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>
73 #include <security/audit/audit.h>
74 #include <security/mac/mac_framework.h>
77 #include <vm/vm_extern.h>
79 #include <vm/vm_map.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_page.h>
83 static fo_rdwr_t vn_read;
84 static fo_rdwr_t vn_write;
85 static fo_rdwr_t vn_io_fault;
86 static fo_truncate_t vn_truncate;
87 static fo_ioctl_t vn_ioctl;
88 static fo_poll_t vn_poll;
89 static fo_kqfilter_t vn_kqfilter;
90 static fo_stat_t vn_statfile;
91 static fo_close_t vn_closefile;
93 struct fileops vnops = {
94 .fo_read = vn_io_fault,
95 .fo_write = vn_io_fault,
96 .fo_truncate = vn_truncate,
99 .fo_kqfilter = vn_kqfilter,
100 .fo_stat = vn_statfile,
101 .fo_close = vn_closefile,
102 .fo_chmod = vn_chmod,
103 .fo_chown = vn_chown,
104 .fo_sendfile = vn_sendfile,
106 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
109 static const int io_hold_cnt = 16;
110 static int vn_io_fault_enable = 1;
111 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
112 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
113 static u_long vn_io_faults_cnt;
114 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
115 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
118 * Returns true if vn_io_fault mode of handling the i/o request should
122 do_vn_io_fault(struct vnode *vp, struct uio *uio)
126 return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
127 (mp = vp->v_mount) != NULL &&
128 (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
132 * Structure used to pass arguments to vn_io_fault1(), to do either
133 * file- or vnode-based I/O calls.
135 struct vn_io_fault_args {
143 struct fop_args_tag {
147 struct vop_args_tag {
153 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
154 struct vn_io_fault_args *args, struct thread *td);
157 vn_open(ndp, flagp, cmode, fp)
158 struct nameidata *ndp;
162 struct thread *td = ndp->ni_cnd.cn_thread;
164 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
168 * Common code for vnode open operations via a name lookup.
169 * Lookup the vnode and invoke VOP_CREATE if needed.
170 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
172 * Note that this does NOT free nameidata for the successful case,
173 * due to the NDINIT being done elsewhere.
176 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
177 struct ucred *cred, struct file *fp)
181 struct thread *td = ndp->ni_cnd.cn_thread;
183 struct vattr *vap = &vat;
188 if (fmode & O_CREAT) {
189 ndp->ni_cnd.cn_nameiop = CREATE;
190 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF;
191 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
192 ndp->ni_cnd.cn_flags |= FOLLOW;
193 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
194 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
195 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
196 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
198 if ((error = namei(ndp)) != 0)
200 if (ndp->ni_vp == NULL) {
203 vap->va_mode = cmode;
205 vap->va_vaflags |= VA_EXCLUSIVE;
206 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
207 NDFREE(ndp, NDF_ONLY_PNBUF);
209 if ((error = vn_start_write(NULL, &mp,
210 V_XSLEEP | PCATCH)) != 0)
215 error = mac_vnode_check_create(cred, ndp->ni_dvp,
219 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
222 vn_finished_write(mp);
224 NDFREE(ndp, NDF_ONLY_PNBUF);
230 if (ndp->ni_dvp == ndp->ni_vp)
236 if (fmode & O_EXCL) {
243 ndp->ni_cnd.cn_nameiop = LOOKUP;
244 ndp->ni_cnd.cn_flags = ISOPEN |
245 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
246 if (!(fmode & FWRITE))
247 ndp->ni_cnd.cn_flags |= LOCKSHARED;
248 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
249 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
250 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
251 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
252 if ((error = namei(ndp)) != 0)
256 error = vn_open_vnode(vp, fmode, cred, td, fp);
262 NDFREE(ndp, NDF_ONLY_PNBUF);
270 * Common code for vnode open operations once a vnode is located.
271 * Check permissions, and call the VOP_OPEN routine.
274 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
275 struct thread *td, struct file *fp)
280 int error, have_flock, lock_flags, type;
282 if (vp->v_type == VLNK)
284 if (vp->v_type == VSOCK)
286 if (vp->v_type != VDIR && fmode & O_DIRECTORY)
289 if (fmode & (FWRITE | O_TRUNC)) {
290 if (vp->v_type == VDIR)
298 if ((fmode & O_APPEND) && (fmode & FWRITE))
301 error = mac_vnode_check_open(cred, vp, accmode);
305 if ((fmode & O_CREAT) == 0) {
306 if (accmode & VWRITE) {
307 error = vn_writechk(vp);
312 error = VOP_ACCESS(vp, accmode, cred, td);
317 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
318 vn_lock(vp, LK_UPGRADE | LK_RETRY);
319 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
322 if (fmode & (O_EXLOCK | O_SHLOCK)) {
323 KASSERT(fp != NULL, ("open with flock requires fp"));
324 lock_flags = VOP_ISLOCKED(vp);
326 lf.l_whence = SEEK_SET;
329 if (fmode & O_EXLOCK)
334 if ((fmode & FNONBLOCK) == 0)
336 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
337 have_flock = (error == 0);
338 vn_lock(vp, lock_flags | LK_RETRY);
339 if (error == 0 && vp->v_iflag & VI_DOOMED)
342 * Another thread might have used this vnode as an
343 * executable while the vnode lock was dropped.
344 * Ensure the vnode is still able to be opened for
345 * writing after the lock has been obtained.
347 if (error == 0 && accmode & VWRITE)
348 error = vn_writechk(vp);
352 lf.l_whence = SEEK_SET;
356 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
359 vn_start_write(vp, &mp, V_WAIT);
360 vn_lock(vp, lock_flags | LK_RETRY);
361 (void)VOP_CLOSE(vp, fmode, cred, td);
362 vn_finished_write(mp);
363 /* Prevent second close from fdrop()->vn_close(). */
365 fp->f_ops= &badfileops;
368 fp->f_flag |= FHASLOCK;
370 if (fmode & FWRITE) {
371 VOP_ADD_WRITECOUNT(vp, 1);
372 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
373 __func__, vp, vp->v_writecount);
375 ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
380 * Check for write permissions on the specified vnode.
381 * Prototype text segments cannot be written.
385 register struct vnode *vp;
388 ASSERT_VOP_LOCKED(vp, "vn_writechk");
390 * If there's shared text associated with
391 * the vnode, try to free it up once. If
392 * we fail, we can't allow writing.
404 vn_close(vp, flags, file_cred, td)
405 register struct vnode *vp;
407 struct ucred *file_cred;
411 int error, lock_flags;
413 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
414 MNT_EXTENDED_SHARED(vp->v_mount))
415 lock_flags = LK_SHARED;
417 lock_flags = LK_EXCLUSIVE;
419 vn_start_write(vp, &mp, V_WAIT);
420 vn_lock(vp, lock_flags | LK_RETRY);
421 if (flags & FWRITE) {
422 VNASSERT(vp->v_writecount > 0, vp,
423 ("vn_close: negative writecount"));
424 VOP_ADD_WRITECOUNT(vp, -1);
425 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
426 __func__, vp, vp->v_writecount);
428 error = VOP_CLOSE(vp, flags, file_cred, td);
430 vn_finished_write(mp);
435 * Heuristic to detect sequential operation.
438 sequential_heuristic(struct uio *uio, struct file *fp)
441 if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
442 return (fp->f_seqcount << IO_SEQSHIFT);
445 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
446 * that the first I/O is normally considered to be slightly
447 * sequential. Seeking to offset 0 doesn't change sequentiality
448 * unless previous seeks have reduced f_seqcount to 0, in which
449 * case offset 0 is not special.
451 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
452 uio->uio_offset == fp->f_nextoff) {
454 * f_seqcount is in units of fixed-size blocks so that it
455 * depends mainly on the amount of sequential I/O and not
456 * much on the number of sequential I/O's. The fixed size
457 * of 16384 is hard-coded here since it is (not quite) just
458 * a magic size that works well here. This size is more
459 * closely related to the best I/O size for real disks than
460 * to any block size used by software.
462 fp->f_seqcount += howmany(uio->uio_resid, 16384);
463 if (fp->f_seqcount > IO_SEQMAX)
464 fp->f_seqcount = IO_SEQMAX;
465 return (fp->f_seqcount << IO_SEQSHIFT);
468 /* Not sequential. Quickly draw-down sequentiality. */
469 if (fp->f_seqcount > 1)
477 * Package up an I/O request on a vnode into a uio and do it.
480 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
481 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
482 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
489 struct vn_io_fault_args args;
490 int error, lock_flags;
492 auio.uio_iov = &aiov;
494 aiov.iov_base = base;
496 auio.uio_resid = len;
497 auio.uio_offset = offset;
498 auio.uio_segflg = segflg;
503 if ((ioflg & IO_NODELOCKED) == 0) {
504 if (rw == UIO_READ) {
505 rl_cookie = vn_rangelock_rlock(vp, offset,
508 rl_cookie = vn_rangelock_wlock(vp, offset,
512 if (rw == UIO_WRITE) {
513 if (vp->v_type != VCHR &&
514 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
517 if (MNT_SHARED_WRITES(mp) ||
518 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
519 lock_flags = LK_SHARED;
521 lock_flags = LK_EXCLUSIVE;
523 lock_flags = LK_SHARED;
524 vn_lock(vp, lock_flags | LK_RETRY);
528 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
530 if ((ioflg & IO_NOMACCHECK) == 0) {
532 error = mac_vnode_check_read(active_cred, file_cred,
535 error = mac_vnode_check_write(active_cred, file_cred,
540 if (file_cred != NULL)
544 if (do_vn_io_fault(vp, &auio)) {
545 args.kind = VN_IO_FAULT_VOP;
548 args.args.vop_args.vp = vp;
549 error = vn_io_fault1(vp, &auio, &args, td);
550 } else if (rw == UIO_READ) {
551 error = VOP_READ(vp, &auio, ioflg, cred);
552 } else /* if (rw == UIO_WRITE) */ {
553 error = VOP_WRITE(vp, &auio, ioflg, cred);
557 *aresid = auio.uio_resid;
559 if (auio.uio_resid && error == 0)
561 if ((ioflg & IO_NODELOCKED) == 0) {
564 vn_finished_write(mp);
567 if (rl_cookie != NULL)
568 vn_rangelock_unlock(vp, rl_cookie);
573 * Package up an I/O request on a vnode into a uio and do it. The I/O
574 * request is split up into smaller chunks and we try to avoid saturating
575 * the buffer cache while potentially holding a vnode locked, so we
576 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
577 * to give other processes a chance to lock the vnode (either other processes
578 * core'ing the same binary, or unrelated processes scanning the directory).
581 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
582 file_cred, aresid, td)
590 struct ucred *active_cred;
591 struct ucred *file_cred;
602 * Force `offset' to a multiple of MAXBSIZE except possibly
603 * for the first chunk, so that filesystems only need to
604 * write full blocks except possibly for the first and last
607 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
611 if (rw != UIO_READ && vp->v_type == VREG)
614 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
615 ioflg, active_cred, file_cred, &iaresid, td);
616 len -= chunk; /* aresid calc already includes length */
620 base = (char *)base + chunk;
621 kern_yield(PRI_USER);
624 *aresid = len + iaresid;
629 foffset_lock(struct file *fp, int flags)
634 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
636 #if OFF_MAX <= LONG_MAX
638 * Caller only wants the current f_offset value. Assume that
639 * the long and shorter integer types reads are atomic.
641 if ((flags & FOF_NOLOCK) != 0)
642 return (fp->f_offset);
646 * According to McKusick the vn lock was protecting f_offset here.
647 * It is now protected by the FOFFSET_LOCKED flag.
649 mtxp = mtx_pool_find(mtxpool_sleep, fp);
651 if ((flags & FOF_NOLOCK) == 0) {
652 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
653 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
654 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
657 fp->f_vnread_flags |= FOFFSET_LOCKED;
665 foffset_unlock(struct file *fp, off_t val, int flags)
669 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
671 #if OFF_MAX <= LONG_MAX
672 if ((flags & FOF_NOLOCK) != 0) {
673 if ((flags & FOF_NOUPDATE) == 0)
675 if ((flags & FOF_NEXTOFF) != 0)
681 mtxp = mtx_pool_find(mtxpool_sleep, fp);
683 if ((flags & FOF_NOUPDATE) == 0)
685 if ((flags & FOF_NEXTOFF) != 0)
687 if ((flags & FOF_NOLOCK) == 0) {
688 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
689 ("Lost FOFFSET_LOCKED"));
690 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
691 wakeup(&fp->f_vnread_flags);
692 fp->f_vnread_flags = 0;
698 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
701 if ((flags & FOF_OFFSET) == 0)
702 uio->uio_offset = foffset_lock(fp, flags);
706 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
709 if ((flags & FOF_OFFSET) == 0)
710 foffset_unlock(fp, uio->uio_offset, flags);
714 get_advice(struct file *fp, struct uio *uio)
719 ret = POSIX_FADV_NORMAL;
720 if (fp->f_advice == NULL)
723 mtxp = mtx_pool_find(mtxpool_sleep, fp);
725 if (uio->uio_offset >= fp->f_advice->fa_start &&
726 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
727 ret = fp->f_advice->fa_advice;
733 * File table vnode read routine.
736 vn_read(fp, uio, active_cred, flags, td)
739 struct ucred *active_cred;
747 off_t offset, start, end;
749 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
751 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
754 if (fp->f_flag & FNONBLOCK)
756 if (fp->f_flag & O_DIRECT)
758 advice = get_advice(fp, uio);
759 vn_lock(vp, LK_SHARED | LK_RETRY);
762 case POSIX_FADV_NORMAL:
763 case POSIX_FADV_SEQUENTIAL:
764 case POSIX_FADV_NOREUSE:
765 ioflag |= sequential_heuristic(uio, fp);
767 case POSIX_FADV_RANDOM:
768 /* Disable read-ahead for random I/O. */
771 offset = uio->uio_offset;
774 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
777 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
778 fp->f_nextoff = uio->uio_offset;
780 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
781 offset != uio->uio_offset) {
783 * Use POSIX_FADV_DONTNEED to flush clean pages and
784 * buffers for the backing file after a
785 * POSIX_FADV_NOREUSE read(2). To optimize the common
786 * case of using POSIX_FADV_NOREUSE with sequential
787 * access, track the previous implicit DONTNEED
788 * request and grow this request to include the
789 * current read(2) in addition to the previous
790 * DONTNEED. With purely sequential access this will
791 * cause the DONTNEED requests to continously grow to
792 * cover all of the previously read regions of the
793 * file. This allows filesystem blocks that are
794 * accessed by multiple calls to read(2) to be flushed
795 * once the last read(2) finishes.
798 end = uio->uio_offset - 1;
799 mtxp = mtx_pool_find(mtxpool_sleep, fp);
801 if (fp->f_advice != NULL &&
802 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
803 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
804 start = fp->f_advice->fa_prevstart;
805 else if (fp->f_advice->fa_prevstart != 0 &&
806 fp->f_advice->fa_prevstart == end + 1)
807 end = fp->f_advice->fa_prevend;
808 fp->f_advice->fa_prevstart = start;
809 fp->f_advice->fa_prevend = end;
812 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
818 * File table vnode write routine.
821 vn_write(fp, uio, active_cred, flags, td)
824 struct ucred *active_cred;
831 int error, ioflag, lock_flags;
833 off_t offset, start, end;
835 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
837 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
839 if (vp->v_type == VREG)
842 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
844 if (fp->f_flag & FNONBLOCK)
846 if (fp->f_flag & O_DIRECT)
848 if ((fp->f_flag & O_FSYNC) ||
849 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
852 if (vp->v_type != VCHR &&
853 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
856 advice = get_advice(fp, uio);
858 if (MNT_SHARED_WRITES(mp) ||
859 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
860 lock_flags = LK_SHARED;
862 lock_flags = LK_EXCLUSIVE;
865 vn_lock(vp, lock_flags | LK_RETRY);
867 case POSIX_FADV_NORMAL:
868 case POSIX_FADV_SEQUENTIAL:
869 case POSIX_FADV_NOREUSE:
870 ioflag |= sequential_heuristic(uio, fp);
872 case POSIX_FADV_RANDOM:
873 /* XXX: Is this correct? */
876 offset = uio->uio_offset;
879 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
882 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
883 fp->f_nextoff = uio->uio_offset;
885 if (vp->v_type != VCHR)
886 vn_finished_write(mp);
887 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
888 offset != uio->uio_offset) {
890 * Use POSIX_FADV_DONTNEED to flush clean pages and
891 * buffers for the backing file after a
892 * POSIX_FADV_NOREUSE write(2). To optimize the
893 * common case of using POSIX_FADV_NOREUSE with
894 * sequential access, track the previous implicit
895 * DONTNEED request and grow this request to include
896 * the current write(2) in addition to the previous
897 * DONTNEED. With purely sequential access this will
898 * cause the DONTNEED requests to continously grow to
899 * cover all of the previously written regions of the
902 * Note that the blocks just written are almost
903 * certainly still dirty, so this only works when
904 * VOP_ADVISE() calls from subsequent writes push out
905 * the data written by this write(2) once the backing
906 * buffers are clean. However, as compared to forcing
907 * IO_DIRECT, this gives much saner behavior. Write
908 * clustering is still allowed, and clean pages are
909 * merely moved to the cache page queue rather than
910 * outright thrown away. This means a subsequent
911 * read(2) can still avoid hitting the disk if the
912 * pages have not been reclaimed.
914 * This does make POSIX_FADV_NOREUSE largely useless
915 * with non-sequential access. However, sequential
916 * access is the more common use case and the flag is
920 end = uio->uio_offset - 1;
921 mtxp = mtx_pool_find(mtxpool_sleep, fp);
923 if (fp->f_advice != NULL &&
924 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
925 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
926 start = fp->f_advice->fa_prevstart;
927 else if (fp->f_advice->fa_prevstart != 0 &&
928 fp->f_advice->fa_prevstart == end + 1)
929 end = fp->f_advice->fa_prevend;
930 fp->f_advice->fa_prevstart = start;
931 fp->f_advice->fa_prevend = end;
934 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
942 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
943 * prevent the following deadlock:
945 * Assume that the thread A reads from the vnode vp1 into userspace
946 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
947 * currently not resident, then system ends up with the call chain
948 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
949 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
950 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
951 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
952 * backed by the pages of vnode vp1, and some page in buf2 is not
953 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
955 * To prevent the lock order reversal and deadlock, vn_io_fault() does
956 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
957 * Instead, it first tries to do the whole range i/o with pagefaults
958 * disabled. If all pages in the i/o buffer are resident and mapped,
959 * VOP will succeed (ignoring the genuine filesystem errors).
960 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
961 * i/o in chunks, with all pages in the chunk prefaulted and held
962 * using vm_fault_quick_hold_pages().
964 * Filesystems using this deadlock avoidance scheme should use the
965 * array of the held pages from uio, saved in the curthread->td_ma,
966 * instead of doing uiomove(). A helper function
967 * vn_io_fault_uiomove() converts uiomove request into
968 * uiomove_fromphys() over td_ma array.
970 * Since vnode locks do not cover the whole i/o anymore, rangelocks
971 * make the current i/o request atomic with respect to other i/os and
976 * Decode vn_io_fault_args and perform the corresponding i/o.
979 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
983 switch (args->kind) {
984 case VN_IO_FAULT_FOP:
985 return ((args->args.fop_args.doio)(args->args.fop_args.fp,
986 uio, args->cred, args->flags, td));
987 case VN_IO_FAULT_VOP:
988 if (uio->uio_rw == UIO_READ) {
989 return (VOP_READ(args->args.vop_args.vp, uio,
990 args->flags, args->cred));
991 } else if (uio->uio_rw == UIO_WRITE) {
992 return (VOP_WRITE(args->args.vop_args.vp, uio,
993 args->flags, args->cred));
997 panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
1002 * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1003 * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1004 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1005 * into args and call vn_io_fault1() to handle faults during the user
1006 * mode buffer accesses.
1009 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1012 vm_page_t ma[io_hold_cnt + 2];
1013 struct uio *uio_clone, short_uio;
1014 struct iovec short_iovec[1];
1015 vm_page_t *prev_td_ma;
1017 vm_offset_t addr, end;
1020 int error, cnt, save, saveheld, prev_td_ma_cnt;
1022 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1025 * The UFS follows IO_UNIT directive and replays back both
1026 * uio_offset and uio_resid if an error is encountered during the
1027 * operation. But, since the iovec may be already advanced,
1028 * uio is still in an inconsistent state.
1030 * Cache a copy of the original uio, which is advanced to the redo
1031 * point using UIO_NOCOPY below.
1033 uio_clone = cloneuio(uio);
1034 resid = uio->uio_resid;
1036 short_uio.uio_segflg = UIO_USERSPACE;
1037 short_uio.uio_rw = uio->uio_rw;
1038 short_uio.uio_td = uio->uio_td;
1040 save = vm_fault_disable_pagefaults();
1041 error = vn_io_fault_doio(args, uio, td);
1042 if (error != EFAULT)
1045 atomic_add_long(&vn_io_faults_cnt, 1);
1046 uio_clone->uio_segflg = UIO_NOCOPY;
1047 uiomove(NULL, resid - uio->uio_resid, uio_clone);
1048 uio_clone->uio_segflg = uio->uio_segflg;
1050 saveheld = curthread_pflags_set(TDP_UIOHELD);
1051 prev_td_ma = td->td_ma;
1052 prev_td_ma_cnt = td->td_ma_cnt;
1054 while (uio_clone->uio_resid != 0) {
1055 len = uio_clone->uio_iov->iov_len;
1057 KASSERT(uio_clone->uio_iovcnt >= 1,
1058 ("iovcnt underflow"));
1059 uio_clone->uio_iov++;
1060 uio_clone->uio_iovcnt--;
1063 if (len > io_hold_cnt * PAGE_SIZE)
1064 len = io_hold_cnt * PAGE_SIZE;
1065 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1066 end = round_page(addr + len);
1071 cnt = atop(end - trunc_page(addr));
1073 * A perfectly misaligned address and length could cause
1074 * both the start and the end of the chunk to use partial
1075 * page. +2 accounts for such a situation.
1077 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1078 addr, len, prot, ma, io_hold_cnt + 2);
1083 short_uio.uio_iov = &short_iovec[0];
1084 short_iovec[0].iov_base = (void *)addr;
1085 short_uio.uio_iovcnt = 1;
1086 short_uio.uio_resid = short_iovec[0].iov_len = len;
1087 short_uio.uio_offset = uio_clone->uio_offset;
1089 td->td_ma_cnt = cnt;
1091 error = vn_io_fault_doio(args, &short_uio, td);
1092 vm_page_unhold_pages(ma, cnt);
1093 adv = len - short_uio.uio_resid;
1095 uio_clone->uio_iov->iov_base =
1096 (char *)uio_clone->uio_iov->iov_base + adv;
1097 uio_clone->uio_iov->iov_len -= adv;
1098 uio_clone->uio_resid -= adv;
1099 uio_clone->uio_offset += adv;
1101 uio->uio_resid -= adv;
1102 uio->uio_offset += adv;
1104 if (error != 0 || adv == 0)
1107 td->td_ma = prev_td_ma;
1108 td->td_ma_cnt = prev_td_ma_cnt;
1109 curthread_pflags_restore(saveheld);
1111 vm_fault_enable_pagefaults(save);
1112 free(uio_clone, M_IOV);
1117 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1118 int flags, struct thread *td)
1123 struct vn_io_fault_args args;
1126 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1128 foffset_lock_uio(fp, uio, flags);
1129 if (do_vn_io_fault(vp, uio)) {
1130 args.kind = VN_IO_FAULT_FOP;
1131 args.args.fop_args.fp = fp;
1132 args.args.fop_args.doio = doio;
1133 args.cred = active_cred;
1134 args.flags = flags | FOF_OFFSET;
1135 if (uio->uio_rw == UIO_READ) {
1136 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1137 uio->uio_offset + uio->uio_resid);
1138 } else if ((fp->f_flag & O_APPEND) != 0 ||
1139 (flags & FOF_OFFSET) == 0) {
1140 /* For appenders, punt and lock the whole range. */
1141 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1143 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1144 uio->uio_offset + uio->uio_resid);
1146 error = vn_io_fault1(vp, uio, &args, td);
1147 vn_rangelock_unlock(vp, rl_cookie);
1149 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1151 foffset_unlock_uio(fp, uio, flags);
1156 * Helper function to perform the requested uiomove operation using
1157 * the held pages for io->uio_iov[0].iov_base buffer instead of
1158 * copyin/copyout. Access to the pages with uiomove_fromphys()
1159 * instead of iov_base prevents page faults that could occur due to
1160 * pmap_collect() invalidating the mapping created by
1161 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1162 * object cleanup revoking the write access from page mappings.
1164 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1165 * instead of plain uiomove().
1168 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1170 struct uio transp_uio;
1171 struct iovec transp_iov[1];
1177 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1178 uio->uio_segflg != UIO_USERSPACE)
1179 return (uiomove(data, xfersize, uio));
1181 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1182 transp_iov[0].iov_base = data;
1183 transp_uio.uio_iov = &transp_iov[0];
1184 transp_uio.uio_iovcnt = 1;
1185 if (xfersize > uio->uio_resid)
1186 xfersize = uio->uio_resid;
1187 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1188 transp_uio.uio_offset = 0;
1189 transp_uio.uio_segflg = UIO_SYSSPACE;
1191 * Since transp_iov points to data, and td_ma page array
1192 * corresponds to original uio->uio_iov, we need to invert the
1193 * direction of the i/o operation as passed to
1194 * uiomove_fromphys().
1196 switch (uio->uio_rw) {
1198 transp_uio.uio_rw = UIO_READ;
1201 transp_uio.uio_rw = UIO_WRITE;
1204 transp_uio.uio_td = uio->uio_td;
1205 error = uiomove_fromphys(td->td_ma,
1206 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1207 xfersize, &transp_uio);
1208 adv = xfersize - transp_uio.uio_resid;
1210 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1211 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1213 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1215 td->td_ma_cnt -= pgadv;
1216 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1217 uio->uio_iov->iov_len -= adv;
1218 uio->uio_resid -= adv;
1219 uio->uio_offset += adv;
1224 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1228 vm_offset_t iov_base;
1232 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1233 uio->uio_segflg != UIO_USERSPACE)
1234 return (uiomove_fromphys(ma, offset, xfersize, uio));
1236 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1237 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1238 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1239 switch (uio->uio_rw) {
1241 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1245 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1249 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1251 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1253 td->td_ma_cnt -= pgadv;
1254 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1255 uio->uio_iov->iov_len -= cnt;
1256 uio->uio_resid -= cnt;
1257 uio->uio_offset += cnt;
1263 * File table truncate routine.
1266 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1278 * Lock the whole range for truncation. Otherwise split i/o
1279 * might happen partly before and partly after the truncation.
1281 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1282 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1285 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1286 if (vp->v_type == VDIR) {
1291 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1295 error = vn_writechk(vp);
1298 vattr.va_size = length;
1299 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1303 vn_finished_write(mp);
1305 vn_rangelock_unlock(vp, rl_cookie);
1310 * File table vnode stat routine.
1313 vn_statfile(fp, sb, active_cred, td)
1316 struct ucred *active_cred;
1319 struct vnode *vp = fp->f_vnode;
1322 vn_lock(vp, LK_SHARED | LK_RETRY);
1323 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1330 * Stat a vnode; implementation for the stat syscall
1333 vn_stat(vp, sb, active_cred, file_cred, td)
1335 register struct stat *sb;
1336 struct ucred *active_cred;
1337 struct ucred *file_cred;
1341 register struct vattr *vap;
1346 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1354 * Initialize defaults for new and unusual fields, so that file
1355 * systems which don't support these fields don't need to know
1358 vap->va_birthtime.tv_sec = -1;
1359 vap->va_birthtime.tv_nsec = 0;
1360 vap->va_fsid = VNOVAL;
1361 vap->va_rdev = NODEV;
1363 error = VOP_GETATTR(vp, vap, active_cred);
1368 * Zero the spare stat fields
1370 bzero(sb, sizeof *sb);
1373 * Copy from vattr table
1375 if (vap->va_fsid != VNOVAL)
1376 sb->st_dev = vap->va_fsid;
1378 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1379 sb->st_ino = vap->va_fileid;
1380 mode = vap->va_mode;
1381 switch (vap->va_type) {
1407 sb->st_nlink = vap->va_nlink;
1408 sb->st_uid = vap->va_uid;
1409 sb->st_gid = vap->va_gid;
1410 sb->st_rdev = vap->va_rdev;
1411 if (vap->va_size > OFF_MAX)
1413 sb->st_size = vap->va_size;
1414 sb->st_atim = vap->va_atime;
1415 sb->st_mtim = vap->va_mtime;
1416 sb->st_ctim = vap->va_ctime;
1417 sb->st_birthtim = vap->va_birthtime;
1420 * According to www.opengroup.org, the meaning of st_blksize is
1421 * "a filesystem-specific preferred I/O block size for this
1422 * object. In some filesystem types, this may vary from file
1424 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1427 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1429 sb->st_flags = vap->va_flags;
1430 if (priv_check(td, PRIV_VFS_GENERATION))
1433 sb->st_gen = vap->va_gen;
1435 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1440 * File table vnode ioctl routine.
1443 vn_ioctl(fp, com, data, active_cred, td)
1447 struct ucred *active_cred;
1455 switch (vp->v_type) {
1460 vn_lock(vp, LK_SHARED | LK_RETRY);
1461 error = VOP_GETATTR(vp, &vattr, active_cred);
1464 *(int *)data = vattr.va_size - fp->f_offset;
1470 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1479 * File table vnode poll routine.
1482 vn_poll(fp, events, active_cred, td)
1485 struct ucred *active_cred;
1493 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1494 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1499 error = VOP_POLL(vp, events, fp->f_cred, td);
1504 * Acquire the requested lock and then check for validity. LK_RETRY
1505 * permits vn_lock to return doomed vnodes.
1508 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1512 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1513 ("vn_lock called with no locktype."));
1515 #ifdef DEBUG_VFS_LOCKS
1516 KASSERT(vp->v_holdcnt != 0,
1517 ("vn_lock %p: zero hold count", vp));
1519 error = VOP_LOCK1(vp, flags, file, line);
1520 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1521 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1522 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1525 * Callers specify LK_RETRY if they wish to get dead vnodes.
1526 * If RETRY is not set, we return ENOENT instead.
1528 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1529 (flags & LK_RETRY) == 0) {
1534 } while (flags & LK_RETRY && error != 0);
1539 * File table vnode close routine.
1542 vn_closefile(fp, td)
1551 fp->f_ops = &badfileops;
1553 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1556 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1558 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1559 lf.l_whence = SEEK_SET;
1562 lf.l_type = F_UNLCK;
1563 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1570 * Preparing to start a filesystem write operation. If the operation is
1571 * permitted, then we bump the count of operations in progress and
1572 * proceed. If a suspend request is in progress, we wait until the
1573 * suspension is over, and then proceed.
1576 vn_start_write_locked(struct mount *mp, int flags)
1580 mtx_assert(MNT_MTX(mp), MA_OWNED);
1584 * Check on status of suspension.
1586 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1587 mp->mnt_susp_owner != curthread) {
1588 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1589 if (flags & V_NOWAIT) {
1590 error = EWOULDBLOCK;
1593 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1594 (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1599 if (flags & V_XSLEEP)
1601 mp->mnt_writeopcount++;
1603 if (error != 0 || (flags & V_XSLEEP) != 0)
1610 vn_start_write(vp, mpp, flags)
1620 * If a vnode is provided, get and return the mount point that
1621 * to which it will write.
1624 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1626 if (error != EOPNOTSUPP)
1631 if ((mp = *mpp) == NULL)
1635 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1637 * As long as a vnode is not provided we need to acquire a
1638 * refcount for the provided mountpoint too, in order to
1639 * emulate a vfs_ref().
1645 return (vn_start_write_locked(mp, flags));
1649 * Secondary suspension. Used by operations such as vop_inactive
1650 * routines that are needed by the higher level functions. These
1651 * are allowed to proceed until all the higher level functions have
1652 * completed (indicated by mnt_writeopcount dropping to zero). At that
1653 * time, these operations are halted until the suspension is over.
1656 vn_start_secondary_write(vp, mpp, flags)
1666 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1668 if (error != EOPNOTSUPP)
1674 * If we are not suspended or have not yet reached suspended
1675 * mode, then let the operation proceed.
1677 if ((mp = *mpp) == NULL)
1681 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1683 * As long as a vnode is not provided we need to acquire a
1684 * refcount for the provided mountpoint too, in order to
1685 * emulate a vfs_ref().
1690 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1691 mp->mnt_secondary_writes++;
1692 mp->mnt_secondary_accwrites++;
1696 if (flags & V_NOWAIT) {
1699 return (EWOULDBLOCK);
1702 * Wait for the suspension to finish.
1704 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1705 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1713 * Filesystem write operation has completed. If we are suspending and this
1714 * operation is the last one, notify the suspender that the suspension is
1718 vn_finished_write(mp)
1725 mp->mnt_writeopcount--;
1726 if (mp->mnt_writeopcount < 0)
1727 panic("vn_finished_write: neg cnt");
1728 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1729 mp->mnt_writeopcount <= 0)
1730 wakeup(&mp->mnt_writeopcount);
1736 * Filesystem secondary write operation has completed. If we are
1737 * suspending and this operation is the last one, notify the suspender
1738 * that the suspension is now in effect.
1741 vn_finished_secondary_write(mp)
1748 mp->mnt_secondary_writes--;
1749 if (mp->mnt_secondary_writes < 0)
1750 panic("vn_finished_secondary_write: neg cnt");
1751 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1752 mp->mnt_secondary_writes <= 0)
1753 wakeup(&mp->mnt_secondary_writes);
1760 * Request a filesystem to suspend write operations.
1763 vfs_write_suspend(struct mount *mp, int flags)
1768 if (mp->mnt_susp_owner == curthread) {
1772 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1773 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1776 * Unmount holds a write reference on the mount point. If we
1777 * own busy reference and drain for writers, we deadlock with
1778 * the reference draining in the unmount path. Callers of
1779 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1780 * vfs_busy() reference is owned and caller is not in the
1783 if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1784 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1789 mp->mnt_kern_flag |= MNTK_SUSPEND;
1790 mp->mnt_susp_owner = curthread;
1791 if (mp->mnt_writeopcount > 0)
1792 (void) msleep(&mp->mnt_writeopcount,
1793 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1796 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1797 vfs_write_resume(mp, 0);
1802 * Request a filesystem to resume write operations.
1805 vfs_write_resume(struct mount *mp, int flags)
1809 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1810 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1811 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1813 mp->mnt_susp_owner = NULL;
1814 wakeup(&mp->mnt_writeopcount);
1815 wakeup(&mp->mnt_flag);
1816 curthread->td_pflags &= ~TDP_IGNSUSP;
1817 if ((flags & VR_START_WRITE) != 0) {
1819 mp->mnt_writeopcount++;
1822 if ((flags & VR_NO_SUSPCLR) == 0)
1824 } else if ((flags & VR_START_WRITE) != 0) {
1826 vn_start_write_locked(mp, 0);
1833 * Implement kqueues for files by translating it to vnode operation.
1836 vn_kqfilter(struct file *fp, struct knote *kn)
1839 return (VOP_KQFILTER(fp->f_vnode, kn));
1843 * Simplified in-kernel wrapper calls for extended attribute access.
1844 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1845 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1848 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1849 const char *attrname, int *buflen, char *buf, struct thread *td)
1855 iov.iov_len = *buflen;
1858 auio.uio_iov = &iov;
1859 auio.uio_iovcnt = 1;
1860 auio.uio_rw = UIO_READ;
1861 auio.uio_segflg = UIO_SYSSPACE;
1863 auio.uio_offset = 0;
1864 auio.uio_resid = *buflen;
1866 if ((ioflg & IO_NODELOCKED) == 0)
1867 vn_lock(vp, LK_SHARED | LK_RETRY);
1869 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1871 /* authorize attribute retrieval as kernel */
1872 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1875 if ((ioflg & IO_NODELOCKED) == 0)
1879 *buflen = *buflen - auio.uio_resid;
1886 * XXX failure mode if partially written?
1889 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1890 const char *attrname, int buflen, char *buf, struct thread *td)
1897 iov.iov_len = buflen;
1900 auio.uio_iov = &iov;
1901 auio.uio_iovcnt = 1;
1902 auio.uio_rw = UIO_WRITE;
1903 auio.uio_segflg = UIO_SYSSPACE;
1905 auio.uio_offset = 0;
1906 auio.uio_resid = buflen;
1908 if ((ioflg & IO_NODELOCKED) == 0) {
1909 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1911 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1914 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1916 /* authorize attribute setting as kernel */
1917 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1919 if ((ioflg & IO_NODELOCKED) == 0) {
1920 vn_finished_write(mp);
1928 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1929 const char *attrname, struct thread *td)
1934 if ((ioflg & IO_NODELOCKED) == 0) {
1935 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1937 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1940 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1942 /* authorize attribute removal as kernel */
1943 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1944 if (error == EOPNOTSUPP)
1945 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1948 if ((ioflg & IO_NODELOCKED) == 0) {
1949 vn_finished_write(mp);
1957 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1963 ltype = VOP_ISLOCKED(vp);
1964 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1965 ("vn_vget_ino: vp not locked"));
1966 error = vfs_busy(mp, MBF_NOWAIT);
1970 error = vfs_busy(mp, 0);
1971 vn_lock(vp, ltype | LK_RETRY);
1975 if (vp->v_iflag & VI_DOOMED) {
1981 error = VFS_VGET(mp, ino, lkflags, rvp);
1983 vn_lock(vp, ltype | LK_RETRY);
1984 if (vp->v_iflag & VI_DOOMED) {
1993 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
1994 const struct thread *td)
1997 if (vp->v_type != VREG || td == NULL)
1999 PROC_LOCK(td->td_proc);
2000 if ((uoff_t)uio->uio_offset + uio->uio_resid >
2001 lim_cur(td->td_proc, RLIMIT_FSIZE)) {
2002 kern_psignal(td->td_proc, SIGXFSZ);
2003 PROC_UNLOCK(td->td_proc);
2006 PROC_UNLOCK(td->td_proc);
2011 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2018 vn_lock(vp, LK_SHARED | LK_RETRY);
2019 AUDIT_ARG_VNODE1(vp);
2022 return (setfmode(td, active_cred, vp, mode));
2026 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2033 vn_lock(vp, LK_SHARED | LK_RETRY);
2034 AUDIT_ARG_VNODE1(vp);
2037 return (setfown(td, active_cred, vp, uid, gid));
2041 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2045 if ((object = vp->v_object) == NULL)
2047 VM_OBJECT_WLOCK(object);
2048 vm_object_page_remove(object, start, end, 0);
2049 VM_OBJECT_WUNLOCK(object);
2053 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2061 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2062 ("Wrong command %lu", cmd));
2064 if (vn_lock(vp, LK_SHARED) != 0)
2066 if (vp->v_type != VREG) {
2070 error = VOP_GETATTR(vp, &va, cred);
2074 if (noff >= va.va_size) {
2078 bsize = vp->v_mount->mnt_stat.f_iosize;
2079 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2080 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2081 if (error == EOPNOTSUPP) {
2085 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2086 (bnp != -1 && cmd == FIOSEEKDATA)) {
2093 if (noff > va.va_size)
2095 /* noff == va.va_size. There is an implicit hole at the end of file. */
2096 if (cmd == FIOSEEKDATA)
2106 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2111 off_t foffset, size;
2114 cred = td->td_ucred;
2116 foffset = foffset_lock(fp, 0);
2117 noneg = (vp->v_type != VCHR);
2123 (offset > 0 && foffset > OFF_MAX - offset))) {
2130 vn_lock(vp, LK_SHARED | LK_RETRY);
2131 error = VOP_GETATTR(vp, &vattr, cred);
2137 * If the file references a disk device, then fetch
2138 * the media size and use that to determine the ending
2141 if (vattr.va_size == 0 && vp->v_type == VCHR &&
2142 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2143 vattr.va_size = size;
2145 (vattr.va_size > OFF_MAX ||
2146 (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2150 offset += vattr.va_size;
2155 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2158 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2163 if (error == 0 && noneg && offset < 0)
2167 VFS_KNOTE_UNLOCKED(vp, 0);
2168 *(off_t *)(td->td_retval) = offset;
2170 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2175 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2180 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2184 * Grant permission if the caller is the owner of the file or
2185 * the super-user. If the time pointer is null, then write
2186 * permission on the file is also sufficient.
2188 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2189 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2190 * will be allowed to set the times [..] to the current
2193 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2194 error = VOP_ACCESS(vp, VWRITE, cred, td);