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/ttycom.h>
61 #include <sys/syslog.h>
62 #include <sys/unistd.h>
64 #include <security/audit/audit.h>
65 #include <security/mac/mac_framework.h>
68 #include <vm/vm_extern.h>
70 #include <vm/vm_map.h>
71 #include <vm/vm_object.h>
72 #include <vm/vm_page.h>
74 static fo_rdwr_t vn_read;
75 static fo_rdwr_t vn_write;
76 static fo_rdwr_t vn_io_fault;
77 static fo_truncate_t vn_truncate;
78 static fo_ioctl_t vn_ioctl;
79 static fo_poll_t vn_poll;
80 static fo_kqfilter_t vn_kqfilter;
81 static fo_stat_t vn_statfile;
82 static fo_close_t vn_closefile;
84 struct fileops vnops = {
85 .fo_read = vn_io_fault,
86 .fo_write = vn_io_fault,
87 .fo_truncate = vn_truncate,
90 .fo_kqfilter = vn_kqfilter,
91 .fo_stat = vn_statfile,
92 .fo_close = vn_closefile,
95 .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE
99 vn_open(ndp, flagp, cmode, fp)
100 struct nameidata *ndp;
104 struct thread *td = ndp->ni_cnd.cn_thread;
106 return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp));
110 * Common code for vnode open operations.
111 * Check permissions, and call the VOP_OPEN or VOP_CREATE routine.
113 * Note that this does NOT free nameidata for the successful case,
114 * due to the NDINIT being done elsewhere.
117 vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags,
118 struct ucred *cred, struct file *fp)
122 struct thread *td = ndp->ni_cnd.cn_thread;
124 struct vattr *vap = &vat;
127 int vfslocked, mpsafe;
129 mpsafe = ndp->ni_cnd.cn_flags & MPSAFE;
133 if (fmode & O_CREAT) {
134 ndp->ni_cnd.cn_nameiop = CREATE;
135 ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF |
137 if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0)
138 ndp->ni_cnd.cn_flags |= FOLLOW;
139 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
140 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
142 if ((error = namei(ndp)) != 0)
144 vfslocked = NDHASGIANT(ndp);
146 ndp->ni_cnd.cn_flags &= ~MPSAFE;
147 if (ndp->ni_vp == NULL) {
150 vap->va_mode = cmode;
152 vap->va_vaflags |= VA_EXCLUSIVE;
153 if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) {
154 NDFREE(ndp, NDF_ONLY_PNBUF);
156 VFS_UNLOCK_GIANT(vfslocked);
157 if ((error = vn_start_write(NULL, &mp,
158 V_XSLEEP | PCATCH)) != 0)
163 error = mac_vnode_check_create(cred, ndp->ni_dvp,
167 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
170 vn_finished_write(mp);
172 VFS_UNLOCK_GIANT(vfslocked);
173 NDFREE(ndp, NDF_ONLY_PNBUF);
179 if (ndp->ni_dvp == ndp->ni_vp)
185 if (fmode & O_EXCL) {
192 ndp->ni_cnd.cn_nameiop = LOOKUP;
193 ndp->ni_cnd.cn_flags = ISOPEN |
194 ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) |
196 if (!(fmode & FWRITE))
197 ndp->ni_cnd.cn_flags |= LOCKSHARED;
198 if (!(vn_open_flags & VN_OPEN_NOAUDIT))
199 ndp->ni_cnd.cn_flags |= AUDITVNODE1;
200 if ((error = namei(ndp)) != 0)
203 ndp->ni_cnd.cn_flags &= ~MPSAFE;
204 vfslocked = NDHASGIANT(ndp);
207 if (vp->v_type == VLNK) {
211 if (vp->v_type == VSOCK) {
215 if (vp->v_type != VDIR && fmode & O_DIRECTORY) {
220 if (fmode & (FWRITE | O_TRUNC)) {
221 if (vp->v_type == VDIR) {
231 if ((fmode & O_APPEND) && (fmode & FWRITE))
234 error = mac_vnode_check_open(cred, vp, accmode);
238 if ((fmode & O_CREAT) == 0) {
239 if (accmode & VWRITE) {
240 error = vn_writechk(vp);
245 error = VOP_ACCESS(vp, accmode, cred, td);
250 if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0)
253 if (fmode & FWRITE) {
255 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
256 __func__, vp, vp->v_writecount);
259 ASSERT_VOP_LOCKED(vp, "vn_open_cred");
261 VFS_UNLOCK_GIANT(vfslocked);
264 NDFREE(ndp, NDF_ONLY_PNBUF);
266 VFS_UNLOCK_GIANT(vfslocked);
273 * Check for write permissions on the specified vnode.
274 * Prototype text segments cannot be written.
278 register struct vnode *vp;
281 ASSERT_VOP_LOCKED(vp, "vn_writechk");
283 * If there's shared text associated with
284 * the vnode, try to free it up once. If
285 * we fail, we can't allow writing.
287 if (vp->v_vflag & VV_TEXT)
297 vn_close(vp, flags, file_cred, td)
298 register struct vnode *vp;
300 struct ucred *file_cred;
304 int error, lock_flags;
306 if (!(flags & FWRITE) && vp->v_mount != NULL &&
307 vp->v_mount->mnt_kern_flag & MNTK_EXTENDED_SHARED)
308 lock_flags = LK_SHARED;
310 lock_flags = LK_EXCLUSIVE;
312 VFS_ASSERT_GIANT(vp->v_mount);
314 vn_start_write(vp, &mp, V_WAIT);
315 vn_lock(vp, lock_flags | LK_RETRY);
316 if (flags & FWRITE) {
317 VNASSERT(vp->v_writecount > 0, vp,
318 ("vn_close: negative writecount"));
320 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
321 __func__, vp, vp->v_writecount);
323 error = VOP_CLOSE(vp, flags, file_cred, td);
325 vn_finished_write(mp);
330 * Heuristic to detect sequential operation.
333 sequential_heuristic(struct uio *uio, struct file *fp)
336 if (atomic_load_acq_int(&(fp->f_flag)) & FRDAHEAD)
337 return (fp->f_seqcount << IO_SEQSHIFT);
340 * Offset 0 is handled specially. open() sets f_seqcount to 1 so
341 * that the first I/O is normally considered to be slightly
342 * sequential. Seeking to offset 0 doesn't change sequentiality
343 * unless previous seeks have reduced f_seqcount to 0, in which
344 * case offset 0 is not special.
346 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
347 uio->uio_offset == fp->f_nextoff) {
349 * f_seqcount is in units of fixed-size blocks so that it
350 * depends mainly on the amount of sequential I/O and not
351 * much on the number of sequential I/O's. The fixed size
352 * of 16384 is hard-coded here since it is (not quite) just
353 * a magic size that works well here. This size is more
354 * closely related to the best I/O size for real disks than
355 * to any block size used by software.
357 fp->f_seqcount += howmany(uio->uio_resid, 16384);
358 if (fp->f_seqcount > IO_SEQMAX)
359 fp->f_seqcount = IO_SEQMAX;
360 return (fp->f_seqcount << IO_SEQSHIFT);
363 /* Not sequential. Quickly draw-down sequentiality. */
364 if (fp->f_seqcount > 1)
372 * Package up an I/O request on a vnode into a uio and do it.
375 vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset,
376 enum uio_seg segflg, int ioflg, struct ucred *active_cred,
377 struct ucred *file_cred, ssize_t *aresid, struct thread *td)
384 int error, lock_flags;
386 VFS_ASSERT_GIANT(vp->v_mount);
388 auio.uio_iov = &aiov;
390 aiov.iov_base = base;
392 auio.uio_resid = len;
393 auio.uio_offset = offset;
394 auio.uio_segflg = segflg;
399 if ((ioflg & IO_NODELOCKED) == 0) {
400 if (rw == UIO_READ) {
401 rl_cookie = vn_rangelock_rlock(vp, offset,
404 rl_cookie = vn_rangelock_wlock(vp, offset,
408 if (rw == UIO_WRITE) {
409 if (vp->v_type != VCHR &&
410 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH))
413 if (MNT_SHARED_WRITES(mp) ||
414 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount)))
415 lock_flags = LK_SHARED;
417 lock_flags = LK_EXCLUSIVE;
419 lock_flags = LK_SHARED;
420 vn_lock(vp, lock_flags | LK_RETRY);
424 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
426 if ((ioflg & IO_NOMACCHECK) == 0) {
428 error = mac_vnode_check_read(active_cred, file_cred,
431 error = mac_vnode_check_write(active_cred, file_cred,
436 if (file_cred != NULL)
441 error = VOP_READ(vp, &auio, ioflg, cred);
443 error = VOP_WRITE(vp, &auio, ioflg, cred);
446 *aresid = auio.uio_resid;
448 if (auio.uio_resid && error == 0)
450 if ((ioflg & IO_NODELOCKED) == 0) {
453 vn_finished_write(mp);
456 if (rl_cookie != NULL)
457 vn_rangelock_unlock(vp, rl_cookie);
462 * Package up an I/O request on a vnode into a uio and do it. The I/O
463 * request is split up into smaller chunks and we try to avoid saturating
464 * the buffer cache while potentially holding a vnode locked, so we
465 * check bwillwrite() before calling vn_rdwr(). We also call kern_yield()
466 * to give other processes a chance to lock the vnode (either other processes
467 * core'ing the same binary, or unrelated processes scanning the directory).
470 vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred,
471 file_cred, aresid, td)
479 struct ucred *active_cred;
480 struct ucred *file_cred;
487 VFS_ASSERT_GIANT(vp->v_mount);
493 * Force `offset' to a multiple of MAXBSIZE except possibly
494 * for the first chunk, so that filesystems only need to
495 * write full blocks except possibly for the first and last
498 chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE;
502 if (rw != UIO_READ && vp->v_type == VREG)
505 error = vn_rdwr(rw, vp, base, chunk, offset, segflg,
506 ioflg, active_cred, file_cred, &iaresid, td);
507 len -= chunk; /* aresid calc already includes length */
511 base = (char *)base + chunk;
512 kern_yield(PRI_USER);
515 *aresid = len + iaresid;
520 * File table vnode read routine.
523 vn_read(fp, uio, active_cred, flags, td)
526 struct ucred *active_cred;
533 int advice, vfslocked;
536 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
541 if (fp->f_flag & FNONBLOCK)
543 if (fp->f_flag & O_DIRECT)
545 advice = POSIX_FADV_NORMAL;
546 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
548 * According to McKusick the vn lock was protecting f_offset here.
549 * It is now protected by the FOFFSET_LOCKED flag.
551 if ((flags & FOF_OFFSET) == 0 || fp->f_advice != NULL) {
552 mtxp = mtx_pool_find(mtxpool_sleep, fp);
554 if ((flags & FOF_OFFSET) == 0) {
555 while (fp->f_vnread_flags & FOFFSET_LOCKED) {
556 fp->f_vnread_flags |= FOFFSET_LOCK_WAITING;
557 msleep(&fp->f_vnread_flags, mtxp, PUSER -1,
558 "vnread offlock", 0);
560 fp->f_vnread_flags |= FOFFSET_LOCKED;
561 uio->uio_offset = fp->f_offset;
563 if (fp->f_advice != NULL &&
564 uio->uio_offset >= fp->f_advice->fa_start &&
565 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
566 advice = fp->f_advice->fa_advice;
569 vn_lock(vp, LK_SHARED | LK_RETRY);
572 case POSIX_FADV_NORMAL:
573 case POSIX_FADV_SEQUENTIAL:
574 case POSIX_FADV_NOREUSE:
575 ioflag |= sequential_heuristic(uio, fp);
577 case POSIX_FADV_RANDOM:
578 /* Disable read-ahead for random I/O. */
581 offset = uio->uio_offset;
584 error = mac_vnode_check_read(active_cred, fp->f_cred, vp);
587 error = VOP_READ(vp, uio, ioflag, fp->f_cred);
588 if ((flags & FOF_OFFSET) == 0) {
589 fp->f_offset = uio->uio_offset;
591 if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING)
592 wakeup(&fp->f_vnread_flags);
593 fp->f_vnread_flags = 0;
596 fp->f_nextoff = uio->uio_offset;
598 if (error == 0 && advice == POSIX_FADV_NOREUSE &&
599 offset != uio->uio_offset)
600 error = VOP_ADVISE(vp, offset, uio->uio_offset - 1,
601 POSIX_FADV_DONTNEED);
602 VFS_UNLOCK_GIANT(vfslocked);
607 * File table vnode write routine.
610 vn_write(fp, uio, active_cred, flags, td)
613 struct ucred *active_cred;
619 int error, ioflag, lock_flags;
621 int advice, vfslocked;
623 KASSERT(uio->uio_td == td, ("uio_td %p is not td %p",
626 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
627 if (vp->v_type == VREG)
630 if (vp->v_type == VREG && (fp->f_flag & O_APPEND))
632 if (fp->f_flag & FNONBLOCK)
634 if (fp->f_flag & O_DIRECT)
636 if ((fp->f_flag & O_FSYNC) ||
637 (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS)))
640 if (vp->v_type != VCHR &&
641 (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0)
644 if ((MNT_SHARED_WRITES(mp) ||
645 ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) &&
646 (flags & FOF_OFFSET) != 0) {
647 lock_flags = LK_SHARED;
649 lock_flags = LK_EXCLUSIVE;
652 vn_lock(vp, lock_flags | LK_RETRY);
653 if ((flags & FOF_OFFSET) == 0)
654 uio->uio_offset = fp->f_offset;
655 advice = POSIX_FADV_NORMAL;
656 if (fp->f_advice != NULL) {
657 mtxp = mtx_pool_find(mtxpool_sleep, fp);
659 if (fp->f_advice != NULL &&
660 uio->uio_offset >= fp->f_advice->fa_start &&
661 uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end)
662 advice = fp->f_advice->fa_advice;
666 case POSIX_FADV_NORMAL:
667 case POSIX_FADV_SEQUENTIAL:
668 ioflag |= sequential_heuristic(uio, fp);
670 case POSIX_FADV_RANDOM:
671 /* XXX: Is this correct? */
673 case POSIX_FADV_NOREUSE:
675 * Request the underlying FS to discard the buffers
676 * and pages after the I/O is complete.
683 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
686 error = VOP_WRITE(vp, uio, ioflag, fp->f_cred);
687 if ((flags & FOF_OFFSET) == 0)
688 fp->f_offset = uio->uio_offset;
689 fp->f_nextoff = uio->uio_offset;
691 if (vp->v_type != VCHR)
692 vn_finished_write(mp);
694 VFS_UNLOCK_GIANT(vfslocked);
698 static const int io_hold_cnt = 16;
701 * The vn_io_fault() is a wrapper around vn_read() and vn_write() to
702 * prevent the following deadlock:
704 * Assume that the thread A reads from the vnode vp1 into userspace
705 * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is
706 * currently not resident, then system ends up with the call chain
707 * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] ->
708 * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2)
709 * which establishes lock order vp1->vn_lock, then vp2->vn_lock.
710 * If, at the same time, thread B reads from vnode vp2 into buffer buf2
711 * backed by the pages of vnode vp1, and some page in buf2 is not
712 * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock.
714 * To prevent the lock order reversal and deadlock, vn_io_fault() does
715 * not allow page faults to happen during VOP_READ() or VOP_WRITE().
716 * Instead, it first tries to do the whole range i/o with pagefaults
717 * disabled. If all pages in the i/o buffer are resident and mapped,
718 * VOP will succeed (ignoring the genuine filesystem errors).
719 * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do
720 * i/o in chunks, with all pages in the chunk prefaulted and held
721 * using vm_fault_quick_hold_pages().
723 * Filesystems using this deadlock avoidance scheme should use the
724 * array of the held pages from uio, saved in the curthread->td_ma,
725 * instead of doing uiomove(). A helper function
726 * vn_io_fault_uiomove() converts uiomove request into
727 * uiomove_fromphys() over td_ma array.
729 * Since vnode locks do not cover the whole i/o anymore, rangelocks
730 * make the current i/o request atomic with respect to other i/os and
734 vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred,
735 int flags, struct thread *td)
737 vm_page_t ma[io_hold_cnt + 2];
738 struct uio *uio_clone, short_uio;
739 struct iovec short_iovec[1];
744 vm_page_t *prev_td_ma;
745 int cnt, error, save, saveheld, prev_td_ma_cnt;
746 vm_offset_t addr, end;
751 if (uio->uio_rw == UIO_READ)
756 if (uio->uio_segflg != UIO_USERSPACE || vp->v_type != VREG ||
757 ((mp = vp->v_mount) != NULL &&
758 (mp->mnt_kern_flag & MNTK_NO_IOPF) == 0))
759 return (doio(fp, uio, active_cred, flags, td));
762 * The UFS follows IO_UNIT directive and replays back both
763 * uio_offset and uio_resid if an error is encountered during the
764 * operation. But, since the iovec may be already advanced,
765 * uio is still in an inconsistent state.
767 * Cache a copy of the original uio, which is advanced to the redo
768 * point using UIO_NOCOPY below.
770 uio_clone = cloneuio(uio);
771 resid = uio->uio_resid;
773 short_uio.uio_segflg = UIO_USERSPACE;
774 short_uio.uio_rw = uio->uio_rw;
775 short_uio.uio_td = uio->uio_td;
777 if (uio->uio_rw == UIO_READ) {
778 prot = VM_PROT_WRITE;
779 rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset,
780 uio->uio_offset + uio->uio_resid);
783 if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0)
784 /* For appenders, punt and lock the whole range. */
785 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
787 rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset,
788 uio->uio_offset + uio->uio_resid);
791 save = vm_fault_disable_pagefaults();
792 error = doio(fp, uio, active_cred, flags, td);
796 uio_clone->uio_segflg = UIO_NOCOPY;
797 uiomove(NULL, resid - uio->uio_resid, uio_clone);
798 uio_clone->uio_segflg = uio->uio_segflg;
800 saveheld = curthread_pflags_set(TDP_UIOHELD);
801 prev_td_ma = td->td_ma;
802 prev_td_ma_cnt = td->td_ma_cnt;
804 while (uio_clone->uio_resid != 0) {
805 len = uio_clone->uio_iov->iov_len;
807 KASSERT(uio_clone->uio_iovcnt >= 1,
808 ("iovcnt underflow"));
809 uio_clone->uio_iov++;
810 uio_clone->uio_iovcnt--;
814 addr = (vm_offset_t)uio_clone->uio_iov->iov_base;
815 end = round_page(addr + len);
816 cnt = howmany(end - trunc_page(addr), PAGE_SIZE);
818 * A perfectly misaligned address and length could cause
819 * both the start and the end of the chunk to use partial
820 * page. +2 accounts for such a situation.
822 if (cnt > io_hold_cnt + 2) {
823 len = io_hold_cnt * PAGE_SIZE;
824 KASSERT(howmany(round_page(addr + len) -
825 trunc_page(addr), PAGE_SIZE) <= io_hold_cnt + 2,
828 cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map,
829 addr, len, prot, ma, io_hold_cnt + 2);
834 short_uio.uio_iov = &short_iovec[0];
835 short_iovec[0].iov_base = (void *)addr;
836 short_uio.uio_iovcnt = 1;
837 short_uio.uio_resid = short_iovec[0].iov_len = len;
838 short_uio.uio_offset = uio_clone->uio_offset;
842 error = doio(fp, &short_uio, active_cred, flags, td);
843 vm_page_unhold_pages(ma, cnt);
844 adv = len - short_uio.uio_resid;
846 uio_clone->uio_iov->iov_base =
847 (char *)uio_clone->uio_iov->iov_base + adv;
848 uio_clone->uio_iov->iov_len -= adv;
849 uio_clone->uio_resid -= adv;
850 uio_clone->uio_offset += adv;
852 uio->uio_resid -= adv;
853 uio->uio_offset += adv;
855 if (error != 0 || adv == 0)
858 td->td_ma = prev_td_ma;
859 td->td_ma_cnt = prev_td_ma_cnt;
860 curthread_pflags_restore(saveheld);
862 vm_fault_enable_pagefaults(save);
863 vn_rangelock_unlock(vp, rl_cookie);
864 free(uio_clone, M_IOV);
869 * Helper function to perform the requested uiomove operation using
870 * the held pages for io->uio_iov[0].iov_base buffer instead of
871 * copyin/copyout. Access to the pages with uiomove_fromphys()
872 * instead of iov_base prevents page faults that could occur due to
873 * pmap_collect() invalidating the mapping created by
874 * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or
875 * object cleanup revoking the write access from page mappings.
877 * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove()
878 * instead of plain uiomove().
881 vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio)
883 struct uio transp_uio;
884 struct iovec transp_iov[1];
890 if ((td->td_pflags & TDP_UIOHELD) == 0 ||
891 uio->uio_segflg != UIO_USERSPACE)
892 return (uiomove(data, xfersize, uio));
894 KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt));
895 transp_iov[0].iov_base = data;
896 transp_uio.uio_iov = &transp_iov[0];
897 transp_uio.uio_iovcnt = 1;
898 if (xfersize > uio->uio_resid)
899 xfersize = uio->uio_resid;
900 transp_uio.uio_resid = transp_iov[0].iov_len = xfersize;
901 transp_uio.uio_offset = 0;
902 transp_uio.uio_segflg = UIO_SYSSPACE;
904 * Since transp_iov points to data, and td_ma page array
905 * corresponds to original uio->uio_iov, we need to invert the
906 * direction of the i/o operation as passed to
907 * uiomove_fromphys().
909 switch (uio->uio_rw) {
911 transp_uio.uio_rw = UIO_READ;
914 transp_uio.uio_rw = UIO_WRITE;
917 transp_uio.uio_td = uio->uio_td;
918 error = uiomove_fromphys(td->td_ma,
919 ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK,
920 xfersize, &transp_uio);
921 adv = xfersize - transp_uio.uio_resid;
923 (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) -
924 (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT);
926 KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt,
928 td->td_ma_cnt -= pgadv;
929 uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv;
930 uio->uio_iov->iov_len -= adv;
931 uio->uio_resid -= adv;
932 uio->uio_offset += adv;
937 * File table truncate routine.
940 vn_truncate(struct file *fp, off_t length, struct ucred *active_cred,
953 * Lock the whole range for truncation. Otherwise split i/o
954 * might happen partly before and partly after the truncation.
956 rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX);
957 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
958 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
961 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
962 if (vp->v_type == VDIR) {
967 error = mac_vnode_check_write(active_cred, fp->f_cred, vp);
971 error = vn_writechk(vp);
974 vattr.va_size = length;
975 error = VOP_SETATTR(vp, &vattr, fp->f_cred);
979 vn_finished_write(mp);
981 VFS_UNLOCK_GIANT(vfslocked);
982 vn_rangelock_unlock(vp, rl_cookie);
987 * File table vnode stat routine.
990 vn_statfile(fp, sb, active_cred, td)
993 struct ucred *active_cred;
996 struct vnode *vp = fp->f_vnode;
1000 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1001 vn_lock(vp, LK_SHARED | LK_RETRY);
1002 error = vn_stat(vp, sb, active_cred, fp->f_cred, td);
1004 VFS_UNLOCK_GIANT(vfslocked);
1010 * Stat a vnode; implementation for the stat syscall
1013 vn_stat(vp, sb, active_cred, file_cred, td)
1015 register struct stat *sb;
1016 struct ucred *active_cred;
1017 struct ucred *file_cred;
1021 register struct vattr *vap;
1026 error = mac_vnode_check_stat(active_cred, file_cred, vp);
1034 * Initialize defaults for new and unusual fields, so that file
1035 * systems which don't support these fields don't need to know
1038 vap->va_birthtime.tv_sec = -1;
1039 vap->va_birthtime.tv_nsec = 0;
1040 vap->va_fsid = VNOVAL;
1041 vap->va_rdev = NODEV;
1043 error = VOP_GETATTR(vp, vap, active_cred);
1048 * Zero the spare stat fields
1050 bzero(sb, sizeof *sb);
1053 * Copy from vattr table
1055 if (vap->va_fsid != VNOVAL)
1056 sb->st_dev = vap->va_fsid;
1058 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1059 sb->st_ino = vap->va_fileid;
1060 mode = vap->va_mode;
1061 switch (vap->va_type) {
1087 sb->st_nlink = vap->va_nlink;
1088 sb->st_uid = vap->va_uid;
1089 sb->st_gid = vap->va_gid;
1090 sb->st_rdev = vap->va_rdev;
1091 if (vap->va_size > OFF_MAX)
1093 sb->st_size = vap->va_size;
1094 sb->st_atim = vap->va_atime;
1095 sb->st_mtim = vap->va_mtime;
1096 sb->st_ctim = vap->va_ctime;
1097 sb->st_birthtim = vap->va_birthtime;
1100 * According to www.opengroup.org, the meaning of st_blksize is
1101 * "a filesystem-specific preferred I/O block size for this
1102 * object. In some filesystem types, this may vary from file
1104 * Use miminum/default of PAGE_SIZE (e.g. for VCHR).
1107 sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize);
1109 sb->st_flags = vap->va_flags;
1110 if (priv_check(td, PRIV_VFS_GENERATION))
1113 sb->st_gen = vap->va_gen;
1115 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1120 * File table vnode ioctl routine.
1123 vn_ioctl(fp, com, data, active_cred, td)
1127 struct ucred *active_cred;
1130 struct vnode *vp = fp->f_vnode;
1135 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1137 switch (vp->v_type) {
1140 if (com == FIONREAD) {
1141 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1142 error = VOP_GETATTR(vp, &vattr, active_cred);
1145 *(int *)data = vattr.va_size - fp->f_offset;
1147 if (com == FIONBIO || com == FIOASYNC) /* XXX */
1150 error = VOP_IOCTL(vp, com, data, fp->f_flag,
1157 VFS_UNLOCK_GIANT(vfslocked);
1162 * File table vnode poll routine.
1165 vn_poll(fp, events, active_cred, td)
1168 struct ucred *active_cred;
1176 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1178 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1179 error = mac_vnode_check_poll(active_cred, fp->f_cred, vp);
1184 error = VOP_POLL(vp, events, fp->f_cred, td);
1185 VFS_UNLOCK_GIANT(vfslocked);
1190 * Acquire the requested lock and then check for validity. LK_RETRY
1191 * permits vn_lock to return doomed vnodes.
1194 _vn_lock(struct vnode *vp, int flags, char *file, int line)
1198 VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
1199 ("vn_lock called with no locktype."));
1201 #ifdef DEBUG_VFS_LOCKS
1202 KASSERT(vp->v_holdcnt != 0,
1203 ("vn_lock %p: zero hold count", vp));
1205 error = VOP_LOCK1(vp, flags, file, line);
1206 flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */
1207 KASSERT((flags & LK_RETRY) == 0 || error == 0,
1208 ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)",
1211 * Callers specify LK_RETRY if they wish to get dead vnodes.
1212 * If RETRY is not set, we return ENOENT instead.
1214 if (error == 0 && vp->v_iflag & VI_DOOMED &&
1215 (flags & LK_RETRY) == 0) {
1220 } while (flags & LK_RETRY && error != 0);
1225 * File table vnode close routine.
1228 vn_closefile(fp, td)
1239 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1240 if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) {
1241 lf.l_whence = SEEK_SET;
1244 lf.l_type = F_UNLCK;
1245 (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK);
1248 fp->f_ops = &badfileops;
1250 error = vn_close(vp, fp->f_flag, fp->f_cred, td);
1251 VFS_UNLOCK_GIANT(vfslocked);
1256 * Preparing to start a filesystem write operation. If the operation is
1257 * permitted, then we bump the count of operations in progress and
1258 * proceed. If a suspend request is in progress, we wait until the
1259 * suspension is over, and then proceed.
1262 vn_start_write(vp, mpp, flags)
1272 * If a vnode is provided, get and return the mount point that
1273 * to which it will write.
1276 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1278 if (error != EOPNOTSUPP)
1283 if ((mp = *mpp) == NULL)
1287 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1289 * As long as a vnode is not provided we need to acquire a
1290 * refcount for the provided mountpoint too, in order to
1291 * emulate a vfs_ref().
1298 * Check on status of suspension.
1300 if ((curthread->td_pflags & TDP_IGNSUSP) == 0 ||
1301 mp->mnt_susp_owner != curthread) {
1302 while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1303 if (flags & V_NOWAIT) {
1304 error = EWOULDBLOCK;
1307 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1308 (PUSER - 1) | (flags & PCATCH), "suspfs", 0);
1313 if (flags & V_XSLEEP)
1315 mp->mnt_writeopcount++;
1317 if (error != 0 || (flags & V_XSLEEP) != 0)
1324 * Secondary suspension. Used by operations such as vop_inactive
1325 * routines that are needed by the higher level functions. These
1326 * are allowed to proceed until all the higher level functions have
1327 * completed (indicated by mnt_writeopcount dropping to zero). At that
1328 * time, these operations are halted until the suspension is over.
1331 vn_start_secondary_write(vp, mpp, flags)
1341 if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) {
1343 if (error != EOPNOTSUPP)
1349 * If we are not suspended or have not yet reached suspended
1350 * mode, then let the operation proceed.
1352 if ((mp = *mpp) == NULL)
1356 * VOP_GETWRITEMOUNT() returns with the mp refcount held through
1358 * As long as a vnode is not provided we need to acquire a
1359 * refcount for the provided mountpoint too, in order to
1360 * emulate a vfs_ref().
1365 if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) {
1366 mp->mnt_secondary_writes++;
1367 mp->mnt_secondary_accwrites++;
1371 if (flags & V_NOWAIT) {
1374 return (EWOULDBLOCK);
1377 * Wait for the suspension to finish.
1379 error = msleep(&mp->mnt_flag, MNT_MTX(mp),
1380 (PUSER - 1) | (flags & PCATCH) | PDROP, "suspfs", 0);
1388 * Filesystem write operation has completed. If we are suspending and this
1389 * operation is the last one, notify the suspender that the suspension is
1393 vn_finished_write(mp)
1400 mp->mnt_writeopcount--;
1401 if (mp->mnt_writeopcount < 0)
1402 panic("vn_finished_write: neg cnt");
1403 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1404 mp->mnt_writeopcount <= 0)
1405 wakeup(&mp->mnt_writeopcount);
1411 * Filesystem secondary write operation has completed. If we are
1412 * suspending and this operation is the last one, notify the suspender
1413 * that the suspension is now in effect.
1416 vn_finished_secondary_write(mp)
1423 mp->mnt_secondary_writes--;
1424 if (mp->mnt_secondary_writes < 0)
1425 panic("vn_finished_secondary_write: neg cnt");
1426 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 &&
1427 mp->mnt_secondary_writes <= 0)
1428 wakeup(&mp->mnt_secondary_writes);
1435 * Request a filesystem to suspend write operations.
1438 vfs_write_suspend(mp)
1444 if (mp->mnt_susp_owner == curthread) {
1448 while (mp->mnt_kern_flag & MNTK_SUSPEND)
1449 msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0);
1450 mp->mnt_kern_flag |= MNTK_SUSPEND;
1451 mp->mnt_susp_owner = curthread;
1452 if (mp->mnt_writeopcount > 0)
1453 (void) msleep(&mp->mnt_writeopcount,
1454 MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0);
1457 if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0)
1458 vfs_write_resume(mp);
1463 * Request a filesystem to resume write operations.
1466 vfs_write_resume(mp)
1471 if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) {
1472 KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner"));
1473 mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 |
1475 mp->mnt_susp_owner = NULL;
1476 wakeup(&mp->mnt_writeopcount);
1477 wakeup(&mp->mnt_flag);
1478 curthread->td_pflags &= ~TDP_IGNSUSP;
1486 * Implement kqueues for files by translating it to vnode operation.
1489 vn_kqfilter(struct file *fp, struct knote *kn)
1494 vfslocked = VFS_LOCK_GIANT(fp->f_vnode->v_mount);
1495 error = VOP_KQFILTER(fp->f_vnode, kn);
1496 VFS_UNLOCK_GIANT(vfslocked);
1502 * Simplified in-kernel wrapper calls for extended attribute access.
1503 * Both calls pass in a NULL credential, authorizing as "kernel" access.
1504 * Set IO_NODELOCKED in ioflg if the vnode is already locked.
1507 vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace,
1508 const char *attrname, int *buflen, char *buf, struct thread *td)
1514 iov.iov_len = *buflen;
1517 auio.uio_iov = &iov;
1518 auio.uio_iovcnt = 1;
1519 auio.uio_rw = UIO_READ;
1520 auio.uio_segflg = UIO_SYSSPACE;
1522 auio.uio_offset = 0;
1523 auio.uio_resid = *buflen;
1525 if ((ioflg & IO_NODELOCKED) == 0)
1526 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1528 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1530 /* authorize attribute retrieval as kernel */
1531 error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL,
1534 if ((ioflg & IO_NODELOCKED) == 0)
1538 *buflen = *buflen - auio.uio_resid;
1545 * XXX failure mode if partially written?
1548 vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace,
1549 const char *attrname, int buflen, char *buf, struct thread *td)
1556 iov.iov_len = buflen;
1559 auio.uio_iov = &iov;
1560 auio.uio_iovcnt = 1;
1561 auio.uio_rw = UIO_WRITE;
1562 auio.uio_segflg = UIO_SYSSPACE;
1564 auio.uio_offset = 0;
1565 auio.uio_resid = buflen;
1567 if ((ioflg & IO_NODELOCKED) == 0) {
1568 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1570 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1573 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1575 /* authorize attribute setting as kernel */
1576 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td);
1578 if ((ioflg & IO_NODELOCKED) == 0) {
1579 vn_finished_write(mp);
1587 vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace,
1588 const char *attrname, struct thread *td)
1593 if ((ioflg & IO_NODELOCKED) == 0) {
1594 if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0)
1596 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1599 ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held");
1601 /* authorize attribute removal as kernel */
1602 error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td);
1603 if (error == EOPNOTSUPP)
1604 error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL,
1607 if ((ioflg & IO_NODELOCKED) == 0) {
1608 vn_finished_write(mp);
1616 vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp)
1622 ltype = VOP_ISLOCKED(vp);
1623 KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED,
1624 ("vn_vget_ino: vp not locked"));
1625 error = vfs_busy(mp, MBF_NOWAIT);
1629 error = vfs_busy(mp, 0);
1630 vn_lock(vp, ltype | LK_RETRY);
1634 if (vp->v_iflag & VI_DOOMED) {
1640 error = VFS_VGET(mp, ino, lkflags, rvp);
1642 vn_lock(vp, ltype | LK_RETRY);
1643 if (vp->v_iflag & VI_DOOMED) {
1652 vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio,
1653 const struct thread *td)
1656 if (vp->v_type != VREG || td == NULL)
1658 PROC_LOCK(td->td_proc);
1659 if ((uoff_t)uio->uio_offset + uio->uio_resid >
1660 lim_cur(td->td_proc, RLIMIT_FSIZE)) {
1661 kern_psignal(td->td_proc, SIGXFSZ);
1662 PROC_UNLOCK(td->td_proc);
1665 PROC_UNLOCK(td->td_proc);
1670 vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
1674 int error, vfslocked;
1677 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1679 vn_lock(vp, LK_SHARED | LK_RETRY);
1680 AUDIT_ARG_VNODE1(vp);
1683 error = setfmode(td, active_cred, vp, mode);
1684 VFS_UNLOCK_GIANT(vfslocked);
1689 vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
1693 int error, vfslocked;
1696 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
1698 vn_lock(vp, LK_SHARED | LK_RETRY);
1699 AUDIT_ARG_VNODE1(vp);
1702 error = setfown(td, active_cred, vp, uid, gid);
1703 VFS_UNLOCK_GIANT(vfslocked);
1708 vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end)
1712 if ((object = vp->v_object) == NULL)
1714 VM_OBJECT_LOCK(object);
1715 vm_object_page_remove(object, start, end, 0);
1716 VM_OBJECT_UNLOCK(object);
1720 vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred)
1728 KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA,
1729 ("Wrong command %lu", cmd));
1731 if (vn_lock(vp, LK_SHARED) != 0)
1733 if (vp->v_type != VREG) {
1737 error = VOP_GETATTR(vp, &va, cred);
1741 if (noff >= va.va_size) {
1745 bsize = vp->v_mount->mnt_stat.f_iosize;
1746 for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) {
1747 error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL);
1748 if (error == EOPNOTSUPP) {
1752 if ((bnp == -1 && cmd == FIOSEEKHOLE) ||
1753 (bnp != -1 && cmd == FIOSEEKDATA)) {
1760 if (noff > va.va_size)
1762 /* noff == va.va_size. There is an implicit hole at the end of file. */
1763 if (cmd == FIOSEEKDATA)