2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org>
13 * Copyright (c) 2013, 2014 The FreeBSD Foundation
15 * Portions of this software were developed by Konstantin Belousov
16 * under sponsorship from the FreeBSD Foundation.
18 * Redistribution and use in source and binary forms, with or without
19 * modification, are permitted provided that the following conditions
21 * 1. Redistributions of source code must retain the above copyright
22 * notice, this list of conditions and the following disclaimer.
23 * 2. Redistributions in binary form must reproduce the above copyright
24 * notice, this list of conditions and the following disclaimer in the
25 * documentation and/or other materials provided with the distribution.
26 * 3. Neither the name of the University nor the names of its contributors
27 * may be used to endorse or promote products derived from this software
28 * without specific prior written permission.
30 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
31 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
34 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
35 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
36 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
37 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
38 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
39 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
42 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
48 #include "opt_hwpmc_hooks.h"
50 #include <sys/param.h>
51 #include <sys/systm.h>
54 #include <sys/fcntl.h>
60 #include <sys/limits.h>
63 #include <sys/mount.h>
64 #include <sys/mutex.h>
65 #include <sys/namei.h>
66 #include <sys/vnode.h>
69 #include <sys/filio.h>
70 #include <sys/resourcevar.h>
71 #include <sys/rwlock.h>
73 #include <sys/sysctl.h>
74 #include <sys/ttycom.h>
76 #include <sys/syslog.h>
77 #include <sys/unistd.h>
80 #include <security/audit/audit.h>
81 #include <security/mac/mac_framework.h>
84 #include <vm/vm_extern.h>
86 #include <vm/vm_map.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_page.h>
89 #include <vm/vnode_pager.h>
92 #include <sys/pmckern.h>
95 static fo_rdwr_t vn_read;
96 static fo_rdwr_t vn_write;
97 static fo_rdwr_t vn_io_fault;
98 static fo_truncate_t vn_truncate;
99 static fo_ioctl_t vn_ioctl;
100 static fo_poll_t vn_poll;
101 static fo_kqfilter_t vn_kqfilter;
102 static fo_stat_t vn_statfile;
103 static fo_close_t vn_closefile;
104 static fo_mmap_t vn_mmap;
106 struct fileops vnops = {
107 .fo_read = vn_io_fault,
108 .fo_write = vn_io_fault,
109 .fo_truncate = vn_truncate,
110 .fo_ioctl = vn_ioctl,
112 .fo_kqfilter = vn_kqfilter,
113 .fo_stat = vn_statfile,
114 .fo_close = vn_closefile,
115 .fo_chmod = vn_chmod,
116 .fo_chown = vn_chown,
117 .fo_sendfile = vn_sendfile,
119 .fo_fill_kinfo = vn_fill_kinfo,
121 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
124 static const int io_hold_cnt = 16;
125 static int vn_io_fault_enable = 1;
126 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
127 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
128 static int vn_io_fault_prefault = 0;
129 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW,
130 &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting");
131 static u_long vn_io_faults_cnt;
132 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
133 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
136 * Returns true if vn_io_fault mode of handling the i/o request should
140 do_vn_io_fault(struct vnode *vp, struct uio *uio)
144 return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
145 (mp = vp->v_mount) != NULL &&
146 (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
150 * Structure used to pass arguments to vn_io_fault1(), to do either
151 * file- or vnode-based I/O calls.
153 struct vn_io_fault_args {
161 struct fop_args_tag {
165 struct vop_args_tag {
171 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
172 struct vn_io_fault_args *args, struct thread *td);
175 vn_open(ndp, flagp, cmode, fp)
176 struct nameidata *ndp;
180 struct thread *td = ndp->ni_cnd.cn_thread;
182 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
186 * Common code for vnode open operations via a name lookup.
187 * Lookup the vnode and invoke VOP_CREATE if needed.
188 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
190 * Note that this does NOT free nameidata for the successful case,
191 * due to the NDINIT being done elsewhere.
194 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
195 struct ucred *cred, struct file *fp)
199 struct thread *td = ndp->ni_cnd.cn_thread;
201 struct vattr *vap = &vat;
206 if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT |
207 O_EXCL | O_DIRECTORY))
209 else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) {
210 ndp->ni_cnd.cn_nameiop = CREATE;
212 * Set NOCACHE to avoid flushing the cache when
213 * rolling in many files at once.
215 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE;
216 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
217 ndp->ni_cnd.cn_flags |= FOLLOW;
218 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
219 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
220 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
221 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
223 if ((error = namei(ndp)) != 0)
225 if (ndp->ni_vp == NULL) {
228 vap->va_mode = cmode;
230 vap->va_vaflags |= VA_EXCLUSIVE;
231 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
232 NDFREE(ndp, NDF_ONLY_PNBUF);
234 if ((error = vn_start_write(NULL, &mp,
235 V_XSLEEP | PCATCH)) != 0)
239 if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0)
240 ndp->ni_cnd.cn_flags |= MAKEENTRY;
242 error = mac_vnode_check_create(cred, ndp->ni_dvp,
246 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
249 vn_finished_write(mp);
251 NDFREE(ndp, NDF_ONLY_PNBUF);
257 if (ndp->ni_dvp == ndp->ni_vp)
263 if (fmode & O_EXCL) {
270 ndp->ni_cnd.cn_nameiop = LOOKUP;
271 ndp->ni_cnd.cn_flags = ISOPEN |
272 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
273 if (!(fmode & FWRITE))
274 ndp->ni_cnd.cn_flags |= LOCKSHARED;
275 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
276 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
277 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
278 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
279 if ((error = namei(ndp)) != 0)
283 error = vn_open_vnode(vp, fmode, cred, td, fp);
289 NDFREE(ndp, NDF_ONLY_PNBUF);
297 * Common code for vnode open operations once a vnode is located.
298 * Check permissions, and call the VOP_OPEN routine.
301 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
302 struct thread *td, struct file *fp)
306 int error, lock_flags, type;
308 if (vp->v_type == VLNK)
310 if (vp->v_type == VSOCK)
312 if (vp->v_type != VDIR && fmode & O_DIRECTORY)
315 if (fmode & (FWRITE | O_TRUNC)) {
316 if (vp->v_type == VDIR)
324 if ((fmode & O_APPEND) && (fmode & FWRITE))
329 if (fmode & O_VERIFY)
331 error = mac_vnode_check_open(cred, vp, accmode);
335 accmode &= ~(VCREAT | VVERIFY);
337 if ((fmode & O_CREAT) == 0) {
338 if (accmode & VWRITE) {
339 error = vn_writechk(vp);
344 error = VOP_ACCESS(vp, accmode, cred, td);
349 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
350 vn_lock(vp, LK_UPGRADE | LK_RETRY);
351 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
354 while ((fmode & (O_EXLOCK | O_SHLOCK)) != 0) {
355 KASSERT(fp != NULL, ("open with flock requires fp"));
356 if (fp->f_type != DTYPE_NONE && fp->f_type != DTYPE_VNODE) {
360 lock_flags = VOP_ISLOCKED(vp);
362 lf.l_whence = SEEK_SET;
365 if (fmode & O_EXLOCK)
370 if ((fmode & FNONBLOCK) == 0)
372 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
374 fp->f_flag |= FHASLOCK;
375 vn_lock(vp, lock_flags | LK_RETRY);
378 if ((vp->v_iflag & VI_DOOMED) != 0) {
384 * Another thread might have used this vnode as an
385 * executable while the vnode lock was dropped.
386 * Ensure the vnode is still able to be opened for
387 * writing after the lock has been obtained.
389 if ((accmode & VWRITE) != 0)
390 error = vn_writechk(vp);
395 fp->f_flag |= FOPENFAILED;
397 if (fp->f_ops == &badfileops) {
398 fp->f_type = DTYPE_VNODE;
402 } else if ((fmode & FWRITE) != 0) {
403 VOP_ADD_WRITECOUNT(vp, 1);
404 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
405 __func__, vp, vp->v_writecount);
407 ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
412 * Check for write permissions on the specified vnode.
413 * Prototype text segments cannot be written.
416 vn_writechk(struct vnode *vp)
419 ASSERT_VOP_LOCKED(vp, "vn_writechk");
421 * If there's shared text associated with
422 * the vnode, try to free it up once. If
423 * we fail, we can't allow writing.
435 vn_close1(struct vnode *vp, int flags, struct ucred *file_cred,
436 struct thread *td, bool keep_ref)
439 int error, lock_flags;
441 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
442 MNT_EXTENDED_SHARED(vp->v_mount))
443 lock_flags = LK_SHARED;
445 lock_flags = LK_EXCLUSIVE;
447 vn_start_write(vp, &mp, V_WAIT);
448 vn_lock(vp, lock_flags | LK_RETRY);
449 AUDIT_ARG_VNODE1(vp);
450 if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) {
451 VNASSERT(vp->v_writecount > 0, vp,
452 ("vn_close: negative writecount"));
453 VOP_ADD_WRITECOUNT(vp, -1);
454 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
455 __func__, vp, vp->v_writecount);
457 error = VOP_CLOSE(vp, flags, file_cred, td);
462 vn_finished_write(mp);
467 vn_close(struct vnode *vp, int flags, struct ucred *file_cred,
471 return (vn_close1(vp, flags, file_cred, td, false));
475 * Heuristic to detect sequential operation.
478 sequential_heuristic(struct uio *uio, struct file *fp)
481 ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
482 if (fp->f_flag & FRDAHEAD)
483 return (fp->f_seqcount << IO_SEQSHIFT);
486 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
487 * that the first I/O is normally considered to be slightly
488 * sequential. Seeking to offset 0 doesn't change sequentiality
489 * unless previous seeks have reduced f_seqcount to 0, in which
490 * case offset 0 is not special.
492 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
493 uio->uio_offset == fp->f_nextoff) {
495 * f_seqcount is in units of fixed-size blocks so that it
496 * depends mainly on the amount of sequential I/O and not
497 * much on the number of sequential I/O's. The fixed size
498 * of 16384 is hard-coded here since it is (not quite) just
499 * a magic size that works well here. This size is more
500 * closely related to the best I/O size for real disks than
501 * to any block size used by software.
503 fp->f_seqcount += howmany(uio->uio_resid, 16384);
504 if (fp->f_seqcount > IO_SEQMAX)
505 fp->f_seqcount = IO_SEQMAX;
506 return (fp->f_seqcount << IO_SEQSHIFT);
509 /* Not sequential. Quickly draw-down sequentiality. */
510 if (fp->f_seqcount > 1)
518 * Package up an I/O request on a vnode into a uio and do it.
521 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
522 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
523 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
530 struct vn_io_fault_args args;
531 int error, lock_flags;
533 auio.uio_iov = &aiov;
535 aiov.iov_base = base;
537 auio.uio_resid = len;
538 auio.uio_offset = offset;
539 auio.uio_segflg = segflg;
544 if ((ioflg & IO_NODELOCKED) == 0) {
545 if ((ioflg & IO_RANGELOCKED) == 0) {
546 if (rw == UIO_READ) {
547 rl_cookie = vn_rangelock_rlock(vp, offset,
550 rl_cookie = vn_rangelock_wlock(vp, offset,
556 if (rw == UIO_WRITE) {
557 if (vp->v_type != VCHR &&
558 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
561 if (MNT_SHARED_WRITES(mp) ||
562 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
563 lock_flags = LK_SHARED;
565 lock_flags = LK_EXCLUSIVE;
567 lock_flags = LK_SHARED;
568 vn_lock(vp, lock_flags | LK_RETRY);
572 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
574 if ((ioflg & IO_NOMACCHECK) == 0) {
576 error = mac_vnode_check_read(active_cred, file_cred,
579 error = mac_vnode_check_write(active_cred, file_cred,
584 if (file_cred != NULL)
588 if (do_vn_io_fault(vp, &auio)) {
589 args.kind = VN_IO_FAULT_VOP;
592 args.args.vop_args.vp = vp;
593 error = vn_io_fault1(vp, &auio, &args, td);
594 } else if (rw == UIO_READ) {
595 error = VOP_READ(vp, &auio, ioflg, cred);
596 } else /* if (rw == UIO_WRITE) */ {
597 error = VOP_WRITE(vp, &auio, ioflg, cred);
601 *aresid = auio.uio_resid;
603 if (auio.uio_resid && error == 0)
605 if ((ioflg & IO_NODELOCKED) == 0) {
608 vn_finished_write(mp);
611 if (rl_cookie != NULL)
612 vn_rangelock_unlock(vp, rl_cookie);
617 * Package up an I/O request on a vnode into a uio and do it. The I/O
618 * request is split up into smaller chunks and we try to avoid saturating
619 * the buffer cache while potentially holding a vnode locked, so we
620 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
621 * to give other processes a chance to lock the vnode (either other processes
622 * core'ing the same binary, or unrelated processes scanning the directory).
625 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
626 file_cred, aresid, td)
634 struct ucred *active_cred;
635 struct ucred *file_cred;
646 * Force `offset' to a multiple of MAXBSIZE except possibly
647 * for the first chunk, so that filesystems only need to
648 * write full blocks except possibly for the first and last
651 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
655 if (rw != UIO_READ && vp->v_type == VREG)
658 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
659 ioflg, active_cred, file_cred, &iaresid, td);
660 len -= chunk; /* aresid calc already includes length */
664 base = (char *)base + chunk;
665 kern_yield(PRI_USER);
668 *aresid = len + iaresid;
673 foffset_lock(struct file *fp, int flags)
678 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
680 #if OFF_MAX <= LONG_MAX
682 * Caller only wants the current f_offset value. Assume that
683 * the long and shorter integer types reads are atomic.
685 if ((flags & FOF_NOLOCK) != 0)
686 return (fp->f_offset);
690 * According to McKusick the vn lock was protecting f_offset here.
691 * It is now protected by the FOFFSET_LOCKED flag.
693 mtxp = mtx_pool_find(mtxpool_sleep, fp);
695 if ((flags & FOF_NOLOCK) == 0) {
696 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
697 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
698 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
701 fp->f_vnread_flags |= FOFFSET_LOCKED;
709 foffset_unlock(struct file *fp, off_t val, int flags)
713 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
715 #if OFF_MAX <= LONG_MAX
716 if ((flags & FOF_NOLOCK) != 0) {
717 if ((flags & FOF_NOUPDATE) == 0)
719 if ((flags & FOF_NEXTOFF) != 0)
725 mtxp = mtx_pool_find(mtxpool_sleep, fp);
727 if ((flags & FOF_NOUPDATE) == 0)
729 if ((flags & FOF_NEXTOFF) != 0)
731 if ((flags & FOF_NOLOCK) == 0) {
732 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
733 ("Lost FOFFSET_LOCKED"));
734 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
735 wakeup(&fp->f_vnread_flags);
736 fp->f_vnread_flags = 0;
742 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
745 if ((flags & FOF_OFFSET) == 0)
746 uio->uio_offset = foffset_lock(fp, flags);
750 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
753 if ((flags & FOF_OFFSET) == 0)
754 foffset_unlock(fp, uio->uio_offset, flags);
758 get_advice(struct file *fp, struct uio *uio)
763 ret = POSIX_FADV_NORMAL;
764 if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG)
767 mtxp = mtx_pool_find(mtxpool_sleep, fp);
769 if (fp->f_advice != NULL &&
770 uio->uio_offset >= fp->f_advice->fa_start &&
771 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
772 ret = fp->f_advice->fa_advice;
778 * File table vnode read routine.
781 vn_read(fp, uio, active_cred, flags, td)
784 struct ucred *active_cred;
793 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
795 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
798 if (fp->f_flag & FNONBLOCK)
800 if (fp->f_flag & O_DIRECT)
802 advice = get_advice(fp, uio);
803 vn_lock(vp, LK_SHARED | LK_RETRY);
806 case POSIX_FADV_NORMAL:
807 case POSIX_FADV_SEQUENTIAL:
808 case POSIX_FADV_NOREUSE:
809 ioflag |= sequential_heuristic(uio, fp);
811 case POSIX_FADV_RANDOM:
812 /* Disable read-ahead for random I/O. */
815 orig_offset = uio->uio_offset;
818 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
821 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
822 fp->f_nextoff = uio->uio_offset;
824 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
825 orig_offset != uio->uio_offset)
827 * Use POSIX_FADV_DONTNEED to flush pages and buffers
828 * for the backing file after a POSIX_FADV_NOREUSE
831 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
832 POSIX_FADV_DONTNEED);
837 * File table vnode write routine.
840 vn_write(fp, uio, active_cred, flags, td)
843 struct ucred *active_cred;
850 int error, ioflag, lock_flags;
853 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
855 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
857 if (vp->v_type == VREG)
860 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
862 if (fp->f_flag & FNONBLOCK)
864 if (fp->f_flag & O_DIRECT)
866 if ((fp->f_flag & O_FSYNC) ||
867 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
870 if (vp->v_type != VCHR &&
871 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
874 advice = get_advice(fp, uio);
876 if (MNT_SHARED_WRITES(mp) ||
877 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
878 lock_flags = LK_SHARED;
880 lock_flags = LK_EXCLUSIVE;
883 vn_lock(vp, lock_flags | LK_RETRY);
885 case POSIX_FADV_NORMAL:
886 case POSIX_FADV_SEQUENTIAL:
887 case POSIX_FADV_NOREUSE:
888 ioflag |= sequential_heuristic(uio, fp);
890 case POSIX_FADV_RANDOM:
891 /* XXX: Is this correct? */
894 orig_offset = uio->uio_offset;
897 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
900 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
901 fp->f_nextoff = uio->uio_offset;
903 if (vp->v_type != VCHR)
904 vn_finished_write(mp);
905 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
906 orig_offset != uio->uio_offset)
908 * Use POSIX_FADV_DONTNEED to flush pages and buffers
909 * for the backing file after a POSIX_FADV_NOREUSE
912 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
913 POSIX_FADV_DONTNEED);
919 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
920 * prevent the following deadlock:
922 * Assume that the thread A reads from the vnode vp1 into userspace
923 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
924 * currently not resident, then system ends up with the call chain
925 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
926 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
927 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
928 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
929 * backed by the pages of vnode vp1, and some page in buf2 is not
930 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
932 * To prevent the lock order reversal and deadlock, vn_io_fault() does
933 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
934 * Instead, it first tries to do the whole range i/o with pagefaults
935 * disabled. If all pages in the i/o buffer are resident and mapped,
936 * VOP will succeed (ignoring the genuine filesystem errors).
937 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
938 * i/o in chunks, with all pages in the chunk prefaulted and held
939 * using vm_fault_quick_hold_pages().
941 * Filesystems using this deadlock avoidance scheme should use the
942 * array of the held pages from uio, saved in the curthread->td_ma,
943 * instead of doing uiomove(). A helper function
944 * vn_io_fault_uiomove() converts uiomove request into
945 * uiomove_fromphys() over td_ma array.
947 * Since vnode locks do not cover the whole i/o anymore, rangelocks
948 * make the current i/o request atomic with respect to other i/os and
953 * Decode vn_io_fault_args and perform the corresponding i/o.
956 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
960 switch (args->kind) {
961 case VN_IO_FAULT_FOP:
962 return ((args->args.fop_args.doio)(args->args.fop_args.fp,
963 uio, args->cred, args->flags, td));
964 case VN_IO_FAULT_VOP:
965 if (uio->uio_rw == UIO_READ) {
966 return (VOP_READ(args->args.vop_args.vp, uio,
967 args->flags, args->cred));
968 } else if (uio->uio_rw == UIO_WRITE) {
969 return (VOP_WRITE(args->args.vop_args.vp, uio,
970 args->flags, args->cred));
974 panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
979 vn_io_fault_touch(char *base, const struct uio *uio)
984 if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1))
990 vn_io_fault_prefault_user(const struct uio *uio)
993 const struct iovec *iov;
998 KASSERT(uio->uio_segflg == UIO_USERSPACE,
999 ("vn_io_fault_prefault userspace"));
1003 resid = uio->uio_resid;
1004 base = iov->iov_base;
1007 error = vn_io_fault_touch(base, uio);
1010 if (len < PAGE_SIZE) {
1012 error = vn_io_fault_touch(base + len - 1, uio);
1017 if (++i >= uio->uio_iovcnt)
1019 iov = uio->uio_iov + i;
1020 base = iov->iov_base;
1032 * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1033 * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1034 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1035 * into args and call vn_io_fault1() to handle faults during the user
1036 * mode buffer accesses.
1039 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1042 vm_page_t ma[io_hold_cnt + 2];
1043 struct uio *uio_clone, short_uio;
1044 struct iovec short_iovec[1];
1045 vm_page_t *prev_td_ma;
1047 vm_offset_t addr, end;
1050 int error, cnt, save, saveheld, prev_td_ma_cnt;
1052 if (vn_io_fault_prefault) {
1053 error = vn_io_fault_prefault_user(uio);
1055 return (error); /* Or ignore ? */
1058 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1061 * The UFS follows IO_UNIT directive and replays back both
1062 * uio_offset and uio_resid if an error is encountered during the
1063 * operation. But, since the iovec may be already advanced,
1064 * uio is still in an inconsistent state.
1066 * Cache a copy of the original uio, which is advanced to the redo
1067 * point using UIO_NOCOPY below.
1069 uio_clone = cloneuio(uio);
1070 resid = uio->uio_resid;
1072 short_uio.uio_segflg = UIO_USERSPACE;
1073 short_uio.uio_rw = uio->uio_rw;
1074 short_uio.uio_td = uio->uio_td;
1076 save = vm_fault_disable_pagefaults();
1077 error = vn_io_fault_doio(args, uio, td);
1078 if (error != EFAULT)
1081 atomic_add_long(&vn_io_faults_cnt, 1);
1082 uio_clone->uio_segflg = UIO_NOCOPY;
1083 uiomove(NULL, resid - uio->uio_resid, uio_clone);
1084 uio_clone->uio_segflg = uio->uio_segflg;
1086 saveheld = curthread_pflags_set(TDP_UIOHELD);
1087 prev_td_ma = td->td_ma;
1088 prev_td_ma_cnt = td->td_ma_cnt;
1090 while (uio_clone->uio_resid != 0) {
1091 len = uio_clone->uio_iov->iov_len;
1093 KASSERT(uio_clone->uio_iovcnt >= 1,
1094 ("iovcnt underflow"));
1095 uio_clone->uio_iov++;
1096 uio_clone->uio_iovcnt--;
1099 if (len > io_hold_cnt * PAGE_SIZE)
1100 len = io_hold_cnt * PAGE_SIZE;
1101 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1102 end = round_page(addr + len);
1107 cnt = atop(end - trunc_page(addr));
1109 * A perfectly misaligned address and length could cause
1110 * both the start and the end of the chunk to use partial
1111 * page. +2 accounts for such a situation.
1113 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1114 addr, len, prot, ma, io_hold_cnt + 2);
1119 short_uio.uio_iov = &short_iovec[0];
1120 short_iovec[0].iov_base = (void *)addr;
1121 short_uio.uio_iovcnt = 1;
1122 short_uio.uio_resid = short_iovec[0].iov_len = len;
1123 short_uio.uio_offset = uio_clone->uio_offset;
1125 td->td_ma_cnt = cnt;
1127 error = vn_io_fault_doio(args, &short_uio, td);
1128 vm_page_unhold_pages(ma, cnt);
1129 adv = len - short_uio.uio_resid;
1131 uio_clone->uio_iov->iov_base =
1132 (char *)uio_clone->uio_iov->iov_base + adv;
1133 uio_clone->uio_iov->iov_len -= adv;
1134 uio_clone->uio_resid -= adv;
1135 uio_clone->uio_offset += adv;
1137 uio->uio_resid -= adv;
1138 uio->uio_offset += adv;
1140 if (error != 0 || adv == 0)
1143 td->td_ma = prev_td_ma;
1144 td->td_ma_cnt = prev_td_ma_cnt;
1145 curthread_pflags_restore(saveheld);
1147 vm_fault_enable_pagefaults(save);
1148 free(uio_clone, M_IOV);
1153 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1154 int flags, struct thread *td)
1159 struct vn_io_fault_args args;
1162 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1164 foffset_lock_uio(fp, uio, flags);
1165 if (do_vn_io_fault(vp, uio)) {
1166 args.kind = VN_IO_FAULT_FOP;
1167 args.args.fop_args.fp = fp;
1168 args.args.fop_args.doio = doio;
1169 args.cred = active_cred;
1170 args.flags = flags | FOF_OFFSET;
1171 if (uio->uio_rw == UIO_READ) {
1172 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1173 uio->uio_offset + uio->uio_resid);
1174 } else if ((fp->f_flag & O_APPEND) != 0 ||
1175 (flags & FOF_OFFSET) == 0) {
1176 /* For appenders, punt and lock the whole range. */
1177 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1179 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1180 uio->uio_offset + uio->uio_resid);
1182 error = vn_io_fault1(vp, uio, &args, td);
1183 vn_rangelock_unlock(vp, rl_cookie);
1185 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1187 foffset_unlock_uio(fp, uio, flags);
1192 * Helper function to perform the requested uiomove operation using
1193 * the held pages for io->uio_iov[0].iov_base buffer instead of
1194 * copyin/copyout. Access to the pages with uiomove_fromphys()
1195 * instead of iov_base prevents page faults that could occur due to
1196 * pmap_collect() invalidating the mapping created by
1197 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1198 * object cleanup revoking the write access from page mappings.
1200 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1201 * instead of plain uiomove().
1204 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1206 struct uio transp_uio;
1207 struct iovec transp_iov[1];
1213 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1214 uio->uio_segflg != UIO_USERSPACE)
1215 return (uiomove(data, xfersize, uio));
1217 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1218 transp_iov[0].iov_base = data;
1219 transp_uio.uio_iov = &transp_iov[0];
1220 transp_uio.uio_iovcnt = 1;
1221 if (xfersize > uio->uio_resid)
1222 xfersize = uio->uio_resid;
1223 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1224 transp_uio.uio_offset = 0;
1225 transp_uio.uio_segflg = UIO_SYSSPACE;
1227 * Since transp_iov points to data, and td_ma page array
1228 * corresponds to original uio->uio_iov, we need to invert the
1229 * direction of the i/o operation as passed to
1230 * uiomove_fromphys().
1232 switch (uio->uio_rw) {
1234 transp_uio.uio_rw = UIO_READ;
1237 transp_uio.uio_rw = UIO_WRITE;
1240 transp_uio.uio_td = uio->uio_td;
1241 error = uiomove_fromphys(td->td_ma,
1242 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1243 xfersize, &transp_uio);
1244 adv = xfersize - transp_uio.uio_resid;
1246 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1247 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1249 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1251 td->td_ma_cnt -= pgadv;
1252 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1253 uio->uio_iov->iov_len -= adv;
1254 uio->uio_resid -= adv;
1255 uio->uio_offset += adv;
1260 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1264 vm_offset_t iov_base;
1268 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1269 uio->uio_segflg != UIO_USERSPACE)
1270 return (uiomove_fromphys(ma, offset, xfersize, uio));
1272 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1273 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1274 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1275 switch (uio->uio_rw) {
1277 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1281 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1285 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1287 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1289 td->td_ma_cnt -= pgadv;
1290 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1291 uio->uio_iov->iov_len -= cnt;
1292 uio->uio_resid -= cnt;
1293 uio->uio_offset += cnt;
1299 * File table truncate routine.
1302 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1314 * Lock the whole range for truncation. Otherwise split i/o
1315 * might happen partly before and partly after the truncation.
1317 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1318 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1321 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1322 AUDIT_ARG_VNODE1(vp);
1323 if (vp->v_type == VDIR) {
1328 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1332 error = vn_writechk(vp);
1335 vattr.va_size = length;
1336 if ((fp->f_flag & O_FSYNC) != 0)
1337 vattr.va_vaflags |= VA_SYNC;
1338 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1342 vn_finished_write(mp);
1344 vn_rangelock_unlock(vp, rl_cookie);
1349 * File table vnode stat routine.
1352 vn_statfile(fp, sb, active_cred, td)
1355 struct ucred *active_cred;
1358 struct vnode *vp = fp->f_vnode;
1361 vn_lock(vp, LK_SHARED | LK_RETRY);
1362 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1369 * Stat a vnode; implementation for the stat syscall
1372 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *active_cred,
1373 struct ucred *file_cred, struct thread *td)
1380 AUDIT_ARG_VNODE1(vp);
1382 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1390 * Initialize defaults for new and unusual fields, so that file
1391 * systems which don't support these fields don't need to know
1394 vap->va_birthtime.tv_sec = -1;
1395 vap->va_birthtime.tv_nsec = 0;
1396 vap->va_fsid = VNOVAL;
1397 vap->va_rdev = NODEV;
1399 error = VOP_GETATTR(vp, vap, active_cred);
1404 * Zero the spare stat fields
1406 bzero(sb, sizeof *sb);
1409 * Copy from vattr table
1411 if (vap->va_fsid != VNOVAL)
1412 sb->st_dev = vap->va_fsid;
1414 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1415 sb->st_ino = vap->va_fileid;
1416 mode = vap->va_mode;
1417 switch (vap->va_type) {
1443 sb->st_nlink = vap->va_nlink;
1444 sb->st_uid = vap->va_uid;
1445 sb->st_gid = vap->va_gid;
1446 sb->st_rdev = vap->va_rdev;
1447 if (vap->va_size > OFF_MAX)
1449 sb->st_size = vap->va_size;
1450 sb->st_atim = vap->va_atime;
1451 sb->st_mtim = vap->va_mtime;
1452 sb->st_ctim = vap->va_ctime;
1453 sb->st_birthtim = vap->va_birthtime;
1456 * According to www.opengroup.org, the meaning of st_blksize is
1457 * "a filesystem-specific preferred I/O block size for this
1458 * object. In some filesystem types, this may vary from file
1460 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1463 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1465 sb->st_flags = vap->va_flags;
1466 if (priv_check(td, PRIV_VFS_GENERATION))
1469 sb->st_gen = vap->va_gen;
1471 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1476 * File table vnode ioctl routine.
1479 vn_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
1487 switch (vp->v_type) {
1492 vn_lock(vp, LK_SHARED | LK_RETRY);
1493 error = VOP_GETATTR(vp, &vattr, active_cred);
1496 *(int *)data = vattr.va_size - fp->f_offset;
1502 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1507 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1515 * File table vnode poll routine.
1518 vn_poll(struct file *fp, int events, struct ucred *active_cred,
1526 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1527 AUDIT_ARG_VNODE1(vp);
1528 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1533 error = VOP_POLL(vp, events, fp->f_cred, td);
1538 * Acquire the requested lock and then check for validity. LK_RETRY
1539 * permits vn_lock to return doomed vnodes.
1542 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1546 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1547 ("vn_lock: no locktype"));
1548 VNASSERT(vp->v_holdcnt != 0, vp, ("vn_lock: zero hold count"));
1550 error = VOP_LOCK1(vp, flags, file, line);
1551 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1552 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1553 ("vn_lock: error %d incompatible with flags %#x", error, flags));
1555 if ((flags & LK_RETRY) == 0) {
1556 if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) {
1560 } else if (error != 0)
1566 * File table vnode close routine.
1569 vn_closefile(struct file *fp, struct thread *td)
1577 fp->f_ops = &badfileops;
1578 ref= (fp->f_flag & FHASLOCK) != 0 && fp->f_type == DTYPE_VNODE;
1580 error = vn_close1(vp, fp->f_flag, fp->f_cred, td, ref);
1582 if (__predict_false(ref)) {
1583 lf.l_whence = SEEK_SET;
1586 lf.l_type = F_UNLCK;
1587 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1594 vn_suspendable(struct mount *mp)
1597 return (mp->mnt_op->vfs_susp_clean != NULL);
1601 * Preparing to start a filesystem write operation. If the operation is
1602 * permitted, then we bump the count of operations in progress and
1603 * proceed. If a suspend request is in progress, we wait until the
1604 * suspension is over, and then proceed.
1607 vn_start_write_locked(struct mount *mp, int flags)
1611 mtx_assert(MNT_MTX(mp), MA_OWNED);
1615 * Check on status of suspension.
1617 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1618 mp->mnt_susp_owner != curthread) {
1619 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
1620 (flags & PCATCH) : 0) | (PUSER - 1);
1621 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1622 if (flags & V_NOWAIT) {
1623 error = EWOULDBLOCK;
1626 error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
1632 if (flags & V_XSLEEP)
1634 mp->mnt_writeopcount++;
1636 if (error != 0 || (flags & V_XSLEEP) != 0)
1643 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
1648 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1649 ("V_MNTREF requires mp"));
1653 * If a vnode is provided, get and return the mount point that
1654 * to which it will write.
1657 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1659 if (error != EOPNOTSUPP)
1664 if ((mp = *mpp) == NULL)
1667 if (!vn_suspendable(mp)) {
1668 if (vp != NULL || (flags & V_MNTREF) != 0)
1674 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1676 * As long as a vnode is not provided we need to acquire a
1677 * refcount for the provided mountpoint too, in order to
1678 * emulate a vfs_ref().
1681 if (vp == NULL && (flags & V_MNTREF) == 0)
1684 return (vn_start_write_locked(mp, flags));
1688 * Secondary suspension. Used by operations such as vop_inactive
1689 * routines that are needed by the higher level functions. These
1690 * are allowed to proceed until all the higher level functions have
1691 * completed (indicated by mnt_writeopcount dropping to zero). At that
1692 * time, these operations are halted until the suspension is over.
1695 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags)
1700 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1701 ("V_MNTREF requires mp"));
1705 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1707 if (error != EOPNOTSUPP)
1713 * If we are not suspended or have not yet reached suspended
1714 * mode, then let the operation proceed.
1716 if ((mp = *mpp) == NULL)
1719 if (!vn_suspendable(mp)) {
1720 if (vp != NULL || (flags & V_MNTREF) != 0)
1726 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1728 * As long as a vnode is not provided we need to acquire a
1729 * refcount for the provided mountpoint too, in order to
1730 * emulate a vfs_ref().
1733 if (vp == NULL && (flags & V_MNTREF) == 0)
1735 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1736 mp->mnt_secondary_writes++;
1737 mp->mnt_secondary_accwrites++;
1741 if (flags & V_NOWAIT) {
1744 return (EWOULDBLOCK);
1747 * Wait for the suspension to finish.
1749 error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
1750 ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
1759 * Filesystem write operation has completed. If we are suspending and this
1760 * operation is the last one, notify the suspender that the suspension is
1764 vn_finished_write(struct mount *mp)
1766 if (mp == NULL || !vn_suspendable(mp))
1770 mp->mnt_writeopcount--;
1771 if (mp->mnt_writeopcount < 0)
1772 panic("vn_finished_write: neg cnt");
1773 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1774 mp->mnt_writeopcount <= 0)
1775 wakeup(&mp->mnt_writeopcount);
1781 * Filesystem secondary write operation has completed. If we are
1782 * suspending and this operation is the last one, notify the suspender
1783 * that the suspension is now in effect.
1786 vn_finished_secondary_write(struct mount *mp)
1788 if (mp == NULL || !vn_suspendable(mp))
1792 mp->mnt_secondary_writes--;
1793 if (mp->mnt_secondary_writes < 0)
1794 panic("vn_finished_secondary_write: neg cnt");
1795 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1796 mp->mnt_secondary_writes <= 0)
1797 wakeup(&mp->mnt_secondary_writes);
1804 * Request a filesystem to suspend write operations.
1807 vfs_write_suspend(struct mount *mp, int flags)
1811 MPASS(vn_suspendable(mp));
1814 if (mp->mnt_susp_owner == curthread) {
1818 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1819 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1822 * Unmount holds a write reference on the mount point. If we
1823 * own busy reference and drain for writers, we deadlock with
1824 * the reference draining in the unmount path. Callers of
1825 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1826 * vfs_busy() reference is owned and caller is not in the
1829 if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1830 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1835 mp->mnt_kern_flag |= MNTK_SUSPEND;
1836 mp->mnt_susp_owner = curthread;
1837 if (mp->mnt_writeopcount > 0)
1838 (void) msleep(&mp->mnt_writeopcount,
1839 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1842 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1843 vfs_write_resume(mp, 0);
1848 * Request a filesystem to resume write operations.
1851 vfs_write_resume(struct mount *mp, int flags)
1854 MPASS(vn_suspendable(mp));
1857 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1858 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1859 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1861 mp->mnt_susp_owner = NULL;
1862 wakeup(&mp->mnt_writeopcount);
1863 wakeup(&mp->mnt_flag);
1864 curthread->td_pflags &= ~TDP_IGNSUSP;
1865 if ((flags & VR_START_WRITE) != 0) {
1867 mp->mnt_writeopcount++;
1870 if ((flags & VR_NO_SUSPCLR) == 0)
1872 } else if ((flags & VR_START_WRITE) != 0) {
1874 vn_start_write_locked(mp, 0);
1881 * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1885 vfs_write_suspend_umnt(struct mount *mp)
1889 MPASS(vn_suspendable(mp));
1890 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1891 ("vfs_write_suspend_umnt: recursed"));
1893 /* dounmount() already called vn_start_write(). */
1895 vn_finished_write(mp);
1896 error = vfs_write_suspend(mp, 0);
1898 vn_start_write(NULL, &mp, V_WAIT);
1902 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1905 vn_start_write(NULL, &mp, V_WAIT);
1907 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1908 wakeup(&mp->mnt_flag);
1910 curthread->td_pflags |= TDP_IGNSUSP;
1915 * Implement kqueues for files by translating it to vnode operation.
1918 vn_kqfilter(struct file *fp, struct knote *kn)
1921 return (VOP_KQFILTER(fp->f_vnode, kn));
1925 * Simplified in-kernel wrapper calls for extended attribute access.
1926 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1927 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1930 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1931 const char *attrname, int *buflen, char *buf, struct thread *td)
1937 iov.iov_len = *buflen;
1940 auio.uio_iov = &iov;
1941 auio.uio_iovcnt = 1;
1942 auio.uio_rw = UIO_READ;
1943 auio.uio_segflg = UIO_SYSSPACE;
1945 auio.uio_offset = 0;
1946 auio.uio_resid = *buflen;
1948 if ((ioflg & IO_NODELOCKED) == 0)
1949 vn_lock(vp, LK_SHARED | LK_RETRY);
1951 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1953 /* authorize attribute retrieval as kernel */
1954 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1957 if ((ioflg & IO_NODELOCKED) == 0)
1961 *buflen = *buflen - auio.uio_resid;
1968 * XXX failure mode if partially written?
1971 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1972 const char *attrname, int buflen, char *buf, struct thread *td)
1979 iov.iov_len = buflen;
1982 auio.uio_iov = &iov;
1983 auio.uio_iovcnt = 1;
1984 auio.uio_rw = UIO_WRITE;
1985 auio.uio_segflg = UIO_SYSSPACE;
1987 auio.uio_offset = 0;
1988 auio.uio_resid = buflen;
1990 if ((ioflg & IO_NODELOCKED) == 0) {
1991 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1993 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1996 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1998 /* authorize attribute setting as kernel */
1999 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
2001 if ((ioflg & IO_NODELOCKED) == 0) {
2002 vn_finished_write(mp);
2010 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
2011 const char *attrname, struct thread *td)
2016 if ((ioflg & IO_NODELOCKED) == 0) {
2017 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2019 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2022 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2024 /* authorize attribute removal as kernel */
2025 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2026 if (error == EOPNOTSUPP)
2027 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2030 if ((ioflg & IO_NODELOCKED) == 0) {
2031 vn_finished_write(mp);
2039 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2043 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2047 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2050 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2055 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2056 int lkflags, struct vnode **rvp)
2061 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2063 ltype = VOP_ISLOCKED(vp);
2064 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2065 ("vn_vget_ino: vp not locked"));
2066 error = vfs_busy(mp, MBF_NOWAIT);
2070 error = vfs_busy(mp, 0);
2071 vn_lock(vp, ltype | LK_RETRY);
2075 if (vp->v_iflag & VI_DOOMED) {
2081 error = alloc(mp, alloc_arg, lkflags, rvp);
2084 vn_lock(vp, ltype | LK_RETRY);
2085 if (vp->v_iflag & VI_DOOMED) {
2098 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2102 if (vp->v_type != VREG || td == NULL)
2104 if ((uoff_t)uio->uio_offset + uio->uio_resid >
2105 lim_cur(td, RLIMIT_FSIZE)) {
2106 PROC_LOCK(td->td_proc);
2107 kern_psignal(td->td_proc, SIGXFSZ);
2108 PROC_UNLOCK(td->td_proc);
2115 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2122 vn_lock(vp, LK_SHARED | LK_RETRY);
2123 AUDIT_ARG_VNODE1(vp);
2126 return (setfmode(td, active_cred, vp, mode));
2130 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2137 vn_lock(vp, LK_SHARED | LK_RETRY);
2138 AUDIT_ARG_VNODE1(vp);
2141 return (setfown(td, active_cred, vp, uid, gid));
2145 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2149 if ((object = vp->v_object) == NULL)
2151 VM_OBJECT_WLOCK(object);
2152 vm_object_page_remove(object, start, end, 0);
2153 VM_OBJECT_WUNLOCK(object);
2157 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2165 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2166 ("Wrong command %lu", cmd));
2168 if (vn_lock(vp, LK_SHARED) != 0)
2170 if (vp->v_type != VREG) {
2174 error = VOP_GETATTR(vp, &va, cred);
2178 if (noff >= va.va_size) {
2182 bsize = vp->v_mount->mnt_stat.f_iosize;
2183 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2184 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2185 if (error == EOPNOTSUPP) {
2189 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2190 (bnp != -1 && cmd == FIOSEEKDATA)) {
2197 if (noff > va.va_size)
2199 /* noff == va.va_size. There is an implicit hole at the end of file. */
2200 if (cmd == FIOSEEKDATA)
2210 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2215 off_t foffset, size;
2218 cred = td->td_ucred;
2220 foffset = foffset_lock(fp, 0);
2221 noneg = (vp->v_type != VCHR);
2227 (offset > 0 && foffset > OFF_MAX - offset))) {
2234 vn_lock(vp, LK_SHARED | LK_RETRY);
2235 error = VOP_GETATTR(vp, &vattr, cred);
2241 * If the file references a disk device, then fetch
2242 * the media size and use that to determine the ending
2245 if (vattr.va_size == 0 && vp->v_type == VCHR &&
2246 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2247 vattr.va_size = size;
2249 (vattr.va_size > OFF_MAX ||
2250 (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2254 offset += vattr.va_size;
2259 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2262 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2267 if (error == 0 && noneg && offset < 0)
2271 VFS_KNOTE_UNLOCKED(vp, 0);
2272 td->td_uretoff.tdu_off = offset;
2274 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2279 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2285 * Grant permission if the caller is the owner of the file, or
2286 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2287 * on the file. If the time pointer is null, then write
2288 * permission on the file is also sufficient.
2290 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2291 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2292 * will be allowed to set the times [..] to the current
2295 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2296 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2297 error = VOP_ACCESS(vp, VWRITE, cred, td);
2302 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2307 if (fp->f_type == DTYPE_FIFO)
2308 kif->kf_type = KF_TYPE_FIFO;
2310 kif->kf_type = KF_TYPE_VNODE;
2313 FILEDESC_SUNLOCK(fdp);
2314 error = vn_fill_kinfo_vnode(vp, kif);
2316 FILEDESC_SLOCK(fdp);
2321 vn_fill_junk(struct kinfo_file *kif)
2326 * Simulate vn_fullpath returning changing values for a given
2327 * vp during e.g. coredump.
2329 len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1;
2330 olen = strlen(kif->kf_path);
2332 strcpy(&kif->kf_path[len - 1], "$");
2334 for (; olen < len; olen++)
2335 strcpy(&kif->kf_path[olen], "A");
2339 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2342 char *fullpath, *freepath;
2345 kif->kf_un.kf_file.kf_file_type = vntype_to_kinfo(vp->v_type);
2348 error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2350 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2352 if (freepath != NULL)
2353 free(freepath, M_TEMP);
2355 KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path,
2360 * Retrieve vnode attributes.
2362 va.va_fsid = VNOVAL;
2364 vn_lock(vp, LK_SHARED | LK_RETRY);
2365 error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2369 if (va.va_fsid != VNOVAL)
2370 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2372 kif->kf_un.kf_file.kf_file_fsid =
2373 vp->v_mount->mnt_stat.f_fsid.val[0];
2374 kif->kf_un.kf_file.kf_file_fsid_freebsd11 =
2375 kif->kf_un.kf_file.kf_file_fsid; /* truncate */
2376 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2377 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2378 kif->kf_un.kf_file.kf_file_size = va.va_size;
2379 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2380 kif->kf_un.kf_file.kf_file_rdev_freebsd11 =
2381 kif->kf_un.kf_file.kf_file_rdev; /* truncate */
2386 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
2387 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
2391 struct pmckern_map_in pkm;
2397 boolean_t writecounted;
2400 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
2401 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
2403 * POSIX shared-memory objects are defined to have
2404 * kernel persistence, and are not defined to support
2405 * read(2)/write(2) -- or even open(2). Thus, we can
2406 * use MAP_ASYNC to trade on-disk coherence for speed.
2407 * The shm_open(3) library routine turns on the FPOSIXSHM
2408 * flag to request this behavior.
2410 if ((fp->f_flag & FPOSIXSHM) != 0)
2411 flags |= MAP_NOSYNC;
2416 * Ensure that file and memory protections are
2417 * compatible. Note that we only worry about
2418 * writability if mapping is shared; in this case,
2419 * current and max prot are dictated by the open file.
2420 * XXX use the vnode instead? Problem is: what
2421 * credentials do we use for determination? What if
2422 * proc does a setuid?
2425 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
2426 maxprot = VM_PROT_NONE;
2427 if ((prot & VM_PROT_EXECUTE) != 0)
2430 maxprot = VM_PROT_EXECUTE;
2431 if ((fp->f_flag & FREAD) != 0)
2432 maxprot |= VM_PROT_READ;
2433 else if ((prot & VM_PROT_READ) != 0)
2437 * If we are sharing potential changes via MAP_SHARED and we
2438 * are trying to get write permission although we opened it
2439 * without asking for it, bail out.
2441 if ((flags & MAP_SHARED) != 0) {
2442 if ((fp->f_flag & FWRITE) != 0)
2443 maxprot |= VM_PROT_WRITE;
2444 else if ((prot & VM_PROT_WRITE) != 0)
2447 maxprot |= VM_PROT_WRITE;
2448 cap_maxprot |= VM_PROT_WRITE;
2450 maxprot &= cap_maxprot;
2453 * For regular files and shared memory, POSIX requires that
2454 * the value of foff be a legitimate offset within the data
2455 * object. In particular, negative offsets are invalid.
2456 * Blocking negative offsets and overflows here avoids
2457 * possible wraparound or user-level access into reserved
2458 * ranges of the data object later. In contrast, POSIX does
2459 * not dictate how offsets are used by device drivers, so in
2460 * the case of a device mapping a negative offset is passed
2467 foff < 0 || foff > OFF_MAX - size)
2470 writecounted = FALSE;
2471 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp,
2472 &foff, &object, &writecounted);
2475 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2476 foff, writecounted, td);
2479 * If this mapping was accounted for in the vnode's
2480 * writecount, then undo that now.
2483 vnode_pager_release_writecount(object, 0, size);
2484 vm_object_deallocate(object);
2487 /* Inform hwpmc(4) if an executable is being mapped. */
2488 if (PMC_HOOK_INSTALLED(PMC_FN_MMAP)) {
2489 if ((prot & VM_PROT_EXECUTE) != 0 && error == 0) {
2491 pkm.pm_address = (uintptr_t) *addr;
2492 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
2500 vn_fsid(struct vnode *vp, struct vattr *va)
2504 f = &vp->v_mount->mnt_stat.f_fsid;
2505 va->va_fsid = (uint32_t)f->val[1];
2506 va->va_fsid <<= sizeof(f->val[1]) * NBBY;
2507 va->va_fsid += (uint32_t)f->val[0];