2 * Copyright (c) 1982, 1986, 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Copyright (c) 2012 Konstantin Belousov <kib@FreeBSD.org>
11 * Copyright (c) 2013, 2014 The FreeBSD Foundation
13 * Portions of this software were developed by Konstantin Belousov
14 * under sponsorship from the FreeBSD Foundation.
16 * Redistribution and use in source and binary forms, with or without
17 * modification, are permitted provided that the following conditions
19 * 1. Redistributions of source code must retain the above copyright
20 * notice, this list of conditions and the following disclaimer.
21 * 2. Redistributions in binary form must reproduce the above copyright
22 * notice, this list of conditions and the following disclaimer in the
23 * documentation and/or other materials provided with the distribution.
24 * 3. Neither the name of the University nor the names of its contributors
25 * may be used to endorse or promote products derived from this software
26 * without specific prior written permission.
28 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
29 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
30 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
31 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
32 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
33 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
34 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
35 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
37 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
40 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94
43 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
46 #include "opt_hwpmc_hooks.h"
48 #include <sys/param.h>
49 #include <sys/systm.h>
52 #include <sys/fcntl.h>
58 #include <sys/limits.h>
61 #include <sys/mount.h>
62 #include <sys/mutex.h>
63 #include <sys/namei.h>
64 #include <sys/vnode.h>
67 #include <sys/filio.h>
68 #include <sys/resourcevar.h>
69 #include <sys/rwlock.h>
71 #include <sys/sysctl.h>
72 #include <sys/ttycom.h>
74 #include <sys/syslog.h>
75 #include <sys/unistd.h>
78 #include <security/audit/audit.h>
79 #include <security/mac/mac_framework.h>
82 #include <vm/vm_extern.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_page.h>
87 #include <vm/vnode_pager.h>
90 #include <sys/pmckern.h>
93 static fo_rdwr_t vn_read;
94 static fo_rdwr_t vn_write;
95 static fo_rdwr_t vn_io_fault;
96 static fo_truncate_t vn_truncate;
97 static fo_ioctl_t vn_ioctl;
98 static fo_poll_t vn_poll;
99 static fo_kqfilter_t vn_kqfilter;
100 static fo_stat_t vn_statfile;
101 static fo_close_t vn_closefile;
102 static fo_mmap_t vn_mmap;
104 struct fileops vnops = {
105 .fo_read = vn_io_fault,
106 .fo_write = vn_io_fault,
107 .fo_truncate = vn_truncate,
108 .fo_ioctl = vn_ioctl,
110 .fo_kqfilter = vn_kqfilter,
111 .fo_stat = vn_statfile,
112 .fo_close = vn_closefile,
113 .fo_chmod = vn_chmod,
114 .fo_chown = vn_chown,
115 .fo_sendfile = vn_sendfile,
117 .fo_fill_kinfo = vn_fill_kinfo,
119 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
122 static const int io_hold_cnt = 16;
123 static int vn_io_fault_enable = 1;
124 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
125 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
126 static int vn_io_fault_prefault = 0;
127 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_prefault, CTLFLAG_RW,
128 &vn_io_fault_prefault, 0, "Enable vn_io_fault prefaulting");
129 static u_long vn_io_faults_cnt;
130 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
131 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
134 * Returns true if vn_io_fault mode of handling the i/o request should
138 do_vn_io_fault(struct vnode *vp, struct uio *uio)
142 return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG &&
143 (mp = vp->v_mount) != NULL &&
144 (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable);
148 * Structure used to pass arguments to vn_io_fault1(), to do either
149 * file- or vnode-based I/O calls.
151 struct vn_io_fault_args {
159 struct fop_args_tag {
163 struct vop_args_tag {
169 static int vn_io_fault1(struct vnode *vp, struct uio *uio,
170 struct vn_io_fault_args *args, struct thread *td);
173 vn_open(ndp, flagp, cmode, fp)
174 struct nameidata *ndp;
178 struct thread *td = ndp->ni_cnd.cn_thread;
180 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
184 * Common code for vnode open operations via a name lookup.
185 * Lookup the vnode and invoke VOP_CREATE if needed.
186 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
188 * Note that this does NOT free nameidata for the successful case,
189 * due to the NDINIT being done elsewhere.
192 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
193 struct ucred *cred, struct file *fp)
197 struct thread *td = ndp->ni_cnd.cn_thread;
199 struct vattr *vap = &vat;
204 if ((fmode & (O_CREAT | O_EXCL | O_DIRECTORY)) == (O_CREAT |
205 O_EXCL | O_DIRECTORY))
207 else if ((fmode & (O_CREAT | O_DIRECTORY)) == O_CREAT) {
208 ndp->ni_cnd.cn_nameiop = CREATE;
210 * Set NOCACHE to avoid flushing the cache when
211 * rolling in many files at once.
213 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE;
214 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
215 ndp->ni_cnd.cn_flags |= FOLLOW;
216 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
217 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
218 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
219 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
221 if ((error = namei(ndp)) != 0)
223 if (ndp->ni_vp == NULL) {
226 vap->va_mode = cmode;
228 vap->va_vaflags |= VA_EXCLUSIVE;
229 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
230 NDFREE(ndp, NDF_ONLY_PNBUF);
232 if ((error = vn_start_write(NULL, &mp,
233 V_XSLEEP | PCATCH)) != 0)
237 if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0)
238 ndp->ni_cnd.cn_flags |= MAKEENTRY;
240 error = mac_vnode_check_create(cred, ndp->ni_dvp,
244 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
247 vn_finished_write(mp);
249 NDFREE(ndp, NDF_ONLY_PNBUF);
255 if (ndp->ni_dvp == ndp->ni_vp)
261 if (fmode & O_EXCL) {
268 ndp->ni_cnd.cn_nameiop = LOOKUP;
269 ndp->ni_cnd.cn_flags = ISOPEN |
270 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF;
271 if (!(fmode & FWRITE))
272 ndp->ni_cnd.cn_flags |= LOCKSHARED;
273 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
274 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
275 if (vn_open_flags & VN_OPEN_NOCAPCHECK)
276 ndp->ni_cnd.cn_flags |= NOCAPCHECK;
277 if ((error = namei(ndp)) != 0)
281 error = vn_open_vnode(vp, fmode, cred, td, fp);
287 NDFREE(ndp, NDF_ONLY_PNBUF);
295 * Common code for vnode open operations once a vnode is located.
296 * Check permissions, and call the VOP_OPEN routine.
299 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
300 struct thread *td, struct file *fp)
304 int error, lock_flags, type;
306 if (vp->v_type == VLNK)
308 if (vp->v_type == VSOCK)
310 if (vp->v_type != VDIR && fmode & O_DIRECTORY)
313 if (fmode & (FWRITE | O_TRUNC)) {
314 if (vp->v_type == VDIR)
322 if ((fmode & O_APPEND) && (fmode & FWRITE))
327 if (fmode & O_VERIFY)
329 error = mac_vnode_check_open(cred, vp, accmode);
333 accmode &= ~(VCREAT | VVERIFY);
335 if ((fmode & O_CREAT) == 0) {
336 if (accmode & VWRITE) {
337 error = vn_writechk(vp);
342 error = VOP_ACCESS(vp, accmode, cred, td);
347 if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
348 vn_lock(vp, LK_UPGRADE | LK_RETRY);
349 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
352 if (fmode & (O_EXLOCK | O_SHLOCK)) {
353 KASSERT(fp != NULL, ("open with flock requires fp"));
354 lock_flags = VOP_ISLOCKED(vp);
356 lf.l_whence = SEEK_SET;
359 if (fmode & O_EXLOCK)
364 if ((fmode & FNONBLOCK) == 0)
366 error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type);
368 fp->f_flag |= FHASLOCK;
369 vn_lock(vp, lock_flags | LK_RETRY);
370 if (error == 0 && vp->v_iflag & VI_DOOMED)
374 * Another thread might have used this vnode as an
375 * executable while the vnode lock was dropped.
376 * Ensure the vnode is still able to be opened for
377 * writing after the lock has been obtained.
379 if (error == 0 && accmode & VWRITE)
380 error = vn_writechk(vp);
383 fp->f_flag |= FOPENFAILED;
385 if (fp->f_ops == &badfileops) {
386 fp->f_type = DTYPE_VNODE;
392 if (error == 0 && fmode & FWRITE) {
393 VOP_ADD_WRITECOUNT(vp, 1);
394 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
395 __func__, vp, vp->v_writecount);
397 ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
402 * Check for write permissions on the specified vnode.
403 * Prototype text segments cannot be written.
407 register struct vnode *vp;
410 ASSERT_VOP_LOCKED(vp, "vn_writechk");
412 * If there's shared text associated with
413 * the vnode, try to free it up once. If
414 * we fail, we can't allow writing.
426 vn_close(vp, flags, file_cred, td)
427 register struct vnode *vp;
429 struct ucred *file_cred;
433 int error, lock_flags;
435 if (vp->v_type != VFIFO && (flags & FWRITE) == 0 &&
436 MNT_EXTENDED_SHARED(vp->v_mount))
437 lock_flags = LK_SHARED;
439 lock_flags = LK_EXCLUSIVE;
441 vn_start_write(vp, &mp, V_WAIT);
442 vn_lock(vp, lock_flags | LK_RETRY);
443 AUDIT_ARG_VNODE1(vp);
444 if ((flags & (FWRITE | FOPENFAILED)) == FWRITE) {
445 VNASSERT(vp->v_writecount > 0, vp,
446 ("vn_close: negative writecount"));
447 VOP_ADD_WRITECOUNT(vp, -1);
448 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
449 __func__, vp, vp->v_writecount);
451 error = VOP_CLOSE(vp, flags, file_cred, td);
453 vn_finished_write(mp);
458 * Heuristic to detect sequential operation.
461 sequential_heuristic(struct uio *uio, struct file *fp)
464 ASSERT_VOP_LOCKED(fp->f_vnode, __func__);
465 if (fp->f_flag & FRDAHEAD)
466 return (fp->f_seqcount << IO_SEQSHIFT);
469 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
470 * that the first I/O is normally considered to be slightly
471 * sequential. Seeking to offset 0 doesn't change sequentiality
472 * unless previous seeks have reduced f_seqcount to 0, in which
473 * case offset 0 is not special.
475 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
476 uio->uio_offset == fp->f_nextoff) {
478 * f_seqcount is in units of fixed-size blocks so that it
479 * depends mainly on the amount of sequential I/O and not
480 * much on the number of sequential I/O's. The fixed size
481 * of 16384 is hard-coded here since it is (not quite) just
482 * a magic size that works well here. This size is more
483 * closely related to the best I/O size for real disks than
484 * to any block size used by software.
486 fp->f_seqcount += howmany(uio->uio_resid, 16384);
487 if (fp->f_seqcount > IO_SEQMAX)
488 fp->f_seqcount = IO_SEQMAX;
489 return (fp->f_seqcount << IO_SEQSHIFT);
492 /* Not sequential. Quickly draw-down sequentiality. */
493 if (fp->f_seqcount > 1)
501 * Package up an I/O request on a vnode into a uio and do it.
504 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
505 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
506 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
513 struct vn_io_fault_args args;
514 int error, lock_flags;
516 auio.uio_iov = &aiov;
518 aiov.iov_base = base;
520 auio.uio_resid = len;
521 auio.uio_offset = offset;
522 auio.uio_segflg = segflg;
527 if ((ioflg & IO_NODELOCKED) == 0) {
528 if ((ioflg & IO_RANGELOCKED) == 0) {
529 if (rw == UIO_READ) {
530 rl_cookie = vn_rangelock_rlock(vp, offset,
533 rl_cookie = vn_rangelock_wlock(vp, offset,
539 if (rw == UIO_WRITE) {
540 if (vp->v_type != VCHR &&
541 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
544 if (MNT_SHARED_WRITES(mp) ||
545 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
546 lock_flags = LK_SHARED;
548 lock_flags = LK_EXCLUSIVE;
550 lock_flags = LK_SHARED;
551 vn_lock(vp, lock_flags | LK_RETRY);
555 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
557 if ((ioflg & IO_NOMACCHECK) == 0) {
559 error = mac_vnode_check_read(active_cred, file_cred,
562 error = mac_vnode_check_write(active_cred, file_cred,
567 if (file_cred != NULL)
571 if (do_vn_io_fault(vp, &auio)) {
572 args.kind = VN_IO_FAULT_VOP;
575 args.args.vop_args.vp = vp;
576 error = vn_io_fault1(vp, &auio, &args, td);
577 } else if (rw == UIO_READ) {
578 error = VOP_READ(vp, &auio, ioflg, cred);
579 } else /* if (rw == UIO_WRITE) */ {
580 error = VOP_WRITE(vp, &auio, ioflg, cred);
584 *aresid = auio.uio_resid;
586 if (auio.uio_resid && error == 0)
588 if ((ioflg & IO_NODELOCKED) == 0) {
591 vn_finished_write(mp);
594 if (rl_cookie != NULL)
595 vn_rangelock_unlock(vp, rl_cookie);
600 * Package up an I/O request on a vnode into a uio and do it. The I/O
601 * request is split up into smaller chunks and we try to avoid saturating
602 * the buffer cache while potentially holding a vnode locked, so we
603 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
604 * to give other processes a chance to lock the vnode (either other processes
605 * core'ing the same binary, or unrelated processes scanning the directory).
608 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
609 file_cred, aresid, td)
617 struct ucred *active_cred;
618 struct ucred *file_cred;
629 * Force `offset' to a multiple of MAXBSIZE except possibly
630 * for the first chunk, so that filesystems only need to
631 * write full blocks except possibly for the first and last
634 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
638 if (rw != UIO_READ && vp->v_type == VREG)
641 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
642 ioflg, active_cred, file_cred, &iaresid, td);
643 len -= chunk; /* aresid calc already includes length */
647 base = (char *)base + chunk;
648 kern_yield(PRI_USER);
651 *aresid = len + iaresid;
656 foffset_lock(struct file *fp, int flags)
661 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
663 #if OFF_MAX <= LONG_MAX
665 * Caller only wants the current f_offset value. Assume that
666 * the long and shorter integer types reads are atomic.
668 if ((flags & FOF_NOLOCK) != 0)
669 return (fp->f_offset);
673 * According to McKusick the vn lock was protecting f_offset here.
674 * It is now protected by the FOFFSET_LOCKED flag.
676 mtxp = mtx_pool_find(mtxpool_sleep, fp);
678 if ((flags & FOF_NOLOCK) == 0) {
679 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
680 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
681 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
684 fp->f_vnread_flags |= FOFFSET_LOCKED;
692 foffset_unlock(struct file *fp, off_t val, int flags)
696 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
698 #if OFF_MAX <= LONG_MAX
699 if ((flags & FOF_NOLOCK) != 0) {
700 if ((flags & FOF_NOUPDATE) == 0)
702 if ((flags & FOF_NEXTOFF) != 0)
708 mtxp = mtx_pool_find(mtxpool_sleep, fp);
710 if ((flags & FOF_NOUPDATE) == 0)
712 if ((flags & FOF_NEXTOFF) != 0)
714 if ((flags & FOF_NOLOCK) == 0) {
715 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
716 ("Lost FOFFSET_LOCKED"));
717 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
718 wakeup(&fp->f_vnread_flags);
719 fp->f_vnread_flags = 0;
725 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
728 if ((flags & FOF_OFFSET) == 0)
729 uio->uio_offset = foffset_lock(fp, flags);
733 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
736 if ((flags & FOF_OFFSET) == 0)
737 foffset_unlock(fp, uio->uio_offset, flags);
741 get_advice(struct file *fp, struct uio *uio)
746 ret = POSIX_FADV_NORMAL;
747 if (fp->f_advice == NULL || fp->f_vnode->v_type != VREG)
750 mtxp = mtx_pool_find(mtxpool_sleep, fp);
752 if (fp->f_advice != NULL &&
753 uio->uio_offset >= fp->f_advice->fa_start &&
754 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
755 ret = fp->f_advice->fa_advice;
761 * File table vnode read routine.
764 vn_read(fp, uio, active_cred, flags, td)
767 struct ucred *active_cred;
776 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
778 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
781 if (fp->f_flag & FNONBLOCK)
783 if (fp->f_flag & O_DIRECT)
785 advice = get_advice(fp, uio);
786 vn_lock(vp, LK_SHARED | LK_RETRY);
789 case POSIX_FADV_NORMAL:
790 case POSIX_FADV_SEQUENTIAL:
791 case POSIX_FADV_NOREUSE:
792 ioflag |= sequential_heuristic(uio, fp);
794 case POSIX_FADV_RANDOM:
795 /* Disable read-ahead for random I/O. */
798 orig_offset = uio->uio_offset;
801 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
804 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
805 fp->f_nextoff = uio->uio_offset;
807 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
808 orig_offset != uio->uio_offset)
810 * Use POSIX_FADV_DONTNEED to flush pages and buffers
811 * for the backing file after a POSIX_FADV_NOREUSE
814 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
815 POSIX_FADV_DONTNEED);
820 * File table vnode write routine.
823 vn_write(fp, uio, active_cred, flags, td)
826 struct ucred *active_cred;
833 int error, ioflag, lock_flags;
836 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
838 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
840 if (vp->v_type == VREG)
843 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
845 if (fp->f_flag & FNONBLOCK)
847 if (fp->f_flag & O_DIRECT)
849 if ((fp->f_flag & O_FSYNC) ||
850 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
853 if (vp->v_type != VCHR &&
854 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
857 advice = get_advice(fp, uio);
859 if (MNT_SHARED_WRITES(mp) ||
860 (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) {
861 lock_flags = LK_SHARED;
863 lock_flags = LK_EXCLUSIVE;
866 vn_lock(vp, lock_flags | LK_RETRY);
868 case POSIX_FADV_NORMAL:
869 case POSIX_FADV_SEQUENTIAL:
870 case POSIX_FADV_NOREUSE:
871 ioflag |= sequential_heuristic(uio, fp);
873 case POSIX_FADV_RANDOM:
874 /* XXX: Is this correct? */
877 orig_offset = uio->uio_offset;
880 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
883 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
884 fp->f_nextoff = uio->uio_offset;
886 if (vp->v_type != VCHR)
887 vn_finished_write(mp);
888 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
889 orig_offset != uio->uio_offset)
891 * Use POSIX_FADV_DONTNEED to flush pages and buffers
892 * for the backing file after a POSIX_FADV_NOREUSE
895 error = VOP_ADVISE(vp, orig_offset, uio->uio_offset - 1,
896 POSIX_FADV_DONTNEED);
902 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
903 * prevent the following deadlock:
905 * Assume that the thread A reads from the vnode vp1 into userspace
906 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
907 * currently not resident, then system ends up with the call chain
908 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
909 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
910 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
911 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
912 * backed by the pages of vnode vp1, and some page in buf2 is not
913 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
915 * To prevent the lock order reversal and deadlock, vn_io_fault() does
916 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
917 * Instead, it first tries to do the whole range i/o with pagefaults
918 * disabled. If all pages in the i/o buffer are resident and mapped,
919 * VOP will succeed (ignoring the genuine filesystem errors).
920 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
921 * i/o in chunks, with all pages in the chunk prefaulted and held
922 * using vm_fault_quick_hold_pages().
924 * Filesystems using this deadlock avoidance scheme should use the
925 * array of the held pages from uio, saved in the curthread->td_ma,
926 * instead of doing uiomove(). A helper function
927 * vn_io_fault_uiomove() converts uiomove request into
928 * uiomove_fromphys() over td_ma array.
930 * Since vnode locks do not cover the whole i/o anymore, rangelocks
931 * make the current i/o request atomic with respect to other i/os and
936 * Decode vn_io_fault_args and perform the corresponding i/o.
939 vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio,
943 switch (args->kind) {
944 case VN_IO_FAULT_FOP:
945 return ((args->args.fop_args.doio)(args->args.fop_args.fp,
946 uio, args->cred, args->flags, td));
947 case VN_IO_FAULT_VOP:
948 if (uio->uio_rw == UIO_READ) {
949 return (VOP_READ(args->args.vop_args.vp, uio,
950 args->flags, args->cred));
951 } else if (uio->uio_rw == UIO_WRITE) {
952 return (VOP_WRITE(args->args.vop_args.vp, uio,
953 args->flags, args->cred));
957 panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind,
962 vn_io_fault_touch(char *base, const struct uio *uio)
967 if (r == -1 || (uio->uio_rw == UIO_READ && subyte(base, r) == -1))
973 vn_io_fault_prefault_user(const struct uio *uio)
976 const struct iovec *iov;
981 KASSERT(uio->uio_segflg == UIO_USERSPACE,
982 ("vn_io_fault_prefault userspace"));
986 resid = uio->uio_resid;
987 base = iov->iov_base;
990 error = vn_io_fault_touch(base, uio);
993 if (len < PAGE_SIZE) {
995 error = vn_io_fault_touch(base + len - 1, uio);
1000 if (++i >= uio->uio_iovcnt)
1002 iov = uio->uio_iov + i;
1003 base = iov->iov_base;
1015 * Common code for vn_io_fault(), agnostic to the kind of i/o request.
1016 * Uses vn_io_fault_doio() to make the call to an actual i/o function.
1017 * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request
1018 * into args and call vn_io_fault1() to handle faults during the user
1019 * mode buffer accesses.
1022 vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args,
1025 vm_page_t ma[io_hold_cnt + 2];
1026 struct uio *uio_clone, short_uio;
1027 struct iovec short_iovec[1];
1028 vm_page_t *prev_td_ma;
1030 vm_offset_t addr, end;
1033 int error, cnt, save, saveheld, prev_td_ma_cnt;
1035 if (vn_io_fault_prefault) {
1036 error = vn_io_fault_prefault_user(uio);
1038 return (error); /* Or ignore ? */
1041 prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ;
1044 * The UFS follows IO_UNIT directive and replays back both
1045 * uio_offset and uio_resid if an error is encountered during the
1046 * operation. But, since the iovec may be already advanced,
1047 * uio is still in an inconsistent state.
1049 * Cache a copy of the original uio, which is advanced to the redo
1050 * point using UIO_NOCOPY below.
1052 uio_clone = cloneuio(uio);
1053 resid = uio->uio_resid;
1055 short_uio.uio_segflg = UIO_USERSPACE;
1056 short_uio.uio_rw = uio->uio_rw;
1057 short_uio.uio_td = uio->uio_td;
1059 save = vm_fault_disable_pagefaults();
1060 error = vn_io_fault_doio(args, uio, td);
1061 if (error != EFAULT)
1064 atomic_add_long(&vn_io_faults_cnt, 1);
1065 uio_clone->uio_segflg = UIO_NOCOPY;
1066 uiomove(NULL, resid - uio->uio_resid, uio_clone);
1067 uio_clone->uio_segflg = uio->uio_segflg;
1069 saveheld = curthread_pflags_set(TDP_UIOHELD);
1070 prev_td_ma = td->td_ma;
1071 prev_td_ma_cnt = td->td_ma_cnt;
1073 while (uio_clone->uio_resid != 0) {
1074 len = uio_clone->uio_iov->iov_len;
1076 KASSERT(uio_clone->uio_iovcnt >= 1,
1077 ("iovcnt underflow"));
1078 uio_clone->uio_iov++;
1079 uio_clone->uio_iovcnt--;
1082 if (len > io_hold_cnt * PAGE_SIZE)
1083 len = io_hold_cnt * PAGE_SIZE;
1084 addr = (uintptr_t)uio_clone->uio_iov->iov_base;
1085 end = round_page(addr + len);
1090 cnt = atop(end - trunc_page(addr));
1092 * A perfectly misaligned address and length could cause
1093 * both the start and the end of the chunk to use partial
1094 * page. +2 accounts for such a situation.
1096 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
1097 addr, len, prot, ma, io_hold_cnt + 2);
1102 short_uio.uio_iov = &short_iovec[0];
1103 short_iovec[0].iov_base = (void *)addr;
1104 short_uio.uio_iovcnt = 1;
1105 short_uio.uio_resid = short_iovec[0].iov_len = len;
1106 short_uio.uio_offset = uio_clone->uio_offset;
1108 td->td_ma_cnt = cnt;
1110 error = vn_io_fault_doio(args, &short_uio, td);
1111 vm_page_unhold_pages(ma, cnt);
1112 adv = len - short_uio.uio_resid;
1114 uio_clone->uio_iov->iov_base =
1115 (char *)uio_clone->uio_iov->iov_base + adv;
1116 uio_clone->uio_iov->iov_len -= adv;
1117 uio_clone->uio_resid -= adv;
1118 uio_clone->uio_offset += adv;
1120 uio->uio_resid -= adv;
1121 uio->uio_offset += adv;
1123 if (error != 0 || adv == 0)
1126 td->td_ma = prev_td_ma;
1127 td->td_ma_cnt = prev_td_ma_cnt;
1128 curthread_pflags_restore(saveheld);
1130 vm_fault_enable_pagefaults(save);
1131 free(uio_clone, M_IOV);
1136 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
1137 int flags, struct thread *td)
1142 struct vn_io_fault_args args;
1145 doio = uio->uio_rw == UIO_READ ? vn_read : vn_write;
1147 foffset_lock_uio(fp, uio, flags);
1148 if (do_vn_io_fault(vp, uio)) {
1149 args.kind = VN_IO_FAULT_FOP;
1150 args.args.fop_args.fp = fp;
1151 args.args.fop_args.doio = doio;
1152 args.cred = active_cred;
1153 args.flags = flags | FOF_OFFSET;
1154 if (uio->uio_rw == UIO_READ) {
1155 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
1156 uio->uio_offset + uio->uio_resid);
1157 } else if ((fp->f_flag & O_APPEND) != 0 ||
1158 (flags & FOF_OFFSET) == 0) {
1159 /* For appenders, punt and lock the whole range. */
1160 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1162 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
1163 uio->uio_offset + uio->uio_resid);
1165 error = vn_io_fault1(vp, uio, &args, td);
1166 vn_rangelock_unlock(vp, rl_cookie);
1168 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
1170 foffset_unlock_uio(fp, uio, flags);
1175 * Helper function to perform the requested uiomove operation using
1176 * the held pages for io->uio_iov[0].iov_base buffer instead of
1177 * copyin/copyout. Access to the pages with uiomove_fromphys()
1178 * instead of iov_base prevents page faults that could occur due to
1179 * pmap_collect() invalidating the mapping created by
1180 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1181 * object cleanup revoking the write access from page mappings.
1183 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1184 * instead of plain uiomove().
1187 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1189 struct uio transp_uio;
1190 struct iovec transp_iov[1];
1196 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1197 uio->uio_segflg != UIO_USERSPACE)
1198 return (uiomove(data, xfersize, uio));
1200 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1201 transp_iov[0].iov_base = data;
1202 transp_uio.uio_iov = &transp_iov[0];
1203 transp_uio.uio_iovcnt = 1;
1204 if (xfersize > uio->uio_resid)
1205 xfersize = uio->uio_resid;
1206 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1207 transp_uio.uio_offset = 0;
1208 transp_uio.uio_segflg = UIO_SYSSPACE;
1210 * Since transp_iov points to data, and td_ma page array
1211 * corresponds to original uio->uio_iov, we need to invert the
1212 * direction of the i/o operation as passed to
1213 * uiomove_fromphys().
1215 switch (uio->uio_rw) {
1217 transp_uio.uio_rw = UIO_READ;
1220 transp_uio.uio_rw = UIO_WRITE;
1223 transp_uio.uio_td = uio->uio_td;
1224 error = uiomove_fromphys(td->td_ma,
1225 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1226 xfersize, &transp_uio);
1227 adv = xfersize - transp_uio.uio_resid;
1229 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1230 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1232 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1234 td->td_ma_cnt -= pgadv;
1235 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1236 uio->uio_iov->iov_len -= adv;
1237 uio->uio_resid -= adv;
1238 uio->uio_offset += adv;
1243 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1247 vm_offset_t iov_base;
1251 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1252 uio->uio_segflg != UIO_USERSPACE)
1253 return (uiomove_fromphys(ma, offset, xfersize, uio));
1255 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1256 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1257 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1258 switch (uio->uio_rw) {
1260 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1264 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1268 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1270 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1272 td->td_ma_cnt -= pgadv;
1273 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1274 uio->uio_iov->iov_len -= cnt;
1275 uio->uio_resid -= cnt;
1276 uio->uio_offset += cnt;
1282 * File table truncate routine.
1285 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1297 * Lock the whole range for truncation. Otherwise split i/o
1298 * might happen partly before and partly after the truncation.
1300 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1301 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1304 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1305 AUDIT_ARG_VNODE1(vp);
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 if ((fp->f_flag & O_FSYNC) != 0)
1320 vattr.va_vaflags |= VA_SYNC;
1321 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1325 vn_finished_write(mp);
1327 vn_rangelock_unlock(vp, rl_cookie);
1332 * File table vnode stat routine.
1335 vn_statfile(fp, sb, active_cred, td)
1338 struct ucred *active_cred;
1341 struct vnode *vp = fp->f_vnode;
1344 vn_lock(vp, LK_SHARED | LK_RETRY);
1345 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1352 * Stat a vnode; implementation for the stat syscall
1355 vn_stat(vp, sb, active_cred, file_cred, td)
1357 register struct stat *sb;
1358 struct ucred *active_cred;
1359 struct ucred *file_cred;
1363 register struct vattr *vap;
1367 AUDIT_ARG_VNODE1(vp);
1369 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1377 * Initialize defaults for new and unusual fields, so that file
1378 * systems which don't support these fields don't need to know
1381 vap->va_birthtime.tv_sec = -1;
1382 vap->va_birthtime.tv_nsec = 0;
1383 vap->va_fsid = VNOVAL;
1384 vap->va_rdev = NODEV;
1386 error = VOP_GETATTR(vp, vap, active_cred);
1391 * Zero the spare stat fields
1393 bzero(sb, sizeof *sb);
1396 * Copy from vattr table
1398 if (vap->va_fsid != VNOVAL)
1399 sb->st_dev = vap->va_fsid;
1401 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1402 sb->st_ino = vap->va_fileid;
1403 mode = vap->va_mode;
1404 switch (vap->va_type) {
1430 sb->st_nlink = vap->va_nlink;
1431 sb->st_uid = vap->va_uid;
1432 sb->st_gid = vap->va_gid;
1433 sb->st_rdev = vap->va_rdev;
1434 if (vap->va_size > OFF_MAX)
1436 sb->st_size = vap->va_size;
1437 sb->st_atim = vap->va_atime;
1438 sb->st_mtim = vap->va_mtime;
1439 sb->st_ctim = vap->va_ctime;
1440 sb->st_birthtim = vap->va_birthtime;
1443 * According to www.opengroup.org, the meaning of st_blksize is
1444 * "a filesystem-specific preferred I/O block size for this
1445 * object. In some filesystem types, this may vary from file
1447 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1450 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1452 sb->st_flags = vap->va_flags;
1453 if (priv_check(td, PRIV_VFS_GENERATION))
1456 sb->st_gen = vap->va_gen;
1458 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1463 * File table vnode ioctl routine.
1466 vn_ioctl(fp, com, data, active_cred, td)
1470 struct ucred *active_cred;
1478 switch (vp->v_type) {
1483 vn_lock(vp, LK_SHARED | LK_RETRY);
1484 error = VOP_GETATTR(vp, &vattr, active_cred);
1487 *(int *)data = vattr.va_size - fp->f_offset;
1493 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1498 return (VOP_IOCTL(vp, com, data, fp->f_flag,
1506 * File table vnode poll routine.
1509 vn_poll(fp, events, active_cred, td)
1512 struct ucred *active_cred;
1520 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1521 AUDIT_ARG_VNODE1(vp);
1522 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1527 error = VOP_POLL(vp, events, fp->f_cred, td);
1532 * Acquire the requested lock and then check for validity. LK_RETRY
1533 * permits vn_lock to return doomed vnodes.
1536 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1540 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1541 ("vn_lock called with no locktype."));
1543 #ifdef DEBUG_VFS_LOCKS
1544 KASSERT(vp->v_holdcnt != 0,
1545 ("vn_lock %p: zero hold count", vp));
1547 error = VOP_LOCK1(vp, flags, file, line);
1548 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1549 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1550 ("LK_RETRY set with incompatible flags (0x%x) or an error occurred (%d)",
1553 * Callers specify LK_RETRY if they wish to get dead vnodes.
1554 * If RETRY is not set, we return ENOENT instead.
1556 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1557 (flags & LK_RETRY) == 0) {
1562 } while (flags & LK_RETRY && error != 0);
1567 * File table vnode close routine.
1570 vn_closefile(fp, td)
1579 fp->f_ops = &badfileops;
1581 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK)
1584 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1586 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1587 lf.l_whence = SEEK_SET;
1590 lf.l_type = F_UNLCK;
1591 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1598 vn_suspendable(struct mount *mp)
1601 return (mp->mnt_op->vfs_susp_clean != NULL);
1605 * Preparing to start a filesystem write operation. If the operation is
1606 * permitted, then we bump the count of operations in progress and
1607 * proceed. If a suspend request is in progress, we wait until the
1608 * suspension is over, and then proceed.
1611 vn_start_write_locked(struct mount *mp, int flags)
1615 mtx_assert(MNT_MTX(mp), MA_OWNED);
1619 * Check on status of suspension.
1621 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1622 mp->mnt_susp_owner != curthread) {
1623 mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ?
1624 (flags & PCATCH) : 0) | (PUSER - 1);
1625 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1626 if (flags & V_NOWAIT) {
1627 error = EWOULDBLOCK;
1630 error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags,
1636 if (flags & V_XSLEEP)
1638 mp->mnt_writeopcount++;
1640 if (error != 0 || (flags & V_XSLEEP) != 0)
1647 vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
1652 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1653 ("V_MNTREF requires mp"));
1657 * If a vnode is provided, get and return the mount point that
1658 * to which it will write.
1661 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1663 if (error != EOPNOTSUPP)
1668 if ((mp = *mpp) == NULL)
1671 if (!vn_suspendable(mp)) {
1672 if (vp != NULL || (flags & V_MNTREF) != 0)
1678 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1680 * As long as a vnode is not provided we need to acquire a
1681 * refcount for the provided mountpoint too, in order to
1682 * emulate a vfs_ref().
1685 if (vp == NULL && (flags & V_MNTREF) == 0)
1688 return (vn_start_write_locked(mp, flags));
1692 * Secondary suspension. Used by operations such as vop_inactive
1693 * routines that are needed by the higher level functions. These
1694 * are allowed to proceed until all the higher level functions have
1695 * completed (indicated by mnt_writeopcount dropping to zero). At that
1696 * time, these operations are halted until the suspension is over.
1699 vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags)
1704 KASSERT((flags & V_MNTREF) == 0 || (*mpp != NULL && vp == NULL),
1705 ("V_MNTREF requires mp"));
1709 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1711 if (error != EOPNOTSUPP)
1717 * If we are not suspended or have not yet reached suspended
1718 * mode, then let the operation proceed.
1720 if ((mp = *mpp) == NULL)
1723 if (!vn_suspendable(mp)) {
1724 if (vp != NULL || (flags & V_MNTREF) != 0)
1730 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1732 * As long as a vnode is not provided we need to acquire a
1733 * refcount for the provided mountpoint too, in order to
1734 * emulate a vfs_ref().
1737 if (vp == NULL && (flags & V_MNTREF) == 0)
1739 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1740 mp->mnt_secondary_writes++;
1741 mp->mnt_secondary_accwrites++;
1745 if (flags & V_NOWAIT) {
1748 return (EWOULDBLOCK);
1751 * Wait for the suspension to finish.
1753 error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP |
1754 ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0),
1763 * Filesystem write operation has completed. If we are suspending and this
1764 * operation is the last one, notify the suspender that the suspension is
1768 vn_finished_write(mp)
1771 if (mp == NULL || !vn_suspendable(mp))
1775 mp->mnt_writeopcount--;
1776 if (mp->mnt_writeopcount < 0)
1777 panic("vn_finished_write: neg cnt");
1778 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1779 mp->mnt_writeopcount <= 0)
1780 wakeup(&mp->mnt_writeopcount);
1786 * Filesystem secondary write operation has completed. If we are
1787 * suspending and this operation is the last one, notify the suspender
1788 * that the suspension is now in effect.
1791 vn_finished_secondary_write(mp)
1794 if (mp == NULL || !vn_suspendable(mp))
1798 mp->mnt_secondary_writes--;
1799 if (mp->mnt_secondary_writes < 0)
1800 panic("vn_finished_secondary_write: neg cnt");
1801 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1802 mp->mnt_secondary_writes <= 0)
1803 wakeup(&mp->mnt_secondary_writes);
1810 * Request a filesystem to suspend write operations.
1813 vfs_write_suspend(struct mount *mp, int flags)
1817 MPASS(vn_suspendable(mp));
1820 if (mp->mnt_susp_owner == curthread) {
1824 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1825 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1828 * Unmount holds a write reference on the mount point. If we
1829 * own busy reference and drain for writers, we deadlock with
1830 * the reference draining in the unmount path. Callers of
1831 * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if
1832 * vfs_busy() reference is owned and caller is not in the
1835 if ((flags & VS_SKIP_UNMOUNT) != 0 &&
1836 (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
1841 mp->mnt_kern_flag |= MNTK_SUSPEND;
1842 mp->mnt_susp_owner = curthread;
1843 if (mp->mnt_writeopcount > 0)
1844 (void) msleep(&mp->mnt_writeopcount,
1845 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1848 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1849 vfs_write_resume(mp, 0);
1854 * Request a filesystem to resume write operations.
1857 vfs_write_resume(struct mount *mp, int flags)
1860 MPASS(vn_suspendable(mp));
1863 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1864 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1865 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1867 mp->mnt_susp_owner = NULL;
1868 wakeup(&mp->mnt_writeopcount);
1869 wakeup(&mp->mnt_flag);
1870 curthread->td_pflags &= ~TDP_IGNSUSP;
1871 if ((flags & VR_START_WRITE) != 0) {
1873 mp->mnt_writeopcount++;
1876 if ((flags & VR_NO_SUSPCLR) == 0)
1878 } else if ((flags & VR_START_WRITE) != 0) {
1880 vn_start_write_locked(mp, 0);
1887 * Helper loop around vfs_write_suspend() for filesystem unmount VFS
1891 vfs_write_suspend_umnt(struct mount *mp)
1895 MPASS(vn_suspendable(mp));
1896 KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0,
1897 ("vfs_write_suspend_umnt: recursed"));
1899 /* dounmount() already called vn_start_write(). */
1901 vn_finished_write(mp);
1902 error = vfs_write_suspend(mp, 0);
1904 vn_start_write(NULL, &mp, V_WAIT);
1908 if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1911 vn_start_write(NULL, &mp, V_WAIT);
1913 mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2);
1914 wakeup(&mp->mnt_flag);
1916 curthread->td_pflags |= TDP_IGNSUSP;
1921 * Implement kqueues for files by translating it to vnode operation.
1924 vn_kqfilter(struct file *fp, struct knote *kn)
1927 return (VOP_KQFILTER(fp->f_vnode, kn));
1931 * Simplified in-kernel wrapper calls for extended attribute access.
1932 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1933 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1936 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1937 const char *attrname, int *buflen, char *buf, struct thread *td)
1943 iov.iov_len = *buflen;
1946 auio.uio_iov = &iov;
1947 auio.uio_iovcnt = 1;
1948 auio.uio_rw = UIO_READ;
1949 auio.uio_segflg = UIO_SYSSPACE;
1951 auio.uio_offset = 0;
1952 auio.uio_resid = *buflen;
1954 if ((ioflg & IO_NODELOCKED) == 0)
1955 vn_lock(vp, LK_SHARED | LK_RETRY);
1957 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1959 /* authorize attribute retrieval as kernel */
1960 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1963 if ((ioflg & IO_NODELOCKED) == 0)
1967 *buflen = *buflen - auio.uio_resid;
1974 * XXX failure mode if partially written?
1977 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1978 const char *attrname, int buflen, char *buf, struct thread *td)
1985 iov.iov_len = buflen;
1988 auio.uio_iov = &iov;
1989 auio.uio_iovcnt = 1;
1990 auio.uio_rw = UIO_WRITE;
1991 auio.uio_segflg = UIO_SYSSPACE;
1993 auio.uio_offset = 0;
1994 auio.uio_resid = buflen;
1996 if ((ioflg & IO_NODELOCKED) == 0) {
1997 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1999 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2002 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2004 /* authorize attribute setting as kernel */
2005 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
2007 if ((ioflg & IO_NODELOCKED) == 0) {
2008 vn_finished_write(mp);
2016 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
2017 const char *attrname, struct thread *td)
2022 if ((ioflg & IO_NODELOCKED) == 0) {
2023 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
2025 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2028 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
2030 /* authorize attribute removal as kernel */
2031 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
2032 if (error == EOPNOTSUPP)
2033 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
2036 if ((ioflg & IO_NODELOCKED) == 0) {
2037 vn_finished_write(mp);
2045 vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags,
2049 return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp));
2053 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
2056 return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino,
2061 vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg,
2062 int lkflags, struct vnode **rvp)
2067 ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get");
2069 ltype = VOP_ISLOCKED(vp);
2070 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
2071 ("vn_vget_ino: vp not locked"));
2072 error = vfs_busy(mp, MBF_NOWAIT);
2076 error = vfs_busy(mp, 0);
2077 vn_lock(vp, ltype | LK_RETRY);
2081 if (vp->v_iflag & VI_DOOMED) {
2087 error = alloc(mp, alloc_arg, lkflags, rvp);
2090 vn_lock(vp, ltype | LK_RETRY);
2091 if (vp->v_iflag & VI_DOOMED) {
2104 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
2108 if (vp->v_type != VREG || td == NULL)
2110 if ((uoff_t)uio->uio_offset + uio->uio_resid >
2111 lim_cur(td, RLIMIT_FSIZE)) {
2112 PROC_LOCK(td->td_proc);
2113 kern_psignal(td->td_proc, SIGXFSZ);
2114 PROC_UNLOCK(td->td_proc);
2121 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
2128 vn_lock(vp, LK_SHARED | LK_RETRY);
2129 AUDIT_ARG_VNODE1(vp);
2132 return (setfmode(td, active_cred, vp, mode));
2136 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
2143 vn_lock(vp, LK_SHARED | LK_RETRY);
2144 AUDIT_ARG_VNODE1(vp);
2147 return (setfown(td, active_cred, vp, uid, gid));
2151 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
2155 if ((object = vp->v_object) == NULL)
2157 VM_OBJECT_WLOCK(object);
2158 vm_object_page_remove(object, start, end, 0);
2159 VM_OBJECT_WUNLOCK(object);
2163 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
2171 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
2172 ("Wrong command %lu", cmd));
2174 if (vn_lock(vp, LK_SHARED) != 0)
2176 if (vp->v_type != VREG) {
2180 error = VOP_GETATTR(vp, &va, cred);
2184 if (noff >= va.va_size) {
2188 bsize = vp->v_mount->mnt_stat.f_iosize;
2189 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
2190 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
2191 if (error == EOPNOTSUPP) {
2195 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
2196 (bnp != -1 && cmd == FIOSEEKDATA)) {
2203 if (noff > va.va_size)
2205 /* noff == va.va_size. There is an implicit hole at the end of file. */
2206 if (cmd == FIOSEEKDATA)
2216 vn_seek(struct file *fp, off_t offset, int whence, struct thread *td)
2221 off_t foffset, size;
2224 cred = td->td_ucred;
2226 foffset = foffset_lock(fp, 0);
2227 noneg = (vp->v_type != VCHR);
2233 (offset > 0 && foffset > OFF_MAX - offset))) {
2240 vn_lock(vp, LK_SHARED | LK_RETRY);
2241 error = VOP_GETATTR(vp, &vattr, cred);
2247 * If the file references a disk device, then fetch
2248 * the media size and use that to determine the ending
2251 if (vattr.va_size == 0 && vp->v_type == VCHR &&
2252 fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0)
2253 vattr.va_size = size;
2255 (vattr.va_size > OFF_MAX ||
2256 (offset > 0 && vattr.va_size > OFF_MAX - offset))) {
2260 offset += vattr.va_size;
2265 error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td);
2268 error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td);
2273 if (error == 0 && noneg && offset < 0)
2277 VFS_KNOTE_UNLOCKED(vp, 0);
2278 td->td_uretoff.tdu_off = offset;
2280 foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0);
2285 vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred,
2291 * Grant permission if the caller is the owner of the file, or
2292 * the super-user, or has ACL_WRITE_ATTRIBUTES permission on
2293 * on the file. If the time pointer is null, then write
2294 * permission on the file is also sufficient.
2296 * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes:
2297 * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES
2298 * will be allowed to set the times [..] to the current
2301 error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td);
2302 if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0)
2303 error = VOP_ACCESS(vp, VWRITE, cred, td);
2308 vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
2313 if (fp->f_type == DTYPE_FIFO)
2314 kif->kf_type = KF_TYPE_FIFO;
2316 kif->kf_type = KF_TYPE_VNODE;
2319 FILEDESC_SUNLOCK(fdp);
2320 error = vn_fill_kinfo_vnode(vp, kif);
2322 FILEDESC_SLOCK(fdp);
2327 vn_fill_junk(struct kinfo_file *kif)
2332 * Simulate vn_fullpath returning changing values for a given
2333 * vp during e.g. coredump.
2335 len = (arc4random() % (sizeof(kif->kf_path) - 2)) + 1;
2336 olen = strlen(kif->kf_path);
2338 strcpy(&kif->kf_path[len - 1], "$");
2340 for (; olen < len; olen++)
2341 strcpy(&kif->kf_path[olen], "A");
2345 vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif)
2348 char *fullpath, *freepath;
2351 kif->kf_vnode_type = vntype_to_kinfo(vp->v_type);
2354 error = vn_fullpath(curthread, vp, &fullpath, &freepath);
2356 strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path));
2358 if (freepath != NULL)
2359 free(freepath, M_TEMP);
2361 KFAIL_POINT_CODE(DEBUG_FP, fill_kinfo_vnode__random_path,
2366 * Retrieve vnode attributes.
2368 va.va_fsid = VNOVAL;
2370 vn_lock(vp, LK_SHARED | LK_RETRY);
2371 error = VOP_GETATTR(vp, &va, curthread->td_ucred);
2375 if (va.va_fsid != VNOVAL)
2376 kif->kf_un.kf_file.kf_file_fsid = va.va_fsid;
2378 kif->kf_un.kf_file.kf_file_fsid =
2379 vp->v_mount->mnt_stat.f_fsid.val[0];
2380 kif->kf_un.kf_file.kf_file_fileid = va.va_fileid;
2381 kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode);
2382 kif->kf_un.kf_file.kf_file_size = va.va_size;
2383 kif->kf_un.kf_file.kf_file_rdev = va.va_rdev;
2388 vn_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
2389 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
2393 struct pmckern_map_in pkm;
2399 boolean_t writecounted;
2402 #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \
2403 defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4)
2405 * POSIX shared-memory objects are defined to have
2406 * kernel persistence, and are not defined to support
2407 * read(2)/write(2) -- or even open(2). Thus, we can
2408 * use MAP_ASYNC to trade on-disk coherence for speed.
2409 * The shm_open(3) library routine turns on the FPOSIXSHM
2410 * flag to request this behavior.
2412 if ((fp->f_flag & FPOSIXSHM) != 0)
2413 flags |= MAP_NOSYNC;
2418 * Ensure that file and memory protections are
2419 * compatible. Note that we only worry about
2420 * writability if mapping is shared; in this case,
2421 * current and max prot are dictated by the open file.
2422 * XXX use the vnode instead? Problem is: what
2423 * credentials do we use for determination? What if
2424 * proc does a setuid?
2427 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0)
2428 maxprot = VM_PROT_NONE;
2430 maxprot = VM_PROT_EXECUTE;
2431 if ((fp->f_flag & FREAD) != 0)
2432 maxprot |= VM_PROT_READ;
2433 else if ((prot & VM_PROT_READ) != 0)
2437 * If we are sharing potential changes via MAP_SHARED and we
2438 * are trying to get write permission although we opened it
2439 * without asking for it, bail out.
2441 if ((flags & MAP_SHARED) != 0) {
2442 if ((fp->f_flag & FWRITE) != 0)
2443 maxprot |= VM_PROT_WRITE;
2444 else if ((prot & VM_PROT_WRITE) != 0)
2447 maxprot |= VM_PROT_WRITE;
2448 cap_maxprot |= VM_PROT_WRITE;
2450 maxprot &= cap_maxprot;
2452 writecounted = FALSE;
2453 error = vm_mmap_vnode(td, size, prot, &maxprot, &flags, vp,
2454 &foff, &object, &writecounted);
2457 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
2458 foff, writecounted, td);
2461 * If this mapping was accounted for in the vnode's
2462 * writecount, then undo that now.
2465 vnode_pager_release_writecount(object, 0, size);
2466 vm_object_deallocate(object);
2469 /* Inform hwpmc(4) if an executable is being mapped. */
2470 if (error == 0 && (prot & VM_PROT_EXECUTE) != 0) {
2472 pkm.pm_address = (uintptr_t) *addr;
2473 PMC_CALL_HOOK(td, PMC_FN_MMAP, (void *) &pkm);