2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org>
11 * Copyright (c) 2013, 2014 The FreeBSD Foundation
13 * Portions of this software were developed by Konstantin Belousov
14 * under sponsorship from the FreeBSD Foundation.
16 * Redistribution and use in source and binary forms, with or without
17 * modification, are permitted provided that the following conditions
19 * 1. Redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer.
21 * 2. Redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution.
24 * 4. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
46 #include <sys/param.h>
47 #include <sys/systm.h>
50 #include <sys/fcntl.h>
56 #include <sys/limits.h>
59 #include <sys/mount.h>
60 #include <sys/mutex.h>
61 #include <sys/namei.h>
62 #include <sys/vnode.h>
65 #include <sys/filio.h>
66 #include <sys/resourcevar.h>
67 #include <sys/rwlock.h>
69 #include <sys/sysctl.h>
70 #include <sys/ttycom.h>
72 #include <sys/syslog.h>
73 #include <sys/unistd.h>
76 #include <security/audit/audit.h>
77 #include <security/mac/mac_framework.h>
80 #include <vm/vm_extern.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_object.h>
84 #include <vm/vm_page.h>
85 #include <vm/vnode_pager.h>
87 static fo_rdwr_t vn_read;
88 static fo_rdwr_t vn_write;
89 static fo_rdwr_t vn_io_fault;
90 static fo_truncate_t vn_truncate;
91 static fo_ioctl_t vn_ioctl;
92 static fo_poll_t vn_poll;
93 static fo_kqfilter_t vn_kqfilter;
94 static fo_stat_t vn_statfile;
95 static fo_close_t vn_closefile;
96 static fo_mmap_t vn_mmap;
98 struct fileops vnops = {
99 .fo_read = vn_io_fault,
100 .fo_write = vn_io_fault,
101 .fo_truncate = vn_truncate,
102 .fo_ioctl = vn_ioctl,
104 .fo_kqfilter = vn_kqfilter,
105 .fo_stat = vn_statfile,
106 .fo_close = vn_closefile,
107 .fo_chmod = vn_chmod,
108 .fo_chown = vn_chown,
109 .fo_sendfile = vn_sendfile,
111 .fo_fill_kinfo = vn_fill_kinfo,
113 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
116 static const int io_hold_cnt = 16;
117 static int vn_io_fault_enable = 1;
118 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
119 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
120 static int vn_io_fault_prefault = 0;
121 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW,
122 &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting");
123 static u_long vn_io_faults_cnt;
124 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
125 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
128 * Returns true if vn_io_fault mode of handling the i/o request should
132 do_vn_io_fault(struct vnode *vp, struct uio *uio)
136 return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
137 (mp = vp->v_mount) != NULL &&
138 (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
142 * Structure used to pass arguments to vn_io_fault1(), to do either
143 * file- or vnode-based I/O calls.
145 struct vn_io_fault_args {
153 struct fop_args_tag {
157 struct vop_args_tag {
163 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
164 struct vn_io_fault_args *args, struct thread *td);
167 vn_open(ndp, flagp, cmode, fp)
168 struct nameidata *ndp;
172 struct thread *td = ndp->ni_cnd.cn_thread;
174 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
178 * Common code for vnode open operations via a name lookup.
179 * Lookup the vnode and invoke VOP_CREATE if needed.
180 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
182 * Note that this does NOT free nameidata for the successful case,
183 * due to the NDINIT being done elsewhere.
186 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
187 struct ucred *cred, struct file *fp)
191 struct thread *td = ndp->ni_cnd.cn_thread;
193 struct vattr *vap = &vat;
198 if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT |
199 O_EXCL | O_DIRECTORY))
201 else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) {
202 ndp->ni_cnd.cn_nameiop = CREATE;
204 * Set NOCACHE to avoid flushing the cache when
205 * rolling in many files at once.
207 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE;
208 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
209 ndp->ni_cnd.cn_flags |= FOLLOW;
210 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
211 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
212 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
213 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
215 if ((error = namei(ndp)) != 0)
217 if (ndp->ni_vp == NULL) {
220 vap->va_mode = cmode;
222 vap->va_vaflags |= VA_EXCLUSIVE;
223 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
224 NDFREE(ndp, NDF_ONLY_PNBUF);
226 if ((error = vn_start_write(NULL, &mp,
227 V_XSLEEP | PCATCH)) != 0)
231 if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0)
232 ndp->ni_cnd.cn_flags |= MAKEENTRY;
234 error = mac_vnode_check_create(cred, ndp->ni_dvp,
238 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
241 vn_finished_write(mp);
243 NDFREE(ndp, NDF_ONLY_PNBUF);
249 if (ndp->ni_dvp == ndp->ni_vp)
255 if (fmode & O_EXCL) {
262 ndp->ni_cnd.cn_nameiop = LOOKUP;
263 ndp->ni_cnd.cn_flags = ISOPEN |
264 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
265 if (!(fmode & FWRITE))
266 ndp->ni_cnd.cn_flags |= LOCKSHARED;
267 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
268 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
269 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
270 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
271 if ((error = namei(ndp)) != 0)
275 error = vn_open_vnode(vp, fmode, cred, td, fp);
281 NDFREE(ndp, NDF_ONLY_PNBUF);
289 * Common code for vnode open operations once a vnode is located.
290 * Check permissions, and call the VOP_OPEN routine.
293 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
294 struct thread *td, struct file *fp)
299 int error, have_flock, lock_flags, type;
301 if (vp->v_type == VLNK)
303 if (vp->v_type == VSOCK)
305 if (vp->v_type != VDIR && fmode & O_DIRECTORY)
308 if (fmode & (FWRITE | O_TRUNC)) {
309 if (vp->v_type == VDIR)
317 if ((fmode & O_APPEND) && (fmode & FWRITE))
322 if (fmode & O_VERIFY)
324 error = mac_vnode_check_open(cred, vp, accmode);
328 accmode &= ~(VCREAT | VVERIFY);
330 if ((fmode & O_CREAT) == 0) {
331 if (accmode & VWRITE) {
332 error = vn_writechk(vp);
337 error = VOP_ACCESS(vp, accmode, cred, td);
342 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
343 vn_lock(vp, LK_UPGRADE | LK_RETRY);
344 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
347 if (fmode & (O_EXLOCK | O_SHLOCK)) {
348 KASSERT(fp != NULL, ("open with flock requires fp"));
349 lock_flags = VOP_ISLOCKED(vp);
351 lf.l_whence = SEEK_SET;
354 if (fmode & O_EXLOCK)
359 if ((fmode & FNONBLOCK) == 0)
361 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
362 have_flock = (error == 0);
363 vn_lock(vp, lock_flags | LK_RETRY);
364 if (error == 0 && vp->v_iflag & VI_DOOMED)
367 * Another thread might have used this vnode as an
368 * executable while the vnode lock was dropped.
369 * Ensure the vnode is still able to be opened for
370 * writing after the lock has been obtained.
372 if (error == 0 && accmode & VWRITE)
373 error = vn_writechk(vp);
377 lf.l_whence = SEEK_SET;
381 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf,
384 vn_start_write(vp, &mp, V_WAIT);
385 vn_lock(vp, lock_flags | LK_RETRY);
386 (void)VOP_CLOSE(vp, fmode, cred, td);
387 vn_finished_write(mp);
388 /* Prevent second close from fdrop()->vn_close(). */
390 fp->f_ops= &badfileops;
393 fp->f_flag |= FHASLOCK;
395 if (fmode & FWRITE) {
396 VOP_ADD_WRITECOUNT(vp, 1);
397 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
398 __func__, vp, vp->v_writecount);
400 ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
405 * Check for write permissions on the specified vnode.
406 * Prototype text segments cannot be written.
410 register struct vnode *vp;
413 ASSERT_VOP_LOCKED(vp, "vn_writechk");
415 * If there's shared text associated with
416 * the vnode, try to free it up once. If
417 * we fail, we can't allow writing.
429 vn_close(vp, flags, file_cred, td)
430 register struct vnode *vp;
432 struct ucred *file_cred;
436 int error, lock_flags;
438 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
439 MNT_EXTENDED_SHARED(vp->v_mount))
440 lock_flags = LK_SHARED;
442 lock_flags = LK_EXCLUSIVE;
444 vn_start_write(vp, &mp, V_WAIT);
445 vn_lock(vp, lock_flags | LK_RETRY);
446 if (flags & FWRITE) {
447 VNASSERT(vp->v_writecount > 0, vp,
448 ("vn_close: negative writecount"));
449 VOP_ADD_WRITECOUNT(vp, -1);
450 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
451 __func__, vp, vp->v_writecount);
453 error = VOP_CLOSE(vp, flags, file_cred, td);
455 vn_finished_write(mp);
460 * Heuristic to detect sequential operation.
463 sequential_heuristic(struct uio *uio, struct file *fp)
466 ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
467 if (fp->f_flag & FRDAHEAD)
468 return (fp->f_seqcount << IO_SEQSHIFT);
471 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
472 * that the first I/O is normally considered to be slightly
473 * sequential. Seeking to offset 0 doesn't change sequentiality
474 * unless previous seeks have reduced f_seqcount to 0, in which
475 * case offset 0 is not special.
477 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
478 uio->uio_offset == fp->f_nextoff) {
480 * f_seqcount is in units of fixed-size blocks so that it
481 * depends mainly on the amount of sequential I/O and not
482 * much on the number of sequential I/O's. The fixed size
483 * of 16384 is hard-coded here since it is (not quite) just
484 * a magic size that works well here. This size is more
485 * closely related to the best I/O size for real disks than
486 * to any block size used by software.
488 fp->f_seqcount += howmany(uio->uio_resid, 16384);
489 if (fp->f_seqcount > IO_SEQMAX)
490 fp->f_seqcount = IO_SEQMAX;
491 return (fp->f_seqcount << IO_SEQSHIFT);
494 /* Not sequential. Quickly draw-down sequentiality. */
495 if (fp->f_seqcount > 1)
503 * Package up an I/O request on a vnode into a uio and do it.
506 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
507 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
508 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
515 struct vn_io_fault_args args;
516 int error, lock_flags;
518 auio.uio_iov = &aiov;
520 aiov.iov_base = base;
522 auio.uio_resid = len;
523 auio.uio_offset = offset;
524 auio.uio_segflg = segflg;
529 if ((ioflg & IO_NODELOCKED) == 0) {
530 if ((ioflg & IO_RANGELOCKED) == 0) {
531 if (rw == UIO_READ) {
532 rl_cookie = vn_rangelock_rlock(vp, offset,
535 rl_cookie = vn_rangelock_wlock(vp, offset,
541 if (rw == UIO_WRITE) {
542 if (vp->v_type != VCHR &&
543 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
546 if (MNT_SHARED_WRITES(mp) ||
547 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
548 lock_flags = LK_SHARED;
550 lock_flags = LK_EXCLUSIVE;
552 lock_flags = LK_SHARED;
553 vn_lock(vp, lock_flags | LK_RETRY);
557 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
559 if ((ioflg & IO_NOMACCHECK) == 0) {
561 error = mac_vnode_check_read(active_cred, file_cred,
564 error = mac_vnode_check_write(active_cred, file_cred,
569 if (file_cred != NULL)
573 if (do_vn_io_fault(vp, &auio)) {
574 args.kind = VN_IO_FAULT_VOP;
577 args.args.vop_args.vp = vp;
578 error = vn_io_fault1(vp, &auio, &args, td);
579 } else if (rw == UIO_READ) {
580 error = VOP_READ(vp, &auio, ioflg, cred);
581 } else /* if (rw == UIO_WRITE) */ {
582 error = VOP_WRITE(vp, &auio, ioflg, cred);
586 *aresid = auio.uio_resid;
588 if (auio.uio_resid && error == 0)
590 if ((ioflg & IO_NODELOCKED) == 0) {
593 vn_finished_write(mp);
596 if (rl_cookie != NULL)
597 vn_rangelock_unlock(vp, rl_cookie);
602 * Package up an I/O request on a vnode into a uio and do it. The I/O
603 * request is split up into smaller chunks and we try to avoid saturating
604 * the buffer cache while potentially holding a vnode locked, so we
605 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
606 * to give other processes a chance to lock the vnode (either other processes
607 * core'ing the same binary, or unrelated processes scanning the directory).
610 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
611 file_cred, aresid, td)
619 struct ucred *active_cred;
620 struct ucred *file_cred;
631 * Force `offset' to a multiple of MAXBSIZE except possibly
632 * for the first chunk, so that filesystems only need to
633 * write full blocks except possibly for the first and last
636 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
640 if (rw != UIO_READ && vp->v_type == VREG)
643 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
644 ioflg, active_cred, file_cred, &iaresid, td);
645 len -= chunk; /* aresid calc already includes length */
649 base = (char *)base + chunk;
650 kern_yield(PRI_USER);
653 *aresid = len + iaresid;
658 foffset_lock(struct file *fp, int flags)
663 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
665 #if OFF_MAX <= LONG_MAX
667 * Caller only wants the current f_offset value. Assume that
668 * the long and shorter integer types reads are atomic.
670 if ((flags & FOF_NOLOCK) != 0)
671 return (fp->f_offset);
675 * According to McKusick the vn lock was protecting f_offset here.
676 * It is now protected by the FOFFSET_LOCKED flag.
678 mtxp = mtx_pool_find(mtxpool_sleep, fp);
680 if ((flags & FOF_NOLOCK) == 0) {
681 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
682 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
683 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
686 fp->f_vnread_flags |= FOFFSET_LOCKED;
694 foffset_unlock(struct file *fp, off_t val, int flags)
698 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
700 #if OFF_MAX <= LONG_MAX
701 if ((flags & FOF_NOLOCK) != 0) {
702 if ((flags & FOF_NOUPDATE) == 0)
704 if ((flags & FOF_NEXTOFF) != 0)
710 mtxp = mtx_pool_find(mtxpool_sleep, fp);
712 if ((flags & FOF_NOUPDATE) == 0)
714 if ((flags & FOF_NEXTOFF) != 0)
716 if ((flags & FOF_NOLOCK) == 0) {
717 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
718 ("Lost FOFFSET_LOCKED"));
719 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
720 wakeup(&fp->f_vnread_flags);
721 fp->f_vnread_flags = 0;
727 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
730 if ((flags & FOF_OFFSET) == 0)
731 uio->uio_offset = foffset_lock(fp, flags);
735 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
738 if ((flags & FOF_OFFSET) == 0)
739 foffset_unlock(fp, uio->uio_offset, flags);
743 get_advice(struct file *fp, struct uio *uio)
748 ret = POSIX_FADV_NORMAL;
749 if (fp->f_advice == NULL)
752 mtxp = mtx_pool_find(mtxpool_sleep, fp);
754 if (uio->uio_offset >= fp->f_advice->fa_start &&
755 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
756 ret = fp->f_advice->fa_advice;
762 * File table vnode read routine.
765 vn_read(fp, uio, active_cred, flags, td)
768 struct ucred *active_cred;
777 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
779 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
782 if (fp->f_flag & FNONBLOCK)
784 if (fp->f_flag & O_DIRECT)
786 advice = get_advice(fp, uio);
787 vn_lock(vp, LK_SHARED | LK_RETRY);
790 case POSIX_FADV_NORMAL:
791 case POSIX_FADV_SEQUENTIAL:
792 case POSIX_FADV_NOREUSE:
793 ioflag |= sequential_heuristic(uio, fp);
795 case POSIX_FADV_RANDOM:
796 /* Disable read-ahead for random I/O. */
799 orig_offset = uio->uio_offset;
802 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
805 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
806 fp->f_nextoff = uio->uio_offset;
808 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
809 orig_offset != uio->uio_offset)
811 * Use POSIX_FADV_DONTNEED to flush pages and buffers
812 * for the backing file after a POSIX_FADV_NOREUSE
815 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
816 POSIX_FADV_DONTNEED);
821 * File table vnode write routine.
824 vn_write(fp, uio, active_cred, flags, td)
827 struct ucred *active_cred;
834 int error, ioflag, lock_flags;
837 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
839 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
841 if (vp->v_type == VREG)
844 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
846 if (fp->f_flag & FNONBLOCK)
848 if (fp->f_flag & O_DIRECT)
850 if ((fp->f_flag & O_FSYNC) ||
851 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
854 if (vp->v_type != VCHR &&
855 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
858 advice = get_advice(fp, uio);
860 if (MNT_SHARED_WRITES(mp) ||
861 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
862 lock_flags = LK_SHARED;
864 lock_flags = LK_EXCLUSIVE;
867 vn_lock(vp, lock_flags | LK_RETRY);
869 case POSIX_FADV_NORMAL:
870 case POSIX_FADV_SEQUENTIAL:
871 case POSIX_FADV_NOREUSE:
872 ioflag |= sequential_heuristic(uio, fp);
874 case POSIX_FADV_RANDOM:
875 /* XXX: Is this correct? */
878 orig_offset = uio->uio_offset;
881 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
884 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
885 fp->f_nextoff = uio->uio_offset;
887 if (vp->v_type != VCHR)
888 vn_finished_write(mp);
889 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
890 orig_offset != uio->uio_offset)
892 * Use POSIX_FADV_DONTNEED to flush pages and buffers
893 * for the backing file after a POSIX_FADV_NOREUSE
896 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
897 POSIX_FADV_DONTNEED);
903 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
904 * prevent the following deadlock:
906 * Assume that the thread A reads from the vnode vp1 into userspace
907 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
908 * currently not resident, then system ends up with the call chain
909 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
910 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
911 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
912 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
913 * backed by the pages of vnode vp1, and some page in buf2 is not
914 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
916 * To prevent the lock order reversal and deadlock, vn_io_fault() does
917 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
918 * Instead, it first tries to do the whole range i/o with pagefaults
919 * disabled. If all pages in the i/o buffer are resident and mapped,
920 * VOP will succeed (ignoring the genuine filesystem errors).
921 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
922 * i/o in chunks, with all pages in the chunk prefaulted and held
923 * using vm_fault_quick_hold_pages().
925 * Filesystems using this deadlock avoidance scheme should use the
926 * array of the held pages from uio, saved in the curthread->td_ma,
927 * instead of doing uiomove(). A helper function
928 * vn_io_fault_uiomove() converts uiomove request into
929 * uiomove_fromphys() over td_ma array.
931 * Since vnode locks do not cover the whole i/o anymore, rangelocks
932 * make the current i/o request atomic with respect to other i/os and
937 * Decode vn_io_fault_args and perform the corresponding i/o.
940 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
944 switch (args->kind) {
945 case VN_IO_FAULT_FOP:
946 return ((args->args.fop_args.doio)(args->args.fop_args.fp,
947 uio, args->cred, args->flags, td));
948 case VN_IO_FAULT_VOP:
949 if (uio->uio_rw == UIO_READ) {
950 return (VOP_READ(args->args.vop_args.vp, uio,
951 args->flags, args->cred));
952 } else if (uio->uio_rw == UIO_WRITE) {
953 return (VOP_WRITE(args->args.vop_args.vp, uio,
954 args->flags, args->cred));
958 panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
963 vn_io_fault_touch(char *base, const struct uio *uio)
968 if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1))
974 vn_io_fault_prefault_user(const struct uio *uio)
977 const struct iovec *iov;
982 KASSERT(uio->uio_segflg == UIO_USERSPACE,
983 ("vn_io_fault_prefault userspace"));
987 resid = uio->uio_resid;
988 base = iov->iov_base;
991 error = vn_io_fault_touch(base, uio);
994 if (len < PAGE_SIZE) {
996 error = vn_io_fault_touch(base + len - 1, uio);
1001 if (++i >= uio->uio_iovcnt)
1003 iov = uio->uio_iov + i;
1004 base = iov->iov_base;
1016 * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1017 * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1018 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1019 * into args and call vn_io_fault1() to handle faults during the user
1020 * mode buffer accesses.
1023 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1026 vm_page_t ma[io_hold_cnt + 2];
1027 struct uio *uio_clone, short_uio;
1028 struct iovec short_iovec[1];
1029 vm_page_t *prev_td_ma;
1031 vm_offset_t addr, end;
1034 int error, cnt, save, saveheld, prev_td_ma_cnt;
1036 if (vn_io_fault_prefault) {
1037 error = vn_io_fault_prefault_user(uio);
1039 return (error); /* Or ignore ? */
1042 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1045 * The UFS follows IO_UNIT directive and replays back both
1046 * uio_offset and uio_resid if an error is encountered during the
1047 * operation. But, since the iovec may be already advanced,
1048 * uio is still in an inconsistent state.
1050 * Cache a copy of the original uio, which is advanced to the redo
1051 * point using UIO_NOCOPY below.
1053 uio_clone = cloneuio(uio);
1054 resid = uio->uio_resid;
1056 short_uio.uio_segflg = UIO_USERSPACE;
1057 short_uio.uio_rw = uio->uio_rw;
1058 short_uio.uio_td = uio->uio_td;
1060 save = vm_fault_disable_pagefaults();
1061 error = vn_io_fault_doio(args, uio, td);
1062 if (error != EFAULT)
1065 atomic_add_long(&vn_io_faults_cnt, 1);
1066 uio_clone->uio_segflg = UIO_NOCOPY;
1067 uiomove(NULL, resid - uio->uio_resid, uio_clone);
1068 uio_clone->uio_segflg = uio->uio_segflg;
1070 saveheld = curthread_pflags_set(TDP_UIOHELD);
1071 prev_td_ma = td->td_ma;
1072 prev_td_ma_cnt = td->td_ma_cnt;
1074 while (uio_clone->uio_resid != 0) {
1075 len = uio_clone->uio_iov->iov_len;
1077 KASSERT(uio_clone->uio_iovcnt >= 1,
1078 ("iovcnt underflow"));
1079 uio_clone->uio_iov++;
1080 uio_clone->uio_iovcnt--;
1083 if (len > io_hold_cnt * PAGE_SIZE)
1084 len = io_hold_cnt * PAGE_SIZE;
1085 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1086 end = round_page(addr + len);
1091 cnt = atop(end - trunc_page(addr));
1093 * A perfectly misaligned address and length could cause
1094 * both the start and the end of the chunk to use partial
1095 * page. +2 accounts for such a situation.
1097 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1098 addr, len, prot, ma, io_hold_cnt + 2);
1103 short_uio.uio_iov = &short_iovec[0];
1104 short_iovec[0].iov_base = (void *)addr;
1105 short_uio.uio_iovcnt = 1;
1106 short_uio.uio_resid = short_iovec[0].iov_len = len;
1107 short_uio.uio_offset = uio_clone->uio_offset;
1109 td->td_ma_cnt = cnt;
1111 error = vn_io_fault_doio(args, &short_uio, td);
1112 vm_page_unhold_pages(ma, cnt);
1113 adv = len - short_uio.uio_resid;
1115 uio_clone->uio_iov->iov_base =
1116 (char *)uio_clone->uio_iov->iov_base + adv;
1117 uio_clone->uio_iov->iov_len -= adv;
1118 uio_clone->uio_resid -= adv;
1119 uio_clone->uio_offset += adv;
1121 uio->uio_resid -= adv;
1122 uio->uio_offset += adv;
1124 if (error != 0 || adv == 0)
1127 td->td_ma = prev_td_ma;
1128 td->td_ma_cnt = prev_td_ma_cnt;
1129 curthread_pflags_restore(saveheld);
1131 vm_fault_enable_pagefaults(save);
1132 free(uio_clone, M_IOV);
1137 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1138 int flags, struct thread *td)
1143 struct vn_io_fault_args args;
1146 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1148 foffset_lock_uio(fp, uio, flags);
1149 if (do_vn_io_fault(vp, uio)) {
1150 args.kind = VN_IO_FAULT_FOP;
1151 args.args.fop_args.fp = fp;
1152 args.args.fop_args.doio = doio;
1153 args.cred = active_cred;
1154 args.flags = flags | FOF_OFFSET;
1155 if (uio->uio_rw == UIO_READ) {
1156 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1157 uio->uio_offset + uio->uio_resid);
1158 } else if ((fp->f_flag & O_APPEND) != 0 ||
1159 (flags & FOF_OFFSET) == 0) {
1160 /* For appenders, punt and lock the whole range. */
1161 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1163 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1164 uio->uio_offset + uio->uio_resid);
1166 error = vn_io_fault1(vp, uio, &args, td);
1167 vn_rangelock_unlock(vp, rl_cookie);
1169 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1171 foffset_unlock_uio(fp, uio, flags);
1176 * Helper function to perform the requested uiomove operation using
1177 * the held pages for io->uio_iov[0].iov_base buffer instead of
1178 * copyin/copyout. Access to the pages with uiomove_fromphys()
1179 * instead of iov_base prevents page faults that could occur due to
1180 * pmap_collect() invalidating the mapping created by
1181 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1182 * object cleanup revoking the write access from page mappings.
1184 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1185 * instead of plain uiomove().
1188 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1190 struct uio transp_uio;
1191 struct iovec transp_iov[1];
1197 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1198 uio->uio_segflg != UIO_USERSPACE)
1199 return (uiomove(data, xfersize, uio));
1201 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1202 transp_iov[0].iov_base = data;
1203 transp_uio.uio_iov = &transp_iov[0];
1204 transp_uio.uio_iovcnt = 1;
1205 if (xfersize > uio->uio_resid)
1206 xfersize = uio->uio_resid;
1207 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1208 transp_uio.uio_offset = 0;
1209 transp_uio.uio_segflg = UIO_SYSSPACE;
1211 * Since transp_iov points to data, and td_ma page array
1212 * corresponds to original uio->uio_iov, we need to invert the
1213 * direction of the i/o operation as passed to
1214 * uiomove_fromphys().
1216 switch (uio->uio_rw) {
1218 transp_uio.uio_rw = UIO_READ;
1221 transp_uio.uio_rw = UIO_WRITE;
1224 transp_uio.uio_td = uio->uio_td;
1225 error = uiomove_fromphys(td->td_ma,
1226 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1227 xfersize, &transp_uio);
1228 adv = xfersize - transp_uio.uio_resid;
1230 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1231 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1233 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1235 td->td_ma_cnt -= pgadv;
1236 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1237 uio->uio_iov->iov_len -= adv;
1238 uio->uio_resid -= adv;
1239 uio->uio_offset += adv;
1244 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1248 vm_offset_t iov_base;
1252 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1253 uio->uio_segflg != UIO_USERSPACE)
1254 return (uiomove_fromphys(ma, offset, xfersize, uio));
1256 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1257 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1258 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1259 switch (uio->uio_rw) {
1261 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1265 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1269 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1271 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1273 td->td_ma_cnt -= pgadv;
1274 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1275 uio->uio_iov->iov_len -= cnt;
1276 uio->uio_resid -= cnt;
1277 uio->uio_offset += cnt;
1283 * File table truncate routine.
1286 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1298 * Lock the whole range for truncation. Otherwise split i/o
1299 * might happen partly before and partly after the truncation.
1301 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1302 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1305 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1306 if (vp->v_type == VDIR) {
1311 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1315 error = vn_writechk(vp);
1318 vattr.va_size = length;
1319 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1323 vn_finished_write(mp);
1325 vn_rangelock_unlock(vp, rl_cookie);
1330 * File table vnode stat routine.
1333 vn_statfile(fp, sb, active_cred, td)
1336 struct ucred *active_cred;
1339 struct vnode *vp = fp->f_vnode;
1342 vn_lock(vp, LK_SHARED | LK_RETRY);
1343 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1350 * Stat a vnode; implementation for the stat syscall
1353 vn_stat(vp, sb, active_cred, file_cred, td)
1355 register struct stat *sb;
1356 struct ucred *active_cred;
1357 struct ucred *file_cred;
1361 register struct vattr *vap;
1366 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1374 * Initialize defaults for new and unusual fields, so that file
1375 * systems which don't support these fields don't need to know
1378 vap->va_birthtime.tv_sec = -1;
1379 vap->va_birthtime.tv_nsec = 0;
1380 vap->va_fsid = VNOVAL;
1381 vap->va_rdev = NODEV;
1383 error = VOP_GETATTR(vp, vap, active_cred);
1388 * Zero the spare stat fields
1390 bzero(sb, sizeof *sb);
1393 * Copy from vattr table
1395 if (vap->va_fsid != VNOVAL)
1396 sb->st_dev = vap->va_fsid;
1398 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1399 sb->st_ino = vap->va_fileid;
1400 mode = vap->va_mode;
1401 switch (vap->va_type) {
1427 sb->st_nlink = vap->va_nlink;
1428 sb->st_uid = vap->va_uid;
1429 sb->st_gid = vap->va_gid;
1430 sb->st_rdev = vap->va_rdev;
1431 if (vap->va_size > OFF_MAX)
1433 sb->st_size = vap->va_size;
1434 sb->st_atim = vap->va_atime;
1435 sb->st_mtim = vap->va_mtime;
1436 sb->st_ctim = vap->va_ctime;
1437 sb->st_birthtim = vap->va_birthtime;
1440 * According to www.opengroup.org, the meaning of st_blksize is
1441 * "a filesystem-specific preferred I/O block size for this
1442 * object. In some filesystem types, this may vary from file
1444 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1447 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1449 sb->st_flags = vap->va_flags;
1450 if (priv_check(td, PRIV_VFS_GENERATION))
1453 sb->st_gen = vap->va_gen;
1455 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1460 * File table vnode ioctl routine.
1463 vn_ioctl(fp, com, data, active_cred, td)
1467 struct ucred *active_cred;
1475 switch (vp->v_type) {
1480 vn_lock(vp, LK_SHARED | LK_RETRY);
1481 error = VOP_GETATTR(vp, &vattr, active_cred);
1484 *(int *)data = vattr.va_size - fp->f_offset;
1490 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1499 * File table vnode poll routine.
1502 vn_poll(fp, events, active_cred, td)
1505 struct ucred *active_cred;
1513 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1514 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1519 error = VOP_POLL(vp, events, fp->f_cred, td);
1524 * Acquire the requested lock and then check for validity. LK_RETRY
1525 * permits vn_lock to return doomed vnodes.
1528 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1532 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1533 ("vn_lock called with no locktype."));
1535 #ifdef DEBUG_VFS_LOCKS
1536 KASSERT(vp->v_holdcnt != 0,
1537 ("vn_lock %p: zero hold count", vp));
1539 error = VOP_LOCK1(vp, flags, file, line);
1540 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1541 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1542 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1545 * Callers specify LK_RETRY if they wish to get dead vnodes.
1546 * If RETRY is not set, we return ENOENT instead.
1548 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1549 (flags & LK_RETRY) == 0) {
1554 } while (flags & LK_RETRY && error != 0);
1559 * File table vnode close routine.
1562 vn_closefile(fp, td)
1571 fp->f_ops = &badfileops;
1573 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1576 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1578 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1579 lf.l_whence = SEEK_SET;
1582 lf.l_type = F_UNLCK;
1583 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1590 vn_suspendable(struct mount *mp)
1593 return (mp->mnt_op->vfs_susp_clean != NULL);
1597 * Preparing to start a filesystem write operation. If the operation is
1598 * permitted, then we bump the count of operations in progress and
1599 * proceed. If a suspend request is in progress, we wait until the
1600 * suspension is over, and then proceed.
1603 vn_start_write_locked(struct mount *mp, int flags)
1607 mtx_assert(MNT_MTX(mp), MA_OWNED);
1611 * Check on status of suspension.
1613 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1614 mp->mnt_susp_owner != curthread) {
1615 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
1616 (flags & PCATCH) : 0) | (PUSER - 1);
1617 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1618 if (flags & V_NOWAIT) {
1619 error = EWOULDBLOCK;
1622 error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
1628 if (flags & V_XSLEEP)
1630 mp->mnt_writeopcount++;
1632 if (error != 0 || (flags & V_XSLEEP) != 0)
1639 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
1644 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1645 ("V_MNTREF requires mp"));
1649 * If a vnode is provided, get and return the mount point that
1650 * to which it will write.
1653 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1655 if (error != EOPNOTSUPP)
1660 if ((mp = *mpp) == NULL)
1663 if (!vn_suspendable(mp)) {
1664 if (vp != NULL || (flags & V_MNTREF) != 0)
1670 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1672 * As long as a vnode is not provided we need to acquire a
1673 * refcount for the provided mountpoint too, in order to
1674 * emulate a vfs_ref().
1677 if (vp == NULL && (flags & V_MNTREF) == 0)
1680 return (vn_start_write_locked(mp, flags));
1684 * Secondary suspension. Used by operations such as vop_inactive
1685 * routines that are needed by the higher level functions. These
1686 * are allowed to proceed until all the higher level functions have
1687 * completed (indicated by mnt_writeopcount dropping to zero). At that
1688 * time, these operations are halted until the suspension is over.
1691 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags)
1696 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1697 ("V_MNTREF requires mp"));
1701 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1703 if (error != EOPNOTSUPP)
1709 * If we are not suspended or have not yet reached suspended
1710 * mode, then let the operation proceed.
1712 if ((mp = *mpp) == NULL)
1715 if (!vn_suspendable(mp)) {
1716 if (vp != NULL || (flags & V_MNTREF) != 0)
1722 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1724 * As long as a vnode is not provided we need to acquire a
1725 * refcount for the provided mountpoint too, in order to
1726 * emulate a vfs_ref().
1729 if (vp == NULL && (flags & V_MNTREF) == 0)
1731 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1732 mp->mnt_secondary_writes++;
1733 mp->mnt_secondary_accwrites++;
1737 if (flags & V_NOWAIT) {
1740 return (EWOULDBLOCK);
1743 * Wait for the suspension to finish.
1745 error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
1746 ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
1755 * Filesystem write operation has completed. If we are suspending and this
1756 * operation is the last one, notify the suspender that the suspension is
1760 vn_finished_write(mp)
1763 if (mp == NULL || !vn_suspendable(mp))
1767 mp->mnt_writeopcount--;
1768 if (mp->mnt_writeopcount < 0)
1769 panic("vn_finished_write: neg cnt");
1770 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1771 mp->mnt_writeopcount <= 0)
1772 wakeup(&mp->mnt_writeopcount);
1778 * Filesystem secondary write operation has completed. If we are
1779 * suspending and this operation is the last one, notify the suspender
1780 * that the suspension is now in effect.
1783 vn_finished_secondary_write(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_vnode_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_fileid = va.va_fileid;
2373 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2374 kif->kf_un.kf_file.kf_file_size = va.va_size;
2375 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2380 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
2381 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
2385 struct pmckern_map_in pkm;
2391 boolean_t writecounted;
2394 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
2395 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
2397 * POSIX shared-memory objects are defined to have
2398 * kernel persistence, and are not defined to support
2399 * read(2)/write(2) -- or even open(2). Thus, we can
2400 * use MAP_ASYNC to trade on-disk coherence for speed.
2401 * The shm_open(3) library routine turns on the FPOSIXSHM
2402 * flag to request this behavior.
2404 if ((fp->f_flag & FPOSIXSHM) != 0)
2405 flags |= MAP_NOSYNC;
2410 * Ensure that file and memory protections are
2411 * compatible. Note that we only worry about
2412 * writability if mapping is shared; in this case,
2413 * current and max prot are dictated by the open file.
2414 * XXX use the vnode instead? Problem is: what
2415 * credentials do we use for determination? What if
2416 * proc does a setuid?
2419 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0)
2420 maxprot = VM_PROT_NONE;
2422 maxprot = VM_PROT_EXECUTE;
2423 if ((fp->f_flag & FREAD) != 0)
2424 maxprot |= VM_PROT_READ;
2425 else if ((prot & VM_PROT_READ) != 0)
2429 * If we are sharing potential changes via MAP_SHARED and we
2430 * are trying to get write permission although we opened it
2431 * without asking for it, bail out.
2433 if ((flags & MAP_SHARED) != 0) {
2434 if ((fp->f_flag & FWRITE) != 0)
2435 maxprot |= VM_PROT_WRITE;
2436 else if ((prot & VM_PROT_WRITE) != 0)
2439 maxprot |= VM_PROT_WRITE;
2440 cap_maxprot |= VM_PROT_WRITE;
2442 maxprot &= cap_maxprot;
2444 writecounted = FALSE;
2445 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp,
2446 &foff, &object, &writecounted);
2449 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2450 foff, writecounted, td);
2453 * If this mapping was accounted for in the vnode's
2454 * writecount, then undo that now.
2457 vnode_pager_release_writecount(object, 0, size);
2458 vm_object_deallocate(object);
2461 /* Inform hwpmc(4) if an executable is being mapped. */
2462 if (error == 0 && (prot & VM_PROT_EXECUTE) != 0) {
2464 pkm.pm_address = (uintptr_t) addr;
2465 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);