1 /* $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $ */
4 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
34 * tmpfs vnode interface.
36 #include <sys/cdefs.h>
38 #include <sys/kernel.h>
39 #include <sys/kern_syscall.h>
40 #include <sys/param.h>
41 #include <sys/fcntl.h>
42 #include <sys/lockf.h>
45 #include <sys/resourcevar.h>
46 #include <sys/sched.h>
47 #include <sys/sfbuf.h>
49 #include <sys/systm.h>
50 #include <sys/unistd.h>
51 #include <sys/vfsops.h>
52 #include <sys/vnode.h>
54 #include <sys/mplock2.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_pager.h>
61 #include <vfs/fifofs/fifo.h>
62 #include <vfs/tmpfs/tmpfs_vnops.h>
63 #include <vfs/tmpfs/tmpfs.h>
65 MALLOC_DECLARE(M_TMPFS
);
67 /* --------------------------------------------------------------------- */
70 tmpfs_nresolve(struct vop_nresolve_args
*v
)
72 struct vnode
*dvp
= v
->a_dvp
;
73 struct vnode
*vp
= NULL
;
74 struct namecache
*ncp
= v
->a_nch
->ncp
;
75 struct ucred
*cred
= v
->a_cred
;
78 struct tmpfs_dirent
*de
;
79 struct tmpfs_node
*dnode
;
81 dnode
= VP_TO_TMPFS_DIR(dvp
);
83 /* Check accessibility of requested node as a first step. */
84 error
= VOP_ACCESS(dvp
, VEXEC
, cred
);
88 if (dnode
->tn_dir
.tn_parent
== NULL
) {
93 de
= tmpfs_dir_lookup(dnode
, NULL
, ncp
);
95 /* The entry was not found in the directory.
96 * This is OK if we are creating or renaming an
97 * entry and are working on the last component of
99 error
= VOP_ACCESS(dvp
, VWRITE
, cred
);
107 struct tmpfs_node
*tnode
;
109 /* The entry was found, so get its associated
113 /* If we are not at the last path component and
114 * found a non-directory or non-link entry (which
115 * may itself be pointing to a directory), raise
117 if (tnode
->tn_links
> 1 &&
118 tnode
->tn_type
!= VDIR
&& tnode
->tn_type
!= VLNK
) {
123 error
= VOP_ACCESS(dvp
, VWRITE
, cred
);
127 /* Allocate a new vnode on the matching entry. */
128 error
= tmpfs_alloc_vp(dvp
->v_mount
, tnode
,
129 LK_EXCLUSIVE
| LK_RETRY
, &vp
);
133 if ((dnode
->tn_mode
& S_ISTXT
) &&
134 VOP_ACCESS(vp
, VWRITE
, cred
)) {
144 /* Store the result of this lookup in the cache. Avoid this if the
145 * request was for creation, as it does not improve timings on
149 cache_setvp(v
->a_nch
, vp
);
152 if (error
== ENOENT
) {
153 cache_setvp(v
->a_nch
, NULL
);
159 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args
*v
)
161 struct vnode
*dvp
= v
->a_dvp
;
162 struct vnode
**vpp
= v
->a_vpp
;
163 struct tmpfs_node
*dnode
= VP_TO_TMPFS_NODE(dvp
);
164 struct ucred
*cred
= v
->a_cred
;
168 /* Check accessibility of requested node as a first step. */
169 error
= VOP_ACCESS(dvp
, VEXEC
, cred
);
173 if (dnode
->tn_dir
.tn_parent
!= NULL
) {
174 /* Allocate a new vnode on the matching entry. */
175 error
= tmpfs_alloc_vp(dvp
->v_mount
, dnode
->tn_dir
.tn_parent
,
176 LK_EXCLUSIVE
| LK_RETRY
, vpp
);
182 return (*vpp
== NULL
) ? ENOENT
: 0;
185 /* --------------------------------------------------------------------- */
188 tmpfs_ncreate(struct vop_ncreate_args
*v
)
190 struct vnode
*dvp
= v
->a_dvp
;
191 struct vnode
**vpp
= v
->a_vpp
;
192 struct namecache
*ncp
= v
->a_nch
->ncp
;
193 struct vattr
*vap
= v
->a_vap
;
194 struct ucred
*cred
= v
->a_cred
;
197 KKASSERT(vap
->va_type
== VREG
|| vap
->va_type
== VSOCK
);
199 error
= tmpfs_alloc_file(dvp
, vpp
, vap
, ncp
, cred
, NULL
);
201 cache_setunresolved(v
->a_nch
);
202 cache_setvp(v
->a_nch
, *vpp
);
207 /* --------------------------------------------------------------------- */
210 tmpfs_nmknod(struct vop_nmknod_args
*v
)
212 struct vnode
*dvp
= v
->a_dvp
;
213 struct vnode
**vpp
= v
->a_vpp
;
214 struct namecache
*ncp
= v
->a_nch
->ncp
;
215 struct vattr
*vap
= v
->a_vap
;
216 struct ucred
*cred
= v
->a_cred
;
219 if (vap
->va_type
!= VBLK
&& vap
->va_type
!= VCHR
&&
220 vap
->va_type
!= VFIFO
)
223 error
= tmpfs_alloc_file(dvp
, vpp
, vap
, ncp
, cred
, NULL
);
225 cache_setunresolved(v
->a_nch
);
226 cache_setvp(v
->a_nch
, *vpp
);
232 /* --------------------------------------------------------------------- */
235 tmpfs_open(struct vop_open_args
*v
)
237 struct vnode
*vp
= v
->a_vp
;
238 int mode
= v
->a_mode
;
241 struct tmpfs_node
*node
;
243 node
= VP_TO_TMPFS_NODE(vp
);
245 /* The file is still active but all its names have been removed
246 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as
247 * it is about to die. */
248 if (node
->tn_links
< 1)
251 /* If the file is marked append-only, deny write requests. */
252 if ((node
->tn_flags
& APPEND
) &&
253 (mode
& (FWRITE
| O_APPEND
)) == FWRITE
) {
256 return (vop_stdopen(v
));
261 /* --------------------------------------------------------------------- */
264 tmpfs_close(struct vop_close_args
*v
)
266 struct vnode
*vp
= v
->a_vp
;
267 struct tmpfs_node
*node
;
269 node
= VP_TO_TMPFS_NODE(vp
);
271 if (node
->tn_links
> 0) {
272 /* Update node times. No need to do it if the node has
273 * been deleted, because it will vanish after we return. */
277 return vop_stdclose(v
);
280 /* --------------------------------------------------------------------- */
283 tmpfs_access(struct vop_access_args
*v
)
285 struct vnode
*vp
= v
->a_vp
;
287 struct tmpfs_node
*node
;
289 node
= VP_TO_TMPFS_NODE(vp
);
291 switch (vp
->v_type
) {
297 if (VWRITE
&& vp
->v_mount
->mnt_flag
& MNT_RDONLY
) {
317 if (VWRITE
&& node
->tn_flags
& IMMUTABLE
) {
322 error
= vop_helper_access(v
, node
->tn_uid
, node
->tn_gid
, node
->tn_mode
, 0);
329 /* --------------------------------------------------------------------- */
332 tmpfs_getattr(struct vop_getattr_args
*v
)
334 struct vnode
*vp
= v
->a_vp
;
335 struct vattr
*vap
= v
->a_vap
;
336 struct tmpfs_node
*node
;
338 node
= VP_TO_TMPFS_NODE(vp
);
342 vap
->va_type
= vp
->v_type
;
343 vap
->va_mode
= node
->tn_mode
;
344 vap
->va_nlink
= node
->tn_links
;
345 vap
->va_uid
= node
->tn_uid
;
346 vap
->va_gid
= node
->tn_gid
;
347 vap
->va_fsid
= vp
->v_mount
->mnt_stat
.f_fsid
.val
[0];
348 vap
->va_fileid
= node
->tn_id
;
349 vap
->va_size
= node
->tn_size
;
350 vap
->va_blocksize
= PAGE_SIZE
;
351 vap
->va_atime
.tv_sec
= node
->tn_atime
;
352 vap
->va_atime
.tv_nsec
= node
->tn_atimensec
;
353 vap
->va_mtime
.tv_sec
= node
->tn_mtime
;
354 vap
->va_mtime
.tv_nsec
= node
->tn_mtimensec
;
355 vap
->va_ctime
.tv_sec
= node
->tn_ctime
;
356 vap
->va_ctime
.tv_nsec
= node
->tn_ctimensec
;
357 vap
->va_gen
= node
->tn_gen
;
358 vap
->va_flags
= node
->tn_flags
;
359 if (vp
->v_type
== VBLK
|| vp
->v_type
== VCHR
)
361 vap
->va_rmajor
= umajor(node
->tn_rdev
);
362 vap
->va_rminor
= uminor(node
->tn_rdev
);
364 vap
->va_bytes
= round_page(node
->tn_size
);
370 /* --------------------------------------------------------------------- */
373 tmpfs_setattr(struct vop_setattr_args
*v
)
375 struct vnode
*vp
= v
->a_vp
;
376 struct vattr
*vap
= v
->a_vap
;
377 struct ucred
*cred
= v
->a_cred
;
380 /* Abort if any unsettable attribute is given. */
381 if (vap
->va_type
!= VNON
||
382 vap
->va_nlink
!= VNOVAL
||
383 vap
->va_fsid
!= VNOVAL
||
384 vap
->va_fileid
!= VNOVAL
||
385 vap
->va_blocksize
!= VNOVAL
||
386 vap
->va_gen
!= VNOVAL
||
387 vap
->va_rmajor
!= VNOVAL
||
388 vap
->va_bytes
!= VNOVAL
)
391 if (error
== 0 && (vap
->va_flags
!= VNOVAL
))
392 error
= tmpfs_chflags(vp
, vap
->va_flags
, cred
);
394 if (error
== 0 && (vap
->va_size
!= VNOVAL
))
395 error
= tmpfs_chsize(vp
, vap
->va_size
, cred
);
397 if (error
== 0 && (vap
->va_uid
!= VNOVAL
|| vap
->va_gid
!= VNOVAL
))
398 error
= tmpfs_chown(vp
, vap
->va_uid
, vap
->va_gid
, cred
);
400 if (error
== 0 && (vap
->va_mode
!= (mode_t
)VNOVAL
))
401 error
= tmpfs_chmod(vp
, vap
->va_mode
, cred
);
403 if (error
== 0 && ((vap
->va_atime
.tv_sec
!= VNOVAL
&&
404 vap
->va_atime
.tv_nsec
!= VNOVAL
) ||
405 (vap
->va_mtime
.tv_sec
!= VNOVAL
&&
406 vap
->va_mtime
.tv_nsec
!= VNOVAL
) ))
407 error
= tmpfs_chtimes(vp
, &vap
->va_atime
, &vap
->va_mtime
,
408 vap
->va_vaflags
, cred
);
410 /* Update the node times. We give preference to the error codes
411 * generated by this function rather than the ones that may arise
412 * from tmpfs_update. */
418 /* --------------------------------------------------------------------- */
421 * fsync is usually a NOP, but we must take action when unmounting or
425 tmpfs_fsync(struct vop_fsync_args
*v
)
427 struct tmpfs_mount
*tmp
;
428 struct tmpfs_node
*node
;
429 struct vnode
*vp
= v
->a_vp
;
431 tmp
= VFS_TO_TMPFS(vp
->v_mount
);
432 node
= VP_TO_TMPFS_NODE(vp
);
435 if (vp
->v_type
== VREG
) {
436 if (tmp
->tm_flags
& TMPFS_FLAG_UNMOUNTING
) {
437 tmpfs_truncate(vp
, 0);
438 } else if (vp
->v_flag
& VRECLAIMED
) {
439 if (node
->tn_links
== 0)
440 tmpfs_truncate(vp
, 0);
442 vfsync(v
->a_vp
, v
->a_waitfor
, 1, NULL
, NULL
);
448 /* --------------------------------------------------------------------- */
451 tmpfs_read (struct vop_read_args
*ap
)
454 struct vnode
*vp
= ap
->a_vp
;
455 struct uio
*uio
= ap
->a_uio
;
456 struct tmpfs_node
*node
;
464 if (uio
->uio_resid
== 0) {
468 node
= VP_TO_TMPFS_NODE(vp
);
470 if (uio
->uio_offset
< 0)
472 if (vp
->v_type
!= VREG
)
476 if(curthread
->td_mpcount
)
484 while (uio
->uio_resid
> 0 && uio
->uio_offset
< node
->tn_size
) {
486 * Use buffer cache I/O (via tmpfs_strategy)
488 offset
= (size_t)uio
->uio_offset
& BMASK
;
489 base_offset
= (off_t
)uio
->uio_offset
- offset
;
490 bp
= getcacheblk(vp
, base_offset
);
493 if (got_mplock
== 0) {
498 error
= bread(vp
, base_offset
, BSIZE
, &bp
);
501 kprintf("tmpfs_read bread error %d\n", error
);
506 if (got_mplock
== 0) {
512 * Figure out how many bytes we can actually copy this loop.
514 len
= BSIZE
- offset
;
515 if (len
> uio
->uio_resid
)
516 len
= uio
->uio_resid
;
517 if (len
> node
->tn_size
- uio
->uio_offset
)
518 len
= (size_t)(node
->tn_size
- uio
->uio_offset
);
520 error
= uiomove((char *)bp
->b_data
+ offset
, len
, uio
);
523 kprintf("tmpfs_read uiomove error %d\n", error
);
531 TMPFS_NODE_LOCK(node
);
532 node
->tn_status
|= TMPFS_NODE_ACCESSED
;
533 TMPFS_NODE_UNLOCK(node
);
539 tmpfs_write (struct vop_write_args
*ap
)
542 struct vnode
*vp
= ap
->a_vp
;
543 struct uio
*uio
= ap
->a_uio
;
544 struct thread
*td
= uio
->uio_td
;
545 struct tmpfs_node
*node
;
557 if (uio
->uio_resid
== 0) {
561 node
= VP_TO_TMPFS_NODE(vp
);
563 if (vp
->v_type
!= VREG
)
566 oldsize
= node
->tn_size
;
567 if (ap
->a_ioflag
& IO_APPEND
)
568 uio
->uio_offset
= node
->tn_size
;
571 * Check for illegal write offsets.
573 if (uio
->uio_offset
+ uio
->uio_resid
>
574 VFS_TO_TMPFS(vp
->v_mount
)->tm_maxfilesize
)
577 if (vp
->v_type
== VREG
&& td
!= NULL
) {
578 error
= kern_getrlimit(RLIMIT_FSIZE
, &limit
);
581 if (uio
->uio_offset
+ uio
->uio_resid
> limit
.rlim_cur
) {
582 ksignal(td
->td_proc
, SIGXFSZ
);
589 * Extend the file's size if necessary
591 extended
= ((uio
->uio_offset
+ uio
->uio_resid
) > node
->tn_size
);
594 if (curthread
->td_mpcount
) {
603 while (uio
->uio_resid
> 0) {
605 * Use buffer cache I/O (via tmpfs_strategy)
607 offset
= (size_t)uio
->uio_offset
& BMASK
;
608 base_offset
= (off_t
)uio
->uio_offset
- offset
;
609 len
= BSIZE
- offset
;
610 if (len
> uio
->uio_resid
)
611 len
= uio
->uio_resid
;
613 if ((uio
->uio_offset
+ len
) > node
->tn_size
) {
614 trivial
= (uio
->uio_offset
<= node
->tn_size
);
615 error
= tmpfs_reg_resize(vp
, uio
->uio_offset
+ len
, trivial
);
621 * Read to fill in any gaps. Theoretically we could
622 * optimize this if the write covers the entire buffer
623 * and is not a UIO_NOCOPY write, however this can lead
624 * to a security violation exposing random kernel memory
625 * (whatever junk was in the backing VM pages before).
627 * So just use bread() to do the right thing.
629 error
= bread(vp
, base_offset
, BSIZE
, &bp
);
630 error
= uiomove((char *)bp
->b_data
+ offset
, len
, uio
);
632 kprintf("tmpfs_write uiomove error %d\n", error
);
637 if (uio
->uio_offset
> node
->tn_size
)
638 node
->tn_size
= uio
->uio_offset
;
641 * The data has been loaded into the buffer, write it out.
643 * We want tmpfs to be able to use all available ram, not
644 * just the buffer cache, so if not explicitly paging we
645 * use buwrite() to leave the buffer clean but mark all the
646 * VM pages valid+dirty.
648 * Be careful, the kernel uses IO_SYNC/IO_ASYNC when it
649 * really really wants to flush the buffer, typically as
650 * part of the paging system, and we'll loop forever if
653 if (ap
->a_ioflag
& IO_SYNC
)
655 else if (ap
->a_ioflag
& IO_ASYNC
)
661 kprintf("tmpfs_write bwrite error %d\n", error
);
671 (void)tmpfs_reg_resize(vp
, oldsize
, trivial
);
675 TMPFS_NODE_LOCK(node
);
676 node
->tn_status
|= TMPFS_NODE_ACCESSED
| TMPFS_NODE_MODIFIED
|
677 (extended
? TMPFS_NODE_CHANGED
: 0);
679 if (node
->tn_mode
& (S_ISUID
| S_ISGID
)) {
680 if (priv_check_cred(ap
->a_cred
, PRIV_VFS_RETAINSUGID
, 0))
681 node
->tn_mode
&= ~(S_ISUID
| S_ISGID
);
683 TMPFS_NODE_UNLOCK(node
);
689 tmpfs_advlock (struct vop_advlock_args
*ap
)
691 struct tmpfs_node
*node
;
692 struct vnode
*vp
= ap
->a_vp
;
694 node
= VP_TO_TMPFS_NODE(vp
);
696 return (lf_advlock(ap
, &node
->tn_advlock
, node
->tn_size
));
701 tmpfs_strategy(struct vop_strategy_args
*ap
)
703 struct bio
*bio
= ap
->a_bio
;
704 struct buf
*bp
= bio
->bio_buf
;
705 struct vnode
*vp
= ap
->a_vp
;
706 struct tmpfs_node
*node
;
709 if (vp
->v_type
!= VREG
) {
710 bp
->b_resid
= bp
->b_bcount
;
711 bp
->b_flags
|= B_ERROR
| B_INVAL
;
712 bp
->b_error
= EINVAL
;
717 node
= VP_TO_TMPFS_NODE(vp
);
719 uobj
= node
->tn_reg
.tn_aobj
;
722 * Call swap_pager_strategy to read or write between the VM
723 * object and the buffer cache.
725 swap_pager_strategy(uobj
, bio
);
731 tmpfs_bmap(struct vop_bmap_args
*ap
)
733 if (ap
->a_doffsetp
!= NULL
)
734 *ap
->a_doffsetp
= ap
->a_loffset
;
735 if (ap
->a_runp
!= NULL
)
737 if (ap
->a_runb
!= NULL
)
743 /* --------------------------------------------------------------------- */
746 tmpfs_nremove(struct vop_nremove_args
*v
)
748 struct vnode
*dvp
= v
->a_dvp
;
749 struct namecache
*ncp
= v
->a_nch
->ncp
;
752 struct tmpfs_dirent
*de
;
753 struct tmpfs_mount
*tmp
;
754 struct tmpfs_node
*dnode
;
755 struct tmpfs_node
*node
;
758 * We have to acquire the vp from v->a_nch because
759 * we will likely unresolve the namecache entry, and
760 * a vrele is needed to trigger the tmpfs_inactive/tmpfs_reclaim
761 * sequence to recover space from the file.
763 error
= cache_vref(v
->a_nch
, v
->a_cred
, &vp
);
764 KKASSERT(error
== 0);
766 if (vp
->v_type
== VDIR
) {
771 dnode
= VP_TO_TMPFS_DIR(dvp
);
772 node
= VP_TO_TMPFS_NODE(vp
);
773 tmp
= VFS_TO_TMPFS(vp
->v_mount
);
774 de
= tmpfs_dir_lookup(dnode
, node
, ncp
);
780 /* Files marked as immutable or append-only cannot be deleted. */
781 if ((node
->tn_flags
& (IMMUTABLE
| APPEND
| NOUNLINK
)) ||
782 (dnode
->tn_flags
& APPEND
)) {
787 /* Remove the entry from the directory; as it is a file, we do not
788 * have to change the number of hard links of the directory. */
789 tmpfs_dir_detach(dnode
, de
);
791 /* Free the directory entry we just deleted. Note that the node
792 * referred by it will not be removed until the vnode is really
794 tmpfs_free_dirent(tmp
, de
);
796 if (node
->tn_links
> 0) {
797 TMPFS_NODE_LOCK(node
);
798 node
->tn_status
|= TMPFS_NODE_ACCESSED
| TMPFS_NODE_CHANGED
| \
800 TMPFS_NODE_UNLOCK(node
);
803 cache_setunresolved(v
->a_nch
);
804 cache_setvp(v
->a_nch
, NULL
);
805 /*cache_inval_vp(vp, CINV_DESTROY);*/
814 /* --------------------------------------------------------------------- */
817 tmpfs_nlink(struct vop_nlink_args
*v
)
819 struct vnode
*dvp
= v
->a_dvp
;
820 struct vnode
*vp
= v
->a_vp
;
821 struct namecache
*ncp
= v
->a_nch
->ncp
;
822 struct tmpfs_dirent
*de
;
823 struct tmpfs_node
*node
;
824 struct tmpfs_node
*dnode
;
827 KKASSERT(dvp
!= vp
); /* XXX When can this be false? */
829 node
= VP_TO_TMPFS_NODE(vp
);
830 dnode
= VP_TO_TMPFS_NODE(dvp
);
832 /* XXX: Why aren't the following two tests done by the caller? */
834 /* Hard links of directories are forbidden. */
835 if (vp
->v_type
== VDIR
) {
840 /* Cannot create cross-device links. */
841 if (dvp
->v_mount
!= vp
->v_mount
) {
846 /* Ensure that we do not overflow the maximum number of links imposed
848 KKASSERT(node
->tn_links
<= LINK_MAX
);
849 if (node
->tn_links
== LINK_MAX
) {
854 /* We cannot create links of files marked immutable or append-only. */
855 if (node
->tn_flags
& (IMMUTABLE
| APPEND
)) {
860 /* Allocate a new directory entry to represent the node. */
861 error
= tmpfs_alloc_dirent(VFS_TO_TMPFS(vp
->v_mount
), node
,
862 ncp
->nc_name
, ncp
->nc_nlen
, &de
);
866 /* Insert the new directory entry into the appropriate directory. */
867 tmpfs_dir_attach(dnode
, de
);
869 /* vp link count has changed, so update node times. */
871 TMPFS_NODE_LOCK(node
);
872 node
->tn_status
|= TMPFS_NODE_CHANGED
;
873 TMPFS_NODE_UNLOCK(node
);
876 cache_setunresolved(v
->a_nch
);
877 cache_setvp(v
->a_nch
, vp
);
884 /* --------------------------------------------------------------------- */
887 tmpfs_nrename(struct vop_nrename_args
*v
)
889 struct vnode
*fdvp
= v
->a_fdvp
;
890 struct namecache
*fncp
= v
->a_fnch
->ncp
;
891 struct vnode
*fvp
= fncp
->nc_vp
;
892 struct vnode
*tdvp
= v
->a_tdvp
;
893 struct namecache
*tncp
= v
->a_tnch
->ncp
;
894 struct vnode
*tvp
= tncp
->nc_vp
;
895 struct tmpfs_dirent
*de
;
896 struct tmpfs_mount
*tmp
;
897 struct tmpfs_node
*fdnode
;
898 struct tmpfs_node
*fnode
;
899 struct tmpfs_node
*tnode
;
900 struct tmpfs_node
*tdnode
;
904 tnode
= (tvp
== NULL
) ? NULL
: VP_TO_TMPFS_NODE(tvp
);
906 /* Disallow cross-device renames.
907 * XXX Why isn't this done by the caller? */
908 if (fvp
->v_mount
!= tdvp
->v_mount
||
909 (tvp
!= NULL
&& fvp
->v_mount
!= tvp
->v_mount
)) {
914 tmp
= VFS_TO_TMPFS(tdvp
->v_mount
);
915 tdnode
= VP_TO_TMPFS_DIR(tdvp
);
917 /* If source and target are the same file, there is nothing to do. */
923 fdnode
= VP_TO_TMPFS_DIR(fdvp
);
924 fnode
= VP_TO_TMPFS_NODE(fvp
);
925 de
= tmpfs_dir_lookup(fdnode
, fnode
, fncp
);
927 /* Avoid manipulating '.' and '..' entries. */
932 KKASSERT(de
->td_node
== fnode
);
934 /* If re-naming a directory to another preexisting directory
935 * ensure that the target directory is empty so that its
936 * removal causes no side effects.
937 * Kern_rename gurantees the destination to be a directory
938 * if the source is one. */
940 KKASSERT(tnode
!= NULL
);
942 if ((tnode
->tn_flags
& (NOUNLINK
| IMMUTABLE
| APPEND
)) ||
943 (tdnode
->tn_flags
& (APPEND
| IMMUTABLE
))) {
948 if (fnode
->tn_type
== VDIR
&& tnode
->tn_type
== VDIR
) {
949 if (tnode
->tn_size
> 0) {
953 } else if (fnode
->tn_type
== VDIR
&& tnode
->tn_type
!= VDIR
) {
956 } else if (fnode
->tn_type
!= VDIR
&& tnode
->tn_type
== VDIR
) {
960 KKASSERT(fnode
->tn_type
!= VDIR
&&
961 tnode
->tn_type
!= VDIR
);
965 if ((fnode
->tn_flags
& (NOUNLINK
| IMMUTABLE
| APPEND
))
966 || (fdnode
->tn_flags
& (APPEND
| IMMUTABLE
))) {
971 /* Ensure that we have enough memory to hold the new name, if it
972 * has to be changed. */
973 if (fncp
->nc_nlen
!= tncp
->nc_nlen
||
974 bcmp(fncp
->nc_name
, tncp
->nc_name
, fncp
->nc_nlen
) != 0) {
975 newname
= kmalloc(tncp
->nc_nlen
+ 1, M_TMPFSNAME
, M_WAITOK
);
979 /* If the node is being moved to another directory, we have to do
981 if (fdnode
!= tdnode
) {
982 /* In case we are moving a directory, we have to adjust its
983 * parent to point to the new parent. */
984 if (de
->td_node
->tn_type
== VDIR
) {
985 struct tmpfs_node
*n
;
987 /* Ensure the target directory is not a child of the
988 * directory being moved. Otherwise, we'd end up
989 * with stale nodes. */
991 /* TMPFS_LOCK garanties that no nodes are freed while
992 * traversing the list. Nodes can only be marked as
993 * removed: tn_parent == NULL. */
996 while (n
!= n
->tn_dir
.tn_parent
) {
997 struct tmpfs_node
*parent
;
1000 TMPFS_NODE_UNLOCK(n
);
1003 if (newname
!= NULL
)
1004 kfree(newname
, M_TMPFSNAME
);
1007 parent
= n
->tn_dir
.tn_parent
;
1008 if (parent
== NULL
) {
1012 TMPFS_NODE_LOCK(parent
);
1013 if (parent
->tn_dir
.tn_parent
== NULL
) {
1014 TMPFS_NODE_UNLOCK(parent
);
1020 TMPFS_NODE_UNLOCK(n
);
1024 if (newname
!= NULL
)
1025 kfree(newname
, M_TMPFSNAME
);
1029 /* Adjust the parent pointer. */
1030 TMPFS_VALIDATE_DIR(fnode
);
1031 TMPFS_NODE_LOCK(de
->td_node
);
1032 de
->td_node
->tn_dir
.tn_parent
= tdnode
;
1034 /* As a result of changing the target of the '..'
1035 * entry, the link count of the source and target
1036 * directories has to be adjusted. */
1037 TMPFS_NODE_LOCK(tdnode
);
1038 TMPFS_ASSERT_LOCKED(tdnode
);
1039 TMPFS_NODE_LOCK(fdnode
);
1040 TMPFS_ASSERT_LOCKED(fdnode
);
1045 TMPFS_NODE_UNLOCK(fdnode
);
1046 TMPFS_NODE_UNLOCK(tdnode
);
1047 TMPFS_NODE_UNLOCK(de
->td_node
);
1050 /* Do the move: just remove the entry from the source directory
1051 * and insert it into the target one. */
1052 tmpfs_dir_detach(fdnode
, de
);
1053 tmpfs_dir_attach(tdnode
, de
);
1056 /* If the name has changed, we need to make it effective by changing
1057 * it in the directory entry. */
1058 if (newname
!= NULL
) {
1060 kfree(de
->td_name
, M_TMPFSNAME
);
1061 de
->td_namelen
= (uint16_t)tncp
->nc_nlen
;
1062 bcopy(tncp
->nc_name
, newname
, tncp
->nc_nlen
);
1063 newname
[tncp
->nc_nlen
] = '\0';
1064 de
->td_name
= newname
;
1066 TMPFS_NODE_LOCK(tdnode
);
1067 TMPFS_NODE_LOCK(fdnode
);
1069 fnode
->tn_status
|= TMPFS_NODE_CHANGED
;
1070 tdnode
->tn_status
|= TMPFS_NODE_MODIFIED
;
1072 TMPFS_NODE_UNLOCK(fdnode
);
1073 TMPFS_NODE_UNLOCK(tdnode
);
1076 /* If we are overwriting an entry, we have to remove the old one
1077 * from the target directory. */
1079 /* Remove the old entry from the target directory. */
1080 de
= tmpfs_dir_lookup(tdnode
, tnode
, tncp
);
1081 tmpfs_dir_detach(tdnode
, de
);
1083 /* Free the directory entry we just deleted. Note that the
1084 * node referred by it will not be removed until the vnode is
1085 * really reclaimed. */
1086 tmpfs_free_dirent(VFS_TO_TMPFS(tvp
->v_mount
), de
);
1087 /*cache_inval_vp(tvp, CINV_DESTROY);*/
1090 cache_rename(v
->a_fnch
, v
->a_tnch
);
1097 /* Release target nodes. */
1098 /* XXX: I don't understand when tdvp can be the same as tvp, but
1099 * other code takes care of this... */
1106 /* --------------------------------------------------------------------- */
1109 tmpfs_nmkdir(struct vop_nmkdir_args
*v
)
1111 struct vnode
*dvp
= v
->a_dvp
;
1112 struct vnode
**vpp
= v
->a_vpp
;
1113 struct namecache
*ncp
= v
->a_nch
->ncp
;
1114 struct vattr
*vap
= v
->a_vap
;
1115 struct ucred
*cred
= v
->a_cred
;
1118 KKASSERT(vap
->va_type
== VDIR
);
1120 error
= tmpfs_alloc_file(dvp
, vpp
, vap
, ncp
, cred
, NULL
);
1122 cache_setunresolved(v
->a_nch
);
1123 cache_setvp(v
->a_nch
, *vpp
);
1129 /* --------------------------------------------------------------------- */
1132 tmpfs_nrmdir(struct vop_nrmdir_args
*v
)
1134 struct vnode
*dvp
= v
->a_dvp
;
1135 struct namecache
*ncp
= v
->a_nch
->ncp
;
1139 struct tmpfs_dirent
*de
;
1140 struct tmpfs_mount
*tmp
;
1141 struct tmpfs_node
*dnode
;
1142 struct tmpfs_node
*node
;
1145 * We have to acquire the vp from v->a_nch because
1146 * we will likely unresolve the namecache entry, and
1147 * a vrele is needed to trigger the tmpfs_inactive/tmpfs_reclaim
1150 error
= cache_vref(v
->a_nch
, v
->a_cred
, &vp
);
1151 KKASSERT(error
== 0);
1153 tmp
= VFS_TO_TMPFS(dvp
->v_mount
);
1154 dnode
= VP_TO_TMPFS_DIR(dvp
);
1155 node
= VP_TO_TMPFS_DIR(vp
);
1157 /* Directories with more than two entries ('.' and '..') cannot be
1159 if (node
->tn_size
> 0) {
1164 if ((dnode
->tn_flags
& APPEND
)
1165 || (node
->tn_flags
& (NOUNLINK
| IMMUTABLE
| APPEND
))) {
1170 /* This invariant holds only if we are not trying to remove "..".
1171 * We checked for that above so this is safe now. */
1172 KKASSERT(node
->tn_dir
.tn_parent
== dnode
);
1174 /* Get the directory entry associated with node (vp). This was
1175 * filled by tmpfs_lookup while looking up the entry. */
1176 de
= tmpfs_dir_lookup(dnode
, node
, ncp
);
1177 KKASSERT(TMPFS_DIRENT_MATCHES(de
,
1181 /* Check flags to see if we are allowed to remove the directory. */
1182 if (dnode
->tn_flags
& APPEND
1183 || node
->tn_flags
& (NOUNLINK
| IMMUTABLE
| APPEND
)) {
1189 /* Detach the directory entry from the directory (dnode). */
1190 tmpfs_dir_detach(dnode
, de
);
1192 /* No vnode should be allocated for this entry from this point */
1193 TMPFS_NODE_LOCK(node
);
1194 TMPFS_ASSERT_ELOCKED(node
);
1195 TMPFS_NODE_LOCK(dnode
);
1196 TMPFS_ASSERT_ELOCKED(dnode
);
1199 /* handled by tmpfs_free_node */
1200 KKASSERT(node
->tn_links
> 0);
1202 node
->tn_dir
.tn_parent
= NULL
;
1204 node
->tn_status
|= TMPFS_NODE_ACCESSED
| TMPFS_NODE_CHANGED
| \
1205 TMPFS_NODE_MODIFIED
;
1208 /* handled by tmpfs_free_node */
1209 KKASSERT(dnode
->tn_links
> 0);
1212 dnode
->tn_status
|= TMPFS_NODE_ACCESSED
| \
1213 TMPFS_NODE_CHANGED
| TMPFS_NODE_MODIFIED
;
1215 TMPFS_NODE_UNLOCK(dnode
);
1216 TMPFS_NODE_UNLOCK(node
);
1218 /* Free the directory entry we just deleted. Note that the node
1219 * referred by it will not be removed until the vnode is really
1221 tmpfs_free_dirent(tmp
, de
);
1223 /* Release the deleted vnode (will destroy the node, notify
1224 * interested parties and clean it from the cache). */
1226 TMPFS_NODE_LOCK(dnode
);
1227 dnode
->tn_status
|= TMPFS_NODE_CHANGED
;
1228 TMPFS_NODE_UNLOCK(dnode
);
1231 cache_setunresolved(v
->a_nch
);
1232 cache_setvp(v
->a_nch
, NULL
);
1233 /*cache_inval_vp(vp, CINV_DESTROY);*/
1242 /* --------------------------------------------------------------------- */
1245 tmpfs_nsymlink(struct vop_nsymlink_args
*v
)
1247 struct vnode
*dvp
= v
->a_dvp
;
1248 struct vnode
**vpp
= v
->a_vpp
;
1249 struct namecache
*ncp
= v
->a_nch
->ncp
;
1250 struct vattr
*vap
= v
->a_vap
;
1251 struct ucred
*cred
= v
->a_cred
;
1252 char *target
= v
->a_target
;
1255 vap
->va_type
= VLNK
;
1256 error
= tmpfs_alloc_file(dvp
, vpp
, vap
, ncp
, cred
, target
);
1258 cache_setunresolved(v
->a_nch
);
1259 cache_setvp(v
->a_nch
, *vpp
);
1265 /* --------------------------------------------------------------------- */
1268 tmpfs_readdir(struct vop_readdir_args
*v
)
1270 struct vnode
*vp
= v
->a_vp
;
1271 struct uio
*uio
= v
->a_uio
;
1272 int *eofflag
= v
->a_eofflag
;
1273 off_t
**cookies
= v
->a_cookies
;
1274 int *ncookies
= v
->a_ncookies
;
1275 struct tmpfs_mount
*tmp
;
1279 struct tmpfs_node
*node
;
1281 /* This operation only makes sense on directory nodes. */
1282 if (vp
->v_type
!= VDIR
)
1285 tmp
= VFS_TO_TMPFS(vp
->v_mount
);
1286 node
= VP_TO_TMPFS_DIR(vp
);
1287 startoff
= uio
->uio_offset
;
1289 if (uio
->uio_offset
== TMPFS_DIRCOOKIE_DOT
) {
1290 error
= tmpfs_dir_getdotdent(node
, uio
);
1296 if (uio
->uio_offset
== TMPFS_DIRCOOKIE_DOTDOT
) {
1297 error
= tmpfs_dir_getdotdotdent(tmp
, node
, uio
);
1303 error
= tmpfs_dir_getdents(node
, uio
, &cnt
);
1306 KKASSERT(error
>= -1);
1311 if (eofflag
!= NULL
)
1313 (error
== 0 && uio
->uio_offset
== TMPFS_DIRCOOKIE_EOF
);
1315 /* Update NFS-related variables. */
1316 if (error
== 0 && cookies
!= NULL
&& ncookies
!= NULL
) {
1318 off_t off
= startoff
;
1319 struct tmpfs_dirent
*de
= NULL
;
1322 *cookies
= kmalloc(cnt
* sizeof(off_t
), M_TEMP
, M_WAITOK
);
1324 for (i
= 0; i
< cnt
; i
++) {
1325 KKASSERT(off
!= TMPFS_DIRCOOKIE_EOF
);
1326 if (off
== TMPFS_DIRCOOKIE_DOT
) {
1327 off
= TMPFS_DIRCOOKIE_DOTDOT
;
1329 if (off
== TMPFS_DIRCOOKIE_DOTDOT
) {
1330 de
= TAILQ_FIRST(&node
->tn_dir
.tn_dirhead
);
1331 } else if (de
!= NULL
) {
1332 de
= TAILQ_NEXT(de
, td_entries
);
1334 de
= tmpfs_dir_lookupbycookie(node
,
1336 KKASSERT(de
!= NULL
);
1337 de
= TAILQ_NEXT(de
, td_entries
);
1340 off
= TMPFS_DIRCOOKIE_EOF
;
1342 off
= tmpfs_dircookie(de
);
1345 (*cookies
)[i
] = off
;
1347 KKASSERT(uio
->uio_offset
== off
);
1353 /* --------------------------------------------------------------------- */
1356 tmpfs_readlink(struct vop_readlink_args
*v
)
1358 struct vnode
*vp
= v
->a_vp
;
1359 struct uio
*uio
= v
->a_uio
;
1362 struct tmpfs_node
*node
;
1364 KKASSERT(uio
->uio_offset
== 0);
1365 KKASSERT(vp
->v_type
== VLNK
);
1367 node
= VP_TO_TMPFS_NODE(vp
);
1369 error
= uiomove(node
->tn_link
, MIN(node
->tn_size
, uio
->uio_resid
),
1371 TMPFS_NODE_LOCK(node
);
1372 node
->tn_status
|= TMPFS_NODE_ACCESSED
;
1373 TMPFS_NODE_UNLOCK(node
);
1378 /* --------------------------------------------------------------------- */
1381 tmpfs_inactive(struct vop_inactive_args
*v
)
1383 struct vnode
*vp
= v
->a_vp
;
1385 struct tmpfs_node
*node
;
1387 node
= VP_TO_TMPFS_NODE(vp
);
1390 * Get rid of unreferenced deleted vnodes sooner rather than
1391 * later so the data memory can be recovered immediately.
1393 TMPFS_NODE_LOCK(node
);
1394 if (node
->tn_links
== 0 &&
1395 (node
->tn_vpstate
& TMPFS_VNODE_ALLOCATING
) == 0) {
1396 node
->tn_vpstate
= TMPFS_VNODE_DOOMED
;
1397 TMPFS_NODE_UNLOCK(node
);
1400 TMPFS_NODE_UNLOCK(node
);
1406 /* --------------------------------------------------------------------- */
1409 tmpfs_reclaim(struct vop_reclaim_args
*v
)
1411 struct vnode
*vp
= v
->a_vp
;
1412 struct tmpfs_mount
*tmp
;
1413 struct tmpfs_node
*node
;
1415 node
= VP_TO_TMPFS_NODE(vp
);
1416 tmp
= VFS_TO_TMPFS(vp
->v_mount
);
1420 /* If the node referenced by this vnode was deleted by the user,
1421 * we must free its associated data structures (now that the vnode
1422 * is being reclaimed). */
1423 TMPFS_NODE_LOCK(node
);
1424 if (node
->tn_links
== 0 &&
1425 (node
->tn_vpstate
& TMPFS_VNODE_ALLOCATING
) == 0) {
1426 node
->tn_vpstate
= TMPFS_VNODE_DOOMED
;
1427 tmpfs_free_node(tmp
, node
);
1430 TMPFS_NODE_UNLOCK(node
);
1433 KKASSERT(vp
->v_data
== NULL
);
1437 /* --------------------------------------------------------------------- */
1440 tmpfs_print(struct vop_print_args
*v
)
1442 struct vnode
*vp
= v
->a_vp
;
1444 struct tmpfs_node
*node
;
1446 node
= VP_TO_TMPFS_NODE(vp
);
1448 kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1449 node
, node
->tn_flags
, node
->tn_links
);
1450 kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1451 node
->tn_mode
, node
->tn_uid
, node
->tn_gid
,
1452 (uintmax_t)node
->tn_size
, node
->tn_status
);
1454 if (vp
->v_type
== VFIFO
)
1462 /* --------------------------------------------------------------------- */
1465 tmpfs_pathconf(struct vop_pathconf_args
*v
)
1467 int name
= v
->a_name
;
1468 register_t
*retval
= v
->a_retval
;
1491 case _PC_CHOWN_RESTRICTED
:
1503 case _PC_FILESIZEBITS
:
1504 *retval
= 0; /* XXX Don't know which value should I return. */
1514 /* --------------------------------------------------------------------- */
1517 * vnode operations vector used for files stored in a tmpfs file system.
1519 struct vop_ops tmpfs_vnode_vops
= {
1520 .vop_default
= vop_defaultop
,
1521 .vop_getpages
= vop_stdgetpages
,
1522 .vop_putpages
= vop_stdputpages
,
1523 .vop_ncreate
= tmpfs_ncreate
,
1524 .vop_nresolve
= tmpfs_nresolve
,
1525 .vop_nlookupdotdot
= tmpfs_nlookupdotdot
,
1526 .vop_nmknod
= tmpfs_nmknod
,
1527 .vop_open
= tmpfs_open
,
1528 .vop_close
= tmpfs_close
,
1529 .vop_access
= tmpfs_access
,
1530 .vop_getattr
= tmpfs_getattr
,
1531 .vop_setattr
= tmpfs_setattr
,
1532 .vop_read
= tmpfs_read
,
1533 .vop_write
= tmpfs_write
,
1534 .vop_fsync
= tmpfs_fsync
,
1535 .vop_nremove
= tmpfs_nremove
,
1536 .vop_nlink
= tmpfs_nlink
,
1537 .vop_nrename
= tmpfs_nrename
,
1538 .vop_nmkdir
= tmpfs_nmkdir
,
1539 .vop_nrmdir
= tmpfs_nrmdir
,
1540 .vop_nsymlink
= tmpfs_nsymlink
,
1541 .vop_readdir
= tmpfs_readdir
,
1542 .vop_readlink
= tmpfs_readlink
,
1543 .vop_inactive
= tmpfs_inactive
,
1544 .vop_reclaim
= tmpfs_reclaim
,
1545 .vop_print
= tmpfs_print
,
1546 .vop_pathconf
= tmpfs_pathconf
,
1547 .vop_bmap
= tmpfs_bmap
,
1548 .vop_bmap
= (void *)vop_eopnotsupp
,
1549 .vop_strategy
= tmpfs_strategy
,
1550 .vop_advlock
= tmpfs_advlock
,