2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org>
11 * Copyright (c) 2013, 2014 The FreeBSD Foundation
13 * Portions of this software were developed by Konstantin Belousov
14 * under sponsorship from the FreeBSD Foundation.
16 * Redistribution and use in source and binary forms, with or without
17 * modification, are permitted provided that the following conditions
19 * 1. Redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer.
21 * 2. Redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
46 #include <sys/param.h>
47 #include <sys/systm.h>
49 #include <sys/fcntl.h>
55 #include <sys/limits.h>
57 #include <sys/mount.h>
58 #include <sys/mutex.h>
59 #include <sys/namei.h>
60 #include <sys/vnode.h>
63 #include <sys/filio.h>
64 #include <sys/resourcevar.h>
65 #include <sys/rwlock.h>
67 #include <sys/sysctl.h>
68 #include <sys/ttycom.h>
70 #include <sys/syslog.h>
71 #include <sys/unistd.h>
74 #include <security/audit/audit.h>
75 #include <security/mac/mac_framework.h>
78 #include <vm/vm_extern.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_page.h>
84 static fo_rdwr_t vn_read;
85 static fo_rdwr_t vn_write;
86 static fo_rdwr_t vn_io_fault;
87 static fo_truncate_t vn_truncate;
88 static fo_ioctl_t vn_ioctl;
89 static fo_poll_t vn_poll;
90 static fo_kqfilter_t vn_kqfilter;
91 static fo_stat_t vn_statfile;
92 static fo_close_t vn_closefile;
94 struct fileops vnops = {
95 .fo_read = vn_io_fault,
96 .fo_write = vn_io_fault,
97 .fo_truncate = vn_truncate,
100 .fo_kqfilter = vn_kqfilter,
101 .fo_stat = vn_statfile,
102 .fo_close = vn_closefile,
103 .fo_chmod = vn_chmod,
104 .fo_chown = vn_chown,
105 .fo_sendfile = vn_sendfile,
107 .fo_fill_kinfo = vn_fill_kinfo,
108 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
111 static const int io_hold_cnt = 16;
112 static int vn_io_fault_enable = 1;
113 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
114 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
115 static u_long vn_io_faults_cnt;
116 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
117 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
120 * Returns true if vn_io_fault mode of handling the i/o request should
124 do_vn_io_fault(struct vnode *vp, struct uio *uio)
128 return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
129 (mp = vp->v_mount) != NULL &&
130 (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
134 * Structure used to pass arguments to vn_io_fault1(), to do either
135 * file- or vnode-based I/O calls.
137 struct vn_io_fault_args {
145 struct fop_args_tag {
149 struct vop_args_tag {
155 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
156 struct vn_io_fault_args *args, struct thread *td);
159 vn_open(ndp, flagp, cmode, fp)
160 struct nameidata *ndp;
164 struct thread *td = ndp->ni_cnd.cn_thread;
166 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
170 * Common code for vnode open operations via a name lookup.
171 * Lookup the vnode and invoke VOP_CREATE if needed.
172 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
174 * Note that this does NOT free nameidata for the successful case,
175 * due to the NDINIT being done elsewhere.
178 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
179 struct ucred *cred, struct file *fp)
183 struct thread *td = ndp->ni_cnd.cn_thread;
185 struct vattr *vap = &vat;
190 if (fmode & O_CREAT) {
191 ndp->ni_cnd.cn_nameiop = CREATE;
192 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF;
193 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
194 ndp->ni_cnd.cn_flags |= FOLLOW;
195 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
196 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
197 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
198 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
200 if ((error = namei(ndp)) != 0)
202 if (ndp->ni_vp == NULL) {
205 vap->va_mode = cmode;
207 vap->va_vaflags |= VA_EXCLUSIVE;
208 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
209 NDFREE(ndp, NDF_ONLY_PNBUF);
211 if ((error = vn_start_write(NULL, &mp,
212 V_XSLEEP | PCATCH)) != 0)
217 error = mac_vnode_check_create(cred, ndp->ni_dvp,
221 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
224 vn_finished_write(mp);
226 NDFREE(ndp, NDF_ONLY_PNBUF);
232 if (ndp->ni_dvp == ndp->ni_vp)
238 if (fmode & O_EXCL) {
245 ndp->ni_cnd.cn_nameiop = LOOKUP;
246 ndp->ni_cnd.cn_flags = ISOPEN |
247 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
248 if (!(fmode & FWRITE))
249 ndp->ni_cnd.cn_flags |= LOCKSHARED;
250 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
251 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
252 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
253 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
254 if ((error = namei(ndp)) != 0)
258 error = vn_open_vnode(vp, fmode, cred, td, fp);
264 NDFREE(ndp, NDF_ONLY_PNBUF);
272 * Common code for vnode open operations once a vnode is located.
273 * Check permissions, and call the VOP_OPEN routine.
276 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
277 struct thread *td, struct file *fp)
282 int error, have_flock, lock_flags, type;
284 if (vp->v_type == VLNK)
286 if (vp->v_type == VSOCK)
288 if (vp->v_type != VDIR && fmode & O_DIRECTORY)
291 if (fmode & (FWRITE | O_TRUNC)) {
292 if (vp->v_type == VDIR)
300 if ((fmode & O_APPEND) && (fmode & FWRITE))
303 error = mac_vnode_check_open(cred, vp, accmode);
307 if ((fmode & O_CREAT) == 0) {
308 if (accmode & VWRITE) {
309 error = vn_writechk(vp);
314 error = VOP_ACCESS(vp, accmode, cred, td);
319 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
320 vn_lock(vp, LK_UPGRADE | LK_RETRY);
321 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
324 if (fmode & (O_EXLOCK | O_SHLOCK)) {
325 KASSERT(fp != NULL, ("open with flock requires fp"));
326 lock_flags = VOP_ISLOCKED(vp);
328 lf.l_whence = SEEK_SET;
331 if (fmode & O_EXLOCK)
336 if ((fmode & FNONBLOCK) == 0)
338 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
339 have_flock = (error == 0);
340 vn_lock(vp, lock_flags | LK_RETRY);
341 if (error == 0 && vp->v_iflag & VI_DOOMED)
344 * Another thread might have used this vnode as an
345 * executable while the vnode lock was dropped.
346 * Ensure the vnode is still able to be opened for
347 * writing after the lock has been obtained.
349 if (error == 0 && accmode & VWRITE)
350 error = vn_writechk(vp);
354 lf.l_whence = SEEK_SET;
358 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
361 vn_start_write(vp, &mp, V_WAIT);
362 vn_lock(vp, lock_flags | LK_RETRY);
363 (void)VOP_CLOSE(vp, fmode, cred, td);
364 vn_finished_write(mp);
365 /* Prevent second close from fdrop()->vn_close(). */
367 fp->f_ops= &badfileops;
370 fp->f_flag |= FHASLOCK;
372 if (fmode & FWRITE) {
373 VOP_ADD_WRITECOUNT(vp, 1);
374 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
375 __func__, vp, vp->v_writecount);
377 ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
382 * Check for write permissions on the specified vnode.
383 * Prototype text segments cannot be written.
387 register struct vnode *vp;
390 ASSERT_VOP_LOCKED(vp, "vn_writechk");
392 * If there's shared text associated with
393 * the vnode, try to free it up once. If
394 * we fail, we can't allow writing.
406 vn_close(vp, flags, file_cred, td)
407 register struct vnode *vp;
409 struct ucred *file_cred;
413 int error, lock_flags;
415 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
416 MNT_EXTENDED_SHARED(vp->v_mount))
417 lock_flags = LK_SHARED;
419 lock_flags = LK_EXCLUSIVE;
421 vn_start_write(vp, &mp, V_WAIT);
422 vn_lock(vp, lock_flags | LK_RETRY);
423 if (flags & FWRITE) {
424 VNASSERT(vp->v_writecount > 0, vp,
425 ("vn_close: negative writecount"));
426 VOP_ADD_WRITECOUNT(vp, -1);
427 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
428 __func__, vp, vp->v_writecount);
430 error = VOP_CLOSE(vp, flags, file_cred, td);
432 vn_finished_write(mp);
437 * Heuristic to detect sequential operation.
440 sequential_heuristic(struct uio *uio, struct file *fp)
443 ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
444 if (fp->f_flag & FRDAHEAD)
445 return (fp->f_seqcount << IO_SEQSHIFT);
448 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
449 * that the first I/O is normally considered to be slightly
450 * sequential. Seeking to offset 0 doesn't change sequentiality
451 * unless previous seeks have reduced f_seqcount to 0, in which
452 * case offset 0 is not special.
454 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
455 uio->uio_offset == fp->f_nextoff) {
457 * f_seqcount is in units of fixed-size blocks so that it
458 * depends mainly on the amount of sequential I/O and not
459 * much on the number of sequential I/O's. The fixed size
460 * of 16384 is hard-coded here since it is (not quite) just
461 * a magic size that works well here. This size is more
462 * closely related to the best I/O size for real disks than
463 * to any block size used by software.
465 fp->f_seqcount += howmany(uio->uio_resid, 16384);
466 if (fp->f_seqcount > IO_SEQMAX)
467 fp->f_seqcount = IO_SEQMAX;
468 return (fp->f_seqcount << IO_SEQSHIFT);
471 /* Not sequential. Quickly draw-down sequentiality. */
472 if (fp->f_seqcount > 1)
480 * Package up an I/O request on a vnode into a uio and do it.
483 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
484 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
485 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
492 struct vn_io_fault_args args;
493 int error, lock_flags;
495 auio.uio_iov = &aiov;
497 aiov.iov_base = base;
499 auio.uio_resid = len;
500 auio.uio_offset = offset;
501 auio.uio_segflg = segflg;
506 if ((ioflg & IO_NODELOCKED) == 0) {
507 if ((ioflg & IO_RANGELOCKED) == 0) {
508 if (rw == UIO_READ) {
509 rl_cookie = vn_rangelock_rlock(vp, offset,
512 rl_cookie = vn_rangelock_wlock(vp, offset,
518 if (rw == UIO_WRITE) {
519 if (vp->v_type != VCHR &&
520 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
523 if (MNT_SHARED_WRITES(mp) ||
524 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
525 lock_flags = LK_SHARED;
527 lock_flags = LK_EXCLUSIVE;
529 lock_flags = LK_SHARED;
530 vn_lock(vp, lock_flags | LK_RETRY);
534 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
536 if ((ioflg & IO_NOMACCHECK) == 0) {
538 error = mac_vnode_check_read(active_cred, file_cred,
541 error = mac_vnode_check_write(active_cred, file_cred,
546 if (file_cred != NULL)
550 if (do_vn_io_fault(vp, &auio)) {
551 args.kind = VN_IO_FAULT_VOP;
554 args.args.vop_args.vp = vp;
555 error = vn_io_fault1(vp, &auio, &args, td);
556 } else if (rw == UIO_READ) {
557 error = VOP_READ(vp, &auio, ioflg, cred);
558 } else /* if (rw == UIO_WRITE) */ {
559 error = VOP_WRITE(vp, &auio, ioflg, cred);
563 *aresid = auio.uio_resid;
565 if (auio.uio_resid && error == 0)
567 if ((ioflg & IO_NODELOCKED) == 0) {
570 vn_finished_write(mp);
573 if (rl_cookie != NULL)
574 vn_rangelock_unlock(vp, rl_cookie);
579 * Package up an I/O request on a vnode into a uio and do it. The I/O
580 * request is split up into smaller chunks and we try to avoid saturating
581 * the buffer cache while potentially holding a vnode locked, so we
582 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
583 * to give other processes a chance to lock the vnode (either other processes
584 * core'ing the same binary, or unrelated processes scanning the directory).
587 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
588 file_cred, aresid, td)
596 struct ucred *active_cred;
597 struct ucred *file_cred;
608 * Force `offset' to a multiple of MAXBSIZE except possibly
609 * for the first chunk, so that filesystems only need to
610 * write full blocks except possibly for the first and last
613 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
617 if (rw != UIO_READ && vp->v_type == VREG)
620 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
621 ioflg, active_cred, file_cred, &iaresid, td);
622 len -= chunk; /* aresid calc already includes length */
626 base = (char *)base + chunk;
627 kern_yield(PRI_USER);
630 *aresid = len + iaresid;
635 foffset_lock(struct file *fp, int flags)
640 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
642 #if OFF_MAX <= LONG_MAX
644 * Caller only wants the current f_offset value. Assume that
645 * the long and shorter integer types reads are atomic.
647 if ((flags & FOF_NOLOCK) != 0)
648 return (fp->f_offset);
652 * According to McKusick the vn lock was protecting f_offset here.
653 * It is now protected by the FOFFSET_LOCKED flag.
655 mtxp = mtx_pool_find(mtxpool_sleep, fp);
657 if ((flags & FOF_NOLOCK) == 0) {
658 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
659 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
660 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
663 fp->f_vnread_flags |= FOFFSET_LOCKED;
671 foffset_unlock(struct file *fp, off_t val, int flags)
675 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
677 #if OFF_MAX <= LONG_MAX
678 if ((flags & FOF_NOLOCK) != 0) {
679 if ((flags & FOF_NOUPDATE) == 0)
681 if ((flags & FOF_NEXTOFF) != 0)
687 mtxp = mtx_pool_find(mtxpool_sleep, fp);
689 if ((flags & FOF_NOUPDATE) == 0)
691 if ((flags & FOF_NEXTOFF) != 0)
693 if ((flags & FOF_NOLOCK) == 0) {
694 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
695 ("Lost FOFFSET_LOCKED"));
696 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
697 wakeup(&fp->f_vnread_flags);
698 fp->f_vnread_flags = 0;
704 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
707 if ((flags & FOF_OFFSET) == 0)
708 uio->uio_offset = foffset_lock(fp, flags);
712 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
715 if ((flags & FOF_OFFSET) == 0)
716 foffset_unlock(fp, uio->uio_offset, flags);
720 get_advice(struct file *fp, struct uio *uio)
725 ret = POSIX_FADV_NORMAL;
726 if (fp->f_advice == NULL)
729 mtxp = mtx_pool_find(mtxpool_sleep, fp);
731 if (uio->uio_offset >= fp->f_advice->fa_start &&
732 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
733 ret = fp->f_advice->fa_advice;
739 * File table vnode read routine.
742 vn_read(fp, uio, active_cred, flags, td)
745 struct ucred *active_cred;
753 off_t offset, start, end;
755 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
757 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
760 if (fp->f_flag & FNONBLOCK)
762 if (fp->f_flag & O_DIRECT)
764 advice = get_advice(fp, uio);
765 vn_lock(vp, LK_SHARED | LK_RETRY);
768 case POSIX_FADV_NORMAL:
769 case POSIX_FADV_SEQUENTIAL:
770 case POSIX_FADV_NOREUSE:
771 ioflag |= sequential_heuristic(uio, fp);
773 case POSIX_FADV_RANDOM:
774 /* Disable read-ahead for random I/O. */
777 offset = uio->uio_offset;
780 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
783 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
784 fp->f_nextoff = uio->uio_offset;
786 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
787 offset != uio->uio_offset) {
789 * Use POSIX_FADV_DONTNEED to flush clean pages and
790 * buffers for the backing file after a
791 * POSIX_FADV_NOREUSE read(2). To optimize the common
792 * case of using POSIX_FADV_NOREUSE with sequential
793 * access, track the previous implicit DONTNEED
794 * request and grow this request to include the
795 * current read(2) in addition to the previous
796 * DONTNEED. With purely sequential access this will
797 * cause the DONTNEED requests to continously grow to
798 * cover all of the previously read regions of the
799 * file. This allows filesystem blocks that are
800 * accessed by multiple calls to read(2) to be flushed
801 * once the last read(2) finishes.
804 end = uio->uio_offset - 1;
805 mtxp = mtx_pool_find(mtxpool_sleep, fp);
807 if (fp->f_advice != NULL &&
808 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
809 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
810 start = fp->f_advice->fa_prevstart;
811 else if (fp->f_advice->fa_prevstart != 0 &&
812 fp->f_advice->fa_prevstart == end + 1)
813 end = fp->f_advice->fa_prevend;
814 fp->f_advice->fa_prevstart = start;
815 fp->f_advice->fa_prevend = end;
818 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
824 * File table vnode write routine.
827 vn_write(fp, uio, active_cred, flags, td)
830 struct ucred *active_cred;
837 int error, ioflag, lock_flags;
839 off_t offset, start, end;
841 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
843 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
845 if (vp->v_type == VREG)
848 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
850 if (fp->f_flag & FNONBLOCK)
852 if (fp->f_flag & O_DIRECT)
854 if ((fp->f_flag & O_FSYNC) ||
855 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
858 if (vp->v_type != VCHR &&
859 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
862 advice = get_advice(fp, uio);
864 if (MNT_SHARED_WRITES(mp) ||
865 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
866 lock_flags = LK_SHARED;
868 lock_flags = LK_EXCLUSIVE;
871 vn_lock(vp, lock_flags | LK_RETRY);
873 case POSIX_FADV_NORMAL:
874 case POSIX_FADV_SEQUENTIAL:
875 case POSIX_FADV_NOREUSE:
876 ioflag |= sequential_heuristic(uio, fp);
878 case POSIX_FADV_RANDOM:
879 /* XXX: Is this correct? */
882 offset = uio->uio_offset;
885 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
888 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
889 fp->f_nextoff = uio->uio_offset;
891 if (vp->v_type != VCHR)
892 vn_finished_write(mp);
893 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
894 offset != uio->uio_offset) {
896 * Use POSIX_FADV_DONTNEED to flush clean pages and
897 * buffers for the backing file after a
898 * POSIX_FADV_NOREUSE write(2). To optimize the
899 * common case of using POSIX_FADV_NOREUSE with
900 * sequential access, track the previous implicit
901 * DONTNEED request and grow this request to include
902 * the current write(2) in addition to the previous
903 * DONTNEED. With purely sequential access this will
904 * cause the DONTNEED requests to continously grow to
905 * cover all of the previously written regions of the
908 * Note that the blocks just written are almost
909 * certainly still dirty, so this only works when
910 * VOP_ADVISE() calls from subsequent writes push out
911 * the data written by this write(2) once the backing
912 * buffers are clean. However, as compared to forcing
913 * IO_DIRECT, this gives much saner behavior. Write
914 * clustering is still allowed, and clean pages are
915 * merely moved to the cache page queue rather than
916 * outright thrown away. This means a subsequent
917 * read(2) can still avoid hitting the disk if the
918 * pages have not been reclaimed.
920 * This does make POSIX_FADV_NOREUSE largely useless
921 * with non-sequential access. However, sequential
922 * access is the more common use case and the flag is
926 end = uio->uio_offset - 1;
927 mtxp = mtx_pool_find(mtxpool_sleep, fp);
929 if (fp->f_advice != NULL &&
930 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
931 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
932 start = fp->f_advice->fa_prevstart;
933 else if (fp->f_advice->fa_prevstart != 0 &&
934 fp->f_advice->fa_prevstart == end + 1)
935 end = fp->f_advice->fa_prevend;
936 fp->f_advice->fa_prevstart = start;
937 fp->f_advice->fa_prevend = end;
940 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
948 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
949 * prevent the following deadlock:
951 * Assume that the thread A reads from the vnode vp1 into userspace
952 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
953 * currently not resident, then system ends up with the call chain
954 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
955 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
956 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
957 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
958 * backed by the pages of vnode vp1, and some page in buf2 is not
959 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
961 * To prevent the lock order reversal and deadlock, vn_io_fault() does
962 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
963 * Instead, it first tries to do the whole range i/o with pagefaults
964 * disabled. If all pages in the i/o buffer are resident and mapped,
965 * VOP will succeed (ignoring the genuine filesystem errors).
966 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
967 * i/o in chunks, with all pages in the chunk prefaulted and held
968 * using vm_fault_quick_hold_pages().
970 * Filesystems using this deadlock avoidance scheme should use the
971 * array of the held pages from uio, saved in the curthread->td_ma,
972 * instead of doing uiomove(). A helper function
973 * vn_io_fault_uiomove() converts uiomove request into
974 * uiomove_fromphys() over td_ma array.
976 * Since vnode locks do not cover the whole i/o anymore, rangelocks
977 * make the current i/o request atomic with respect to other i/os and
982 * Decode vn_io_fault_args and perform the corresponding i/o.
985 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
989 switch (args->kind) {
990 case VN_IO_FAULT_FOP:
991 return ((args->args.fop_args.doio)(args->args.fop_args.fp,
992 uio, args->cred, args->flags, td));
993 case VN_IO_FAULT_VOP:
994 if (uio->uio_rw == UIO_READ) {
995 return (VOP_READ(args->args.vop_args.vp, uio,
996 args->flags, args->cred));
997 } else if (uio->uio_rw == UIO_WRITE) {
998 return (VOP_WRITE(args->args.vop_args.vp, uio,
999 args->flags, args->cred));
1003 panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
1008 * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1009 * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1010 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1011 * into args and call vn_io_fault1() to handle faults during the user
1012 * mode buffer accesses.
1015 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1018 vm_page_t ma[io_hold_cnt + 2];
1019 struct uio *uio_clone, short_uio;
1020 struct iovec short_iovec[1];
1021 vm_page_t *prev_td_ma;
1023 vm_offset_t addr, end;
1026 int error, cnt, save, saveheld, prev_td_ma_cnt;
1028 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1031 * The UFS follows IO_UNIT directive and replays back both
1032 * uio_offset and uio_resid if an error is encountered during the
1033 * operation. But, since the iovec may be already advanced,
1034 * uio is still in an inconsistent state.
1036 * Cache a copy of the original uio, which is advanced to the redo
1037 * point using UIO_NOCOPY below.
1039 uio_clone = cloneuio(uio);
1040 resid = uio->uio_resid;
1042 short_uio.uio_segflg = UIO_USERSPACE;
1043 short_uio.uio_rw = uio->uio_rw;
1044 short_uio.uio_td = uio->uio_td;
1046 save = vm_fault_disable_pagefaults();
1047 error = vn_io_fault_doio(args, uio, td);
1048 if (error != EFAULT)
1051 atomic_add_long(&vn_io_faults_cnt, 1);
1052 uio_clone->uio_segflg = UIO_NOCOPY;
1053 uiomove(NULL, resid - uio->uio_resid, uio_clone);
1054 uio_clone->uio_segflg = uio->uio_segflg;
1056 saveheld = curthread_pflags_set(TDP_UIOHELD);
1057 prev_td_ma = td->td_ma;
1058 prev_td_ma_cnt = td->td_ma_cnt;
1060 while (uio_clone->uio_resid != 0) {
1061 len = uio_clone->uio_iov->iov_len;
1063 KASSERT(uio_clone->uio_iovcnt >= 1,
1064 ("iovcnt underflow"));
1065 uio_clone->uio_iov++;
1066 uio_clone->uio_iovcnt--;
1069 if (len > io_hold_cnt * PAGE_SIZE)
1070 len = io_hold_cnt * PAGE_SIZE;
1071 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1072 end = round_page(addr + len);
1077 cnt = atop(end - trunc_page(addr));
1079 * A perfectly misaligned address and length could cause
1080 * both the start and the end of the chunk to use partial
1081 * page. +2 accounts for such a situation.
1083 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1084 addr, len, prot, ma, io_hold_cnt + 2);
1089 short_uio.uio_iov = &short_iovec[0];
1090 short_iovec[0].iov_base = (void *)addr;
1091 short_uio.uio_iovcnt = 1;
1092 short_uio.uio_resid = short_iovec[0].iov_len = len;
1093 short_uio.uio_offset = uio_clone->uio_offset;
1095 td->td_ma_cnt = cnt;
1097 error = vn_io_fault_doio(args, &short_uio, td);
1098 vm_page_unhold_pages(ma, cnt);
1099 adv = len - short_uio.uio_resid;
1101 uio_clone->uio_iov->iov_base =
1102 (char *)uio_clone->uio_iov->iov_base + adv;
1103 uio_clone->uio_iov->iov_len -= adv;
1104 uio_clone->uio_resid -= adv;
1105 uio_clone->uio_offset += adv;
1107 uio->uio_resid -= adv;
1108 uio->uio_offset += adv;
1110 if (error != 0 || adv == 0)
1113 td->td_ma = prev_td_ma;
1114 td->td_ma_cnt = prev_td_ma_cnt;
1115 curthread_pflags_restore(saveheld);
1117 vm_fault_enable_pagefaults(save);
1118 free(uio_clone, M_IOV);
1123 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1124 int flags, struct thread *td)
1129 struct vn_io_fault_args args;
1132 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1134 foffset_lock_uio(fp, uio, flags);
1135 if (do_vn_io_fault(vp, uio)) {
1136 args.kind = VN_IO_FAULT_FOP;
1137 args.args.fop_args.fp = fp;
1138 args.args.fop_args.doio = doio;
1139 args.cred = active_cred;
1140 args.flags = flags | FOF_OFFSET;
1141 if (uio->uio_rw == UIO_READ) {
1142 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1143 uio->uio_offset + uio->uio_resid);
1144 } else if ((fp->f_flag & O_APPEND) != 0 ||
1145 (flags & FOF_OFFSET) == 0) {
1146 /* For appenders, punt and lock the whole range. */
1147 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1149 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1150 uio->uio_offset + uio->uio_resid);
1152 error = vn_io_fault1(vp, uio, &args, td);
1153 vn_rangelock_unlock(vp, rl_cookie);
1155 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1157 foffset_unlock_uio(fp, uio, flags);
1162 * Helper function to perform the requested uiomove operation using
1163 * the held pages for io->uio_iov[0].iov_base buffer instead of
1164 * copyin/copyout. Access to the pages with uiomove_fromphys()
1165 * instead of iov_base prevents page faults that could occur due to
1166 * pmap_collect() invalidating the mapping created by
1167 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1168 * object cleanup revoking the write access from page mappings.
1170 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1171 * instead of plain uiomove().
1174 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1176 struct uio transp_uio;
1177 struct iovec transp_iov[1];
1183 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1184 uio->uio_segflg != UIO_USERSPACE)
1185 return (uiomove(data, xfersize, uio));
1187 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1188 transp_iov[0].iov_base = data;
1189 transp_uio.uio_iov = &transp_iov[0];
1190 transp_uio.uio_iovcnt = 1;
1191 if (xfersize > uio->uio_resid)
1192 xfersize = uio->uio_resid;
1193 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1194 transp_uio.uio_offset = 0;
1195 transp_uio.uio_segflg = UIO_SYSSPACE;
1197 * Since transp_iov points to data, and td_ma page array
1198 * corresponds to original uio->uio_iov, we need to invert the
1199 * direction of the i/o operation as passed to
1200 * uiomove_fromphys().
1202 switch (uio->uio_rw) {
1204 transp_uio.uio_rw = UIO_READ;
1207 transp_uio.uio_rw = UIO_WRITE;
1210 transp_uio.uio_td = uio->uio_td;
1211 error = uiomove_fromphys(td->td_ma,
1212 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1213 xfersize, &transp_uio);
1214 adv = xfersize - transp_uio.uio_resid;
1216 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1217 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1219 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1221 td->td_ma_cnt -= pgadv;
1222 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1223 uio->uio_iov->iov_len -= adv;
1224 uio->uio_resid -= adv;
1225 uio->uio_offset += adv;
1230 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1234 vm_offset_t iov_base;
1238 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1239 uio->uio_segflg != UIO_USERSPACE)
1240 return (uiomove_fromphys(ma, offset, xfersize, uio));
1242 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1243 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1244 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1245 switch (uio->uio_rw) {
1247 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1251 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1255 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1257 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1259 td->td_ma_cnt -= pgadv;
1260 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1261 uio->uio_iov->iov_len -= cnt;
1262 uio->uio_resid -= cnt;
1263 uio->uio_offset += cnt;
1269 * File table truncate routine.
1272 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1284 * Lock the whole range for truncation. Otherwise split i/o
1285 * might happen partly before and partly after the truncation.
1287 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1288 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1291 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1292 if (vp->v_type == VDIR) {
1297 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1301 error = vn_writechk(vp);
1304 vattr.va_size = length;
1305 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1309 vn_finished_write(mp);
1311 vn_rangelock_unlock(vp, rl_cookie);
1316 * File table vnode stat routine.
1319 vn_statfile(fp, sb, active_cred, td)
1322 struct ucred *active_cred;
1325 struct vnode *vp = fp->f_vnode;
1328 vn_lock(vp, LK_SHARED | LK_RETRY);
1329 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1336 * Stat a vnode; implementation for the stat syscall
1339 vn_stat(vp, sb, active_cred, file_cred, td)
1341 register struct stat *sb;
1342 struct ucred *active_cred;
1343 struct ucred *file_cred;
1347 register struct vattr *vap;
1352 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1360 * Initialize defaults for new and unusual fields, so that file
1361 * systems which don't support these fields don't need to know
1364 vap->va_birthtime.tv_sec = -1;
1365 vap->va_birthtime.tv_nsec = 0;
1366 vap->va_fsid = VNOVAL;
1367 vap->va_rdev = NODEV;
1369 error = VOP_GETATTR(vp, vap, active_cred);
1374 * Zero the spare stat fields
1376 bzero(sb, sizeof *sb);
1379 * Copy from vattr table
1381 if (vap->va_fsid != VNOVAL)
1382 sb->st_dev = vap->va_fsid;
1384 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1385 sb->st_ino = vap->va_fileid;
1386 mode = vap->va_mode;
1387 switch (vap->va_type) {
1413 sb->st_nlink = vap->va_nlink;
1414 sb->st_uid = vap->va_uid;
1415 sb->st_gid = vap->va_gid;
1416 sb->st_rdev = vap->va_rdev;
1417 if (vap->va_size > OFF_MAX)
1419 sb->st_size = vap->va_size;
1420 sb->st_atim = vap->va_atime;
1421 sb->st_mtim = vap->va_mtime;
1422 sb->st_ctim = vap->va_ctime;
1423 sb->st_birthtim = vap->va_birthtime;
1426 * According to www.opengroup.org, the meaning of st_blksize is
1427 * "a filesystem-specific preferred I/O block size for this
1428 * object. In some filesystem types, this may vary from file
1430 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1433 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1435 sb->st_flags = vap->va_flags;
1436 if (priv_check(td, PRIV_VFS_GENERATION))
1439 sb->st_gen = vap->va_gen;
1441 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1446 * File table vnode ioctl routine.
1449 vn_ioctl(fp, com, data, active_cred, td)
1453 struct ucred *active_cred;
1461 switch (vp->v_type) {
1466 vn_lock(vp, LK_SHARED | LK_RETRY);
1467 error = VOP_GETATTR(vp, &vattr, active_cred);
1470 *(int *)data = vattr.va_size - fp->f_offset;
1476 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1485 * File table vnode poll routine.
1488 vn_poll(fp, events, active_cred, td)
1491 struct ucred *active_cred;
1499 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1500 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1505 error = VOP_POLL(vp, events, fp->f_cred, td);
1510 * Acquire the requested lock and then check for validity. LK_RETRY
1511 * permits vn_lock to return doomed vnodes.
1514 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1518 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1519 ("vn_lock called with no locktype."));
1521 #ifdef DEBUG_VFS_LOCKS
1522 KASSERT(vp->v_holdcnt != 0,
1523 ("vn_lock %p: zero hold count", vp));
1525 error = VOP_LOCK1(vp, flags, file, line);
1526 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1527 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1528 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1531 * Callers specify LK_RETRY if they wish to get dead vnodes.
1532 * If RETRY is not set, we return ENOENT instead.
1534 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1535 (flags & LK_RETRY) == 0) {
1540 } while (flags & LK_RETRY && error != 0);
1545 * File table vnode close routine.
1548 vn_closefile(fp, td)
1557 fp->f_ops = &badfileops;
1559 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1562 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1564 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1565 lf.l_whence = SEEK_SET;
1568 lf.l_type = F_UNLCK;
1569 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1576 vn_suspendable_mp(struct mount *mp)
1579 return ((mp->mnt_kern_flag & MNTK_SUSPENDABLE) != 0);
1583 vn_suspendable(struct vnode *vp, struct mount **mpp)
1591 return (vn_suspendable_mp(*mpp));
1595 * Preparing to start a filesystem write operation. If the operation is
1596 * permitted, then we bump the count of operations in progress and
1597 * proceed. If a suspend request is in progress, we wait until the
1598 * suspension is over, and then proceed.
1601 vn_start_write_locked(struct mount *mp, int flags)
1605 mtx_assert(MNT_MTX(mp), MA_OWNED);
1609 * Check on status of suspension.
1611 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1612 mp->mnt_susp_owner != curthread) {
1613 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1614 if (flags & V_NOWAIT) {
1615 error = EWOULDBLOCK;
1618 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1619 (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1624 if (flags & V_XSLEEP)
1626 mp->mnt_writeopcount++;
1628 if (error != 0 || (flags & V_XSLEEP) != 0)
1635 vn_start_write(vp, mpp, flags)
1643 if (!vn_suspendable(vp, mpp))
1648 * If a vnode is provided, get and return the mount point that
1649 * to which it will write.
1652 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1654 if (error != EOPNOTSUPP)
1659 if ((mp = *mpp) == NULL)
1663 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1665 * As long as a vnode is not provided we need to acquire a
1666 * refcount for the provided mountpoint too, in order to
1667 * emulate a vfs_ref().
1673 return (vn_start_write_locked(mp, flags));
1677 * Secondary suspension. Used by operations such as vop_inactive
1678 * routines that are needed by the higher level functions. These
1679 * are allowed to proceed until all the higher level functions have
1680 * completed (indicated by mnt_writeopcount dropping to zero). At that
1681 * time, these operations are halted until the suspension is over.
1684 vn_start_secondary_write(vp, mpp, flags)
1692 if (!vn_suspendable(vp, mpp))
1697 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1699 if (error != EOPNOTSUPP)
1705 * If we are not suspended or have not yet reached suspended
1706 * mode, then let the operation proceed.
1708 if ((mp = *mpp) == NULL)
1712 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1714 * As long as a vnode is not provided we need to acquire a
1715 * refcount for the provided mountpoint too, in order to
1716 * emulate a vfs_ref().
1721 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1722 mp->mnt_secondary_writes++;
1723 mp->mnt_secondary_accwrites++;
1727 if (flags & V_NOWAIT) {
1730 return (EWOULDBLOCK);
1733 * Wait for the suspension to finish.
1735 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1736 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1744 * Filesystem write operation has completed. If we are suspending and this
1745 * operation is the last one, notify the suspender that the suspension is
1749 vn_finished_write(mp)
1752 if (mp == NULL || !vn_suspendable_mp(mp))
1756 mp->mnt_writeopcount--;
1757 if (mp->mnt_writeopcount < 0)
1758 panic("vn_finished_write: neg cnt");
1759 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1760 mp->mnt_writeopcount <= 0)
1761 wakeup(&mp->mnt_writeopcount);
1767 * Filesystem secondary write operation has completed. If we are
1768 * suspending and this operation is the last one, notify the suspender
1769 * that the suspension is now in effect.
1772 vn_finished_secondary_write(mp)
1775 if (mp == NULL || !vn_suspendable_mp(mp))
1779 mp->mnt_secondary_writes--;
1780 if (mp->mnt_secondary_writes < 0)
1781 panic("vn_finished_secondary_write: neg cnt");
1782 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1783 mp->mnt_secondary_writes <= 0)
1784 wakeup(&mp->mnt_secondary_writes);
1791 * Request a filesystem to suspend write operations.
1794 vfs_write_suspend(struct mount *mp, int flags)
1798 MPASS(vn_suspendable_mp(mp));
1801 if (mp->mnt_susp_owner == curthread) {
1805 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1806 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1809 * Unmount holds a write reference on the mount point. If we
1810 * own busy reference and drain for writers, we deadlock with
1811 * the reference draining in the unmount path. Callers of
1812 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1813 * vfs_busy() reference is owned and caller is not in the
1816 if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1817 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1822 mp->mnt_kern_flag |= MNTK_SUSPEND;
1823 mp->mnt_susp_owner = curthread;
1824 if (mp->mnt_writeopcount > 0)
1825 (void) msleep(&mp->mnt_writeopcount,
1826 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1829 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1830 vfs_write_resume(mp, 0);
1835 * Request a filesystem to resume write operations.
1838 vfs_write_resume(struct mount *mp, int flags)
1841 MPASS(vn_suspendable_mp(mp));
1844 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1845 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1846 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1848 mp->mnt_susp_owner = NULL;
1849 wakeup(&mp->mnt_writeopcount);
1850 wakeup(&mp->mnt_flag);
1851 curthread->td_pflags &= ~TDP_IGNSUSP;
1852 if ((flags & VR_START_WRITE) != 0) {
1854 mp->mnt_writeopcount++;
1857 if ((flags & VR_NO_SUSPCLR) == 0)
1859 } else if ((flags & VR_START_WRITE) != 0) {
1861 vn_start_write_locked(mp, 0);
1868 * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1872 vfs_write_suspend_umnt(struct mount *mp)
1876 MPASS(vn_suspendable_mp(mp));
1877 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1878 ("vfs_write_suspend_umnt: recursed"));
1880 /* dounmount() already called vn_start_write(). */
1882 vn_finished_write(mp);
1883 error = vfs_write_suspend(mp, 0);
1887 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1890 vn_start_write(NULL, &mp, V_WAIT);
1892 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1893 wakeup(&mp->mnt_flag);
1895 curthread->td_pflags |= TDP_IGNSUSP;
1900 * Implement kqueues for files by translating it to vnode operation.
1903 vn_kqfilter(struct file *fp, struct knote *kn)
1906 return (VOP_KQFILTER(fp->f_vnode, kn));
1910 * Simplified in-kernel wrapper calls for extended attribute access.
1911 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1912 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1915 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1916 const char *attrname, int *buflen, char *buf, struct thread *td)
1922 iov.iov_len = *buflen;
1925 auio.uio_iov = &iov;
1926 auio.uio_iovcnt = 1;
1927 auio.uio_rw = UIO_READ;
1928 auio.uio_segflg = UIO_SYSSPACE;
1930 auio.uio_offset = 0;
1931 auio.uio_resid = *buflen;
1933 if ((ioflg & IO_NODELOCKED) == 0)
1934 vn_lock(vp, LK_SHARED | LK_RETRY);
1936 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1938 /* authorize attribute retrieval as kernel */
1939 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1942 if ((ioflg & IO_NODELOCKED) == 0)
1946 *buflen = *buflen - auio.uio_resid;
1953 * XXX failure mode if partially written?
1956 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1957 const char *attrname, int buflen, char *buf, struct thread *td)
1964 iov.iov_len = buflen;
1967 auio.uio_iov = &iov;
1968 auio.uio_iovcnt = 1;
1969 auio.uio_rw = UIO_WRITE;
1970 auio.uio_segflg = UIO_SYSSPACE;
1972 auio.uio_offset = 0;
1973 auio.uio_resid = buflen;
1975 if ((ioflg & IO_NODELOCKED) == 0) {
1976 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1978 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1981 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1983 /* authorize attribute setting as kernel */
1984 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1986 if ((ioflg & IO_NODELOCKED) == 0) {
1987 vn_finished_write(mp);
1995 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1996 const char *attrname, struct thread *td)
2001 if ((ioflg & IO_NODELOCKED) == 0) {
2002 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2004 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2007 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2009 /* authorize attribute removal as kernel */
2010 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2011 if (error == EOPNOTSUPP)
2012 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2015 if ((ioflg & IO_NODELOCKED) == 0) {
2016 vn_finished_write(mp);
2024 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2028 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2032 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2035 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2040 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2041 int lkflags, struct vnode **rvp)
2046 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2048 ltype = VOP_ISLOCKED(vp);
2049 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2050 ("vn_vget_ino: vp not locked"));
2051 error = vfs_busy(mp, MBF_NOWAIT);
2055 error = vfs_busy(mp, 0);
2056 vn_lock(vp, ltype | LK_RETRY);
2060 if (vp->v_iflag & VI_DOOMED) {
2066 error = alloc(mp, alloc_arg, lkflags, rvp);
2069 vn_lock(vp, ltype | LK_RETRY);
2070 if (vp->v_iflag & VI_DOOMED) {
2083 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2084 const struct thread *td)
2087 if (vp->v_type != VREG || td == NULL)
2089 PROC_LOCK(td->td_proc);
2090 if ((uoff_t)uio->uio_offset + uio->uio_resid >
2091 lim_cur(td->td_proc, RLIMIT_FSIZE)) {
2092 kern_psignal(td->td_proc, SIGXFSZ);
2093 PROC_UNLOCK(td->td_proc);
2096 PROC_UNLOCK(td->td_proc);
2101 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2108 vn_lock(vp, LK_SHARED | LK_RETRY);
2109 AUDIT_ARG_VNODE1(vp);
2112 return (setfmode(td, active_cred, vp, mode));
2116 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2123 vn_lock(vp, LK_SHARED | LK_RETRY);
2124 AUDIT_ARG_VNODE1(vp);
2127 return (setfown(td, active_cred, vp, uid, gid));
2131 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2135 if ((object = vp->v_object) == NULL)
2137 VM_OBJECT_WLOCK(object);
2138 vm_object_page_remove(object, start, end, 0);
2139 VM_OBJECT_WUNLOCK(object);
2143 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2151 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2152 ("Wrong command %lu", cmd));
2154 if (vn_lock(vp, LK_SHARED) != 0)
2156 if (vp->v_type != VREG) {
2160 error = VOP_GETATTR(vp, &va, cred);
2164 if (noff >= va.va_size) {
2168 bsize = vp->v_mount->mnt_stat.f_iosize;
2169 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2170 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2171 if (error == EOPNOTSUPP) {
2175 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2176 (bnp != -1 && cmd == FIOSEEKDATA)) {
2183 if (noff > va.va_size)
2185 /* noff == va.va_size. There is an implicit hole at the end of file. */
2186 if (cmd == FIOSEEKDATA)
2196 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2201 off_t foffset, size;
2204 cred = td->td_ucred;
2206 foffset = foffset_lock(fp, 0);
2207 noneg = (vp->v_type != VCHR);
2213 (offset > 0 && foffset > OFF_MAX - offset))) {
2220 vn_lock(vp, LK_SHARED | LK_RETRY);
2221 error = VOP_GETATTR(vp, &vattr, cred);
2227 * If the file references a disk device, then fetch
2228 * the media size and use that to determine the ending
2231 if (vattr.va_size == 0 && vp->v_type == VCHR &&
2232 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2233 vattr.va_size = size;
2235 (vattr.va_size > OFF_MAX ||
2236 (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2240 offset += vattr.va_size;
2245 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2248 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2253 if (error == 0 && noneg && offset < 0)
2257 VFS_KNOTE_UNLOCKED(vp, 0);
2258 td->td_uretoff.tdu_off = offset;
2260 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2265 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2271 * Grant permission if the caller is the owner of the file, or
2272 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2273 * on the file. If the time pointer is null, then write
2274 * permission on the file is also sufficient.
2276 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2277 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2278 * will be allowed to set the times [..] to the current
2281 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2282 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2283 error = VOP_ACCESS(vp, VWRITE, cred, td);
2288 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2293 if (fp->f_type == DTYPE_FIFO)
2294 kif->kf_type = KF_TYPE_FIFO;
2296 kif->kf_type = KF_TYPE_VNODE;
2299 FILEDESC_SUNLOCK(fdp);
2300 error = vn_fill_kinfo_vnode(vp, kif);
2302 FILEDESC_SLOCK(fdp);
2307 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2310 char *fullpath, *freepath;
2313 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
2316 error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2318 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2320 if (freepath != NULL)
2321 free(freepath, M_TEMP);
2324 * Retrieve vnode attributes.
2326 va.va_fsid = VNOVAL;
2328 vn_lock(vp, LK_SHARED | LK_RETRY);
2329 error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2333 if (va.va_fsid != VNOVAL)
2334 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2336 kif->kf_un.kf_file.kf_file_fsid =
2337 vp->v_mount->mnt_stat.f_fsid.val[0];
2338 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2339 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2340 kif->kf_un.kf_file.kf_file_size = va.va_size;
2341 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;