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)
255 VOP_ADD_WRITECOUNT(vp, 1);
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.
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"));
317 VOP_ADD_WRITECOUNT(vp, -1);
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 foffset_lock(struct file *fp, int flags)
521 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
523 #if OFF_MAX <= LONG_MAX
525 * Caller only wants the current f_offset value. Assume that
526 * the long and shorter integer types reads are atomic.
528 if ((flags & FOF_NOLOCK) != 0)
529 return (fp->f_offset);
533 * According to McKusick the vn lock was protecting f_offset here.
534 * It is now protected by the FOFFSET_LOCKED flag.
536 mtxp = mtx_pool_find(mtxpool_sleep, fp);
538 if ((flags & FOF_NOLOCK) == 0) {
539 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
540 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
541 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
544 fp->f_vnread_flags |= FOFFSET_LOCKED;
552 foffset_unlock(struct file *fp, off_t val, int flags)
556 KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed"));
558 #if OFF_MAX <= LONG_MAX
559 if ((flags & FOF_NOLOCK) != 0) {
560 if ((flags & FOF_NOUPDATE) == 0)
562 if ((flags & FOF_NEXTOFF) != 0)
568 mtxp = mtx_pool_find(mtxpool_sleep, fp);
570 if ((flags & FOF_NOUPDATE) == 0)
572 if ((flags & FOF_NEXTOFF) != 0)
574 if ((flags & FOF_NOLOCK) == 0) {
575 KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0,
576 ("Lost FOFFSET_LOCKED"));
577 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
578 wakeup(&fp->f_vnread_flags);
579 fp->f_vnread_flags = 0;
585 foffset_lock_uio(struct file *fp, struct uio *uio, int flags)
588 if ((flags & FOF_OFFSET) == 0)
589 uio->uio_offset = foffset_lock(fp, flags);
593 foffset_unlock_uio(struct file *fp, struct uio *uio, int flags)
596 if ((flags & FOF_OFFSET) == 0)
597 foffset_unlock(fp, uio->uio_offset, flags);
601 get_advice(struct file *fp, struct uio *uio)
606 ret = POSIX_FADV_NORMAL;
607 if (fp->f_advice == NULL)
610 mtxp = mtx_pool_find(mtxpool_sleep, fp);
612 if (uio->uio_offset >= fp->f_advice->fa_start &&
613 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
614 ret = fp->f_advice->fa_advice;
620 * File table vnode read routine.
623 vn_read(fp, uio, active_cred, flags, td)
626 struct ucred *active_cred;
633 int advice, vfslocked;
634 off_t offset, start, end;
636 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
638 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
641 if (fp->f_flag & FNONBLOCK)
643 if (fp->f_flag & O_DIRECT)
645 advice = get_advice(fp, uio);
646 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
647 vn_lock(vp, LK_SHARED | LK_RETRY);
650 case POSIX_FADV_NORMAL:
651 case POSIX_FADV_SEQUENTIAL:
652 case POSIX_FADV_NOREUSE:
653 ioflag |= sequential_heuristic(uio, fp);
655 case POSIX_FADV_RANDOM:
656 /* Disable read-ahead for random I/O. */
659 offset = uio->uio_offset;
662 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
665 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
666 fp->f_nextoff = uio->uio_offset;
668 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
669 offset != uio->uio_offset) {
671 * Use POSIX_FADV_DONTNEED to flush clean pages and
672 * buffers for the backing file after a
673 * POSIX_FADV_NOREUSE read(2). To optimize the common
674 * case of using POSIX_FADV_NOREUSE with sequential
675 * access, track the previous implicit DONTNEED
676 * request and grow this request to include the
677 * current read(2) in addition to the previous
678 * DONTNEED. With purely sequential access this will
679 * cause the DONTNEED requests to continously grow to
680 * cover all of the previously read regions of the
681 * file. This allows filesystem blocks that are
682 * accessed by multiple calls to read(2) to be flushed
683 * once the last read(2) finishes.
686 end = uio->uio_offset - 1;
687 mtxp = mtx_pool_find(mtxpool_sleep, fp);
689 if (fp->f_advice != NULL &&
690 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
691 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
692 start = fp->f_advice->fa_prevstart;
693 else if (fp->f_advice->fa_prevstart != 0 &&
694 fp->f_advice->fa_prevstart == end + 1)
695 end = fp->f_advice->fa_prevend;
696 fp->f_advice->fa_prevstart = start;
697 fp->f_advice->fa_prevend = end;
700 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
702 VFS_UNLOCK_GIANT(vfslocked);
707 * File table vnode write routine.
710 vn_write(fp, uio, active_cred, flags, td)
713 struct ucred *active_cred;
720 int error, ioflag, lock_flags;
721 int advice, vfslocked;
722 off_t offset, start, end;
724 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
726 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
728 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
729 if (vp->v_type == VREG)
732 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
734 if (fp->f_flag & FNONBLOCK)
736 if (fp->f_flag & O_DIRECT)
738 if ((fp->f_flag & O_FSYNC) ||
739 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
742 if (vp->v_type != VCHR &&
743 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
746 advice = get_advice(fp, uio);
748 if ((MNT_SHARED_WRITES(mp) ||
749 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) &&
750 (flags & FOF_OFFSET) != 0) {
751 lock_flags = LK_SHARED;
753 lock_flags = LK_EXCLUSIVE;
756 vn_lock(vp, lock_flags | LK_RETRY);
758 case POSIX_FADV_NORMAL:
759 case POSIX_FADV_SEQUENTIAL:
760 case POSIX_FADV_NOREUSE:
761 ioflag |= sequential_heuristic(uio, fp);
763 case POSIX_FADV_RANDOM:
764 /* XXX: Is this correct? */
767 offset = uio->uio_offset;
770 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
773 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
774 fp->f_nextoff = uio->uio_offset;
776 if (vp->v_type != VCHR)
777 vn_finished_write(mp);
778 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
779 offset != uio->uio_offset) {
781 * Use POSIX_FADV_DONTNEED to flush clean pages and
782 * buffers for the backing file after a
783 * POSIX_FADV_NOREUSE write(2). To optimize the
784 * common case of using POSIX_FADV_NOREUSE with
785 * sequential access, track the previous implicit
786 * DONTNEED request and grow this request to include
787 * the current write(2) in addition to the previous
788 * DONTNEED. With purely sequential access this will
789 * cause the DONTNEED requests to continously grow to
790 * cover all of the previously written regions of the
793 * Note that the blocks just written are almost
794 * certainly still dirty, so this only works when
795 * VOP_ADVISE() calls from subsequent writes push out
796 * the data written by this write(2) once the backing
797 * buffers are clean. However, as compared to forcing
798 * IO_DIRECT, this gives much saner behavior. Write
799 * clustering is still allowed, and clean pages are
800 * merely moved to the cache page queue rather than
801 * outright thrown away. This means a subsequent
802 * read(2) can still avoid hitting the disk if the
803 * pages have not been reclaimed.
805 * This does make POSIX_FADV_NOREUSE largely useless
806 * with non-sequential access. However, sequential
807 * access is the more common use case and the flag is
811 end = uio->uio_offset - 1;
812 mtxp = mtx_pool_find(mtxpool_sleep, fp);
814 if (fp->f_advice != NULL &&
815 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
816 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
817 start = fp->f_advice->fa_prevstart;
818 else if (fp->f_advice->fa_prevstart != 0 &&
819 fp->f_advice->fa_prevstart == end + 1)
820 end = fp->f_advice->fa_prevend;
821 fp->f_advice->fa_prevstart = start;
822 fp->f_advice->fa_prevend = end;
825 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
829 VFS_UNLOCK_GIANT(vfslocked);
833 static const int io_hold_cnt = 16;
834 static int vn_io_fault_enable = 0;
835 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
836 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
837 static u_long vn_io_faults_cnt;
838 SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
839 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
842 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
843 * prevent the following deadlock:
845 * Assume that the thread A reads from the vnode vp1 into userspace
846 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
847 * currently not resident, then system ends up with the call chain
848 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
849 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
850 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
851 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
852 * backed by the pages of vnode vp1, and some page in buf2 is not
853 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
855 * To prevent the lock order reversal and deadlock, vn_io_fault() does
856 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
857 * Instead, it first tries to do the whole range i/o with pagefaults
858 * disabled. If all pages in the i/o buffer are resident and mapped,
859 * VOP will succeed (ignoring the genuine filesystem errors).
860 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
861 * i/o in chunks, with all pages in the chunk prefaulted and held
862 * using vm_fault_quick_hold_pages().
864 * Filesystems using this deadlock avoidance scheme should use the
865 * array of the held pages from uio, saved in the curthread->td_ma,
866 * instead of doing uiomove(). A helper function
867 * vn_io_fault_uiomove() converts uiomove request into
868 * uiomove_fromphys() over td_ma array.
870 * Since vnode locks do not cover the whole i/o anymore, rangelocks
871 * make the current i/o request atomic with respect to other i/os and
875 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
876 int flags, struct thread *td)
878 vm_page_t ma[io_hold_cnt + 2];
879 struct uio *uio_clone, short_uio;
880 struct iovec short_iovec[1];
885 vm_page_t *prev_td_ma;
886 int cnt, error, save, saveheld, prev_td_ma_cnt;
887 vm_offset_t addr, end;
892 if (uio->uio_rw == UIO_READ)
897 foffset_lock_uio(fp, uio, flags);
899 if (uio->uio_segflg != UIO_USERSPACE || vp->v_type != VREG ||
900 ((mp = vp->v_mount) != NULL &&
901 (mp->mnt_kern_flag & MNTK_NO_IOPF) == 0) ||
902 !vn_io_fault_enable) {
903 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
908 * The UFS follows IO_UNIT directive and replays back both
909 * uio_offset and uio_resid if an error is encountered during the
910 * operation. But, since the iovec may be already advanced,
911 * uio is still in an inconsistent state.
913 * Cache a copy of the original uio, which is advanced to the redo
914 * point using UIO_NOCOPY below.
916 uio_clone = cloneuio(uio);
917 resid = uio->uio_resid;
919 short_uio.uio_segflg = UIO_USERSPACE;
920 short_uio.uio_rw = uio->uio_rw;
921 short_uio.uio_td = uio->uio_td;
923 if (uio->uio_rw == UIO_READ) {
924 prot = VM_PROT_WRITE;
925 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
926 uio->uio_offset + uio->uio_resid);
929 if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0)
930 /* For appenders, punt and lock the whole range. */
931 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
933 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
934 uio->uio_offset + uio->uio_resid);
937 save = vm_fault_disable_pagefaults();
938 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
942 atomic_add_long(&vn_io_faults_cnt, 1);
943 uio_clone->uio_segflg = UIO_NOCOPY;
944 uiomove(NULL, resid - uio->uio_resid, uio_clone);
945 uio_clone->uio_segflg = uio->uio_segflg;
947 saveheld = curthread_pflags_set(TDP_UIOHELD);
948 prev_td_ma = td->td_ma;
949 prev_td_ma_cnt = td->td_ma_cnt;
951 while (uio_clone->uio_resid != 0) {
952 len = uio_clone->uio_iov->iov_len;
954 KASSERT(uio_clone->uio_iovcnt >= 1,
955 ("iovcnt underflow"));
956 uio_clone->uio_iov++;
957 uio_clone->uio_iovcnt--;
961 addr = (vm_offset_t)uio_clone->uio_iov->iov_base;
962 end = round_page(addr + len);
963 cnt = howmany(end - trunc_page(addr), PAGE_SIZE);
965 * A perfectly misaligned address and length could cause
966 * both the start and the end of the chunk to use partial
967 * page. +2 accounts for such a situation.
969 if (cnt > io_hold_cnt + 2) {
970 len = io_hold_cnt * PAGE_SIZE;
971 KASSERT(howmany(round_page(addr + len) -
972 trunc_page(addr), PAGE_SIZE) <= io_hold_cnt + 2,
975 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
976 addr, len, prot, ma, io_hold_cnt + 2);
981 short_uio.uio_iov = &short_iovec[0];
982 short_iovec[0].iov_base = (void *)addr;
983 short_uio.uio_iovcnt = 1;
984 short_uio.uio_resid = short_iovec[0].iov_len = len;
985 short_uio.uio_offset = uio_clone->uio_offset;
989 error = doio(fp, &short_uio, active_cred, flags | FOF_OFFSET,
991 vm_page_unhold_pages(ma, cnt);
992 adv = len - short_uio.uio_resid;
994 uio_clone->uio_iov->iov_base =
995 (char *)uio_clone->uio_iov->iov_base + adv;
996 uio_clone->uio_iov->iov_len -= adv;
997 uio_clone->uio_resid -= adv;
998 uio_clone->uio_offset += adv;
1000 uio->uio_resid -= adv;
1001 uio->uio_offset += adv;
1003 if (error != 0 || adv == 0)
1006 td->td_ma = prev_td_ma;
1007 td->td_ma_cnt = prev_td_ma_cnt;
1008 curthread_pflags_restore(saveheld);
1010 vm_fault_enable_pagefaults(save);
1011 vn_rangelock_unlock(vp, rl_cookie);
1012 free(uio_clone, M_IOV);
1014 foffset_unlock_uio(fp, uio, flags);
1019 * Helper function to perform the requested uiomove operation using
1020 * the held pages for io->uio_iov[0].iov_base buffer instead of
1021 * copyin/copyout. Access to the pages with uiomove_fromphys()
1022 * instead of iov_base prevents page faults that could occur due to
1023 * pmap_collect() invalidating the mapping created by
1024 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
1025 * object cleanup revoking the write access from page mappings.
1027 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
1028 * instead of plain uiomove().
1031 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1033 struct uio transp_uio;
1034 struct iovec transp_iov[1];
1040 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1041 uio->uio_segflg != UIO_USERSPACE)
1042 return (uiomove(data, xfersize, uio));
1044 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1045 transp_iov[0].iov_base = data;
1046 transp_uio.uio_iov = &transp_iov[0];
1047 transp_uio.uio_iovcnt = 1;
1048 if (xfersize > uio->uio_resid)
1049 xfersize = uio->uio_resid;
1050 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1051 transp_uio.uio_offset = 0;
1052 transp_uio.uio_segflg = UIO_SYSSPACE;
1054 * Since transp_iov points to data, and td_ma page array
1055 * corresponds to original uio->uio_iov, we need to invert the
1056 * direction of the i/o operation as passed to
1057 * uiomove_fromphys().
1059 switch (uio->uio_rw) {
1061 transp_uio.uio_rw = UIO_READ;
1064 transp_uio.uio_rw = UIO_WRITE;
1067 transp_uio.uio_td = uio->uio_td;
1068 error = uiomove_fromphys(td->td_ma,
1069 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1070 xfersize, &transp_uio);
1071 adv = xfersize - transp_uio.uio_resid;
1073 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1074 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1076 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1078 td->td_ma_cnt -= pgadv;
1079 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1080 uio->uio_iov->iov_len -= adv;
1081 uio->uio_resid -= adv;
1082 uio->uio_offset += adv;
1087 * File table truncate routine.
1090 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1103 * Lock the whole range for truncation. Otherwise split i/o
1104 * might happen partly before and partly after the truncation.
1106 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1107 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1108 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1111 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1112 if (vp->v_type == VDIR) {
1117 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1121 error = vn_writechk(vp);
1124 vattr.va_size = length;
1125 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1129 vn_finished_write(mp);
1131 VFS_UNLOCK_GIANT(vfslocked);
1132 vn_rangelock_unlock(vp, rl_cookie);
1137 * File table vnode stat routine.
1140 vn_statfile(fp, sb, active_cred, td)
1143 struct ucred *active_cred;
1146 struct vnode *vp = fp->f_vnode;
1150 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1151 vn_lock(vp, LK_SHARED | LK_RETRY);
1152 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1154 VFS_UNLOCK_GIANT(vfslocked);
1160 * Stat a vnode; implementation for the stat syscall
1163 vn_stat(vp, sb, active_cred, file_cred, td)
1165 register struct stat *sb;
1166 struct ucred *active_cred;
1167 struct ucred *file_cred;
1171 register struct vattr *vap;
1176 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1184 * Initialize defaults for new and unusual fields, so that file
1185 * systems which don't support these fields don't need to know
1188 vap->va_birthtime.tv_sec = -1;
1189 vap->va_birthtime.tv_nsec = 0;
1190 vap->va_fsid = VNOVAL;
1191 vap->va_rdev = NODEV;
1193 error = VOP_GETATTR(vp, vap, active_cred);
1198 * Zero the spare stat fields
1200 bzero(sb, sizeof *sb);
1203 * Copy from vattr table
1205 if (vap->va_fsid != VNOVAL)
1206 sb->st_dev = vap->va_fsid;
1208 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1209 sb->st_ino = vap->va_fileid;
1210 mode = vap->va_mode;
1211 switch (vap->va_type) {
1237 sb->st_nlink = vap->va_nlink;
1238 sb->st_uid = vap->va_uid;
1239 sb->st_gid = vap->va_gid;
1240 sb->st_rdev = vap->va_rdev;
1241 if (vap->va_size > OFF_MAX)
1243 sb->st_size = vap->va_size;
1244 sb->st_atim = vap->va_atime;
1245 sb->st_mtim = vap->va_mtime;
1246 sb->st_ctim = vap->va_ctime;
1247 sb->st_birthtim = vap->va_birthtime;
1250 * According to www.opengroup.org, the meaning of st_blksize is
1251 * "a filesystem-specific preferred I/O block size for this
1252 * object. In some filesystem types, this may vary from file
1254 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1257 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1259 sb->st_flags = vap->va_flags;
1260 if (priv_check(td, PRIV_VFS_GENERATION))
1263 sb->st_gen = vap->va_gen;
1265 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1270 * File table vnode ioctl routine.
1273 vn_ioctl(fp, com, data, active_cred, td)
1277 struct ucred *active_cred;
1280 struct vnode *vp = fp->f_vnode;
1285 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1287 switch (vp->v_type) {
1290 if (com == FIONREAD) {
1291 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1292 error = VOP_GETATTR(vp, &vattr, active_cred);
1295 *(int *)data = vattr.va_size - fp->f_offset;
1297 if (com == FIONBIO || com == FIOASYNC) /* XXX */
1300 error = VOP_IOCTL(vp, com, data, fp->f_flag,
1307 VFS_UNLOCK_GIANT(vfslocked);
1312 * File table vnode poll routine.
1315 vn_poll(fp, events, active_cred, td)
1318 struct ucred *active_cred;
1326 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1328 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1329 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1334 error = VOP_POLL(vp, events, fp->f_cred, td);
1335 VFS_UNLOCK_GIANT(vfslocked);
1340 * Acquire the requested lock and then check for validity. LK_RETRY
1341 * permits vn_lock to return doomed vnodes.
1344 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1348 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1349 ("vn_lock called with no locktype."));
1351 #ifdef DEBUG_VFS_LOCKS
1352 KASSERT(vp->v_holdcnt != 0,
1353 ("vn_lock %p: zero hold count", vp));
1355 error = VOP_LOCK1(vp, flags, file, line);
1356 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1357 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1358 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1361 * Callers specify LK_RETRY if they wish to get dead vnodes.
1362 * If RETRY is not set, we return ENOENT instead.
1364 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1365 (flags & LK_RETRY) == 0) {
1370 } while (flags & LK_RETRY && error != 0);
1375 * File table vnode close routine.
1378 vn_closefile(fp, td)
1389 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1390 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1391 lf.l_whence = SEEK_SET;
1394 lf.l_type = F_UNLCK;
1395 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1398 fp->f_ops = &badfileops;
1400 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1401 VFS_UNLOCK_GIANT(vfslocked);
1406 * Preparing to start a filesystem write operation. If the operation is
1407 * permitted, then we bump the count of operations in progress and
1408 * proceed. If a suspend request is in progress, we wait until the
1409 * suspension is over, and then proceed.
1412 vn_start_write_locked(struct mount *mp, int flags)
1416 mtx_assert(MNT_MTX(mp), MA_OWNED);
1420 * Check on status of suspension.
1422 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1423 mp->mnt_susp_owner != curthread) {
1424 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1425 if (flags & V_NOWAIT) {
1426 error = EWOULDBLOCK;
1429 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1430 (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1435 if (flags & V_XSLEEP)
1437 mp->mnt_writeopcount++;
1439 if (error != 0 || (flags & V_XSLEEP) != 0)
1446 vn_start_write(vp, mpp, flags)
1456 * If a vnode is provided, get and return the mount point that
1457 * to which it will write.
1460 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1462 if (error != EOPNOTSUPP)
1467 if ((mp = *mpp) == NULL)
1471 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1473 * As long as a vnode is not provided we need to acquire a
1474 * refcount for the provided mountpoint too, in order to
1475 * emulate a vfs_ref().
1481 return (vn_start_write_locked(mp, flags));
1485 * Secondary suspension. Used by operations such as vop_inactive
1486 * routines that are needed by the higher level functions. These
1487 * are allowed to proceed until all the higher level functions have
1488 * completed (indicated by mnt_writeopcount dropping to zero). At that
1489 * time, these operations are halted until the suspension is over.
1492 vn_start_secondary_write(vp, mpp, flags)
1502 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1504 if (error != EOPNOTSUPP)
1510 * If we are not suspended or have not yet reached suspended
1511 * mode, then let the operation proceed.
1513 if ((mp = *mpp) == NULL)
1517 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1519 * As long as a vnode is not provided we need to acquire a
1520 * refcount for the provided mountpoint too, in order to
1521 * emulate a vfs_ref().
1526 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1527 mp->mnt_secondary_writes++;
1528 mp->mnt_secondary_accwrites++;
1532 if (flags & V_NOWAIT) {
1535 return (EWOULDBLOCK);
1538 * Wait for the suspension to finish.
1540 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1541 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1549 * Filesystem write operation has completed. If we are suspending and this
1550 * operation is the last one, notify the suspender that the suspension is
1554 vn_finished_write(mp)
1561 mp->mnt_writeopcount--;
1562 if (mp->mnt_writeopcount < 0)
1563 panic("vn_finished_write: neg cnt");
1564 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1565 mp->mnt_writeopcount <= 0)
1566 wakeup(&mp->mnt_writeopcount);
1572 * Filesystem secondary write operation has completed. If we are
1573 * suspending and this operation is the last one, notify the suspender
1574 * that the suspension is now in effect.
1577 vn_finished_secondary_write(mp)
1584 mp->mnt_secondary_writes--;
1585 if (mp->mnt_secondary_writes < 0)
1586 panic("vn_finished_secondary_write: neg cnt");
1587 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1588 mp->mnt_secondary_writes <= 0)
1589 wakeup(&mp->mnt_secondary_writes);
1596 * Request a filesystem to suspend write operations.
1599 vfs_write_suspend(mp)
1605 if (mp->mnt_susp_owner == curthread) {
1609 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1610 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1611 mp->mnt_kern_flag |= MNTK_SUSPEND;
1612 mp->mnt_susp_owner = curthread;
1613 if (mp->mnt_writeopcount > 0)
1614 (void) msleep(&mp->mnt_writeopcount,
1615 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1618 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1619 vfs_write_resume(mp);
1624 * Request a filesystem to resume write operations.
1627 vfs_write_resume_flags(struct mount *mp, int flags)
1631 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1632 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1633 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1635 mp->mnt_susp_owner = NULL;
1636 wakeup(&mp->mnt_writeopcount);
1637 wakeup(&mp->mnt_flag);
1638 curthread->td_pflags &= ~TDP_IGNSUSP;
1639 if ((flags & VR_START_WRITE) != 0) {
1641 mp->mnt_writeopcount++;
1644 if ((flags & VR_NO_SUSPCLR) == 0)
1646 } else if ((flags & VR_START_WRITE) != 0) {
1648 vn_start_write_locked(mp, 0);
1655 vfs_write_resume(struct mount *mp)
1658 vfs_write_resume_flags(mp, 0);
1662 * Implement kqueues for files by translating it to vnode operation.
1665 vn_kqfilter(struct file *fp, struct knote *kn)
1670 vfslocked = VFS_LOCK_GIANT(fp->f_vnode->v_mount);
1671 error = VOP_KQFILTER(fp->f_vnode, kn);
1672 VFS_UNLOCK_GIANT(vfslocked);
1678 * Simplified in-kernel wrapper calls for extended attribute access.
1679 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1680 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1683 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1684 const char *attrname, int *buflen, char *buf, struct thread *td)
1690 iov.iov_len = *buflen;
1693 auio.uio_iov = &iov;
1694 auio.uio_iovcnt = 1;
1695 auio.uio_rw = UIO_READ;
1696 auio.uio_segflg = UIO_SYSSPACE;
1698 auio.uio_offset = 0;
1699 auio.uio_resid = *buflen;
1701 if ((ioflg & IO_NODELOCKED) == 0)
1702 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1704 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1706 /* authorize attribute retrieval as kernel */
1707 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1710 if ((ioflg & IO_NODELOCKED) == 0)
1714 *buflen = *buflen - auio.uio_resid;
1721 * XXX failure mode if partially written?
1724 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1725 const char *attrname, int buflen, char *buf, struct thread *td)
1732 iov.iov_len = buflen;
1735 auio.uio_iov = &iov;
1736 auio.uio_iovcnt = 1;
1737 auio.uio_rw = UIO_WRITE;
1738 auio.uio_segflg = UIO_SYSSPACE;
1740 auio.uio_offset = 0;
1741 auio.uio_resid = buflen;
1743 if ((ioflg & IO_NODELOCKED) == 0) {
1744 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1746 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1749 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1751 /* authorize attribute setting as kernel */
1752 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1754 if ((ioflg & IO_NODELOCKED) == 0) {
1755 vn_finished_write(mp);
1763 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1764 const char *attrname, struct thread *td)
1769 if ((ioflg & IO_NODELOCKED) == 0) {
1770 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1772 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1775 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1777 /* authorize attribute removal as kernel */
1778 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1779 if (error == EOPNOTSUPP)
1780 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1783 if ((ioflg & IO_NODELOCKED) == 0) {
1784 vn_finished_write(mp);
1792 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1798 ltype = VOP_ISLOCKED(vp);
1799 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1800 ("vn_vget_ino: vp not locked"));
1801 error = vfs_busy(mp, MBF_NOWAIT);
1805 error = vfs_busy(mp, 0);
1806 vn_lock(vp, ltype | LK_RETRY);
1810 if (vp->v_iflag & VI_DOOMED) {
1816 error = VFS_VGET(mp, ino, lkflags, rvp);
1818 vn_lock(vp, ltype | LK_RETRY);
1819 if (vp->v_iflag & VI_DOOMED) {
1828 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
1829 const struct thread *td)
1832 if (vp->v_type != VREG || td == NULL)
1834 PROC_LOCK(td->td_proc);
1835 if ((uoff_t)uio->uio_offset + uio->uio_resid >
1836 lim_cur(td->td_proc, RLIMIT_FSIZE)) {
1837 kern_psignal(td->td_proc, SIGXFSZ);
1838 PROC_UNLOCK(td->td_proc);
1841 PROC_UNLOCK(td->td_proc);
1846 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
1850 int error, vfslocked;
1853 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1855 vn_lock(vp, LK_SHARED | LK_RETRY);
1856 AUDIT_ARG_VNODE1(vp);
1859 error = setfmode(td, active_cred, vp, mode);
1860 VFS_UNLOCK_GIANT(vfslocked);
1865 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
1869 int error, vfslocked;
1872 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1874 vn_lock(vp, LK_SHARED | LK_RETRY);
1875 AUDIT_ARG_VNODE1(vp);
1878 error = setfown(td, active_cred, vp, uid, gid);
1879 VFS_UNLOCK_GIANT(vfslocked);
1884 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
1888 if ((object = vp->v_object) == NULL)
1890 VM_OBJECT_LOCK(object);
1891 vm_object_page_remove(object, start, end, 0);
1892 VM_OBJECT_UNLOCK(object);
1896 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
1904 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
1905 ("Wrong command %lu", cmd));
1907 if (vn_lock(vp, LK_SHARED) != 0)
1909 if (vp->v_type != VREG) {
1913 error = VOP_GETATTR(vp, &va, cred);
1917 if (noff >= va.va_size) {
1921 bsize = vp->v_mount->mnt_stat.f_iosize;
1922 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
1923 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
1924 if (error == EOPNOTSUPP) {
1928 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
1929 (bnp != -1 && cmd == FIOSEEKDATA)) {
1936 if (noff > va.va_size)
1938 /* noff == va.va_size. There is an implicit hole at the end of file. */
1939 if (cmd == FIOSEEKDATA)