2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org>
11 * Copyright (c) 2013, 2014 The FreeBSD Foundation
13 * Portions of this software were developed by Konstantin Belousov
14 * under sponsorship from the FreeBSD Foundation.
16 * Redistribution and use in source and binary forms, with or without
17 * modification, are permitted provided that the following conditions
19 * 1. Redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer.
21 * 2. Redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
46 #include <sys/param.h>
47 #include <sys/systm.h>
49 #include <sys/fcntl.h>
55 #include <sys/limits.h>
57 #include <sys/mount.h>
58 #include <sys/mutex.h>
59 #include <sys/namei.h>
60 #include <sys/vnode.h>
63 #include <sys/filio.h>
64 #include <sys/resourcevar.h>
65 #include <sys/rwlock.h>
67 #include <sys/sysctl.h>
68 #include <sys/ttycom.h>
70 #include <sys/syslog.h>
71 #include <sys/unistd.h>
74 #include <security/audit/audit.h>
75 #include <security/mac/mac_framework.h>
78 #include <vm/vm_extern.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_page.h>
84 static fo_rdwr_t vn_read;
85 static fo_rdwr_t vn_write;
86 static fo_rdwr_t vn_io_fault;
87 static fo_truncate_t vn_truncate;
88 static fo_ioctl_t vn_ioctl;
89 static fo_poll_t vn_poll;
90 static fo_kqfilter_t vn_kqfilter;
91 static fo_stat_t vn_statfile;
92 static fo_close_t vn_closefile;
94 struct fileops vnops = {
95 .fo_read = vn_io_fault,
96 .fo_write = vn_io_fault,
97 .fo_truncate = vn_truncate,
100 .fo_kqfilter = vn_kqfilter,
101 .fo_stat = vn_statfile,
102 .fo_close = vn_closefile,
103 .fo_chmod = vn_chmod,
104 .fo_chown = vn_chown,
105 .fo_sendfile = vn_sendfile,
107 .fo_fill_kinfo = vn_fill_kinfo,
108 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
111 static const int io_hold_cnt = 16;
112 static int vn_io_fault_enable = 1;
113 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
114 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
115 static u_long vn_io_faults_cnt;
116 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
117 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
120 * Returns true if vn_io_fault mode of handling the i/o request should
124 do_vn_io_fault(struct vnode *vp, struct uio *uio)
128 return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
129 (mp = vp->v_mount) != NULL &&
130 (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
134 * Structure used to pass arguments to vn_io_fault1(), to do either
135 * file- or vnode-based I/O calls.
137 struct vn_io_fault_args {
145 struct fop_args_tag {
149 struct vop_args_tag {
155 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
156 struct vn_io_fault_args *args, struct thread *td);
159 vn_open(ndp, flagp, cmode, fp)
160 struct nameidata *ndp;
164 struct thread *td = ndp->ni_cnd.cn_thread;
166 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
170 * Common code for vnode open operations via a name lookup.
171 * Lookup the vnode and invoke VOP_CREATE if needed.
172 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
174 * Note that this does NOT free nameidata for the successful case,
175 * due to the NDINIT being done elsewhere.
178 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
179 struct ucred *cred, struct file *fp)
183 struct thread *td = ndp->ni_cnd.cn_thread;
185 struct vattr *vap = &vat;
190 if (fmode & O_CREAT) {
191 ndp->ni_cnd.cn_nameiop = CREATE;
193 * Set NOCACHE to avoid flushing the cache when
194 * rolling in many files at once.
196 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE;
197 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
198 ndp->ni_cnd.cn_flags |= FOLLOW;
199 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
200 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
201 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
202 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
204 if ((error = namei(ndp)) != 0)
206 if (ndp->ni_vp == NULL) {
209 vap->va_mode = cmode;
211 vap->va_vaflags |= VA_EXCLUSIVE;
212 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
213 NDFREE(ndp, NDF_ONLY_PNBUF);
215 if ((error = vn_start_write(NULL, &mp,
216 V_XSLEEP | PCATCH)) != 0)
220 if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0)
221 ndp->ni_cnd.cn_flags |= MAKEENTRY;
223 error = mac_vnode_check_create(cred, ndp->ni_dvp,
227 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
230 vn_finished_write(mp);
232 NDFREE(ndp, NDF_ONLY_PNBUF);
238 if (ndp->ni_dvp == ndp->ni_vp)
244 if (fmode & O_EXCL) {
251 ndp->ni_cnd.cn_nameiop = LOOKUP;
252 ndp->ni_cnd.cn_flags = ISOPEN |
253 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
254 if (!(fmode & FWRITE))
255 ndp->ni_cnd.cn_flags |= LOCKSHARED;
256 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
257 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
258 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
259 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
260 if ((error = namei(ndp)) != 0)
264 error = vn_open_vnode(vp, fmode, cred, td, fp);
270 NDFREE(ndp, NDF_ONLY_PNBUF);
278 * Common code for vnode open operations once a vnode is located.
279 * Check permissions, and call the VOP_OPEN routine.
282 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
283 struct thread *td, struct file *fp)
288 int error, have_flock, lock_flags, type;
290 if (vp->v_type == VLNK)
292 if (vp->v_type == VSOCK)
294 if (vp->v_type != VDIR && fmode & O_DIRECTORY)
297 if (fmode & (FWRITE | O_TRUNC)) {
298 if (vp->v_type == VDIR)
306 if ((fmode & O_APPEND) && (fmode & FWRITE))
311 if (fmode & O_VERIFY)
313 error = mac_vnode_check_open(cred, vp, accmode);
317 accmode &= ~(VCREAT | VVERIFY);
319 if ((fmode & O_CREAT) == 0) {
320 if (accmode & VWRITE) {
321 error = vn_writechk(vp);
326 error = VOP_ACCESS(vp, accmode, cred, td);
331 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
332 vn_lock(vp, LK_UPGRADE | LK_RETRY);
333 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
336 if (fmode & (O_EXLOCK | O_SHLOCK)) {
337 KASSERT(fp != NULL, ("open with flock requires fp"));
338 lock_flags = VOP_ISLOCKED(vp);
340 lf.l_whence = SEEK_SET;
343 if (fmode & O_EXLOCK)
348 if ((fmode & FNONBLOCK) == 0)
350 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
351 have_flock = (error == 0);
352 vn_lock(vp, lock_flags | LK_RETRY);
353 if (error == 0 && vp->v_iflag & VI_DOOMED)
356 * Another thread might have used this vnode as an
357 * executable while the vnode lock was dropped.
358 * Ensure the vnode is still able to be opened for
359 * writing after the lock has been obtained.
361 if (error == 0 && accmode & VWRITE)
362 error = vn_writechk(vp);
366 lf.l_whence = SEEK_SET;
370 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
373 vn_start_write(vp, &mp, V_WAIT);
374 vn_lock(vp, lock_flags | LK_RETRY);
375 (void)VOP_CLOSE(vp, fmode, cred, td);
376 vn_finished_write(mp);
377 /* Prevent second close from fdrop()->vn_close(). */
379 fp->f_ops= &badfileops;
382 fp->f_flag |= FHASLOCK;
384 if (fmode & FWRITE) {
385 VOP_ADD_WRITECOUNT(vp, 1);
386 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
387 __func__, vp, vp->v_writecount);
389 ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
394 * Check for write permissions on the specified vnode.
395 * Prototype text segments cannot be written.
399 register struct vnode *vp;
402 ASSERT_VOP_LOCKED(vp, "vn_writechk");
404 * If there's shared text associated with
405 * the vnode, try to free it up once. If
406 * we fail, we can't allow writing.
418 vn_close(vp, flags, file_cred, td)
419 register struct vnode *vp;
421 struct ucred *file_cred;
425 int error, lock_flags;
427 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
428 MNT_EXTENDED_SHARED(vp->v_mount))
429 lock_flags = LK_SHARED;
431 lock_flags = LK_EXCLUSIVE;
433 vn_start_write(vp, &mp, V_WAIT);
434 vn_lock(vp, lock_flags | LK_RETRY);
435 if (flags & FWRITE) {
436 VNASSERT(vp->v_writecount > 0, vp,
437 ("vn_close: negative writecount"));
438 VOP_ADD_WRITECOUNT(vp, -1);
439 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
440 __func__, vp, vp->v_writecount);
442 error = VOP_CLOSE(vp, flags, file_cred, td);
444 vn_finished_write(mp);
449 * Heuristic to detect sequential operation.
452 sequential_heuristic(struct uio *uio, struct file *fp)
455 ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
456 if (fp->f_flag & FRDAHEAD)
457 return (fp->f_seqcount << IO_SEQSHIFT);
460 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
461 * that the first I/O is normally considered to be slightly
462 * sequential. Seeking to offset 0 doesn't change sequentiality
463 * unless previous seeks have reduced f_seqcount to 0, in which
464 * case offset 0 is not special.
466 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
467 uio->uio_offset == fp->f_nextoff) {
469 * f_seqcount is in units of fixed-size blocks so that it
470 * depends mainly on the amount of sequential I/O and not
471 * much on the number of sequential I/O's. The fixed size
472 * of 16384 is hard-coded here since it is (not quite) just
473 * a magic size that works well here. This size is more
474 * closely related to the best I/O size for real disks than
475 * to any block size used by software.
477 fp->f_seqcount += howmany(uio->uio_resid, 16384);
478 if (fp->f_seqcount > IO_SEQMAX)
479 fp->f_seqcount = IO_SEQMAX;
480 return (fp->f_seqcount << IO_SEQSHIFT);
483 /* Not sequential. Quickly draw-down sequentiality. */
484 if (fp->f_seqcount > 1)
492 * Package up an I/O request on a vnode into a uio and do it.
495 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
496 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
497 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
504 struct vn_io_fault_args args;
505 int error, lock_flags;
507 auio.uio_iov = &aiov;
509 aiov.iov_base = base;
511 auio.uio_resid = len;
512 auio.uio_offset = offset;
513 auio.uio_segflg = segflg;
518 if ((ioflg & IO_NODELOCKED) == 0) {
519 if ((ioflg & IO_RANGELOCKED) == 0) {
520 if (rw == UIO_READ) {
521 rl_cookie = vn_rangelock_rlock(vp, offset,
524 rl_cookie = vn_rangelock_wlock(vp, offset,
530 if (rw == UIO_WRITE) {
531 if (vp->v_type != VCHR &&
532 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
535 if (MNT_SHARED_WRITES(mp) ||
536 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
537 lock_flags = LK_SHARED;
539 lock_flags = LK_EXCLUSIVE;
541 lock_flags = LK_SHARED;
542 vn_lock(vp, lock_flags | LK_RETRY);
546 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
548 if ((ioflg & IO_NOMACCHECK) == 0) {
550 error = mac_vnode_check_read(active_cred, file_cred,
553 error = mac_vnode_check_write(active_cred, file_cred,
558 if (file_cred != NULL)
562 if (do_vn_io_fault(vp, &auio)) {
563 args.kind = VN_IO_FAULT_VOP;
566 args.args.vop_args.vp = vp;
567 error = vn_io_fault1(vp, &auio, &args, td);
568 } else if (rw == UIO_READ) {
569 error = VOP_READ(vp, &auio, ioflg, cred);
570 } else /* if (rw == UIO_WRITE) */ {
571 error = VOP_WRITE(vp, &auio, ioflg, cred);
575 *aresid = auio.uio_resid;
577 if (auio.uio_resid && error == 0)
579 if ((ioflg & IO_NODELOCKED) == 0) {
582 vn_finished_write(mp);
585 if (rl_cookie != NULL)
586 vn_rangelock_unlock(vp, rl_cookie);
591 * Package up an I/O request on a vnode into a uio and do it. The I/O
592 * request is split up into smaller chunks and we try to avoid saturating
593 * the buffer cache while potentially holding a vnode locked, so we
594 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
595 * to give other processes a chance to lock the vnode (either other processes
596 * core'ing the same binary, or unrelated processes scanning the directory).
599 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
600 file_cred, aresid, td)
608 struct ucred *active_cred;
609 struct ucred *file_cred;
620 * Force `offset' to a multiple of MAXBSIZE except possibly
621 * for the first chunk, so that filesystems only need to
622 * write full blocks except possibly for the first and last
625 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
629 if (rw != UIO_READ && vp->v_type == VREG)
632 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
633 ioflg, active_cred, file_cred, &iaresid, td);
634 len -= chunk; /* aresid calc already includes length */
638 base = (char *)base + chunk;
639 kern_yield(PRI_USER);
642 *aresid = len + iaresid;
647 foffset_lock(struct file *fp, int flags)
652 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
654 #if OFF_MAX <= LONG_MAX
656 * Caller only wants the current f_offset value. Assume that
657 * the long and shorter integer types reads are atomic.
659 if ((flags & FOF_NOLOCK) != 0)
660 return (fp->f_offset);
664 * According to McKusick the vn lock was protecting f_offset here.
665 * It is now protected by the FOFFSET_LOCKED flag.
667 mtxp = mtx_pool_find(mtxpool_sleep, fp);
669 if ((flags & FOF_NOLOCK) == 0) {
670 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
671 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
672 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
675 fp->f_vnread_flags |= FOFFSET_LOCKED;
683 foffset_unlock(struct file *fp, off_t val, int flags)
687 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
689 #if OFF_MAX <= LONG_MAX
690 if ((flags & FOF_NOLOCK) != 0) {
691 if ((flags & FOF_NOUPDATE) == 0)
693 if ((flags & FOF_NEXTOFF) != 0)
699 mtxp = mtx_pool_find(mtxpool_sleep, fp);
701 if ((flags & FOF_NOUPDATE) == 0)
703 if ((flags & FOF_NEXTOFF) != 0)
705 if ((flags & FOF_NOLOCK) == 0) {
706 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
707 ("Lost FOFFSET_LOCKED"));
708 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
709 wakeup(&fp->f_vnread_flags);
710 fp->f_vnread_flags = 0;
716 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
719 if ((flags & FOF_OFFSET) == 0)
720 uio->uio_offset = foffset_lock(fp, flags);
724 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
727 if ((flags & FOF_OFFSET) == 0)
728 foffset_unlock(fp, uio->uio_offset, flags);
732 get_advice(struct file *fp, struct uio *uio)
737 ret = POSIX_FADV_NORMAL;
738 if (fp->f_advice == NULL)
741 mtxp = mtx_pool_find(mtxpool_sleep, fp);
743 if (uio->uio_offset >= fp->f_advice->fa_start &&
744 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
745 ret = fp->f_advice->fa_advice;
751 * File table vnode read routine.
754 vn_read(fp, uio, active_cred, flags, td)
757 struct ucred *active_cred;
765 off_t offset, start, end;
767 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
769 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
772 if (fp->f_flag & FNONBLOCK)
774 if (fp->f_flag & O_DIRECT)
776 advice = get_advice(fp, uio);
777 vn_lock(vp, LK_SHARED | LK_RETRY);
780 case POSIX_FADV_NORMAL:
781 case POSIX_FADV_SEQUENTIAL:
782 case POSIX_FADV_NOREUSE:
783 ioflag |= sequential_heuristic(uio, fp);
785 case POSIX_FADV_RANDOM:
786 /* Disable read-ahead for random I/O. */
789 offset = uio->uio_offset;
792 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
795 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
796 fp->f_nextoff = uio->uio_offset;
798 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
799 offset != uio->uio_offset) {
801 * Use POSIX_FADV_DONTNEED to flush clean pages and
802 * buffers for the backing file after a
803 * POSIX_FADV_NOREUSE read(2). To optimize the common
804 * case of using POSIX_FADV_NOREUSE with sequential
805 * access, track the previous implicit DONTNEED
806 * request and grow this request to include the
807 * current read(2) in addition to the previous
808 * DONTNEED. With purely sequential access this will
809 * cause the DONTNEED requests to continously grow to
810 * cover all of the previously read regions of the
811 * file. This allows filesystem blocks that are
812 * accessed by multiple calls to read(2) to be flushed
813 * once the last read(2) finishes.
816 end = uio->uio_offset - 1;
817 mtxp = mtx_pool_find(mtxpool_sleep, fp);
819 if (fp->f_advice != NULL &&
820 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
821 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
822 start = fp->f_advice->fa_prevstart;
823 else if (fp->f_advice->fa_prevstart != 0 &&
824 fp->f_advice->fa_prevstart == end + 1)
825 end = fp->f_advice->fa_prevend;
826 fp->f_advice->fa_prevstart = start;
827 fp->f_advice->fa_prevend = end;
830 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
836 * File table vnode write routine.
839 vn_write(fp, uio, active_cred, flags, td)
842 struct ucred *active_cred;
849 int error, ioflag, lock_flags;
851 off_t offset, start, end;
853 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
855 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
857 if (vp->v_type == VREG)
860 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
862 if (fp->f_flag & FNONBLOCK)
864 if (fp->f_flag & O_DIRECT)
866 if ((fp->f_flag & O_FSYNC) ||
867 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
870 if (vp->v_type != VCHR &&
871 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
874 advice = get_advice(fp, uio);
876 if (MNT_SHARED_WRITES(mp) ||
877 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
878 lock_flags = LK_SHARED;
880 lock_flags = LK_EXCLUSIVE;
883 vn_lock(vp, lock_flags | LK_RETRY);
885 case POSIX_FADV_NORMAL:
886 case POSIX_FADV_SEQUENTIAL:
887 case POSIX_FADV_NOREUSE:
888 ioflag |= sequential_heuristic(uio, fp);
890 case POSIX_FADV_RANDOM:
891 /* XXX: Is this correct? */
894 offset = uio->uio_offset;
897 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
900 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
901 fp->f_nextoff = uio->uio_offset;
903 if (vp->v_type != VCHR)
904 vn_finished_write(mp);
905 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
906 offset != uio->uio_offset) {
908 * Use POSIX_FADV_DONTNEED to flush clean pages and
909 * buffers for the backing file after a
910 * POSIX_FADV_NOREUSE write(2). To optimize the
911 * common case of using POSIX_FADV_NOREUSE with
912 * sequential access, track the previous implicit
913 * DONTNEED request and grow this request to include
914 * the current write(2) in addition to the previous
915 * DONTNEED. With purely sequential access this will
916 * cause the DONTNEED requests to continously grow to
917 * cover all of the previously written regions of the
920 * Note that the blocks just written are almost
921 * certainly still dirty, so this only works when
922 * VOP_ADVISE() calls from subsequent writes push out
923 * the data written by this write(2) once the backing
924 * buffers are clean. However, as compared to forcing
925 * IO_DIRECT, this gives much saner behavior. Write
926 * clustering is still allowed, and clean pages are
927 * merely moved to the cache page queue rather than
928 * outright thrown away. This means a subsequent
929 * read(2) can still avoid hitting the disk if the
930 * pages have not been reclaimed.
932 * This does make POSIX_FADV_NOREUSE largely useless
933 * with non-sequential access. However, sequential
934 * access is the more common use case and the flag is
938 end = uio->uio_offset - 1;
939 mtxp = mtx_pool_find(mtxpool_sleep, fp);
941 if (fp->f_advice != NULL &&
942 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
943 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
944 start = fp->f_advice->fa_prevstart;
945 else if (fp->f_advice->fa_prevstart != 0 &&
946 fp->f_advice->fa_prevstart == end + 1)
947 end = fp->f_advice->fa_prevend;
948 fp->f_advice->fa_prevstart = start;
949 fp->f_advice->fa_prevend = end;
952 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
960 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
961 * prevent the following deadlock:
963 * Assume that the thread A reads from the vnode vp1 into userspace
964 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
965 * currently not resident, then system ends up with the call chain
966 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
967 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
968 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
969 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
970 * backed by the pages of vnode vp1, and some page in buf2 is not
971 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
973 * To prevent the lock order reversal and deadlock, vn_io_fault() does
974 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
975 * Instead, it first tries to do the whole range i/o with pagefaults
976 * disabled. If all pages in the i/o buffer are resident and mapped,
977 * VOP will succeed (ignoring the genuine filesystem errors).
978 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
979 * i/o in chunks, with all pages in the chunk prefaulted and held
980 * using vm_fault_quick_hold_pages().
982 * Filesystems using this deadlock avoidance scheme should use the
983 * array of the held pages from uio, saved in the curthread->td_ma,
984 * instead of doing uiomove(). A helper function
985 * vn_io_fault_uiomove() converts uiomove request into
986 * uiomove_fromphys() over td_ma array.
988 * Since vnode locks do not cover the whole i/o anymore, rangelocks
989 * make the current i/o request atomic with respect to other i/os and
994 * Decode vn_io_fault_args and perform the corresponding i/o.
997 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
1001 switch (args->kind) {
1002 case VN_IO_FAULT_FOP:
1003 return ((args->args.fop_args.doio)(args->args.fop_args.fp,
1004 uio, args->cred, args->flags, td));
1005 case VN_IO_FAULT_VOP:
1006 if (uio->uio_rw == UIO_READ) {
1007 return (VOP_READ(args->args.vop_args.vp, uio,
1008 args->flags, args->cred));
1009 } else if (uio->uio_rw == UIO_WRITE) {
1010 return (VOP_WRITE(args->args.vop_args.vp, uio,
1011 args->flags, args->cred));
1015 panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
1020 * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1021 * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1022 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1023 * into args and call vn_io_fault1() to handle faults during the user
1024 * mode buffer accesses.
1027 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1030 vm_page_t ma[io_hold_cnt + 2];
1031 struct uio *uio_clone, short_uio;
1032 struct iovec short_iovec[1];
1033 vm_page_t *prev_td_ma;
1035 vm_offset_t addr, end;
1038 int error, cnt, save, saveheld, prev_td_ma_cnt;
1040 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1043 * The UFS follows IO_UNIT directive and replays back both
1044 * uio_offset and uio_resid if an error is encountered during the
1045 * operation. But, since the iovec may be already advanced,
1046 * uio is still in an inconsistent state.
1048 * Cache a copy of the original uio, which is advanced to the redo
1049 * point using UIO_NOCOPY below.
1051 uio_clone = cloneuio(uio);
1052 resid = uio->uio_resid;
1054 short_uio.uio_segflg = UIO_USERSPACE;
1055 short_uio.uio_rw = uio->uio_rw;
1056 short_uio.uio_td = uio->uio_td;
1058 save = vm_fault_disable_pagefaults();
1059 error = vn_io_fault_doio(args, uio, td);
1060 if (error != EFAULT)
1063 atomic_add_long(&vn_io_faults_cnt, 1);
1064 uio_clone->uio_segflg = UIO_NOCOPY;
1065 uiomove(NULL, resid - uio->uio_resid, uio_clone);
1066 uio_clone->uio_segflg = uio->uio_segflg;
1068 saveheld = curthread_pflags_set(TDP_UIOHELD);
1069 prev_td_ma = td->td_ma;
1070 prev_td_ma_cnt = td->td_ma_cnt;
1072 while (uio_clone->uio_resid != 0) {
1073 len = uio_clone->uio_iov->iov_len;
1075 KASSERT(uio_clone->uio_iovcnt >= 1,
1076 ("iovcnt underflow"));
1077 uio_clone->uio_iov++;
1078 uio_clone->uio_iovcnt--;
1081 if (len > io_hold_cnt * PAGE_SIZE)
1082 len = io_hold_cnt * PAGE_SIZE;
1083 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1084 end = round_page(addr + len);
1089 cnt = atop(end - trunc_page(addr));
1091 * A perfectly misaligned address and length could cause
1092 * both the start and the end of the chunk to use partial
1093 * page. +2 accounts for such a situation.
1095 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1096 addr, len, prot, ma, io_hold_cnt + 2);
1101 short_uio.uio_iov = &short_iovec[0];
1102 short_iovec[0].iov_base = (void *)addr;
1103 short_uio.uio_iovcnt = 1;
1104 short_uio.uio_resid = short_iovec[0].iov_len = len;
1105 short_uio.uio_offset = uio_clone->uio_offset;
1107 td->td_ma_cnt = cnt;
1109 error = vn_io_fault_doio(args, &short_uio, td);
1110 vm_page_unhold_pages(ma, cnt);
1111 adv = len - short_uio.uio_resid;
1113 uio_clone->uio_iov->iov_base =
1114 (char *)uio_clone->uio_iov->iov_base + adv;
1115 uio_clone->uio_iov->iov_len -= adv;
1116 uio_clone->uio_resid -= adv;
1117 uio_clone->uio_offset += adv;
1119 uio->uio_resid -= adv;
1120 uio->uio_offset += adv;
1122 if (error != 0 || adv == 0)
1125 td->td_ma = prev_td_ma;
1126 td->td_ma_cnt = prev_td_ma_cnt;
1127 curthread_pflags_restore(saveheld);
1129 vm_fault_enable_pagefaults(save);
1130 free(uio_clone, M_IOV);
1135 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1136 int flags, struct thread *td)
1141 struct vn_io_fault_args args;
1144 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1146 foffset_lock_uio(fp, uio, flags);
1147 if (do_vn_io_fault(vp, uio)) {
1148 args.kind = VN_IO_FAULT_FOP;
1149 args.args.fop_args.fp = fp;
1150 args.args.fop_args.doio = doio;
1151 args.cred = active_cred;
1152 args.flags = flags | FOF_OFFSET;
1153 if (uio->uio_rw == UIO_READ) {
1154 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1155 uio->uio_offset + uio->uio_resid);
1156 } else if ((fp->f_flag & O_APPEND) != 0 ||
1157 (flags & FOF_OFFSET) == 0) {
1158 /* For appenders, punt and lock the whole range. */
1159 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1161 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1162 uio->uio_offset + uio->uio_resid);
1164 error = vn_io_fault1(vp, uio, &args, td);
1165 vn_rangelock_unlock(vp, rl_cookie);
1167 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1169 foffset_unlock_uio(fp, uio, flags);
1174 * Helper function to perform the requested uiomove operation using
1175 * the held pages for io->uio_iov[0].iov_base buffer instead of
1176 * copyin/copyout. Access to the pages with uiomove_fromphys()
1177 * instead of iov_base prevents page faults that could occur due to
1178 * pmap_collect() invalidating the mapping created by
1179 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1180 * object cleanup revoking the write access from page mappings.
1182 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1183 * instead of plain uiomove().
1186 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1188 struct uio transp_uio;
1189 struct iovec transp_iov[1];
1195 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1196 uio->uio_segflg != UIO_USERSPACE)
1197 return (uiomove(data, xfersize, uio));
1199 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1200 transp_iov[0].iov_base = data;
1201 transp_uio.uio_iov = &transp_iov[0];
1202 transp_uio.uio_iovcnt = 1;
1203 if (xfersize > uio->uio_resid)
1204 xfersize = uio->uio_resid;
1205 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1206 transp_uio.uio_offset = 0;
1207 transp_uio.uio_segflg = UIO_SYSSPACE;
1209 * Since transp_iov points to data, and td_ma page array
1210 * corresponds to original uio->uio_iov, we need to invert the
1211 * direction of the i/o operation as passed to
1212 * uiomove_fromphys().
1214 switch (uio->uio_rw) {
1216 transp_uio.uio_rw = UIO_READ;
1219 transp_uio.uio_rw = UIO_WRITE;
1222 transp_uio.uio_td = uio->uio_td;
1223 error = uiomove_fromphys(td->td_ma,
1224 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1225 xfersize, &transp_uio);
1226 adv = xfersize - transp_uio.uio_resid;
1228 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1229 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1231 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1233 td->td_ma_cnt -= pgadv;
1234 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1235 uio->uio_iov->iov_len -= adv;
1236 uio->uio_resid -= adv;
1237 uio->uio_offset += adv;
1242 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1246 vm_offset_t iov_base;
1250 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1251 uio->uio_segflg != UIO_USERSPACE)
1252 return (uiomove_fromphys(ma, offset, xfersize, uio));
1254 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1255 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1256 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1257 switch (uio->uio_rw) {
1259 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1263 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1267 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1269 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1271 td->td_ma_cnt -= pgadv;
1272 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1273 uio->uio_iov->iov_len -= cnt;
1274 uio->uio_resid -= cnt;
1275 uio->uio_offset += cnt;
1281 * File table truncate routine.
1284 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1296 * Lock the whole range for truncation. Otherwise split i/o
1297 * might happen partly before and partly after the truncation.
1299 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1300 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1303 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1304 if (vp->v_type == VDIR) {
1309 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1313 error = vn_writechk(vp);
1316 vattr.va_size = length;
1317 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1321 vn_finished_write(mp);
1323 vn_rangelock_unlock(vp, rl_cookie);
1328 * File table vnode stat routine.
1331 vn_statfile(fp, sb, active_cred, td)
1334 struct ucred *active_cred;
1337 struct vnode *vp = fp->f_vnode;
1340 vn_lock(vp, LK_SHARED | LK_RETRY);
1341 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1348 * Stat a vnode; implementation for the stat syscall
1351 vn_stat(vp, sb, active_cred, file_cred, td)
1353 register struct stat *sb;
1354 struct ucred *active_cred;
1355 struct ucred *file_cred;
1359 register struct vattr *vap;
1364 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1372 * Initialize defaults for new and unusual fields, so that file
1373 * systems which don't support these fields don't need to know
1376 vap->va_birthtime.tv_sec = -1;
1377 vap->va_birthtime.tv_nsec = 0;
1378 vap->va_fsid = VNOVAL;
1379 vap->va_rdev = NODEV;
1381 error = VOP_GETATTR(vp, vap, active_cred);
1386 * Zero the spare stat fields
1388 bzero(sb, sizeof *sb);
1391 * Copy from vattr table
1393 if (vap->va_fsid != VNOVAL)
1394 sb->st_dev = vap->va_fsid;
1396 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1397 sb->st_ino = vap->va_fileid;
1398 mode = vap->va_mode;
1399 switch (vap->va_type) {
1425 sb->st_nlink = vap->va_nlink;
1426 sb->st_uid = vap->va_uid;
1427 sb->st_gid = vap->va_gid;
1428 sb->st_rdev = vap->va_rdev;
1429 if (vap->va_size > OFF_MAX)
1431 sb->st_size = vap->va_size;
1432 sb->st_atim = vap->va_atime;
1433 sb->st_mtim = vap->va_mtime;
1434 sb->st_ctim = vap->va_ctime;
1435 sb->st_birthtim = vap->va_birthtime;
1438 * According to www.opengroup.org, the meaning of st_blksize is
1439 * "a filesystem-specific preferred I/O block size for this
1440 * object. In some filesystem types, this may vary from file
1442 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1445 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1447 sb->st_flags = vap->va_flags;
1448 if (priv_check(td, PRIV_VFS_GENERATION))
1451 sb->st_gen = vap->va_gen;
1453 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1458 * File table vnode ioctl routine.
1461 vn_ioctl(fp, com, data, active_cred, td)
1465 struct ucred *active_cred;
1473 switch (vp->v_type) {
1478 vn_lock(vp, LK_SHARED | LK_RETRY);
1479 error = VOP_GETATTR(vp, &vattr, active_cred);
1482 *(int *)data = vattr.va_size - fp->f_offset;
1488 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1497 * File table vnode poll routine.
1500 vn_poll(fp, events, active_cred, td)
1503 struct ucred *active_cred;
1511 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1512 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1517 error = VOP_POLL(vp, events, fp->f_cred, td);
1522 * Acquire the requested lock and then check for validity. LK_RETRY
1523 * permits vn_lock to return doomed vnodes.
1526 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1530 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1531 ("vn_lock called with no locktype."));
1533 #ifdef DEBUG_VFS_LOCKS
1534 KASSERT(vp->v_holdcnt != 0,
1535 ("vn_lock %p: zero hold count", vp));
1537 error = VOP_LOCK1(vp, flags, file, line);
1538 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1539 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1540 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1543 * Callers specify LK_RETRY if they wish to get dead vnodes.
1544 * If RETRY is not set, we return ENOENT instead.
1546 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1547 (flags & LK_RETRY) == 0) {
1552 } while (flags & LK_RETRY && error != 0);
1557 * File table vnode close routine.
1560 vn_closefile(fp, td)
1569 fp->f_ops = &badfileops;
1571 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1574 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1576 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1577 lf.l_whence = SEEK_SET;
1580 lf.l_type = F_UNLCK;
1581 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1588 vn_suspendable_mp(struct mount *mp)
1591 return ((mp->mnt_kern_flag & MNTK_SUSPENDABLE) != 0);
1595 vn_suspendable(struct vnode *vp, struct mount **mpp)
1603 return (vn_suspendable_mp(*mpp));
1607 * Preparing to start a filesystem write operation. If the operation is
1608 * permitted, then we bump the count of operations in progress and
1609 * proceed. If a suspend request is in progress, we wait until the
1610 * suspension is over, and then proceed.
1613 vn_start_write_locked(struct mount *mp, int flags)
1617 mtx_assert(MNT_MTX(mp), MA_OWNED);
1621 * Check on status of suspension.
1623 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1624 mp->mnt_susp_owner != curthread) {
1625 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
1626 (flags & PCATCH) : 0) | (PUSER - 1);
1627 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1628 if (flags & V_NOWAIT) {
1629 error = EWOULDBLOCK;
1632 error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
1638 if (flags & V_XSLEEP)
1640 mp->mnt_writeopcount++;
1642 if (error != 0 || (flags & V_XSLEEP) != 0)
1649 vn_start_write(vp, mpp, flags)
1657 if (!vn_suspendable(vp, mpp))
1662 * If a vnode is provided, get and return the mount point that
1663 * to which it will write.
1666 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1668 if (error != EOPNOTSUPP)
1673 if ((mp = *mpp) == NULL)
1677 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1679 * As long as a vnode is not provided we need to acquire a
1680 * refcount for the provided mountpoint too, in order to
1681 * emulate a vfs_ref().
1687 return (vn_start_write_locked(mp, flags));
1691 * Secondary suspension. Used by operations such as vop_inactive
1692 * routines that are needed by the higher level functions. These
1693 * are allowed to proceed until all the higher level functions have
1694 * completed (indicated by mnt_writeopcount dropping to zero). At that
1695 * time, these operations are halted until the suspension is over.
1698 vn_start_secondary_write(vp, mpp, flags)
1706 if (!vn_suspendable(vp, mpp))
1711 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1713 if (error != EOPNOTSUPP)
1719 * If we are not suspended or have not yet reached suspended
1720 * mode, then let the operation proceed.
1722 if ((mp = *mpp) == NULL)
1726 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1728 * As long as a vnode is not provided we need to acquire a
1729 * refcount for the provided mountpoint too, in order to
1730 * emulate a vfs_ref().
1735 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1736 mp->mnt_secondary_writes++;
1737 mp->mnt_secondary_accwrites++;
1741 if (flags & V_NOWAIT) {
1744 return (EWOULDBLOCK);
1747 * Wait for the suspension to finish.
1749 error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
1750 ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
1759 * Filesystem write operation has completed. If we are suspending and this
1760 * operation is the last one, notify the suspender that the suspension is
1764 vn_finished_write(mp)
1767 if (mp == NULL || !vn_suspendable_mp(mp))
1771 mp->mnt_writeopcount--;
1772 if (mp->mnt_writeopcount < 0)
1773 panic("vn_finished_write: neg cnt");
1774 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1775 mp->mnt_writeopcount <= 0)
1776 wakeup(&mp->mnt_writeopcount);
1782 * Filesystem secondary write operation has completed. If we are
1783 * suspending and this operation is the last one, notify the suspender
1784 * that the suspension is now in effect.
1787 vn_finished_secondary_write(mp)
1790 if (mp == NULL || !vn_suspendable_mp(mp))
1794 mp->mnt_secondary_writes--;
1795 if (mp->mnt_secondary_writes < 0)
1796 panic("vn_finished_secondary_write: neg cnt");
1797 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1798 mp->mnt_secondary_writes <= 0)
1799 wakeup(&mp->mnt_secondary_writes);
1806 * Request a filesystem to suspend write operations.
1809 vfs_write_suspend(struct mount *mp, int flags)
1813 MPASS(vn_suspendable_mp(mp));
1816 if (mp->mnt_susp_owner == curthread) {
1820 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1821 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1824 * Unmount holds a write reference on the mount point. If we
1825 * own busy reference and drain for writers, we deadlock with
1826 * the reference draining in the unmount path. Callers of
1827 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1828 * vfs_busy() reference is owned and caller is not in the
1831 if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1832 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1837 mp->mnt_kern_flag |= MNTK_SUSPEND;
1838 mp->mnt_susp_owner = curthread;
1839 if (mp->mnt_writeopcount > 0)
1840 (void) msleep(&mp->mnt_writeopcount,
1841 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1844 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1845 vfs_write_resume(mp, 0);
1850 * Request a filesystem to resume write operations.
1853 vfs_write_resume(struct mount *mp, int flags)
1856 MPASS(vn_suspendable_mp(mp));
1859 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1860 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1861 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1863 mp->mnt_susp_owner = NULL;
1864 wakeup(&mp->mnt_writeopcount);
1865 wakeup(&mp->mnt_flag);
1866 curthread->td_pflags &= ~TDP_IGNSUSP;
1867 if ((flags & VR_START_WRITE) != 0) {
1869 mp->mnt_writeopcount++;
1872 if ((flags & VR_NO_SUSPCLR) == 0)
1874 } else if ((flags & VR_START_WRITE) != 0) {
1876 vn_start_write_locked(mp, 0);
1883 * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1887 vfs_write_suspend_umnt(struct mount *mp)
1891 MPASS(vn_suspendable_mp(mp));
1892 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1893 ("vfs_write_suspend_umnt: recursed"));
1895 /* dounmount() already called vn_start_write(). */
1897 vn_finished_write(mp);
1898 error = vfs_write_suspend(mp, 0);
1900 vn_start_write(NULL, &mp, V_WAIT);
1904 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1907 vn_start_write(NULL, &mp, V_WAIT);
1909 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1910 wakeup(&mp->mnt_flag);
1912 curthread->td_pflags |= TDP_IGNSUSP;
1917 * Implement kqueues for files by translating it to vnode operation.
1920 vn_kqfilter(struct file *fp, struct knote *kn)
1923 return (VOP_KQFILTER(fp->f_vnode, kn));
1927 * Simplified in-kernel wrapper calls for extended attribute access.
1928 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1929 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1932 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1933 const char *attrname, int *buflen, char *buf, struct thread *td)
1939 iov.iov_len = *buflen;
1942 auio.uio_iov = &iov;
1943 auio.uio_iovcnt = 1;
1944 auio.uio_rw = UIO_READ;
1945 auio.uio_segflg = UIO_SYSSPACE;
1947 auio.uio_offset = 0;
1948 auio.uio_resid = *buflen;
1950 if ((ioflg & IO_NODELOCKED) == 0)
1951 vn_lock(vp, LK_SHARED | LK_RETRY);
1953 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1955 /* authorize attribute retrieval as kernel */
1956 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1959 if ((ioflg & IO_NODELOCKED) == 0)
1963 *buflen = *buflen - auio.uio_resid;
1970 * XXX failure mode if partially written?
1973 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1974 const char *attrname, int buflen, char *buf, struct thread *td)
1981 iov.iov_len = buflen;
1984 auio.uio_iov = &iov;
1985 auio.uio_iovcnt = 1;
1986 auio.uio_rw = UIO_WRITE;
1987 auio.uio_segflg = UIO_SYSSPACE;
1989 auio.uio_offset = 0;
1990 auio.uio_resid = buflen;
1992 if ((ioflg & IO_NODELOCKED) == 0) {
1993 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1995 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1998 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2000 /* authorize attribute setting as kernel */
2001 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
2003 if ((ioflg & IO_NODELOCKED) == 0) {
2004 vn_finished_write(mp);
2012 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
2013 const char *attrname, struct thread *td)
2018 if ((ioflg & IO_NODELOCKED) == 0) {
2019 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2021 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2024 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2026 /* authorize attribute removal as kernel */
2027 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2028 if (error == EOPNOTSUPP)
2029 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2032 if ((ioflg & IO_NODELOCKED) == 0) {
2033 vn_finished_write(mp);
2041 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2045 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2049 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2052 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2057 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2058 int lkflags, struct vnode **rvp)
2063 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2065 ltype = VOP_ISLOCKED(vp);
2066 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2067 ("vn_vget_ino: vp not locked"));
2068 error = vfs_busy(mp, MBF_NOWAIT);
2072 error = vfs_busy(mp, 0);
2073 vn_lock(vp, ltype | LK_RETRY);
2077 if (vp->v_iflag & VI_DOOMED) {
2083 error = alloc(mp, alloc_arg, lkflags, rvp);
2086 vn_lock(vp, ltype | LK_RETRY);
2087 if (vp->v_iflag & VI_DOOMED) {
2100 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2101 const struct thread *td)
2104 if (vp->v_type != VREG || td == NULL)
2106 PROC_LOCK(td->td_proc);
2107 if ((uoff_t)uio->uio_offset + uio->uio_resid >
2108 lim_cur(td->td_proc, RLIMIT_FSIZE)) {
2109 kern_psignal(td->td_proc, SIGXFSZ);
2110 PROC_UNLOCK(td->td_proc);
2113 PROC_UNLOCK(td->td_proc);
2118 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2125 vn_lock(vp, LK_SHARED | LK_RETRY);
2126 AUDIT_ARG_VNODE1(vp);
2129 return (setfmode(td, active_cred, vp, mode));
2133 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2140 vn_lock(vp, LK_SHARED | LK_RETRY);
2141 AUDIT_ARG_VNODE1(vp);
2144 return (setfown(td, active_cred, vp, uid, gid));
2148 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2152 if ((object = vp->v_object) == NULL)
2154 VM_OBJECT_WLOCK(object);
2155 vm_object_page_remove(object, start, end, 0);
2156 VM_OBJECT_WUNLOCK(object);
2160 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2168 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2169 ("Wrong command %lu", cmd));
2171 if (vn_lock(vp, LK_SHARED) != 0)
2173 if (vp->v_type != VREG) {
2177 error = VOP_GETATTR(vp, &va, cred);
2181 if (noff >= va.va_size) {
2185 bsize = vp->v_mount->mnt_stat.f_iosize;
2186 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2187 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2188 if (error == EOPNOTSUPP) {
2192 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2193 (bnp != -1 && cmd == FIOSEEKDATA)) {
2200 if (noff > va.va_size)
2202 /* noff == va.va_size. There is an implicit hole at the end of file. */
2203 if (cmd == FIOSEEKDATA)
2213 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2218 off_t foffset, size;
2221 cred = td->td_ucred;
2223 foffset = foffset_lock(fp, 0);
2224 noneg = (vp->v_type != VCHR);
2230 (offset > 0 && foffset > OFF_MAX - offset))) {
2237 vn_lock(vp, LK_SHARED | LK_RETRY);
2238 error = VOP_GETATTR(vp, &vattr, cred);
2244 * If the file references a disk device, then fetch
2245 * the media size and use that to determine the ending
2248 if (vattr.va_size == 0 && vp->v_type == VCHR &&
2249 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2250 vattr.va_size = size;
2252 (vattr.va_size > OFF_MAX ||
2253 (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2257 offset += vattr.va_size;
2262 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2265 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2270 if (error == 0 && noneg && offset < 0)
2274 VFS_KNOTE_UNLOCKED(vp, 0);
2275 td->td_uretoff.tdu_off = offset;
2277 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2282 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2288 * Grant permission if the caller is the owner of the file, or
2289 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2290 * on the file. If the time pointer is null, then write
2291 * permission on the file is also sufficient.
2293 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2294 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2295 * will be allowed to set the times [..] to the current
2298 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2299 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2300 error = VOP_ACCESS(vp, VWRITE, cred, td);
2305 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2310 if (fp->f_type == DTYPE_FIFO)
2311 kif->kf_type = KF_TYPE_FIFO;
2313 kif->kf_type = KF_TYPE_VNODE;
2316 FILEDESC_SUNLOCK(fdp);
2317 error = vn_fill_kinfo_vnode(vp, kif);
2319 FILEDESC_SLOCK(fdp);
2324 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2327 char *fullpath, *freepath;
2330 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
2333 error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2335 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2337 if (freepath != NULL)
2338 free(freepath, M_TEMP);
2341 * Retrieve vnode attributes.
2343 va.va_fsid = VNOVAL;
2345 vn_lock(vp, LK_SHARED | LK_RETRY);
2346 error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2350 if (va.va_fsid != VNOVAL)
2351 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2353 kif->kf_un.kf_file.kf_file_fsid =
2354 vp->v_mount->mnt_stat.f_fsid.val[0];
2355 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2356 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2357 kif->kf_un.kf_file.kf_file_size = va.va_size;
2358 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;