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 * 3. 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.
414 vn_writechk(struct vnode *vp)
417 ASSERT_VOP_LOCKED(vp, "vn_writechk");
419 * If there's shared text associated with
420 * the vnode, try to free it up once. If
421 * we fail, we can't allow writing.
433 vn_close1(struct vnode *vp, int flags, struct ucred *file_cred,
434 struct thread *td, bool keep_ref)
437 int error, lock_flags;
439 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
440 MNT_EXTENDED_SHARED(vp->v_mount))
441 lock_flags = LK_SHARED;
443 lock_flags = LK_EXCLUSIVE;
445 vn_start_write(vp, &mp, V_WAIT);
446 vn_lock(vp, lock_flags | LK_RETRY);
447 AUDIT_ARG_VNODE1(vp);
448 if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) {
449 VNASSERT(vp->v_writecount > 0, vp,
450 ("vn_close: negative writecount"));
451 VOP_ADD_WRITECOUNT(vp, -1);
452 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
453 __func__, vp, vp->v_writecount);
455 error = VOP_CLOSE(vp, flags, file_cred, td);
460 vn_finished_write(mp);
465 vn_close(struct vnode *vp, int flags, struct ucred *file_cred,
469 return (vn_close1(vp, flags, file_cred, td, false));
473 * Heuristic to detect sequential operation.
476 sequential_heuristic(struct uio *uio, struct file *fp)
479 ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
480 if (fp->f_flag & FRDAHEAD)
481 return (fp->f_seqcount << IO_SEQSHIFT);
484 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
485 * that the first I/O is normally considered to be slightly
486 * sequential. Seeking to offset 0 doesn't change sequentiality
487 * unless previous seeks have reduced f_seqcount to 0, in which
488 * case offset 0 is not special.
490 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
491 uio->uio_offset == fp->f_nextoff) {
493 * f_seqcount is in units of fixed-size blocks so that it
494 * depends mainly on the amount of sequential I/O and not
495 * much on the number of sequential I/O's. The fixed size
496 * of 16384 is hard-coded here since it is (not quite) just
497 * a magic size that works well here. This size is more
498 * closely related to the best I/O size for real disks than
499 * to any block size used by software.
501 fp->f_seqcount += howmany(uio->uio_resid, 16384);
502 if (fp->f_seqcount > IO_SEQMAX)
503 fp->f_seqcount = IO_SEQMAX;
504 return (fp->f_seqcount << IO_SEQSHIFT);
507 /* Not sequential. Quickly draw-down sequentiality. */
508 if (fp->f_seqcount > 1)
516 * Package up an I/O request on a vnode into a uio and do it.
519 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
520 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
521 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
528 struct vn_io_fault_args args;
529 int error, lock_flags;
531 auio.uio_iov = &aiov;
533 aiov.iov_base = base;
535 auio.uio_resid = len;
536 auio.uio_offset = offset;
537 auio.uio_segflg = segflg;
542 if ((ioflg & IO_NODELOCKED) == 0) {
543 if ((ioflg & IO_RANGELOCKED) == 0) {
544 if (rw == UIO_READ) {
545 rl_cookie = vn_rangelock_rlock(vp, offset,
548 rl_cookie = vn_rangelock_wlock(vp, offset,
554 if (rw == UIO_WRITE) {
555 if (vp->v_type != VCHR &&
556 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
559 if (MNT_SHARED_WRITES(mp) ||
560 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
561 lock_flags = LK_SHARED;
563 lock_flags = LK_EXCLUSIVE;
565 lock_flags = LK_SHARED;
566 vn_lock(vp, lock_flags | LK_RETRY);
570 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
572 if ((ioflg & IO_NOMACCHECK) == 0) {
574 error = mac_vnode_check_read(active_cred, file_cred,
577 error = mac_vnode_check_write(active_cred, file_cred,
582 if (file_cred != NULL)
586 if (do_vn_io_fault(vp, &auio)) {
587 args.kind = VN_IO_FAULT_VOP;
590 args.args.vop_args.vp = vp;
591 error = vn_io_fault1(vp, &auio, &args, td);
592 } else if (rw == UIO_READ) {
593 error = VOP_READ(vp, &auio, ioflg, cred);
594 } else /* if (rw == UIO_WRITE) */ {
595 error = VOP_WRITE(vp, &auio, ioflg, cred);
599 *aresid = auio.uio_resid;
601 if (auio.uio_resid && error == 0)
603 if ((ioflg & IO_NODELOCKED) == 0) {
606 vn_finished_write(mp);
609 if (rl_cookie != NULL)
610 vn_rangelock_unlock(vp, rl_cookie);
615 * Package up an I/O request on a vnode into a uio and do it. The I/O
616 * request is split up into smaller chunks and we try to avoid saturating
617 * the buffer cache while potentially holding a vnode locked, so we
618 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
619 * to give other processes a chance to lock the vnode (either other processes
620 * core'ing the same binary, or unrelated processes scanning the directory).
623 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
624 file_cred, aresid, td)
632 struct ucred *active_cred;
633 struct ucred *file_cred;
644 * Force `offset' to a multiple of MAXBSIZE except possibly
645 * for the first chunk, so that filesystems only need to
646 * write full blocks except possibly for the first and last
649 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
653 if (rw != UIO_READ && vp->v_type == VREG)
656 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
657 ioflg, active_cred, file_cred, &iaresid, td);
658 len -= chunk; /* aresid calc already includes length */
662 base = (char *)base + chunk;
663 kern_yield(PRI_USER);
666 *aresid = len + iaresid;
671 foffset_lock(struct file *fp, int flags)
676 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
678 #if OFF_MAX <= LONG_MAX
680 * Caller only wants the current f_offset value. Assume that
681 * the long and shorter integer types reads are atomic.
683 if ((flags & FOF_NOLOCK) != 0)
684 return (fp->f_offset);
688 * According to McKusick the vn lock was protecting f_offset here.
689 * It is now protected by the FOFFSET_LOCKED flag.
691 mtxp = mtx_pool_find(mtxpool_sleep, fp);
693 if ((flags & FOF_NOLOCK) == 0) {
694 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
695 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
696 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
699 fp->f_vnread_flags |= FOFFSET_LOCKED;
707 foffset_unlock(struct file *fp, off_t val, int flags)
711 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
713 #if OFF_MAX <= LONG_MAX
714 if ((flags & FOF_NOLOCK) != 0) {
715 if ((flags & FOF_NOUPDATE) == 0)
717 if ((flags & FOF_NEXTOFF) != 0)
723 mtxp = mtx_pool_find(mtxpool_sleep, fp);
725 if ((flags & FOF_NOUPDATE) == 0)
727 if ((flags & FOF_NEXTOFF) != 0)
729 if ((flags & FOF_NOLOCK) == 0) {
730 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
731 ("Lost FOFFSET_LOCKED"));
732 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
733 wakeup(&fp->f_vnread_flags);
734 fp->f_vnread_flags = 0;
740 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
743 if ((flags & FOF_OFFSET) == 0)
744 uio->uio_offset = foffset_lock(fp, flags);
748 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
751 if ((flags & FOF_OFFSET) == 0)
752 foffset_unlock(fp, uio->uio_offset, flags);
756 get_advice(struct file *fp, struct uio *uio)
761 ret = POSIX_FADV_NORMAL;
762 if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG)
765 mtxp = mtx_pool_find(mtxpool_sleep, fp);
767 if (fp->f_advice != NULL &&
768 uio->uio_offset >= fp->f_advice->fa_start &&
769 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
770 ret = fp->f_advice->fa_advice;
776 * File table vnode read routine.
779 vn_read(fp, uio, active_cred, flags, td)
782 struct ucred *active_cred;
791 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
793 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
796 if (fp->f_flag & FNONBLOCK)
798 if (fp->f_flag & O_DIRECT)
800 advice = get_advice(fp, uio);
801 vn_lock(vp, LK_SHARED | LK_RETRY);
804 case POSIX_FADV_NORMAL:
805 case POSIX_FADV_SEQUENTIAL:
806 case POSIX_FADV_NOREUSE:
807 ioflag |= sequential_heuristic(uio, fp);
809 case POSIX_FADV_RANDOM:
810 /* Disable read-ahead for random I/O. */
813 orig_offset = uio->uio_offset;
816 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
819 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
820 fp->f_nextoff = uio->uio_offset;
822 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
823 orig_offset != uio->uio_offset)
825 * Use POSIX_FADV_DONTNEED to flush pages and buffers
826 * for the backing file after a POSIX_FADV_NOREUSE
829 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
830 POSIX_FADV_DONTNEED);
835 * File table vnode write routine.
838 vn_write(fp, uio, active_cred, flags, td)
841 struct ucred *active_cred;
848 int error, ioflag, lock_flags;
851 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
853 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
855 if (vp->v_type == VREG)
858 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
860 if (fp->f_flag & FNONBLOCK)
862 if (fp->f_flag & O_DIRECT)
864 if ((fp->f_flag & O_FSYNC) ||
865 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
868 if (vp->v_type != VCHR &&
869 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
872 advice = get_advice(fp, uio);
874 if (MNT_SHARED_WRITES(mp) ||
875 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
876 lock_flags = LK_SHARED;
878 lock_flags = LK_EXCLUSIVE;
881 vn_lock(vp, lock_flags | LK_RETRY);
883 case POSIX_FADV_NORMAL:
884 case POSIX_FADV_SEQUENTIAL:
885 case POSIX_FADV_NOREUSE:
886 ioflag |= sequential_heuristic(uio, fp);
888 case POSIX_FADV_RANDOM:
889 /* XXX: Is this correct? */
892 orig_offset = uio->uio_offset;
895 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
898 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
899 fp->f_nextoff = uio->uio_offset;
901 if (vp->v_type != VCHR)
902 vn_finished_write(mp);
903 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
904 orig_offset != uio->uio_offset)
906 * Use POSIX_FADV_DONTNEED to flush pages and buffers
907 * for the backing file after a POSIX_FADV_NOREUSE
910 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
911 POSIX_FADV_DONTNEED);
917 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
918 * prevent the following deadlock:
920 * Assume that the thread A reads from the vnode vp1 into userspace
921 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
922 * currently not resident, then system ends up with the call chain
923 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
924 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
925 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
926 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
927 * backed by the pages of vnode vp1, and some page in buf2 is not
928 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
930 * To prevent the lock order reversal and deadlock, vn_io_fault() does
931 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
932 * Instead, it first tries to do the whole range i/o with pagefaults
933 * disabled. If all pages in the i/o buffer are resident and mapped,
934 * VOP will succeed (ignoring the genuine filesystem errors).
935 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
936 * i/o in chunks, with all pages in the chunk prefaulted and held
937 * using vm_fault_quick_hold_pages().
939 * Filesystems using this deadlock avoidance scheme should use the
940 * array of the held pages from uio, saved in the curthread->td_ma,
941 * instead of doing uiomove(). A helper function
942 * vn_io_fault_uiomove() converts uiomove request into
943 * uiomove_fromphys() over td_ma array.
945 * Since vnode locks do not cover the whole i/o anymore, rangelocks
946 * make the current i/o request atomic with respect to other i/os and
951 * Decode vn_io_fault_args and perform the corresponding i/o.
954 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
958 switch (args->kind) {
959 case VN_IO_FAULT_FOP:
960 return ((args->args.fop_args.doio)(args->args.fop_args.fp,
961 uio, args->cred, args->flags, td));
962 case VN_IO_FAULT_VOP:
963 if (uio->uio_rw == UIO_READ) {
964 return (VOP_READ(args->args.vop_args.vp, uio,
965 args->flags, args->cred));
966 } else if (uio->uio_rw == UIO_WRITE) {
967 return (VOP_WRITE(args->args.vop_args.vp, uio,
968 args->flags, args->cred));
972 panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
977 vn_io_fault_touch(char *base, const struct uio *uio)
982 if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1))
988 vn_io_fault_prefault_user(const struct uio *uio)
991 const struct iovec *iov;
996 KASSERT(uio->uio_segflg == UIO_USERSPACE,
997 ("vn_io_fault_prefault userspace"));
1001 resid = uio->uio_resid;
1002 base = iov->iov_base;
1005 error = vn_io_fault_touch(base, uio);
1008 if (len < PAGE_SIZE) {
1010 error = vn_io_fault_touch(base + len - 1, uio);
1015 if (++i >= uio->uio_iovcnt)
1017 iov = uio->uio_iov + i;
1018 base = iov->iov_base;
1030 * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1031 * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1032 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1033 * into args and call vn_io_fault1() to handle faults during the user
1034 * mode buffer accesses.
1037 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1040 vm_page_t ma[io_hold_cnt + 2];
1041 struct uio *uio_clone, short_uio;
1042 struct iovec short_iovec[1];
1043 vm_page_t *prev_td_ma;
1045 vm_offset_t addr, end;
1048 int error, cnt, save, saveheld, prev_td_ma_cnt;
1050 if (vn_io_fault_prefault) {
1051 error = vn_io_fault_prefault_user(uio);
1053 return (error); /* Or ignore ? */
1056 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1059 * The UFS follows IO_UNIT directive and replays back both
1060 * uio_offset and uio_resid if an error is encountered during the
1061 * operation. But, since the iovec may be already advanced,
1062 * uio is still in an inconsistent state.
1064 * Cache a copy of the original uio, which is advanced to the redo
1065 * point using UIO_NOCOPY below.
1067 uio_clone = cloneuio(uio);
1068 resid = uio->uio_resid;
1070 short_uio.uio_segflg = UIO_USERSPACE;
1071 short_uio.uio_rw = uio->uio_rw;
1072 short_uio.uio_td = uio->uio_td;
1074 save = vm_fault_disable_pagefaults();
1075 error = vn_io_fault_doio(args, uio, td);
1076 if (error != EFAULT)
1079 atomic_add_long(&vn_io_faults_cnt, 1);
1080 uio_clone->uio_segflg = UIO_NOCOPY;
1081 uiomove(NULL, resid - uio->uio_resid, uio_clone);
1082 uio_clone->uio_segflg = uio->uio_segflg;
1084 saveheld = curthread_pflags_set(TDP_UIOHELD);
1085 prev_td_ma = td->td_ma;
1086 prev_td_ma_cnt = td->td_ma_cnt;
1088 while (uio_clone->uio_resid != 0) {
1089 len = uio_clone->uio_iov->iov_len;
1091 KASSERT(uio_clone->uio_iovcnt >= 1,
1092 ("iovcnt underflow"));
1093 uio_clone->uio_iov++;
1094 uio_clone->uio_iovcnt--;
1097 if (len > io_hold_cnt * PAGE_SIZE)
1098 len = io_hold_cnt * PAGE_SIZE;
1099 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1100 end = round_page(addr + len);
1105 cnt = atop(end - trunc_page(addr));
1107 * A perfectly misaligned address and length could cause
1108 * both the start and the end of the chunk to use partial
1109 * page. +2 accounts for such a situation.
1111 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1112 addr, len, prot, ma, io_hold_cnt + 2);
1117 short_uio.uio_iov = &short_iovec[0];
1118 short_iovec[0].iov_base = (void *)addr;
1119 short_uio.uio_iovcnt = 1;
1120 short_uio.uio_resid = short_iovec[0].iov_len = len;
1121 short_uio.uio_offset = uio_clone->uio_offset;
1123 td->td_ma_cnt = cnt;
1125 error = vn_io_fault_doio(args, &short_uio, td);
1126 vm_page_unhold_pages(ma, cnt);
1127 adv = len - short_uio.uio_resid;
1129 uio_clone->uio_iov->iov_base =
1130 (char *)uio_clone->uio_iov->iov_base + adv;
1131 uio_clone->uio_iov->iov_len -= adv;
1132 uio_clone->uio_resid -= adv;
1133 uio_clone->uio_offset += adv;
1135 uio->uio_resid -= adv;
1136 uio->uio_offset += adv;
1138 if (error != 0 || adv == 0)
1141 td->td_ma = prev_td_ma;
1142 td->td_ma_cnt = prev_td_ma_cnt;
1143 curthread_pflags_restore(saveheld);
1145 vm_fault_enable_pagefaults(save);
1146 free(uio_clone, M_IOV);
1151 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1152 int flags, struct thread *td)
1157 struct vn_io_fault_args args;
1160 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1162 foffset_lock_uio(fp, uio, flags);
1163 if (do_vn_io_fault(vp, uio)) {
1164 args.kind = VN_IO_FAULT_FOP;
1165 args.args.fop_args.fp = fp;
1166 args.args.fop_args.doio = doio;
1167 args.cred = active_cred;
1168 args.flags = flags | FOF_OFFSET;
1169 if (uio->uio_rw == UIO_READ) {
1170 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1171 uio->uio_offset + uio->uio_resid);
1172 } else if ((fp->f_flag & O_APPEND) != 0 ||
1173 (flags & FOF_OFFSET) == 0) {
1174 /* For appenders, punt and lock the whole range. */
1175 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1177 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1178 uio->uio_offset + uio->uio_resid);
1180 error = vn_io_fault1(vp, uio, &args, td);
1181 vn_rangelock_unlock(vp, rl_cookie);
1183 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1185 foffset_unlock_uio(fp, uio, flags);
1190 * Helper function to perform the requested uiomove operation using
1191 * the held pages for io->uio_iov[0].iov_base buffer instead of
1192 * copyin/copyout. Access to the pages with uiomove_fromphys()
1193 * instead of iov_base prevents page faults that could occur due to
1194 * pmap_collect() invalidating the mapping created by
1195 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1196 * object cleanup revoking the write access from page mappings.
1198 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1199 * instead of plain uiomove().
1202 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1204 struct uio transp_uio;
1205 struct iovec transp_iov[1];
1211 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1212 uio->uio_segflg != UIO_USERSPACE)
1213 return (uiomove(data, xfersize, uio));
1215 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1216 transp_iov[0].iov_base = data;
1217 transp_uio.uio_iov = &transp_iov[0];
1218 transp_uio.uio_iovcnt = 1;
1219 if (xfersize > uio->uio_resid)
1220 xfersize = uio->uio_resid;
1221 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1222 transp_uio.uio_offset = 0;
1223 transp_uio.uio_segflg = UIO_SYSSPACE;
1225 * Since transp_iov points to data, and td_ma page array
1226 * corresponds to original uio->uio_iov, we need to invert the
1227 * direction of the i/o operation as passed to
1228 * uiomove_fromphys().
1230 switch (uio->uio_rw) {
1232 transp_uio.uio_rw = UIO_READ;
1235 transp_uio.uio_rw = UIO_WRITE;
1238 transp_uio.uio_td = uio->uio_td;
1239 error = uiomove_fromphys(td->td_ma,
1240 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1241 xfersize, &transp_uio);
1242 adv = xfersize - transp_uio.uio_resid;
1244 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1245 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1247 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1249 td->td_ma_cnt -= pgadv;
1250 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1251 uio->uio_iov->iov_len -= adv;
1252 uio->uio_resid -= adv;
1253 uio->uio_offset += adv;
1258 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1262 vm_offset_t iov_base;
1266 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1267 uio->uio_segflg != UIO_USERSPACE)
1268 return (uiomove_fromphys(ma, offset, xfersize, uio));
1270 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1271 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1272 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1273 switch (uio->uio_rw) {
1275 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1279 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1283 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1285 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1287 td->td_ma_cnt -= pgadv;
1288 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1289 uio->uio_iov->iov_len -= cnt;
1290 uio->uio_resid -= cnt;
1291 uio->uio_offset += cnt;
1297 * File table truncate routine.
1300 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1312 * Lock the whole range for truncation. Otherwise split i/o
1313 * might happen partly before and partly after the truncation.
1315 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1316 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1319 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1320 AUDIT_ARG_VNODE1(vp);
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(struct vnode *vp, struct stat *sb, struct ucred *active_cred,
1371 struct ucred *file_cred, struct thread *td)
1378 AUDIT_ARG_VNODE1(vp);
1380 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1388 * Initialize defaults for new and unusual fields, so that file
1389 * systems which don't support these fields don't need to know
1392 vap->va_birthtime.tv_sec = -1;
1393 vap->va_birthtime.tv_nsec = 0;
1394 vap->va_fsid = VNOVAL;
1395 vap->va_rdev = NODEV;
1397 error = VOP_GETATTR(vp, vap, active_cred);
1402 * Zero the spare stat fields
1404 bzero(sb, sizeof *sb);
1407 * Copy from vattr table
1409 if (vap->va_fsid != VNOVAL)
1410 sb->st_dev = vap->va_fsid;
1412 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1413 sb->st_ino = vap->va_fileid;
1414 mode = vap->va_mode;
1415 switch (vap->va_type) {
1441 sb->st_nlink = vap->va_nlink;
1442 sb->st_uid = vap->va_uid;
1443 sb->st_gid = vap->va_gid;
1444 sb->st_rdev = vap->va_rdev;
1445 if (vap->va_size > OFF_MAX)
1447 sb->st_size = vap->va_size;
1448 sb->st_atim = vap->va_atime;
1449 sb->st_mtim = vap->va_mtime;
1450 sb->st_ctim = vap->va_ctime;
1451 sb->st_birthtim = vap->va_birthtime;
1454 * According to www.opengroup.org, the meaning of st_blksize is
1455 * "a filesystem-specific preferred I/O block size for this
1456 * object. In some filesystem types, this may vary from file
1458 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1461 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1463 sb->st_flags = vap->va_flags;
1464 if (priv_check(td, PRIV_VFS_GENERATION))
1467 sb->st_gen = vap->va_gen;
1469 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1474 * File table vnode ioctl routine.
1477 vn_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred,
1485 switch (vp->v_type) {
1490 vn_lock(vp, LK_SHARED | LK_RETRY);
1491 error = VOP_GETATTR(vp, &vattr, active_cred);
1494 *(int *)data = vattr.va_size - fp->f_offset;
1500 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1505 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1513 * File table vnode poll routine.
1516 vn_poll(struct file *fp, int events, struct ucred *active_cred,
1524 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1525 AUDIT_ARG_VNODE1(vp);
1526 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1531 error = VOP_POLL(vp, events, fp->f_cred, td);
1536 * Acquire the requested lock and then check for validity. LK_RETRY
1537 * permits vn_lock to return doomed vnodes.
1540 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1544 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1545 ("vn_lock: no locktype"));
1546 VNASSERT(vp->v_holdcnt != 0, vp, ("vn_lock: zero hold count"));
1548 error = VOP_LOCK1(vp, flags, file, line);
1549 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1550 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1551 ("vn_lock: error %d incompatible with flags %#x", error, flags));
1553 if ((flags & LK_RETRY) == 0) {
1554 if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0) {
1558 } else if (error != 0)
1564 * File table vnode close routine.
1567 vn_closefile(struct file *fp, struct thread *td)
1575 fp->f_ops = &badfileops;
1576 ref= (fp->f_flag & FHASLOCK) != 0 && fp->f_type == DTYPE_VNODE;
1578 error = vn_close1(vp, fp->f_flag, fp->f_cred, td, ref);
1580 if (__predict_false(ref)) {
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(struct mount *mp)
1764 if (mp == NULL || !vn_suspendable(mp))
1768 mp->mnt_writeopcount--;
1769 if (mp->mnt_writeopcount < 0)
1770 panic("vn_finished_write: neg cnt");
1771 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1772 mp->mnt_writeopcount <= 0)
1773 wakeup(&mp->mnt_writeopcount);
1779 * Filesystem secondary write operation has completed. If we are
1780 * suspending and this operation is the last one, notify the suspender
1781 * that the suspension is now in effect.
1784 vn_finished_secondary_write(struct mount *mp)
1786 if (mp == NULL || !vn_suspendable(mp))
1790 mp->mnt_secondary_writes--;
1791 if (mp->mnt_secondary_writes < 0)
1792 panic("vn_finished_secondary_write: neg cnt");
1793 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1794 mp->mnt_secondary_writes <= 0)
1795 wakeup(&mp->mnt_secondary_writes);
1802 * Request a filesystem to suspend write operations.
1805 vfs_write_suspend(struct mount *mp, int flags)
1809 MPASS(vn_suspendable(mp));
1812 if (mp->mnt_susp_owner == curthread) {
1816 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1817 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1820 * Unmount holds a write reference on the mount point. If we
1821 * own busy reference and drain for writers, we deadlock with
1822 * the reference draining in the unmount path. Callers of
1823 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1824 * vfs_busy() reference is owned and caller is not in the
1827 if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1828 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1833 mp->mnt_kern_flag |= MNTK_SUSPEND;
1834 mp->mnt_susp_owner = curthread;
1835 if (mp->mnt_writeopcount > 0)
1836 (void) msleep(&mp->mnt_writeopcount,
1837 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1840 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1841 vfs_write_resume(mp, 0);
1846 * Request a filesystem to resume write operations.
1849 vfs_write_resume(struct mount *mp, int flags)
1852 MPASS(vn_suspendable(mp));
1855 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1856 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1857 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1859 mp->mnt_susp_owner = NULL;
1860 wakeup(&mp->mnt_writeopcount);
1861 wakeup(&mp->mnt_flag);
1862 curthread->td_pflags &= ~TDP_IGNSUSP;
1863 if ((flags & VR_START_WRITE) != 0) {
1865 mp->mnt_writeopcount++;
1868 if ((flags & VR_NO_SUSPCLR) == 0)
1870 } else if ((flags & VR_START_WRITE) != 0) {
1872 vn_start_write_locked(mp, 0);
1879 * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1883 vfs_write_suspend_umnt(struct mount *mp)
1887 MPASS(vn_suspendable(mp));
1888 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1889 ("vfs_write_suspend_umnt: recursed"));
1891 /* dounmount() already called vn_start_write(). */
1893 vn_finished_write(mp);
1894 error = vfs_write_suspend(mp, 0);
1896 vn_start_write(NULL, &mp, V_WAIT);
1900 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1903 vn_start_write(NULL, &mp, V_WAIT);
1905 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1906 wakeup(&mp->mnt_flag);
1908 curthread->td_pflags |= TDP_IGNSUSP;
1913 * Implement kqueues for files by translating it to vnode operation.
1916 vn_kqfilter(struct file *fp, struct knote *kn)
1919 return (VOP_KQFILTER(fp->f_vnode, kn));
1923 * Simplified in-kernel wrapper calls for extended attribute access.
1924 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1925 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1928 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1929 const char *attrname, int *buflen, char *buf, struct thread *td)
1935 iov.iov_len = *buflen;
1938 auio.uio_iov = &iov;
1939 auio.uio_iovcnt = 1;
1940 auio.uio_rw = UIO_READ;
1941 auio.uio_segflg = UIO_SYSSPACE;
1943 auio.uio_offset = 0;
1944 auio.uio_resid = *buflen;
1946 if ((ioflg & IO_NODELOCKED) == 0)
1947 vn_lock(vp, LK_SHARED | LK_RETRY);
1949 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1951 /* authorize attribute retrieval as kernel */
1952 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1955 if ((ioflg & IO_NODELOCKED) == 0)
1959 *buflen = *buflen - auio.uio_resid;
1966 * XXX failure mode if partially written?
1969 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1970 const char *attrname, int buflen, char *buf, struct thread *td)
1977 iov.iov_len = buflen;
1980 auio.uio_iov = &iov;
1981 auio.uio_iovcnt = 1;
1982 auio.uio_rw = UIO_WRITE;
1983 auio.uio_segflg = UIO_SYSSPACE;
1985 auio.uio_offset = 0;
1986 auio.uio_resid = buflen;
1988 if ((ioflg & IO_NODELOCKED) == 0) {
1989 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1991 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1994 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1996 /* authorize attribute setting as kernel */
1997 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1999 if ((ioflg & IO_NODELOCKED) == 0) {
2000 vn_finished_write(mp);
2008 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
2009 const char *attrname, struct thread *td)
2014 if ((ioflg & IO_NODELOCKED) == 0) {
2015 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2017 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2020 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2022 /* authorize attribute removal as kernel */
2023 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2024 if (error == EOPNOTSUPP)
2025 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2028 if ((ioflg & IO_NODELOCKED) == 0) {
2029 vn_finished_write(mp);
2037 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2041 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2045 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2048 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2053 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2054 int lkflags, struct vnode **rvp)
2059 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2061 ltype = VOP_ISLOCKED(vp);
2062 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2063 ("vn_vget_ino: vp not locked"));
2064 error = vfs_busy(mp, MBF_NOWAIT);
2068 error = vfs_busy(mp, 0);
2069 vn_lock(vp, ltype | LK_RETRY);
2073 if (vp->v_iflag & VI_DOOMED) {
2079 error = alloc(mp, alloc_arg, lkflags, rvp);
2082 vn_lock(vp, ltype | LK_RETRY);
2083 if (vp->v_iflag & VI_DOOMED) {
2096 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2100 if (vp->v_type != VREG || td == NULL)
2102 if ((uoff_t)uio->uio_offset + uio->uio_resid >
2103 lim_cur(td, RLIMIT_FSIZE)) {
2104 PROC_LOCK(td->td_proc);
2105 kern_psignal(td->td_proc, SIGXFSZ);
2106 PROC_UNLOCK(td->td_proc);
2113 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2120 vn_lock(vp, LK_SHARED | LK_RETRY);
2121 AUDIT_ARG_VNODE1(vp);
2124 return (setfmode(td, active_cred, vp, mode));
2128 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2135 vn_lock(vp, LK_SHARED | LK_RETRY);
2136 AUDIT_ARG_VNODE1(vp);
2139 return (setfown(td, active_cred, vp, uid, gid));
2143 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2147 if ((object = vp->v_object) == NULL)
2149 VM_OBJECT_WLOCK(object);
2150 vm_object_page_remove(object, start, end, 0);
2151 VM_OBJECT_WUNLOCK(object);
2155 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2163 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2164 ("Wrong command %lu", cmd));
2166 if (vn_lock(vp, LK_SHARED) != 0)
2168 if (vp->v_type != VREG) {
2172 error = VOP_GETATTR(vp, &va, cred);
2176 if (noff >= va.va_size) {
2180 bsize = vp->v_mount->mnt_stat.f_iosize;
2181 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2182 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2183 if (error == EOPNOTSUPP) {
2187 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2188 (bnp != -1 && cmd == FIOSEEKDATA)) {
2195 if (noff > va.va_size)
2197 /* noff == va.va_size. There is an implicit hole at the end of file. */
2198 if (cmd == FIOSEEKDATA)
2208 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2213 off_t foffset, size;
2216 cred = td->td_ucred;
2218 foffset = foffset_lock(fp, 0);
2219 noneg = (vp->v_type != VCHR);
2225 (offset > 0 && foffset > OFF_MAX - offset))) {
2232 vn_lock(vp, LK_SHARED | LK_RETRY);
2233 error = VOP_GETATTR(vp, &vattr, cred);
2239 * If the file references a disk device, then fetch
2240 * the media size and use that to determine the ending
2243 if (vattr.va_size == 0 && vp->v_type == VCHR &&
2244 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2245 vattr.va_size = size;
2247 (vattr.va_size > OFF_MAX ||
2248 (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2252 offset += vattr.va_size;
2257 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2260 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2265 if (error == 0 && noneg && offset < 0)
2269 VFS_KNOTE_UNLOCKED(vp, 0);
2270 td->td_uretoff.tdu_off = offset;
2272 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2277 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2283 * Grant permission if the caller is the owner of the file, or
2284 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2285 * on the file. If the time pointer is null, then write
2286 * permission on the file is also sufficient.
2288 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2289 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2290 * will be allowed to set the times [..] to the current
2293 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2294 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2295 error = VOP_ACCESS(vp, VWRITE, cred, td);
2300 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2305 if (fp->f_type == DTYPE_FIFO)
2306 kif->kf_type = KF_TYPE_FIFO;
2308 kif->kf_type = KF_TYPE_VNODE;
2311 FILEDESC_SUNLOCK(fdp);
2312 error = vn_fill_kinfo_vnode(vp, kif);
2314 FILEDESC_SLOCK(fdp);
2319 vn_fill_junk(struct kinfo_file *kif)
2324 * Simulate vn_fullpath returning changing values for a given
2325 * vp during e.g. coredump.
2327 len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1;
2328 olen = strlen(kif->kf_path);
2330 strcpy(&kif->kf_path[len - 1], "$");
2332 for (; olen < len; olen++)
2333 strcpy(&kif->kf_path[olen], "A");
2337 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2340 char *fullpath, *freepath;
2343 kif->kf_un.kf_file.kf_file_type = vntype_to_kinfo(vp->v_type);
2346 error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2348 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2350 if (freepath != NULL)
2351 free(freepath, M_TEMP);
2353 KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path,
2358 * Retrieve vnode attributes.
2360 va.va_fsid = VNOVAL;
2362 vn_lock(vp, LK_SHARED | LK_RETRY);
2363 error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2367 if (va.va_fsid != VNOVAL)
2368 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2370 kif->kf_un.kf_file.kf_file_fsid =
2371 vp->v_mount->mnt_stat.f_fsid.val[0];
2372 kif->kf_un.kf_file.kf_file_fsid_freebsd11 =
2373 kif->kf_un.kf_file.kf_file_fsid; /* truncate */
2374 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2375 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2376 kif->kf_un.kf_file.kf_file_size = va.va_size;
2377 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2378 kif->kf_un.kf_file.kf_file_rdev_freebsd11 =
2379 kif->kf_un.kf_file.kf_file_rdev; /* truncate */
2384 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
2385 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
2389 struct pmckern_map_in pkm;
2395 boolean_t writecounted;
2398 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
2399 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
2401 * POSIX shared-memory objects are defined to have
2402 * kernel persistence, and are not defined to support
2403 * read(2)/write(2) -- or even open(2). Thus, we can
2404 * use MAP_ASYNC to trade on-disk coherence for speed.
2405 * The shm_open(3) library routine turns on the FPOSIXSHM
2406 * flag to request this behavior.
2408 if ((fp->f_flag & FPOSIXSHM) != 0)
2409 flags |= MAP_NOSYNC;
2414 * Ensure that file and memory protections are
2415 * compatible. Note that we only worry about
2416 * writability if mapping is shared; in this case,
2417 * current and max prot are dictated by the open file.
2418 * XXX use the vnode instead? Problem is: what
2419 * credentials do we use for determination? What if
2420 * proc does a setuid?
2423 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
2424 maxprot = VM_PROT_NONE;
2425 if ((prot & VM_PROT_EXECUTE) != 0)
2428 maxprot = VM_PROT_EXECUTE;
2429 if ((fp->f_flag & FREAD) != 0)
2430 maxprot |= VM_PROT_READ;
2431 else if ((prot & VM_PROT_READ) != 0)
2435 * If we are sharing potential changes via MAP_SHARED and we
2436 * are trying to get write permission although we opened it
2437 * without asking for it, bail out.
2439 if ((flags & MAP_SHARED) != 0) {
2440 if ((fp->f_flag & FWRITE) != 0)
2441 maxprot |= VM_PROT_WRITE;
2442 else if ((prot & VM_PROT_WRITE) != 0)
2445 maxprot |= VM_PROT_WRITE;
2446 cap_maxprot |= VM_PROT_WRITE;
2448 maxprot &= cap_maxprot;
2451 * For regular files and shared memory, POSIX requires that
2452 * the value of foff be a legitimate offset within the data
2453 * object. In particular, negative offsets are invalid.
2454 * Blocking negative offsets and overflows here avoids
2455 * possible wraparound or user-level access into reserved
2456 * ranges of the data object later. In contrast, POSIX does
2457 * not dictate how offsets are used by device drivers, so in
2458 * the case of a device mapping a negative offset is passed
2465 foff < 0 || foff > OFF_MAX - size)
2468 writecounted = FALSE;
2469 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp,
2470 &foff, &object, &writecounted);
2473 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2474 foff, writecounted, td);
2477 * If this mapping was accounted for in the vnode's
2478 * writecount, then undo that now.
2481 vnode_pager_release_writecount(object, 0, size);
2482 vm_object_deallocate(object);
2485 /* Inform hwpmc(4) if an executable is being mapped. */
2486 if (PMC_HOOK_INSTALLED(PMC_FN_MMAP)) {
2487 if ((prot & VM_PROT_EXECUTE) != 0 && error == 0) {
2489 pkm.pm_address = (uintptr_t) *addr;
2490 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);
2498 vn_fsid(struct vnode *vp, struct vattr *va)
2502 f = &vp->v_mount->mnt_stat.f_fsid;
2503 va->va_fsid = (uint32_t)f->val[1];
2504 va->va_fsid <<= sizeof(f->val[1]) * NBBY;
2505 va->va_fsid += (uint32_t)f->val[0];