2 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc.
5 * This code is derived from software contributed to The NetBSD Foundation
6 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
19 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
20 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
21 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
22 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28 * POSSIBILITY OF SUCH DAMAGE.
30 * $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $
34 * tmpfs vnode interface.
37 #include <sys/kernel.h>
38 #include <sys/kern_syscall.h>
39 #include <sys/param.h>
40 #include <sys/fcntl.h>
41 #include <sys/lockf.h>
44 #include <sys/resourcevar.h>
45 #include <sys/sched.h>
47 #include <sys/systm.h>
48 #include <sys/unistd.h>
49 #include <sys/vfsops.h>
50 #include <sys/vnode.h>
52 #include <sys/mplock2.h>
55 #include <vm/vm_object.h>
56 #include <vm/vm_page.h>
57 #include <vm/vm_pager.h>
58 #include <vm/swap_pager.h>
60 #include <vfs/fifofs/fifo.h>
61 #include <vfs/tmpfs/tmpfs_vnops.h>
62 #include <vfs/tmpfs/tmpfs.h>
64 MALLOC_DECLARE(M_TMPFS);
68 tmpfs_knote(struct vnode *vp, int flags)
71 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
75 /* --------------------------------------------------------------------- */
78 tmpfs_nresolve(struct vop_nresolve_args *v)
80 struct vnode *dvp = v->a_dvp;
81 struct vnode *vp = NULL;
82 struct namecache *ncp = v->a_nch->ncp;
83 struct tmpfs_node *tnode;
86 struct tmpfs_dirent *de;
87 struct tmpfs_node *dnode;
89 dnode = VP_TO_TMPFS_DIR(dvp);
91 de = tmpfs_dir_lookup(dnode, NULL, ncp);
96 * Allocate a vnode for the node we found.
99 error = tmpfs_alloc_vp(dvp->v_mount, tnode,
100 LK_EXCLUSIVE | LK_RETRY, &vp);
108 * Store the result of this lookup in the cache. Avoid this if the
109 * request was for creation, as it does not improve timings on
114 cache_setvp(v->a_nch, vp);
116 } else if (error == ENOENT) {
117 cache_setvp(v->a_nch, NULL);
123 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v)
125 struct vnode *dvp = v->a_dvp;
126 struct vnode **vpp = v->a_vpp;
127 struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp);
128 struct ucred *cred = v->a_cred;
132 /* Check accessibility of requested node as a first step. */
133 error = VOP_ACCESS(dvp, VEXEC, cred);
137 if (dnode->tn_dir.tn_parent != NULL) {
138 /* Allocate a new vnode on the matching entry. */
139 error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent,
140 LK_EXCLUSIVE | LK_RETRY, vpp);
146 return (*vpp == NULL) ? ENOENT : 0;
149 /* --------------------------------------------------------------------- */
152 tmpfs_ncreate(struct vop_ncreate_args *v)
154 struct vnode *dvp = v->a_dvp;
155 struct vnode **vpp = v->a_vpp;
156 struct namecache *ncp = v->a_nch->ncp;
157 struct vattr *vap = v->a_vap;
158 struct ucred *cred = v->a_cred;
161 KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK);
163 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
165 cache_setunresolved(v->a_nch);
166 cache_setvp(v->a_nch, *vpp);
167 tmpfs_knote(dvp, NOTE_WRITE);
172 /* --------------------------------------------------------------------- */
175 tmpfs_nmknod(struct vop_nmknod_args *v)
177 struct vnode *dvp = v->a_dvp;
178 struct vnode **vpp = v->a_vpp;
179 struct namecache *ncp = v->a_nch->ncp;
180 struct vattr *vap = v->a_vap;
181 struct ucred *cred = v->a_cred;
184 if (vap->va_type != VBLK && vap->va_type != VCHR &&
185 vap->va_type != VFIFO)
188 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
190 cache_setunresolved(v->a_nch);
191 cache_setvp(v->a_nch, *vpp);
192 tmpfs_knote(dvp, NOTE_WRITE);
198 /* --------------------------------------------------------------------- */
201 tmpfs_open(struct vop_open_args *v)
203 struct vnode *vp = v->a_vp;
204 int mode = v->a_mode;
207 struct tmpfs_node *node;
209 node = VP_TO_TMPFS_NODE(vp);
211 /* The file is still active but all its names have been removed
212 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as
213 * it is about to die. */
214 if (node->tn_links < 1)
217 /* If the file is marked append-only, deny write requests. */
218 if ((node->tn_flags & APPEND) &&
219 (mode & (FWRITE | O_APPEND)) == FWRITE) {
222 return (vop_stdopen(v));
227 /* --------------------------------------------------------------------- */
230 tmpfs_close(struct vop_close_args *v)
232 struct vnode *vp = v->a_vp;
233 struct tmpfs_node *node;
235 node = VP_TO_TMPFS_NODE(vp);
237 if (node->tn_links > 0) {
238 /* Update node times. No need to do it if the node has
239 * been deleted, because it will vanish after we return. */
243 return vop_stdclose(v);
246 /* --------------------------------------------------------------------- */
249 tmpfs_access(struct vop_access_args *v)
251 struct vnode *vp = v->a_vp;
253 struct tmpfs_node *node;
255 node = VP_TO_TMPFS_NODE(vp);
257 switch (vp->v_type) {
263 if ((v->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
283 if ((v->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) {
288 error = vop_helper_access(v, node->tn_uid, node->tn_gid, node->tn_mode, 0);
295 /* --------------------------------------------------------------------- */
298 tmpfs_getattr(struct vop_getattr_args *v)
300 struct vnode *vp = v->a_vp;
301 struct vattr *vap = v->a_vap;
302 struct tmpfs_node *node;
304 node = VP_TO_TMPFS_NODE(vp);
306 lwkt_gettoken(&vp->v_mount->mnt_token);
309 vap->va_type = vp->v_type;
310 vap->va_mode = node->tn_mode;
311 vap->va_nlink = node->tn_links;
312 vap->va_uid = node->tn_uid;
313 vap->va_gid = node->tn_gid;
314 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
315 vap->va_fileid = node->tn_id;
316 vap->va_size = node->tn_size;
317 vap->va_blocksize = PAGE_SIZE;
318 vap->va_atime.tv_sec = node->tn_atime;
319 vap->va_atime.tv_nsec = node->tn_atimensec;
320 vap->va_mtime.tv_sec = node->tn_mtime;
321 vap->va_mtime.tv_nsec = node->tn_mtimensec;
322 vap->va_ctime.tv_sec = node->tn_ctime;
323 vap->va_ctime.tv_nsec = node->tn_ctimensec;
324 vap->va_gen = node->tn_gen;
325 vap->va_flags = node->tn_flags;
326 if (vp->v_type == VBLK || vp->v_type == VCHR)
328 vap->va_rmajor = umajor(node->tn_rdev);
329 vap->va_rminor = uminor(node->tn_rdev);
331 vap->va_bytes = round_page(node->tn_size);
334 lwkt_reltoken(&vp->v_mount->mnt_token);
339 /* --------------------------------------------------------------------- */
342 tmpfs_setattr(struct vop_setattr_args *v)
344 struct vnode *vp = v->a_vp;
345 struct vattr *vap = v->a_vap;
346 struct ucred *cred = v->a_cred;
347 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
351 if (error == 0 && (vap->va_flags != VNOVAL)) {
352 error = tmpfs_chflags(vp, vap->va_flags, cred);
353 kflags |= NOTE_ATTRIB;
356 if (error == 0 && (vap->va_size != VNOVAL)) {
357 if (vap->va_size > node->tn_size)
358 kflags |= NOTE_WRITE | NOTE_EXTEND;
360 kflags |= NOTE_WRITE;
361 error = tmpfs_chsize(vp, vap->va_size, cred);
364 if (error == 0 && (vap->va_uid != (uid_t)VNOVAL ||
365 vap->va_gid != (gid_t)VNOVAL)) {
366 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred);
367 kflags |= NOTE_ATTRIB;
370 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) {
371 error = tmpfs_chmod(vp, vap->va_mode, cred);
372 kflags |= NOTE_ATTRIB;
375 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
376 vap->va_atime.tv_nsec != VNOVAL) ||
377 (vap->va_mtime.tv_sec != VNOVAL &&
378 vap->va_mtime.tv_nsec != VNOVAL) )) {
379 error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime,
380 vap->va_vaflags, cred);
381 kflags |= NOTE_ATTRIB;
384 /* Update the node times. We give preference to the error codes
385 * generated by this function rather than the ones that may arise
386 * from tmpfs_update. */
388 tmpfs_knote(vp, kflags);
393 /* --------------------------------------------------------------------- */
396 * fsync is usually a NOP, but we must take action when unmounting or
400 tmpfs_fsync(struct vop_fsync_args *v)
402 struct tmpfs_mount *tmp;
403 struct tmpfs_node *node;
404 struct vnode *vp = v->a_vp;
406 tmp = VFS_TO_TMPFS(vp->v_mount);
407 node = VP_TO_TMPFS_NODE(vp);
410 if (vp->v_type == VREG) {
411 if (vp->v_flag & VRECLAIMED) {
412 if (node->tn_links == 0)
413 tmpfs_truncate(vp, 0);
415 vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL);
421 /* --------------------------------------------------------------------- */
424 tmpfs_read (struct vop_read_args *ap)
427 struct vnode *vp = ap->a_vp;
428 struct uio *uio = ap->a_uio;
429 struct tmpfs_node *node;
436 if (uio->uio_resid == 0) {
440 node = VP_TO_TMPFS_NODE(vp);
442 if (uio->uio_offset < 0)
444 if (vp->v_type != VREG)
447 while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) {
449 * Use buffer cache I/O (via tmpfs_strategy)
451 offset = (size_t)uio->uio_offset & BMASK;
452 base_offset = (off_t)uio->uio_offset - offset;
453 bp = getcacheblk(vp, base_offset, BSIZE);
456 lwkt_gettoken(&vp->v_mount->mnt_token);
457 error = bread(vp, base_offset, BSIZE, &bp);
460 lwkt_reltoken(&vp->v_mount->mnt_token);
461 kprintf("tmpfs_read bread error %d\n", error);
464 lwkt_reltoken(&vp->v_mount->mnt_token);
468 * Figure out how many bytes we can actually copy this loop.
470 len = BSIZE - offset;
471 if (len > uio->uio_resid)
472 len = uio->uio_resid;
473 if (len > node->tn_size - uio->uio_offset)
474 len = (size_t)(node->tn_size - uio->uio_offset);
476 error = uiomove((char *)bp->b_data + offset, len, uio);
479 kprintf("tmpfs_read uiomove error %d\n", error);
484 TMPFS_NODE_LOCK(node);
485 node->tn_status |= TMPFS_NODE_ACCESSED;
486 TMPFS_NODE_UNLOCK(node);
492 tmpfs_write (struct vop_write_args *ap)
495 struct vnode *vp = ap->a_vp;
496 struct uio *uio = ap->a_uio;
497 struct thread *td = uio->uio_td;
498 struct tmpfs_node *node;
510 if (uio->uio_resid == 0) {
514 node = VP_TO_TMPFS_NODE(vp);
516 if (vp->v_type != VREG)
519 lwkt_gettoken(&vp->v_mount->mnt_token);
521 oldsize = node->tn_size;
522 if (ap->a_ioflag & IO_APPEND)
523 uio->uio_offset = node->tn_size;
526 * Check for illegal write offsets.
528 if (uio->uio_offset + uio->uio_resid >
529 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) {
530 lwkt_reltoken(&vp->v_mount->mnt_token);
534 if (vp->v_type == VREG && td != NULL) {
535 error = kern_getrlimit(RLIMIT_FSIZE, &limit);
537 lwkt_reltoken(&vp->v_mount->mnt_token);
540 if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) {
541 ksignal(td->td_proc, SIGXFSZ);
542 lwkt_reltoken(&vp->v_mount->mnt_token);
549 * Extend the file's size if necessary
551 extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size);
555 while (uio->uio_resid > 0) {
557 * Use buffer cache I/O (via tmpfs_strategy)
559 offset = (size_t)uio->uio_offset & BMASK;
560 base_offset = (off_t)uio->uio_offset - offset;
561 len = BSIZE - offset;
562 if (len > uio->uio_resid)
563 len = uio->uio_resid;
565 if ((uio->uio_offset + len) > node->tn_size) {
566 trivial = (uio->uio_offset <= node->tn_size);
567 error = tmpfs_reg_resize(vp, uio->uio_offset + len, trivial);
573 * Read to fill in any gaps. Theoretically we could
574 * optimize this if the write covers the entire buffer
575 * and is not a UIO_NOCOPY write, however this can lead
576 * to a security violation exposing random kernel memory
577 * (whatever junk was in the backing VM pages before).
579 * So just use bread() to do the right thing.
581 error = bread(vp, base_offset, BSIZE, &bp);
582 error = uiomove((char *)bp->b_data + offset, len, uio);
584 kprintf("tmpfs_write uiomove error %d\n", error);
589 if (uio->uio_offset > node->tn_size) {
590 node->tn_size = uio->uio_offset;
591 kflags |= NOTE_EXTEND;
593 kflags |= NOTE_WRITE;
596 * The data has been loaded into the buffer, write it out.
598 * We want tmpfs to be able to use all available ram, not
599 * just the buffer cache, so if not explicitly paging we
600 * use buwrite() to leave the buffer clean but mark all the
601 * VM pages valid+dirty.
603 * When the kernel is paging, either via normal pageout
604 * operation or when cleaning the object during a recycle,
605 * the underlying VM pages are going to get thrown away
606 * so we MUST write them to swap.
608 * XXX unfortunately this catches msync() system calls too
611 if (vm_swap_size == 0) {
613 * if swap isn't configured yet, force a buwrite() to
614 * avoid problems further down the line, due to flushing
619 if (ap->a_ioflag & IO_SYNC) {
621 } else if ((ap->a_ioflag & IO_ASYNC) ||
622 (uio->uio_segflg == UIO_NOCOPY)) {
630 kprintf("tmpfs_write bwrite error %d\n", bp->b_error);
639 (void)tmpfs_reg_resize(vp, oldsize, trivial);
640 kflags &= ~NOTE_EXTEND;
645 TMPFS_NODE_LOCK(node);
646 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
647 (extended? TMPFS_NODE_CHANGED : 0);
649 if (node->tn_mode & (S_ISUID | S_ISGID)) {
650 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0))
651 node->tn_mode &= ~(S_ISUID | S_ISGID);
653 TMPFS_NODE_UNLOCK(node);
656 tmpfs_knote(vp, kflags);
659 lwkt_reltoken(&vp->v_mount->mnt_token);
664 tmpfs_advlock (struct vop_advlock_args *ap)
666 struct tmpfs_node *node;
667 struct vnode *vp = ap->a_vp;
669 node = VP_TO_TMPFS_NODE(vp);
671 return (lf_advlock(ap, &node->tn_advlock, node->tn_size));
675 tmpfs_strategy(struct vop_strategy_args *ap)
677 struct bio *bio = ap->a_bio;
678 struct buf *bp = bio->bio_buf;
679 struct vnode *vp = ap->a_vp;
680 struct tmpfs_node *node;
683 if (vp->v_type != VREG) {
684 bp->b_resid = bp->b_bcount;
685 bp->b_flags |= B_ERROR | B_INVAL;
686 bp->b_error = EINVAL;
691 lwkt_gettoken(&vp->v_mount->mnt_token);
692 node = VP_TO_TMPFS_NODE(vp);
694 uobj = node->tn_reg.tn_aobj;
697 * Call swap_pager_strategy to read or write between the VM
698 * object and the buffer cache.
700 swap_pager_strategy(uobj, bio);
702 lwkt_reltoken(&vp->v_mount->mnt_token);
707 tmpfs_bmap(struct vop_bmap_args *ap)
709 if (ap->a_doffsetp != NULL)
710 *ap->a_doffsetp = ap->a_loffset;
711 if (ap->a_runp != NULL)
713 if (ap->a_runb != NULL)
719 /* --------------------------------------------------------------------- */
722 tmpfs_nremove(struct vop_nremove_args *v)
724 struct vnode *dvp = v->a_dvp;
725 struct namecache *ncp = v->a_nch->ncp;
728 struct tmpfs_dirent *de;
729 struct tmpfs_mount *tmp;
730 struct tmpfs_node *dnode;
731 struct tmpfs_node *node;
734 * We have to acquire the vp from v->a_nch because
735 * we will likely unresolve the namecache entry, and
736 * a vrele is needed to trigger the tmpfs_inactive/tmpfs_reclaim
737 * sequence to recover space from the file.
739 error = cache_vref(v->a_nch, v->a_cred, &vp);
740 KKASSERT(error == 0);
742 if (vp->v_type == VDIR) {
747 dnode = VP_TO_TMPFS_DIR(dvp);
748 node = VP_TO_TMPFS_NODE(vp);
749 tmp = VFS_TO_TMPFS(vp->v_mount);
750 de = tmpfs_dir_lookup(dnode, node, ncp);
756 /* Files marked as immutable or append-only cannot be deleted. */
757 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
758 (dnode->tn_flags & APPEND)) {
763 /* Remove the entry from the directory; as it is a file, we do not
764 * have to change the number of hard links of the directory. */
765 tmpfs_dir_detach(dnode, de);
767 /* Free the directory entry we just deleted. Note that the node
768 * referred by it will not be removed until the vnode is really
770 tmpfs_free_dirent(tmp, de);
772 if (node->tn_links > 0) {
773 TMPFS_NODE_LOCK(node);
774 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
776 TMPFS_NODE_UNLOCK(node);
779 cache_setunresolved(v->a_nch);
780 cache_setvp(v->a_nch, NULL);
781 tmpfs_knote(vp, NOTE_DELETE);
782 /*cache_inval_vp(vp, CINV_DESTROY);*/
783 tmpfs_knote(dvp, NOTE_WRITE);
792 /* --------------------------------------------------------------------- */
795 tmpfs_nlink(struct vop_nlink_args *v)
797 struct vnode *dvp = v->a_dvp;
798 struct vnode *vp = v->a_vp;
799 struct namecache *ncp = v->a_nch->ncp;
800 struct tmpfs_dirent *de;
801 struct tmpfs_node *node;
802 struct tmpfs_node *dnode;
805 KKASSERT(dvp != vp); /* XXX When can this be false? */
807 node = VP_TO_TMPFS_NODE(vp);
808 dnode = VP_TO_TMPFS_NODE(dvp);
810 /* XXX: Why aren't the following two tests done by the caller? */
812 /* Hard links of directories are forbidden. */
813 if (vp->v_type == VDIR) {
818 /* Cannot create cross-device links. */
819 if (dvp->v_mount != vp->v_mount) {
824 /* Ensure that we do not overflow the maximum number of links imposed
826 KKASSERT(node->tn_links <= LINK_MAX);
827 if (node->tn_links == LINK_MAX) {
832 /* We cannot create links of files marked immutable or append-only. */
833 if (node->tn_flags & (IMMUTABLE | APPEND)) {
838 /* Allocate a new directory entry to represent the node. */
839 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
840 ncp->nc_name, ncp->nc_nlen, &de);
844 /* Insert the new directory entry into the appropriate directory. */
845 tmpfs_dir_attach(dnode, de);
847 /* vp link count has changed, so update node times. */
849 TMPFS_NODE_LOCK(node);
850 node->tn_status |= TMPFS_NODE_CHANGED;
851 TMPFS_NODE_UNLOCK(node);
854 tmpfs_knote(vp, NOTE_LINK);
855 cache_setunresolved(v->a_nch);
856 cache_setvp(v->a_nch, vp);
857 tmpfs_knote(dvp, NOTE_WRITE);
864 /* --------------------------------------------------------------------- */
867 tmpfs_nrename(struct vop_nrename_args *v)
869 struct vnode *fdvp = v->a_fdvp;
870 struct namecache *fncp = v->a_fnch->ncp;
871 struct vnode *fvp = fncp->nc_vp;
872 struct vnode *tdvp = v->a_tdvp;
873 struct namecache *tncp = v->a_tnch->ncp;
874 struct vnode *tvp = tncp->nc_vp;
875 struct tmpfs_dirent *de;
876 struct tmpfs_mount *tmp;
877 struct tmpfs_node *fdnode;
878 struct tmpfs_node *fnode;
879 struct tmpfs_node *tnode;
880 struct tmpfs_node *tdnode;
885 tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
887 /* Disallow cross-device renames.
888 * XXX Why isn't this done by the caller? */
889 if (fvp->v_mount != tdvp->v_mount ||
890 (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
895 tmp = VFS_TO_TMPFS(tdvp->v_mount);
896 tdnode = VP_TO_TMPFS_DIR(tdvp);
898 /* If source and target are the same file, there is nothing to do. */
904 fdnode = VP_TO_TMPFS_DIR(fdvp);
905 fnode = VP_TO_TMPFS_NODE(fvp);
906 de = tmpfs_dir_lookup(fdnode, fnode, fncp);
908 /* Avoid manipulating '.' and '..' entries. */
913 KKASSERT(de->td_node == fnode);
916 * If replacing an entry in the target directory and that entry
917 * is a directory, it must be empty.
919 * Kern_rename gurantees the destination to be a directory
920 * if the source is one (it does?).
923 KKASSERT(tnode != NULL);
925 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
926 (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
931 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
932 if (tnode->tn_size > 0) {
936 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
939 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
943 KKASSERT(fnode->tn_type != VDIR &&
944 tnode->tn_type != VDIR);
948 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
949 (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
955 * Ensure that we have enough memory to hold the new name, if it
958 if (fncp->nc_nlen != tncp->nc_nlen ||
959 bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) {
960 newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone,
961 M_WAITOK | M_NULLOK);
962 if (newname == NULL) {
966 bcopy(tncp->nc_name, newname, tncp->nc_nlen);
967 newname[tncp->nc_nlen] = '\0';
973 * Unlink entry from source directory. Note that the kernel has
974 * already checked for illegal recursion cases (renaming a directory
975 * into a subdirectory of itself).
977 if (fdnode != tdnode)
978 tmpfs_dir_detach(fdnode, de);
981 * Handle any name change. Swap with newname, we will
982 * deallocate it at the end.
984 if (newname != NULL) {
986 TMPFS_NODE_LOCK(fnode);
987 fnode->tn_status |= TMPFS_NODE_CHANGED;
988 TMPFS_NODE_UNLOCK(fnode);
990 oldname = de->td_name;
991 de->td_name = newname;
992 de->td_namelen = (uint16_t)tncp->nc_nlen;
997 * Link entry to target directory. If the entry
998 * represents a directory move the parent linkage
1001 if (fdnode != tdnode) {
1002 if (de->td_node->tn_type == VDIR) {
1003 TMPFS_VALIDATE_DIR(fnode);
1005 TMPFS_NODE_LOCK(tdnode);
1007 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1008 TMPFS_NODE_UNLOCK(tdnode);
1010 TMPFS_NODE_LOCK(fnode);
1011 fnode->tn_dir.tn_parent = tdnode;
1012 fnode->tn_status |= TMPFS_NODE_CHANGED;
1013 TMPFS_NODE_UNLOCK(fnode);
1015 TMPFS_NODE_LOCK(fdnode);
1017 fdnode->tn_status |= TMPFS_NODE_MODIFIED;
1018 TMPFS_NODE_UNLOCK(fdnode);
1020 tmpfs_dir_attach(tdnode, de);
1022 TMPFS_NODE_LOCK(tdnode);
1023 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1024 TMPFS_NODE_UNLOCK(tdnode);
1028 * If we are overwriting an entry, we have to remove the old one
1029 * from the target directory.
1032 /* Remove the old entry from the target directory. */
1033 de = tmpfs_dir_lookup(tdnode, tnode, tncp);
1034 tmpfs_dir_detach(tdnode, de);
1035 tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE);
1038 * Free the directory entry we just deleted. Note that the
1039 * node referred by it will not be removed until the vnode is
1042 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), de);
1043 /*cache_inval_vp(tvp, CINV_DESTROY);*/
1050 kfree(newname, tmp->tm_name_zone);
1053 cache_rename(v->a_fnch, v->a_tnch);
1054 tmpfs_knote(v->a_fdvp, NOTE_WRITE);
1055 tmpfs_knote(v->a_tdvp, NOTE_WRITE);
1056 if (fnode->tn_vnode)
1057 tmpfs_knote(fnode->tn_vnode, NOTE_RENAME);
1064 /* Release target nodes. */
1065 /* XXX: I don't understand when tdvp can be the same as tvp, but
1066 * other code takes care of this... */
1073 /* --------------------------------------------------------------------- */
1076 tmpfs_nmkdir(struct vop_nmkdir_args *v)
1078 struct vnode *dvp = v->a_dvp;
1079 struct vnode **vpp = v->a_vpp;
1080 struct namecache *ncp = v->a_nch->ncp;
1081 struct vattr *vap = v->a_vap;
1082 struct ucred *cred = v->a_cred;
1085 KKASSERT(vap->va_type == VDIR);
1087 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL);
1089 cache_setunresolved(v->a_nch);
1090 cache_setvp(v->a_nch, *vpp);
1091 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1097 /* --------------------------------------------------------------------- */
1100 tmpfs_nrmdir(struct vop_nrmdir_args *v)
1102 struct vnode *dvp = v->a_dvp;
1103 struct namecache *ncp = v->a_nch->ncp;
1105 struct tmpfs_dirent *de;
1106 struct tmpfs_mount *tmp;
1107 struct tmpfs_node *dnode;
1108 struct tmpfs_node *node;
1112 * We have to acquire the vp from v->a_nch because
1113 * we will likely unresolve the namecache entry, and
1114 * a vrele is needed to trigger the tmpfs_inactive/tmpfs_reclaim
1117 error = cache_vref(v->a_nch, v->a_cred, &vp);
1118 KKASSERT(error == 0);
1121 * Prevalidate so we don't hit an assertion later
1123 if (vp->v_type != VDIR) {
1128 tmp = VFS_TO_TMPFS(dvp->v_mount);
1129 dnode = VP_TO_TMPFS_DIR(dvp);
1130 node = VP_TO_TMPFS_DIR(vp);
1132 /* Directories with more than two entries ('.' and '..') cannot be
1134 if (node->tn_size > 0) {
1139 if ((dnode->tn_flags & APPEND)
1140 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1145 /* This invariant holds only if we are not trying to remove "..".
1146 * We checked for that above so this is safe now. */
1147 KKASSERT(node->tn_dir.tn_parent == dnode);
1149 /* Get the directory entry associated with node (vp). This was
1150 * filled by tmpfs_lookup while looking up the entry. */
1151 de = tmpfs_dir_lookup(dnode, node, ncp);
1152 KKASSERT(TMPFS_DIRENT_MATCHES(de,
1156 /* Check flags to see if we are allowed to remove the directory. */
1157 if ((dnode->tn_flags & APPEND) ||
1158 node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) {
1164 /* Detach the directory entry from the directory (dnode). */
1165 tmpfs_dir_detach(dnode, de);
1167 /* No vnode should be allocated for this entry from this point */
1168 TMPFS_NODE_LOCK(node);
1169 TMPFS_ASSERT_ELOCKED(node);
1170 TMPFS_NODE_LOCK(dnode);
1171 TMPFS_ASSERT_ELOCKED(dnode);
1174 /* handled by tmpfs_free_node */
1175 KKASSERT(node->tn_links > 0);
1177 node->tn_dir.tn_parent = NULL;
1179 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1180 TMPFS_NODE_MODIFIED;
1183 /* handled by tmpfs_free_node */
1184 KKASSERT(dnode->tn_links > 0);
1187 dnode->tn_status |= TMPFS_NODE_ACCESSED | \
1188 TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1190 TMPFS_NODE_UNLOCK(dnode);
1191 TMPFS_NODE_UNLOCK(node);
1193 /* Free the directory entry we just deleted. Note that the node
1194 * referred by it will not be removed until the vnode is really
1196 tmpfs_free_dirent(tmp, de);
1198 /* Release the deleted vnode (will destroy the node, notify
1199 * interested parties and clean it from the cache). */
1201 TMPFS_NODE_LOCK(dnode);
1202 dnode->tn_status |= TMPFS_NODE_CHANGED;
1203 TMPFS_NODE_UNLOCK(dnode);
1206 cache_setunresolved(v->a_nch);
1207 cache_setvp(v->a_nch, NULL);
1208 /*cache_inval_vp(vp, CINV_DESTROY);*/
1209 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK);
1218 /* --------------------------------------------------------------------- */
1221 tmpfs_nsymlink(struct vop_nsymlink_args *v)
1223 struct vnode *dvp = v->a_dvp;
1224 struct vnode **vpp = v->a_vpp;
1225 struct namecache *ncp = v->a_nch->ncp;
1226 struct vattr *vap = v->a_vap;
1227 struct ucred *cred = v->a_cred;
1228 char *target = v->a_target;
1231 vap->va_type = VLNK;
1232 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target);
1234 tmpfs_knote(*vpp, NOTE_WRITE);
1235 cache_setunresolved(v->a_nch);
1236 cache_setvp(v->a_nch, *vpp);
1242 /* --------------------------------------------------------------------- */
1245 tmpfs_readdir(struct vop_readdir_args *v)
1247 struct vnode *vp = v->a_vp;
1248 struct uio *uio = v->a_uio;
1249 int *eofflag = v->a_eofflag;
1250 off_t **cookies = v->a_cookies;
1251 int *ncookies = v->a_ncookies;
1252 struct tmpfs_mount *tmp;
1256 struct tmpfs_node *node;
1258 /* This operation only makes sense on directory nodes. */
1259 if (vp->v_type != VDIR)
1262 tmp = VFS_TO_TMPFS(vp->v_mount);
1263 node = VP_TO_TMPFS_DIR(vp);
1264 startoff = uio->uio_offset;
1266 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) {
1267 error = tmpfs_dir_getdotdent(node, uio);
1273 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) {
1274 error = tmpfs_dir_getdotdotdent(tmp, node, uio);
1280 error = tmpfs_dir_getdents(node, uio, &cnt);
1283 KKASSERT(error >= -1);
1288 if (eofflag != NULL)
1290 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1292 /* Update NFS-related variables. */
1293 if (error == 0 && cookies != NULL && ncookies != NULL) {
1295 off_t off = startoff;
1296 struct tmpfs_dirent *de = NULL;
1299 *cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK);
1301 for (i = 0; i < cnt; i++) {
1302 KKASSERT(off != TMPFS_DIRCOOKIE_EOF);
1303 if (off == TMPFS_DIRCOOKIE_DOT) {
1304 off = TMPFS_DIRCOOKIE_DOTDOT;
1306 if (off == TMPFS_DIRCOOKIE_DOTDOT) {
1307 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
1308 } else if (de != NULL) {
1309 de = TAILQ_NEXT(de, td_entries);
1311 de = tmpfs_dir_lookupbycookie(node,
1313 KKASSERT(de != NULL);
1314 de = TAILQ_NEXT(de, td_entries);
1317 off = TMPFS_DIRCOOKIE_EOF;
1319 off = tmpfs_dircookie(de);
1322 (*cookies)[i] = off;
1324 KKASSERT(uio->uio_offset == off);
1330 /* --------------------------------------------------------------------- */
1333 tmpfs_readlink(struct vop_readlink_args *v)
1335 struct vnode *vp = v->a_vp;
1336 struct uio *uio = v->a_uio;
1339 struct tmpfs_node *node;
1341 KKASSERT(uio->uio_offset == 0);
1342 KKASSERT(vp->v_type == VLNK);
1344 node = VP_TO_TMPFS_NODE(vp);
1346 error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1348 TMPFS_NODE_LOCK(node);
1349 node->tn_status |= TMPFS_NODE_ACCESSED;
1350 TMPFS_NODE_UNLOCK(node);
1355 /* --------------------------------------------------------------------- */
1358 tmpfs_inactive(struct vop_inactive_args *v)
1360 struct vnode *vp = v->a_vp;
1362 struct tmpfs_node *node;
1364 node = VP_TO_TMPFS_NODE(vp);
1367 * Get rid of unreferenced deleted vnodes sooner rather than
1368 * later so the data memory can be recovered immediately.
1370 * We must truncate the vnode to prevent the normal reclamation
1371 * path from flushing the data for the removed file to disk.
1373 TMPFS_NODE_LOCK(node);
1374 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1375 (node->tn_links == 0 ||
1376 (node->tn_links == 1 && node->tn_type == VDIR &&
1377 node->tn_dir.tn_parent)))
1379 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1380 TMPFS_NODE_UNLOCK(node);
1381 if (node->tn_type == VREG)
1382 tmpfs_truncate(vp, 0);
1385 TMPFS_NODE_UNLOCK(node);
1391 /* --------------------------------------------------------------------- */
1394 tmpfs_reclaim(struct vop_reclaim_args *v)
1396 struct vnode *vp = v->a_vp;
1397 struct tmpfs_mount *tmp;
1398 struct tmpfs_node *node;
1400 node = VP_TO_TMPFS_NODE(vp);
1401 tmp = VFS_TO_TMPFS(vp->v_mount);
1406 * If the node referenced by this vnode was deleted by the
1407 * user, we must free its associated data structures now that
1408 * the vnode is being reclaimed.
1410 * Directories have an extra link ref.
1412 TMPFS_NODE_LOCK(node);
1413 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 &&
1414 (node->tn_links == 0 ||
1415 (node->tn_links == 1 && node->tn_type == VDIR &&
1416 node->tn_dir.tn_parent)))
1418 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1419 tmpfs_free_node(tmp, node);
1422 TMPFS_NODE_UNLOCK(node);
1425 KKASSERT(vp->v_data == NULL);
1429 /* --------------------------------------------------------------------- */
1432 tmpfs_print(struct vop_print_args *v)
1434 struct vnode *vp = v->a_vp;
1436 struct tmpfs_node *node;
1438 node = VP_TO_TMPFS_NODE(vp);
1440 kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n",
1441 node, node->tn_flags, node->tn_links);
1442 kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n",
1443 node->tn_mode, node->tn_uid, node->tn_gid,
1444 (uintmax_t)node->tn_size, node->tn_status);
1446 if (vp->v_type == VFIFO)
1454 /* --------------------------------------------------------------------- */
1457 tmpfs_pathconf(struct vop_pathconf_args *v)
1459 int name = v->a_name;
1460 register_t *retval = v->a_retval;
1483 case _PC_CHOWN_RESTRICTED:
1495 case _PC_FILESIZEBITS:
1496 *retval = 0; /* XXX Don't know which value should I return. */
1506 /************************************************************************
1508 ************************************************************************/
1510 static void filt_tmpfsdetach(struct knote *kn);
1511 static int filt_tmpfsread(struct knote *kn, long hint);
1512 static int filt_tmpfswrite(struct knote *kn, long hint);
1513 static int filt_tmpfsvnode(struct knote *kn, long hint);
1515 static struct filterops tmpfsread_filtops =
1516 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsread };
1517 static struct filterops tmpfswrite_filtops =
1518 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfswrite };
1519 static struct filterops tmpfsvnode_filtops =
1520 { FILTEROP_ISFD, NULL, filt_tmpfsdetach, filt_tmpfsvnode };
1523 tmpfs_kqfilter (struct vop_kqfilter_args *ap)
1525 struct vnode *vp = ap->a_vp;
1526 struct knote *kn = ap->a_kn;
1528 switch (kn->kn_filter) {
1530 kn->kn_fop = &tmpfsread_filtops;
1533 kn->kn_fop = &tmpfswrite_filtops;
1536 kn->kn_fop = &tmpfsvnode_filtops;
1539 return (EOPNOTSUPP);
1542 kn->kn_hook = (caddr_t)vp;
1544 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1550 filt_tmpfsdetach(struct knote *kn)
1552 struct vnode *vp = (void *)kn->kn_hook;
1554 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
1558 filt_tmpfsread(struct knote *kn, long hint)
1560 struct vnode *vp = (void *)kn->kn_hook;
1561 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp);
1564 if (hint == NOTE_REVOKE) {
1565 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
1568 off = node->tn_size - kn->kn_fp->f_offset;
1569 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1570 if (kn->kn_sfflags & NOTE_OLDAPI)
1574 * Handle possible MP race interlock on filter check/write
1576 if (kn->kn_data == 0) {
1578 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
1581 return (kn->kn_data != 0);
1585 filt_tmpfswrite(struct knote *kn, long hint)
1587 if (hint == NOTE_REVOKE)
1588 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
1594 filt_tmpfsvnode(struct knote *kn, long hint)
1596 if (kn->kn_sfflags & hint)
1597 kn->kn_fflags |= hint;
1598 if (hint == NOTE_REVOKE) {
1599 kn->kn_flags |= EV_EOF;
1602 return (kn->kn_fflags != 0);
1606 /* --------------------------------------------------------------------- */
1609 * vnode operations vector used for files stored in a tmpfs file system.
1611 struct vop_ops tmpfs_vnode_vops = {
1612 .vop_default = vop_defaultop,
1613 .vop_getpages = vop_stdgetpages,
1614 .vop_putpages = vop_stdputpages,
1615 .vop_ncreate = tmpfs_ncreate,
1616 .vop_nresolve = tmpfs_nresolve,
1617 .vop_nlookupdotdot = tmpfs_nlookupdotdot,
1618 .vop_nmknod = tmpfs_nmknod,
1619 .vop_open = tmpfs_open,
1620 .vop_close = tmpfs_close,
1621 .vop_access = tmpfs_access,
1622 .vop_getattr = tmpfs_getattr,
1623 .vop_setattr = tmpfs_setattr,
1624 .vop_read = tmpfs_read,
1625 .vop_write = tmpfs_write,
1626 .vop_fsync = tmpfs_fsync,
1627 .vop_nremove = tmpfs_nremove,
1628 .vop_nlink = tmpfs_nlink,
1629 .vop_nrename = tmpfs_nrename,
1630 .vop_nmkdir = tmpfs_nmkdir,
1631 .vop_nrmdir = tmpfs_nrmdir,
1632 .vop_nsymlink = tmpfs_nsymlink,
1633 .vop_readdir = tmpfs_readdir,
1634 .vop_readlink = tmpfs_readlink,
1635 .vop_inactive = tmpfs_inactive,
1636 .vop_reclaim = tmpfs_reclaim,
1637 .vop_print = tmpfs_print,
1638 .vop_pathconf = tmpfs_pathconf,
1639 .vop_bmap = tmpfs_bmap,
1640 .vop_strategy = tmpfs_strategy,
1641 .vop_advlock = tmpfs_advlock,
1642 .vop_kqfilter = tmpfs_kqfilter