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 "opt_hwpmc_hooks.h"
48 #include <sys/param.h>
49 #include <sys/systm.h>
52 #include <sys/fcntl.h>
58 #include <sys/limits.h>
61 #include <sys/mount.h>
62 #include <sys/mutex.h>
63 #include <sys/namei.h>
64 #include <sys/vnode.h>
67 #include <sys/filio.h>
68 #include <sys/resourcevar.h>
69 #include <sys/rwlock.h>
71 #include <sys/sysctl.h>
72 #include <sys/ttycom.h>
74 #include <sys/syslog.h>
75 #include <sys/unistd.h>
78 #include <security/audit/audit.h>
79 #include <security/mac/mac_framework.h>
82 #include <vm/vm_extern.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_page.h>
87 #include <vm/vnode_pager.h>
90 #include <sys/pmckern.h>
93 static fo_rdwr_t vn_read;
94 static fo_rdwr_t vn_write;
95 static fo_rdwr_t vn_io_fault;
96 static fo_truncate_t vn_truncate;
97 static fo_ioctl_t vn_ioctl;
98 static fo_poll_t vn_poll;
99 static fo_kqfilter_t vn_kqfilter;
100 static fo_stat_t vn_statfile;
101 static fo_close_t vn_closefile;
102 static fo_mmap_t vn_mmap;
104 struct fileops vnops = {
105 .fo_read = vn_io_fault,
106 .fo_write = vn_io_fault,
107 .fo_truncate = vn_truncate,
108 .fo_ioctl = vn_ioctl,
110 .fo_kqfilter = vn_kqfilter,
111 .fo_stat = vn_statfile,
112 .fo_close = vn_closefile,
113 .fo_chmod = vn_chmod,
114 .fo_chown = vn_chown,
115 .fo_sendfile = vn_sendfile,
117 .fo_fill_kinfo = vn_fill_kinfo,
119 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
122 static const int io_hold_cnt = 16;
123 static int vn_io_fault_enable = 1;
124 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
125 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
126 static int vn_io_fault_prefault = 0;
127 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW,
128 &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting");
129 static u_long vn_io_faults_cnt;
130 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
131 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
134 * Returns true if vn_io_fault mode of handling the i/o request should
138 do_vn_io_fault(struct vnode *vp, struct uio *uio)
142 return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
143 (mp = vp->v_mount) != NULL &&
144 (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
148 * Structure used to pass arguments to vn_io_fault1(), to do either
149 * file- or vnode-based I/O calls.
151 struct vn_io_fault_args {
159 struct fop_args_tag {
163 struct vop_args_tag {
169 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
170 struct vn_io_fault_args *args, struct thread *td);
173 vn_open(ndp, flagp, cmode, fp)
174 struct nameidata *ndp;
178 struct thread *td = ndp->ni_cnd.cn_thread;
180 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
184 * Common code for vnode open operations via a name lookup.
185 * Lookup the vnode and invoke VOP_CREATE if needed.
186 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
188 * Note that this does NOT free nameidata for the successful case,
189 * due to the NDINIT being done elsewhere.
192 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
193 struct ucred *cred, struct file *fp)
197 struct thread *td = ndp->ni_cnd.cn_thread;
199 struct vattr *vap = &vat;
204 if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT |
205 O_EXCL | O_DIRECTORY))
207 else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) {
208 ndp->ni_cnd.cn_nameiop = CREATE;
210 * Set NOCACHE to avoid flushing the cache when
211 * rolling in many files at once.
213 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE;
214 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
215 ndp->ni_cnd.cn_flags |= FOLLOW;
216 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
217 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
218 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
219 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
221 if ((error = namei(ndp)) != 0)
223 if (ndp->ni_vp == NULL) {
226 vap->va_mode = cmode;
228 vap->va_vaflags |= VA_EXCLUSIVE;
229 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
230 NDFREE(ndp, NDF_ONLY_PNBUF);
232 if ((error = vn_start_write(NULL, &mp,
233 V_XSLEEP | PCATCH)) != 0)
237 if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0)
238 ndp->ni_cnd.cn_flags |= MAKEENTRY;
240 error = mac_vnode_check_create(cred, ndp->ni_dvp,
244 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
247 vn_finished_write(mp);
249 NDFREE(ndp, NDF_ONLY_PNBUF);
255 if (ndp->ni_dvp == ndp->ni_vp)
261 if (fmode & O_EXCL) {
268 ndp->ni_cnd.cn_nameiop = LOOKUP;
269 ndp->ni_cnd.cn_flags = ISOPEN |
270 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
271 if (!(fmode & FWRITE))
272 ndp->ni_cnd.cn_flags |= LOCKSHARED;
273 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
274 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
275 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
276 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
277 if ((error = namei(ndp)) != 0)
281 error = vn_open_vnode(vp, fmode, cred, td, fp);
287 NDFREE(ndp, NDF_ONLY_PNBUF);
295 * Common code for vnode open operations once a vnode is located.
296 * Check permissions, and call the VOP_OPEN routine.
299 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
300 struct thread *td, struct file *fp)
304 int error, lock_flags, type;
306 if (vp->v_type == VLNK)
308 if (vp->v_type == VSOCK)
310 if (vp->v_type != VDIR && fmode & O_DIRECTORY)
313 if (fmode & (FWRITE | O_TRUNC)) {
314 if (vp->v_type == VDIR)
322 if ((fmode & O_APPEND) && (fmode & FWRITE))
327 if (fmode & O_VERIFY)
329 error = mac_vnode_check_open(cred, vp, accmode);
333 accmode &= ~(VCREAT | VVERIFY);
335 if ((fmode & O_CREAT) == 0) {
336 if (accmode & VWRITE) {
337 error = vn_writechk(vp);
342 error = VOP_ACCESS(vp, accmode, cred, td);
347 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
348 vn_lock(vp, LK_UPGRADE | LK_RETRY);
349 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
352 while ((fmode & (O_EXLOCK | O_SHLOCK)) != 0) {
353 KASSERT(fp != NULL, ("open with flock requires fp"));
354 if (fp->f_type != DTYPE_NONE && fp->f_type != DTYPE_VNODE) {
358 lock_flags = VOP_ISLOCKED(vp);
360 lf.l_whence = SEEK_SET;
363 if (fmode & O_EXLOCK)
368 if ((fmode & FNONBLOCK) == 0)
370 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
372 fp->f_flag |= FHASLOCK;
373 vn_lock(vp, lock_flags | LK_RETRY);
376 if ((vp->v_iflag & VI_DOOMED) != 0) {
382 * Another thread might have used this vnode as an
383 * executable while the vnode lock was dropped.
384 * Ensure the vnode is still able to be opened for
385 * writing after the lock has been obtained.
387 if ((accmode & VWRITE) != 0)
388 error = vn_writechk(vp);
393 fp->f_flag |= FOPENFAILED;
395 if (fp->f_ops == &badfileops) {
396 fp->f_type = DTYPE_VNODE;
400 } else if ((fmode & FWRITE) != 0) {
401 VOP_ADD_WRITECOUNT(vp, 1);
402 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
403 __func__, vp, vp->v_writecount);
405 ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
410 * Check for write permissions on the specified vnode.
411 * Prototype text segments cannot be written.
415 register struct vnode *vp;
418 ASSERT_VOP_LOCKED(vp, "vn_writechk");
420 * If there's shared text associated with
421 * the vnode, try to free it up once. If
422 * we fail, we can't allow writing.
434 vn_close1(struct vnode *vp, int flags, struct ucred *file_cred,
435 struct thread *td, bool keep_ref)
438 int error, lock_flags;
440 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
441 MNT_EXTENDED_SHARED(vp->v_mount))
442 lock_flags = LK_SHARED;
444 lock_flags = LK_EXCLUSIVE;
446 vn_start_write(vp, &mp, V_WAIT);
447 vn_lock(vp, lock_flags | LK_RETRY);
448 AUDIT_ARG_VNODE1(vp);
449 if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) {
450 VNASSERT(vp->v_writecount > 0, vp,
451 ("vn_close: negative writecount"));
452 VOP_ADD_WRITECOUNT(vp, -1);
453 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
454 __func__, vp, vp->v_writecount);
456 error = VOP_CLOSE(vp, flags, file_cred, td);
461 vn_finished_write(mp);
466 vn_close(struct vnode *vp, int flags, struct ucred *file_cred,
470 return (vn_close1(vp, flags, file_cred, td, false));
474 * Heuristic to detect sequential operation.
477 sequential_heuristic(struct uio *uio, struct file *fp)
480 ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
481 if (fp->f_flag & FRDAHEAD)
482 return (fp->f_seqcount << IO_SEQSHIFT);
485 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
486 * that the first I/O is normally considered to be slightly
487 * sequential. Seeking to offset 0 doesn't change sequentiality
488 * unless previous seeks have reduced f_seqcount to 0, in which
489 * case offset 0 is not special.
491 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
492 uio->uio_offset == fp->f_nextoff) {
494 * f_seqcount is in units of fixed-size blocks so that it
495 * depends mainly on the amount of sequential I/O and not
496 * much on the number of sequential I/O's. The fixed size
497 * of 16384 is hard-coded here since it is (not quite) just
498 * a magic size that works well here. This size is more
499 * closely related to the best I/O size for real disks than
500 * to any block size used by software.
502 fp->f_seqcount += howmany(uio->uio_resid, 16384);
503 if (fp->f_seqcount > IO_SEQMAX)
504 fp->f_seqcount = IO_SEQMAX;
505 return (fp->f_seqcount << IO_SEQSHIFT);
508 /* Not sequential. Quickly draw-down sequentiality. */
509 if (fp->f_seqcount > 1)
517 * Package up an I/O request on a vnode into a uio and do it.
520 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
521 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
522 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
529 struct vn_io_fault_args args;
530 int error, lock_flags;
532 auio.uio_iov = &aiov;
534 aiov.iov_base = base;
536 auio.uio_resid = len;
537 auio.uio_offset = offset;
538 auio.uio_segflg = segflg;
543 if ((ioflg & IO_NODELOCKED) == 0) {
544 if ((ioflg & IO_RANGELOCKED) == 0) {
545 if (rw == UIO_READ) {
546 rl_cookie = vn_rangelock_rlock(vp, offset,
549 rl_cookie = vn_rangelock_wlock(vp, offset,
555 if (rw == UIO_WRITE) {
556 if (vp->v_type != VCHR &&
557 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
560 if (MNT_SHARED_WRITES(mp) ||
561 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
562 lock_flags = LK_SHARED;
564 lock_flags = LK_EXCLUSIVE;
566 lock_flags = LK_SHARED;
567 vn_lock(vp, lock_flags | LK_RETRY);
571 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
573 if ((ioflg & IO_NOMACCHECK) == 0) {
575 error = mac_vnode_check_read(active_cred, file_cred,
578 error = mac_vnode_check_write(active_cred, file_cred,
583 if (file_cred != NULL)
587 if (do_vn_io_fault(vp, &auio)) {
588 args.kind = VN_IO_FAULT_VOP;
591 args.args.vop_args.vp = vp;
592 error = vn_io_fault1(vp, &auio, &args, td);
593 } else if (rw == UIO_READ) {
594 error = VOP_READ(vp, &auio, ioflg, cred);
595 } else /* if (rw == UIO_WRITE) */ {
596 error = VOP_WRITE(vp, &auio, ioflg, cred);
600 *aresid = auio.uio_resid;
602 if (auio.uio_resid && error == 0)
604 if ((ioflg & IO_NODELOCKED) == 0) {
607 vn_finished_write(mp);
610 if (rl_cookie != NULL)
611 vn_rangelock_unlock(vp, rl_cookie);
616 * Package up an I/O request on a vnode into a uio and do it. The I/O
617 * request is split up into smaller chunks and we try to avoid saturating
618 * the buffer cache while potentially holding a vnode locked, so we
619 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
620 * to give other processes a chance to lock the vnode (either other processes
621 * core'ing the same binary, or unrelated processes scanning the directory).
624 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
625 file_cred, aresid, td)
633 struct ucred *active_cred;
634 struct ucred *file_cred;
645 * Force `offset' to a multiple of MAXBSIZE except possibly
646 * for the first chunk, so that filesystems only need to
647 * write full blocks except possibly for the first and last
650 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
654 if (rw != UIO_READ && vp->v_type == VREG)
657 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
658 ioflg, active_cred, file_cred, &iaresid, td);
659 len -= chunk; /* aresid calc already includes length */
663 base = (char *)base + chunk;
664 kern_yield(PRI_USER);
667 *aresid = len + iaresid;
672 foffset_lock(struct file *fp, int flags)
677 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
679 #if OFF_MAX <= LONG_MAX
681 * Caller only wants the current f_offset value. Assume that
682 * the long and shorter integer types reads are atomic.
684 if ((flags & FOF_NOLOCK) != 0)
685 return (fp->f_offset);
689 * According to McKusick the vn lock was protecting f_offset here.
690 * It is now protected by the FOFFSET_LOCKED flag.
692 mtxp = mtx_pool_find(mtxpool_sleep, fp);
694 if ((flags & FOF_NOLOCK) == 0) {
695 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
696 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
697 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
700 fp->f_vnread_flags |= FOFFSET_LOCKED;
708 foffset_unlock(struct file *fp, off_t val, int flags)
712 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
714 #if OFF_MAX <= LONG_MAX
715 if ((flags & FOF_NOLOCK) != 0) {
716 if ((flags & FOF_NOUPDATE) == 0)
718 if ((flags & FOF_NEXTOFF) != 0)
724 mtxp = mtx_pool_find(mtxpool_sleep, fp);
726 if ((flags & FOF_NOUPDATE) == 0)
728 if ((flags & FOF_NEXTOFF) != 0)
730 if ((flags & FOF_NOLOCK) == 0) {
731 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
732 ("Lost FOFFSET_LOCKED"));
733 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
734 wakeup(&fp->f_vnread_flags);
735 fp->f_vnread_flags = 0;
741 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
744 if ((flags & FOF_OFFSET) == 0)
745 uio->uio_offset = foffset_lock(fp, flags);
749 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
752 if ((flags & FOF_OFFSET) == 0)
753 foffset_unlock(fp, uio->uio_offset, flags);
757 get_advice(struct file *fp, struct uio *uio)
762 ret = POSIX_FADV_NORMAL;
763 if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG)
766 mtxp = mtx_pool_find(mtxpool_sleep, fp);
768 if (fp->f_advice != NULL &&
769 uio->uio_offset >= fp->f_advice->fa_start &&
770 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
771 ret = fp->f_advice->fa_advice;
777 * File table vnode read routine.
780 vn_read(fp, uio, active_cred, flags, td)
783 struct ucred *active_cred;
792 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
794 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
797 if (fp->f_flag & FNONBLOCK)
799 if (fp->f_flag & O_DIRECT)
801 advice = get_advice(fp, uio);
802 vn_lock(vp, LK_SHARED | LK_RETRY);
805 case POSIX_FADV_NORMAL:
806 case POSIX_FADV_SEQUENTIAL:
807 case POSIX_FADV_NOREUSE:
808 ioflag |= sequential_heuristic(uio, fp);
810 case POSIX_FADV_RANDOM:
811 /* Disable read-ahead for random I/O. */
814 orig_offset = uio->uio_offset;
817 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
820 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
821 fp->f_nextoff = uio->uio_offset;
823 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
824 orig_offset != uio->uio_offset)
826 * Use POSIX_FADV_DONTNEED to flush pages and buffers
827 * for the backing file after a POSIX_FADV_NOREUSE
830 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
831 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;
852 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
854 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
856 if (vp->v_type == VREG)
859 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
861 if (fp->f_flag & FNONBLOCK)
863 if (fp->f_flag & O_DIRECT)
865 if ((fp->f_flag & O_FSYNC) ||
866 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
869 if (vp->v_type != VCHR &&
870 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
873 advice = get_advice(fp, uio);
875 if (MNT_SHARED_WRITES(mp) ||
876 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
877 lock_flags = LK_SHARED;
879 lock_flags = LK_EXCLUSIVE;
882 vn_lock(vp, lock_flags | LK_RETRY);
884 case POSIX_FADV_NORMAL:
885 case POSIX_FADV_SEQUENTIAL:
886 case POSIX_FADV_NOREUSE:
887 ioflag |= sequential_heuristic(uio, fp);
889 case POSIX_FADV_RANDOM:
890 /* XXX: Is this correct? */
893 orig_offset = uio->uio_offset;
896 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
899 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
900 fp->f_nextoff = uio->uio_offset;
902 if (vp->v_type != VCHR)
903 vn_finished_write(mp);
904 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
905 orig_offset != uio->uio_offset)
907 * Use POSIX_FADV_DONTNEED to flush pages and buffers
908 * for the backing file after a POSIX_FADV_NOREUSE
911 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
912 POSIX_FADV_DONTNEED);
918 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
919 * prevent the following deadlock:
921 * Assume that the thread A reads from the vnode vp1 into userspace
922 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
923 * currently not resident, then system ends up with the call chain
924 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
925 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
926 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
927 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
928 * backed by the pages of vnode vp1, and some page in buf2 is not
929 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
931 * To prevent the lock order reversal and deadlock, vn_io_fault() does
932 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
933 * Instead, it first tries to do the whole range i/o with pagefaults
934 * disabled. If all pages in the i/o buffer are resident and mapped,
935 * VOP will succeed (ignoring the genuine filesystem errors).
936 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
937 * i/o in chunks, with all pages in the chunk prefaulted and held
938 * using vm_fault_quick_hold_pages().
940 * Filesystems using this deadlock avoidance scheme should use the
941 * array of the held pages from uio, saved in the curthread->td_ma,
942 * instead of doing uiomove(). A helper function
943 * vn_io_fault_uiomove() converts uiomove request into
944 * uiomove_fromphys() over td_ma array.
946 * Since vnode locks do not cover the whole i/o anymore, rangelocks
947 * make the current i/o request atomic with respect to other i/os and
952 * Decode vn_io_fault_args and perform the corresponding i/o.
955 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
959 switch (args->kind) {
960 case VN_IO_FAULT_FOP:
961 return ((args->args.fop_args.doio)(args->args.fop_args.fp,
962 uio, args->cred, args->flags, td));
963 case VN_IO_FAULT_VOP:
964 if (uio->uio_rw == UIO_READ) {
965 return (VOP_READ(args->args.vop_args.vp, uio,
966 args->flags, args->cred));
967 } else if (uio->uio_rw == UIO_WRITE) {
968 return (VOP_WRITE(args->args.vop_args.vp, uio,
969 args->flags, args->cred));
973 panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
978 vn_io_fault_touch(char *base, const struct uio *uio)
983 if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1))
989 vn_io_fault_prefault_user(const struct uio *uio)
992 const struct iovec *iov;
997 KASSERT(uio->uio_segflg == UIO_USERSPACE,
998 ("vn_io_fault_prefault userspace"));
1002 resid = uio->uio_resid;
1003 base = iov->iov_base;
1006 error = vn_io_fault_touch(base, uio);
1009 if (len < PAGE_SIZE) {
1011 error = vn_io_fault_touch(base + len - 1, uio);
1016 if (++i >= uio->uio_iovcnt)
1018 iov = uio->uio_iov + i;
1019 base = iov->iov_base;
1031 * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1032 * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1033 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1034 * into args and call vn_io_fault1() to handle faults during the user
1035 * mode buffer accesses.
1038 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1041 vm_page_t ma[io_hold_cnt + 2];
1042 struct uio *uio_clone, short_uio;
1043 struct iovec short_iovec[1];
1044 vm_page_t *prev_td_ma;
1046 vm_offset_t addr, end;
1049 int error, cnt, save, saveheld, prev_td_ma_cnt;
1051 if (vn_io_fault_prefault) {
1052 error = vn_io_fault_prefault_user(uio);
1054 return (error); /* Or ignore ? */
1057 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1060 * The UFS follows IO_UNIT directive and replays back both
1061 * uio_offset and uio_resid if an error is encountered during the
1062 * operation. But, since the iovec may be already advanced,
1063 * uio is still in an inconsistent state.
1065 * Cache a copy of the original uio, which is advanced to the redo
1066 * point using UIO_NOCOPY below.
1068 uio_clone = cloneuio(uio);
1069 resid = uio->uio_resid;
1071 short_uio.uio_segflg = UIO_USERSPACE;
1072 short_uio.uio_rw = uio->uio_rw;
1073 short_uio.uio_td = uio->uio_td;
1075 save = vm_fault_disable_pagefaults();
1076 error = vn_io_fault_doio(args, uio, td);
1077 if (error != EFAULT)
1080 atomic_add_long(&vn_io_faults_cnt, 1);
1081 uio_clone->uio_segflg = UIO_NOCOPY;
1082 uiomove(NULL, resid - uio->uio_resid, uio_clone);
1083 uio_clone->uio_segflg = uio->uio_segflg;
1085 saveheld = curthread_pflags_set(TDP_UIOHELD);
1086 prev_td_ma = td->td_ma;
1087 prev_td_ma_cnt = td->td_ma_cnt;
1089 while (uio_clone->uio_resid != 0) {
1090 len = uio_clone->uio_iov->iov_len;
1092 KASSERT(uio_clone->uio_iovcnt >= 1,
1093 ("iovcnt underflow"));
1094 uio_clone->uio_iov++;
1095 uio_clone->uio_iovcnt--;
1098 if (len > io_hold_cnt * PAGE_SIZE)
1099 len = io_hold_cnt * PAGE_SIZE;
1100 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1101 end = round_page(addr + len);
1106 cnt = atop(end - trunc_page(addr));
1108 * A perfectly misaligned address and length could cause
1109 * both the start and the end of the chunk to use partial
1110 * page. +2 accounts for such a situation.
1112 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1113 addr, len, prot, ma, io_hold_cnt + 2);
1118 short_uio.uio_iov = &short_iovec[0];
1119 short_iovec[0].iov_base = (void *)addr;
1120 short_uio.uio_iovcnt = 1;
1121 short_uio.uio_resid = short_iovec[0].iov_len = len;
1122 short_uio.uio_offset = uio_clone->uio_offset;
1124 td->td_ma_cnt = cnt;
1126 error = vn_io_fault_doio(args, &short_uio, td);
1127 vm_page_unhold_pages(ma, cnt);
1128 adv = len - short_uio.uio_resid;
1130 uio_clone->uio_iov->iov_base =
1131 (char *)uio_clone->uio_iov->iov_base + adv;
1132 uio_clone->uio_iov->iov_len -= adv;
1133 uio_clone->uio_resid -= adv;
1134 uio_clone->uio_offset += adv;
1136 uio->uio_resid -= adv;
1137 uio->uio_offset += adv;
1139 if (error != 0 || adv == 0)
1142 td->td_ma = prev_td_ma;
1143 td->td_ma_cnt = prev_td_ma_cnt;
1144 curthread_pflags_restore(saveheld);
1146 vm_fault_enable_pagefaults(save);
1147 free(uio_clone, M_IOV);
1152 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1153 int flags, struct thread *td)
1158 struct vn_io_fault_args args;
1161 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1163 foffset_lock_uio(fp, uio, flags);
1164 if (do_vn_io_fault(vp, uio)) {
1165 args.kind = VN_IO_FAULT_FOP;
1166 args.args.fop_args.fp = fp;
1167 args.args.fop_args.doio = doio;
1168 args.cred = active_cred;
1169 args.flags = flags | FOF_OFFSET;
1170 if (uio->uio_rw == UIO_READ) {
1171 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1172 uio->uio_offset + uio->uio_resid);
1173 } else if ((fp->f_flag & O_APPEND) != 0 ||
1174 (flags & FOF_OFFSET) == 0) {
1175 /* For appenders, punt and lock the whole range. */
1176 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1178 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1179 uio->uio_offset + uio->uio_resid);
1181 error = vn_io_fault1(vp, uio, &args, td);
1182 vn_rangelock_unlock(vp, rl_cookie);
1184 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1186 foffset_unlock_uio(fp, uio, flags);
1191 * Helper function to perform the requested uiomove operation using
1192 * the held pages for io->uio_iov[0].iov_base buffer instead of
1193 * copyin/copyout. Access to the pages with uiomove_fromphys()
1194 * instead of iov_base prevents page faults that could occur due to
1195 * pmap_collect() invalidating the mapping created by
1196 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1197 * object cleanup revoking the write access from page mappings.
1199 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1200 * instead of plain uiomove().
1203 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1205 struct uio transp_uio;
1206 struct iovec transp_iov[1];
1212 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1213 uio->uio_segflg != UIO_USERSPACE)
1214 return (uiomove(data, xfersize, uio));
1216 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1217 transp_iov[0].iov_base = data;
1218 transp_uio.uio_iov = &transp_iov[0];
1219 transp_uio.uio_iovcnt = 1;
1220 if (xfersize > uio->uio_resid)
1221 xfersize = uio->uio_resid;
1222 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1223 transp_uio.uio_offset = 0;
1224 transp_uio.uio_segflg = UIO_SYSSPACE;
1226 * Since transp_iov points to data, and td_ma page array
1227 * corresponds to original uio->uio_iov, we need to invert the
1228 * direction of the i/o operation as passed to
1229 * uiomove_fromphys().
1231 switch (uio->uio_rw) {
1233 transp_uio.uio_rw = UIO_READ;
1236 transp_uio.uio_rw = UIO_WRITE;
1239 transp_uio.uio_td = uio->uio_td;
1240 error = uiomove_fromphys(td->td_ma,
1241 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1242 xfersize, &transp_uio);
1243 adv = xfersize - transp_uio.uio_resid;
1245 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1246 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1248 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1250 td->td_ma_cnt -= pgadv;
1251 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1252 uio->uio_iov->iov_len -= adv;
1253 uio->uio_resid -= adv;
1254 uio->uio_offset += adv;
1259 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1263 vm_offset_t iov_base;
1267 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1268 uio->uio_segflg != UIO_USERSPACE)
1269 return (uiomove_fromphys(ma, offset, xfersize, uio));
1271 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1272 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1273 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1274 switch (uio->uio_rw) {
1276 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1280 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1284 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1286 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1288 td->td_ma_cnt -= pgadv;
1289 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1290 uio->uio_iov->iov_len -= cnt;
1291 uio->uio_resid -= cnt;
1292 uio->uio_offset += cnt;
1298 * File table truncate routine.
1301 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1313 * Lock the whole range for truncation. Otherwise split i/o
1314 * might happen partly before and partly after the truncation.
1316 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1317 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1320 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1321 if (vp->v_type == VDIR) {
1326 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1330 error = vn_writechk(vp);
1333 vattr.va_size = length;
1334 if ((fp->f_flag & O_FSYNC) != 0)
1335 vattr.va_vaflags |= VA_SYNC;
1336 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1340 vn_finished_write(mp);
1342 vn_rangelock_unlock(vp, rl_cookie);
1347 * File table vnode stat routine.
1350 vn_statfile(fp, sb, active_cred, td)
1353 struct ucred *active_cred;
1356 struct vnode *vp = fp->f_vnode;
1359 vn_lock(vp, LK_SHARED | LK_RETRY);
1360 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1367 * Stat a vnode; implementation for the stat syscall
1370 vn_stat(vp, sb, active_cred, file_cred, td)
1372 register struct stat *sb;
1373 struct ucred *active_cred;
1374 struct ucred *file_cred;
1378 register struct vattr *vap;
1382 AUDIT_ARG_VNODE1(vp);
1384 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1392 * Initialize defaults for new and unusual fields, so that file
1393 * systems which don't support these fields don't need to know
1396 vap->va_birthtime.tv_sec = -1;
1397 vap->va_birthtime.tv_nsec = 0;
1398 vap->va_fsid = VNOVAL;
1399 vap->va_rdev = NODEV;
1401 error = VOP_GETATTR(vp, vap, active_cred);
1406 * Zero the spare stat fields
1408 bzero(sb, sizeof *sb);
1411 * Copy from vattr table
1413 if (vap->va_fsid != VNOVAL)
1414 sb->st_dev = vap->va_fsid;
1416 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1417 sb->st_ino = vap->va_fileid;
1418 mode = vap->va_mode;
1419 switch (vap->va_type) {
1445 sb->st_nlink = vap->va_nlink;
1446 sb->st_uid = vap->va_uid;
1447 sb->st_gid = vap->va_gid;
1448 sb->st_rdev = vap->va_rdev;
1449 if (vap->va_size > OFF_MAX)
1451 sb->st_size = vap->va_size;
1452 sb->st_atim = vap->va_atime;
1453 sb->st_mtim = vap->va_mtime;
1454 sb->st_ctim = vap->va_ctime;
1455 sb->st_birthtim = vap->va_birthtime;
1458 * According to www.opengroup.org, the meaning of st_blksize is
1459 * "a filesystem-specific preferred I/O block size for this
1460 * object. In some filesystem types, this may vary from file
1462 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1465 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1467 sb->st_flags = vap->va_flags;
1468 if (priv_check(td, PRIV_VFS_GENERATION))
1471 sb->st_gen = vap->va_gen;
1473 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1478 * File table vnode ioctl routine.
1481 vn_ioctl(fp, com, data, active_cred, td)
1485 struct ucred *active_cred;
1493 switch (vp->v_type) {
1498 vn_lock(vp, LK_SHARED | LK_RETRY);
1499 error = VOP_GETATTR(vp, &vattr, active_cred);
1502 *(int *)data = vattr.va_size - fp->f_offset;
1508 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1517 * File table vnode poll routine.
1520 vn_poll(fp, events, active_cred, td)
1523 struct ucred *active_cred;
1531 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1532 AUDIT_ARG_VNODE1(vp);
1533 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1538 error = VOP_POLL(vp, events, fp->f_cred, td);
1543 * Acquire the requested lock and then check for validity. LK_RETRY
1544 * permits vn_lock to return doomed vnodes.
1547 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1551 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1552 ("vn_lock called with no locktype."));
1554 #ifdef DEBUG_VFS_LOCKS
1555 KASSERT(vp->v_holdcnt != 0,
1556 ("vn_lock %p: zero hold count", vp));
1558 error = VOP_LOCK1(vp, flags, file, line);
1559 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1560 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1561 ("LK_RETRY set with incompatible flags (0x%x) or an error occurred (%d)",
1564 * Callers specify LK_RETRY if they wish to get dead vnodes.
1565 * If RETRY is not set, we return ENOENT instead.
1567 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1568 (flags & LK_RETRY) == 0) {
1573 } while (flags & LK_RETRY && error != 0);
1578 * File table vnode close routine.
1581 vn_closefile(fp, td)
1591 fp->f_ops = &badfileops;
1592 ref= (fp->f_flag & FHASLOCK) != 0 && fp->f_type == DTYPE_VNODE;
1594 error = vn_close1(vp, fp->f_flag, fp->f_cred, td, ref);
1596 if (__predict_false(ref)) {
1597 lf.l_whence = SEEK_SET;
1600 lf.l_type = F_UNLCK;
1601 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1608 vn_suspendable(struct mount *mp)
1611 return (mp->mnt_op->vfs_susp_clean != NULL);
1615 * Preparing to start a filesystem write operation. If the operation is
1616 * permitted, then we bump the count of operations in progress and
1617 * proceed. If a suspend request is in progress, we wait until the
1618 * suspension is over, and then proceed.
1621 vn_start_write_locked(struct mount *mp, int flags)
1625 mtx_assert(MNT_MTX(mp), MA_OWNED);
1629 * Check on status of suspension.
1631 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1632 mp->mnt_susp_owner != curthread) {
1633 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
1634 (flags & PCATCH) : 0) | (PUSER - 1);
1635 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1636 if (flags & V_NOWAIT) {
1637 error = EWOULDBLOCK;
1640 error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
1646 if (flags & V_XSLEEP)
1648 mp->mnt_writeopcount++;
1650 if (error != 0 || (flags & V_XSLEEP) != 0)
1657 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
1662 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1663 ("V_MNTREF requires mp"));
1667 * If a vnode is provided, get and return the mount point that
1668 * to which it will write.
1671 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1673 if (error != EOPNOTSUPP)
1678 if ((mp = *mpp) == NULL)
1681 if (!vn_suspendable(mp)) {
1682 if (vp != NULL || (flags & V_MNTREF) != 0)
1688 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1690 * As long as a vnode is not provided we need to acquire a
1691 * refcount for the provided mountpoint too, in order to
1692 * emulate a vfs_ref().
1695 if (vp == NULL && (flags & V_MNTREF) == 0)
1698 return (vn_start_write_locked(mp, flags));
1702 * Secondary suspension. Used by operations such as vop_inactive
1703 * routines that are needed by the higher level functions. These
1704 * are allowed to proceed until all the higher level functions have
1705 * completed (indicated by mnt_writeopcount dropping to zero). At that
1706 * time, these operations are halted until the suspension is over.
1709 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags)
1714 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1715 ("V_MNTREF requires mp"));
1719 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1721 if (error != EOPNOTSUPP)
1727 * If we are not suspended or have not yet reached suspended
1728 * mode, then let the operation proceed.
1730 if ((mp = *mpp) == NULL)
1733 if (!vn_suspendable(mp)) {
1734 if (vp != NULL || (flags & V_MNTREF) != 0)
1740 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1742 * As long as a vnode is not provided we need to acquire a
1743 * refcount for the provided mountpoint too, in order to
1744 * emulate a vfs_ref().
1747 if (vp == NULL && (flags & V_MNTREF) == 0)
1749 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1750 mp->mnt_secondary_writes++;
1751 mp->mnt_secondary_accwrites++;
1755 if (flags & V_NOWAIT) {
1758 return (EWOULDBLOCK);
1761 * Wait for the suspension to finish.
1763 error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
1764 ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
1773 * Filesystem write operation has completed. If we are suspending and this
1774 * operation is the last one, notify the suspender that the suspension is
1778 vn_finished_write(mp)
1781 if (mp == NULL || !vn_suspendable(mp))
1785 mp->mnt_writeopcount--;
1786 if (mp->mnt_writeopcount < 0)
1787 panic("vn_finished_write: neg cnt");
1788 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1789 mp->mnt_writeopcount <= 0)
1790 wakeup(&mp->mnt_writeopcount);
1796 * Filesystem secondary write operation has completed. If we are
1797 * suspending and this operation is the last one, notify the suspender
1798 * that the suspension is now in effect.
1801 vn_finished_secondary_write(mp)
1804 if (mp == NULL || !vn_suspendable(mp))
1808 mp->mnt_secondary_writes--;
1809 if (mp->mnt_secondary_writes < 0)
1810 panic("vn_finished_secondary_write: neg cnt");
1811 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1812 mp->mnt_secondary_writes <= 0)
1813 wakeup(&mp->mnt_secondary_writes);
1820 * Request a filesystem to suspend write operations.
1823 vfs_write_suspend(struct mount *mp, int flags)
1827 MPASS(vn_suspendable(mp));
1830 if (mp->mnt_susp_owner == curthread) {
1834 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1835 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1838 * Unmount holds a write reference on the mount point. If we
1839 * own busy reference and drain for writers, we deadlock with
1840 * the reference draining in the unmount path. Callers of
1841 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1842 * vfs_busy() reference is owned and caller is not in the
1845 if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1846 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1851 mp->mnt_kern_flag |= MNTK_SUSPEND;
1852 mp->mnt_susp_owner = curthread;
1853 if (mp->mnt_writeopcount > 0)
1854 (void) msleep(&mp->mnt_writeopcount,
1855 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1858 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1859 vfs_write_resume(mp, 0);
1864 * Request a filesystem to resume write operations.
1867 vfs_write_resume(struct mount *mp, int flags)
1870 MPASS(vn_suspendable(mp));
1873 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1874 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1875 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1877 mp->mnt_susp_owner = NULL;
1878 wakeup(&mp->mnt_writeopcount);
1879 wakeup(&mp->mnt_flag);
1880 curthread->td_pflags &= ~TDP_IGNSUSP;
1881 if ((flags & VR_START_WRITE) != 0) {
1883 mp->mnt_writeopcount++;
1886 if ((flags & VR_NO_SUSPCLR) == 0)
1888 } else if ((flags & VR_START_WRITE) != 0) {
1890 vn_start_write_locked(mp, 0);
1897 * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1901 vfs_write_suspend_umnt(struct mount *mp)
1905 MPASS(vn_suspendable(mp));
1906 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1907 ("vfs_write_suspend_umnt: recursed"));
1909 /* dounmount() already called vn_start_write(). */
1911 vn_finished_write(mp);
1912 error = vfs_write_suspend(mp, 0);
1914 vn_start_write(NULL, &mp, V_WAIT);
1918 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1921 vn_start_write(NULL, &mp, V_WAIT);
1923 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1924 wakeup(&mp->mnt_flag);
1926 curthread->td_pflags |= TDP_IGNSUSP;
1931 * Implement kqueues for files by translating it to vnode operation.
1934 vn_kqfilter(struct file *fp, struct knote *kn)
1937 return (VOP_KQFILTER(fp->f_vnode, kn));
1941 * Simplified in-kernel wrapper calls for extended attribute access.
1942 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1943 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1946 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1947 const char *attrname, int *buflen, char *buf, struct thread *td)
1953 iov.iov_len = *buflen;
1956 auio.uio_iov = &iov;
1957 auio.uio_iovcnt = 1;
1958 auio.uio_rw = UIO_READ;
1959 auio.uio_segflg = UIO_SYSSPACE;
1961 auio.uio_offset = 0;
1962 auio.uio_resid = *buflen;
1964 if ((ioflg & IO_NODELOCKED) == 0)
1965 vn_lock(vp, LK_SHARED | LK_RETRY);
1967 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1969 /* authorize attribute retrieval as kernel */
1970 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1973 if ((ioflg & IO_NODELOCKED) == 0)
1977 *buflen = *buflen - auio.uio_resid;
1984 * XXX failure mode if partially written?
1987 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1988 const char *attrname, int buflen, char *buf, struct thread *td)
1995 iov.iov_len = buflen;
1998 auio.uio_iov = &iov;
1999 auio.uio_iovcnt = 1;
2000 auio.uio_rw = UIO_WRITE;
2001 auio.uio_segflg = UIO_SYSSPACE;
2003 auio.uio_offset = 0;
2004 auio.uio_resid = buflen;
2006 if ((ioflg & IO_NODELOCKED) == 0) {
2007 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2009 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2012 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2014 /* authorize attribute setting as kernel */
2015 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
2017 if ((ioflg & IO_NODELOCKED) == 0) {
2018 vn_finished_write(mp);
2026 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
2027 const char *attrname, struct thread *td)
2032 if ((ioflg & IO_NODELOCKED) == 0) {
2033 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2035 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2038 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2040 /* authorize attribute removal as kernel */
2041 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2042 if (error == EOPNOTSUPP)
2043 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2046 if ((ioflg & IO_NODELOCKED) == 0) {
2047 vn_finished_write(mp);
2055 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2059 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2063 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2066 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2071 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2072 int lkflags, struct vnode **rvp)
2077 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2079 ltype = VOP_ISLOCKED(vp);
2080 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2081 ("vn_vget_ino: vp not locked"));
2082 error = vfs_busy(mp, MBF_NOWAIT);
2086 error = vfs_busy(mp, 0);
2087 vn_lock(vp, ltype | LK_RETRY);
2091 if (vp->v_iflag & VI_DOOMED) {
2097 error = alloc(mp, alloc_arg, lkflags, rvp);
2100 vn_lock(vp, ltype | LK_RETRY);
2101 if (vp->v_iflag & VI_DOOMED) {
2114 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2118 if (vp->v_type != VREG || td == NULL)
2120 if ((uoff_t)uio->uio_offset + uio->uio_resid >
2121 lim_cur(td, RLIMIT_FSIZE)) {
2122 PROC_LOCK(td->td_proc);
2123 kern_psignal(td->td_proc, SIGXFSZ);
2124 PROC_UNLOCK(td->td_proc);
2131 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2138 vn_lock(vp, LK_SHARED | LK_RETRY);
2139 AUDIT_ARG_VNODE1(vp);
2142 return (setfmode(td, active_cred, vp, mode));
2146 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2153 vn_lock(vp, LK_SHARED | LK_RETRY);
2154 AUDIT_ARG_VNODE1(vp);
2157 return (setfown(td, active_cred, vp, uid, gid));
2161 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2165 if ((object = vp->v_object) == NULL)
2167 VM_OBJECT_WLOCK(object);
2168 vm_object_page_remove(object, start, end, 0);
2169 VM_OBJECT_WUNLOCK(object);
2173 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2181 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2182 ("Wrong command %lu", cmd));
2184 if (vn_lock(vp, LK_SHARED) != 0)
2186 if (vp->v_type != VREG) {
2190 error = VOP_GETATTR(vp, &va, cred);
2194 if (noff >= va.va_size) {
2198 bsize = vp->v_mount->mnt_stat.f_iosize;
2199 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2200 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2201 if (error == EOPNOTSUPP) {
2205 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2206 (bnp != -1 && cmd == FIOSEEKDATA)) {
2213 if (noff > va.va_size)
2215 /* noff == va.va_size. There is an implicit hole at the end of file. */
2216 if (cmd == FIOSEEKDATA)
2226 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2231 off_t foffset, size;
2234 cred = td->td_ucred;
2236 foffset = foffset_lock(fp, 0);
2237 noneg = (vp->v_type != VCHR);
2243 (offset > 0 && foffset > OFF_MAX - offset))) {
2250 vn_lock(vp, LK_SHARED | LK_RETRY);
2251 error = VOP_GETATTR(vp, &vattr, cred);
2257 * If the file references a disk device, then fetch
2258 * the media size and use that to determine the ending
2261 if (vattr.va_size == 0 && vp->v_type == VCHR &&
2262 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2263 vattr.va_size = size;
2265 (vattr.va_size > OFF_MAX ||
2266 (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2270 offset += vattr.va_size;
2275 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2278 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2283 if (error == 0 && noneg && offset < 0)
2287 VFS_KNOTE_UNLOCKED(vp, 0);
2288 td->td_uretoff.tdu_off = offset;
2290 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2295 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2301 * Grant permission if the caller is the owner of the file, or
2302 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2303 * on the file. If the time pointer is null, then write
2304 * permission on the file is also sufficient.
2306 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2307 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2308 * will be allowed to set the times [..] to the current
2311 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2312 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2313 error = VOP_ACCESS(vp, VWRITE, cred, td);
2318 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2323 if (fp->f_type == DTYPE_FIFO)
2324 kif->kf_type = KF_TYPE_FIFO;
2326 kif->kf_type = KF_TYPE_VNODE;
2329 FILEDESC_SUNLOCK(fdp);
2330 error = vn_fill_kinfo_vnode(vp, kif);
2332 FILEDESC_SLOCK(fdp);
2337 vn_fill_junk(struct kinfo_file *kif)
2342 * Simulate vn_fullpath returning changing values for a given
2343 * vp during e.g. coredump.
2345 len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1;
2346 olen = strlen(kif->kf_path);
2348 strcpy(&kif->kf_path[len - 1], "$");
2350 for (; olen < len; olen++)
2351 strcpy(&kif->kf_path[olen], "A");
2355 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2358 char *fullpath, *freepath;
2361 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
2364 error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2366 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2368 if (freepath != NULL)
2369 free(freepath, M_TEMP);
2371 KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path,
2376 * Retrieve vnode attributes.
2378 va.va_fsid = VNOVAL;
2380 vn_lock(vp, LK_SHARED | LK_RETRY);
2381 error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2385 if (va.va_fsid != VNOVAL)
2386 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2388 kif->kf_un.kf_file.kf_file_fsid =
2389 vp->v_mount->mnt_stat.f_fsid.val[0];
2390 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2391 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2392 kif->kf_un.kf_file.kf_file_size = va.va_size;
2393 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2398 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
2399 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
2403 struct pmckern_map_in pkm;
2409 boolean_t writecounted;
2412 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
2413 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
2415 * POSIX shared-memory objects are defined to have
2416 * kernel persistence, and are not defined to support
2417 * read(2)/write(2) -- or even open(2). Thus, we can
2418 * use MAP_ASYNC to trade on-disk coherence for speed.
2419 * The shm_open(3) library routine turns on the FPOSIXSHM
2420 * flag to request this behavior.
2422 if ((fp->f_flag & FPOSIXSHM) != 0)
2423 flags |= MAP_NOSYNC;
2428 * Ensure that file and memory protections are
2429 * compatible. Note that we only worry about
2430 * writability if mapping is shared; in this case,
2431 * current and max prot are dictated by the open file.
2432 * XXX use the vnode instead? Problem is: what
2433 * credentials do we use for determination? What if
2434 * proc does a setuid?
2437 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
2438 maxprot = VM_PROT_NONE;
2439 if ((prot & VM_PROT_EXECUTE) != 0)
2442 maxprot = VM_PROT_EXECUTE;
2443 if ((fp->f_flag & FREAD) != 0)
2444 maxprot |= VM_PROT_READ;
2445 else if ((prot & VM_PROT_READ) != 0)
2449 * If we are sharing potential changes via MAP_SHARED and we
2450 * are trying to get write permission although we opened it
2451 * without asking for it, bail out.
2453 if ((flags & MAP_SHARED) != 0) {
2454 if ((fp->f_flag & FWRITE) != 0)
2455 maxprot |= VM_PROT_WRITE;
2456 else if ((prot & VM_PROT_WRITE) != 0)
2459 maxprot |= VM_PROT_WRITE;
2460 cap_maxprot |= VM_PROT_WRITE;
2462 maxprot &= cap_maxprot;
2464 writecounted = FALSE;
2465 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp,
2466 &foff, &object, &writecounted);
2469 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2470 foff, writecounted, td);
2473 * If this mapping was accounted for in the vnode's
2474 * writecount, then undo that now.
2477 vnode_pager_release_writecount(object, 0, size);
2478 vm_object_deallocate(object);
2481 /* Inform hwpmc(4) if an executable is being mapped. */
2482 if (PMC_HOOK_INSTALLED(PMC_FN_MMAP)) {
2483 if ((prot & VM_PROT_EXECUTE) != 0 && error == 0) {
2485 pkm.pm_address = (uintptr_t) *addr;
2486 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);