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 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/fcntl.h>
48 #include <sys/limits.h>
50 #include <sys/mount.h>
51 #include <sys/mutex.h>
52 #include <sys/namei.h>
53 #include <sys/vnode.h>
56 #include <sys/filio.h>
57 #include <sys/resourcevar.h>
59 #include <sys/sysctl.h>
60 #include <sys/ttycom.h>
62 #include <sys/syslog.h>
63 #include <sys/unistd.h>
65 #include <security/audit/audit.h>
66 #include <security/mac/mac_framework.h>
69 #include <vm/vm_extern.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_page.h>
75 static fo_rdwr_t vn_read;
76 static fo_rdwr_t vn_write;
77 static fo_rdwr_t vn_io_fault;
78 static fo_truncate_t vn_truncate;
79 static fo_ioctl_t vn_ioctl;
80 static fo_poll_t vn_poll;
81 static fo_kqfilter_t vn_kqfilter;
82 static fo_stat_t vn_statfile;
83 static fo_close_t vn_closefile;
85 struct fileops vnops = {
86 .fo_read = vn_io_fault,
87 .fo_write = vn_io_fault,
88 .fo_truncate = vn_truncate,
91 .fo_kqfilter = vn_kqfilter,
92 .fo_stat = vn_statfile,
93 .fo_close = vn_closefile,
96 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
100 vn_open(ndp, flagp, cmode, fp)
101 struct nameidata *ndp;
105 struct thread *td = ndp->ni_cnd.cn_thread;
107 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
111 * Common code for vnode open operations.
112 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
114 * Note that this does NOT free nameidata for the successful case,
115 * due to the NDINIT being done elsewhere.
118 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
119 struct ucred *cred, struct file *fp)
123 struct thread *td = ndp->ni_cnd.cn_thread;
125 struct vattr *vap = &vat;
128 int vfslocked, mpsafe;
130 mpsafe = ndp->ni_cnd.cn_flags & MPSAFE;
134 if (fmode & O_CREAT) {
135 ndp->ni_cnd.cn_nameiop = CREATE;
136 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF |
138 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
139 ndp->ni_cnd.cn_flags |= FOLLOW;
140 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
141 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
143 if ((error = namei(ndp)) != 0)
145 vfslocked = NDHASGIANT(ndp);
147 ndp->ni_cnd.cn_flags &= ~MPSAFE;
148 if (ndp->ni_vp == NULL) {
151 vap->va_mode = cmode;
153 vap->va_vaflags |= VA_EXCLUSIVE;
154 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
155 NDFREE(ndp, NDF_ONLY_PNBUF);
157 VFS_UNLOCK_GIANT(vfslocked);
158 if ((error = vn_start_write(NULL, &mp,
159 V_XSLEEP | PCATCH)) != 0)
164 error = mac_vnode_check_create(cred, ndp->ni_dvp,
168 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
171 vn_finished_write(mp);
173 VFS_UNLOCK_GIANT(vfslocked);
174 NDFREE(ndp, NDF_ONLY_PNBUF);
180 if (ndp->ni_dvp == ndp->ni_vp)
186 if (fmode & O_EXCL) {
193 ndp->ni_cnd.cn_nameiop = LOOKUP;
194 ndp->ni_cnd.cn_flags = ISOPEN |
195 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) |
197 if (!(fmode & FWRITE))
198 ndp->ni_cnd.cn_flags |= LOCKSHARED;
199 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
200 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
201 if ((error = namei(ndp)) != 0)
204 ndp->ni_cnd.cn_flags &= ~MPSAFE;
205 vfslocked = NDHASGIANT(ndp);
208 if (vp->v_type == VLNK) {
212 if (vp->v_type == VSOCK) {
216 if (vp->v_type != VDIR && fmode & O_DIRECTORY) {
221 if (fmode & (FWRITE | O_TRUNC)) {
222 if (vp->v_type == VDIR) {
232 if ((fmode & O_APPEND) && (fmode & FWRITE))
235 error = mac_vnode_check_open(cred, vp, accmode);
239 if ((fmode & O_CREAT) == 0) {
240 if (accmode & VWRITE) {
241 error = vn_writechk(vp);
246 error = VOP_ACCESS(vp, accmode, cred, td);
251 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
257 ASSERT_VOP_LOCKED(vp, "vn_open_cred");
259 VFS_UNLOCK_GIANT(vfslocked);
262 NDFREE(ndp, NDF_ONLY_PNBUF);
264 VFS_UNLOCK_GIANT(vfslocked);
271 * Check for write permissions on the specified vnode.
272 * Prototype text segments cannot be written.
276 register struct vnode *vp;
279 ASSERT_VOP_LOCKED(vp, "vn_writechk");
281 * If there's shared text associated with
282 * the vnode, try to free it up once. If
283 * we fail, we can't allow writing.
285 if (vp->v_vflag & VV_TEXT)
295 vn_close(vp, flags, file_cred, td)
296 register struct vnode *vp;
298 struct ucred *file_cred;
302 int error, lock_flags;
304 if (!(flags & FWRITE) && vp->v_mount != NULL &&
305 vp->v_mount->mnt_kern_flag & MNTK_EXTENDED_SHARED)
306 lock_flags = LK_SHARED;
308 lock_flags = LK_EXCLUSIVE;
310 VFS_ASSERT_GIANT(vp->v_mount);
312 vn_start_write(vp, &mp, V_WAIT);
313 vn_lock(vp, lock_flags | LK_RETRY);
314 if (flags & FWRITE) {
315 VNASSERT(vp->v_writecount > 0, vp,
316 ("vn_close: negative writecount"));
319 error = VOP_CLOSE(vp, flags, file_cred, td);
321 vn_finished_write(mp);
326 * Heuristic to detect sequential operation.
329 sequential_heuristic(struct uio *uio, struct file *fp)
332 if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
333 return (fp->f_seqcount << IO_SEQSHIFT);
336 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
337 * that the first I/O is normally considered to be slightly
338 * sequential. Seeking to offset 0 doesn't change sequentiality
339 * unless previous seeks have reduced f_seqcount to 0, in which
340 * case offset 0 is not special.
342 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
343 uio->uio_offset == fp->f_nextoff) {
345 * f_seqcount is in units of fixed-size blocks so that it
346 * depends mainly on the amount of sequential I/O and not
347 * much on the number of sequential I/O's. The fixed size
348 * of 16384 is hard-coded here since it is (not quite) just
349 * a magic size that works well here. This size is more
350 * closely related to the best I/O size for real disks than
351 * to any block size used by software.
353 fp->f_seqcount += howmany(uio->uio_resid, 16384);
354 if (fp->f_seqcount > IO_SEQMAX)
355 fp->f_seqcount = IO_SEQMAX;
356 return (fp->f_seqcount << IO_SEQSHIFT);
359 /* Not sequential. Quickly draw-down sequentiality. */
360 if (fp->f_seqcount > 1)
368 * Package up an I/O request on a vnode into a uio and do it.
371 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
372 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
373 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
380 int error, lock_flags;
382 VFS_ASSERT_GIANT(vp->v_mount);
384 auio.uio_iov = &aiov;
386 aiov.iov_base = base;
388 auio.uio_resid = len;
389 auio.uio_offset = offset;
390 auio.uio_segflg = segflg;
395 if ((ioflg & IO_NODELOCKED) == 0) {
396 if (rw == UIO_READ) {
397 rl_cookie = vn_rangelock_rlock(vp, offset,
400 rl_cookie = vn_rangelock_wlock(vp, offset,
404 if (rw == UIO_WRITE) {
405 if (vp->v_type != VCHR &&
406 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
409 if (MNT_SHARED_WRITES(mp) ||
410 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
411 lock_flags = LK_SHARED;
413 lock_flags = LK_EXCLUSIVE;
415 lock_flags = LK_SHARED;
416 vn_lock(vp, lock_flags | LK_RETRY);
420 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
422 if ((ioflg & IO_NOMACCHECK) == 0) {
424 error = mac_vnode_check_read(active_cred, file_cred,
427 error = mac_vnode_check_write(active_cred, file_cred,
432 if (file_cred != NULL)
437 error = VOP_READ(vp, &auio, ioflg, cred);
439 error = VOP_WRITE(vp, &auio, ioflg, cred);
442 *aresid = auio.uio_resid;
444 if (auio.uio_resid && error == 0)
446 if ((ioflg & IO_NODELOCKED) == 0) {
449 vn_finished_write(mp);
452 if (rl_cookie != NULL)
453 vn_rangelock_unlock(vp, rl_cookie);
458 * Package up an I/O request on a vnode into a uio and do it. The I/O
459 * request is split up into smaller chunks and we try to avoid saturating
460 * the buffer cache while potentially holding a vnode locked, so we
461 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
462 * to give other processes a chance to lock the vnode (either other processes
463 * core'ing the same binary, or unrelated processes scanning the directory).
466 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
467 file_cred, aresid, td)
475 struct ucred *active_cred;
476 struct ucred *file_cred;
483 VFS_ASSERT_GIANT(vp->v_mount);
489 * Force `offset' to a multiple of MAXBSIZE except possibly
490 * for the first chunk, so that filesystems only need to
491 * write full blocks except possibly for the first and last
494 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
498 if (rw != UIO_READ && vp->v_type == VREG)
501 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
502 ioflg, active_cred, file_cred, &iaresid, td);
503 len -= chunk; /* aresid calc already includes length */
507 base = (char *)base + chunk;
508 kern_yield(PRI_USER);
511 *aresid = len + iaresid;
516 * File table vnode read routine.
519 vn_read(fp, uio, active_cred, flags, td)
522 struct ucred *active_cred;
529 int advice, vfslocked;
530 off_t offset, start, end;
532 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
537 if (fp->f_flag & FNONBLOCK)
539 if (fp->f_flag & O_DIRECT)
541 advice = POSIX_FADV_NORMAL;
542 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
544 * According to McKusick the vn lock was protecting f_offset here.
545 * It is now protected by the FOFFSET_LOCKED flag.
547 if ((flags & FOF_OFFSET) == 0 || fp->f_advice != NULL) {
548 mtxp = mtx_pool_find(mtxpool_sleep, fp);
550 if ((flags & FOF_OFFSET) == 0) {
551 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
552 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
553 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
554 "vnread offlock", 0);
556 fp->f_vnread_flags |= FOFFSET_LOCKED;
557 uio->uio_offset = fp->f_offset;
559 if (fp->f_advice != NULL &&
560 uio->uio_offset >= fp->f_advice->fa_start &&
561 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
562 advice = fp->f_advice->fa_advice;
565 vn_lock(vp, LK_SHARED | LK_RETRY);
568 case POSIX_FADV_NORMAL:
569 case POSIX_FADV_SEQUENTIAL:
570 case POSIX_FADV_NOREUSE:
571 ioflag |= sequential_heuristic(uio, fp);
573 case POSIX_FADV_RANDOM:
574 /* Disable read-ahead for random I/O. */
577 offset = uio->uio_offset;
580 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
583 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
584 if ((flags & FOF_OFFSET) == 0) {
585 fp->f_offset = uio->uio_offset;
587 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
588 wakeup(&fp->f_vnread_flags);
589 fp->f_vnread_flags = 0;
592 fp->f_nextoff = uio->uio_offset;
594 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
595 offset != uio->uio_offset) {
597 * Use POSIX_FADV_DONTNEED to flush clean pages and
598 * buffers for the backing file after a
599 * POSIX_FADV_NOREUSE read(2). To optimize the common
600 * case of using POSIX_FADV_NOREUSE with sequential
601 * access, track the previous implicit DONTNEED
602 * request and grow this request to include the
603 * current read(2) in addition to the previous
604 * DONTNEED. With purely sequential access this will
605 * cause the DONTNEED requests to continously grow to
606 * cover all of the previously read regions of the
607 * file. This allows filesystem blocks that are
608 * accessed by multiple calls to read(2) to be flushed
609 * once the last read(2) finishes.
612 end = uio->uio_offset - 1;
614 if (fp->f_advice != NULL &&
615 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
616 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
617 start = fp->f_advice->fa_prevstart;
618 else if (fp->f_advice->fa_prevstart != 0 &&
619 fp->f_advice->fa_prevstart == end + 1)
620 end = fp->f_advice->fa_prevend;
621 fp->f_advice->fa_prevstart = start;
622 fp->f_advice->fa_prevend = end;
625 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
627 VFS_UNLOCK_GIANT(vfslocked);
632 * File table vnode write routine.
635 vn_write(fp, uio, active_cred, flags, td)
638 struct ucred *active_cred;
644 int error, ioflag, lock_flags;
646 int advice, vfslocked;
647 off_t offset, start, end;
649 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
652 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
653 if (vp->v_type == VREG)
656 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
658 if (fp->f_flag & FNONBLOCK)
660 if (fp->f_flag & O_DIRECT)
662 if ((fp->f_flag & O_FSYNC) ||
663 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
666 if (vp->v_type != VCHR &&
667 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
670 if ((MNT_SHARED_WRITES(mp) ||
671 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) &&
672 (flags & FOF_OFFSET) != 0) {
673 lock_flags = LK_SHARED;
675 lock_flags = LK_EXCLUSIVE;
678 vn_lock(vp, lock_flags | LK_RETRY);
679 if ((flags & FOF_OFFSET) == 0)
680 uio->uio_offset = fp->f_offset;
681 advice = POSIX_FADV_NORMAL;
683 if (fp->f_advice != NULL) {
684 mtxp = mtx_pool_find(mtxpool_sleep, fp);
686 if (fp->f_advice != NULL &&
687 uio->uio_offset >= fp->f_advice->fa_start &&
688 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
689 advice = fp->f_advice->fa_advice;
693 case POSIX_FADV_NORMAL:
694 case POSIX_FADV_SEQUENTIAL:
695 case POSIX_FADV_NOREUSE:
696 ioflag |= sequential_heuristic(uio, fp);
698 case POSIX_FADV_RANDOM:
699 /* XXX: Is this correct? */
702 offset = uio->uio_offset;
705 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
708 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
709 if ((flags & FOF_OFFSET) == 0)
710 fp->f_offset = uio->uio_offset;
711 fp->f_nextoff = uio->uio_offset;
713 if (vp->v_type != VCHR)
714 vn_finished_write(mp);
715 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
716 offset != uio->uio_offset) {
718 * Use POSIX_FADV_DONTNEED to flush clean pages and
719 * buffers for the backing file after a
720 * POSIX_FADV_NOREUSE write(2). To optimize the
721 * common case of using POSIX_FADV_NOREUSE with
722 * sequential access, track the previous implicit
723 * DONTNEED request and grow this request to include
724 * the current write(2) in addition to the previous
725 * DONTNEED. With purely sequential access this will
726 * cause the DONTNEED requests to continously grow to
727 * cover all of the previously written regions of the
730 * Note that the blocks just written are almost
731 * certainly still dirty, so this only works when
732 * VOP_ADVISE() calls from subsequent writes push out
733 * the data written by this write(2) once the backing
734 * buffers are clean. However, as compared to forcing
735 * IO_DIRECT, this gives much saner behavior. Write
736 * clustering is still allowed, and clean pages are
737 * merely moved to the cache page queue rather than
738 * outright thrown away. This means a subsequent
739 * read(2) can still avoid hitting the disk if the
740 * pages have not been reclaimed.
742 * This does make POSIX_FADV_NOREUSE largely useless
743 * with non-sequential access. However, sequential
744 * access is the more common use case and the flag is
748 end = uio->uio_offset - 1;
750 if (fp->f_advice != NULL &&
751 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
752 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
753 start = fp->f_advice->fa_prevstart;
754 else if (fp->f_advice->fa_prevstart != 0 &&
755 fp->f_advice->fa_prevstart == end + 1)
756 end = fp->f_advice->fa_prevend;
757 fp->f_advice->fa_prevstart = start;
758 fp->f_advice->fa_prevend = end;
761 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
765 VFS_UNLOCK_GIANT(vfslocked);
769 static const int io_hold_cnt = 16;
770 static int vn_io_fault_enable = 0;
771 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
772 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
773 static unsigned long vn_io_faults_cnt;
774 SYSCTL_LONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
775 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
778 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
779 * prevent the following deadlock:
781 * Assume that the thread A reads from the vnode vp1 into userspace
782 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
783 * currently not resident, then system ends up with the call chain
784 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
785 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
786 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
787 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
788 * backed by the pages of vnode vp1, and some page in buf2 is not
789 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
791 * To prevent the lock order reversal and deadlock, vn_io_fault() does
792 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
793 * Instead, it first tries to do the whole range i/o with pagefaults
794 * disabled. If all pages in the i/o buffer are resident and mapped,
795 * VOP will succeed (ignoring the genuine filesystem errors).
796 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
797 * i/o in chunks, with all pages in the chunk prefaulted and held
798 * using vm_fault_quick_hold_pages().
800 * Filesystems using this deadlock avoidance scheme should use the
801 * array of the held pages from uio, saved in the curthread->td_ma,
802 * instead of doing uiomove(). A helper function
803 * vn_io_fault_uiomove() converts uiomove request into
804 * uiomove_fromphys() over td_ma array.
806 * Since vnode locks do not cover the whole i/o anymore, rangelocks
807 * make the current i/o request atomic with respect to other i/os and
811 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
812 int flags, struct thread *td)
814 vm_page_t ma[io_hold_cnt + 2];
815 struct uio *uio_clone, short_uio;
816 struct iovec short_iovec[1];
821 vm_page_t *prev_td_ma;
822 int cnt, error, save, saveheld, prev_td_ma_cnt;
823 vm_offset_t addr, end;
828 if (uio->uio_rw == UIO_READ)
833 if (uio->uio_segflg != UIO_USERSPACE || vp->v_type != VREG ||
834 ((mp = vp->v_mount) != NULL &&
835 (mp->mnt_kern_flag & MNTK_NO_IOPF) == 0) ||
837 return (doio(fp, uio, active_cred, flags, td));
840 * The UFS follows IO_UNIT directive and replays back both
841 * uio_offset and uio_resid if an error is encountered during the
842 * operation. But, since the iovec may be already advanced,
843 * uio is still in an inconsistent state.
845 * Cache a copy of the original uio, which is advanced to the redo
846 * point using UIO_NOCOPY below.
848 uio_clone = cloneuio(uio);
849 resid = uio->uio_resid;
851 short_uio.uio_segflg = UIO_USERSPACE;
852 short_uio.uio_rw = uio->uio_rw;
853 short_uio.uio_td = uio->uio_td;
855 if (uio->uio_rw == UIO_READ) {
856 prot = VM_PROT_WRITE;
857 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
858 uio->uio_offset + uio->uio_resid);
861 if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0)
862 /* For appenders, punt and lock the whole range. */
863 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
865 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
866 uio->uio_offset + uio->uio_resid);
869 save = vm_fault_disable_pagefaults();
870 error = doio(fp, uio, active_cred, flags, td);
874 atomic_add_long(&vn_io_faults_cnt, 1);
875 uio_clone->uio_segflg = UIO_NOCOPY;
876 uiomove(NULL, resid - uio->uio_resid, uio_clone);
877 uio_clone->uio_segflg = uio->uio_segflg;
879 saveheld = curthread_pflags_set(TDP_UIOHELD);
880 prev_td_ma = td->td_ma;
881 prev_td_ma_cnt = td->td_ma_cnt;
883 while (uio_clone->uio_resid != 0) {
884 len = uio_clone->uio_iov->iov_len;
886 KASSERT(uio_clone->uio_iovcnt >= 1,
887 ("iovcnt underflow"));
888 uio_clone->uio_iov++;
889 uio_clone->uio_iovcnt--;
893 addr = (vm_offset_t)uio_clone->uio_iov->iov_base;
894 end = round_page(addr + len);
895 cnt = howmany(end - trunc_page(addr), PAGE_SIZE);
897 * A perfectly misaligned address and length could cause
898 * both the start and the end of the chunk to use partial
899 * page. +2 accounts for such a situation.
901 if (cnt > io_hold_cnt + 2) {
902 len = io_hold_cnt * PAGE_SIZE;
903 KASSERT(howmany(round_page(addr + len) -
904 trunc_page(addr), PAGE_SIZE) <= io_hold_cnt + 2,
907 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
908 addr, len, prot, ma, io_hold_cnt + 2);
913 short_uio.uio_iov = &short_iovec[0];
914 short_iovec[0].iov_base = (void *)addr;
915 short_uio.uio_iovcnt = 1;
916 short_uio.uio_resid = short_iovec[0].iov_len = len;
917 short_uio.uio_offset = uio_clone->uio_offset;
921 error = doio(fp, &short_uio, active_cred, flags, td);
922 vm_page_unhold_pages(ma, cnt);
923 adv = len - short_uio.uio_resid;
925 uio_clone->uio_iov->iov_base =
926 (char *)uio_clone->uio_iov->iov_base + adv;
927 uio_clone->uio_iov->iov_len -= adv;
928 uio_clone->uio_resid -= adv;
929 uio_clone->uio_offset += adv;
931 uio->uio_resid -= adv;
932 uio->uio_offset += adv;
934 if (error != 0 || adv == 0)
937 td->td_ma = prev_td_ma;
938 td->td_ma_cnt = prev_td_ma_cnt;
939 curthread_pflags_restore(saveheld);
941 vm_fault_enable_pagefaults(save);
942 vn_rangelock_unlock(vp, rl_cookie);
943 free(uio_clone, M_IOV);
948 * Helper function to perform the requested uiomove operation using
949 * the held pages for io->uio_iov[0].iov_base buffer instead of
950 * copyin/copyout. Access to the pages with uiomove_fromphys()
951 * instead of iov_base prevents page faults that could occur due to
952 * pmap_collect() invalidating the mapping created by
953 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
954 * object cleanup revoking the write access from page mappings.
956 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
957 * instead of plain uiomove().
960 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
962 struct uio transp_uio;
963 struct iovec transp_iov[1];
969 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
970 uio->uio_segflg != UIO_USERSPACE)
971 return (uiomove(data, xfersize, uio));
973 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
974 transp_iov[0].iov_base = data;
975 transp_uio.uio_iov = &transp_iov[0];
976 transp_uio.uio_iovcnt = 1;
977 if (xfersize > uio->uio_resid)
978 xfersize = uio->uio_resid;
979 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
980 transp_uio.uio_offset = 0;
981 transp_uio.uio_segflg = UIO_SYSSPACE;
983 * Since transp_iov points to data, and td_ma page array
984 * corresponds to original uio->uio_iov, we need to invert the
985 * direction of the i/o operation as passed to
986 * uiomove_fromphys().
988 switch (uio->uio_rw) {
990 transp_uio.uio_rw = UIO_READ;
993 transp_uio.uio_rw = UIO_WRITE;
996 transp_uio.uio_td = uio->uio_td;
997 error = uiomove_fromphys(td->td_ma,
998 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
999 xfersize, &transp_uio);
1000 adv = xfersize - transp_uio.uio_resid;
1002 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1003 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1005 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1007 td->td_ma_cnt -= pgadv;
1008 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1009 uio->uio_iov->iov_len -= adv;
1010 uio->uio_resid -= adv;
1011 uio->uio_offset += adv;
1016 * File table truncate routine.
1019 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1032 * Lock the whole range for truncation. Otherwise split i/o
1033 * might happen partly before and partly after the truncation.
1035 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1036 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1037 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1040 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1041 if (vp->v_type == VDIR) {
1046 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1050 error = vn_writechk(vp);
1053 vattr.va_size = length;
1054 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1058 vn_finished_write(mp);
1060 VFS_UNLOCK_GIANT(vfslocked);
1061 vn_rangelock_unlock(vp, rl_cookie);
1066 * File table vnode stat routine.
1069 vn_statfile(fp, sb, active_cred, td)
1072 struct ucred *active_cred;
1075 struct vnode *vp = fp->f_vnode;
1079 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1080 vn_lock(vp, LK_SHARED | LK_RETRY);
1081 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1083 VFS_UNLOCK_GIANT(vfslocked);
1089 * Stat a vnode; implementation for the stat syscall
1092 vn_stat(vp, sb, active_cred, file_cred, td)
1094 register struct stat *sb;
1095 struct ucred *active_cred;
1096 struct ucred *file_cred;
1100 register struct vattr *vap;
1105 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1113 * Initialize defaults for new and unusual fields, so that file
1114 * systems which don't support these fields don't need to know
1117 vap->va_birthtime.tv_sec = -1;
1118 vap->va_birthtime.tv_nsec = 0;
1119 vap->va_fsid = VNOVAL;
1120 vap->va_rdev = NODEV;
1122 error = VOP_GETATTR(vp, vap, active_cred);
1127 * Zero the spare stat fields
1129 bzero(sb, sizeof *sb);
1132 * Copy from vattr table
1134 if (vap->va_fsid != VNOVAL)
1135 sb->st_dev = vap->va_fsid;
1137 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1138 sb->st_ino = vap->va_fileid;
1139 mode = vap->va_mode;
1140 switch (vap->va_type) {
1166 sb->st_nlink = vap->va_nlink;
1167 sb->st_uid = vap->va_uid;
1168 sb->st_gid = vap->va_gid;
1169 sb->st_rdev = vap->va_rdev;
1170 if (vap->va_size > OFF_MAX)
1172 sb->st_size = vap->va_size;
1173 sb->st_atim = vap->va_atime;
1174 sb->st_mtim = vap->va_mtime;
1175 sb->st_ctim = vap->va_ctime;
1176 sb->st_birthtim = vap->va_birthtime;
1179 * According to www.opengroup.org, the meaning of st_blksize is
1180 * "a filesystem-specific preferred I/O block size for this
1181 * object. In some filesystem types, this may vary from file
1183 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1186 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1188 sb->st_flags = vap->va_flags;
1189 if (priv_check(td, PRIV_VFS_GENERATION))
1192 sb->st_gen = vap->va_gen;
1194 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1199 * File table vnode ioctl routine.
1202 vn_ioctl(fp, com, data, active_cred, td)
1206 struct ucred *active_cred;
1209 struct vnode *vp = fp->f_vnode;
1214 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1216 switch (vp->v_type) {
1219 if (com == FIONREAD) {
1220 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1221 error = VOP_GETATTR(vp, &vattr, active_cred);
1224 *(int *)data = vattr.va_size - fp->f_offset;
1226 if (com == FIONBIO || com == FIOASYNC) /* XXX */
1229 error = VOP_IOCTL(vp, com, data, fp->f_flag,
1236 VFS_UNLOCK_GIANT(vfslocked);
1241 * File table vnode poll routine.
1244 vn_poll(fp, events, active_cred, td)
1247 struct ucred *active_cred;
1255 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1257 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1258 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1263 error = VOP_POLL(vp, events, fp->f_cred, td);
1264 VFS_UNLOCK_GIANT(vfslocked);
1269 * Acquire the requested lock and then check for validity. LK_RETRY
1270 * permits vn_lock to return doomed vnodes.
1273 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1277 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1278 ("vn_lock called with no locktype."));
1280 #ifdef DEBUG_VFS_LOCKS
1281 KASSERT(vp->v_holdcnt != 0,
1282 ("vn_lock %p: zero hold count", vp));
1284 error = VOP_LOCK1(vp, flags, file, line);
1285 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1286 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1287 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1290 * Callers specify LK_RETRY if they wish to get dead vnodes.
1291 * If RETRY is not set, we return ENOENT instead.
1293 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1294 (flags & LK_RETRY) == 0) {
1299 } while (flags & LK_RETRY && error != 0);
1304 * File table vnode close routine.
1307 vn_closefile(fp, td)
1318 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1319 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1320 lf.l_whence = SEEK_SET;
1323 lf.l_type = F_UNLCK;
1324 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1327 fp->f_ops = &badfileops;
1329 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1330 VFS_UNLOCK_GIANT(vfslocked);
1335 * Preparing to start a filesystem write operation. If the operation is
1336 * permitted, then we bump the count of operations in progress and
1337 * proceed. If a suspend request is in progress, we wait until the
1338 * suspension is over, and then proceed.
1341 vn_start_write(vp, mpp, flags)
1351 * If a vnode is provided, get and return the mount point that
1352 * to which it will write.
1355 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1357 if (error != EOPNOTSUPP)
1362 if ((mp = *mpp) == NULL)
1366 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1368 * As long as a vnode is not provided we need to acquire a
1369 * refcount for the provided mountpoint too, in order to
1370 * emulate a vfs_ref().
1377 * Check on status of suspension.
1379 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1380 mp->mnt_susp_owner != curthread) {
1381 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1382 if (flags & V_NOWAIT) {
1383 error = EWOULDBLOCK;
1386 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1387 (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1392 if (flags & V_XSLEEP)
1394 mp->mnt_writeopcount++;
1396 if (error != 0 || (flags & V_XSLEEP) != 0)
1403 * Secondary suspension. Used by operations such as vop_inactive
1404 * routines that are needed by the higher level functions. These
1405 * are allowed to proceed until all the higher level functions have
1406 * completed (indicated by mnt_writeopcount dropping to zero). At that
1407 * time, these operations are halted until the suspension is over.
1410 vn_start_secondary_write(vp, mpp, flags)
1420 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1422 if (error != EOPNOTSUPP)
1428 * If we are not suspended or have not yet reached suspended
1429 * mode, then let the operation proceed.
1431 if ((mp = *mpp) == NULL)
1435 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1437 * As long as a vnode is not provided we need to acquire a
1438 * refcount for the provided mountpoint too, in order to
1439 * emulate a vfs_ref().
1444 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1445 mp->mnt_secondary_writes++;
1446 mp->mnt_secondary_accwrites++;
1450 if (flags & V_NOWAIT) {
1453 return (EWOULDBLOCK);
1456 * Wait for the suspension to finish.
1458 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1459 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1467 * Filesystem write operation has completed. If we are suspending and this
1468 * operation is the last one, notify the suspender that the suspension is
1472 vn_finished_write(mp)
1479 mp->mnt_writeopcount--;
1480 if (mp->mnt_writeopcount < 0)
1481 panic("vn_finished_write: neg cnt");
1482 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1483 mp->mnt_writeopcount <= 0)
1484 wakeup(&mp->mnt_writeopcount);
1490 * Filesystem secondary write operation has completed. If we are
1491 * suspending and this operation is the last one, notify the suspender
1492 * that the suspension is now in effect.
1495 vn_finished_secondary_write(mp)
1502 mp->mnt_secondary_writes--;
1503 if (mp->mnt_secondary_writes < 0)
1504 panic("vn_finished_secondary_write: neg cnt");
1505 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1506 mp->mnt_secondary_writes <= 0)
1507 wakeup(&mp->mnt_secondary_writes);
1514 * Request a filesystem to suspend write operations.
1517 vfs_write_suspend(mp)
1523 if (mp->mnt_susp_owner == curthread) {
1527 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1528 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1529 mp->mnt_kern_flag |= MNTK_SUSPEND;
1530 mp->mnt_susp_owner = curthread;
1531 if (mp->mnt_writeopcount > 0)
1532 (void) msleep(&mp->mnt_writeopcount,
1533 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1536 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1537 vfs_write_resume(mp);
1542 * Request a filesystem to resume write operations.
1545 vfs_write_resume(mp)
1550 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1551 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1552 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1554 mp->mnt_susp_owner = NULL;
1555 wakeup(&mp->mnt_writeopcount);
1556 wakeup(&mp->mnt_flag);
1557 curthread->td_pflags &= ~TDP_IGNSUSP;
1565 * Implement kqueues for files by translating it to vnode operation.
1568 vn_kqfilter(struct file *fp, struct knote *kn)
1573 vfslocked = VFS_LOCK_GIANT(fp->f_vnode->v_mount);
1574 error = VOP_KQFILTER(fp->f_vnode, kn);
1575 VFS_UNLOCK_GIANT(vfslocked);
1581 * Simplified in-kernel wrapper calls for extended attribute access.
1582 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1583 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1586 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1587 const char *attrname, int *buflen, char *buf, struct thread *td)
1593 iov.iov_len = *buflen;
1596 auio.uio_iov = &iov;
1597 auio.uio_iovcnt = 1;
1598 auio.uio_rw = UIO_READ;
1599 auio.uio_segflg = UIO_SYSSPACE;
1601 auio.uio_offset = 0;
1602 auio.uio_resid = *buflen;
1604 if ((ioflg & IO_NODELOCKED) == 0)
1605 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1607 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1609 /* authorize attribute retrieval as kernel */
1610 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1613 if ((ioflg & IO_NODELOCKED) == 0)
1617 *buflen = *buflen - auio.uio_resid;
1624 * XXX failure mode if partially written?
1627 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1628 const char *attrname, int buflen, char *buf, struct thread *td)
1635 iov.iov_len = buflen;
1638 auio.uio_iov = &iov;
1639 auio.uio_iovcnt = 1;
1640 auio.uio_rw = UIO_WRITE;
1641 auio.uio_segflg = UIO_SYSSPACE;
1643 auio.uio_offset = 0;
1644 auio.uio_resid = buflen;
1646 if ((ioflg & IO_NODELOCKED) == 0) {
1647 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1649 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1652 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1654 /* authorize attribute setting as kernel */
1655 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1657 if ((ioflg & IO_NODELOCKED) == 0) {
1658 vn_finished_write(mp);
1666 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1667 const char *attrname, struct thread *td)
1672 if ((ioflg & IO_NODELOCKED) == 0) {
1673 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1675 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1678 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1680 /* authorize attribute removal as kernel */
1681 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1682 if (error == EOPNOTSUPP)
1683 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1686 if ((ioflg & IO_NODELOCKED) == 0) {
1687 vn_finished_write(mp);
1695 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1701 ltype = VOP_ISLOCKED(vp);
1702 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1703 ("vn_vget_ino: vp not locked"));
1704 error = vfs_busy(mp, MBF_NOWAIT);
1708 error = vfs_busy(mp, 0);
1709 vn_lock(vp, ltype | LK_RETRY);
1713 if (vp->v_iflag & VI_DOOMED) {
1719 error = VFS_VGET(mp, ino, lkflags, rvp);
1721 vn_lock(vp, ltype | LK_RETRY);
1722 if (vp->v_iflag & VI_DOOMED) {
1731 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
1732 const struct thread *td)
1735 if (vp->v_type != VREG || td == NULL)
1737 PROC_LOCK(td->td_proc);
1738 if ((uoff_t)uio->uio_offset + uio->uio_resid >
1739 lim_cur(td->td_proc, RLIMIT_FSIZE)) {
1740 kern_psignal(td->td_proc, SIGXFSZ);
1741 PROC_UNLOCK(td->td_proc);
1744 PROC_UNLOCK(td->td_proc);
1749 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
1753 int error, vfslocked;
1756 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1758 vn_lock(vp, LK_SHARED | LK_RETRY);
1759 AUDIT_ARG_VNODE1(vp);
1762 error = setfmode(td, active_cred, vp, mode);
1763 VFS_UNLOCK_GIANT(vfslocked);
1768 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
1772 int error, vfslocked;
1775 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1777 vn_lock(vp, LK_SHARED | LK_RETRY);
1778 AUDIT_ARG_VNODE1(vp);
1781 error = setfown(td, active_cred, vp, uid, gid);
1782 VFS_UNLOCK_GIANT(vfslocked);
1787 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
1791 if ((object = vp->v_object) == NULL)
1793 VM_OBJECT_LOCK(object);
1794 vm_object_page_remove(object, start, end, 0);
1795 VM_OBJECT_UNLOCK(object);
1799 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
1807 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
1808 ("Wrong command %lu", cmd));
1810 if (vn_lock(vp, LK_SHARED) != 0)
1812 if (vp->v_type != VREG) {
1816 error = VOP_GETATTR(vp, &va, cred);
1820 if (noff >= va.va_size) {
1824 bsize = vp->v_mount->mnt_stat.f_iosize;
1825 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
1826 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
1827 if (error == EOPNOTSUPP) {
1831 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
1832 (bnp != -1 && cmd == FIOSEEKDATA)) {
1839 if (noff > va.va_size)
1841 /* noff == va.va_size. There is an implicit hole at the end of file. */
1842 if (cmd == FIOSEEKDATA)