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 ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
442 if (fp->f_flag & FRDAHEAD)
443 return (fp->f_seqcount << IO_SEQSHIFT);
446 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
447 * that the first I/O is normally considered to be slightly
448 * sequential. Seeking to offset 0 doesn't change sequentiality
449 * unless previous seeks have reduced f_seqcount to 0, in which
450 * case offset 0 is not special.
452 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
453 uio->uio_offset == fp->f_nextoff) {
455 * f_seqcount is in units of fixed-size blocks so that it
456 * depends mainly on the amount of sequential I/O and not
457 * much on the number of sequential I/O's. The fixed size
458 * of 16384 is hard-coded here since it is (not quite) just
459 * a magic size that works well here. This size is more
460 * closely related to the best I/O size for real disks than
461 * to any block size used by software.
463 fp->f_seqcount += howmany(uio->uio_resid, 16384);
464 if (fp->f_seqcount > IO_SEQMAX)
465 fp->f_seqcount = IO_SEQMAX;
466 return (fp->f_seqcount << IO_SEQSHIFT);
469 /* Not sequential. Quickly draw-down sequentiality. */
470 if (fp->f_seqcount > 1)
478 * Package up an I/O request on a vnode into a uio and do it.
481 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
482 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
483 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
490 struct vn_io_fault_args args;
491 int error, lock_flags;
493 auio.uio_iov = &aiov;
495 aiov.iov_base = base;
497 auio.uio_resid = len;
498 auio.uio_offset = offset;
499 auio.uio_segflg = segflg;
504 if ((ioflg & IO_NODELOCKED) == 0) {
505 if ((ioflg & IO_RANGELOCKED) == 0) {
506 if (rw == UIO_READ) {
507 rl_cookie = vn_rangelock_rlock(vp, offset,
510 rl_cookie = vn_rangelock_wlock(vp, offset,
516 if (rw == UIO_WRITE) {
517 if (vp->v_type != VCHR &&
518 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
521 if (MNT_SHARED_WRITES(mp) ||
522 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
523 lock_flags = LK_SHARED;
525 lock_flags = LK_EXCLUSIVE;
527 lock_flags = LK_SHARED;
528 vn_lock(vp, lock_flags | LK_RETRY);
532 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
534 if ((ioflg & IO_NOMACCHECK) == 0) {
536 error = mac_vnode_check_read(active_cred, file_cred,
539 error = mac_vnode_check_write(active_cred, file_cred,
544 if (file_cred != NULL)
548 if (do_vn_io_fault(vp, &auio)) {
549 args.kind = VN_IO_FAULT_VOP;
552 args.args.vop_args.vp = vp;
553 error = vn_io_fault1(vp, &auio, &args, td);
554 } else if (rw == UIO_READ) {
555 error = VOP_READ(vp, &auio, ioflg, cred);
556 } else /* if (rw == UIO_WRITE) */ {
557 error = VOP_WRITE(vp, &auio, ioflg, cred);
561 *aresid = auio.uio_resid;
563 if (auio.uio_resid && error == 0)
565 if ((ioflg & IO_NODELOCKED) == 0) {
568 vn_finished_write(mp);
571 if (rl_cookie != NULL)
572 vn_rangelock_unlock(vp, rl_cookie);
577 * Package up an I/O request on a vnode into a uio and do it. The I/O
578 * request is split up into smaller chunks and we try to avoid saturating
579 * the buffer cache while potentially holding a vnode locked, so we
580 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
581 * to give other processes a chance to lock the vnode (either other processes
582 * core'ing the same binary, or unrelated processes scanning the directory).
585 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
586 file_cred, aresid, td)
594 struct ucred *active_cred;
595 struct ucred *file_cred;
606 * Force `offset' to a multiple of MAXBSIZE except possibly
607 * for the first chunk, so that filesystems only need to
608 * write full blocks except possibly for the first and last
611 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
615 if (rw != UIO_READ && vp->v_type == VREG)
618 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
619 ioflg, active_cred, file_cred, &iaresid, td);
620 len -= chunk; /* aresid calc already includes length */
624 base = (char *)base + chunk;
625 kern_yield(PRI_USER);
628 *aresid = len + iaresid;
633 foffset_lock(struct file *fp, int flags)
638 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
640 #if OFF_MAX <= LONG_MAX
642 * Caller only wants the current f_offset value. Assume that
643 * the long and shorter integer types reads are atomic.
645 if ((flags & FOF_NOLOCK) != 0)
646 return (fp->f_offset);
650 * According to McKusick the vn lock was protecting f_offset here.
651 * It is now protected by the FOFFSET_LOCKED flag.
653 mtxp = mtx_pool_find(mtxpool_sleep, fp);
655 if ((flags & FOF_NOLOCK) == 0) {
656 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
657 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
658 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
661 fp->f_vnread_flags |= FOFFSET_LOCKED;
669 foffset_unlock(struct file *fp, off_t val, int flags)
673 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
675 #if OFF_MAX <= LONG_MAX
676 if ((flags & FOF_NOLOCK) != 0) {
677 if ((flags & FOF_NOUPDATE) == 0)
679 if ((flags & FOF_NEXTOFF) != 0)
685 mtxp = mtx_pool_find(mtxpool_sleep, fp);
687 if ((flags & FOF_NOUPDATE) == 0)
689 if ((flags & FOF_NEXTOFF) != 0)
691 if ((flags & FOF_NOLOCK) == 0) {
692 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
693 ("Lost FOFFSET_LOCKED"));
694 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
695 wakeup(&fp->f_vnread_flags);
696 fp->f_vnread_flags = 0;
702 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
705 if ((flags & FOF_OFFSET) == 0)
706 uio->uio_offset = foffset_lock(fp, flags);
710 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
713 if ((flags & FOF_OFFSET) == 0)
714 foffset_unlock(fp, uio->uio_offset, flags);
718 get_advice(struct file *fp, struct uio *uio)
723 ret = POSIX_FADV_NORMAL;
724 if (fp->f_advice == NULL)
727 mtxp = mtx_pool_find(mtxpool_sleep, fp);
729 if (uio->uio_offset >= fp->f_advice->fa_start &&
730 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
731 ret = fp->f_advice->fa_advice;
737 * File table vnode read routine.
740 vn_read(fp, uio, active_cred, flags, td)
743 struct ucred *active_cred;
751 off_t offset, start, end;
753 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
755 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
758 if (fp->f_flag & FNONBLOCK)
760 if (fp->f_flag & O_DIRECT)
762 advice = get_advice(fp, uio);
763 vn_lock(vp, LK_SHARED | LK_RETRY);
766 case POSIX_FADV_NORMAL:
767 case POSIX_FADV_SEQUENTIAL:
768 case POSIX_FADV_NOREUSE:
769 ioflag |= sequential_heuristic(uio, fp);
771 case POSIX_FADV_RANDOM:
772 /* Disable read-ahead for random I/O. */
775 offset = uio->uio_offset;
778 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
781 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
782 fp->f_nextoff = uio->uio_offset;
784 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
785 offset != uio->uio_offset) {
787 * Use POSIX_FADV_DONTNEED to flush clean pages and
788 * buffers for the backing file after a
789 * POSIX_FADV_NOREUSE read(2). To optimize the common
790 * case of using POSIX_FADV_NOREUSE with sequential
791 * access, track the previous implicit DONTNEED
792 * request and grow this request to include the
793 * current read(2) in addition to the previous
794 * DONTNEED. With purely sequential access this will
795 * cause the DONTNEED requests to continously grow to
796 * cover all of the previously read regions of the
797 * file. This allows filesystem blocks that are
798 * accessed by multiple calls to read(2) to be flushed
799 * once the last read(2) finishes.
802 end = uio->uio_offset - 1;
803 mtxp = mtx_pool_find(mtxpool_sleep, fp);
805 if (fp->f_advice != NULL &&
806 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
807 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
808 start = fp->f_advice->fa_prevstart;
809 else if (fp->f_advice->fa_prevstart != 0 &&
810 fp->f_advice->fa_prevstart == end + 1)
811 end = fp->f_advice->fa_prevend;
812 fp->f_advice->fa_prevstart = start;
813 fp->f_advice->fa_prevend = end;
816 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
822 * File table vnode write routine.
825 vn_write(fp, uio, active_cred, flags, td)
828 struct ucred *active_cred;
835 int error, ioflag, lock_flags;
837 off_t offset, start, end;
839 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
841 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
843 if (vp->v_type == VREG)
846 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
848 if (fp->f_flag & FNONBLOCK)
850 if (fp->f_flag & O_DIRECT)
852 if ((fp->f_flag & O_FSYNC) ||
853 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
856 if (vp->v_type != VCHR &&
857 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
860 advice = get_advice(fp, uio);
862 if (MNT_SHARED_WRITES(mp) ||
863 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
864 lock_flags = LK_SHARED;
866 lock_flags = LK_EXCLUSIVE;
869 vn_lock(vp, lock_flags | LK_RETRY);
871 case POSIX_FADV_NORMAL:
872 case POSIX_FADV_SEQUENTIAL:
873 case POSIX_FADV_NOREUSE:
874 ioflag |= sequential_heuristic(uio, fp);
876 case POSIX_FADV_RANDOM:
877 /* XXX: Is this correct? */
880 offset = uio->uio_offset;
883 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
886 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
887 fp->f_nextoff = uio->uio_offset;
889 if (vp->v_type != VCHR)
890 vn_finished_write(mp);
891 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
892 offset != uio->uio_offset) {
894 * Use POSIX_FADV_DONTNEED to flush clean pages and
895 * buffers for the backing file after a
896 * POSIX_FADV_NOREUSE write(2). To optimize the
897 * common case of using POSIX_FADV_NOREUSE with
898 * sequential access, track the previous implicit
899 * DONTNEED request and grow this request to include
900 * the current write(2) in addition to the previous
901 * DONTNEED. With purely sequential access this will
902 * cause the DONTNEED requests to continously grow to
903 * cover all of the previously written regions of the
906 * Note that the blocks just written are almost
907 * certainly still dirty, so this only works when
908 * VOP_ADVISE() calls from subsequent writes push out
909 * the data written by this write(2) once the backing
910 * buffers are clean. However, as compared to forcing
911 * IO_DIRECT, this gives much saner behavior. Write
912 * clustering is still allowed, and clean pages are
913 * merely moved to the cache page queue rather than
914 * outright thrown away. This means a subsequent
915 * read(2) can still avoid hitting the disk if the
916 * pages have not been reclaimed.
918 * This does make POSIX_FADV_NOREUSE largely useless
919 * with non-sequential access. However, sequential
920 * access is the more common use case and the flag is
924 end = uio->uio_offset - 1;
925 mtxp = mtx_pool_find(mtxpool_sleep, fp);
927 if (fp->f_advice != NULL &&
928 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
929 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
930 start = fp->f_advice->fa_prevstart;
931 else if (fp->f_advice->fa_prevstart != 0 &&
932 fp->f_advice->fa_prevstart == end + 1)
933 end = fp->f_advice->fa_prevend;
934 fp->f_advice->fa_prevstart = start;
935 fp->f_advice->fa_prevend = end;
938 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
946 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
947 * prevent the following deadlock:
949 * Assume that the thread A reads from the vnode vp1 into userspace
950 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
951 * currently not resident, then system ends up with the call chain
952 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
953 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
954 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
955 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
956 * backed by the pages of vnode vp1, and some page in buf2 is not
957 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
959 * To prevent the lock order reversal and deadlock, vn_io_fault() does
960 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
961 * Instead, it first tries to do the whole range i/o with pagefaults
962 * disabled. If all pages in the i/o buffer are resident and mapped,
963 * VOP will succeed (ignoring the genuine filesystem errors).
964 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
965 * i/o in chunks, with all pages in the chunk prefaulted and held
966 * using vm_fault_quick_hold_pages().
968 * Filesystems using this deadlock avoidance scheme should use the
969 * array of the held pages from uio, saved in the curthread->td_ma,
970 * instead of doing uiomove(). A helper function
971 * vn_io_fault_uiomove() converts uiomove request into
972 * uiomove_fromphys() over td_ma array.
974 * Since vnode locks do not cover the whole i/o anymore, rangelocks
975 * make the current i/o request atomic with respect to other i/os and
980 * Decode vn_io_fault_args and perform the corresponding i/o.
983 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
987 switch (args->kind) {
988 case VN_IO_FAULT_FOP:
989 return ((args->args.fop_args.doio)(args->args.fop_args.fp,
990 uio, args->cred, args->flags, td));
991 case VN_IO_FAULT_VOP:
992 if (uio->uio_rw == UIO_READ) {
993 return (VOP_READ(args->args.vop_args.vp, uio,
994 args->flags, args->cred));
995 } else if (uio->uio_rw == UIO_WRITE) {
996 return (VOP_WRITE(args->args.vop_args.vp, uio,
997 args->flags, args->cred));
1001 panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
1006 * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1007 * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1008 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1009 * into args and call vn_io_fault1() to handle faults during the user
1010 * mode buffer accesses.
1013 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1016 vm_page_t ma[io_hold_cnt + 2];
1017 struct uio *uio_clone, short_uio;
1018 struct iovec short_iovec[1];
1019 vm_page_t *prev_td_ma;
1021 vm_offset_t addr, end;
1024 int error, cnt, save, saveheld, prev_td_ma_cnt;
1026 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1029 * The UFS follows IO_UNIT directive and replays back both
1030 * uio_offset and uio_resid if an error is encountered during the
1031 * operation. But, since the iovec may be already advanced,
1032 * uio is still in an inconsistent state.
1034 * Cache a copy of the original uio, which is advanced to the redo
1035 * point using UIO_NOCOPY below.
1037 uio_clone = cloneuio(uio);
1038 resid = uio->uio_resid;
1040 short_uio.uio_segflg = UIO_USERSPACE;
1041 short_uio.uio_rw = uio->uio_rw;
1042 short_uio.uio_td = uio->uio_td;
1044 save = vm_fault_disable_pagefaults();
1045 error = vn_io_fault_doio(args, uio, td);
1046 if (error != EFAULT)
1049 atomic_add_long(&vn_io_faults_cnt, 1);
1050 uio_clone->uio_segflg = UIO_NOCOPY;
1051 uiomove(NULL, resid - uio->uio_resid, uio_clone);
1052 uio_clone->uio_segflg = uio->uio_segflg;
1054 saveheld = curthread_pflags_set(TDP_UIOHELD);
1055 prev_td_ma = td->td_ma;
1056 prev_td_ma_cnt = td->td_ma_cnt;
1058 while (uio_clone->uio_resid != 0) {
1059 len = uio_clone->uio_iov->iov_len;
1061 KASSERT(uio_clone->uio_iovcnt >= 1,
1062 ("iovcnt underflow"));
1063 uio_clone->uio_iov++;
1064 uio_clone->uio_iovcnt--;
1067 if (len > io_hold_cnt * PAGE_SIZE)
1068 len = io_hold_cnt * PAGE_SIZE;
1069 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1070 end = round_page(addr + len);
1075 cnt = atop(end - trunc_page(addr));
1077 * A perfectly misaligned address and length could cause
1078 * both the start and the end of the chunk to use partial
1079 * page. +2 accounts for such a situation.
1081 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1082 addr, len, prot, ma, io_hold_cnt + 2);
1087 short_uio.uio_iov = &short_iovec[0];
1088 short_iovec[0].iov_base = (void *)addr;
1089 short_uio.uio_iovcnt = 1;
1090 short_uio.uio_resid = short_iovec[0].iov_len = len;
1091 short_uio.uio_offset = uio_clone->uio_offset;
1093 td->td_ma_cnt = cnt;
1095 error = vn_io_fault_doio(args, &short_uio, td);
1096 vm_page_unhold_pages(ma, cnt);
1097 adv = len - short_uio.uio_resid;
1099 uio_clone->uio_iov->iov_base =
1100 (char *)uio_clone->uio_iov->iov_base + adv;
1101 uio_clone->uio_iov->iov_len -= adv;
1102 uio_clone->uio_resid -= adv;
1103 uio_clone->uio_offset += adv;
1105 uio->uio_resid -= adv;
1106 uio->uio_offset += adv;
1108 if (error != 0 || adv == 0)
1111 td->td_ma = prev_td_ma;
1112 td->td_ma_cnt = prev_td_ma_cnt;
1113 curthread_pflags_restore(saveheld);
1115 vm_fault_enable_pagefaults(save);
1116 free(uio_clone, M_IOV);
1121 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1122 int flags, struct thread *td)
1127 struct vn_io_fault_args args;
1130 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1132 foffset_lock_uio(fp, uio, flags);
1133 if (do_vn_io_fault(vp, uio)) {
1134 args.kind = VN_IO_FAULT_FOP;
1135 args.args.fop_args.fp = fp;
1136 args.args.fop_args.doio = doio;
1137 args.cred = active_cred;
1138 args.flags = flags | FOF_OFFSET;
1139 if (uio->uio_rw == UIO_READ) {
1140 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1141 uio->uio_offset + uio->uio_resid);
1142 } else if ((fp->f_flag & O_APPEND) != 0 ||
1143 (flags & FOF_OFFSET) == 0) {
1144 /* For appenders, punt and lock the whole range. */
1145 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1147 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1148 uio->uio_offset + uio->uio_resid);
1150 error = vn_io_fault1(vp, uio, &args, td);
1151 vn_rangelock_unlock(vp, rl_cookie);
1153 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1155 foffset_unlock_uio(fp, uio, flags);
1160 * Helper function to perform the requested uiomove operation using
1161 * the held pages for io->uio_iov[0].iov_base buffer instead of
1162 * copyin/copyout. Access to the pages with uiomove_fromphys()
1163 * instead of iov_base prevents page faults that could occur due to
1164 * pmap_collect() invalidating the mapping created by
1165 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1166 * object cleanup revoking the write access from page mappings.
1168 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1169 * instead of plain uiomove().
1172 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1174 struct uio transp_uio;
1175 struct iovec transp_iov[1];
1181 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1182 uio->uio_segflg != UIO_USERSPACE)
1183 return (uiomove(data, xfersize, uio));
1185 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1186 transp_iov[0].iov_base = data;
1187 transp_uio.uio_iov = &transp_iov[0];
1188 transp_uio.uio_iovcnt = 1;
1189 if (xfersize > uio->uio_resid)
1190 xfersize = uio->uio_resid;
1191 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1192 transp_uio.uio_offset = 0;
1193 transp_uio.uio_segflg = UIO_SYSSPACE;
1195 * Since transp_iov points to data, and td_ma page array
1196 * corresponds to original uio->uio_iov, we need to invert the
1197 * direction of the i/o operation as passed to
1198 * uiomove_fromphys().
1200 switch (uio->uio_rw) {
1202 transp_uio.uio_rw = UIO_READ;
1205 transp_uio.uio_rw = UIO_WRITE;
1208 transp_uio.uio_td = uio->uio_td;
1209 error = uiomove_fromphys(td->td_ma,
1210 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1211 xfersize, &transp_uio);
1212 adv = xfersize - transp_uio.uio_resid;
1214 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1215 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1217 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1219 td->td_ma_cnt -= pgadv;
1220 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1221 uio->uio_iov->iov_len -= adv;
1222 uio->uio_resid -= adv;
1223 uio->uio_offset += adv;
1228 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1232 vm_offset_t iov_base;
1236 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1237 uio->uio_segflg != UIO_USERSPACE)
1238 return (uiomove_fromphys(ma, offset, xfersize, uio));
1240 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1241 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1242 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1243 switch (uio->uio_rw) {
1245 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1249 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1253 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1255 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1257 td->td_ma_cnt -= pgadv;
1258 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1259 uio->uio_iov->iov_len -= cnt;
1260 uio->uio_resid -= cnt;
1261 uio->uio_offset += cnt;
1267 * File table truncate routine.
1270 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1282 * Lock the whole range for truncation. Otherwise split i/o
1283 * might happen partly before and partly after the truncation.
1285 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1286 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1289 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1290 if (vp->v_type == VDIR) {
1295 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1299 error = vn_writechk(vp);
1302 vattr.va_size = length;
1303 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1307 vn_finished_write(mp);
1309 vn_rangelock_unlock(vp, rl_cookie);
1314 * File table vnode stat routine.
1317 vn_statfile(fp, sb, active_cred, td)
1320 struct ucred *active_cred;
1323 struct vnode *vp = fp->f_vnode;
1326 vn_lock(vp, LK_SHARED | LK_RETRY);
1327 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1334 * Stat a vnode; implementation for the stat syscall
1337 vn_stat(vp, sb, active_cred, file_cred, td)
1339 register struct stat *sb;
1340 struct ucred *active_cred;
1341 struct ucred *file_cred;
1345 register struct vattr *vap;
1350 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1358 * Initialize defaults for new and unusual fields, so that file
1359 * systems which don't support these fields don't need to know
1362 vap->va_birthtime.tv_sec = -1;
1363 vap->va_birthtime.tv_nsec = 0;
1364 vap->va_fsid = VNOVAL;
1365 vap->va_rdev = NODEV;
1367 error = VOP_GETATTR(vp, vap, active_cred);
1372 * Zero the spare stat fields
1374 bzero(sb, sizeof *sb);
1377 * Copy from vattr table
1379 if (vap->va_fsid != VNOVAL)
1380 sb->st_dev = vap->va_fsid;
1382 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1383 sb->st_ino = vap->va_fileid;
1384 mode = vap->va_mode;
1385 switch (vap->va_type) {
1411 sb->st_nlink = vap->va_nlink;
1412 sb->st_uid = vap->va_uid;
1413 sb->st_gid = vap->va_gid;
1414 sb->st_rdev = vap->va_rdev;
1415 if (vap->va_size > OFF_MAX)
1417 sb->st_size = vap->va_size;
1418 sb->st_atim = vap->va_atime;
1419 sb->st_mtim = vap->va_mtime;
1420 sb->st_ctim = vap->va_ctime;
1421 sb->st_birthtim = vap->va_birthtime;
1424 * According to www.opengroup.org, the meaning of st_blksize is
1425 * "a filesystem-specific preferred I/O block size for this
1426 * object. In some filesystem types, this may vary from file
1428 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1431 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1433 sb->st_flags = vap->va_flags;
1434 if (priv_check(td, PRIV_VFS_GENERATION))
1437 sb->st_gen = vap->va_gen;
1439 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1444 * File table vnode ioctl routine.
1447 vn_ioctl(fp, com, data, active_cred, td)
1451 struct ucred *active_cred;
1459 switch (vp->v_type) {
1464 vn_lock(vp, LK_SHARED | LK_RETRY);
1465 error = VOP_GETATTR(vp, &vattr, active_cred);
1468 *(int *)data = vattr.va_size - fp->f_offset;
1474 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1483 * File table vnode poll routine.
1486 vn_poll(fp, events, active_cred, td)
1489 struct ucred *active_cred;
1497 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1498 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1503 error = VOP_POLL(vp, events, fp->f_cred, td);
1508 * Acquire the requested lock and then check for validity. LK_RETRY
1509 * permits vn_lock to return doomed vnodes.
1512 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1516 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1517 ("vn_lock called with no locktype."));
1519 #ifdef DEBUG_VFS_LOCKS
1520 KASSERT(vp->v_holdcnt != 0,
1521 ("vn_lock %p: zero hold count", vp));
1523 error = VOP_LOCK1(vp, flags, file, line);
1524 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1525 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1526 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1529 * Callers specify LK_RETRY if they wish to get dead vnodes.
1530 * If RETRY is not set, we return ENOENT instead.
1532 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1533 (flags & LK_RETRY) == 0) {
1538 } while (flags & LK_RETRY && error != 0);
1543 * File table vnode close routine.
1546 vn_closefile(fp, td)
1555 fp->f_ops = &badfileops;
1557 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1560 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1562 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1563 lf.l_whence = SEEK_SET;
1566 lf.l_type = F_UNLCK;
1567 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1574 * Preparing to start a filesystem write operation. If the operation is
1575 * permitted, then we bump the count of operations in progress and
1576 * proceed. If a suspend request is in progress, we wait until the
1577 * suspension is over, and then proceed.
1580 vn_start_write_locked(struct mount *mp, int flags)
1584 mtx_assert(MNT_MTX(mp), MA_OWNED);
1588 * Check on status of suspension.
1590 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1591 mp->mnt_susp_owner != curthread) {
1592 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1593 if (flags & V_NOWAIT) {
1594 error = EWOULDBLOCK;
1597 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1598 (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1603 if (flags & V_XSLEEP)
1605 mp->mnt_writeopcount++;
1607 if (error != 0 || (flags & V_XSLEEP) != 0)
1614 vn_start_write(vp, mpp, flags)
1624 * If a vnode is provided, get and return the mount point that
1625 * to which it will write.
1628 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1630 if (error != EOPNOTSUPP)
1635 if ((mp = *mpp) == NULL)
1639 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1641 * As long as a vnode is not provided we need to acquire a
1642 * refcount for the provided mountpoint too, in order to
1643 * emulate a vfs_ref().
1649 return (vn_start_write_locked(mp, flags));
1653 * Secondary suspension. Used by operations such as vop_inactive
1654 * routines that are needed by the higher level functions. These
1655 * are allowed to proceed until all the higher level functions have
1656 * completed (indicated by mnt_writeopcount dropping to zero). At that
1657 * time, these operations are halted until the suspension is over.
1660 vn_start_secondary_write(vp, mpp, flags)
1670 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1672 if (error != EOPNOTSUPP)
1678 * If we are not suspended or have not yet reached suspended
1679 * mode, then let the operation proceed.
1681 if ((mp = *mpp) == NULL)
1685 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1687 * As long as a vnode is not provided we need to acquire a
1688 * refcount for the provided mountpoint too, in order to
1689 * emulate a vfs_ref().
1694 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1695 mp->mnt_secondary_writes++;
1696 mp->mnt_secondary_accwrites++;
1700 if (flags & V_NOWAIT) {
1703 return (EWOULDBLOCK);
1706 * Wait for the suspension to finish.
1708 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1709 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1717 * Filesystem write operation has completed. If we are suspending and this
1718 * operation is the last one, notify the suspender that the suspension is
1722 vn_finished_write(mp)
1729 mp->mnt_writeopcount--;
1730 if (mp->mnt_writeopcount < 0)
1731 panic("vn_finished_write: neg cnt");
1732 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1733 mp->mnt_writeopcount <= 0)
1734 wakeup(&mp->mnt_writeopcount);
1740 * Filesystem secondary write operation has completed. If we are
1741 * suspending and this operation is the last one, notify the suspender
1742 * that the suspension is now in effect.
1745 vn_finished_secondary_write(mp)
1752 mp->mnt_secondary_writes--;
1753 if (mp->mnt_secondary_writes < 0)
1754 panic("vn_finished_secondary_write: neg cnt");
1755 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1756 mp->mnt_secondary_writes <= 0)
1757 wakeup(&mp->mnt_secondary_writes);
1764 * Request a filesystem to suspend write operations.
1767 vfs_write_suspend(struct mount *mp, int flags)
1772 if (mp->mnt_susp_owner == curthread) {
1776 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1777 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1780 * Unmount holds a write reference on the mount point. If we
1781 * own busy reference and drain for writers, we deadlock with
1782 * the reference draining in the unmount path. Callers of
1783 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1784 * vfs_busy() reference is owned and caller is not in the
1787 if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1788 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1793 mp->mnt_kern_flag |= MNTK_SUSPEND;
1794 mp->mnt_susp_owner = curthread;
1795 if (mp->mnt_writeopcount > 0)
1796 (void) msleep(&mp->mnt_writeopcount,
1797 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1800 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1801 vfs_write_resume(mp, 0);
1806 * Request a filesystem to resume write operations.
1809 vfs_write_resume(struct mount *mp, int flags)
1813 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1814 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1815 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1817 mp->mnt_susp_owner = NULL;
1818 wakeup(&mp->mnt_writeopcount);
1819 wakeup(&mp->mnt_flag);
1820 curthread->td_pflags &= ~TDP_IGNSUSP;
1821 if ((flags & VR_START_WRITE) != 0) {
1823 mp->mnt_writeopcount++;
1826 if ((flags & VR_NO_SUSPCLR) == 0)
1828 } else if ((flags & VR_START_WRITE) != 0) {
1830 vn_start_write_locked(mp, 0);
1837 * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1841 vfs_write_suspend_umnt(struct mount *mp)
1845 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1846 ("vfs_write_suspend_umnt: recursed"));
1848 /* dounmount() already called vn_start_write(). */
1850 vn_finished_write(mp);
1851 error = vfs_write_suspend(mp, 0);
1853 vn_start_write(NULL, &mp, V_WAIT);
1857 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1860 vn_start_write(NULL, &mp, V_WAIT);
1862 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1863 wakeup(&mp->mnt_flag);
1865 curthread->td_pflags |= TDP_IGNSUSP;
1870 * Implement kqueues for files by translating it to vnode operation.
1873 vn_kqfilter(struct file *fp, struct knote *kn)
1876 return (VOP_KQFILTER(fp->f_vnode, kn));
1880 * Simplified in-kernel wrapper calls for extended attribute access.
1881 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1882 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1885 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1886 const char *attrname, int *buflen, char *buf, struct thread *td)
1892 iov.iov_len = *buflen;
1895 auio.uio_iov = &iov;
1896 auio.uio_iovcnt = 1;
1897 auio.uio_rw = UIO_READ;
1898 auio.uio_segflg = UIO_SYSSPACE;
1900 auio.uio_offset = 0;
1901 auio.uio_resid = *buflen;
1903 if ((ioflg & IO_NODELOCKED) == 0)
1904 vn_lock(vp, LK_SHARED | LK_RETRY);
1906 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1908 /* authorize attribute retrieval as kernel */
1909 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1912 if ((ioflg & IO_NODELOCKED) == 0)
1916 *buflen = *buflen - auio.uio_resid;
1923 * XXX failure mode if partially written?
1926 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1927 const char *attrname, int buflen, char *buf, struct thread *td)
1934 iov.iov_len = buflen;
1937 auio.uio_iov = &iov;
1938 auio.uio_iovcnt = 1;
1939 auio.uio_rw = UIO_WRITE;
1940 auio.uio_segflg = UIO_SYSSPACE;
1942 auio.uio_offset = 0;
1943 auio.uio_resid = buflen;
1945 if ((ioflg & IO_NODELOCKED) == 0) {
1946 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1948 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1951 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1953 /* authorize attribute setting as kernel */
1954 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1956 if ((ioflg & IO_NODELOCKED) == 0) {
1957 vn_finished_write(mp);
1965 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1966 const char *attrname, struct thread *td)
1971 if ((ioflg & IO_NODELOCKED) == 0) {
1972 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1974 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1977 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1979 /* authorize attribute removal as kernel */
1980 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1981 if (error == EOPNOTSUPP)
1982 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1985 if ((ioflg & IO_NODELOCKED) == 0) {
1986 vn_finished_write(mp);
1994 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
1998 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2002 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2005 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2010 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2011 int lkflags, struct vnode **rvp)
2016 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2018 ltype = VOP_ISLOCKED(vp);
2019 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2020 ("vn_vget_ino: vp not locked"));
2021 error = vfs_busy(mp, MBF_NOWAIT);
2025 error = vfs_busy(mp, 0);
2026 vn_lock(vp, ltype | LK_RETRY);
2030 if (vp->v_iflag & VI_DOOMED) {
2036 error = alloc(mp, alloc_arg, lkflags, rvp);
2039 vn_lock(vp, ltype | LK_RETRY);
2040 if (vp->v_iflag & VI_DOOMED) {
2053 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2054 const struct thread *td)
2057 if (vp->v_type != VREG || td == NULL)
2059 PROC_LOCK(td->td_proc);
2060 if ((uoff_t)uio->uio_offset + uio->uio_resid >
2061 lim_cur(td->td_proc, RLIMIT_FSIZE)) {
2062 kern_psignal(td->td_proc, SIGXFSZ);
2063 PROC_UNLOCK(td->td_proc);
2066 PROC_UNLOCK(td->td_proc);
2071 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2078 vn_lock(vp, LK_SHARED | LK_RETRY);
2079 AUDIT_ARG_VNODE1(vp);
2082 return (setfmode(td, active_cred, vp, mode));
2086 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2093 vn_lock(vp, LK_SHARED | LK_RETRY);
2094 AUDIT_ARG_VNODE1(vp);
2097 return (setfown(td, active_cred, vp, uid, gid));
2101 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2105 if ((object = vp->v_object) == NULL)
2107 VM_OBJECT_WLOCK(object);
2108 vm_object_page_remove(object, start, end, 0);
2109 VM_OBJECT_WUNLOCK(object);
2113 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2121 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2122 ("Wrong command %lu", cmd));
2124 if (vn_lock(vp, LK_SHARED) != 0)
2126 if (vp->v_type != VREG) {
2130 error = VOP_GETATTR(vp, &va, cred);
2134 if (noff >= va.va_size) {
2138 bsize = vp->v_mount->mnt_stat.f_iosize;
2139 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2140 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2141 if (error == EOPNOTSUPP) {
2145 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2146 (bnp != -1 && cmd == FIOSEEKDATA)) {
2153 if (noff > va.va_size)
2155 /* noff == va.va_size. There is an implicit hole at the end of file. */
2156 if (cmd == FIOSEEKDATA)
2166 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2171 off_t foffset, size;
2174 cred = td->td_ucred;
2176 foffset = foffset_lock(fp, 0);
2177 noneg = (vp->v_type != VCHR);
2183 (offset > 0 && foffset > OFF_MAX - offset))) {
2190 vn_lock(vp, LK_SHARED | LK_RETRY);
2191 error = VOP_GETATTR(vp, &vattr, cred);
2197 * If the file references a disk device, then fetch
2198 * the media size and use that to determine the ending
2201 if (vattr.va_size == 0 && vp->v_type == VCHR &&
2202 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2203 vattr.va_size = size;
2205 (vattr.va_size > OFF_MAX ||
2206 (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2210 offset += vattr.va_size;
2215 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2218 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2223 if (error == 0 && noneg && offset < 0)
2227 VFS_KNOTE_UNLOCKED(vp, 0);
2228 *(off_t *)(td->td_retval) = offset;
2230 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2235 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2241 * Grant permission if the caller is the owner of the file, or
2242 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2243 * on the file. If the time pointer is null, then write
2244 * permission on the file is also sufficient.
2246 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2247 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2248 * will be allowed to set the times [..] to the current
2251 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2252 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2253 error = VOP_ACCESS(vp, VWRITE, cred, td);