io_hold_cnt + 2);
1033 short_uio.uio_iov = &short_iovec[0];
1034 short_iovec[0].iov_base = (void *)addr;
1035 short_uio.uio_iovcnt = 1;
1036 short_uio.uio_resid = short_iovec[0].iov_len = len;
1037 short_uio.uio_offset = uio_clone->uio_offset;
1039 td->td_ma_cnt = cnt;
1041 error = doio(fp, &short_uio, active_cred, flags | FOF_OFFSET,
1043 vm_page_unhold_pages(ma, cnt);
1044 adv = len - short_uio.uio_resid;
1046 uio_clone->uio_iov->iov_base =
1047 (char *)uio_clone->uio_iov->iov_base + adv;
1048 uio_clone->uio_iov->iov_len -= adv;
1049 uio_clone->uio_resid -= adv;
1050 uio_clone->uio_offset += adv;
1052 uio->uio_resid -= adv;
1053 uio->uio_offset += adv;
1055 if (error != 0 || adv == 0)
1058 td->td_ma = prev_td_ma;
1059 td->td_ma_cnt = prev_td_ma_cnt;
1060 curthread_pflags_restore(saveheld);
1062 vm_fault_enable_pagefaults(save);
1063 vn_rangelock_unlock(vp, rl_cookie);
1064 free(uio_clone, M_IOV);
1066 foffset_unlock_uio(fp, uio, flags);
1071 * Helper function to perform the requested uiomove operation using
1072 * the held pages for io->uio_iov[0].iov_base buffer instead of
1073 * copyin/copyout. Access to the pages with uiomove_fromphys()
1074 * instead of iov_base prevents page faults that could occur due to
1075 * pmap_collect() invalidating the mapping created by
1076 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1077 * object cleanup revoking the write access from page mappings.
1079 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1080 * instead of plain uiomove().
1083 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1085 struct uio transp_uio;
1086 struct iovec transp_iov[1];
1092 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1093 uio->uio_segflg != UIO_USERSPACE)
1094 return (uiomove(data, xfersize, uio));
1096 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1097 transp_iov[0].iov_base = data;
1098 transp_uio.uio_iov = &transp_iov[0];
1099 transp_uio.uio_iovcnt = 1;
1100 if (xfersize > uio->uio_resid)
1101 xfersize = uio->uio_resid;
1102 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1103 transp_uio.uio_offset = 0;
1104 transp_uio.uio_segflg = UIO_SYSSPACE;
1106 * Since transp_iov points to data, and td_ma page array
1107 * corresponds to original uio->uio_iov, we need to invert the
1108 * direction of the i/o operation as passed to
1109 * uiomove_fromphys().
1111 switch (uio->uio_rw) {
1113 transp_uio.uio_rw = UIO_READ;
1116 transp_uio.uio_rw = UIO_WRITE;
1119 transp_uio.uio_td = uio->uio_td;
1120 error = uiomove_fromphys(td->td_ma,
1121 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1122 xfersize, &transp_uio);
1123 adv = xfersize - transp_uio.uio_resid;
1125 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1126 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1128 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1130 td->td_ma_cnt -= pgadv;
1131 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1132 uio->uio_iov->iov_len -= adv;
1133 uio->uio_resid -= adv;
1134 uio->uio_offset += adv;
1139 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1143 vm_offset_t iov_base;
1147 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1148 uio->uio_segflg != UIO_USERSPACE)
1149 return (uiomove_fromphys(ma, offset, xfersize, uio));
1151 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1152 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1153 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1154 switch (uio->uio_rw) {
1156 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1160 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1164 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1166 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1168 td->td_ma_cnt -= pgadv;
1169 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1170 uio->uio_iov->iov_len -= cnt;
1171 uio->uio_resid -= cnt;
1172 uio->uio_offset += cnt;
1178 * File table truncate routine.
1181 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1193 * Lock the whole range for truncation. Otherwise split i/o
1194 * might happen partly before and partly after the truncation.
1196 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1197 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1200 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1201 if (vp->v_type == VDIR) {
1206 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1210 error = vn_writechk(vp);
1213 vattr.va_size = length;
1214 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1218 vn_finished_write(mp);
1220 vn_rangelock_unlock(vp, rl_cookie);
1225 * File table vnode stat routine.
1228 vn_statfile(fp, sb, active_cred, td)
1231 struct ucred *active_cred;
1234 struct vnode *vp = fp->f_vnode;
1237 vn_lock(vp, LK_SHARED | LK_RETRY);
1238 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1245 * Stat a vnode; implementation for the stat syscall
1248 vn_stat(vp, sb, active_cred, file_cred, td)
1250 register struct stat *sb;
1251 struct ucred *active_cred;
1252 struct ucred *file_cred;
1256 register struct vattr *vap;
1261 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1269 * Initialize defaults for new and unusual fields, so that file
1270 * systems which don't support these fields don't need to know
1273 vap->va_birthtime.tv_sec = -1;
1274 vap->va_birthtime.tv_nsec = 0;
1275 vap->va_fsid = VNOVAL;
1276 vap->va_rdev = NODEV;
1278 error = VOP_GETATTR(vp, vap, active_cred);
1283 * Zero the spare stat fields
1285 bzero(sb, sizeof *sb);
1288 * Copy from vattr table
1290 if (vap->va_fsid != VNOVAL)
1291 sb->st_dev = vap->va_fsid;
1293 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1294 sb->st_ino = vap->va_fileid;
1295 mode = vap->va_mode;
1296 switch (vap->va_type) {
1322 sb->st_nlink = vap->va_nlink;
1323 sb->st_uid = vap->va_uid;
1324 sb->st_gid = vap->va_gid;
1325 sb->st_rdev = vap->va_rdev;
1326 if (vap->va_size > OFF_MAX)
1328 sb->st_size = vap->va_size;
1329 sb->st_atim = vap->va_atime;
1330 sb->st_mtim = vap->va_mtime;
1331 sb->st_ctim = vap->va_ctime;
1332 sb->st_birthtim = vap->va_birthtime;
1335 * According to www.opengroup.org, the meaning of st_blksize is
1336 * "a filesystem-specific preferred I/O block size for this
1337 * object. In some filesystem types, this may vary from file
1339 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1342 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1344 sb->st_flags = vap->va_flags;
1345 if (priv_check(td, PRIV_VFS_GENERATION))
1348 sb->st_gen = vap->va_gen;
1350 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1355 * File table vnode ioctl routine.
1358 vn_ioctl(fp, com, data, active_cred, td)
1362 struct ucred *active_cred;
1370 switch (vp->v_type) {
1375 vn_lock(vp, LK_SHARED | LK_RETRY);
1376 error = VOP_GETATTR(vp, &vattr, active_cred);
1379 *(int *)data = vattr.va_size - fp->f_offset;
1385 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1394 * File table vnode poll routine.
1397 vn_poll(fp, events, active_cred, td)
1400 struct ucred *active_cred;
1408 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1409 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1414 error = VOP_POLL(vp, events, fp->f_cred, td);
1419 * Acquire the requested lock and then check for validity. LK_RETRY
1420 * permits vn_lock to return doomed vnodes.
1423 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1427 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1428 ("vn_lock called with no locktype."));
1430 #ifdef DEBUG_VFS_LOCKS
1431 KASSERT(vp->v_holdcnt != 0,
1432 ("vn_lock %p: zero hold count", vp));
1434 error = VOP_LOCK1(vp, flags, file, line);
1435 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1436 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1437 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1440 * Callers specify LK_RETRY if they wish to get dead vnodes.
1441 * If RETRY is not set, we return ENOENT instead.
1443 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1444 (flags & LK_RETRY) == 0) {
1449 } while (flags & LK_RETRY && error != 0);
1454 * File table vnode close routine.
1457 vn_closefile(fp, td)
1466 fp->f_ops = &badfileops;
1468 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1471 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1473 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1474 lf.l_whence = SEEK_SET;
1477 lf.l_type = F_UNLCK;
1478 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1485 * Preparing to start a filesystem write operation. If the operation is
1486 * permitted, then we bump the count of operations in progress and
1487 * proceed. If a suspend request is in progress, we wait until the
1488 * suspension is over, and then proceed.
1491 vn_start_write_locked(struct mount *mp, int flags)
1495 mtx_assert(MNT_MTX(mp), MA_OWNED);
1499 * Check on status of suspension.
1501 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1502 mp->mnt_susp_owner != curthread) {
1503 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1504 if (flags & V_NOWAIT) {
1505 error = EWOULDBLOCK;
1508 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1509 (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1514 if (flags & V_XSLEEP)
1516 mp->mnt_writeopcount++;
1518 if (error != 0 || (flags & V_XSLEEP) != 0)
1525 vn_start_write(vp, mpp, flags)
1535 * If a vnode is provided, get and return the mount point that
1536 * to which it will write.
1539 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1541 if (error != EOPNOTSUPP)
1546 if ((mp = *mpp) == NULL)
1550 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1552 * As long as a vnode is not provided we need to acquire a
1553 * refcount for the provided mountpoint too, in order to
1554 * emulate a vfs_ref().
1560 return (vn_start_write_locked(mp, flags));
1564 * Secondary suspension. Used by operations such as vop_inactive
1565 * routines that are needed by the higher level functions. These
1566 * are allowed to proceed until all the higher level functions have
1567 * completed (indicated by mnt_writeopcount dropping to zero). At that
1568 * time, these operations are halted until the suspension is over.
1571 vn_start_secondary_write(vp, mpp, flags)
1581 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1583 if (error != EOPNOTSUPP)
1589 * If we are not suspended or have not yet reached suspended
1590 * mode, then let the operation proceed.
1592 if ((mp = *mpp) == NULL)
1596 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1598 * As long as a vnode is not provided we need to acquire a
1599 * refcount for the provided mountpoint too, in order to
1600 * emulate a vfs_ref().
1605 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1606 mp->mnt_secondary_writes++;
1607 mp->mnt_secondary_accwrites++;
1611 if (flags & V_NOWAIT) {
1614 return (EWOULDBLOCK);
1617 * Wait for the suspension to finish.
1619 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1620 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1628 * Filesystem write operation has completed. If we are suspending and this
1629 * operation is the last one, notify the suspender that the suspension is
1633 vn_finished_write(mp)
1640 mp->mnt_writeopcount--;
1641 if (mp->mnt_writeopcount < 0)
1642 panic("vn_finished_write: neg cnt");
1643 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1644 mp->mnt_writeopcount <= 0)
1645 wakeup(&mp->mnt_writeopcount);
1651 * Filesystem secondary write operation has completed. If we are
1652 * suspending and this operation is the last one, notify the suspender
1653 * that the suspension is now in effect.
1656 vn_finished_secondary_write(mp)
1663 mp->mnt_secondary_writes--;
1664 if (mp->mnt_secondary_writes < 0)
1665 panic("vn_finished_secondary_write: neg cnt");
1666 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1667 mp->mnt_secondary_writes <= 0)
1668 wakeup(&mp->mnt_secondary_writes);
1675 * Request a filesystem to suspend write operations.
1678 vfs_write_suspend(struct mount *mp, int flags)
1683 if (mp->mnt_susp_owner == curthread) {
1687 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1688 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1691 * Unmount holds a write reference on the mount point. If we
1692 * own busy reference and drain for writers, we deadlock with
1693 * the reference draining in the unmount path. Callers of
1694 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1695 * vfs_busy() reference is owned and caller is not in the
1698 if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1699 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1704 mp->mnt_kern_flag |= MNTK_SUSPEND;
1705 mp->mnt_susp_owner = curthread;
1706 if (mp->mnt_writeopcount > 0)
1707 (void) msleep(&mp->mnt_writeopcount,
1708 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1711 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1712 vfs_write_resume(mp, 0);
1717 * Request a filesystem to resume write operations.
1720 vfs_write_resume(struct mount *mp, int flags)
1724 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1725 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1726 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1728 mp->mnt_susp_owner = NULL;
1729 wakeup(&mp->mnt_writeopcount);
1730 wakeup(&mp->mnt_flag);
1731 curthread->td_pflags &= ~TDP_IGNSUSP;
1732 if ((flags & VR_START_WRITE) != 0) {
1734 mp->mnt_writeopcount++;
1737 if ((flags & VR_NO_SUSPCLR) == 0)
1739 } else if ((flags & VR_START_WRITE) != 0) {
1741 vn_start_write_locked(mp, 0);
1748 * Implement kqueues for files by translating it to vnode operation.
1751 vn_kqfilter(struct file *fp, struct knote *kn)
1754 return (VOP_KQFILTER(fp->f_vnode, kn));
1758 * Simplified in-kernel wrapper calls for extended attribute access.
1759 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1760 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1763 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1764 const char *attrname, int *buflen, char *buf, struct thread *td)
1770 iov.iov_len = *buflen;
1773 auio.uio_iov = &iov;
1774 auio.uio_iovcnt = 1;
1775 auio.uio_rw = UIO_READ;
1776 auio.uio_segflg = UIO_SYSSPACE;
1778 auio.uio_offset = 0;
1779 auio.uio_resid = *buflen;
1781 if ((ioflg & IO_NODELOCKED) == 0)
1782 vn_lock(vp, LK_SHARED | LK_RETRY);
1784 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1786 /* authorize attribute retrieval as kernel */
1787 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1790 if ((ioflg & IO_NODELOCKED) == 0)
1794 *buflen = *buflen - auio.uio_resid;
1801 * XXX failure mode if partially written?
1804 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1805 const char *attrname, int buflen, char *buf, struct thread *td)
1812 iov.iov_len = buflen;
1815 auio.uio_iov = &iov;
1816 auio.uio_iovcnt = 1;
1817 auio.uio_rw = UIO_WRITE;
1818 auio.uio_segflg = UIO_SYSSPACE;
1820 auio.uio_offset = 0;
1821 auio.uio_resid = buflen;
1823 if ((ioflg & IO_NODELOCKED) == 0) {
1824 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1826 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1829 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1831 /* authorize attribute setting as kernel */
1832 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1834 if ((ioflg & IO_NODELOCKED) == 0) {
1835 vn_finished_write(mp);
1843 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1844 const char *attrname, struct thread *td)
1849 if ((ioflg & IO_NODELOCKED) == 0) {
1850 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1852 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1855 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1857 /* authorize attribute removal as kernel */
1858 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1859 if (error == EOPNOTSUPP)
1860 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1863 if ((ioflg & IO_NODELOCKED) == 0) {
1864 vn_finished_write(mp);
1872 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1878 ltype = VOP_ISLOCKED(vp);
1879 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1880 ("vn_vget_ino: vp not locked"));
1881 error = vfs_busy(mp, MBF_NOWAIT);
1885 error = vfs_busy(mp, 0);
1886 vn_lock(vp, ltype | LK_RETRY);
1890 if (vp->v_iflag & VI_DOOMED) {
1896 error = VFS_VGET(mp, ino, lkflags, rvp);
1898 vn_lock(vp, ltype | LK_RETRY);
1899 if (vp->v_iflag & VI_DOOMED) {
1908 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
1909 const struct thread *td)
1912 if (vp->v_type != VREG || td == NULL)
1914 PROC_LOCK(td->td_proc);
1915 if ((uoff_t)uio->uio_offset + uio->uio_resid >
1916 lim_cur(td->td_proc, RLIMIT_FSIZE)) {
1917 kern_psignal(td->td_proc, SIGXFSZ);
1918 PROC_UNLOCK(td->td_proc);
1921 PROC_UNLOCK(td->td_proc);
1926 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
1933 vn_lock(vp, LK_SHARED | LK_RETRY);
1934 AUDIT_ARG_VNODE1(vp);
1937 return (setfmode(td, active_cred, vp, mode));
1941 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
1948 vn_lock(vp, LK_SHARED | LK_RETRY);
1949 AUDIT_ARG_VNODE1(vp);
1952 return (setfown(td, active_cred, vp, uid, gid));
1956 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
1960 if ((object = vp->v_object) == NULL)
1962 VM_OBJECT_WLOCK(object);
1963 vm_object_page_remove(object, start, end, 0);
1964 VM_OBJECT_WUNLOCK(object);
1968 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
1976 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
1977 ("Wrong command %lu", cmd));
1979 if (vn_lock(vp, LK_SHARED) != 0)
1981 if (vp->v_type != VREG) {
1985 error = VOP_GETATTR(vp, &va, cred);
1989 if (noff >= va.va_size) {
1993 bsize = vp->v_mount->mnt_stat.f_iosize;
1994 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
1995 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
1996 if (error == EOPNOTSUPP) {
2000 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2001 (bnp != -1 && cmd == FIOSEEKDATA)) {
2008 if (noff > va.va_size)
2010 /* noff == va.va_size. There is an implicit hole at the end of file. */
2011 if (cmd == FIOSEEKDATA)
2021 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2026 off_t foffset, size;
2029 cred = td->td_ucred;
2031 foffset = foffset_lock(fp, 0);
2032 noneg = (vp->v_type != VCHR);
2038 (offset > 0 && foffset > OFF_MAX - offset))) {
2045 vn_lock(vp, LK_SHARED | LK_RETRY);
2046 error = VOP_GETATTR(vp, &vattr, cred);
2052 * If the file references a disk device, then fetch
2053 * the media size and use that to determine the ending
2056 if (vattr.va_size == 0 && vp->v_type == VCHR &&
2057 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2058 vattr.va_size = size;
2060 (vattr.va_size > OFF_MAX ||
2061 (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2065 offset += vattr.va_size;
2070 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2073 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2078 if (error == 0 && noneg && offset < 0)
2082 VFS_KNOTE_UNLOCKED(vp, 0);
2083 *(off_t *)(td->td_retval) = offset;
2085 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);