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>
58 #include <sys/mount.h>
59 #include <sys/mutex.h>
60 #include <sys/namei.h>
61 #include <sys/vnode.h>
64 #include <sys/filio.h>
65 #include <sys/resourcevar.h>
66 #include <sys/rwlock.h>
68 #include <sys/sysctl.h>
69 #include <sys/ttycom.h>
71 #include <sys/syslog.h>
72 #include <sys/unistd.h>
75 #include <security/audit/audit.h>
76 #include <security/mac/mac_framework.h>
79 #include <vm/vm_extern.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_object.h>
83 #include <vm/vm_page.h>
84 #include <vm/vnode_pager.h>
86 static fo_rdwr_t vn_read;
87 static fo_rdwr_t vn_write;
88 static fo_rdwr_t vn_io_fault;
89 static fo_truncate_t vn_truncate;
90 static fo_ioctl_t vn_ioctl;
91 static fo_poll_t vn_poll;
92 static fo_kqfilter_t vn_kqfilter;
93 static fo_stat_t vn_statfile;
94 static fo_close_t vn_closefile;
95 static fo_mmap_t vn_mmap;
97 struct fileops vnops = {
98 .fo_read = vn_io_fault,
99 .fo_write = vn_io_fault,
100 .fo_truncate = vn_truncate,
101 .fo_ioctl = vn_ioctl,
103 .fo_kqfilter = vn_kqfilter,
104 .fo_stat = vn_statfile,
105 .fo_close = vn_closefile,
106 .fo_chmod = vn_chmod,
107 .fo_chown = vn_chown,
108 .fo_sendfile = vn_sendfile,
110 .fo_fill_kinfo = vn_fill_kinfo,
112 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
115 static const int io_hold_cnt = 16;
116 static int vn_io_fault_enable = 1;
117 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
118 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
119 static u_long vn_io_faults_cnt;
120 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
121 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
124 * Returns true if vn_io_fault mode of handling the i/o request should
128 do_vn_io_fault(struct vnode *vp, struct uio *uio)
132 return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
133 (mp = vp->v_mount) != NULL &&
134 (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
138 * Structure used to pass arguments to vn_io_fault1(), to do either
139 * file- or vnode-based I/O calls.
141 struct vn_io_fault_args {
149 struct fop_args_tag {
153 struct vop_args_tag {
159 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
160 struct vn_io_fault_args *args, struct thread *td);
163 vn_open(ndp, flagp, cmode, fp)
164 struct nameidata *ndp;
168 struct thread *td = ndp->ni_cnd.cn_thread;
170 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
174 * Common code for vnode open operations via a name lookup.
175 * Lookup the vnode and invoke VOP_CREATE if needed.
176 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
178 * Note that this does NOT free nameidata for the successful case,
179 * due to the NDINIT being done elsewhere.
182 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
183 struct ucred *cred, struct file *fp)
187 struct thread *td = ndp->ni_cnd.cn_thread;
189 struct vattr *vap = &vat;
194 if (fmode & O_CREAT) {
195 ndp->ni_cnd.cn_nameiop = CREATE;
197 * Set NOCACHE to avoid flushing the cache when
198 * rolling in many files at once.
200 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE;
201 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
202 ndp->ni_cnd.cn_flags |= FOLLOW;
203 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
204 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
205 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
206 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
208 if ((error = namei(ndp)) != 0)
210 if (ndp->ni_vp == NULL) {
213 vap->va_mode = cmode;
215 vap->va_vaflags |= VA_EXCLUSIVE;
216 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
217 NDFREE(ndp, NDF_ONLY_PNBUF);
219 if ((error = vn_start_write(NULL, &mp,
220 V_XSLEEP | PCATCH)) != 0)
224 if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0)
225 ndp->ni_cnd.cn_flags |= MAKEENTRY;
227 error = mac_vnode_check_create(cred, ndp->ni_dvp,
231 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
234 vn_finished_write(mp);
236 NDFREE(ndp, NDF_ONLY_PNBUF);
242 if (ndp->ni_dvp == ndp->ni_vp)
248 if (fmode & O_EXCL) {
255 ndp->ni_cnd.cn_nameiop = LOOKUP;
256 ndp->ni_cnd.cn_flags = ISOPEN |
257 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
258 if (!(fmode & FWRITE))
259 ndp->ni_cnd.cn_flags |= LOCKSHARED;
260 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
261 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
262 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
263 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
264 if ((error = namei(ndp)) != 0)
268 error = vn_open_vnode(vp, fmode, cred, td, fp);
274 NDFREE(ndp, NDF_ONLY_PNBUF);
282 * Common code for vnode open operations once a vnode is located.
283 * Check permissions, and call the VOP_OPEN routine.
286 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
287 struct thread *td, struct file *fp)
292 int error, have_flock, lock_flags, type;
294 if (vp->v_type == VLNK)
296 if (vp->v_type == VSOCK)
298 if (vp->v_type != VDIR && fmode & O_DIRECTORY)
301 if (fmode & (FWRITE | O_TRUNC)) {
302 if (vp->v_type == VDIR)
310 if ((fmode & O_APPEND) && (fmode & FWRITE))
315 if (fmode & O_VERIFY)
317 error = mac_vnode_check_open(cred, vp, accmode);
321 accmode &= ~(VCREAT | VVERIFY);
323 if ((fmode & O_CREAT) == 0) {
324 if (accmode & VWRITE) {
325 error = vn_writechk(vp);
330 error = VOP_ACCESS(vp, accmode, cred, td);
335 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
336 vn_lock(vp, LK_UPGRADE | LK_RETRY);
337 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
340 if (fmode & (O_EXLOCK | O_SHLOCK)) {
341 KASSERT(fp != NULL, ("open with flock requires fp"));
342 lock_flags = VOP_ISLOCKED(vp);
344 lf.l_whence = SEEK_SET;
347 if (fmode & O_EXLOCK)
352 if ((fmode & FNONBLOCK) == 0)
354 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
355 have_flock = (error == 0);
356 vn_lock(vp, lock_flags | LK_RETRY);
357 if (error == 0 && vp->v_iflag & VI_DOOMED)
360 * Another thread might have used this vnode as an
361 * executable while the vnode lock was dropped.
362 * Ensure the vnode is still able to be opened for
363 * writing after the lock has been obtained.
365 if (error == 0 && accmode & VWRITE)
366 error = vn_writechk(vp);
370 lf.l_whence = SEEK_SET;
374 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
377 vn_start_write(vp, &mp, V_WAIT);
378 vn_lock(vp, lock_flags | LK_RETRY);
379 (void)VOP_CLOSE(vp, fmode, cred, td);
380 vn_finished_write(mp);
381 /* Prevent second close from fdrop()->vn_close(). */
383 fp->f_ops= &badfileops;
386 fp->f_flag |= FHASLOCK;
388 if (fmode & FWRITE) {
389 VOP_ADD_WRITECOUNT(vp, 1);
390 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
391 __func__, vp, vp->v_writecount);
393 ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
398 * Check for write permissions on the specified vnode.
399 * Prototype text segments cannot be written.
403 register struct vnode *vp;
406 ASSERT_VOP_LOCKED(vp, "vn_writechk");
408 * If there's shared text associated with
409 * the vnode, try to free it up once. If
410 * we fail, we can't allow writing.
422 vn_close(vp, flags, file_cred, td)
423 register struct vnode *vp;
425 struct ucred *file_cred;
429 int error, lock_flags;
431 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
432 MNT_EXTENDED_SHARED(vp->v_mount))
433 lock_flags = LK_SHARED;
435 lock_flags = LK_EXCLUSIVE;
437 vn_start_write(vp, &mp, V_WAIT);
438 vn_lock(vp, lock_flags | LK_RETRY);
439 if (flags & FWRITE) {
440 VNASSERT(vp->v_writecount > 0, vp,
441 ("vn_close: negative writecount"));
442 VOP_ADD_WRITECOUNT(vp, -1);
443 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
444 __func__, vp, vp->v_writecount);
446 error = VOP_CLOSE(vp, flags, file_cred, td);
448 vn_finished_write(mp);
453 * Heuristic to detect sequential operation.
456 sequential_heuristic(struct uio *uio, struct file *fp)
459 ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
460 if (fp->f_flag & FRDAHEAD)
461 return (fp->f_seqcount << IO_SEQSHIFT);
464 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
465 * that the first I/O is normally considered to be slightly
466 * sequential. Seeking to offset 0 doesn't change sequentiality
467 * unless previous seeks have reduced f_seqcount to 0, in which
468 * case offset 0 is not special.
470 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
471 uio->uio_offset == fp->f_nextoff) {
473 * f_seqcount is in units of fixed-size blocks so that it
474 * depends mainly on the amount of sequential I/O and not
475 * much on the number of sequential I/O's. The fixed size
476 * of 16384 is hard-coded here since it is (not quite) just
477 * a magic size that works well here. This size is more
478 * closely related to the best I/O size for real disks than
479 * to any block size used by software.
481 fp->f_seqcount += howmany(uio->uio_resid, 16384);
482 if (fp->f_seqcount > IO_SEQMAX)
483 fp->f_seqcount = IO_SEQMAX;
484 return (fp->f_seqcount << IO_SEQSHIFT);
487 /* Not sequential. Quickly draw-down sequentiality. */
488 if (fp->f_seqcount > 1)
496 * Package up an I/O request on a vnode into a uio and do it.
499 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
500 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
501 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
508 struct vn_io_fault_args args;
509 int error, lock_flags;
511 auio.uio_iov = &aiov;
513 aiov.iov_base = base;
515 auio.uio_resid = len;
516 auio.uio_offset = offset;
517 auio.uio_segflg = segflg;
522 if ((ioflg & IO_NODELOCKED) == 0) {
523 if ((ioflg & IO_RANGELOCKED) == 0) {
524 if (rw == UIO_READ) {
525 rl_cookie = vn_rangelock_rlock(vp, offset,
528 rl_cookie = vn_rangelock_wlock(vp, offset,
534 if (rw == UIO_WRITE) {
535 if (vp->v_type != VCHR &&
536 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
539 if (MNT_SHARED_WRITES(mp) ||
540 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
541 lock_flags = LK_SHARED;
543 lock_flags = LK_EXCLUSIVE;
545 lock_flags = LK_SHARED;
546 vn_lock(vp, lock_flags | LK_RETRY);
550 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
552 if ((ioflg & IO_NOMACCHECK) == 0) {
554 error = mac_vnode_check_read(active_cred, file_cred,
557 error = mac_vnode_check_write(active_cred, file_cred,
562 if (file_cred != NULL)
566 if (do_vn_io_fault(vp, &auio)) {
567 args.kind = VN_IO_FAULT_VOP;
570 args.args.vop_args.vp = vp;
571 error = vn_io_fault1(vp, &auio, &args, td);
572 } else if (rw == UIO_READ) {
573 error = VOP_READ(vp, &auio, ioflg, cred);
574 } else /* if (rw == UIO_WRITE) */ {
575 error = VOP_WRITE(vp, &auio, ioflg, cred);
579 *aresid = auio.uio_resid;
581 if (auio.uio_resid && error == 0)
583 if ((ioflg & IO_NODELOCKED) == 0) {
586 vn_finished_write(mp);
589 if (rl_cookie != NULL)
590 vn_rangelock_unlock(vp, rl_cookie);
595 * Package up an I/O request on a vnode into a uio and do it. The I/O
596 * request is split up into smaller chunks and we try to avoid saturating
597 * the buffer cache while potentially holding a vnode locked, so we
598 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
599 * to give other processes a chance to lock the vnode (either other processes
600 * core'ing the same binary, or unrelated processes scanning the directory).
603 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
604 file_cred, aresid, td)
612 struct ucred *active_cred;
613 struct ucred *file_cred;
624 * Force `offset' to a multiple of MAXBSIZE except possibly
625 * for the first chunk, so that filesystems only need to
626 * write full blocks except possibly for the first and last
629 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
633 if (rw != UIO_READ && vp->v_type == VREG)
636 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
637 ioflg, active_cred, file_cred, &iaresid, td);
638 len -= chunk; /* aresid calc already includes length */
642 base = (char *)base + chunk;
643 kern_yield(PRI_USER);
646 *aresid = len + iaresid;
651 foffset_lock(struct file *fp, int flags)
656 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
658 #if OFF_MAX <= LONG_MAX
660 * Caller only wants the current f_offset value. Assume that
661 * the long and shorter integer types reads are atomic.
663 if ((flags & FOF_NOLOCK) != 0)
664 return (fp->f_offset);
668 * According to McKusick the vn lock was protecting f_offset here.
669 * It is now protected by the FOFFSET_LOCKED flag.
671 mtxp = mtx_pool_find(mtxpool_sleep, fp);
673 if ((flags & FOF_NOLOCK) == 0) {
674 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
675 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
676 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
679 fp->f_vnread_flags |= FOFFSET_LOCKED;
687 foffset_unlock(struct file *fp, off_t val, int flags)
691 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
693 #if OFF_MAX <= LONG_MAX
694 if ((flags & FOF_NOLOCK) != 0) {
695 if ((flags & FOF_NOUPDATE) == 0)
697 if ((flags & FOF_NEXTOFF) != 0)
703 mtxp = mtx_pool_find(mtxpool_sleep, fp);
705 if ((flags & FOF_NOUPDATE) == 0)
707 if ((flags & FOF_NEXTOFF) != 0)
709 if ((flags & FOF_NOLOCK) == 0) {
710 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
711 ("Lost FOFFSET_LOCKED"));
712 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
713 wakeup(&fp->f_vnread_flags);
714 fp->f_vnread_flags = 0;
720 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
723 if ((flags & FOF_OFFSET) == 0)
724 uio->uio_offset = foffset_lock(fp, flags);
728 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
731 if ((flags & FOF_OFFSET) == 0)
732 foffset_unlock(fp, uio->uio_offset, flags);
736 get_advice(struct file *fp, struct uio *uio)
741 ret = POSIX_FADV_NORMAL;
742 if (fp->f_advice == NULL)
745 mtxp = mtx_pool_find(mtxpool_sleep, fp);
747 if (uio->uio_offset >= fp->f_advice->fa_start &&
748 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
749 ret = fp->f_advice->fa_advice;
755 * File table vnode read routine.
758 vn_read(fp, uio, active_cred, flags, td)
761 struct ucred *active_cred;
769 off_t offset, start, end;
771 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
773 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
776 if (fp->f_flag & FNONBLOCK)
778 if (fp->f_flag & O_DIRECT)
780 advice = get_advice(fp, uio);
781 vn_lock(vp, LK_SHARED | LK_RETRY);
784 case POSIX_FADV_NORMAL:
785 case POSIX_FADV_SEQUENTIAL:
786 case POSIX_FADV_NOREUSE:
787 ioflag |= sequential_heuristic(uio, fp);
789 case POSIX_FADV_RANDOM:
790 /* Disable read-ahead for random I/O. */
793 offset = uio->uio_offset;
796 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
799 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
800 fp->f_nextoff = uio->uio_offset;
802 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
803 offset != uio->uio_offset) {
805 * Use POSIX_FADV_DONTNEED to flush clean pages and
806 * buffers for the backing file after a
807 * POSIX_FADV_NOREUSE read(2). To optimize the common
808 * case of using POSIX_FADV_NOREUSE with sequential
809 * access, track the previous implicit DONTNEED
810 * request and grow this request to include the
811 * current read(2) in addition to the previous
812 * DONTNEED. With purely sequential access this will
813 * cause the DONTNEED requests to continously grow to
814 * cover all of the previously read regions of the
815 * file. This allows filesystem blocks that are
816 * accessed by multiple calls to read(2) to be flushed
817 * once the last read(2) finishes.
820 end = uio->uio_offset - 1;
821 mtxp = mtx_pool_find(mtxpool_sleep, fp);
823 if (fp->f_advice != NULL &&
824 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
825 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
826 start = fp->f_advice->fa_prevstart;
827 else if (fp->f_advice->fa_prevstart != 0 &&
828 fp->f_advice->fa_prevstart == end + 1)
829 end = fp->f_advice->fa_prevend;
830 fp->f_advice->fa_prevstart = start;
831 fp->f_advice->fa_prevend = end;
834 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
840 * File table vnode write routine.
843 vn_write(fp, uio, active_cred, flags, td)
846 struct ucred *active_cred;
853 int error, ioflag, lock_flags;
855 off_t offset, start, end;
857 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
859 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
861 if (vp->v_type == VREG)
864 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
866 if (fp->f_flag & FNONBLOCK)
868 if (fp->f_flag & O_DIRECT)
870 if ((fp->f_flag & O_FSYNC) ||
871 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
874 if (vp->v_type != VCHR &&
875 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
878 advice = get_advice(fp, uio);
880 if (MNT_SHARED_WRITES(mp) ||
881 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
882 lock_flags = LK_SHARED;
884 lock_flags = LK_EXCLUSIVE;
887 vn_lock(vp, lock_flags | LK_RETRY);
889 case POSIX_FADV_NORMAL:
890 case POSIX_FADV_SEQUENTIAL:
891 case POSIX_FADV_NOREUSE:
892 ioflag |= sequential_heuristic(uio, fp);
894 case POSIX_FADV_RANDOM:
895 /* XXX: Is this correct? */
898 offset = uio->uio_offset;
901 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
904 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
905 fp->f_nextoff = uio->uio_offset;
907 if (vp->v_type != VCHR)
908 vn_finished_write(mp);
909 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
910 offset != uio->uio_offset) {
912 * Use POSIX_FADV_DONTNEED to flush clean pages and
913 * buffers for the backing file after a
914 * POSIX_FADV_NOREUSE write(2). To optimize the
915 * common case of using POSIX_FADV_NOREUSE with
916 * sequential access, track the previous implicit
917 * DONTNEED request and grow this request to include
918 * the current write(2) in addition to the previous
919 * DONTNEED. With purely sequential access this will
920 * cause the DONTNEED requests to continously grow to
921 * cover all of the previously written regions of the
924 * Note that the blocks just written are almost
925 * certainly still dirty, so this only works when
926 * VOP_ADVISE() calls from subsequent writes push out
927 * the data written by this write(2) once the backing
928 * buffers are clean. However, as compared to forcing
929 * IO_DIRECT, this gives much saner behavior. Write
930 * clustering is still allowed, and clean pages are
931 * merely moved to the cache page queue rather than
932 * outright thrown away. This means a subsequent
933 * read(2) can still avoid hitting the disk if the
934 * pages have not been reclaimed.
936 * This does make POSIX_FADV_NOREUSE largely useless
937 * with non-sequential access. However, sequential
938 * access is the more common use case and the flag is
942 end = uio->uio_offset - 1;
943 mtxp = mtx_pool_find(mtxpool_sleep, fp);
945 if (fp->f_advice != NULL &&
946 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
947 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
948 start = fp->f_advice->fa_prevstart;
949 else if (fp->f_advice->fa_prevstart != 0 &&
950 fp->f_advice->fa_prevstart == end + 1)
951 end = fp->f_advice->fa_prevend;
952 fp->f_advice->fa_prevstart = start;
953 fp->f_advice->fa_prevend = end;
956 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
964 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
965 * prevent the following deadlock:
967 * Assume that the thread A reads from the vnode vp1 into userspace
968 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
969 * currently not resident, then system ends up with the call chain
970 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
971 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
972 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
973 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
974 * backed by the pages of vnode vp1, and some page in buf2 is not
975 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
977 * To prevent the lock order reversal and deadlock, vn_io_fault() does
978 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
979 * Instead, it first tries to do the whole range i/o with pagefaults
980 * disabled. If all pages in the i/o buffer are resident and mapped,
981 * VOP will succeed (ignoring the genuine filesystem errors).
982 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
983 * i/o in chunks, with all pages in the chunk prefaulted and held
984 * using vm_fault_quick_hold_pages().
986 * Filesystems using this deadlock avoidance scheme should use the
987 * array of the held pages from uio, saved in the curthread->td_ma,
988 * instead of doing uiomove(). A helper function
989 * vn_io_fault_uiomove() converts uiomove request into
990 * uiomove_fromphys() over td_ma array.
992 * Since vnode locks do not cover the whole i/o anymore, rangelocks
993 * make the current i/o request atomic with respect to other i/os and
998 * Decode vn_io_fault_args and perform the corresponding i/o.
1001 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
1005 switch (args->kind) {
1006 case VN_IO_FAULT_FOP:
1007 return ((args->args.fop_args.doio)(args->args.fop_args.fp,
1008 uio, args->cred, args->flags, td));
1009 case VN_IO_FAULT_VOP:
1010 if (uio->uio_rw == UIO_READ) {
1011 return (VOP_READ(args->args.vop_args.vp, uio,
1012 args->flags, args->cred));
1013 } else if (uio->uio_rw == UIO_WRITE) {
1014 return (VOP_WRITE(args->args.vop_args.vp, uio,
1015 args->flags, args->cred));
1019 panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
1024 * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1025 * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1026 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1027 * into args and call vn_io_fault1() to handle faults during the user
1028 * mode buffer accesses.
1031 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1034 vm_page_t ma[io_hold_cnt + 2];
1035 struct uio *uio_clone, short_uio;
1036 struct iovec short_iovec[1];
1037 vm_page_t *prev_td_ma;
1039 vm_offset_t addr, end;
1042 int error, cnt, save, saveheld, prev_td_ma_cnt;
1044 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1047 * The UFS follows IO_UNIT directive and replays back both
1048 * uio_offset and uio_resid if an error is encountered during the
1049 * operation. But, since the iovec may be already advanced,
1050 * uio is still in an inconsistent state.
1052 * Cache a copy of the original uio, which is advanced to the redo
1053 * point using UIO_NOCOPY below.
1055 uio_clone = cloneuio(uio);
1056 resid = uio->uio_resid;
1058 short_uio.uio_segflg = UIO_USERSPACE;
1059 short_uio.uio_rw = uio->uio_rw;
1060 short_uio.uio_td = uio->uio_td;
1062 save = vm_fault_disable_pagefaults();
1063 error = vn_io_fault_doio(args, uio, td);
1064 if (error != EFAULT)
1067 atomic_add_long(&vn_io_faults_cnt, 1);
1068 uio_clone->uio_segflg = UIO_NOCOPY;
1069 uiomove(NULL, resid - uio->uio_resid, uio_clone);
1070 uio_clone->uio_segflg = uio->uio_segflg;
1072 saveheld = curthread_pflags_set(TDP_UIOHELD);
1073 prev_td_ma = td->td_ma;
1074 prev_td_ma_cnt = td->td_ma_cnt;
1076 while (uio_clone->uio_resid != 0) {
1077 len = uio_clone->uio_iov->iov_len;
1079 KASSERT(uio_clone->uio_iovcnt >= 1,
1080 ("iovcnt underflow"));
1081 uio_clone->uio_iov++;
1082 uio_clone->uio_iovcnt--;
1085 if (len > io_hold_cnt * PAGE_SIZE)
1086 len = io_hold_cnt * PAGE_SIZE;
1087 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1088 end = round_page(addr + len);
1093 cnt = atop(end - trunc_page(addr));
1095 * A perfectly misaligned address and length could cause
1096 * both the start and the end of the chunk to use partial
1097 * page. +2 accounts for such a situation.
1099 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1100 addr, len, prot, ma, io_hold_cnt + 2);
1105 short_uio.uio_iov = &short_iovec[0];
1106 short_iovec[0].iov_base = (void *)addr;
1107 short_uio.uio_iovcnt = 1;
1108 short_uio.uio_resid = short_iovec[0].iov_len = len;
1109 short_uio.uio_offset = uio_clone->uio_offset;
1111 td->td_ma_cnt = cnt;
1113 error = vn_io_fault_doio(args, &short_uio, td);
1114 vm_page_unhold_pages(ma, cnt);
1115 adv = len - short_uio.uio_resid;
1117 uio_clone->uio_iov->iov_base =
1118 (char *)uio_clone->uio_iov->iov_base + adv;
1119 uio_clone->uio_iov->iov_len -= adv;
1120 uio_clone->uio_resid -= adv;
1121 uio_clone->uio_offset += adv;
1123 uio->uio_resid -= adv;
1124 uio->uio_offset += adv;
1126 if (error != 0 || adv == 0)
1129 td->td_ma = prev_td_ma;
1130 td->td_ma_cnt = prev_td_ma_cnt;
1131 curthread_pflags_restore(saveheld);
1133 vm_fault_enable_pagefaults(save);
1134 free(uio_clone, M_IOV);
1139 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1140 int flags, struct thread *td)
1145 struct vn_io_fault_args args;
1148 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1150 foffset_lock_uio(fp, uio, flags);
1151 if (do_vn_io_fault(vp, uio)) {
1152 args.kind = VN_IO_FAULT_FOP;
1153 args.args.fop_args.fp = fp;
1154 args.args.fop_args.doio = doio;
1155 args.cred = active_cred;
1156 args.flags = flags | FOF_OFFSET;
1157 if (uio->uio_rw == UIO_READ) {
1158 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1159 uio->uio_offset + uio->uio_resid);
1160 } else if ((fp->f_flag & O_APPEND) != 0 ||
1161 (flags & FOF_OFFSET) == 0) {
1162 /* For appenders, punt and lock the whole range. */
1163 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1165 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1166 uio->uio_offset + uio->uio_resid);
1168 error = vn_io_fault1(vp, uio, &args, td);
1169 vn_rangelock_unlock(vp, rl_cookie);
1171 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1173 foffset_unlock_uio(fp, uio, flags);
1178 * Helper function to perform the requested uiomove operation using
1179 * the held pages for io->uio_iov[0].iov_base buffer instead of
1180 * copyin/copyout. Access to the pages with uiomove_fromphys()
1181 * instead of iov_base prevents page faults that could occur due to
1182 * pmap_collect() invalidating the mapping created by
1183 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1184 * object cleanup revoking the write access from page mappings.
1186 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1187 * instead of plain uiomove().
1190 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1192 struct uio transp_uio;
1193 struct iovec transp_iov[1];
1199 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1200 uio->uio_segflg != UIO_USERSPACE)
1201 return (uiomove(data, xfersize, uio));
1203 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1204 transp_iov[0].iov_base = data;
1205 transp_uio.uio_iov = &transp_iov[0];
1206 transp_uio.uio_iovcnt = 1;
1207 if (xfersize > uio->uio_resid)
1208 xfersize = uio->uio_resid;
1209 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1210 transp_uio.uio_offset = 0;
1211 transp_uio.uio_segflg = UIO_SYSSPACE;
1213 * Since transp_iov points to data, and td_ma page array
1214 * corresponds to original uio->uio_iov, we need to invert the
1215 * direction of the i/o operation as passed to
1216 * uiomove_fromphys().
1218 switch (uio->uio_rw) {
1220 transp_uio.uio_rw = UIO_READ;
1223 transp_uio.uio_rw = UIO_WRITE;
1226 transp_uio.uio_td = uio->uio_td;
1227 error = uiomove_fromphys(td->td_ma,
1228 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1229 xfersize, &transp_uio);
1230 adv = xfersize - transp_uio.uio_resid;
1232 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1233 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1235 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1237 td->td_ma_cnt -= pgadv;
1238 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1239 uio->uio_iov->iov_len -= adv;
1240 uio->uio_resid -= adv;
1241 uio->uio_offset += adv;
1246 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1250 vm_offset_t iov_base;
1254 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1255 uio->uio_segflg != UIO_USERSPACE)
1256 return (uiomove_fromphys(ma, offset, xfersize, uio));
1258 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1259 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1260 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1261 switch (uio->uio_rw) {
1263 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1267 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1271 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1273 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1275 td->td_ma_cnt -= pgadv;
1276 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1277 uio->uio_iov->iov_len -= cnt;
1278 uio->uio_resid -= cnt;
1279 uio->uio_offset += cnt;
1285 * File table truncate routine.
1288 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1300 * Lock the whole range for truncation. Otherwise split i/o
1301 * might happen partly before and partly after the truncation.
1303 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1304 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1307 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1308 if (vp->v_type == VDIR) {
1313 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1317 error = vn_writechk(vp);
1320 vattr.va_size = length;
1321 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1325 vn_finished_write(mp);
1327 vn_rangelock_unlock(vp, rl_cookie);
1332 * File table vnode stat routine.
1335 vn_statfile(fp, sb, active_cred, td)
1338 struct ucred *active_cred;
1341 struct vnode *vp = fp->f_vnode;
1344 vn_lock(vp, LK_SHARED | LK_RETRY);
1345 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1352 * Stat a vnode; implementation for the stat syscall
1355 vn_stat(vp, sb, active_cred, file_cred, td)
1357 register struct stat *sb;
1358 struct ucred *active_cred;
1359 struct ucred *file_cred;
1363 register struct vattr *vap;
1368 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1376 * Initialize defaults for new and unusual fields, so that file
1377 * systems which don't support these fields don't need to know
1380 vap->va_birthtime.tv_sec = -1;
1381 vap->va_birthtime.tv_nsec = 0;
1382 vap->va_fsid = VNOVAL;
1383 vap->va_rdev = NODEV;
1385 error = VOP_GETATTR(vp, vap, active_cred);
1390 * Zero the spare stat fields
1392 bzero(sb, sizeof *sb);
1395 * Copy from vattr table
1397 if (vap->va_fsid != VNOVAL)
1398 sb->st_dev = vap->va_fsid;
1400 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1401 sb->st_ino = vap->va_fileid;
1402 mode = vap->va_mode;
1403 switch (vap->va_type) {
1429 sb->st_nlink = vap->va_nlink;
1430 sb->st_uid = vap->va_uid;
1431 sb->st_gid = vap->va_gid;
1432 sb->st_rdev = vap->va_rdev;
1433 if (vap->va_size > OFF_MAX)
1435 sb->st_size = vap->va_size;
1436 sb->st_atim = vap->va_atime;
1437 sb->st_mtim = vap->va_mtime;
1438 sb->st_ctim = vap->va_ctime;
1439 sb->st_birthtim = vap->va_birthtime;
1442 * According to www.opengroup.org, the meaning of st_blksize is
1443 * "a filesystem-specific preferred I/O block size for this
1444 * object. In some filesystem types, this may vary from file
1446 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1449 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1451 sb->st_flags = vap->va_flags;
1452 if (priv_check(td, PRIV_VFS_GENERATION))
1455 sb->st_gen = vap->va_gen;
1457 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1462 * File table vnode ioctl routine.
1465 vn_ioctl(fp, com, data, active_cred, td)
1469 struct ucred *active_cred;
1477 switch (vp->v_type) {
1482 vn_lock(vp, LK_SHARED | LK_RETRY);
1483 error = VOP_GETATTR(vp, &vattr, active_cred);
1486 *(int *)data = vattr.va_size - fp->f_offset;
1492 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1501 * File table vnode poll routine.
1504 vn_poll(fp, events, active_cred, td)
1507 struct ucred *active_cred;
1515 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1516 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1521 error = VOP_POLL(vp, events, fp->f_cred, td);
1526 * Acquire the requested lock and then check for validity. LK_RETRY
1527 * permits vn_lock to return doomed vnodes.
1530 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1534 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1535 ("vn_lock called with no locktype."));
1537 #ifdef DEBUG_VFS_LOCKS
1538 KASSERT(vp->v_holdcnt != 0,
1539 ("vn_lock %p: zero hold count", vp));
1541 error = VOP_LOCK1(vp, flags, file, line);
1542 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1543 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1544 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1547 * Callers specify LK_RETRY if they wish to get dead vnodes.
1548 * If RETRY is not set, we return ENOENT instead.
1550 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1551 (flags & LK_RETRY) == 0) {
1556 } while (flags & LK_RETRY && error != 0);
1561 * File table vnode close routine.
1564 vn_closefile(fp, td)
1573 fp->f_ops = &badfileops;
1575 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1578 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1580 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1581 lf.l_whence = SEEK_SET;
1584 lf.l_type = F_UNLCK;
1585 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1592 vn_suspendable(struct mount *mp)
1595 return (mp->mnt_op->vfs_susp_clean != NULL);
1599 * Preparing to start a filesystem write operation. If the operation is
1600 * permitted, then we bump the count of operations in progress and
1601 * proceed. If a suspend request is in progress, we wait until the
1602 * suspension is over, and then proceed.
1605 vn_start_write_locked(struct mount *mp, int flags)
1609 mtx_assert(MNT_MTX(mp), MA_OWNED);
1613 * Check on status of suspension.
1615 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1616 mp->mnt_susp_owner != curthread) {
1617 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
1618 (flags & PCATCH) : 0) | (PUSER - 1);
1619 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1620 if (flags & V_NOWAIT) {
1621 error = EWOULDBLOCK;
1624 error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
1630 if (flags & V_XSLEEP)
1632 mp->mnt_writeopcount++;
1634 if (error != 0 || (flags & V_XSLEEP) != 0)
1641 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
1646 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1647 ("V_MNTREF requires mp"));
1651 * If a vnode is provided, get and return the mount point that
1652 * to which it will write.
1655 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1657 if (error != EOPNOTSUPP)
1662 if ((mp = *mpp) == NULL)
1665 if (!vn_suspendable(mp)) {
1666 if (vp != NULL || (flags & V_MNTREF) != 0)
1672 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1674 * As long as a vnode is not provided we need to acquire a
1675 * refcount for the provided mountpoint too, in order to
1676 * emulate a vfs_ref().
1679 if (vp == NULL && (flags & V_MNTREF) == 0)
1682 return (vn_start_write_locked(mp, flags));
1686 * Secondary suspension. Used by operations such as vop_inactive
1687 * routines that are needed by the higher level functions. These
1688 * are allowed to proceed until all the higher level functions have
1689 * completed (indicated by mnt_writeopcount dropping to zero). At that
1690 * time, these operations are halted until the suspension is over.
1693 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags)
1698 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1699 ("V_MNTREF requires mp"));
1703 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1705 if (error != EOPNOTSUPP)
1711 * If we are not suspended or have not yet reached suspended
1712 * mode, then let the operation proceed.
1714 if ((mp = *mpp) == NULL)
1717 if (!vn_suspendable(mp)) {
1718 if (vp != NULL || (flags & V_MNTREF) != 0)
1724 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1726 * As long as a vnode is not provided we need to acquire a
1727 * refcount for the provided mountpoint too, in order to
1728 * emulate a vfs_ref().
1731 if (vp == NULL && (flags & V_MNTREF) == 0)
1733 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1734 mp->mnt_secondary_writes++;
1735 mp->mnt_secondary_accwrites++;
1739 if (flags & V_NOWAIT) {
1742 return (EWOULDBLOCK);
1745 * Wait for the suspension to finish.
1747 error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
1748 ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
1757 * Filesystem write operation has completed. If we are suspending and this
1758 * operation is the last one, notify the suspender that the suspension is
1762 vn_finished_write(mp)
1765 if (mp == NULL || !vn_suspendable(mp))
1769 mp->mnt_writeopcount--;
1770 if (mp->mnt_writeopcount < 0)
1771 panic("vn_finished_write: neg cnt");
1772 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1773 mp->mnt_writeopcount <= 0)
1774 wakeup(&mp->mnt_writeopcount);
1780 * Filesystem secondary write operation has completed. If we are
1781 * suspending and this operation is the last one, notify the suspender
1782 * that the suspension is now in effect.
1785 vn_finished_secondary_write(mp)
1788 if (mp == NULL || !vn_suspendable(mp))
1792 mp->mnt_secondary_writes--;
1793 if (mp->mnt_secondary_writes < 0)
1794 panic("vn_finished_secondary_write: neg cnt");
1795 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1796 mp->mnt_secondary_writes <= 0)
1797 wakeup(&mp->mnt_secondary_writes);
1804 * Request a filesystem to suspend write operations.
1807 vfs_write_suspend(struct mount *mp, int flags)
1811 MPASS(vn_suspendable(mp));
1814 if (mp->mnt_susp_owner == curthread) {
1818 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1819 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1822 * Unmount holds a write reference on the mount point. If we
1823 * own busy reference and drain for writers, we deadlock with
1824 * the reference draining in the unmount path. Callers of
1825 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1826 * vfs_busy() reference is owned and caller is not in the
1829 if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1830 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1835 mp->mnt_kern_flag |= MNTK_SUSPEND;
1836 mp->mnt_susp_owner = curthread;
1837 if (mp->mnt_writeopcount > 0)
1838 (void) msleep(&mp->mnt_writeopcount,
1839 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1842 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1843 vfs_write_resume(mp, 0);
1848 * Request a filesystem to resume write operations.
1851 vfs_write_resume(struct mount *mp, int flags)
1854 MPASS(vn_suspendable(mp));
1857 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1858 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1859 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1861 mp->mnt_susp_owner = NULL;
1862 wakeup(&mp->mnt_writeopcount);
1863 wakeup(&mp->mnt_flag);
1864 curthread->td_pflags &= ~TDP_IGNSUSP;
1865 if ((flags & VR_START_WRITE) != 0) {
1867 mp->mnt_writeopcount++;
1870 if ((flags & VR_NO_SUSPCLR) == 0)
1872 } else if ((flags & VR_START_WRITE) != 0) {
1874 vn_start_write_locked(mp, 0);
1881 * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1885 vfs_write_suspend_umnt(struct mount *mp)
1889 MPASS(vn_suspendable(mp));
1890 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1891 ("vfs_write_suspend_umnt: recursed"));
1893 /* dounmount() already called vn_start_write(). */
1895 vn_finished_write(mp);
1896 error = vfs_write_suspend(mp, 0);
1898 vn_start_write(NULL, &mp, V_WAIT);
1902 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1905 vn_start_write(NULL, &mp, V_WAIT);
1907 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1908 wakeup(&mp->mnt_flag);
1910 curthread->td_pflags |= TDP_IGNSUSP;
1915 * Implement kqueues for files by translating it to vnode operation.
1918 vn_kqfilter(struct file *fp, struct knote *kn)
1921 return (VOP_KQFILTER(fp->f_vnode, kn));
1925 * Simplified in-kernel wrapper calls for extended attribute access.
1926 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1927 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1930 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1931 const char *attrname, int *buflen, char *buf, struct thread *td)
1937 iov.iov_len = *buflen;
1940 auio.uio_iov = &iov;
1941 auio.uio_iovcnt = 1;
1942 auio.uio_rw = UIO_READ;
1943 auio.uio_segflg = UIO_SYSSPACE;
1945 auio.uio_offset = 0;
1946 auio.uio_resid = *buflen;
1948 if ((ioflg & IO_NODELOCKED) == 0)
1949 vn_lock(vp, LK_SHARED | LK_RETRY);
1951 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1953 /* authorize attribute retrieval as kernel */
1954 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1957 if ((ioflg & IO_NODELOCKED) == 0)
1961 *buflen = *buflen - auio.uio_resid;
1968 * XXX failure mode if partially written?
1971 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1972 const char *attrname, int buflen, char *buf, struct thread *td)
1979 iov.iov_len = buflen;
1982 auio.uio_iov = &iov;
1983 auio.uio_iovcnt = 1;
1984 auio.uio_rw = UIO_WRITE;
1985 auio.uio_segflg = UIO_SYSSPACE;
1987 auio.uio_offset = 0;
1988 auio.uio_resid = buflen;
1990 if ((ioflg & IO_NODELOCKED) == 0) {
1991 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1993 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1996 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1998 /* authorize attribute setting as kernel */
1999 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
2001 if ((ioflg & IO_NODELOCKED) == 0) {
2002 vn_finished_write(mp);
2010 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
2011 const char *attrname, struct thread *td)
2016 if ((ioflg & IO_NODELOCKED) == 0) {
2017 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2019 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2022 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2024 /* authorize attribute removal as kernel */
2025 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2026 if (error == EOPNOTSUPP)
2027 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2030 if ((ioflg & IO_NODELOCKED) == 0) {
2031 vn_finished_write(mp);
2039 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2043 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2047 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2050 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2055 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2056 int lkflags, struct vnode **rvp)
2061 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2063 ltype = VOP_ISLOCKED(vp);
2064 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2065 ("vn_vget_ino: vp not locked"));
2066 error = vfs_busy(mp, MBF_NOWAIT);
2070 error = vfs_busy(mp, 0);
2071 vn_lock(vp, ltype | LK_RETRY);
2075 if (vp->v_iflag & VI_DOOMED) {
2081 error = alloc(mp, alloc_arg, lkflags, rvp);
2084 vn_lock(vp, ltype | LK_RETRY);
2085 if (vp->v_iflag & VI_DOOMED) {
2098 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2102 if (vp->v_type != VREG || td == NULL)
2104 if ((uoff_t)uio->uio_offset + uio->uio_resid >
2105 lim_cur(td, RLIMIT_FSIZE)) {
2106 PROC_LOCK(td->td_proc);
2107 kern_psignal(td->td_proc, SIGXFSZ);
2108 PROC_UNLOCK(td->td_proc);
2115 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2122 vn_lock(vp, LK_SHARED | LK_RETRY);
2123 AUDIT_ARG_VNODE1(vp);
2126 return (setfmode(td, active_cred, vp, mode));
2130 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2137 vn_lock(vp, LK_SHARED | LK_RETRY);
2138 AUDIT_ARG_VNODE1(vp);
2141 return (setfown(td, active_cred, vp, uid, gid));
2145 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2149 if ((object = vp->v_object) == NULL)
2151 VM_OBJECT_WLOCK(object);
2152 vm_object_page_remove(object, start, end, 0);
2153 VM_OBJECT_WUNLOCK(object);
2157 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2165 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2166 ("Wrong command %lu", cmd));
2168 if (vn_lock(vp, LK_SHARED) != 0)
2170 if (vp->v_type != VREG) {
2174 error = VOP_GETATTR(vp, &va, cred);
2178 if (noff >= va.va_size) {
2182 bsize = vp->v_mount->mnt_stat.f_iosize;
2183 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2184 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2185 if (error == EOPNOTSUPP) {
2189 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2190 (bnp != -1 && cmd == FIOSEEKDATA)) {
2197 if (noff > va.va_size)
2199 /* noff == va.va_size. There is an implicit hole at the end of file. */
2200 if (cmd == FIOSEEKDATA)
2210 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2215 off_t foffset, size;
2218 cred = td->td_ucred;
2220 foffset = foffset_lock(fp, 0);
2221 noneg = (vp->v_type != VCHR);
2227 (offset > 0 && foffset > OFF_MAX - offset))) {
2234 vn_lock(vp, LK_SHARED | LK_RETRY);
2235 error = VOP_GETATTR(vp, &vattr, cred);
2241 * If the file references a disk device, then fetch
2242 * the media size and use that to determine the ending
2245 if (vattr.va_size == 0 && vp->v_type == VCHR &&
2246 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2247 vattr.va_size = size;
2249 (vattr.va_size > OFF_MAX ||
2250 (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2254 offset += vattr.va_size;
2259 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2262 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2267 if (error == 0 && noneg && offset < 0)
2271 VFS_KNOTE_UNLOCKED(vp, 0);
2272 td->td_uretoff.tdu_off = offset;
2274 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2279 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2285 * Grant permission if the caller is the owner of the file, or
2286 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2287 * on the file. If the time pointer is null, then write
2288 * permission on the file is also sufficient.
2290 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2291 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2292 * will be allowed to set the times [..] to the current
2295 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2296 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2297 error = VOP_ACCESS(vp, VWRITE, cred, td);
2302 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2307 if (fp->f_type == DTYPE_FIFO)
2308 kif->kf_type = KF_TYPE_FIFO;
2310 kif->kf_type = KF_TYPE_VNODE;
2313 FILEDESC_SUNLOCK(fdp);
2314 error = vn_fill_kinfo_vnode(vp, kif);
2316 FILEDESC_SLOCK(fdp);
2321 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2324 char *fullpath, *freepath;
2327 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
2330 error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2332 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2334 if (freepath != NULL)
2335 free(freepath, M_TEMP);
2338 * Retrieve vnode attributes.
2340 va.va_fsid = VNOVAL;
2342 vn_lock(vp, LK_SHARED | LK_RETRY);
2343 error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2347 if (va.va_fsid != VNOVAL)
2348 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2350 kif->kf_un.kf_file.kf_file_fsid =
2351 vp->v_mount->mnt_stat.f_fsid.val[0];
2352 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2353 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2354 kif->kf_un.kf_file.kf_file_size = va.va_size;
2355 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2360 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
2361 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
2365 struct pmckern_map_in pkm;
2371 boolean_t writecounted;
2374 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
2375 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
2377 * POSIX shared-memory objects are defined to have
2378 * kernel persistence, and are not defined to support
2379 * read(2)/write(2) -- or even open(2). Thus, we can
2380 * use MAP_ASYNC to trade on-disk coherence for speed.
2381 * The shm_open(3) library routine turns on the FPOSIXSHM
2382 * flag to request this behavior.
2384 if ((fp->f_flag & FPOSIXSHM) != 0)
2385 flags |= MAP_NOSYNC;
2390 * Ensure that file and memory protections are
2391 * compatible. Note that we only worry about
2392 * writability if mapping is shared; in this case,
2393 * current and max prot are dictated by the open file.
2394 * XXX use the vnode instead? Problem is: what
2395 * credentials do we use for determination? What if
2396 * proc does a setuid?
2399 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0)
2400 maxprot = VM_PROT_NONE;
2402 maxprot = VM_PROT_EXECUTE;
2403 if ((fp->f_flag & FREAD) != 0)
2404 maxprot |= VM_PROT_READ;
2405 else if ((prot & VM_PROT_READ) != 0)
2409 * If we are sharing potential changes via MAP_SHARED and we
2410 * are trying to get write permission although we opened it
2411 * without asking for it, bail out.
2413 if ((flags & MAP_SHARED) != 0) {
2414 if ((fp->f_flag & FWRITE) != 0)
2415 maxprot |= VM_PROT_WRITE;
2416 else if ((prot & VM_PROT_WRITE) != 0)
2419 maxprot |= VM_PROT_WRITE;
2420 cap_maxprot |= VM_PROT_WRITE;
2422 maxprot &= cap_maxprot;
2424 writecounted = FALSE;
2425 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp,
2426 &foff, &object, &writecounted);
2429 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2430 foff, writecounted, td);
2433 * If this mapping was accounted for in the vnode's
2434 * writecount, then undo that now.
2437 vnode_pager_release_writecount(object, 0, size);
2438 vm_object_deallocate(object);
2441 /* Inform hwpmc(4) if an executable is being mapped. */
2442 if (error == 0 && (prot & VM_PROT_EXECUTE) != 0) {
2444 pkm.pm_address = (uintptr_t) addr;
2445 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);