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 via a name lookup.
112 * Lookup the vnode and invoke VOP_CREATE if needed.
113 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
115 * Note that this does NOT free nameidata for the successful case,
116 * due to the NDINIT being done elsewhere.
119 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
120 struct ucred *cred, struct file *fp)
124 struct thread *td = ndp->ni_cnd.cn_thread;
126 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 error = vn_open_vnode(vp, fmode, cred, td, fp);
213 VFS_UNLOCK_GIANT(vfslocked);
216 NDFREE(ndp, NDF_ONLY_PNBUF);
218 VFS_UNLOCK_GIANT(vfslocked);
225 * Common code for vnode open operations once a vnode is located.
226 * Check permissions, and call the VOP_OPEN routine.
229 vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred,
230 struct thread *td, struct file *fp)
235 VFS_ASSERT_GIANT(vp->v_mount);
236 if (vp->v_type == VLNK)
238 if (vp->v_type == VSOCK)
240 if (vp->v_type != VDIR && fmode & O_DIRECTORY)
243 if (fmode & (FWRITE | O_TRUNC)) {
244 if (vp->v_type == VDIR)
252 if ((fmode & O_APPEND) && (fmode & FWRITE))
255 error = mac_vnode_check_open(cred, vp, accmode);
259 if ((fmode & O_CREAT) == 0) {
260 if (accmode & VWRITE) {
261 error = vn_writechk(vp);
266 error = VOP_ACCESS(vp, accmode, cred, td);
271 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
274 if (fmode & FWRITE) {
276 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
277 __func__, vp, vp->v_writecount);
279 ASSERT_VOP_LOCKED(vp, "vn_open_vnode");
284 * Check for write permissions on the specified vnode.
285 * Prototype text segments cannot be written.
289 register struct vnode *vp;
292 ASSERT_VOP_LOCKED(vp, "vn_writechk");
294 * If there's shared text associated with
295 * the vnode, try to free it up once. If
296 * we fail, we can't allow writing.
298 if (vp->v_vflag & VV_TEXT)
308 vn_close(vp, flags, file_cred, td)
309 register struct vnode *vp;
311 struct ucred *file_cred;
315 int error, lock_flags;
317 if (!(flags & FWRITE) && vp->v_mount != NULL &&
318 vp->v_mount->mnt_kern_flag & MNTK_EXTENDED_SHARED)
319 lock_flags = LK_SHARED;
321 lock_flags = LK_EXCLUSIVE;
323 VFS_ASSERT_GIANT(vp->v_mount);
325 vn_start_write(vp, &mp, V_WAIT);
326 vn_lock(vp, lock_flags | LK_RETRY);
327 if (flags & FWRITE) {
328 VNASSERT(vp->v_writecount > 0, vp,
329 ("vn_close: negative writecount"));
331 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
332 __func__, vp, vp->v_writecount);
334 error = VOP_CLOSE(vp, flags, file_cred, td);
336 vn_finished_write(mp);
341 * Heuristic to detect sequential operation.
344 sequential_heuristic(struct uio *uio, struct file *fp)
347 if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
348 return (fp->f_seqcount << IO_SEQSHIFT);
351 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
352 * that the first I/O is normally considered to be slightly
353 * sequential. Seeking to offset 0 doesn't change sequentiality
354 * unless previous seeks have reduced f_seqcount to 0, in which
355 * case offset 0 is not special.
357 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
358 uio->uio_offset == fp->f_nextoff) {
360 * f_seqcount is in units of fixed-size blocks so that it
361 * depends mainly on the amount of sequential I/O and not
362 * much on the number of sequential I/O's. The fixed size
363 * of 16384 is hard-coded here since it is (not quite) just
364 * a magic size that works well here. This size is more
365 * closely related to the best I/O size for real disks than
366 * to any block size used by software.
368 fp->f_seqcount += howmany(uio->uio_resid, 16384);
369 if (fp->f_seqcount > IO_SEQMAX)
370 fp->f_seqcount = IO_SEQMAX;
371 return (fp->f_seqcount << IO_SEQSHIFT);
374 /* Not sequential. Quickly draw-down sequentiality. */
375 if (fp->f_seqcount > 1)
383 * Package up an I/O request on a vnode into a uio and do it.
386 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
387 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
388 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
395 int error, lock_flags;
397 VFS_ASSERT_GIANT(vp->v_mount);
399 auio.uio_iov = &aiov;
401 aiov.iov_base = base;
403 auio.uio_resid = len;
404 auio.uio_offset = offset;
405 auio.uio_segflg = segflg;
410 if ((ioflg & IO_NODELOCKED) == 0) {
411 if (rw == UIO_READ) {
412 rl_cookie = vn_rangelock_rlock(vp, offset,
415 rl_cookie = vn_rangelock_wlock(vp, offset,
419 if (rw == UIO_WRITE) {
420 if (vp->v_type != VCHR &&
421 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
424 if (MNT_SHARED_WRITES(mp) ||
425 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
426 lock_flags = LK_SHARED;
428 lock_flags = LK_EXCLUSIVE;
430 lock_flags = LK_SHARED;
431 vn_lock(vp, lock_flags | LK_RETRY);
435 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
437 if ((ioflg & IO_NOMACCHECK) == 0) {
439 error = mac_vnode_check_read(active_cred, file_cred,
442 error = mac_vnode_check_write(active_cred, file_cred,
447 if (file_cred != NULL)
452 error = VOP_READ(vp, &auio, ioflg, cred);
454 error = VOP_WRITE(vp, &auio, ioflg, cred);
457 *aresid = auio.uio_resid;
459 if (auio.uio_resid && error == 0)
461 if ((ioflg & IO_NODELOCKED) == 0) {
464 vn_finished_write(mp);
467 if (rl_cookie != NULL)
468 vn_rangelock_unlock(vp, rl_cookie);
473 * Package up an I/O request on a vnode into a uio and do it. The I/O
474 * request is split up into smaller chunks and we try to avoid saturating
475 * the buffer cache while potentially holding a vnode locked, so we
476 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
477 * to give other processes a chance to lock the vnode (either other processes
478 * core'ing the same binary, or unrelated processes scanning the directory).
481 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
482 file_cred, aresid, td)
490 struct ucred *active_cred;
491 struct ucred *file_cred;
498 VFS_ASSERT_GIANT(vp->v_mount);
504 * Force `offset' to a multiple of MAXBSIZE except possibly
505 * for the first chunk, so that filesystems only need to
506 * write full blocks except possibly for the first and last
509 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
513 if (rw != UIO_READ && vp->v_type == VREG)
516 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
517 ioflg, active_cred, file_cred, &iaresid, td);
518 len -= chunk; /* aresid calc already includes length */
522 base = (char *)base + chunk;
523 kern_yield(PRI_USER);
526 *aresid = len + iaresid;
531 foffset_lock(struct file *fp, struct uio *uio, int flags)
535 if ((flags & FOF_OFFSET) != 0)
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 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
545 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
546 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
547 "vnread offlock", 0);
549 fp->f_vnread_flags |= FOFFSET_LOCKED;
550 uio->uio_offset = fp->f_offset;
555 get_advice(struct file *fp, struct uio *uio)
560 ret = POSIX_FADV_NORMAL;
561 if (fp->f_advice == NULL)
564 mtxp = mtx_pool_find(mtxpool_sleep, fp);
566 if (uio->uio_offset >= fp->f_advice->fa_start &&
567 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
568 ret = fp->f_advice->fa_advice;
574 foffset_unlock(struct file *fp, struct uio *uio, int flags)
578 if ((flags & FOF_OFFSET) != 0)
581 fp->f_offset = uio->uio_offset;
582 mtxp = mtx_pool_find(mtxpool_sleep, fp);
584 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
585 wakeup(&fp->f_vnread_flags);
586 fp->f_vnread_flags = 0;
591 * File table vnode read routine.
594 vn_read(fp, uio, active_cred, flags, td)
597 struct ucred *active_cred;
604 int advice, vfslocked;
605 off_t offset, start, end;
607 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
609 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
612 if (fp->f_flag & FNONBLOCK)
614 if (fp->f_flag & O_DIRECT)
616 advice = get_advice(fp, uio);
617 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
618 vn_lock(vp, LK_SHARED | LK_RETRY);
621 case POSIX_FADV_NORMAL:
622 case POSIX_FADV_SEQUENTIAL:
623 case POSIX_FADV_NOREUSE:
624 ioflag |= sequential_heuristic(uio, fp);
626 case POSIX_FADV_RANDOM:
627 /* Disable read-ahead for random I/O. */
630 offset = uio->uio_offset;
633 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
636 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
637 fp->f_nextoff = uio->uio_offset;
639 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
640 offset != uio->uio_offset) {
642 * Use POSIX_FADV_DONTNEED to flush clean pages and
643 * buffers for the backing file after a
644 * POSIX_FADV_NOREUSE read(2). To optimize the common
645 * case of using POSIX_FADV_NOREUSE with sequential
646 * access, track the previous implicit DONTNEED
647 * request and grow this request to include the
648 * current read(2) in addition to the previous
649 * DONTNEED. With purely sequential access this will
650 * cause the DONTNEED requests to continously grow to
651 * cover all of the previously read regions of the
652 * file. This allows filesystem blocks that are
653 * accessed by multiple calls to read(2) to be flushed
654 * once the last read(2) finishes.
657 end = uio->uio_offset - 1;
658 mtxp = mtx_pool_find(mtxpool_sleep, fp);
660 if (fp->f_advice != NULL &&
661 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
662 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
663 start = fp->f_advice->fa_prevstart;
664 else if (fp->f_advice->fa_prevstart != 0 &&
665 fp->f_advice->fa_prevstart == end + 1)
666 end = fp->f_advice->fa_prevend;
667 fp->f_advice->fa_prevstart = start;
668 fp->f_advice->fa_prevend = end;
671 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
673 VFS_UNLOCK_GIANT(vfslocked);
678 * File table vnode write routine.
681 vn_write(fp, uio, active_cred, flags, td)
684 struct ucred *active_cred;
691 int error, ioflag, lock_flags;
692 int advice, vfslocked;
693 off_t offset, start, end;
695 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
697 KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET"));
699 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
700 if (vp->v_type == VREG)
703 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
705 if (fp->f_flag & FNONBLOCK)
707 if (fp->f_flag & O_DIRECT)
709 if ((fp->f_flag & O_FSYNC) ||
710 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
713 if (vp->v_type != VCHR &&
714 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
717 advice = get_advice(fp, uio);
719 if ((MNT_SHARED_WRITES(mp) ||
720 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) &&
721 (flags & FOF_OFFSET) != 0) {
722 lock_flags = LK_SHARED;
724 lock_flags = LK_EXCLUSIVE;
727 vn_lock(vp, lock_flags | LK_RETRY);
729 case POSIX_FADV_NORMAL:
730 case POSIX_FADV_SEQUENTIAL:
731 case POSIX_FADV_NOREUSE:
732 ioflag |= sequential_heuristic(uio, fp);
734 case POSIX_FADV_RANDOM:
735 /* XXX: Is this correct? */
738 offset = uio->uio_offset;
741 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
744 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
745 fp->f_nextoff = uio->uio_offset;
747 if (vp->v_type != VCHR)
748 vn_finished_write(mp);
749 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
750 offset != uio->uio_offset) {
752 * Use POSIX_FADV_DONTNEED to flush clean pages and
753 * buffers for the backing file after a
754 * POSIX_FADV_NOREUSE write(2). To optimize the
755 * common case of using POSIX_FADV_NOREUSE with
756 * sequential access, track the previous implicit
757 * DONTNEED request and grow this request to include
758 * the current write(2) in addition to the previous
759 * DONTNEED. With purely sequential access this will
760 * cause the DONTNEED requests to continously grow to
761 * cover all of the previously written regions of the
764 * Note that the blocks just written are almost
765 * certainly still dirty, so this only works when
766 * VOP_ADVISE() calls from subsequent writes push out
767 * the data written by this write(2) once the backing
768 * buffers are clean. However, as compared to forcing
769 * IO_DIRECT, this gives much saner behavior. Write
770 * clustering is still allowed, and clean pages are
771 * merely moved to the cache page queue rather than
772 * outright thrown away. This means a subsequent
773 * read(2) can still avoid hitting the disk if the
774 * pages have not been reclaimed.
776 * This does make POSIX_FADV_NOREUSE largely useless
777 * with non-sequential access. However, sequential
778 * access is the more common use case and the flag is
782 end = uio->uio_offset - 1;
783 mtxp = mtx_pool_find(mtxpool_sleep, fp);
785 if (fp->f_advice != NULL &&
786 fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) {
787 if (start != 0 && fp->f_advice->fa_prevend + 1 == start)
788 start = fp->f_advice->fa_prevstart;
789 else if (fp->f_advice->fa_prevstart != 0 &&
790 fp->f_advice->fa_prevstart == end + 1)
791 end = fp->f_advice->fa_prevend;
792 fp->f_advice->fa_prevstart = start;
793 fp->f_advice->fa_prevend = end;
796 error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED);
800 VFS_UNLOCK_GIANT(vfslocked);
804 static const int io_hold_cnt = 16;
805 static int vn_io_fault_enable = 1;
806 SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW,
807 &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance");
808 static unsigned long vn_io_faults_cnt;
809 SYSCTL_LONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD,
810 &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers");
813 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
814 * prevent the following deadlock:
816 * Assume that the thread A reads from the vnode vp1 into userspace
817 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
818 * currently not resident, then system ends up with the call chain
819 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
820 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
821 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
822 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
823 * backed by the pages of vnode vp1, and some page in buf2 is not
824 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
826 * To prevent the lock order reversal and deadlock, vn_io_fault() does
827 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
828 * Instead, it first tries to do the whole range i/o with pagefaults
829 * disabled. If all pages in the i/o buffer are resident and mapped,
830 * VOP will succeed (ignoring the genuine filesystem errors).
831 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
832 * i/o in chunks, with all pages in the chunk prefaulted and held
833 * using vm_fault_quick_hold_pages().
835 * Filesystems using this deadlock avoidance scheme should use the
836 * array of the held pages from uio, saved in the curthread->td_ma,
837 * instead of doing uiomove(). A helper function
838 * vn_io_fault_uiomove() converts uiomove request into
839 * uiomove_fromphys() over td_ma array.
841 * Since vnode locks do not cover the whole i/o anymore, rangelocks
842 * make the current i/o request atomic with respect to other i/os and
846 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
847 int flags, struct thread *td)
849 vm_page_t ma[io_hold_cnt + 2];
850 struct uio *uio_clone, short_uio;
851 struct iovec short_iovec[1];
856 vm_page_t *prev_td_ma;
857 int cnt, error, save, saveheld, prev_td_ma_cnt;
858 vm_offset_t addr, end;
863 if (uio->uio_rw == UIO_READ)
868 foffset_lock(fp, uio, flags);
870 if (uio->uio_segflg != UIO_USERSPACE || vp->v_type != VREG ||
871 ((mp = vp->v_mount) != NULL &&
872 (mp->mnt_kern_flag & MNTK_NO_IOPF) == 0) ||
873 !vn_io_fault_enable) {
874 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
879 * The UFS follows IO_UNIT directive and replays back both
880 * uio_offset and uio_resid if an error is encountered during the
881 * operation. But, since the iovec may be already advanced,
882 * uio is still in an inconsistent state.
884 * Cache a copy of the original uio, which is advanced to the redo
885 * point using UIO_NOCOPY below.
887 uio_clone = cloneuio(uio);
888 resid = uio->uio_resid;
890 short_uio.uio_segflg = UIO_USERSPACE;
891 short_uio.uio_rw = uio->uio_rw;
892 short_uio.uio_td = uio->uio_td;
894 if (uio->uio_rw == UIO_READ) {
895 prot = VM_PROT_WRITE;
896 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
897 uio->uio_offset + uio->uio_resid);
900 if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0)
901 /* For appenders, punt and lock the whole range. */
902 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
904 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
905 uio->uio_offset + uio->uio_resid);
908 save = vm_fault_disable_pagefaults();
909 error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td);
913 atomic_add_long(&vn_io_faults_cnt, 1);
914 uio_clone->uio_segflg = UIO_NOCOPY;
915 uiomove(NULL, resid - uio->uio_resid, uio_clone);
916 uio_clone->uio_segflg = uio->uio_segflg;
918 saveheld = curthread_pflags_set(TDP_UIOHELD);
919 prev_td_ma = td->td_ma;
920 prev_td_ma_cnt = td->td_ma_cnt;
922 while (uio_clone->uio_resid != 0) {
923 len = uio_clone->uio_iov->iov_len;
925 KASSERT(uio_clone->uio_iovcnt >= 1,
926 ("iovcnt underflow"));
927 uio_clone->uio_iov++;
928 uio_clone->uio_iovcnt--;
932 addr = (vm_offset_t)uio_clone->uio_iov->iov_base;
933 end = round_page(addr + len);
934 cnt = howmany(end - trunc_page(addr), PAGE_SIZE);
936 * A perfectly misaligned address and length could cause
937 * both the start and the end of the chunk to use partial
938 * page. +2 accounts for such a situation.
940 if (cnt > io_hold_cnt + 2) {
941 len = io_hold_cnt * PAGE_SIZE;
942 KASSERT(howmany(round_page(addr + len) -
943 trunc_page(addr), PAGE_SIZE) <= io_hold_cnt + 2,
946 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
947 addr, len, prot, ma, io_hold_cnt + 2);
952 short_uio.uio_iov = &short_iovec[0];
953 short_iovec[0].iov_base = (void *)addr;
954 short_uio.uio_iovcnt = 1;
955 short_uio.uio_resid = short_iovec[0].iov_len = len;
956 short_uio.uio_offset = uio_clone->uio_offset;
960 error = doio(fp, &short_uio, active_cred, flags | FOF_OFFSET,
962 vm_page_unhold_pages(ma, cnt);
963 adv = len - short_uio.uio_resid;
965 uio_clone->uio_iov->iov_base =
966 (char *)uio_clone->uio_iov->iov_base + adv;
967 uio_clone->uio_iov->iov_len -= adv;
968 uio_clone->uio_resid -= adv;
969 uio_clone->uio_offset += adv;
971 uio->uio_resid -= adv;
972 uio->uio_offset += adv;
974 if (error != 0 || adv == 0)
977 td->td_ma = prev_td_ma;
978 td->td_ma_cnt = prev_td_ma_cnt;
979 curthread_pflags_restore(saveheld);
981 vm_fault_enable_pagefaults(save);
982 vn_rangelock_unlock(vp, rl_cookie);
983 free(uio_clone, M_IOV);
985 foffset_unlock(fp, uio, flags);
990 * Helper function to perform the requested uiomove operation using
991 * the held pages for io->uio_iov[0].iov_base buffer instead of
992 * copyin/copyout. Access to the pages with uiomove_fromphys()
993 * instead of iov_base prevents page faults that could occur due to
994 * pmap_collect() invalidating the mapping created by
995 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
996 * object cleanup revoking the write access from page mappings.
998 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
999 * instead of plain uiomove().
1002 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
1004 struct uio transp_uio;
1005 struct iovec transp_iov[1];
1011 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
1012 uio->uio_segflg != UIO_USERSPACE)
1013 return (uiomove(data, xfersize, uio));
1015 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
1016 transp_iov[0].iov_base = data;
1017 transp_uio.uio_iov = &transp_iov[0];
1018 transp_uio.uio_iovcnt = 1;
1019 if (xfersize > uio->uio_resid)
1020 xfersize = uio->uio_resid;
1021 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
1022 transp_uio.uio_offset = 0;
1023 transp_uio.uio_segflg = UIO_SYSSPACE;
1025 * Since transp_iov points to data, and td_ma page array
1026 * corresponds to original uio->uio_iov, we need to invert the
1027 * direction of the i/o operation as passed to
1028 * uiomove_fromphys().
1030 switch (uio->uio_rw) {
1032 transp_uio.uio_rw = UIO_READ;
1035 transp_uio.uio_rw = UIO_WRITE;
1038 transp_uio.uio_td = uio->uio_td;
1039 error = uiomove_fromphys(td->td_ma,
1040 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
1041 xfersize, &transp_uio);
1042 adv = xfersize - transp_uio.uio_resid;
1044 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
1045 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
1047 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
1049 td->td_ma_cnt -= pgadv;
1050 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
1051 uio->uio_iov->iov_len -= adv;
1052 uio->uio_resid -= adv;
1053 uio->uio_offset += adv;
1058 * File table truncate routine.
1061 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
1074 * Lock the whole range for truncation. Otherwise split i/o
1075 * might happen partly before and partly after the truncation.
1077 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
1078 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1079 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
1082 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1083 if (vp->v_type == VDIR) {
1088 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
1092 error = vn_writechk(vp);
1095 vattr.va_size = length;
1096 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
1100 vn_finished_write(mp);
1102 VFS_UNLOCK_GIANT(vfslocked);
1103 vn_rangelock_unlock(vp, rl_cookie);
1108 * File table vnode stat routine.
1111 vn_statfile(fp, sb, active_cred, td)
1114 struct ucred *active_cred;
1117 struct vnode *vp = fp->f_vnode;
1121 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1122 vn_lock(vp, LK_SHARED | LK_RETRY);
1123 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1125 VFS_UNLOCK_GIANT(vfslocked);
1131 * Stat a vnode; implementation for the stat syscall
1134 vn_stat(vp, sb, active_cred, file_cred, td)
1136 register struct stat *sb;
1137 struct ucred *active_cred;
1138 struct ucred *file_cred;
1142 register struct vattr *vap;
1147 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1155 * Initialize defaults for new and unusual fields, so that file
1156 * systems which don't support these fields don't need to know
1159 vap->va_birthtime.tv_sec = -1;
1160 vap->va_birthtime.tv_nsec = 0;
1161 vap->va_fsid = VNOVAL;
1162 vap->va_rdev = NODEV;
1164 error = VOP_GETATTR(vp, vap, active_cred);
1169 * Zero the spare stat fields
1171 bzero(sb, sizeof *sb);
1174 * Copy from vattr table
1176 if (vap->va_fsid != VNOVAL)
1177 sb->st_dev = vap->va_fsid;
1179 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1180 sb->st_ino = vap->va_fileid;
1181 mode = vap->va_mode;
1182 switch (vap->va_type) {
1208 sb->st_nlink = vap->va_nlink;
1209 sb->st_uid = vap->va_uid;
1210 sb->st_gid = vap->va_gid;
1211 sb->st_rdev = vap->va_rdev;
1212 if (vap->va_size > OFF_MAX)
1214 sb->st_size = vap->va_size;
1215 sb->st_atim = vap->va_atime;
1216 sb->st_mtim = vap->va_mtime;
1217 sb->st_ctim = vap->va_ctime;
1218 sb->st_birthtim = vap->va_birthtime;
1221 * According to www.opengroup.org, the meaning of st_blksize is
1222 * "a filesystem-specific preferred I/O block size for this
1223 * object. In some filesystem types, this may vary from file
1225 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1228 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1230 sb->st_flags = vap->va_flags;
1231 if (priv_check(td, PRIV_VFS_GENERATION))
1234 sb->st_gen = vap->va_gen;
1236 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1241 * File table vnode ioctl routine.
1244 vn_ioctl(fp, com, data, active_cred, td)
1248 struct ucred *active_cred;
1251 struct vnode *vp = fp->f_vnode;
1256 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1258 switch (vp->v_type) {
1261 if (com == FIONREAD) {
1262 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1263 error = VOP_GETATTR(vp, &vattr, active_cred);
1266 *(int *)data = vattr.va_size - fp->f_offset;
1268 if (com == FIONBIO || com == FIOASYNC) /* XXX */
1271 error = VOP_IOCTL(vp, com, data, fp->f_flag,
1278 VFS_UNLOCK_GIANT(vfslocked);
1283 * File table vnode poll routine.
1286 vn_poll(fp, events, active_cred, td)
1289 struct ucred *active_cred;
1297 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1299 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1300 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1305 error = VOP_POLL(vp, events, fp->f_cred, td);
1306 VFS_UNLOCK_GIANT(vfslocked);
1311 * Acquire the requested lock and then check for validity. LK_RETRY
1312 * permits vn_lock to return doomed vnodes.
1315 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1319 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1320 ("vn_lock called with no locktype."));
1322 #ifdef DEBUG_VFS_LOCKS
1323 KASSERT(vp->v_holdcnt != 0,
1324 ("vn_lock %p: zero hold count", vp));
1326 error = VOP_LOCK1(vp, flags, file, line);
1327 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1328 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1329 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1332 * Callers specify LK_RETRY if they wish to get dead vnodes.
1333 * If RETRY is not set, we return ENOENT instead.
1335 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1336 (flags & LK_RETRY) == 0) {
1341 } while (flags & LK_RETRY && error != 0);
1346 * File table vnode close routine.
1349 vn_closefile(fp, td)
1360 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1361 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1362 lf.l_whence = SEEK_SET;
1365 lf.l_type = F_UNLCK;
1366 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1369 fp->f_ops = &badfileops;
1371 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1372 VFS_UNLOCK_GIANT(vfslocked);
1377 * Preparing to start a filesystem write operation. If the operation is
1378 * permitted, then we bump the count of operations in progress and
1379 * proceed. If a suspend request is in progress, we wait until the
1380 * suspension is over, and then proceed.
1383 vn_start_write(vp, mpp, flags)
1393 * If a vnode is provided, get and return the mount point that
1394 * to which it will write.
1397 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1399 if (error != EOPNOTSUPP)
1404 if ((mp = *mpp) == NULL)
1408 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1410 * As long as a vnode is not provided we need to acquire a
1411 * refcount for the provided mountpoint too, in order to
1412 * emulate a vfs_ref().
1419 * Check on status of suspension.
1421 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1422 mp->mnt_susp_owner != curthread) {
1423 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1424 if (flags & V_NOWAIT) {
1425 error = EWOULDBLOCK;
1428 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1429 (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1434 if (flags & V_XSLEEP)
1436 mp->mnt_writeopcount++;
1438 if (error != 0 || (flags & V_XSLEEP) != 0)
1445 * Secondary suspension. Used by operations such as vop_inactive
1446 * routines that are needed by the higher level functions. These
1447 * are allowed to proceed until all the higher level functions have
1448 * completed (indicated by mnt_writeopcount dropping to zero). At that
1449 * time, these operations are halted until the suspension is over.
1452 vn_start_secondary_write(vp, mpp, flags)
1462 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1464 if (error != EOPNOTSUPP)
1470 * If we are not suspended or have not yet reached suspended
1471 * mode, then let the operation proceed.
1473 if ((mp = *mpp) == NULL)
1477 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1479 * As long as a vnode is not provided we need to acquire a
1480 * refcount for the provided mountpoint too, in order to
1481 * emulate a vfs_ref().
1486 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1487 mp->mnt_secondary_writes++;
1488 mp->mnt_secondary_accwrites++;
1492 if (flags & V_NOWAIT) {
1495 return (EWOULDBLOCK);
1498 * Wait for the suspension to finish.
1500 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1501 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1509 * Filesystem write operation has completed. If we are suspending and this
1510 * operation is the last one, notify the suspender that the suspension is
1514 vn_finished_write(mp)
1521 mp->mnt_writeopcount--;
1522 if (mp->mnt_writeopcount < 0)
1523 panic("vn_finished_write: neg cnt");
1524 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1525 mp->mnt_writeopcount <= 0)
1526 wakeup(&mp->mnt_writeopcount);
1532 * Filesystem secondary write operation has completed. If we are
1533 * suspending and this operation is the last one, notify the suspender
1534 * that the suspension is now in effect.
1537 vn_finished_secondary_write(mp)
1544 mp->mnt_secondary_writes--;
1545 if (mp->mnt_secondary_writes < 0)
1546 panic("vn_finished_secondary_write: neg cnt");
1547 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1548 mp->mnt_secondary_writes <= 0)
1549 wakeup(&mp->mnt_secondary_writes);
1556 * Request a filesystem to suspend write operations.
1559 vfs_write_suspend(mp)
1565 if (mp->mnt_susp_owner == curthread) {
1569 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1570 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1571 mp->mnt_kern_flag |= MNTK_SUSPEND;
1572 mp->mnt_susp_owner = curthread;
1573 if (mp->mnt_writeopcount > 0)
1574 (void) msleep(&mp->mnt_writeopcount,
1575 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1578 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1579 vfs_write_resume(mp);
1584 * Request a filesystem to resume write operations.
1587 vfs_write_resume(mp)
1592 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1593 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1594 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1596 mp->mnt_susp_owner = NULL;
1597 wakeup(&mp->mnt_writeopcount);
1598 wakeup(&mp->mnt_flag);
1599 curthread->td_pflags &= ~TDP_IGNSUSP;
1607 * Implement kqueues for files by translating it to vnode operation.
1610 vn_kqfilter(struct file *fp, struct knote *kn)
1615 vfslocked = VFS_LOCK_GIANT(fp->f_vnode->v_mount);
1616 error = VOP_KQFILTER(fp->f_vnode, kn);
1617 VFS_UNLOCK_GIANT(vfslocked);
1623 * Simplified in-kernel wrapper calls for extended attribute access.
1624 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1625 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1628 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1629 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_READ;
1641 auio.uio_segflg = UIO_SYSSPACE;
1643 auio.uio_offset = 0;
1644 auio.uio_resid = *buflen;
1646 if ((ioflg & IO_NODELOCKED) == 0)
1647 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1649 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1651 /* authorize attribute retrieval as kernel */
1652 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1655 if ((ioflg & IO_NODELOCKED) == 0)
1659 *buflen = *buflen - auio.uio_resid;
1666 * XXX failure mode if partially written?
1669 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1670 const char *attrname, int buflen, char *buf, struct thread *td)
1677 iov.iov_len = buflen;
1680 auio.uio_iov = &iov;
1681 auio.uio_iovcnt = 1;
1682 auio.uio_rw = UIO_WRITE;
1683 auio.uio_segflg = UIO_SYSSPACE;
1685 auio.uio_offset = 0;
1686 auio.uio_resid = buflen;
1688 if ((ioflg & IO_NODELOCKED) == 0) {
1689 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1691 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1694 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1696 /* authorize attribute setting as kernel */
1697 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1699 if ((ioflg & IO_NODELOCKED) == 0) {
1700 vn_finished_write(mp);
1708 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1709 const char *attrname, struct thread *td)
1714 if ((ioflg & IO_NODELOCKED) == 0) {
1715 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1717 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1720 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1722 /* authorize attribute removal as kernel */
1723 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1724 if (error == EOPNOTSUPP)
1725 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1728 if ((ioflg & IO_NODELOCKED) == 0) {
1729 vn_finished_write(mp);
1737 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1743 ltype = VOP_ISLOCKED(vp);
1744 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1745 ("vn_vget_ino: vp not locked"));
1746 error = vfs_busy(mp, MBF_NOWAIT);
1750 error = vfs_busy(mp, 0);
1751 vn_lock(vp, ltype | LK_RETRY);
1755 if (vp->v_iflag & VI_DOOMED) {
1761 error = VFS_VGET(mp, ino, lkflags, rvp);
1763 vn_lock(vp, ltype | LK_RETRY);
1764 if (vp->v_iflag & VI_DOOMED) {
1773 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
1774 const struct thread *td)
1777 if (vp->v_type != VREG || td == NULL)
1779 PROC_LOCK(td->td_proc);
1780 if ((uoff_t)uio->uio_offset + uio->uio_resid >
1781 lim_cur(td->td_proc, RLIMIT_FSIZE)) {
1782 kern_psignal(td->td_proc, SIGXFSZ);
1783 PROC_UNLOCK(td->td_proc);
1786 PROC_UNLOCK(td->td_proc);
1791 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
1795 int error, vfslocked;
1798 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1800 vn_lock(vp, LK_SHARED | LK_RETRY);
1801 AUDIT_ARG_VNODE1(vp);
1804 error = setfmode(td, active_cred, vp, mode);
1805 VFS_UNLOCK_GIANT(vfslocked);
1810 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
1814 int error, vfslocked;
1817 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1819 vn_lock(vp, LK_SHARED | LK_RETRY);
1820 AUDIT_ARG_VNODE1(vp);
1823 error = setfown(td, active_cred, vp, uid, gid);
1824 VFS_UNLOCK_GIANT(vfslocked);
1829 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
1833 if ((object = vp->v_object) == NULL)
1835 VM_OBJECT_LOCK(object);
1836 vm_object_page_remove(object, start, end, 0);
1837 VM_OBJECT_UNLOCK(object);
1841 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
1849 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
1850 ("Wrong command %lu", cmd));
1852 if (vn_lock(vp, LK_SHARED) != 0)
1854 if (vp->v_type != VREG) {
1858 error = VOP_GETATTR(vp, &va, cred);
1862 if (noff >= va.va_size) {
1866 bsize = vp->v_mount->mnt_stat.f_iosize;
1867 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
1868 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
1869 if (error == EOPNOTSUPP) {
1873 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
1874 (bnp != -1 && cmd == FIOSEEKDATA)) {
1881 if (noff > va.va_size)
1883 /* noff == va.va_size. There is an implicit hole at the end of file. */
1884 if (cmd == FIOSEEKDATA)