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(struct nameidata *ndp, int *flagp, int cmode, struct file *fp)
177 struct thread *td = ndp->ni_cnd.cn_thread;
179 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
183 * Common code for vnode open operations via a name lookup.
184 * Lookup the vnode and invoke VOP_CREATE if needed.
185 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
187 * Note that this does NOT free nameidata for the successful case,
188 * due to the NDINIT being done elsewhere.
191 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
192 struct ucred *cred, struct file *fp)
196 struct thread *td = ndp->ni_cnd.cn_thread;
198 struct vattr *vap = &vat;
203 if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT |
204 O_EXCL | O_DIRECTORY))
206 else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) {
207 ndp->ni_cnd.cn_nameiop = CREATE;
209 * Set NOCACHE to avoid flushing the cache when
210 * rolling in many files at once.
212 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE;
213 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
214 ndp->ni_cnd.cn_flags |= FOLLOW;
215 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
216 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
217 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
218 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
220 if ((error = namei(ndp)) != 0)
222 if (ndp->ni_vp == NULL) {
225 vap->va_mode = cmode;
227 vap->va_vaflags |= VA_EXCLUSIVE;
228 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
229 NDFREE(ndp, NDF_ONLY_PNBUF);
231 if ((error = vn_start_write(NULL, &mp,
232 V_XSLEEP | PCATCH)) != 0)
236 if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0)
237 ndp->ni_cnd.cn_flags |= MAKEENTRY;
239 error = mac_vnode_check_create(cred, ndp->ni_dvp,
243 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
246 vn_finished_write(mp);
248 NDFREE(ndp, NDF_ONLY_PNBUF);
254 if (ndp->ni_dvp == ndp->ni_vp)
260 if (fmode & O_EXCL) {
267 ndp->ni_cnd.cn_nameiop = LOOKUP;
268 ndp->ni_cnd.cn_flags = ISOPEN |
269 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
270 if (!(fmode & FWRITE))
271 ndp->ni_cnd.cn_flags |= LOCKSHARED;
272 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
273 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
274 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
275 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
276 if ((error = namei(ndp)) != 0)
280 error = vn_open_vnode(vp, fmode, cred, td, fp);
286 NDFREE(ndp, NDF_ONLY_PNBUF);
294 * Common code for vnode open operations once a vnode is located.
295 * Check permissions, and call the VOP_OPEN routine.
298 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
299 struct thread *td, struct file *fp)
303 int error, lock_flags, type;
305 if (vp->v_type == VLNK)
307 if (vp->v_type == VSOCK)
309 if (vp->v_type != VDIR && fmode & O_DIRECTORY)
312 if (fmode & (FWRITE | O_TRUNC)) {
313 if (vp->v_type == VDIR)
321 if ((fmode & O_APPEND) && (fmode & FWRITE))
326 if (fmode & O_VERIFY)
328 error = mac_vnode_check_open(cred, vp, accmode);
332 accmode &= ~(VCREAT | VVERIFY);
334 if ((fmode & O_CREAT) == 0) {
335 if (accmode & VWRITE) {
336 error = vn_writechk(vp);
341 error = VOP_ACCESS(vp, accmode, cred, td);
346 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
347 vn_lock(vp, LK_UPGRADE | LK_RETRY);
348 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
351 while ((fmode & (O_EXLOCK | O_SHLOCK)) != 0) {
352 KASSERT(fp != NULL, ("open with flock requires fp"));
353 if (fp->f_type != DTYPE_NONE && fp->f_type != DTYPE_VNODE) {
357 lock_flags = VOP_ISLOCKED(vp);
359 lf.l_whence = SEEK_SET;
362 if (fmode & O_EXLOCK)
367 if ((fmode & FNONBLOCK) == 0)
369 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
371 fp->f_flag |= FHASLOCK;
372 vn_lock(vp, lock_flags | LK_RETRY);
375 if ((vp->v_iflag & VI_DOOMED) != 0) {
381 * Another thread might have used this vnode as an
382 * executable while the vnode lock was dropped.
383 * Ensure the vnode is still able to be opened for
384 * writing after the lock has been obtained.
386 if ((accmode & VWRITE) != 0)
387 error = vn_writechk(vp);
392 fp->f_flag |= FOPENFAILED;
394 if (fp->f_ops == &badfileops) {
395 fp->f_type = DTYPE_VNODE;
399 } else if ((fmode & FWRITE) != 0) {
400 VOP_ADD_WRITECOUNT(vp, 1);
401 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
402 __func__, vp, vp->v_writecount);
404 ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
409 * Check for write permissions on the specified vnode.
410 * Prototype text segments cannot be written.
413 vn_writechk(struct vnode *vp)
416 ASSERT_VOP_LOCKED(vp, "vn_writechk");
418 * If there's shared text associated with
419 * the vnode, try to free it up once. If
420 * we fail, we can't allow writing.
432 vn_close1(struct vnode *vp, int flags, struct ucred *file_cred,
433 struct thread *td, bool keep_ref)
436 int error, lock_flags;
438 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
439 MNT_EXTENDED_SHARED(vp->v_mount))
440 lock_flags = LK_SHARED;
442 lock_flags = LK_EXCLUSIVE;
444 vn_start_write(vp, &mp, V_WAIT);
445 vn_lock(vp, lock_flags | LK_RETRY);
446 AUDIT_ARG_VNODE1(vp);
447 if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) {
448 VNASSERT(vp->v_writecount > 0, vp,
449 ("vn_close: negative writecount"));
450 VOP_ADD_WRITECOUNT(vp, -1);
451 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
452 __func__, vp, vp->v_writecount);
454 error = VOP_CLOSE(vp, flags, file_cred, td);
459 vn_finished_write(mp);
464 vn_close(struct vnode *vp, int flags, struct ucred *file_cred,
468 return (vn_close1(vp, flags, file_cred, td, false));
472 * Heuristic to detect sequential operation.
475 sequential_heuristic(struct uio *uio, struct file *fp)
478 ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
479 if (fp->f_flag & FRDAHEAD)
480 return (fp->f_seqcount << IO_SEQSHIFT);
483 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
484 * that the first I/O is normally considered to be slightly
485 * sequential. Seeking to offset 0 doesn't change sequentiality
486 * unless previous seeks have reduced f_seqcount to 0, in which
487 * case offset 0 is not special.
489 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
490 uio->uio_offset == fp->f_nextoff) {
492 * f_seqcount is in units of fixed-size blocks so that it
493 * depends mainly on the amount of sequential I/O and not
494 * much on the number of sequential I/O's. The fixed size
495 * of 16384 is hard-coded here since it is (not quite) just
496 * a magic size that works well here. This size is more
497 * closely related to the best I/O size for real disks than
498 * to any block size used by software.
500 fp->f_seqcount += howmany(uio->uio_resid, 16384);
501 if (fp->f_seqcount > IO_SEQMAX)
502 fp->f_seqcount = IO_SEQMAX;
503 return (fp->f_seqcount << IO_SEQSHIFT);
506 /* Not sequential. Quickly draw-down sequentiality. */
507 if (fp->f_seqcount > 1)
515 * Package up an I/O request on a vnode into a uio and do it.
518 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
519 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
520 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
527 struct vn_io_fault_args args;
528 int error, lock_flags;
530 if (offset < 0 && vp->v_type != VCHR)
532 auio.uio_iov = &aiov;
534 aiov.iov_base = base;
536 auio.uio_resid = len;
537 auio.uio_offset = offset;
538 auio.uio_segflg = segflg;
543 if ((ioflg & IO_NODELOCKED) == 0) {
544 if ((ioflg & IO_RANGELOCKED) == 0) {
545 if (rw == UIO_READ) {
546 rl_cookie = vn_rangelock_rlock(vp, offset,
549 rl_cookie = vn_rangelock_wlock(vp, offset,
555 if (rw == UIO_WRITE) {
556 if (vp->v_type != VCHR &&
557 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
560 if (MNT_SHARED_WRITES(mp) ||
561 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
562 lock_flags = LK_SHARED;
564 lock_flags = LK_EXCLUSIVE;
566 lock_flags = LK_SHARED;
567 vn_lock(vp, lock_flags | LK_RETRY);
571 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
573 if ((ioflg & IO_NOMACCHECK) == 0) {
575 error = mac_vnode_check_read(active_cred, file_cred,
578 error = mac_vnode_check_write(active_cred, file_cred,
583 if (file_cred != NULL)
587 if (do_vn_io_fault(vp, &auio)) {
588 args.kind = VN_IO_FAULT_VOP;
591 args.args.vop_args.vp = vp;
592 error = vn_io_fault1(vp, &auio, &args, td);
593 } else if (rw == UIO_READ) {
594 error = VOP_READ(vp, &auio, ioflg, cred);
595 } else /* if (rw == UIO_WRITE) */ {
596 error = VOP_WRITE(vp, &auio, ioflg, cred);
600 *aresid = auio.uio_resid;
602 if (auio.uio_resid && error == 0)
604 if ((ioflg & IO_NODELOCKED) == 0) {
607 vn_finished_write(mp);
610 if (rl_cookie != NULL)
611 vn_rangelock_unlock(vp, rl_cookie);
616 * Package up an I/O request on a vnode into a uio and do it. The I/O
617 * request is split up into smaller chunks and we try to avoid saturating
618 * the buffer cache while potentially holding a vnode locked, so we
619 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
620 * to give other processes a chance to lock the vnode (either other processes
621 * core'ing the same binary, or unrelated processes scanning the directory).
624 vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, void *base, size_t len,
625 off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred,
626 struct ucred *file_cred, size_t *aresid, struct thread *td)
635 * Force `offset' to a multiple of MAXBSIZE except possibly
636 * for the first chunk, so that filesystems only need to
637 * write full blocks except possibly for the first and last
640 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
644 if (rw != UIO_READ && vp->v_type == VREG)
647 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
648 ioflg, active_cred, file_cred, &iaresid, td);
649 len -= chunk; /* aresid calc already includes length */
653 base = (char *)base + chunk;
654 kern_yield(PRI_USER);
657 *aresid = len + iaresid;
662 foffset_lock(struct file *fp, int flags)
667 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
669 #if OFF_MAX <= LONG_MAX
671 * Caller only wants the current f_offset value. Assume that
672 * the long and shorter integer types reads are atomic.
674 if ((flags & FOF_NOLOCK) != 0)
675 return (fp->f_offset);
679 * According to McKusick the vn lock was protecting f_offset here.
680 * It is now protected by the FOFFSET_LOCKED flag.
682 mtxp = mtx_pool_find(mtxpool_sleep, fp);
684 if ((flags & FOF_NOLOCK) == 0) {
685 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
686 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
687 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
690 fp->f_vnread_flags |= FOFFSET_LOCKED;
698 foffset_unlock(struct file *fp, off_t val, int flags)
702 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
704 #if OFF_MAX <= LONG_MAX
705 if ((flags & FOF_NOLOCK) != 0) {
706 if ((flags & FOF_NOUPDATE) == 0)
708 if ((flags & FOF_NEXTOFF) != 0)
714 mtxp = mtx_pool_find(mtxpool_sleep, fp);
716 if ((flags & FOF_NOUPDATE) == 0)
718 if ((flags & FOF_NEXTOFF) != 0)
720 if ((flags & FOF_NOLOCK) == 0) {
721 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
722 ("Lost FOFFSET_LOCKED"));
723 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
724 wakeup(&fp->f_vnread_flags);
725 fp->f_vnread_flags = 0;
731 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
734 if ((flags & FOF_OFFSET) == 0)
735 uio->uio_offset = foffset_lock(fp, flags);
739 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
742 if ((flags & FOF_OFFSET) == 0)
743 foffset_unlock(fp, uio->uio_offset, flags);
747 get_advice(struct file *fp, struct uio *uio)
752 ret = POSIX_FADV_NORMAL;
753 if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG)
756 mtxp = mtx_pool_find(mtxpool_sleep, fp);
758 if (fp->f_advice != NULL &&
759 uio->uio_offset >= fp->f_advice->fa_start &&
760 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
761 ret = fp->f_advice->fa_advice;
767 * File table vnode read routine.
770 vn_read(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags,
778 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
780 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
783 if (fp->f_flag & FNONBLOCK)
785 if (fp->f_flag & O_DIRECT)
787 advice = get_advice(fp, uio);
788 vn_lock(vp, LK_SHARED | LK_RETRY);
791 case POSIX_FADV_NORMAL:
792 case POSIX_FADV_SEQUENTIAL:
793 case POSIX_FADV_NOREUSE:
794 ioflag |= sequential_heuristic(uio, fp);
796 case POSIX_FADV_RANDOM:
797 /* Disable read-ahead for random I/O. */
800 orig_offset = uio->uio_offset;
803 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
806 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
807 fp->f_nextoff = uio->uio_offset;
809 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
810 orig_offset != uio->uio_offset)
812 * Use POSIX_FADV_DONTNEED to flush pages and buffers
813 * for the backing file after a POSIX_FADV_NOREUSE
816 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
817 POSIX_FADV_DONTNEED);
822 * File table vnode write routine.
825 vn_write(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags,
831 int error, ioflag, lock_flags;
834 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
836 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
838 if (vp->v_type == VREG)
841 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
843 if (fp->f_flag & FNONBLOCK)
845 if (fp->f_flag & O_DIRECT)
847 if ((fp->f_flag & O_FSYNC) ||
848 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
851 if (vp->v_type != VCHR &&
852 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
855 advice = get_advice(fp, uio);
857 if (MNT_SHARED_WRITES(mp) ||
858 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
859 lock_flags = LK_SHARED;
861 lock_flags = LK_EXCLUSIVE;
864 vn_lock(vp, lock_flags | LK_RETRY);
866 case POSIX_FADV_NORMAL:
867 case POSIX_FADV_SEQUENTIAL:
868 case POSIX_FADV_NOREUSE:
869 ioflag |= sequential_heuristic(uio, fp);
871 case POSIX_FADV_RANDOM:
872 /* XXX: Is this correct? */
875 orig_offset = uio->uio_offset;
878 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
881 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
882 fp->f_nextoff = uio->uio_offset;
884 if (vp->v_type != VCHR)
885 vn_finished_write(mp);
886 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
887 orig_offset != uio->uio_offset)
889 * Use POSIX_FADV_DONTNEED to flush pages and buffers
890 * for the backing file after a POSIX_FADV_NOREUSE
893 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
894 POSIX_FADV_DONTNEED);
900 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
901 * prevent the following deadlock:
903 * Assume that the thread A reads from the vnode vp1 into userspace
904 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
905 * currently not resident, then system ends up with the call chain
906 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
907 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
908 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
909 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
910 * backed by the pages of vnode vp1, and some page in buf2 is not
911 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
913 * To prevent the lock order reversal and deadlock, vn_io_fault() does
914 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
915 * Instead, it first tries to do the whole range i/o with pagefaults
916 * disabled. If all pages in the i/o buffer are resident and mapped,
917 * VOP will succeed (ignoring the genuine filesystem errors).
918 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
919 * i/o in chunks, with all pages in the chunk prefaulted and held
920 * using vm_fault_quick_hold_pages().
922 * Filesystems using this deadlock avoidance scheme should use the
923 * array of the held pages from uio, saved in the curthread->td_ma,
924 * instead of doing uiomove(). A helper function
925 * vn_io_fault_uiomove() converts uiomove request into
926 * uiomove_fromphys() over td_ma array.
928 * Since vnode locks do not cover the whole i/o anymore, rangelocks
929 * make the current i/o request atomic with respect to other i/os and
934 * Decode vn_io_fault_args and perform the corresponding i/o.
937 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
943 save = vm_fault_disable_pagefaults();
944 switch (args->kind) {
945 case VN_IO_FAULT_FOP:
946 error = (args->args.fop_args.doio)(args->args.fop_args.fp,
947 uio, args->cred, args->flags, td);
949 case VN_IO_FAULT_VOP:
950 if (uio->uio_rw == UIO_READ) {
951 error = VOP_READ(args->args.vop_args.vp, uio,
952 args->flags, args->cred);
953 } else if (uio->uio_rw == UIO_WRITE) {
954 error = VOP_WRITE(args->args.vop_args.vp, uio,
955 args->flags, args->cred);
959 panic("vn_io_fault_doio: unknown kind of io %d %d",
960 args->kind, uio->uio_rw);
962 vm_fault_enable_pagefaults(save);
967 vn_io_fault_touch(char *base, const struct uio *uio)
972 if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1))
978 vn_io_fault_prefault_user(const struct uio *uio)
981 const struct iovec *iov;
986 KASSERT(uio->uio_segflg == UIO_USERSPACE,
987 ("vn_io_fault_prefault userspace"));
991 resid = uio->uio_resid;
992 base = iov->iov_base;
995 error = vn_io_fault_touch(base, uio);
998 if (len < PAGE_SIZE) {
1000 error = vn_io_fault_touch(base + len - 1, uio);
1005 if (++i >= uio->uio_iovcnt)
1007 iov = uio->uio_iov + i;
1008 base = iov->iov_base;
1020 * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1021 * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1022 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1023 * into args and call vn_io_fault1() to handle faults during the user
1024 * mode buffer accesses.
1027 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1030 vm_page_t ma[io_hold_cnt + 2];
1031 struct uio *uio_clone, short_uio;
1032 struct iovec short_iovec[1];
1033 vm_page_t *prev_td_ma;
1035 vm_offset_t addr, end;
1038 int error, cnt, saveheld, prev_td_ma_cnt;
1040 if (vn_io_fault_prefault) {
1041 error = vn_io_fault_prefault_user(uio);
1043 return (error); /* Or ignore ? */
1046 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1049 * The UFS follows IO_UNIT directive and replays back both
1050 * uio_offset and uio_resid if an error is encountered during the
1051 * operation. But, since the iovec may be already advanced,
1052 * uio is still in an inconsistent state.
1054 * Cache a copy of the original uio, which is advanced to the redo
1055 * point using UIO_NOCOPY below.
1057 uio_clone = cloneuio(uio);
1058 resid = uio->uio_resid;
1060 short_uio.uio_segflg = UIO_USERSPACE;
1061 short_uio.uio_rw = uio->uio_rw;
1062 short_uio.uio_td = uio->uio_td;
1064 error = vn_io_fault_doio(args, uio, td);
1065 if (error != EFAULT)
1068 atomic_add_long(&vn_io_faults_cnt, 1);
1069 uio_clone->uio_segflg = UIO_NOCOPY;
1070 uiomove(NULL, resid - uio->uio_resid, uio_clone);
1071 uio_clone->uio_segflg = uio->uio_segflg;
1073 saveheld = curthread_pflags_set(TDP_UIOHELD);
1074 prev_td_ma = td->td_ma;
1075 prev_td_ma_cnt = td->td_ma_cnt;
1077 while (uio_clone->uio_resid != 0) {
1078 len = uio_clone->uio_iov->iov_len;
1080 KASSERT(uio_clone->uio_iovcnt >= 1,
1081 ("iovcnt underflow"));
1082 uio_clone->uio_iov++;
1083 uio_clone->uio_iovcnt--;
1086 if (len > io_hold_cnt * PAGE_SIZE)
1087 len = io_hold_cnt * PAGE_SIZE;
1088 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1089 end = round_page(addr + len);
1094 cnt = atop(end - trunc_page(addr));
1096 * A perfectly misaligned address and length could cause
1097 * both the start and the end of the chunk to use partial
1098 * page. +2 accounts for such a situation.
1100 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1101 addr, len, prot, ma, io_hold_cnt + 2);
1106 short_uio.uio_iov = &short_iovec[0];
1107 short_iovec[0].iov_base = (void *)addr;
1108 short_uio.uio_iovcnt = 1;
1109 short_uio.uio_resid = short_iovec[0].iov_len = len;
1110 short_uio.uio_offset = uio_clone->uio_offset;
1112 td->td_ma_cnt = cnt;
1114 error = vn_io_fault_doio(args, &short_uio, td);
1115 vm_page_unhold_pages(ma, cnt);
1116 adv = len - short_uio.uio_resid;
1118 uio_clone->uio_iov->iov_base =
1119 (char *)uio_clone->uio_iov->iov_base + adv;
1120 uio_clone->uio_iov->iov_len -= adv;
1121 uio_clone->uio_resid -= adv;
1122 uio_clone->uio_offset += adv;
1124 uio->uio_resid -= adv;
1125 uio->uio_offset += adv;
1127 if (error != 0 || adv == 0)
1130 td->td_ma = prev_td_ma;
1131 td->td_ma_cnt = prev_td_ma_cnt;
1132 curthread_pflags_restore(saveheld);
1134 free(uio_clone, M_IOV);
1139 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1140 int flags, struct thread *td)
1145 struct vn_io_fault_args args;
1148 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1150 foffset_lock_uio(fp, uio, flags);
1151 if (do_vn_io_fault(vp, uio)) {
1152 args.kind = VN_IO_FAULT_FOP;
1153 args.args.fop_args.fp = fp;
1154 args.args.fop_args.doio = doio;
1155 args.cred = active_cred;
1156 args.flags = flags | FOF_OFFSET;
1157 if (uio->uio_rw == UIO_READ) {
1158 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1159 uio->uio_offset + uio->uio_resid);
1160 } else if ((fp->f_flag & O_APPEND) != 0 ||
1161 (flags & FOF_OFFSET) == 0) {
1162 /* For appenders, punt and lock the whole range. */
1163 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1165 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1166 uio->uio_offset + uio->uio_resid);
1168 error = vn_io_fault1(vp, uio, &args, td);
1169 vn_rangelock_unlock(vp, rl_cookie);
1171 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1173 foffset_unlock_uio(fp, uio, flags);
1178 * Helper function to perform the requested uiomove operation using
1179 * the held pages for io->uio_iov[0].iov_base buffer instead of
1180 * copyin/copyout. Access to the pages with uiomove_fromphys()
1181 * instead of iov_base prevents page faults that could occur due to
1182 * pmap_collect() invalidating the mapping created by
1183 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1184 * object cleanup revoking the write access from page mappings.
1186 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1187 * instead of plain uiomove().
1190 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1192 struct uio transp_uio;
1193 struct iovec transp_iov[1];
1199 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1200 uio->uio_segflg != UIO_USERSPACE)
1201 return (uiomove(data, xfersize, uio));
1203 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1204 transp_iov[0].iov_base = data;
1205 transp_uio.uio_iov = &transp_iov[0];
1206 transp_uio.uio_iovcnt = 1;
1207 if (xfersize > uio->uio_resid)
1208 xfersize = uio->uio_resid;
1209 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1210 transp_uio.uio_offset = 0;
1211 transp_uio.uio_segflg = UIO_SYSSPACE;
1213 * Since transp_iov points to data, and td_ma page array
1214 * corresponds to original uio->uio_iov, we need to invert the
1215 * direction of the i/o operation as passed to
1216 * uiomove_fromphys().
1218 switch (uio->uio_rw) {
1220 transp_uio.uio_rw = UIO_READ;
1223 transp_uio.uio_rw = UIO_WRITE;
1226 transp_uio.uio_td = uio->uio_td;
1227 error = uiomove_fromphys(td->td_ma,
1228 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1229 xfersize, &transp_uio);
1230 adv = xfersize - transp_uio.uio_resid;
1232 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1233 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1235 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1237 td->td_ma_cnt -= pgadv;
1238 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1239 uio->uio_iov->iov_len -= adv;
1240 uio->uio_resid -= adv;
1241 uio->uio_offset += adv;
1246 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1250 vm_offset_t iov_base;
1254 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1255 uio->uio_segflg != UIO_USERSPACE)
1256 return (uiomove_fromphys(ma, offset, xfersize, uio));
1258 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1259 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1260 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1261 switch (uio->uio_rw) {
1263 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1267 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1271 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1273 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1275 td->td_ma_cnt -= pgadv;
1276 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1277 uio->uio_iov->iov_len -= cnt;
1278 uio->uio_resid -= cnt;
1279 uio->uio_offset += cnt;
1285 * File table truncate routine.
1288 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1300 * Lock the whole range for truncation. Otherwise split i/o
1301 * might happen partly before and partly after the truncation.
1303 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1304 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1307 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1308 AUDIT_ARG_VNODE1(vp);
1309 if (vp->v_type == VDIR) {
1314 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1318 error = vn_writechk(vp);
1321 vattr.va_size = length;
1322 if ((fp->f_flag & O_FSYNC) != 0)
1323 vattr.va_vaflags |= VA_SYNC;
1324 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1328 vn_finished_write(mp);
1330 vn_rangelock_unlock(vp, rl_cookie);
1335 * File table vnode stat routine.
1338 vn_statfile(struct file *fp, struct stat *sb, struct ucred *active_cred,
1341 struct vnode *vp = fp->f_vnode;
1344 vn_lock(vp, LK_SHARED | LK_RETRY);
1345 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1352 * Stat a vnode; implementation for the stat syscall
1355 vn_stat(struct vnode *vp, struct stat *sb, struct ucred *active_cred,
1356 struct ucred *file_cred, struct thread *td)
1363 AUDIT_ARG_VNODE1(vp);
1365 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1373 * Initialize defaults for new and unusual fields, so that file
1374 * systems which don't support these fields don't need to know
1377 vap->va_birthtime.tv_sec = -1;
1378 vap->va_birthtime.tv_nsec = 0;
1379 vap->va_fsid = VNOVAL;
1380 vap->va_rdev = NODEV;
1382 error = VOP_GETATTR(vp, vap, active_cred);
1387 * Zero the spare stat fields
1389 bzero(sb, sizeof *sb);
1392 * Copy from vattr table
1394 if (vap->va_fsid != VNOVAL)
1395 sb->st_dev = vap->va_fsid;
1397 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1398 sb->st_ino = vap->va_fileid;
1399 mode = vap->va_mode;
1400 switch (vap->va_type) {
1426 sb->st_nlink = vap->va_nlink;
1427 sb->st_uid = vap->va_uid;
1428 sb->st_gid = vap->va_gid;
1429 sb->st_rdev = vap->va_rdev;
1430 if (vap->va_size > OFF_MAX)
1432 sb->st_size = vap->va_size;
1433 sb->st_atim = vap->va_atime;
1434 sb->st_mtim = vap->va_mtime;
1435 sb->st_ctim = vap->va_ctime;
1436 sb->st_birthtim = vap->va_birthtime;
1439 * According to www.opengroup.org, the meaning of st_blksize is
1440 * "a filesystem-specific preferred I/O block size for this
1441 * object. In some filesystem types, this may vary from file
1443 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1446 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1448 sb->st_flags = vap->va_flags;
1449 if (priv_check(td, PRIV_VFS_GENERATION))
1452 sb->st_gen = vap->va_gen;
1454 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1459 * File table vnode ioctl routine.
1462 vn_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
1470 switch (vp->v_type) {
1475 vn_lock(vp, LK_SHARED | LK_RETRY);
1476 error = VOP_GETATTR(vp, &vattr, active_cred);
1479 *(int *)data = vattr.va_size - fp->f_offset;
1485 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1490 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1498 * File table vnode poll routine.
1501 vn_poll(struct file *fp, int events, struct ucred *active_cred,
1509 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1510 AUDIT_ARG_VNODE1(vp);
1511 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1516 error = VOP_POLL(vp, events, fp->f_cred, td);
1521 * Acquire the requested lock and then check for validity. LK_RETRY
1522 * permits vn_lock to return doomed vnodes.
1525 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1529 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1530 ("vn_lock: no locktype"));
1531 VNASSERT(vp->v_holdcnt != 0, vp, ("vn_lock: zero hold count"));
1533 error = VOP_LOCK1(vp, flags, file, line);
1534 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1535 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1536 ("vn_lock: error %d incompatible with flags %#x", error, flags));
1538 if ((flags & LK_RETRY) == 0) {
1539 if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) {
1543 } else if (error != 0)
1549 * File table vnode close routine.
1552 vn_closefile(struct file *fp, struct thread *td)
1560 fp->f_ops = &badfileops;
1561 ref= (fp->f_flag & FHASLOCK) != 0 && fp->f_type == DTYPE_VNODE;
1563 error = vn_close1(vp, fp->f_flag, fp->f_cred, td, ref);
1565 if (__predict_false(ref)) {
1566 lf.l_whence = SEEK_SET;
1569 lf.l_type = F_UNLCK;
1570 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1577 vn_suspendable(struct mount *mp)
1580 return (mp->mnt_op->vfs_susp_clean != NULL);
1584 * Preparing to start a filesystem write operation. If the operation is
1585 * permitted, then we bump the count of operations in progress and
1586 * proceed. If a suspend request is in progress, we wait until the
1587 * suspension is over, and then proceed.
1590 vn_start_write_locked(struct mount *mp, int flags)
1594 mtx_assert(MNT_MTX(mp), MA_OWNED);
1598 * Check on status of suspension.
1600 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1601 mp->mnt_susp_owner != curthread) {
1602 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
1603 (flags & PCATCH) : 0) | (PUSER - 1);
1604 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1605 if (flags & V_NOWAIT) {
1606 error = EWOULDBLOCK;
1609 error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
1615 if (flags & V_XSLEEP)
1617 mp->mnt_writeopcount++;
1619 if (error != 0 || (flags & V_XSLEEP) != 0)
1626 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
1631 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1632 ("V_MNTREF requires mp"));
1636 * If a vnode is provided, get and return the mount point that
1637 * to which it will write.
1640 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1642 if (error != EOPNOTSUPP)
1647 if ((mp = *mpp) == NULL)
1650 if (!vn_suspendable(mp)) {
1651 if (vp != NULL || (flags & V_MNTREF) != 0)
1657 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1659 * As long as a vnode is not provided we need to acquire a
1660 * refcount for the provided mountpoint too, in order to
1661 * emulate a vfs_ref().
1664 if (vp == NULL && (flags & V_MNTREF) == 0)
1667 return (vn_start_write_locked(mp, flags));
1671 * Secondary suspension. Used by operations such as vop_inactive
1672 * routines that are needed by the higher level functions. These
1673 * are allowed to proceed until all the higher level functions have
1674 * completed (indicated by mnt_writeopcount dropping to zero). At that
1675 * time, these operations are halted until the suspension is over.
1678 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags)
1683 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1684 ("V_MNTREF requires mp"));
1688 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1690 if (error != EOPNOTSUPP)
1696 * If we are not suspended or have not yet reached suspended
1697 * mode, then let the operation proceed.
1699 if ((mp = *mpp) == NULL)
1702 if (!vn_suspendable(mp)) {
1703 if (vp != NULL || (flags & V_MNTREF) != 0)
1709 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1711 * As long as a vnode is not provided we need to acquire a
1712 * refcount for the provided mountpoint too, in order to
1713 * emulate a vfs_ref().
1716 if (vp == NULL && (flags & V_MNTREF) == 0)
1718 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1719 mp->mnt_secondary_writes++;
1720 mp->mnt_secondary_accwrites++;
1724 if (flags & V_NOWAIT) {
1727 return (EWOULDBLOCK);
1730 * Wait for the suspension to finish.
1732 error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
1733 ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
1742 * Filesystem write operation has completed. If we are suspending and this
1743 * operation is the last one, notify the suspender that the suspension is
1747 vn_finished_write(struct mount *mp)
1749 if (mp == NULL || !vn_suspendable(mp))
1753 mp->mnt_writeopcount--;
1754 if (mp->mnt_writeopcount < 0)
1755 panic("vn_finished_write: neg cnt");
1756 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1757 mp->mnt_writeopcount <= 0)
1758 wakeup(&mp->mnt_writeopcount);
1764 * Filesystem secondary write operation has completed. If we are
1765 * suspending and this operation is the last one, notify the suspender
1766 * that the suspension is now in effect.
1769 vn_finished_secondary_write(struct mount *mp)
1771 if (mp == NULL || !vn_suspendable(mp))
1775 mp->mnt_secondary_writes--;
1776 if (mp->mnt_secondary_writes < 0)
1777 panic("vn_finished_secondary_write: neg cnt");
1778 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1779 mp->mnt_secondary_writes <= 0)
1780 wakeup(&mp->mnt_secondary_writes);
1787 * Request a filesystem to suspend write operations.
1790 vfs_write_suspend(struct mount *mp, int flags)
1794 MPASS(vn_suspendable(mp));
1797 if (mp->mnt_susp_owner == curthread) {
1801 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1802 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1805 * Unmount holds a write reference on the mount point. If we
1806 * own busy reference and drain for writers, we deadlock with
1807 * the reference draining in the unmount path. Callers of
1808 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1809 * vfs_busy() reference is owned and caller is not in the
1812 if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1813 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1818 mp->mnt_kern_flag |= MNTK_SUSPEND;
1819 mp->mnt_susp_owner = curthread;
1820 if (mp->mnt_writeopcount > 0)
1821 (void) msleep(&mp->mnt_writeopcount,
1822 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1825 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1826 vfs_write_resume(mp, 0);
1831 * Request a filesystem to resume write operations.
1834 vfs_write_resume(struct mount *mp, int flags)
1837 MPASS(vn_suspendable(mp));
1840 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1841 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1842 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1844 mp->mnt_susp_owner = NULL;
1845 wakeup(&mp->mnt_writeopcount);
1846 wakeup(&mp->mnt_flag);
1847 curthread->td_pflags &= ~TDP_IGNSUSP;
1848 if ((flags & VR_START_WRITE) != 0) {
1850 mp->mnt_writeopcount++;
1853 if ((flags & VR_NO_SUSPCLR) == 0)
1855 } else if ((flags & VR_START_WRITE) != 0) {
1857 vn_start_write_locked(mp, 0);
1864 * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1868 vfs_write_suspend_umnt(struct mount *mp)
1872 MPASS(vn_suspendable(mp));
1873 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1874 ("vfs_write_suspend_umnt: recursed"));
1876 /* dounmount() already called vn_start_write(). */
1878 vn_finished_write(mp);
1879 error = vfs_write_suspend(mp, 0);
1881 vn_start_write(NULL, &mp, V_WAIT);
1885 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1888 vn_start_write(NULL, &mp, V_WAIT);
1890 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1891 wakeup(&mp->mnt_flag);
1893 curthread->td_pflags |= TDP_IGNSUSP;
1898 * Implement kqueues for files by translating it to vnode operation.
1901 vn_kqfilter(struct file *fp, struct knote *kn)
1904 return (VOP_KQFILTER(fp->f_vnode, kn));
1908 * Simplified in-kernel wrapper calls for extended attribute access.
1909 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1910 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1913 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1914 const char *attrname, int *buflen, char *buf, struct thread *td)
1920 iov.iov_len = *buflen;
1923 auio.uio_iov = &iov;
1924 auio.uio_iovcnt = 1;
1925 auio.uio_rw = UIO_READ;
1926 auio.uio_segflg = UIO_SYSSPACE;
1928 auio.uio_offset = 0;
1929 auio.uio_resid = *buflen;
1931 if ((ioflg & IO_NODELOCKED) == 0)
1932 vn_lock(vp, LK_SHARED | LK_RETRY);
1934 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1936 /* authorize attribute retrieval as kernel */
1937 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1940 if ((ioflg & IO_NODELOCKED) == 0)
1944 *buflen = *buflen - auio.uio_resid;
1951 * XXX failure mode if partially written?
1954 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1955 const char *attrname, int buflen, char *buf, struct thread *td)
1962 iov.iov_len = buflen;
1965 auio.uio_iov = &iov;
1966 auio.uio_iovcnt = 1;
1967 auio.uio_rw = UIO_WRITE;
1968 auio.uio_segflg = UIO_SYSSPACE;
1970 auio.uio_offset = 0;
1971 auio.uio_resid = buflen;
1973 if ((ioflg & IO_NODELOCKED) == 0) {
1974 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1976 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1979 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1981 /* authorize attribute setting as kernel */
1982 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1984 if ((ioflg & IO_NODELOCKED) == 0) {
1985 vn_finished_write(mp);
1993 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1994 const char *attrname, struct thread *td)
1999 if ((ioflg & IO_NODELOCKED) == 0) {
2000 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2002 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2005 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2007 /* authorize attribute removal as kernel */
2008 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2009 if (error == EOPNOTSUPP)
2010 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2013 if ((ioflg & IO_NODELOCKED) == 0) {
2014 vn_finished_write(mp);
2022 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2026 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2030 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2033 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2038 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2039 int lkflags, struct vnode **rvp)
2044 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2046 ltype = VOP_ISLOCKED(vp);
2047 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2048 ("vn_vget_ino: vp not locked"));
2049 error = vfs_busy(mp, MBF_NOWAIT);
2053 error = vfs_busy(mp, 0);
2054 vn_lock(vp, ltype | LK_RETRY);
2058 if (vp->v_iflag & VI_DOOMED) {
2064 error = alloc(mp, alloc_arg, lkflags, rvp);
2067 vn_lock(vp, ltype | LK_RETRY);
2068 if (vp->v_iflag & VI_DOOMED) {
2081 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2085 if (vp->v_type != VREG || td == NULL)
2087 if ((uoff_t)uio->uio_offset + uio->uio_resid >
2088 lim_cur(td, RLIMIT_FSIZE)) {
2089 PROC_LOCK(td->td_proc);
2090 kern_psignal(td->td_proc, SIGXFSZ);
2091 PROC_UNLOCK(td->td_proc);
2098 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2105 vn_lock(vp, LK_SHARED | LK_RETRY);
2106 AUDIT_ARG_VNODE1(vp);
2109 return (setfmode(td, active_cred, vp, mode));
2113 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2120 vn_lock(vp, LK_SHARED | LK_RETRY);
2121 AUDIT_ARG_VNODE1(vp);
2124 return (setfown(td, active_cred, vp, uid, gid));
2128 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2132 if ((object = vp->v_object) == NULL)
2134 VM_OBJECT_WLOCK(object);
2135 vm_object_page_remove(object, start, end, 0);
2136 VM_OBJECT_WUNLOCK(object);
2140 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2148 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2149 ("Wrong command %lu", cmd));
2151 if (vn_lock(vp, LK_SHARED) != 0)
2153 if (vp->v_type != VREG) {
2157 error = VOP_GETATTR(vp, &va, cred);
2161 if (noff >= va.va_size) {
2165 bsize = vp->v_mount->mnt_stat.f_iosize;
2166 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2167 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2168 if (error == EOPNOTSUPP) {
2172 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2173 (bnp != -1 && cmd == FIOSEEKDATA)) {
2180 if (noff > va.va_size)
2182 /* noff == va.va_size. There is an implicit hole at the end of file. */
2183 if (cmd == FIOSEEKDATA)
2193 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2198 off_t foffset, size;
2201 cred = td->td_ucred;
2203 foffset = foffset_lock(fp, 0);
2204 noneg = (vp->v_type != VCHR);
2210 (offset > 0 && foffset > OFF_MAX - offset))) {
2217 vn_lock(vp, LK_SHARED | LK_RETRY);
2218 error = VOP_GETATTR(vp, &vattr, cred);
2224 * If the file references a disk device, then fetch
2225 * the media size and use that to determine the ending
2228 if (vattr.va_size == 0 && vp->v_type == VCHR &&
2229 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2230 vattr.va_size = size;
2232 (vattr.va_size > OFF_MAX ||
2233 (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2237 offset += vattr.va_size;
2242 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2245 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2250 if (error == 0 && noneg && offset < 0)
2254 VFS_KNOTE_UNLOCKED(vp, 0);
2255 td->td_uretoff.tdu_off = offset;
2257 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2262 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2268 * Grant permission if the caller is the owner of the file, or
2269 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2270 * on the file. If the time pointer is null, then write
2271 * permission on the file is also sufficient.
2273 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2274 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2275 * will be allowed to set the times [..] to the current
2278 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2279 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2280 error = VOP_ACCESS(vp, VWRITE, cred, td);
2285 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2290 if (fp->f_type == DTYPE_FIFO)
2291 kif->kf_type = KF_TYPE_FIFO;
2293 kif->kf_type = KF_TYPE_VNODE;
2296 FILEDESC_SUNLOCK(fdp);
2297 error = vn_fill_kinfo_vnode(vp, kif);
2299 FILEDESC_SLOCK(fdp);
2304 vn_fill_junk(struct kinfo_file *kif)
2309 * Simulate vn_fullpath returning changing values for a given
2310 * vp during e.g. coredump.
2312 len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1;
2313 olen = strlen(kif->kf_path);
2315 strcpy(&kif->kf_path[len - 1], "$");
2317 for (; olen < len; olen++)
2318 strcpy(&kif->kf_path[olen], "A");
2322 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2325 char *fullpath, *freepath;
2328 kif->kf_un.kf_file.kf_file_type = vntype_to_kinfo(vp->v_type);
2331 error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2333 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2335 if (freepath != NULL)
2336 free(freepath, M_TEMP);
2338 KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path,
2343 * Retrieve vnode attributes.
2345 va.va_fsid = VNOVAL;
2347 vn_lock(vp, LK_SHARED | LK_RETRY);
2348 error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2352 if (va.va_fsid != VNOVAL)
2353 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2355 kif->kf_un.kf_file.kf_file_fsid =
2356 vp->v_mount->mnt_stat.f_fsid.val[0];
2357 kif->kf_un.kf_file.kf_file_fsid_freebsd11 =
2358 kif->kf_un.kf_file.kf_file_fsid; /* truncate */
2359 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2360 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2361 kif->kf_un.kf_file.kf_file_size = va.va_size;
2362 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2363 kif->kf_un.kf_file.kf_file_rdev_freebsd11 =
2364 kif->kf_un.kf_file.kf_file_rdev; /* truncate */
2369 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
2370 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
2374 struct pmckern_map_in pkm;
2380 boolean_t writecounted;
2383 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
2384 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
2386 * POSIX shared-memory objects are defined to have
2387 * kernel persistence, and are not defined to support
2388 * read(2)/write(2) -- or even open(2). Thus, we can
2389 * use MAP_ASYNC to trade on-disk coherence for speed.
2390 * The shm_open(3) library routine turns on the FPOSIXSHM
2391 * flag to request this behavior.
2393 if ((fp->f_flag & FPOSIXSHM) != 0)
2394 flags |= MAP_NOSYNC;
2399 * Ensure that file and memory protections are
2400 * compatible. Note that we only worry about
2401 * writability if mapping is shared; in this case,
2402 * current and max prot are dictated by the open file.
2403 * XXX use the vnode instead? Problem is: what
2404 * credentials do we use for determination? What if
2405 * proc does a setuid?
2408 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
2409 maxprot = VM_PROT_NONE;
2410 if ((prot & VM_PROT_EXECUTE) != 0)
2413 maxprot = VM_PROT_EXECUTE;
2414 if ((fp->f_flag & FREAD) != 0)
2415 maxprot |= VM_PROT_READ;
2416 else if ((prot & VM_PROT_READ) != 0)
2420 * If we are sharing potential changes via MAP_SHARED and we
2421 * are trying to get write permission although we opened it
2422 * without asking for it, bail out.
2424 if ((flags & MAP_SHARED) != 0) {
2425 if ((fp->f_flag & FWRITE) != 0)
2426 maxprot |= VM_PROT_WRITE;
2427 else if ((prot & VM_PROT_WRITE) != 0)
2430 maxprot |= VM_PROT_WRITE;
2431 cap_maxprot |= VM_PROT_WRITE;
2433 maxprot &= cap_maxprot;
2436 * For regular files and shared memory, POSIX requires that
2437 * the value of foff be a legitimate offset within the data
2438 * object. In particular, negative offsets are invalid.
2439 * Blocking negative offsets and overflows here avoids
2440 * possible wraparound or user-level access into reserved
2441 * ranges of the data object later. In contrast, POSIX does
2442 * not dictate how offsets are used by device drivers, so in
2443 * the case of a device mapping a negative offset is passed
2450 foff < 0 || foff > OFF_MAX - size)
2453 writecounted = FALSE;
2454 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp,
2455 &foff, &object, &writecounted);
2458 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2459 foff, writecounted, td);
2462 * If this mapping was accounted for in the vnode's
2463 * writecount, then undo that now.
2466 vnode_pager_release_writecount(object, 0, size);
2467 vm_object_deallocate(object);
2470 /* Inform hwpmc(4) if an executable is being mapped. */
2471 if (PMC_HOOK_INSTALLED(PMC_FN_MMAP)) {
2472 if ((prot & VM_PROT_EXECUTE) != 0 && error == 0) {
2474 pkm.pm_address = (uintptr_t) *addr;
2475 PMC_CALL_HOOK_UNLOCKED(td, PMC_FN_MMAP, (void *) &pkm);
2483 vn_fsid(struct vnode *vp, struct vattr *va)
2487 f = &vp->v_mount->mnt_stat.f_fsid;
2488 va->va_fsid = (uint32_t)f->val[1];
2489 va->va_fsid <<= sizeof(f->val[1]) * NBBY;
2490 va->va_fsid += (uint32_t)f->val[0];