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 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>
48 #include <sys/fcntl.h>
54 #include <sys/limits.h>
56 #include <sys/mount.h>
57 #include <sys/mutex.h>
58 #include <sys/namei.h>
59 #include <sys/vnode.h>
62 #include <sys/filio.h>
63 #include <sys/resourcevar.h>
65 #include <sys/sysctl.h>
66 #include <sys/ttycom.h>
68 #include <sys/syslog.h>
69 #include <sys/unistd.h>
71 #include <security/audit/audit.h>
72 #include <security/mac/mac_framework.h>
75 #include <vm/vm_extern.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_object.h>
79 #include <vm/vm_page.h>
81 static fo_rdwr_t vn_read;
82 static fo_rdwr_t vn_write;
83 static fo_rdwr_t vn_io_fault;
84 static fo_truncate_t vn_truncate;
85 static fo_ioctl_t vn_ioctl;
86 static fo_poll_t vn_poll;
87 static fo_kqfilter_t vn_kqfilter;
88 static fo_stat_t vn_statfile;
89 static fo_close_t vn_closefile;
91 struct fileops vnops = {
92 .fo_read = vn_io_fault,
93 .fo_write = vn_io_fault,
94 .fo_truncate = vn_truncate,
97 .fo_kqfilter = vn_kqfilter,
98 .fo_stat = vn_statfile,
99 .fo_close = vn_closefile,
100 .fo_chmod = vn_chmod,
101 .fo_chown = vn_chown,
102 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
106 vn_open(ndp, flagp, cmode, fp)
107 struct nameidata *ndp;
111 struct thread *td = ndp->ni_cnd.cn_thread;
113 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
117 * Common code for vnode open operations.
118 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
120 * Note that this does NOT free nameidata for the successful case,
121 * due to the NDINIT being done elsewhere.
124 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
125 struct ucred *cred, struct file *fp)
129 struct thread *td = ndp->ni_cnd.cn_thread;
131 struct vattr *vap = &vat;
134 int vfslocked, mpsafe;
136 mpsafe = ndp->ni_cnd.cn_flags & MPSAFE;
140 if (fmode & O_CREAT) {
141 ndp->ni_cnd.cn_nameiop = CREATE;
142 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF |
144 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
145 ndp->ni_cnd.cn_flags |= FOLLOW;
146 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
147 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
149 if ((error = namei(ndp)) != 0)
151 vfslocked = NDHASGIANT(ndp);
153 ndp->ni_cnd.cn_flags &= ~MPSAFE;
154 if (ndp->ni_vp == NULL) {
157 vap->va_mode = cmode;
159 vap->va_vaflags |= VA_EXCLUSIVE;
160 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
161 NDFREE(ndp, NDF_ONLY_PNBUF);
163 VFS_UNLOCK_GIANT(vfslocked);
164 if ((error = vn_start_write(NULL, &mp,
165 V_XSLEEP | PCATCH)) != 0)
170 error = mac_vnode_check_create(cred, ndp->ni_dvp,
174 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
177 vn_finished_write(mp);
179 VFS_UNLOCK_GIANT(vfslocked);
180 NDFREE(ndp, NDF_ONLY_PNBUF);
186 if (ndp->ni_dvp == ndp->ni_vp)
192 if (fmode & O_EXCL) {
199 ndp->ni_cnd.cn_nameiop = LOOKUP;
200 ndp->ni_cnd.cn_flags = ISOPEN |
201 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) |
203 if (!(fmode & FWRITE))
204 ndp->ni_cnd.cn_flags |= LOCKSHARED;
205 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
206 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
207 if ((error = namei(ndp)) != 0)
210 ndp->ni_cnd.cn_flags &= ~MPSAFE;
211 vfslocked = NDHASGIANT(ndp);
214 if (vp->v_type == VLNK) {
218 if (vp->v_type == VSOCK) {
222 if (vp->v_type != VDIR && fmode & O_DIRECTORY) {
227 if (fmode & (FWRITE | O_TRUNC)) {
228 if (vp->v_type == VDIR) {
238 if ((fmode & O_APPEND) && (fmode & FWRITE))
241 error = mac_vnode_check_open(cred, vp, accmode);
245 if ((fmode & O_CREAT) == 0) {
246 if (accmode & VWRITE) {
247 error = vn_writechk(vp);
252 error = VOP_ACCESS(vp, accmode, cred, td);
257 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
261 VOP_ADD_WRITECOUNT(vp, 1);
263 ASSERT_VOP_LOCKED(vp, "vn_open_cred");
265 VFS_UNLOCK_GIANT(vfslocked);
268 NDFREE(ndp, NDF_ONLY_PNBUF);
270 VFS_UNLOCK_GIANT(vfslocked);
277 * Check for write permissions on the specified vnode.
278 * Prototype text segments cannot be written.
282 register struct vnode *vp;
285 ASSERT_VOP_LOCKED(vp, "vn_writechk");
287 * If there's shared text associated with
288 * the vnode, try to free it up once. If
289 * we fail, we can't allow writing.
301 vn_close(vp, flags, file_cred, td)
302 register struct vnode *vp;
304 struct ucred *file_cred;
308 int error, lock_flags;
310 if (!(flags & FWRITE) && vp->v_mount != NULL &&
311 vp->v_mount->mnt_kern_flag & MNTK_EXTENDED_SHARED)
312 lock_flags = LK_SHARED;
314 lock_flags = LK_EXCLUSIVE;
316 VFS_ASSERT_GIANT(vp->v_mount);
318 vn_start_write(vp, &mp, V_WAIT);
319 vn_lock(vp, lock_flags | LK_RETRY);
320 if (flags & FWRITE) {
321 VNASSERT(vp->v_writecount > 0, vp,
322 ("vn_close: negative writecount"));
323 VOP_ADD_WRITECOUNT(vp, -1);
325 error = VOP_CLOSE(vp, flags, file_cred, td);
327 vn_finished_write(mp);
332 * Heuristic to detect sequential operation.
335 sequential_heuristic(struct uio *uio, struct file *fp)
338 if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
339 return (fp->f_seqcount << IO_SEQSHIFT);
342 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
343 * that the first I/O is normally considered to be slightly
344 * sequential. Seeking to offset 0 doesn't change sequentiality
345 * unless previous seeks have reduced f_seqcount to 0, in which
346 * case offset 0 is not special.
348 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
349 uio->uio_offset == fp->f_nextoff) {
351 * f_seqcount is in units of fixed-size blocks so that it
352 * depends mainly on the amount of sequential I/O and not
353 * much on the number of sequential I/O's. The fixed size
354 * of 16384 is hard-coded here since it is (not quite) just
355 * a magic size that works well here. This size is more
356 * closely related to the best I/O size for real disks than
357 * to any block size used by software.
359 fp->f_seqcount += howmany(uio->uio_resid, 16384);
360 if (fp->f_seqcount > IO_SEQMAX)
361 fp->f_seqcount = IO_SEQMAX;
362 return (fp->f_seqcount << IO_SEQSHIFT);
365 /* Not sequential. Quickly draw-down sequentiality. */
366 if (fp->f_seqcount > 1)
374 * Package up an I/O request on a vnode into a uio and do it.
377 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
378 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
379 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
386 int error, lock_flags;
388 VFS_ASSERT_GIANT(vp->v_mount);
390 auio.uio_iov = &aiov;
392 aiov.iov_base = base;
394 auio.uio_resid = len;
395 auio.uio_offset = offset;
396 auio.uio_segflg = segflg;
401 if ((ioflg & IO_NODELOCKED) == 0) {
402 if (rw == UIO_READ) {
403 rl_cookie = vn_rangelock_rlock(vp, offset,
406 rl_cookie = vn_rangelock_wlock(vp, offset,
410 if (rw == UIO_WRITE) {
411 if (vp->v_type != VCHR &&
412 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
415 if (MNT_SHARED_WRITES(mp) ||
416 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
417 lock_flags = LK_SHARED;
419 lock_flags = LK_EXCLUSIVE;
421 lock_flags = LK_SHARED;
422 vn_lock(vp, lock_flags | LK_RETRY);
426 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
428 if ((ioflg & IO_NOMACCHECK) == 0) {
430 error = mac_vnode_check_read(active_cred, file_cred,
433 error = mac_vnode_check_write(active_cred, file_cred,
438 if (file_cred != NULL)
443 error = VOP_READ(vp, &auio, ioflg, cred);
445 error = VOP_WRITE(vp, &auio, ioflg, cred);
448 *aresid = auio.uio_resid;
450 if (auio.uio_resid && error == 0)
452 if ((ioflg & IO_NODELOCKED) == 0) {
455 vn_finished_write(mp);
458 if (rl_cookie != NULL)
459 vn_rangelock_unlock(vp, rl_cookie);
464 * Package up an I/O request on a vnode into a uio and do it. The I/O
465 * request is split up into smaller chunks and we try to avoid saturating
466 * the buffer cache while potentially holding a vnode locked, so we
467 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
468 * to give other processes a chance to lock the vnode (either other processes
469 * core'ing the same binary, or unrelated processes scanning the directory).
472 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
473 file_cred, aresid, td)
481 struct ucred *active_cred;
482 struct ucred *file_cred;
489 VFS_ASSERT_GIANT(vp->v_mount);
495 * Force `offset' to a multiple of MAXBSIZE except possibly
496 * for the first chunk, so that filesystems only need to
497 * write full blocks except possibly for the first and last
500 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
504 if (rw != UIO_READ && vp->v_type == VREG)
507 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
508 ioflg, active_cred, file_cred, &iaresid, td);
509 len -= chunk; /* aresid calc already includes length */
513 base = (char *)base + chunk;
514 kern_yield(PRI_USER);
517 *aresid = len + iaresid;
522 foffset_lock(struct file *fp, int flags)
527 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
529 #if OFF_MAX <= LONG_MAX
531 * Caller only wants the current f_offset value. Assume that
532 * the long and shorter integer types reads are atomic.
534 if ((flags & FOF_NOLOCK) != 0)
535 return (fp->f_offset);
539 * According to McKusick the vn lock was protecting f_offset here.
540 * It is now protected by the FOFFSET_LOCKED flag.
542 mtxp = mtx_pool_find(mtxpool_sleep, fp);
544 if ((flags & FOF_NOLOCK) == 0) {
545 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
546 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
547 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
550 fp->f_vnread_flags |= FOFFSET_LOCKED;
558 foffset_unlock(struct file *fp, off_t val, int flags)
562 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
564 #if OFF_MAX <= LONG_MAX
565 if ((flags & FOF_NOLOCK) != 0) {
566 if ((flags & FOF_NOUPDATE) == 0)
568 if ((flags & FOF_NEXTOFF) != 0)
574 mtxp = mtx_pool_find(mtxpool_sleep, fp);
576 if ((flags & FOF_NOUPDATE) == 0)
578 if ((flags & FOF_NEXTOFF) != 0)
580 if ((flags & FOF_NOLOCK) == 0) {
581 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
582 ("Lost FOFFSET_LOCKED"));
583 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
584 wakeup(&fp->f_vnread_flags);
585 fp->f_vnread_flags = 0;
591 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
594 if ((flags & FOF_OFFSET) == 0)
595 uio->uio_offset = foffset_lock(fp, flags);
599 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
602 if ((flags & FOF_OFFSET) == 0)
603 foffset_unlock(fp, uio->uio_offset, flags);
607 get_advice(struct file *fp, struct uio *uio)
612 ret = POSIX_FADV_NORMAL;
613 if (fp->f_advice == NULL)
616 mtxp = mtx_pool_find(mtxpool_sleep, fp);
618 if (uio->uio_offset >= fp->f_advice->fa_start &&
619 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
620 ret = fp->f_advice->fa_advice;
626 * File table vnode read routine.
629 vn_read(fp, uio, active_cred, flags, td)
632 struct ucred *active_cred;
639 int advice, vfslocked;
640 off_t offset, start, end;
642 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
644 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
647 if (fp->f_flag & FNONBLOCK)
649 if (fp->f_flag & O_DIRECT)
651 advice = get_advice(fp, uio);
652 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
653 vn_lock(vp, LK_SHARED | LK_RETRY);
656 case POSIX_FADV_NORMAL:
657 case POSIX_FADV_SEQUENTIAL:
658 case POSIX_FADV_NOREUSE:
659 ioflag |= sequential_heuristic(uio, fp);
661 case POSIX_FADV_RANDOM:
662 /* Disable read-ahead for random I/O. */
665 offset = uio->uio_offset;
668 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
671 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
672 fp->f_nextoff = uio->uio_offset;
674 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
675 offset != uio->uio_offset) {
677 * Use POSIX_FADV_DONTNEED to flush clean pages and
678 * buffers for the backing file after a
679 * POSIX_FADV_NOREUSE read(2). To optimize the common
680 * case of using POSIX_FADV_NOREUSE with sequential
681 * access, track the previous implicit DONTNEED
682 * request and grow this request to include the
683 * current read(2) in addition to the previous
684 * DONTNEED. With purely sequential access this will
685 * cause the DONTNEED requests to continously grow to
686 * cover all of the previously read regions of the
687 * file. This allows filesystem blocks that are
688 * accessed by multiple calls to read(2) to be flushed
689 * once the last read(2) finishes.
692 end = uio->uio_offset - 1;
693 mtxp = mtx_pool_find(mtxpool_sleep, fp);
695 if (fp->f_advice != NULL &&
696 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
697 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
698 start = fp->f_advice->fa_prevstart;
699 else if (fp->f_advice->fa_prevstart != 0 &&
700 fp->f_advice->fa_prevstart == end + 1)
701 end = fp->f_advice->fa_prevend;
702 fp->f_advice->fa_prevstart = start;
703 fp->f_advice->fa_prevend = end;
706 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
708 VFS_UNLOCK_GIANT(vfslocked);
713 * File table vnode write routine.
716 vn_write(fp, uio, active_cred, flags, td)
719 struct ucred *active_cred;
726 int error, ioflag, lock_flags;
727 int advice, vfslocked;
728 off_t offset, start, end;
730 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
732 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
734 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
735 if (vp->v_type == VREG)
738 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
740 if (fp->f_flag & FNONBLOCK)
742 if (fp->f_flag & O_DIRECT)
744 if ((fp->f_flag & O_FSYNC) ||
745 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
748 if (vp->v_type != VCHR &&
749 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
752 advice = get_advice(fp, uio);
754 if ((MNT_SHARED_WRITES(mp) ||
755 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) &&
756 (flags & FOF_OFFSET) != 0) {
757 lock_flags = LK_SHARED;
759 lock_flags = LK_EXCLUSIVE;
762 vn_lock(vp, lock_flags | LK_RETRY);
764 case POSIX_FADV_NORMAL:
765 case POSIX_FADV_SEQUENTIAL:
766 case POSIX_FADV_NOREUSE:
767 ioflag |= sequential_heuristic(uio, fp);
769 case POSIX_FADV_RANDOM:
770 /* XXX: Is this correct? */
773 offset = uio->uio_offset;
776 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
779 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
780 fp->f_nextoff = uio->uio_offset;
782 if (vp->v_type != VCHR)
783 vn_finished_write(mp);
784 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
785 offset != uio->uio_offset) {
787 * Use POSIX_FADV_DONTNEED to flush clean pages and
788 * buffers for the backing file after a
789 * POSIX_FADV_NOREUSE write(2). To optimize the
790 * common case of using POSIX_FADV_NOREUSE with
791 * sequential access, track the previous implicit
792 * DONTNEED request and grow this request to include
793 * the current write(2) in addition to the previous
794 * DONTNEED. With purely sequential access this will
795 * cause the DONTNEED requests to continously grow to
796 * cover all of the previously written regions of the
799 * Note that the blocks just written are almost
800 * certainly still dirty, so this only works when
801 * VOP_ADVISE() calls from subsequent writes push out
802 * the data written by this write(2) once the backing
803 * buffers are clean. However, as compared to forcing
804 * IO_DIRECT, this gives much saner behavior. Write
805 * clustering is still allowed, and clean pages are
806 * merely moved to the cache page queue rather than
807 * outright thrown away. This means a subsequent
808 * read(2) can still avoid hitting the disk if the
809 * pages have not been reclaimed.
811 * This does make POSIX_FADV_NOREUSE largely useless
812 * with non-sequential access. However, sequential
813 * access is the more common use case and the flag is
817 end = uio->uio_offset - 1;
818 mtxp = mtx_pool_find(mtxpool_sleep, fp);
820 if (fp->f_advice != NULL &&
821 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
822 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
823 start = fp->f_advice->fa_prevstart;
824 else if (fp->f_advice->fa_prevstart != 0 &&
825 fp->f_advice->fa_prevstart == end + 1)
826 end = fp->f_advice->fa_prevend;
827 fp->f_advice->fa_prevstart = start;
828 fp->f_advice->fa_prevend = end;
831 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
835 VFS_UNLOCK_GIANT(vfslocked);
839 static const int io_hold_cnt = 16;
840 static int vn_io_fault_enable = 0;
841 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
842 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
843 static u_long vn_io_faults_cnt;
844 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
845 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
848 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
849 * prevent the following deadlock:
851 * Assume that the thread A reads from the vnode vp1 into userspace
852 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
853 * currently not resident, then system ends up with the call chain
854 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
855 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
856 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
857 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
858 * backed by the pages of vnode vp1, and some page in buf2 is not
859 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
861 * To prevent the lock order reversal and deadlock, vn_io_fault() does
862 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
863 * Instead, it first tries to do the whole range i/o with pagefaults
864 * disabled. If all pages in the i/o buffer are resident and mapped,
865 * VOP will succeed (ignoring the genuine filesystem errors).
866 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
867 * i/o in chunks, with all pages in the chunk prefaulted and held
868 * using vm_fault_quick_hold_pages().
870 * Filesystems using this deadlock avoidance scheme should use the
871 * array of the held pages from uio, saved in the curthread->td_ma,
872 * instead of doing uiomove(). A helper function
873 * vn_io_fault_uiomove() converts uiomove request into
874 * uiomove_fromphys() over td_ma array.
876 * Since vnode locks do not cover the whole i/o anymore, rangelocks
877 * make the current i/o request atomic with respect to other i/os and
881 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
882 int flags, struct thread *td)
884 vm_page_t ma[io_hold_cnt + 2];
885 struct uio *uio_clone, short_uio;
886 struct iovec short_iovec[1];
891 vm_page_t *prev_td_ma;
892 int cnt, error, save, saveheld, prev_td_ma_cnt;
893 vm_offset_t addr, end;
898 if (uio->uio_rw == UIO_READ)
903 foffset_lock_uio(fp, uio, flags);
905 if (uio->uio_segflg != UIO_USERSPACE || vp->v_type != VREG ||
906 ((mp = vp->v_mount) != NULL &&
907 (mp->mnt_kern_flag & MNTK_NO_IOPF) == 0) ||
908 !vn_io_fault_enable) {
909 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
914 * The UFS follows IO_UNIT directive and replays back both
915 * uio_offset and uio_resid if an error is encountered during the
916 * operation. But, since the iovec may be already advanced,
917 * uio is still in an inconsistent state.
919 * Cache a copy of the original uio, which is advanced to the redo
920 * point using UIO_NOCOPY below.
922 uio_clone = cloneuio(uio);
923 resid = uio->uio_resid;
925 short_uio.uio_segflg = UIO_USERSPACE;
926 short_uio.uio_rw = uio->uio_rw;
927 short_uio.uio_td = uio->uio_td;
929 if (uio->uio_rw == UIO_READ) {
930 prot = VM_PROT_WRITE;
931 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
932 uio->uio_offset + uio->uio_resid);
935 if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0)
936 /* For appenders, punt and lock the whole range. */
937 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
939 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
940 uio->uio_offset + uio->uio_resid);
943 save = vm_fault_disable_pagefaults();
944 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
948 atomic_add_long(&vn_io_faults_cnt, 1);
949 uio_clone->uio_segflg = UIO_NOCOPY;
950 uiomove(NULL, resid - uio->uio_resid, uio_clone);
951 uio_clone->uio_segflg = uio->uio_segflg;
953 saveheld = curthread_pflags_set(TDP_UIOHELD);
954 prev_td_ma = td->td_ma;
955 prev_td_ma_cnt = td->td_ma_cnt;
957 while (uio_clone->uio_resid != 0) {
958 len = uio_clone->uio_iov->iov_len;
960 KASSERT(uio_clone->uio_iovcnt >= 1,
961 ("iovcnt underflow"));
962 uio_clone->uio_iov++;
963 uio_clone->uio_iovcnt--;
967 addr = (vm_offset_t)uio_clone->uio_iov->iov_base;
968 end = round_page(addr + len);
969 cnt = howmany(end - trunc_page(addr), PAGE_SIZE);
971 * A perfectly misaligned address and length could cause
972 * both the start and the end of the chunk to use partial
973 * page. +2 accounts for such a situation.
975 if (cnt > io_hold_cnt + 2) {
976 len = io_hold_cnt * PAGE_SIZE;
977 KASSERT(howmany(round_page(addr + len) -
978 trunc_page(addr), PAGE_SIZE) <= io_hold_cnt + 2,
981 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
982 addr, len, prot, ma, io_hold_cnt + 2);
987 short_uio.uio_iov = &short_iovec[0];
988 short_iovec[0].iov_base = (void *)addr;
989 short_uio.uio_iovcnt = 1;
990 short_uio.uio_resid = short_iovec[0].iov_len = len;
991 short_uio.uio_offset = uio_clone->uio_offset;
995 error = doio(fp, &short_uio, active_cred, flags | FOF_OFFSET,
997 vm_page_unhold_pages(ma, cnt);
998 adv = len - short_uio.uio_resid;
1000 uio_clone->uio_iov->iov_base =
1001 (char *)uio_clone->uio_iov->iov_base + adv;
1002 uio_clone->uio_iov->iov_len -= adv;
1003 uio_clone->uio_resid -= adv;
1004 uio_clone->uio_offset += adv;
1006 uio->uio_resid -= adv;
1007 uio->uio_offset += adv;
1009 if (error != 0 || adv == 0)
1012 td->td_ma = prev_td_ma;
1013 td->td_ma_cnt = prev_td_ma_cnt;
1014 curthread_pflags_restore(saveheld);
1016 vm_fault_enable_pagefaults(save);
1017 vn_rangelock_unlock(vp, rl_cookie);
1018 free(uio_clone, M_IOV);
1020 foffset_unlock_uio(fp, uio, flags);
1025 * Helper function to perform the requested uiomove operation using
1026 * the held pages for io->uio_iov[0].iov_base buffer instead of
1027 * copyin/copyout. Access to the pages with uiomove_fromphys()
1028 * instead of iov_base prevents page faults that could occur due to
1029 * pmap_collect() invalidating the mapping created by
1030 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1031 * object cleanup revoking the write access from page mappings.
1033 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1034 * instead of plain uiomove().
1037 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1039 struct uio transp_uio;
1040 struct iovec transp_iov[1];
1046 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1047 uio->uio_segflg != UIO_USERSPACE)
1048 return (uiomove(data, xfersize, uio));
1050 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1051 transp_iov[0].iov_base = data;
1052 transp_uio.uio_iov = &transp_iov[0];
1053 transp_uio.uio_iovcnt = 1;
1054 if (xfersize > uio->uio_resid)
1055 xfersize = uio->uio_resid;
1056 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1057 transp_uio.uio_offset = 0;
1058 transp_uio.uio_segflg = UIO_SYSSPACE;
1060 * Since transp_iov points to data, and td_ma page array
1061 * corresponds to original uio->uio_iov, we need to invert the
1062 * direction of the i/o operation as passed to
1063 * uiomove_fromphys().
1065 switch (uio->uio_rw) {
1067 transp_uio.uio_rw = UIO_READ;
1070 transp_uio.uio_rw = UIO_WRITE;
1073 transp_uio.uio_td = uio->uio_td;
1074 error = uiomove_fromphys(td->td_ma,
1075 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1076 xfersize, &transp_uio);
1077 adv = xfersize - transp_uio.uio_resid;
1079 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1080 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1082 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1084 td->td_ma_cnt -= pgadv;
1085 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1086 uio->uio_iov->iov_len -= adv;
1087 uio->uio_resid -= adv;
1088 uio->uio_offset += adv;
1093 vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize,
1097 vm_offset_t iov_base;
1101 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1102 uio->uio_segflg != UIO_USERSPACE)
1103 return (uiomove_fromphys(ma, offset, xfersize, uio));
1105 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1106 cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize;
1107 iov_base = (vm_offset_t)uio->uio_iov->iov_base;
1108 switch (uio->uio_rw) {
1110 pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma,
1114 pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK,
1118 pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT);
1120 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1122 td->td_ma_cnt -= pgadv;
1123 uio->uio_iov->iov_base = (char *)(iov_base + cnt);
1124 uio->uio_iov->iov_len -= cnt;
1125 uio->uio_resid -= cnt;
1126 uio->uio_offset += cnt;
1132 * File table truncate routine.
1135 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1148 * Lock the whole range for truncation. Otherwise split i/o
1149 * might happen partly before and partly after the truncation.
1151 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1152 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1153 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1156 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1157 if (vp->v_type == VDIR) {
1162 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1166 error = vn_writechk(vp);
1169 vattr.va_size = length;
1170 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1174 vn_finished_write(mp);
1176 VFS_UNLOCK_GIANT(vfslocked);
1177 vn_rangelock_unlock(vp, rl_cookie);
1182 * File table vnode stat routine.
1185 vn_statfile(fp, sb, active_cred, td)
1188 struct ucred *active_cred;
1191 struct vnode *vp = fp->f_vnode;
1195 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1196 vn_lock(vp, LK_SHARED | LK_RETRY);
1197 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1199 VFS_UNLOCK_GIANT(vfslocked);
1205 * Stat a vnode; implementation for the stat syscall
1208 vn_stat(vp, sb, active_cred, file_cred, td)
1210 register struct stat *sb;
1211 struct ucred *active_cred;
1212 struct ucred *file_cred;
1216 register struct vattr *vap;
1221 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1229 * Initialize defaults for new and unusual fields, so that file
1230 * systems which don't support these fields don't need to know
1233 vap->va_birthtime.tv_sec = -1;
1234 vap->va_birthtime.tv_nsec = 0;
1235 vap->va_fsid = VNOVAL;
1236 vap->va_rdev = NODEV;
1238 error = VOP_GETATTR(vp, vap, active_cred);
1243 * Zero the spare stat fields
1245 bzero(sb, sizeof *sb);
1248 * Copy from vattr table
1250 if (vap->va_fsid != VNOVAL)
1251 sb->st_dev = vap->va_fsid;
1253 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1254 sb->st_ino = vap->va_fileid;
1255 mode = vap->va_mode;
1256 switch (vap->va_type) {
1282 sb->st_nlink = vap->va_nlink;
1283 sb->st_uid = vap->va_uid;
1284 sb->st_gid = vap->va_gid;
1285 sb->st_rdev = vap->va_rdev;
1286 if (vap->va_size > OFF_MAX)
1288 sb->st_size = vap->va_size;
1289 sb->st_atim = vap->va_atime;
1290 sb->st_mtim = vap->va_mtime;
1291 sb->st_ctim = vap->va_ctime;
1292 sb->st_birthtim = vap->va_birthtime;
1295 * According to www.opengroup.org, the meaning of st_blksize is
1296 * "a filesystem-specific preferred I/O block size for this
1297 * object. In some filesystem types, this may vary from file
1299 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1302 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1304 sb->st_flags = vap->va_flags;
1305 if (priv_check(td, PRIV_VFS_GENERATION))
1308 sb->st_gen = vap->va_gen;
1310 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1315 * File table vnode ioctl routine.
1318 vn_ioctl(fp, com, data, active_cred, td)
1322 struct ucred *active_cred;
1325 struct vnode *vp = fp->f_vnode;
1330 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1332 switch (vp->v_type) {
1335 if (com == FIONREAD) {
1336 vn_lock(vp, LK_SHARED | LK_RETRY);
1337 error = VOP_GETATTR(vp, &vattr, active_cred);
1340 *(int *)data = vattr.va_size - fp->f_offset;
1341 } else if (com == FIONBIO || com == FIOASYNC) /* XXX */
1344 error = VOP_IOCTL(vp, com, data, fp->f_flag,
1351 VFS_UNLOCK_GIANT(vfslocked);
1356 * File table vnode poll routine.
1359 vn_poll(fp, events, active_cred, td)
1362 struct ucred *active_cred;
1370 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1372 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1373 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1378 error = VOP_POLL(vp, events, fp->f_cred, td);
1379 VFS_UNLOCK_GIANT(vfslocked);
1384 * Acquire the requested lock and then check for validity. LK_RETRY
1385 * permits vn_lock to return doomed vnodes.
1388 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1392 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1393 ("vn_lock called with no locktype."));
1395 #ifdef DEBUG_VFS_LOCKS
1396 KASSERT(vp->v_holdcnt != 0,
1397 ("vn_lock %p: zero hold count", vp));
1399 error = VOP_LOCK1(vp, flags, file, line);
1400 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1401 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1402 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1405 * Callers specify LK_RETRY if they wish to get dead vnodes.
1406 * If RETRY is not set, we return ENOENT instead.
1408 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1409 (flags & LK_RETRY) == 0) {
1414 } while (flags & LK_RETRY && error != 0);
1419 * File table vnode close routine.
1422 vn_closefile(fp, td)
1433 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1434 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1435 lf.l_whence = SEEK_SET;
1438 lf.l_type = F_UNLCK;
1439 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1442 fp->f_ops = &badfileops;
1444 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1445 VFS_UNLOCK_GIANT(vfslocked);
1450 * Preparing to start a filesystem write operation. If the operation is
1451 * permitted, then we bump the count of operations in progress and
1452 * proceed. If a suspend request is in progress, we wait until the
1453 * suspension is over, and then proceed.
1456 vn_start_write_locked(struct mount *mp, int flags)
1460 mtx_assert(MNT_MTX(mp), MA_OWNED);
1464 * Check on status of suspension.
1466 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1467 mp->mnt_susp_owner != curthread) {
1468 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1469 if (flags & V_NOWAIT) {
1470 error = EWOULDBLOCK;
1473 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1474 (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1479 if (flags & V_XSLEEP)
1481 mp->mnt_writeopcount++;
1483 if (error != 0 || (flags & V_XSLEEP) != 0)
1490 vn_start_write(vp, mpp, flags)
1500 * If a vnode is provided, get and return the mount point that
1501 * to which it will write.
1504 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1506 if (error != EOPNOTSUPP)
1511 if ((mp = *mpp) == NULL)
1515 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1517 * As long as a vnode is not provided we need to acquire a
1518 * refcount for the provided mountpoint too, in order to
1519 * emulate a vfs_ref().
1525 return (vn_start_write_locked(mp, flags));
1529 * Secondary suspension. Used by operations such as vop_inactive
1530 * routines that are needed by the higher level functions. These
1531 * are allowed to proceed until all the higher level functions have
1532 * completed (indicated by mnt_writeopcount dropping to zero). At that
1533 * time, these operations are halted until the suspension is over.
1536 vn_start_secondary_write(vp, mpp, flags)
1546 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1548 if (error != EOPNOTSUPP)
1554 * If we are not suspended or have not yet reached suspended
1555 * mode, then let the operation proceed.
1557 if ((mp = *mpp) == NULL)
1561 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1563 * As long as a vnode is not provided we need to acquire a
1564 * refcount for the provided mountpoint too, in order to
1565 * emulate a vfs_ref().
1570 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1571 mp->mnt_secondary_writes++;
1572 mp->mnt_secondary_accwrites++;
1576 if (flags & V_NOWAIT) {
1579 return (EWOULDBLOCK);
1582 * Wait for the suspension to finish.
1584 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1585 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1593 * Filesystem write operation has completed. If we are suspending and this
1594 * operation is the last one, notify the suspender that the suspension is
1598 vn_finished_write(mp)
1605 mp->mnt_writeopcount--;
1606 if (mp->mnt_writeopcount < 0)
1607 panic("vn_finished_write: neg cnt");
1608 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1609 mp->mnt_writeopcount <= 0)
1610 wakeup(&mp->mnt_writeopcount);
1616 * Filesystem secondary write operation has completed. If we are
1617 * suspending and this operation is the last one, notify the suspender
1618 * that the suspension is now in effect.
1621 vn_finished_secondary_write(mp)
1628 mp->mnt_secondary_writes--;
1629 if (mp->mnt_secondary_writes < 0)
1630 panic("vn_finished_secondary_write: neg cnt");
1631 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1632 mp->mnt_secondary_writes <= 0)
1633 wakeup(&mp->mnt_secondary_writes);
1640 * Request a filesystem to suspend write operations.
1643 vfs_write_suspend(mp)
1649 if (mp->mnt_susp_owner == curthread) {
1653 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1654 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1655 mp->mnt_kern_flag |= MNTK_SUSPEND;
1656 mp->mnt_susp_owner = curthread;
1657 if (mp->mnt_writeopcount > 0)
1658 (void) msleep(&mp->mnt_writeopcount,
1659 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1662 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1663 vfs_write_resume(mp);
1668 * Request a filesystem to resume write operations.
1671 vfs_write_resume_flags(struct mount *mp, int flags)
1675 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1676 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1677 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1679 mp->mnt_susp_owner = NULL;
1680 wakeup(&mp->mnt_writeopcount);
1681 wakeup(&mp->mnt_flag);
1682 curthread->td_pflags &= ~TDP_IGNSUSP;
1683 if ((flags & VR_START_WRITE) != 0) {
1685 mp->mnt_writeopcount++;
1688 if ((flags & VR_NO_SUSPCLR) == 0)
1690 } else if ((flags & VR_START_WRITE) != 0) {
1692 vn_start_write_locked(mp, 0);
1699 vfs_write_resume(struct mount *mp)
1702 vfs_write_resume_flags(mp, 0);
1706 * Implement kqueues for files by translating it to vnode operation.
1709 vn_kqfilter(struct file *fp, struct knote *kn)
1714 vfslocked = VFS_LOCK_GIANT(fp->f_vnode->v_mount);
1715 error = VOP_KQFILTER(fp->f_vnode, kn);
1716 VFS_UNLOCK_GIANT(vfslocked);
1722 * Simplified in-kernel wrapper calls for extended attribute access.
1723 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1724 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1727 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1728 const char *attrname, int *buflen, char *buf, struct thread *td)
1734 iov.iov_len = *buflen;
1737 auio.uio_iov = &iov;
1738 auio.uio_iovcnt = 1;
1739 auio.uio_rw = UIO_READ;
1740 auio.uio_segflg = UIO_SYSSPACE;
1742 auio.uio_offset = 0;
1743 auio.uio_resid = *buflen;
1745 if ((ioflg & IO_NODELOCKED) == 0)
1746 vn_lock(vp, LK_SHARED | LK_RETRY);
1748 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1750 /* authorize attribute retrieval as kernel */
1751 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1754 if ((ioflg & IO_NODELOCKED) == 0)
1758 *buflen = *buflen - auio.uio_resid;
1765 * XXX failure mode if partially written?
1768 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1769 const char *attrname, int buflen, char *buf, struct thread *td)
1776 iov.iov_len = buflen;
1779 auio.uio_iov = &iov;
1780 auio.uio_iovcnt = 1;
1781 auio.uio_rw = UIO_WRITE;
1782 auio.uio_segflg = UIO_SYSSPACE;
1784 auio.uio_offset = 0;
1785 auio.uio_resid = buflen;
1787 if ((ioflg & IO_NODELOCKED) == 0) {
1788 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1790 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1793 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1795 /* authorize attribute setting as kernel */
1796 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1798 if ((ioflg & IO_NODELOCKED) == 0) {
1799 vn_finished_write(mp);
1807 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1808 const char *attrname, struct thread *td)
1813 if ((ioflg & IO_NODELOCKED) == 0) {
1814 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1816 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1819 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1821 /* authorize attribute removal as kernel */
1822 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1823 if (error == EOPNOTSUPP)
1824 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1827 if ((ioflg & IO_NODELOCKED) == 0) {
1828 vn_finished_write(mp);
1836 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1842 ltype = VOP_ISLOCKED(vp);
1843 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1844 ("vn_vget_ino: vp not locked"));
1845 error = vfs_busy(mp, MBF_NOWAIT);
1849 error = vfs_busy(mp, 0);
1850 vn_lock(vp, ltype | LK_RETRY);
1854 if (vp->v_iflag & VI_DOOMED) {
1860 error = VFS_VGET(mp, ino, lkflags, rvp);
1862 vn_lock(vp, ltype | LK_RETRY);
1863 if (vp->v_iflag & VI_DOOMED) {
1872 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
1873 const struct thread *td)
1876 if (vp->v_type != VREG || td == NULL)
1878 PROC_LOCK(td->td_proc);
1879 if ((uoff_t)uio->uio_offset + uio->uio_resid >
1880 lim_cur(td->td_proc, RLIMIT_FSIZE)) {
1881 kern_psignal(td->td_proc, SIGXFSZ);
1882 PROC_UNLOCK(td->td_proc);
1885 PROC_UNLOCK(td->td_proc);
1890 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
1894 int error, vfslocked;
1897 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1899 vn_lock(vp, LK_SHARED | LK_RETRY);
1900 AUDIT_ARG_VNODE1(vp);
1903 error = setfmode(td, active_cred, vp, mode);
1904 VFS_UNLOCK_GIANT(vfslocked);
1909 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
1913 int error, vfslocked;
1916 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1918 vn_lock(vp, LK_SHARED | LK_RETRY);
1919 AUDIT_ARG_VNODE1(vp);
1922 error = setfown(td, active_cred, vp, uid, gid);
1923 VFS_UNLOCK_GIANT(vfslocked);
1928 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
1932 if ((object = vp->v_object) == NULL)
1934 VM_OBJECT_LOCK(object);
1935 vm_object_page_remove(object, start, end, 0);
1936 VM_OBJECT_UNLOCK(object);
1940 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
1948 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
1949 ("Wrong command %lu", cmd));
1951 if (vn_lock(vp, LK_SHARED) != 0)
1953 if (vp->v_type != VREG) {
1957 error = VOP_GETATTR(vp, &va, cred);
1961 if (noff >= va.va_size) {
1965 bsize = vp->v_mount->mnt_stat.f_iosize;
1966 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
1967 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
1968 if (error == EOPNOTSUPP) {
1972 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
1973 (bnp != -1 && cmd == FIOSEEKDATA)) {
1980 if (noff > va.va_size)
1982 /* noff == va.va_size. There is an implicit hole at the end of file. */
1983 if (cmd == FIOSEEKDATA)