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)
305 int error, have_flock, lock_flags, type;
307 if (vp->v_type == VLNK)
309 if (vp->v_type == VSOCK)
311 if (vp->v_type != VDIR && fmode & O_DIRECTORY)
314 if (fmode & (FWRITE | O_TRUNC)) {
315 if (vp->v_type == VDIR)
323 if ((fmode & O_APPEND) && (fmode & FWRITE))
328 if (fmode & O_VERIFY)
330 error = mac_vnode_check_open(cred, vp, accmode);
334 accmode &= ~(VCREAT | VVERIFY);
336 if ((fmode & O_CREAT) == 0) {
337 if (accmode & VWRITE) {
338 error = vn_writechk(vp);
343 error = VOP_ACCESS(vp, accmode, cred, td);
348 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
349 vn_lock(vp, LK_UPGRADE | LK_RETRY);
350 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
353 if (fmode & (O_EXLOCK | O_SHLOCK)) {
354 KASSERT(fp != NULL, ("open with flock requires fp"));
355 lock_flags = VOP_ISLOCKED(vp);
357 lf.l_whence = SEEK_SET;
360 if (fmode & O_EXLOCK)
365 if ((fmode & FNONBLOCK) == 0)
367 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
368 have_flock = (error == 0);
369 vn_lock(vp, lock_flags | LK_RETRY);
370 if (error == 0 && vp->v_iflag & VI_DOOMED)
373 * Another thread might have used this vnode as an
374 * executable while the vnode lock was dropped.
375 * Ensure the vnode is still able to be opened for
376 * writing after the lock has been obtained.
378 if (error == 0 && accmode & VWRITE)
379 error = vn_writechk(vp);
383 lf.l_whence = SEEK_SET;
387 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
390 vn_start_write(vp, &mp, V_WAIT);
391 vn_lock(vp, lock_flags | LK_RETRY);
392 (void)VOP_CLOSE(vp, fmode, cred, td);
393 vn_finished_write(mp);
394 /* Prevent second close from fdrop()->vn_close(). */
396 fp->f_ops= &badfileops;
399 fp->f_flag |= FHASLOCK;
401 if (fmode & FWRITE) {
402 VOP_ADD_WRITECOUNT(vp, 1);
403 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
404 __func__, vp, vp->v_writecount);
406 ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
411 * Check for write permissions on the specified vnode.
412 * Prototype text segments cannot be written.
416 register struct vnode *vp;
419 ASSERT_VOP_LOCKED(vp, "vn_writechk");
421 * If there's shared text associated with
422 * the vnode, try to free it up once. If
423 * we fail, we can't allow writing.
435 vn_close(vp, flags, file_cred, td)
436 register struct vnode *vp;
438 struct ucred *file_cred;
442 int error, lock_flags;
444 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
445 MNT_EXTENDED_SHARED(vp->v_mount))
446 lock_flags = LK_SHARED;
448 lock_flags = LK_EXCLUSIVE;
450 vn_start_write(vp, &mp, V_WAIT);
451 vn_lock(vp, lock_flags | LK_RETRY);
452 if (flags & FWRITE) {
453 VNASSERT(vp->v_writecount > 0, vp,
454 ("vn_close: negative writecount"));
455 VOP_ADD_WRITECOUNT(vp, -1);
456 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
457 __func__, vp, vp->v_writecount);
459 error = VOP_CLOSE(vp, flags, file_cred, td);
461 vn_finished_write(mp);
466 * Heuristic to detect sequential operation.
469 sequential_heuristic(struct uio *uio, struct file *fp)
472 ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
473 if (fp->f_flag & FRDAHEAD)
474 return (fp->f_seqcount << IO_SEQSHIFT);
477 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
478 * that the first I/O is normally considered to be slightly
479 * sequential. Seeking to offset 0 doesn't change sequentiality
480 * unless previous seeks have reduced f_seqcount to 0, in which
481 * case offset 0 is not special.
483 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
484 uio->uio_offset == fp->f_nextoff) {
486 * f_seqcount is in units of fixed-size blocks so that it
487 * depends mainly on the amount of sequential I/O and not
488 * much on the number of sequential I/O's. The fixed size
489 * of 16384 is hard-coded here since it is (not quite) just
490 * a magic size that works well here. This size is more
491 * closely related to the best I/O size for real disks than
492 * to any block size used by software.
494 fp->f_seqcount += howmany(uio->uio_resid, 16384);
495 if (fp->f_seqcount > IO_SEQMAX)
496 fp->f_seqcount = IO_SEQMAX;
497 return (fp->f_seqcount << IO_SEQSHIFT);
500 /* Not sequential. Quickly draw-down sequentiality. */
501 if (fp->f_seqcount > 1)
509 * Package up an I/O request on a vnode into a uio and do it.
512 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
513 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
514 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
521 struct vn_io_fault_args args;
522 int error, lock_flags;
524 auio.uio_iov = &aiov;
526 aiov.iov_base = base;
528 auio.uio_resid = len;
529 auio.uio_offset = offset;
530 auio.uio_segflg = segflg;
535 if ((ioflg & IO_NODELOCKED) == 0) {
536 if ((ioflg & IO_RANGELOCKED) == 0) {
537 if (rw == UIO_READ) {
538 rl_cookie = vn_rangelock_rlock(vp, offset,
541 rl_cookie = vn_rangelock_wlock(vp, offset,
547 if (rw == UIO_WRITE) {
548 if (vp->v_type != VCHR &&
549 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
552 if (MNT_SHARED_WRITES(mp) ||
553 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
554 lock_flags = LK_SHARED;
556 lock_flags = LK_EXCLUSIVE;
558 lock_flags = LK_SHARED;
559 vn_lock(vp, lock_flags | LK_RETRY);
563 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
565 if ((ioflg & IO_NOMACCHECK) == 0) {
567 error = mac_vnode_check_read(active_cred, file_cred,
570 error = mac_vnode_check_write(active_cred, file_cred,
575 if (file_cred != NULL)
579 if (do_vn_io_fault(vp, &auio)) {
580 args.kind = VN_IO_FAULT_VOP;
583 args.args.vop_args.vp = vp;
584 error = vn_io_fault1(vp, &auio, &args, td);
585 } else if (rw == UIO_READ) {
586 error = VOP_READ(vp, &auio, ioflg, cred);
587 } else /* if (rw == UIO_WRITE) */ {
588 error = VOP_WRITE(vp, &auio, ioflg, cred);
592 *aresid = auio.uio_resid;
594 if (auio.uio_resid && error == 0)
596 if ((ioflg & IO_NODELOCKED) == 0) {
599 vn_finished_write(mp);
602 if (rl_cookie != NULL)
603 vn_rangelock_unlock(vp, rl_cookie);
608 * Package up an I/O request on a vnode into a uio and do it. The I/O
609 * request is split up into smaller chunks and we try to avoid saturating
610 * the buffer cache while potentially holding a vnode locked, so we
611 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
612 * to give other processes a chance to lock the vnode (either other processes
613 * core'ing the same binary, or unrelated processes scanning the directory).
616 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
617 file_cred, aresid, td)
625 struct ucred *active_cred;
626 struct ucred *file_cred;
637 * Force `offset' to a multiple of MAXBSIZE except possibly
638 * for the first chunk, so that filesystems only need to
639 * write full blocks except possibly for the first and last
642 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
646 if (rw != UIO_READ && vp->v_type == VREG)
649 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
650 ioflg, active_cred, file_cred, &iaresid, td);
651 len -= chunk; /* aresid calc already includes length */
655 base = (char *)base + chunk;
656 kern_yield(PRI_USER);
659 *aresid = len + iaresid;
664 foffset_lock(struct file *fp, int flags)
669 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
671 #if OFF_MAX <= LONG_MAX
673 * Caller only wants the current f_offset value. Assume that
674 * the long and shorter integer types reads are atomic.
676 if ((flags & FOF_NOLOCK) != 0)
677 return (fp->f_offset);
681 * According to McKusick the vn lock was protecting f_offset here.
682 * It is now protected by the FOFFSET_LOCKED flag.
684 mtxp = mtx_pool_find(mtxpool_sleep, fp);
686 if ((flags & FOF_NOLOCK) == 0) {
687 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
688 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
689 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
692 fp->f_vnread_flags |= FOFFSET_LOCKED;
700 foffset_unlock(struct file *fp, off_t val, int flags)
704 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
706 #if OFF_MAX <= LONG_MAX
707 if ((flags & FOF_NOLOCK) != 0) {
708 if ((flags & FOF_NOUPDATE) == 0)
710 if ((flags & FOF_NEXTOFF) != 0)
716 mtxp = mtx_pool_find(mtxpool_sleep, fp);
718 if ((flags & FOF_NOUPDATE) == 0)
720 if ((flags & FOF_NEXTOFF) != 0)
722 if ((flags & FOF_NOLOCK) == 0) {
723 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
724 ("Lost FOFFSET_LOCKED"));
725 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
726 wakeup(&fp->f_vnread_flags);
727 fp->f_vnread_flags = 0;
733 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
736 if ((flags & FOF_OFFSET) == 0)
737 uio->uio_offset = foffset_lock(fp, flags);
741 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
744 if ((flags & FOF_OFFSET) == 0)
745 foffset_unlock(fp, uio->uio_offset, flags);
749 get_advice(struct file *fp, struct uio *uio)
754 ret = POSIX_FADV_NORMAL;
755 if (fp->f_advice == NULL)
758 mtxp = mtx_pool_find(mtxpool_sleep, fp);
760 if (uio->uio_offset >= fp->f_advice->fa_start &&
761 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
762 ret = fp->f_advice->fa_advice;
768 * File table vnode read routine.
771 vn_read(fp, uio, active_cred, flags, td)
774 struct ucred *active_cred;
783 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
785 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
788 if (fp->f_flag & FNONBLOCK)
790 if (fp->f_flag & O_DIRECT)
792 advice = get_advice(fp, uio);
793 vn_lock(vp, LK_SHARED | LK_RETRY);
796 case POSIX_FADV_NORMAL:
797 case POSIX_FADV_SEQUENTIAL:
798 case POSIX_FADV_NOREUSE:
799 ioflag |= sequential_heuristic(uio, fp);
801 case POSIX_FADV_RANDOM:
802 /* Disable read-ahead for random I/O. */
805 orig_offset = uio->uio_offset;
808 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
811 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
812 fp->f_nextoff = uio->uio_offset;
814 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
815 orig_offset != uio->uio_offset)
817 * Use POSIX_FADV_DONTNEED to flush pages and buffers
818 * for the backing file after a POSIX_FADV_NOREUSE
821 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
822 POSIX_FADV_DONTNEED);
827 * File table vnode write routine.
830 vn_write(fp, uio, active_cred, flags, td)
833 struct ucred *active_cred;
840 int error, ioflag, lock_flags;
843 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
845 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
847 if (vp->v_type == VREG)
850 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
852 if (fp->f_flag & FNONBLOCK)
854 if (fp->f_flag & O_DIRECT)
856 if ((fp->f_flag & O_FSYNC) ||
857 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
860 if (vp->v_type != VCHR &&
861 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
864 advice = get_advice(fp, uio);
866 if (MNT_SHARED_WRITES(mp) ||
867 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
868 lock_flags = LK_SHARED;
870 lock_flags = LK_EXCLUSIVE;
873 vn_lock(vp, lock_flags | LK_RETRY);
875 case POSIX_FADV_NORMAL:
876 case POSIX_FADV_SEQUENTIAL:
877 case POSIX_FADV_NOREUSE:
878 ioflag |= sequential_heuristic(uio, fp);
880 case POSIX_FADV_RANDOM:
881 /* XXX: Is this correct? */
884 orig_offset = uio->uio_offset;
887 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
890 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
891 fp->f_nextoff = uio->uio_offset;
893 if (vp->v_type != VCHR)
894 vn_finished_write(mp);
895 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
896 orig_offset != uio->uio_offset)
898 * Use POSIX_FADV_DONTNEED to flush pages and buffers
899 * for the backing file after a POSIX_FADV_NOREUSE
902 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
903 POSIX_FADV_DONTNEED);
909 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
910 * prevent the following deadlock:
912 * Assume that the thread A reads from the vnode vp1 into userspace
913 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
914 * currently not resident, then system ends up with the call chain
915 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
916 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
917 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
918 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
919 * backed by the pages of vnode vp1, and some page in buf2 is not
920 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
922 * To prevent the lock order reversal and deadlock, vn_io_fault() does
923 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
924 * Instead, it first tries to do the whole range i/o with pagefaults
925 * disabled. If all pages in the i/o buffer are resident and mapped,
926 * VOP will succeed (ignoring the genuine filesystem errors).
927 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
928 * i/o in chunks, with all pages in the chunk prefaulted and held
929 * using vm_fault_quick_hold_pages().
931 * Filesystems using this deadlock avoidance scheme should use the
932 * array of the held pages from uio, saved in the curthread->td_ma,
933 * instead of doing uiomove(). A helper function
934 * vn_io_fault_uiomove() converts uiomove request into
935 * uiomove_fromphys() over td_ma array.
937 * Since vnode locks do not cover the whole i/o anymore, rangelocks
938 * make the current i/o request atomic with respect to other i/os and
943 * Decode vn_io_fault_args and perform the corresponding i/o.
946 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
950 switch (args->kind) {
951 case VN_IO_FAULT_FOP:
952 return ((args->args.fop_args.doio)(args->args.fop_args.fp,
953 uio, args->cred, args->flags, td));
954 case VN_IO_FAULT_VOP:
955 if (uio->uio_rw == UIO_READ) {
956 return (VOP_READ(args->args.vop_args.vp, uio,
957 args->flags, args->cred));
958 } else if (uio->uio_rw == UIO_WRITE) {
959 return (VOP_WRITE(args->args.vop_args.vp, uio,
960 args->flags, args->cred));
964 panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
969 vn_io_fault_touch(char *base, const struct uio *uio)
974 if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1))
980 vn_io_fault_prefault_user(const struct uio *uio)
983 const struct iovec *iov;
988 KASSERT(uio->uio_segflg == UIO_USERSPACE,
989 ("vn_io_fault_prefault userspace"));
993 resid = uio->uio_resid;
994 base = iov->iov_base;
997 error = vn_io_fault_touch(base, uio);
1000 if (len < PAGE_SIZE) {
1002 error = vn_io_fault_touch(base + len - 1, uio);
1007 if (++i >= uio->uio_iovcnt)
1009 iov = uio->uio_iov + i;
1010 base = iov->iov_base;
1022 * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1023 * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1024 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1025 * into args and call vn_io_fault1() to handle faults during the user
1026 * mode buffer accesses.
1029 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1032 vm_page_t ma[io_hold_cnt + 2];
1033 struct uio *uio_clone, short_uio;
1034 struct iovec short_iovec[1];
1035 vm_page_t *prev_td_ma;
1037 vm_offset_t addr, end;
1040 int error, cnt, save, saveheld, prev_td_ma_cnt;
1042 if (vn_io_fault_prefault) {
1043 error = vn_io_fault_prefault_user(uio);
1045 return (error); /* Or ignore ? */
1048 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1051 * The UFS follows IO_UNIT directive and replays back both
1052 * uio_offset and uio_resid if an error is encountered during the
1053 * operation. But, since the iovec may be already advanced,
1054 * uio is still in an inconsistent state.
1056 * Cache a copy of the original uio, which is advanced to the redo
1057 * point using UIO_NOCOPY below.
1059 uio_clone = cloneuio(uio);
1060 resid = uio->uio_resid;
1062 short_uio.uio_segflg = UIO_USERSPACE;
1063 short_uio.uio_rw = uio->uio_rw;
1064 short_uio.uio_td = uio->uio_td;
1066 save = vm_fault_disable_pagefaults();
1067 error = vn_io_fault_doio(args, uio, td);
1068 if (error != EFAULT)
1071 atomic_add_long(&vn_io_faults_cnt, 1);
1072 uio_clone->uio_segflg = UIO_NOCOPY;
1073 uiomove(NULL, resid - uio->uio_resid, uio_clone);
1074 uio_clone->uio_segflg = uio->uio_segflg;
1076 saveheld = curthread_pflags_set(TDP_UIOHELD);
1077 prev_td_ma = td->td_ma;
1078 prev_td_ma_cnt = td->td_ma_cnt;
1080 while (uio_clone->uio_resid != 0) {
1081 len = uio_clone->uio_iov->iov_len;
1083 KASSERT(uio_clone->uio_iovcnt >= 1,
1084 ("iovcnt underflow"));
1085 uio_clone->uio_iov++;
1086 uio_clone->uio_iovcnt--;
1089 if (len > io_hold_cnt * PAGE_SIZE)
1090 len = io_hold_cnt * PAGE_SIZE;
1091 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1092 end = round_page(addr + len);
1097 cnt = atop(end - trunc_page(addr));
1099 * A perfectly misaligned address and length could cause
1100 * both the start and the end of the chunk to use partial
1101 * page. +2 accounts for such a situation.
1103 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1104 addr, len, prot, ma, io_hold_cnt + 2);
1109 short_uio.uio_iov = &short_iovec[0];
1110 short_iovec[0].iov_base = (void *)addr;
1111 short_uio.uio_iovcnt = 1;
1112 short_uio.uio_resid = short_iovec[0].iov_len = len;
1113 short_uio.uio_offset = uio_clone->uio_offset;
1115 td->td_ma_cnt = cnt;
1117 error = vn_io_fault_doio(args, &short_uio, td);
1118 vm_page_unhold_pages(ma, cnt);
1119 adv = len - short_uio.uio_resid;
1121 uio_clone->uio_iov->iov_base =
1122 (char *)uio_clone->uio_iov->iov_base + adv;
1123 uio_clone->uio_iov->iov_len -= adv;
1124 uio_clone->uio_resid -= adv;
1125 uio_clone->uio_offset += adv;
1127 uio->uio_resid -= adv;
1128 uio->uio_offset += adv;
1130 if (error != 0 || adv == 0)
1133 td->td_ma = prev_td_ma;
1134 td->td_ma_cnt = prev_td_ma_cnt;
1135 curthread_pflags_restore(saveheld);
1137 vm_fault_enable_pagefaults(save);
1138 free(uio_clone, M_IOV);
1143 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1144 int flags, struct thread *td)
1149 struct vn_io_fault_args args;
1152 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1154 foffset_lock_uio(fp, uio, flags);
1155 if (do_vn_io_fault(vp, uio)) {
1156 args.kind = VN_IO_FAULT_FOP;
1157 args.args.fop_args.fp = fp;
1158 args.args.fop_args.doio = doio;
1159 args.cred = active_cred;
1160 args.flags = flags | FOF_OFFSET;
1161 if (uio->uio_rw == UIO_READ) {
1162 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1163 uio->uio_offset + uio->uio_resid);
1164 } else if ((fp->f_flag & O_APPEND) != 0 ||
1165 (flags & FOF_OFFSET) == 0) {
1166 /* For appenders, punt and lock the whole range. */
1167 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1169 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1170 uio->uio_offset + uio->uio_resid);
1172 error = vn_io_fault1(vp, uio, &args, td);
1173 vn_rangelock_unlock(vp, rl_cookie);
1175 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1177 foffset_unlock_uio(fp, uio, flags);
1182 * Helper function to perform the requested uiomove operation using
1183 * the held pages for io->uio_iov[0].iov_base buffer instead of
1184 * copyin/copyout. Access to the pages with uiomove_fromphys()
1185 * instead of iov_base prevents page faults that could occur due to
1186 * pmap_collect() invalidating the mapping created by
1187 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1188 * object cleanup revoking the write access from page mappings.
1190 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1191 * instead of plain uiomove().
1194 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1196 struct uio transp_uio;
1197 struct iovec transp_iov[1];
1203 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1204 uio->uio_segflg != UIO_USERSPACE)
1205 return (uiomove(data, xfersize, uio));
1207 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1208 transp_iov[0].iov_base = data;
1209 transp_uio.uio_iov = &transp_iov[0];
1210 transp_uio.uio_iovcnt = 1;
1211 if (xfersize > uio->uio_resid)
1212 xfersize = uio->uio_resid;
1213 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1214 transp_uio.uio_offset = 0;
1215 transp_uio.uio_segflg = UIO_SYSSPACE;
1217 * Since transp_iov points to data, and td_ma page array
1218 * corresponds to original uio->uio_iov, we need to invert the
1219 * direction of the i/o operation as passed to
1220 * uiomove_fromphys().
1222 switch (uio->uio_rw) {
1224 transp_uio.uio_rw = UIO_READ;
1227 transp_uio.uio_rw = UIO_WRITE;
1230 transp_uio.uio_td = uio->uio_td;
1231 error = uiomove_fromphys(td->td_ma,
1232 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1233 xfersize, &transp_uio);
1234 adv = xfersize - transp_uio.uio_resid;
1236 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1237 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1239 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1241 td->td_ma_cnt -= pgadv;
1242 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1243 uio->uio_iov->iov_len -= adv;
1244 uio->uio_resid -= adv;
1245 uio->uio_offset += adv;
1250 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1254 vm_offset_t iov_base;
1258 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1259 uio->uio_segflg != UIO_USERSPACE)
1260 return (uiomove_fromphys(ma, offset, xfersize, uio));
1262 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1263 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1264 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1265 switch (uio->uio_rw) {
1267 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1271 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1275 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1277 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1279 td->td_ma_cnt -= pgadv;
1280 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1281 uio->uio_iov->iov_len -= cnt;
1282 uio->uio_resid -= cnt;
1283 uio->uio_offset += cnt;
1289 * File table truncate routine.
1292 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1304 * Lock the whole range for truncation. Otherwise split i/o
1305 * might happen partly before and partly after the truncation.
1307 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1308 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1311 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1312 if (vp->v_type == VDIR) {
1317 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1321 error = vn_writechk(vp);
1324 vattr.va_size = length;
1325 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1329 vn_finished_write(mp);
1331 vn_rangelock_unlock(vp, rl_cookie);
1336 * File table vnode stat routine.
1339 vn_statfile(fp, sb, active_cred, td)
1342 struct ucred *active_cred;
1345 struct vnode *vp = fp->f_vnode;
1348 vn_lock(vp, LK_SHARED | LK_RETRY);
1349 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1356 * Stat a vnode; implementation for the stat syscall
1359 vn_stat(vp, sb, active_cred, file_cred, td)
1361 register struct stat *sb;
1362 struct ucred *active_cred;
1363 struct ucred *file_cred;
1367 register struct vattr *vap;
1372 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1380 * Initialize defaults for new and unusual fields, so that file
1381 * systems which don't support these fields don't need to know
1384 vap->va_birthtime.tv_sec = -1;
1385 vap->va_birthtime.tv_nsec = 0;
1386 vap->va_fsid = VNOVAL;
1387 vap->va_rdev = NODEV;
1389 error = VOP_GETATTR(vp, vap, active_cred);
1394 * Zero the spare stat fields
1396 bzero(sb, sizeof *sb);
1399 * Copy from vattr table
1401 if (vap->va_fsid != VNOVAL)
1402 sb->st_dev = vap->va_fsid;
1404 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1405 sb->st_ino = vap->va_fileid;
1406 mode = vap->va_mode;
1407 switch (vap->va_type) {
1433 sb->st_nlink = vap->va_nlink;
1434 sb->st_uid = vap->va_uid;
1435 sb->st_gid = vap->va_gid;
1436 sb->st_rdev = vap->va_rdev;
1437 if (vap->va_size > OFF_MAX)
1439 sb->st_size = vap->va_size;
1440 sb->st_atim = vap->va_atime;
1441 sb->st_mtim = vap->va_mtime;
1442 sb->st_ctim = vap->va_ctime;
1443 sb->st_birthtim = vap->va_birthtime;
1446 * According to www.opengroup.org, the meaning of st_blksize is
1447 * "a filesystem-specific preferred I/O block size for this
1448 * object. In some filesystem types, this may vary from file
1450 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1453 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1455 sb->st_flags = vap->va_flags;
1456 if (priv_check(td, PRIV_VFS_GENERATION))
1459 sb->st_gen = vap->va_gen;
1461 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1466 * File table vnode ioctl routine.
1469 vn_ioctl(fp, com, data, active_cred, td)
1473 struct ucred *active_cred;
1481 switch (vp->v_type) {
1486 vn_lock(vp, LK_SHARED | LK_RETRY);
1487 error = VOP_GETATTR(vp, &vattr, active_cred);
1490 *(int *)data = vattr.va_size - fp->f_offset;
1496 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1505 * File table vnode poll routine.
1508 vn_poll(fp, events, active_cred, td)
1511 struct ucred *active_cred;
1519 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1520 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1525 error = VOP_POLL(vp, events, fp->f_cred, td);
1530 * Acquire the requested lock and then check for validity. LK_RETRY
1531 * permits vn_lock to return doomed vnodes.
1534 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1538 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1539 ("vn_lock called with no locktype."));
1541 #ifdef DEBUG_VFS_LOCKS
1542 KASSERT(vp->v_holdcnt != 0,
1543 ("vn_lock %p: zero hold count", vp));
1545 error = VOP_LOCK1(vp, flags, file, line);
1546 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1547 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1548 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1551 * Callers specify LK_RETRY if they wish to get dead vnodes.
1552 * If RETRY is not set, we return ENOENT instead.
1554 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1555 (flags & LK_RETRY) == 0) {
1560 } while (flags & LK_RETRY && error != 0);
1565 * File table vnode close routine.
1568 vn_closefile(fp, td)
1577 fp->f_ops = &badfileops;
1579 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1582 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1584 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1585 lf.l_whence = SEEK_SET;
1588 lf.l_type = F_UNLCK;
1589 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1596 vn_suspendable(struct mount *mp)
1599 return (mp->mnt_op->vfs_susp_clean != NULL);
1603 * Preparing to start a filesystem write operation. If the operation is
1604 * permitted, then we bump the count of operations in progress and
1605 * proceed. If a suspend request is in progress, we wait until the
1606 * suspension is over, and then proceed.
1609 vn_start_write_locked(struct mount *mp, int flags)
1613 mtx_assert(MNT_MTX(mp), MA_OWNED);
1617 * Check on status of suspension.
1619 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1620 mp->mnt_susp_owner != curthread) {
1621 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
1622 (flags & PCATCH) : 0) | (PUSER - 1);
1623 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1624 if (flags & V_NOWAIT) {
1625 error = EWOULDBLOCK;
1628 error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
1634 if (flags & V_XSLEEP)
1636 mp->mnt_writeopcount++;
1638 if (error != 0 || (flags & V_XSLEEP) != 0)
1645 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
1650 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1651 ("V_MNTREF requires mp"));
1655 * If a vnode is provided, get and return the mount point that
1656 * to which it will write.
1659 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1661 if (error != EOPNOTSUPP)
1666 if ((mp = *mpp) == NULL)
1669 if (!vn_suspendable(mp)) {
1670 if (vp != NULL || (flags & V_MNTREF) != 0)
1676 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1678 * As long as a vnode is not provided we need to acquire a
1679 * refcount for the provided mountpoint too, in order to
1680 * emulate a vfs_ref().
1683 if (vp == NULL && (flags & V_MNTREF) == 0)
1686 return (vn_start_write_locked(mp, flags));
1690 * Secondary suspension. Used by operations such as vop_inactive
1691 * routines that are needed by the higher level functions. These
1692 * are allowed to proceed until all the higher level functions have
1693 * completed (indicated by mnt_writeopcount dropping to zero). At that
1694 * time, these operations are halted until the suspension is over.
1697 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags)
1702 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1703 ("V_MNTREF requires mp"));
1707 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1709 if (error != EOPNOTSUPP)
1715 * If we are not suspended or have not yet reached suspended
1716 * mode, then let the operation proceed.
1718 if ((mp = *mpp) == NULL)
1721 if (!vn_suspendable(mp)) {
1722 if (vp != NULL || (flags & V_MNTREF) != 0)
1728 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1730 * As long as a vnode is not provided we need to acquire a
1731 * refcount for the provided mountpoint too, in order to
1732 * emulate a vfs_ref().
1735 if (vp == NULL && (flags & V_MNTREF) == 0)
1737 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1738 mp->mnt_secondary_writes++;
1739 mp->mnt_secondary_accwrites++;
1743 if (flags & V_NOWAIT) {
1746 return (EWOULDBLOCK);
1749 * Wait for the suspension to finish.
1751 error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
1752 ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
1761 * Filesystem write operation has completed. If we are suspending and this
1762 * operation is the last one, notify the suspender that the suspension is
1766 vn_finished_write(mp)
1769 if (mp == NULL || !vn_suspendable(mp))
1773 mp->mnt_writeopcount--;
1774 if (mp->mnt_writeopcount < 0)
1775 panic("vn_finished_write: neg cnt");
1776 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1777 mp->mnt_writeopcount <= 0)
1778 wakeup(&mp->mnt_writeopcount);
1784 * Filesystem secondary write operation has completed. If we are
1785 * suspending and this operation is the last one, notify the suspender
1786 * that the suspension is now in effect.
1789 vn_finished_secondary_write(mp)
1792 if (mp == NULL || !vn_suspendable(mp))
1796 mp->mnt_secondary_writes--;
1797 if (mp->mnt_secondary_writes < 0)
1798 panic("vn_finished_secondary_write: neg cnt");
1799 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1800 mp->mnt_secondary_writes <= 0)
1801 wakeup(&mp->mnt_secondary_writes);
1808 * Request a filesystem to suspend write operations.
1811 vfs_write_suspend(struct mount *mp, int flags)
1815 MPASS(vn_suspendable(mp));
1818 if (mp->mnt_susp_owner == curthread) {
1822 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1823 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1826 * Unmount holds a write reference on the mount point. If we
1827 * own busy reference and drain for writers, we deadlock with
1828 * the reference draining in the unmount path. Callers of
1829 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1830 * vfs_busy() reference is owned and caller is not in the
1833 if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1834 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1839 mp->mnt_kern_flag |= MNTK_SUSPEND;
1840 mp->mnt_susp_owner = curthread;
1841 if (mp->mnt_writeopcount > 0)
1842 (void) msleep(&mp->mnt_writeopcount,
1843 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1846 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1847 vfs_write_resume(mp, 0);
1852 * Request a filesystem to resume write operations.
1855 vfs_write_resume(struct mount *mp, int flags)
1858 MPASS(vn_suspendable(mp));
1861 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1862 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1863 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1865 mp->mnt_susp_owner = NULL;
1866 wakeup(&mp->mnt_writeopcount);
1867 wakeup(&mp->mnt_flag);
1868 curthread->td_pflags &= ~TDP_IGNSUSP;
1869 if ((flags & VR_START_WRITE) != 0) {
1871 mp->mnt_writeopcount++;
1874 if ((flags & VR_NO_SUSPCLR) == 0)
1876 } else if ((flags & VR_START_WRITE) != 0) {
1878 vn_start_write_locked(mp, 0);
1885 * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1889 vfs_write_suspend_umnt(struct mount *mp)
1893 MPASS(vn_suspendable(mp));
1894 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1895 ("vfs_write_suspend_umnt: recursed"));
1897 /* dounmount() already called vn_start_write(). */
1899 vn_finished_write(mp);
1900 error = vfs_write_suspend(mp, 0);
1902 vn_start_write(NULL, &mp, V_WAIT);
1906 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1909 vn_start_write(NULL, &mp, V_WAIT);
1911 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1912 wakeup(&mp->mnt_flag);
1914 curthread->td_pflags |= TDP_IGNSUSP;
1919 * Implement kqueues for files by translating it to vnode operation.
1922 vn_kqfilter(struct file *fp, struct knote *kn)
1925 return (VOP_KQFILTER(fp->f_vnode, kn));
1929 * Simplified in-kernel wrapper calls for extended attribute access.
1930 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1931 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1934 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1935 const char *attrname, int *buflen, char *buf, struct thread *td)
1941 iov.iov_len = *buflen;
1944 auio.uio_iov = &iov;
1945 auio.uio_iovcnt = 1;
1946 auio.uio_rw = UIO_READ;
1947 auio.uio_segflg = UIO_SYSSPACE;
1949 auio.uio_offset = 0;
1950 auio.uio_resid = *buflen;
1952 if ((ioflg & IO_NODELOCKED) == 0)
1953 vn_lock(vp, LK_SHARED | LK_RETRY);
1955 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1957 /* authorize attribute retrieval as kernel */
1958 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1961 if ((ioflg & IO_NODELOCKED) == 0)
1965 *buflen = *buflen - auio.uio_resid;
1972 * XXX failure mode if partially written?
1975 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1976 const char *attrname, int buflen, char *buf, struct thread *td)
1983 iov.iov_len = buflen;
1986 auio.uio_iov = &iov;
1987 auio.uio_iovcnt = 1;
1988 auio.uio_rw = UIO_WRITE;
1989 auio.uio_segflg = UIO_SYSSPACE;
1991 auio.uio_offset = 0;
1992 auio.uio_resid = buflen;
1994 if ((ioflg & IO_NODELOCKED) == 0) {
1995 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1997 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2000 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2002 /* authorize attribute setting as kernel */
2003 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
2005 if ((ioflg & IO_NODELOCKED) == 0) {
2006 vn_finished_write(mp);
2014 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
2015 const char *attrname, struct thread *td)
2020 if ((ioflg & IO_NODELOCKED) == 0) {
2021 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2023 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2026 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2028 /* authorize attribute removal as kernel */
2029 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2030 if (error == EOPNOTSUPP)
2031 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2034 if ((ioflg & IO_NODELOCKED) == 0) {
2035 vn_finished_write(mp);
2043 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2047 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2051 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2054 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2059 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2060 int lkflags, struct vnode **rvp)
2065 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2067 ltype = VOP_ISLOCKED(vp);
2068 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2069 ("vn_vget_ino: vp not locked"));
2070 error = vfs_busy(mp, MBF_NOWAIT);
2074 error = vfs_busy(mp, 0);
2075 vn_lock(vp, ltype | LK_RETRY);
2079 if (vp->v_iflag & VI_DOOMED) {
2085 error = alloc(mp, alloc_arg, lkflags, rvp);
2088 vn_lock(vp, ltype | LK_RETRY);
2089 if (vp->v_iflag & VI_DOOMED) {
2102 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2106 if (vp->v_type != VREG || td == NULL)
2108 if ((uoff_t)uio->uio_offset + uio->uio_resid >
2109 lim_cur(td, RLIMIT_FSIZE)) {
2110 PROC_LOCK(td->td_proc);
2111 kern_psignal(td->td_proc, SIGXFSZ);
2112 PROC_UNLOCK(td->td_proc);
2119 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2126 vn_lock(vp, LK_SHARED | LK_RETRY);
2127 AUDIT_ARG_VNODE1(vp);
2130 return (setfmode(td, active_cred, vp, mode));
2134 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2141 vn_lock(vp, LK_SHARED | LK_RETRY);
2142 AUDIT_ARG_VNODE1(vp);
2145 return (setfown(td, active_cred, vp, uid, gid));
2149 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2153 if ((object = vp->v_object) == NULL)
2155 VM_OBJECT_WLOCK(object);
2156 vm_object_page_remove(object, start, end, 0);
2157 VM_OBJECT_WUNLOCK(object);
2161 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2169 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2170 ("Wrong command %lu", cmd));
2172 if (vn_lock(vp, LK_SHARED) != 0)
2174 if (vp->v_type != VREG) {
2178 error = VOP_GETATTR(vp, &va, cred);
2182 if (noff >= va.va_size) {
2186 bsize = vp->v_mount->mnt_stat.f_iosize;
2187 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2188 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2189 if (error == EOPNOTSUPP) {
2193 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2194 (bnp != -1 && cmd == FIOSEEKDATA)) {
2201 if (noff > va.va_size)
2203 /* noff == va.va_size. There is an implicit hole at the end of file. */
2204 if (cmd == FIOSEEKDATA)
2214 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2219 off_t foffset, size;
2222 cred = td->td_ucred;
2224 foffset = foffset_lock(fp, 0);
2225 noneg = (vp->v_type != VCHR);
2231 (offset > 0 && foffset > OFF_MAX - offset))) {
2238 vn_lock(vp, LK_SHARED | LK_RETRY);
2239 error = VOP_GETATTR(vp, &vattr, cred);
2245 * If the file references a disk device, then fetch
2246 * the media size and use that to determine the ending
2249 if (vattr.va_size == 0 && vp->v_type == VCHR &&
2250 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2251 vattr.va_size = size;
2253 (vattr.va_size > OFF_MAX ||
2254 (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2258 offset += vattr.va_size;
2263 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2266 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2271 if (error == 0 && noneg && offset < 0)
2275 VFS_KNOTE_UNLOCKED(vp, 0);
2276 td->td_uretoff.tdu_off = offset;
2278 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2283 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2289 * Grant permission if the caller is the owner of the file, or
2290 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2291 * on the file. If the time pointer is null, then write
2292 * permission on the file is also sufficient.
2294 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2295 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2296 * will be allowed to set the times [..] to the current
2299 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2300 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2301 error = VOP_ACCESS(vp, VWRITE, cred, td);
2306 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2311 if (fp->f_type == DTYPE_FIFO)
2312 kif->kf_type = KF_TYPE_FIFO;
2314 kif->kf_type = KF_TYPE_VNODE;
2317 FILEDESC_SUNLOCK(fdp);
2318 error = vn_fill_kinfo_vnode(vp, kif);
2320 FILEDESC_SLOCK(fdp);
2325 vn_fill_junk(struct kinfo_file *kif)
2330 * Simulate vn_fullpath returning changing values for a given
2331 * vp during e.g. coredump.
2333 len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1;
2334 olen = strlen(kif->kf_path);
2336 strcpy(&kif->kf_path[len - 1], "$");
2338 for (; olen < len; olen++)
2339 strcpy(&kif->kf_path[olen], "A");
2343 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2346 char *fullpath, *freepath;
2349 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
2352 error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2354 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2356 if (freepath != NULL)
2357 free(freepath, M_TEMP);
2359 KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path,
2364 * Retrieve vnode attributes.
2366 va.va_fsid = VNOVAL;
2368 vn_lock(vp, LK_SHARED | LK_RETRY);
2369 error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2373 if (va.va_fsid != VNOVAL)
2374 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2376 kif->kf_un.kf_file.kf_file_fsid =
2377 vp->v_mount->mnt_stat.f_fsid.val[0];
2378 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2379 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2380 kif->kf_un.kf_file.kf_file_size = va.va_size;
2381 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2386 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
2387 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
2391 struct pmckern_map_in pkm;
2397 boolean_t writecounted;
2400 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
2401 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
2403 * POSIX shared-memory objects are defined to have
2404 * kernel persistence, and are not defined to support
2405 * read(2)/write(2) -- or even open(2). Thus, we can
2406 * use MAP_ASYNC to trade on-disk coherence for speed.
2407 * The shm_open(3) library routine turns on the FPOSIXSHM
2408 * flag to request this behavior.
2410 if ((fp->f_flag & FPOSIXSHM) != 0)
2411 flags |= MAP_NOSYNC;
2416 * Ensure that file and memory protections are
2417 * compatible. Note that we only worry about
2418 * writability if mapping is shared; in this case,
2419 * current and max prot are dictated by the open file.
2420 * XXX use the vnode instead? Problem is: what
2421 * credentials do we use for determination? What if
2422 * proc does a setuid?
2425 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0)
2426 maxprot = VM_PROT_NONE;
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;
2450 writecounted = FALSE;
2451 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp,
2452 &foff, &object, &writecounted);
2455 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2456 foff, writecounted, td);
2459 * If this mapping was accounted for in the vnode's
2460 * writecount, then undo that now.
2463 vnode_pager_release_writecount(object, 0, size);
2464 vm_object_deallocate(object);
2467 /* Inform hwpmc(4) if an executable is being mapped. */
2468 if (error == 0 && (prot & VM_PROT_EXECUTE) != 0) {
2470 pkm.pm_address = (uintptr_t) *addr;
2471 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);