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>
37 __FBSDID("$FreeBSD$");
39 #include <sys/param.h>
40 #include <sys/fcntl.h>
41 #include <sys/lockf.h>
43 #include <sys/namei.h>
46 #include <sys/rwlock.h>
47 #include <sys/sched.h>
49 #include <sys/systm.h>
50 #include <sys/sysctl.h>
51 #include <sys/unistd.h>
52 #include <sys/vnode.h>
55 #include <vm/vm_param.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_pager.h>
60 #include <fs/tmpfs/tmpfs_vnops.h>
61 #include <fs/tmpfs/tmpfs.h>
63 SYSCTL_DECL(_vfs_tmpfs);
65 static volatile int tmpfs_rename_restarts;
66 SYSCTL_INT(_vfs_tmpfs, OID_AUTO, rename_restarts, CTLFLAG_RD,
67 __DEVOLATILE(int *, &tmpfs_rename_restarts), 0,
68 "Times rename had to restart due to lock contention");
71 tmpfs_vn_get_ino_alloc(struct mount *mp, void *arg, int lkflags,
75 return (tmpfs_alloc_vp(mp, arg, lkflags, rvp));
79 tmpfs_lookup1(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp)
81 struct tmpfs_dirent *de;
82 struct tmpfs_node *dnode, *pnode;
83 struct tmpfs_mount *tm;
86 dnode = VP_TO_TMPFS_DIR(dvp);
89 /* Check accessibility of requested node as a first step. */
90 error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
94 /* We cannot be requesting the parent directory of the root node. */
95 MPASS(IMPLIES(dnode->tn_type == VDIR &&
96 dnode->tn_dir.tn_parent == dnode,
97 !(cnp->cn_flags & ISDOTDOT)));
99 TMPFS_ASSERT_LOCKED(dnode);
100 if (dnode->tn_dir.tn_parent == NULL) {
104 if (cnp->cn_flags & ISDOTDOT) {
105 tm = VFS_TO_TMPFS(dvp->v_mount);
106 pnode = dnode->tn_dir.tn_parent;
107 tmpfs_ref_node(pnode);
108 error = vn_vget_ino_gen(dvp, tmpfs_vn_get_ino_alloc,
109 pnode, cnp->cn_lkflags, vpp);
110 tmpfs_free_node(tm, pnode);
113 } else if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
118 de = tmpfs_dir_lookup(dnode, NULL, cnp);
119 if (de != NULL && de->td_node == NULL)
120 cnp->cn_flags |= ISWHITEOUT;
121 if (de == NULL || de->td_node == NULL) {
123 * The entry was not found in the directory.
124 * This is OK if we are creating or renaming an
125 * entry and are working on the last component of
128 if ((cnp->cn_flags & ISLASTCN) &&
129 (cnp->cn_nameiop == CREATE || \
130 cnp->cn_nameiop == RENAME ||
131 (cnp->cn_nameiop == DELETE &&
132 cnp->cn_flags & DOWHITEOUT &&
133 cnp->cn_flags & ISWHITEOUT))) {
134 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
140 * Keep the component name in the buffer for
143 cnp->cn_flags |= SAVENAME;
149 struct tmpfs_node *tnode;
152 * The entry was found, so get its associated
158 * If we are not at the last path component and
159 * found a non-directory or non-link entry (which
160 * may itself be pointing to a directory), raise
163 if ((tnode->tn_type != VDIR &&
164 tnode->tn_type != VLNK) &&
165 !(cnp->cn_flags & ISLASTCN)) {
171 * If we are deleting or renaming the entry, keep
172 * track of its tmpfs_dirent so that it can be
173 * easily deleted later.
175 if ((cnp->cn_flags & ISLASTCN) &&
176 (cnp->cn_nameiop == DELETE ||
177 cnp->cn_nameiop == RENAME)) {
178 error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred,
183 /* Allocate a new vnode on the matching entry. */
184 error = tmpfs_alloc_vp(dvp->v_mount, tnode,
185 cnp->cn_lkflags, vpp);
189 if ((dnode->tn_mode & S_ISTXT) &&
190 VOP_ACCESS(dvp, VADMIN, cnp->cn_cred,
191 cnp->cn_thread) && VOP_ACCESS(*vpp, VADMIN,
192 cnp->cn_cred, cnp->cn_thread)) {
198 cnp->cn_flags |= SAVENAME;
200 error = tmpfs_alloc_vp(dvp->v_mount, tnode,
201 cnp->cn_lkflags, vpp);
209 * Store the result of this lookup in the cache. Avoid this if the
210 * request was for creation, as it does not improve timings on
213 if ((cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp))
214 cache_enter(dvp, *vpp, cnp);
218 * If there were no errors, *vpp cannot be null and it must be
221 MPASS(IFF(error == 0, *vpp != NULLVP && VOP_ISLOCKED(*vpp)));
227 tmpfs_cached_lookup(struct vop_cachedlookup_args *v)
230 return (tmpfs_lookup1(v->a_dvp, v->a_vpp, v->a_cnp));
234 tmpfs_lookup(struct vop_lookup_args *v)
237 return (tmpfs_lookup1(v->a_dvp, v->a_vpp, v->a_cnp));
241 tmpfs_create(struct vop_create_args *v)
243 struct vnode *dvp = v->a_dvp;
244 struct vnode **vpp = v->a_vpp;
245 struct componentname *cnp = v->a_cnp;
246 struct vattr *vap = v->a_vap;
249 MPASS(vap->va_type == VREG || vap->va_type == VSOCK);
251 error = tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
252 if (error == 0 && (cnp->cn_flags & MAKEENTRY) != 0 && tmpfs_use_nc(dvp))
253 cache_enter(dvp, *vpp, cnp);
258 tmpfs_mknod(struct vop_mknod_args *v)
260 struct vnode *dvp = v->a_dvp;
261 struct vnode **vpp = v->a_vpp;
262 struct componentname *cnp = v->a_cnp;
263 struct vattr *vap = v->a_vap;
265 if (vap->va_type != VBLK && vap->va_type != VCHR &&
266 vap->va_type != VFIFO)
269 return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
273 tmpfs_open(struct vop_open_args *v)
275 struct vnode *vp = v->a_vp;
276 int mode = v->a_mode;
279 struct tmpfs_node *node;
281 MPASS(VOP_ISLOCKED(vp));
283 node = VP_TO_TMPFS_NODE(vp);
285 /* The file is still active but all its names have been removed
286 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as
287 * it is about to die. */
288 if (node->tn_links < 1)
291 /* If the file is marked append-only, deny write requests. */
292 if (node->tn_flags & APPEND && (mode & (FWRITE | O_APPEND)) == FWRITE)
296 /* For regular files, the call below is nop. */
297 KASSERT(vp->v_type != VREG || (node->tn_reg.tn_aobj->flags &
298 OBJ_DEAD) == 0, ("dead object"));
299 vnode_create_vobject(vp, node->tn_size, v->a_td);
302 MPASS(VOP_ISLOCKED(vp));
307 tmpfs_close(struct vop_close_args *v)
309 struct vnode *vp = v->a_vp;
311 /* Update node times. */
318 tmpfs_access(struct vop_access_args *v)
320 struct vnode *vp = v->a_vp;
321 accmode_t accmode = v->a_accmode;
322 struct ucred *cred = v->a_cred;
325 struct tmpfs_node *node;
327 MPASS(VOP_ISLOCKED(vp));
329 node = VP_TO_TMPFS_NODE(vp);
331 switch (vp->v_type) {
337 if (accmode & VWRITE && vp->v_mount->mnt_flag & MNT_RDONLY) {
357 if (accmode & VWRITE && node->tn_flags & IMMUTABLE) {
362 error = vaccess(vp->v_type, node->tn_mode, node->tn_uid,
363 node->tn_gid, accmode, cred, NULL);
366 MPASS(VOP_ISLOCKED(vp));
372 tmpfs_getattr(struct vop_getattr_args *v)
374 struct vnode *vp = v->a_vp;
375 struct vattr *vap = v->a_vap;
377 struct tmpfs_node *node;
379 node = VP_TO_TMPFS_NODE(vp);
383 vap->va_type = vp->v_type;
384 vap->va_mode = node->tn_mode;
385 vap->va_nlink = node->tn_links;
386 vap->va_uid = node->tn_uid;
387 vap->va_gid = node->tn_gid;
388 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
389 vap->va_fileid = node->tn_id;
390 vap->va_size = node->tn_size;
391 vap->va_blocksize = PAGE_SIZE;
392 vap->va_atime = node->tn_atime;
393 vap->va_mtime = node->tn_mtime;
394 vap->va_ctime = node->tn_ctime;
395 vap->va_birthtime = node->tn_birthtime;
396 vap->va_gen = node->tn_gen;
397 vap->va_flags = node->tn_flags;
398 vap->va_rdev = (vp->v_type == VBLK || vp->v_type == VCHR) ?
399 node->tn_rdev : NODEV;
400 if (vp->v_type == VREG) {
401 obj = node->tn_reg.tn_aobj;
402 vap->va_bytes = (u_quad_t)obj->resident_page_count * PAGE_SIZE;
404 vap->va_bytes = node->tn_size;
411 tmpfs_setattr(struct vop_setattr_args *v)
413 struct vnode *vp = v->a_vp;
414 struct vattr *vap = v->a_vap;
415 struct ucred *cred = v->a_cred;
416 struct thread *td = curthread;
420 MPASS(VOP_ISLOCKED(vp));
424 /* Abort if any unsettable attribute is given. */
425 if (vap->va_type != VNON ||
426 vap->va_nlink != VNOVAL ||
427 vap->va_fsid != VNOVAL ||
428 vap->va_fileid != VNOVAL ||
429 vap->va_blocksize != VNOVAL ||
430 vap->va_gen != VNOVAL ||
431 vap->va_rdev != VNOVAL ||
432 vap->va_bytes != VNOVAL)
435 if (error == 0 && (vap->va_flags != VNOVAL))
436 error = tmpfs_chflags(vp, vap->va_flags, cred, td);
438 if (error == 0 && (vap->va_size != VNOVAL))
439 error = tmpfs_chsize(vp, vap->va_size, cred, td);
441 if (error == 0 && (vap->va_uid != VNOVAL || vap->va_gid != VNOVAL))
442 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred, td);
444 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL))
445 error = tmpfs_chmod(vp, vap->va_mode, cred, td);
447 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL &&
448 vap->va_atime.tv_nsec != VNOVAL) ||
449 (vap->va_mtime.tv_sec != VNOVAL &&
450 vap->va_mtime.tv_nsec != VNOVAL) ||
451 (vap->va_birthtime.tv_sec != VNOVAL &&
452 vap->va_birthtime.tv_nsec != VNOVAL)))
453 error = tmpfs_chtimes(vp, vap, cred, td);
455 /* Update the node times. We give preference to the error codes
456 * generated by this function rather than the ones that may arise
457 * from tmpfs_update. */
460 MPASS(VOP_ISLOCKED(vp));
466 tmpfs_read(struct vop_read_args *v)
470 struct tmpfs_node *node;
473 if (vp->v_type != VREG)
476 if (uio->uio_offset < 0)
478 node = VP_TO_TMPFS_NODE(vp);
479 tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
480 return (uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio));
484 tmpfs_write(struct vop_write_args *v)
488 struct tmpfs_node *node;
494 ioflag = v->a_ioflag;
496 node = VP_TO_TMPFS_NODE(vp);
497 oldsize = node->tn_size;
499 if (uio->uio_offset < 0 || vp->v_type != VREG)
501 if (uio->uio_resid == 0)
503 if (ioflag & IO_APPEND)
504 uio->uio_offset = node->tn_size;
505 if (uio->uio_offset + uio->uio_resid >
506 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
508 if (vn_rlimit_fsize(vp, uio, uio->uio_td))
510 if (uio->uio_offset + uio->uio_resid > node->tn_size) {
511 error = tmpfs_reg_resize(vp, uio->uio_offset + uio->uio_resid,
517 error = uiomove_object(node->tn_reg.tn_aobj, node->tn_size, uio);
518 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
520 if (node->tn_mode & (S_ISUID | S_ISGID)) {
521 if (priv_check_cred(v->a_cred, PRIV_VFS_RETAINSUGID, 0))
522 node->tn_mode &= ~(S_ISUID | S_ISGID);
525 (void)tmpfs_reg_resize(vp, oldsize, TRUE);
528 MPASS(IMPLIES(error == 0, uio->uio_resid == 0));
529 MPASS(IMPLIES(error != 0, oldsize == node->tn_size));
535 tmpfs_fsync(struct vop_fsync_args *v)
537 struct vnode *vp = v->a_vp;
539 MPASS(VOP_ISLOCKED(vp));
541 tmpfs_check_mtime(vp);
548 tmpfs_remove(struct vop_remove_args *v)
550 struct vnode *dvp = v->a_dvp;
551 struct vnode *vp = v->a_vp;
554 struct tmpfs_dirent *de;
555 struct tmpfs_mount *tmp;
556 struct tmpfs_node *dnode;
557 struct tmpfs_node *node;
559 MPASS(VOP_ISLOCKED(dvp));
560 MPASS(VOP_ISLOCKED(vp));
562 if (vp->v_type == VDIR) {
567 dnode = VP_TO_TMPFS_DIR(dvp);
568 node = VP_TO_TMPFS_NODE(vp);
569 tmp = VFS_TO_TMPFS(vp->v_mount);
570 de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
573 /* Files marked as immutable or append-only cannot be deleted. */
574 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) ||
575 (dnode->tn_flags & APPEND)) {
580 /* Remove the entry from the directory; as it is a file, we do not
581 * have to change the number of hard links of the directory. */
582 tmpfs_dir_detach(dvp, de);
583 if (v->a_cnp->cn_flags & DOWHITEOUT)
584 tmpfs_dir_whiteout_add(dvp, v->a_cnp);
586 /* Free the directory entry we just deleted. Note that the node
587 * referred by it will not be removed until the vnode is really
589 tmpfs_free_dirent(tmp, de);
591 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED;
600 tmpfs_link(struct vop_link_args *v)
602 struct vnode *dvp = v->a_tdvp;
603 struct vnode *vp = v->a_vp;
604 struct componentname *cnp = v->a_cnp;
607 struct tmpfs_dirent *de;
608 struct tmpfs_node *node;
610 MPASS(VOP_ISLOCKED(dvp));
611 MPASS(cnp->cn_flags & HASBUF);
612 MPASS(dvp != vp); /* XXX When can this be false? */
613 node = VP_TO_TMPFS_NODE(vp);
615 /* Ensure that we do not overflow the maximum number of links imposed
617 MPASS(node->tn_links <= LINK_MAX);
618 if (node->tn_links == LINK_MAX) {
623 /* We cannot create links of files marked immutable or append-only. */
624 if (node->tn_flags & (IMMUTABLE | APPEND)) {
629 /* Allocate a new directory entry to represent the node. */
630 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node,
631 cnp->cn_nameptr, cnp->cn_namelen, &de);
635 /* Insert the new directory entry into the appropriate directory. */
636 if (cnp->cn_flags & ISWHITEOUT)
637 tmpfs_dir_whiteout_remove(dvp, cnp);
638 tmpfs_dir_attach(dvp, de);
640 /* vp link count has changed, so update node times. */
641 node->tn_status |= TMPFS_NODE_CHANGED;
651 * We acquire all but fdvp locks using non-blocking acquisitions. If we
652 * fail to acquire any lock in the path we will drop all held locks,
653 * acquire the new lock in a blocking fashion, and then release it and
654 * restart the rename. This acquire/release step ensures that we do not
655 * spin on a lock waiting for release. On error release all vnode locks
656 * and decrement references the way tmpfs_rename() would do.
659 tmpfs_rename_relock(struct vnode *fdvp, struct vnode **fvpp,
660 struct vnode *tdvp, struct vnode **tvpp,
661 struct componentname *fcnp, struct componentname *tcnp)
665 struct tmpfs_dirent *de;
666 int error, restarts = 0;
669 if (*tvpp != NULL && *tvpp != tdvp)
670 VOP_UNLOCK(*tvpp, 0);
675 error = vn_lock(fdvp, LK_EXCLUSIVE);
678 if (vn_lock(tdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
680 error = vn_lock(tdvp, LK_EXCLUSIVE);
687 * Re-resolve fvp to be certain it still exists and fetch the
690 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(fdvp), NULL, fcnp);
694 if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
695 (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
701 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE | LK_NOWAIT, &nvp);
707 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE, &nvp);
712 * Concurrent rename race.
725 VOP_UNLOCK(*fvpp, 0);
727 * Re-resolve tvp and acquire the vnode lock if present.
729 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(tdvp), NULL, tcnp);
731 * If tvp disappeared we just carry on.
733 if (de == NULL && *tvpp != NULL) {
738 * Get the tvp ino if the lookup succeeded. We may have to restart
739 * if the non-blocking acquire fails.
743 error = tmpfs_alloc_vp(mp, de->td_node,
744 LK_EXCLUSIVE | LK_NOWAIT, &nvp);
753 error = tmpfs_alloc_vp(mp, de->td_node, LK_EXCLUSIVE,
759 * fdvp contains fvp, thus tvp (=fdvp) is not empty.
768 tmpfs_rename_restarts += restarts;
778 tmpfs_rename_restarts += restarts;
784 tmpfs_rename(struct vop_rename_args *v)
786 struct vnode *fdvp = v->a_fdvp;
787 struct vnode *fvp = v->a_fvp;
788 struct componentname *fcnp = v->a_fcnp;
789 struct vnode *tdvp = v->a_tdvp;
790 struct vnode *tvp = v->a_tvp;
791 struct componentname *tcnp = v->a_tcnp;
792 struct mount *mp = NULL;
796 struct tmpfs_dirent *de;
797 struct tmpfs_mount *tmp;
798 struct tmpfs_node *fdnode;
799 struct tmpfs_node *fnode;
800 struct tmpfs_node *tnode;
801 struct tmpfs_node *tdnode;
803 MPASS(VOP_ISLOCKED(tdvp));
804 MPASS(IMPLIES(tvp != NULL, VOP_ISLOCKED(tvp)));
805 MPASS(fcnp->cn_flags & HASBUF);
806 MPASS(tcnp->cn_flags & HASBUF);
808 /* Disallow cross-device renames.
809 * XXX Why isn't this done by the caller? */
810 if (fvp->v_mount != tdvp->v_mount ||
811 (tvp != NULL && fvp->v_mount != tvp->v_mount)) {
816 /* If source and target are the same file, there is nothing to do. */
822 /* If we need to move the directory between entries, lock the
823 * source so that we can safely operate on it. */
824 if (fdvp != tdvp && fdvp != tvp) {
825 if (vn_lock(fdvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
827 error = vfs_busy(mp, 0);
832 error = tmpfs_rename_relock(fdvp, &fvp, tdvp, &tvp,
838 ASSERT_VOP_ELOCKED(fdvp,
839 "tmpfs_rename: fdvp not locked");
840 ASSERT_VOP_ELOCKED(tdvp,
841 "tmpfs_rename: tdvp not locked");
843 ASSERT_VOP_ELOCKED(tvp,
844 "tmpfs_rename: tvp not locked");
852 tmp = VFS_TO_TMPFS(tdvp->v_mount);
853 tdnode = VP_TO_TMPFS_DIR(tdvp);
854 tnode = (tvp == NULL) ? NULL : VP_TO_TMPFS_NODE(tvp);
855 fdnode = VP_TO_TMPFS_DIR(fdvp);
856 fnode = VP_TO_TMPFS_NODE(fvp);
857 de = tmpfs_dir_lookup(fdnode, fnode, fcnp);
859 /* Entry can disappear before we lock fdvp,
860 * also avoid manipulating '.' and '..' entries. */
862 if ((fcnp->cn_flags & ISDOTDOT) != 0 ||
863 (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.'))
869 MPASS(de->td_node == fnode);
871 /* If re-naming a directory to another preexisting directory
872 * ensure that the target directory is empty so that its
873 * removal causes no side effects.
874 * Kern_rename guarantees the destination to be a directory
875 * if the source is one. */
877 MPASS(tnode != NULL);
879 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) ||
880 (tdnode->tn_flags & (APPEND | IMMUTABLE))) {
885 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) {
886 if (tnode->tn_size > 0) {
890 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) {
893 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) {
897 MPASS(fnode->tn_type != VDIR &&
898 tnode->tn_type != VDIR);
902 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))
903 || (fdnode->tn_flags & (APPEND | IMMUTABLE))) {
908 /* Ensure that we have enough memory to hold the new name, if it
909 * has to be changed. */
910 if (fcnp->cn_namelen != tcnp->cn_namelen ||
911 bcmp(fcnp->cn_nameptr, tcnp->cn_nameptr, fcnp->cn_namelen) != 0) {
912 newname = malloc(tcnp->cn_namelen, M_TMPFSNAME, M_WAITOK);
916 /* If the node is being moved to another directory, we have to do
918 if (fdnode != tdnode) {
919 /* In case we are moving a directory, we have to adjust its
920 * parent to point to the new parent. */
921 if (de->td_node->tn_type == VDIR) {
922 struct tmpfs_node *n;
924 /* Ensure the target directory is not a child of the
925 * directory being moved. Otherwise, we'd end up
926 * with stale nodes. */
928 /* TMPFS_LOCK garanties that no nodes are freed while
929 * traversing the list. Nodes can only be marked as
930 * removed: tn_parent == NULL. */
933 while (n != n->tn_dir.tn_parent) {
934 struct tmpfs_node *parent;
937 TMPFS_NODE_UNLOCK(n);
941 free(newname, M_TMPFSNAME);
944 parent = n->tn_dir.tn_parent;
945 TMPFS_NODE_UNLOCK(n);
946 if (parent == NULL) {
950 TMPFS_NODE_LOCK(parent);
951 if (parent->tn_dir.tn_parent == NULL) {
952 TMPFS_NODE_UNLOCK(parent);
962 free(newname, M_TMPFSNAME);
965 TMPFS_NODE_UNLOCK(n);
967 /* Adjust the parent pointer. */
968 TMPFS_VALIDATE_DIR(fnode);
969 TMPFS_NODE_LOCK(de->td_node);
970 de->td_node->tn_dir.tn_parent = tdnode;
971 TMPFS_NODE_UNLOCK(de->td_node);
973 /* As a result of changing the target of the '..'
974 * entry, the link count of the source and target
975 * directories has to be adjusted. */
976 TMPFS_NODE_LOCK(tdnode);
977 TMPFS_ASSERT_LOCKED(tdnode);
979 TMPFS_NODE_UNLOCK(tdnode);
981 TMPFS_NODE_LOCK(fdnode);
982 TMPFS_ASSERT_LOCKED(fdnode);
984 TMPFS_NODE_UNLOCK(fdnode);
988 /* Do the move: just remove the entry from the source directory
989 * and insert it into the target one. */
990 tmpfs_dir_detach(fdvp, de);
992 if (fcnp->cn_flags & DOWHITEOUT)
993 tmpfs_dir_whiteout_add(fdvp, fcnp);
994 if (tcnp->cn_flags & ISWHITEOUT)
995 tmpfs_dir_whiteout_remove(tdvp, tcnp);
997 /* If the name has changed, we need to make it effective by changing
998 * it in the directory entry. */
999 if (newname != NULL) {
1000 MPASS(tcnp->cn_namelen <= MAXNAMLEN);
1002 free(de->ud.td_name, M_TMPFSNAME);
1003 de->ud.td_name = newname;
1004 tmpfs_dirent_init(de, tcnp->cn_nameptr, tcnp->cn_namelen);
1006 fnode->tn_status |= TMPFS_NODE_CHANGED;
1007 tdnode->tn_status |= TMPFS_NODE_MODIFIED;
1010 /* If we are overwriting an entry, we have to remove the old one
1011 * from the target directory. */
1013 struct tmpfs_dirent *tde;
1015 /* Remove the old entry from the target directory. */
1016 tde = tmpfs_dir_lookup(tdnode, tnode, tcnp);
1017 tmpfs_dir_detach(tdvp, tde);
1019 /* Free the directory entry we just deleted. Note that the
1020 * node referred by it will not be removed until the vnode is
1021 * really reclaimed. */
1022 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde);
1025 tmpfs_dir_attach(tdvp, de);
1027 if (tmpfs_use_nc(fvp)) {
1031 cache_purge_negative(tdvp);
1037 if (fdvp != tdvp && fdvp != tvp)
1038 VOP_UNLOCK(fdvp, 0);
1041 /* Release target nodes. */
1042 /* XXX: I don't understand when tdvp can be the same as tvp, but
1043 * other code takes care of this... */
1051 /* Release source nodes. */
1062 tmpfs_mkdir(struct vop_mkdir_args *v)
1064 struct vnode *dvp = v->a_dvp;
1065 struct vnode **vpp = v->a_vpp;
1066 struct componentname *cnp = v->a_cnp;
1067 struct vattr *vap = v->a_vap;
1069 MPASS(vap->va_type == VDIR);
1071 return tmpfs_alloc_file(dvp, vpp, vap, cnp, NULL);
1075 tmpfs_rmdir(struct vop_rmdir_args *v)
1077 struct vnode *dvp = v->a_dvp;
1078 struct vnode *vp = v->a_vp;
1081 struct tmpfs_dirent *de;
1082 struct tmpfs_mount *tmp;
1083 struct tmpfs_node *dnode;
1084 struct tmpfs_node *node;
1086 MPASS(VOP_ISLOCKED(dvp));
1087 MPASS(VOP_ISLOCKED(vp));
1089 tmp = VFS_TO_TMPFS(dvp->v_mount);
1090 dnode = VP_TO_TMPFS_DIR(dvp);
1091 node = VP_TO_TMPFS_DIR(vp);
1093 /* Directories with more than two entries ('.' and '..') cannot be
1095 if (node->tn_size > 0) {
1100 if ((dnode->tn_flags & APPEND)
1101 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) {
1106 /* This invariant holds only if we are not trying to remove "..".
1107 * We checked for that above so this is safe now. */
1108 MPASS(node->tn_dir.tn_parent == dnode);
1110 /* Get the directory entry associated with node (vp). This was
1111 * filled by tmpfs_lookup while looking up the entry. */
1112 de = tmpfs_dir_lookup(dnode, node, v->a_cnp);
1113 MPASS(TMPFS_DIRENT_MATCHES(de,
1114 v->a_cnp->cn_nameptr,
1115 v->a_cnp->cn_namelen));
1117 /* Check flags to see if we are allowed to remove the directory. */
1118 if ((dnode->tn_flags & APPEND) != 0 ||
1119 (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) != 0) {
1125 /* Detach the directory entry from the directory (dnode). */
1126 tmpfs_dir_detach(dvp, de);
1127 if (v->a_cnp->cn_flags & DOWHITEOUT)
1128 tmpfs_dir_whiteout_add(dvp, v->a_cnp);
1130 /* No vnode should be allocated for this entry from this point */
1131 TMPFS_NODE_LOCK(node);
1133 node->tn_dir.tn_parent = NULL;
1134 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
1135 TMPFS_NODE_MODIFIED;
1137 TMPFS_NODE_UNLOCK(node);
1139 TMPFS_NODE_LOCK(dnode);
1141 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
1142 TMPFS_NODE_MODIFIED;
1143 TMPFS_NODE_UNLOCK(dnode);
1145 if (tmpfs_use_nc(dvp)) {
1150 /* Free the directory entry we just deleted. Note that the node
1151 * referred by it will not be removed until the vnode is really
1153 tmpfs_free_dirent(tmp, de);
1155 /* Release the deleted vnode (will destroy the node, notify
1156 * interested parties and clean it from the cache). */
1158 dnode->tn_status |= TMPFS_NODE_CHANGED;
1168 tmpfs_symlink(struct vop_symlink_args *v)
1170 struct vnode *dvp = v->a_dvp;
1171 struct vnode **vpp = v->a_vpp;
1172 struct componentname *cnp = v->a_cnp;
1173 struct vattr *vap = v->a_vap;
1174 char *target = v->a_target;
1176 #ifdef notyet /* XXX FreeBSD BUG: kern_symlink is not setting VLNK */
1177 MPASS(vap->va_type == VLNK);
1179 vap->va_type = VLNK;
1182 return tmpfs_alloc_file(dvp, vpp, vap, cnp, target);
1186 tmpfs_readdir(struct vop_readdir_args *v)
1188 struct vnode *vp = v->a_vp;
1189 struct uio *uio = v->a_uio;
1190 int *eofflag = v->a_eofflag;
1191 u_long **cookies = v->a_cookies;
1192 int *ncookies = v->a_ncookies;
1197 struct tmpfs_node *node;
1199 /* This operation only makes sense on directory nodes. */
1200 if (vp->v_type != VDIR)
1204 node = VP_TO_TMPFS_DIR(vp);
1206 startresid = uio->uio_resid;
1208 /* Allocate cookies for NFS and compat modules. */
1209 if (cookies != NULL && ncookies != NULL) {
1210 maxcookies = howmany(node->tn_size,
1211 sizeof(struct tmpfs_dirent)) + 2;
1212 *cookies = malloc(maxcookies * sizeof(**cookies), M_TEMP,
1217 if (cookies == NULL)
1218 error = tmpfs_dir_getdents(node, uio, 0, NULL, NULL);
1220 error = tmpfs_dir_getdents(node, uio, maxcookies, *cookies,
1223 /* Buffer was filled without hitting EOF. */
1224 if (error == EJUSTRETURN)
1225 error = (uio->uio_resid != startresid) ? 0 : EINVAL;
1227 if (error != 0 && cookies != NULL && ncookies != NULL) {
1228 free(*cookies, M_TEMP);
1233 if (eofflag != NULL)
1235 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF);
1241 tmpfs_readlink(struct vop_readlink_args *v)
1243 struct vnode *vp = v->a_vp;
1244 struct uio *uio = v->a_uio;
1247 struct tmpfs_node *node;
1249 MPASS(uio->uio_offset == 0);
1250 MPASS(vp->v_type == VLNK);
1252 node = VP_TO_TMPFS_NODE(vp);
1254 error = uiomove(node->tn_link, MIN(node->tn_size, uio->uio_resid),
1256 tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
1262 tmpfs_inactive(struct vop_inactive_args *v)
1265 struct tmpfs_node *node;
1268 node = VP_TO_TMPFS_NODE(vp);
1269 if (node->tn_links == 0)
1272 tmpfs_check_mtime(vp);
1277 tmpfs_reclaim(struct vop_reclaim_args *v)
1279 struct vnode *vp = v->a_vp;
1281 struct tmpfs_mount *tmp;
1282 struct tmpfs_node *node;
1284 node = VP_TO_TMPFS_NODE(vp);
1285 tmp = VFS_TO_TMPFS(vp->v_mount);
1287 if (vp->v_type == VREG)
1288 tmpfs_destroy_vobject(vp, node->tn_reg.tn_aobj);
1290 vnode_destroy_vobject(vp);
1291 vp->v_object = NULL;
1292 if (tmpfs_use_nc(vp))
1295 TMPFS_NODE_LOCK(node);
1298 /* If the node referenced by this vnode was deleted by the user,
1299 * we must free its associated data structures (now that the vnode
1300 * is being reclaimed). */
1301 if (node->tn_links == 0 &&
1302 (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0) {
1303 node->tn_vpstate = TMPFS_VNODE_DOOMED;
1304 TMPFS_NODE_UNLOCK(node);
1305 tmpfs_free_node(tmp, node);
1307 TMPFS_NODE_UNLOCK(node);
1309 MPASS(vp->v_data == NULL);
1314 tmpfs_print(struct vop_print_args *v)
1316 struct vnode *vp = v->a_vp;
1318 struct tmpfs_node *node;
1320 node = VP_TO_TMPFS_NODE(vp);
1322 printf("tag VT_TMPFS, tmpfs_node %p, flags 0x%lx, links %jd\n",
1323 node, node->tn_flags, (uintmax_t)node->tn_links);
1324 printf("\tmode 0%o, owner %d, group %d, size %jd, status 0x%x\n",
1325 node->tn_mode, node->tn_uid, node->tn_gid,
1326 (intmax_t)node->tn_size, node->tn_status);
1328 if (vp->v_type == VFIFO)
1337 tmpfs_pathconf(struct vop_pathconf_args *v)
1339 int name = v->a_name;
1340 register_t *retval = v->a_retval;
1363 case _PC_CHOWN_RESTRICTED:
1375 case _PC_FILESIZEBITS:
1376 *retval = 0; /* XXX Don't know which value should I return. */
1387 tmpfs_vptofh(struct vop_vptofh_args *ap)
1389 struct tmpfs_fid *tfhp;
1390 struct tmpfs_node *node;
1392 tfhp = (struct tmpfs_fid *)ap->a_fhp;
1393 node = VP_TO_TMPFS_NODE(ap->a_vp);
1395 tfhp->tf_len = sizeof(struct tmpfs_fid);
1396 tfhp->tf_id = node->tn_id;
1397 tfhp->tf_gen = node->tn_gen;
1403 tmpfs_whiteout(struct vop_whiteout_args *ap)
1405 struct vnode *dvp = ap->a_dvp;
1406 struct componentname *cnp = ap->a_cnp;
1407 struct tmpfs_dirent *de;
1409 switch (ap->a_flags) {
1413 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1415 return (de->td_node == NULL ? 0 : EEXIST);
1416 return (tmpfs_dir_whiteout_add(dvp, cnp));
1418 tmpfs_dir_whiteout_remove(dvp, cnp);
1421 panic("tmpfs_whiteout: unknown op");
1426 tmpfs_vptocnp_dir(struct tmpfs_node *tn, struct tmpfs_node *tnp,
1427 struct tmpfs_dirent **pde)
1429 struct tmpfs_dir_cursor dc;
1430 struct tmpfs_dirent *de;
1432 for (de = tmpfs_dir_first(tnp, &dc); de != NULL;
1433 de = tmpfs_dir_next(tnp, &dc)) {
1434 if (de->td_node == tn) {
1443 tmpfs_vptocnp_fill(struct vnode *vp, struct tmpfs_node *tn,
1444 struct tmpfs_node *tnp, char *buf, int *buflen, struct vnode **dvp)
1446 struct tmpfs_dirent *de;
1449 error = vn_vget_ino_gen(vp, tmpfs_vn_get_ino_alloc, tnp, LK_SHARED,
1453 error = tmpfs_vptocnp_dir(tn, tnp, &de);
1456 i -= de->td_namelen;
1460 bcopy(de->ud.td_name, buf + i, de->td_namelen);
1466 VOP_UNLOCK(*dvp, 0);
1477 tmpfs_vptocnp(struct vop_vptocnp_args *ap)
1479 struct vnode *vp, **dvp;
1480 struct tmpfs_node *tn, *tnp, *tnp1;
1481 struct tmpfs_dirent *de;
1482 struct tmpfs_mount *tm;
1490 buflen = ap->a_buflen;
1492 tm = VFS_TO_TMPFS(vp->v_mount);
1493 tn = VP_TO_TMPFS_NODE(vp);
1494 if (tn->tn_type == VDIR) {
1495 tnp = tn->tn_dir.tn_parent;
1498 tmpfs_ref_node(tnp);
1499 error = tmpfs_vptocnp_fill(vp, tn, tn->tn_dir.tn_parent, buf,
1501 tmpfs_free_node(tm, tnp);
1506 LIST_FOREACH_SAFE(tnp, &tm->tm_nodes_used, tn_entries, tnp1) {
1507 if (tnp->tn_type != VDIR)
1509 TMPFS_NODE_LOCK(tnp);
1510 tmpfs_ref_node_locked(tnp);
1513 * tn_vnode cannot be instantiated while we hold the
1514 * node lock, so the directory cannot be changed while
1515 * we iterate over it. Do this to avoid instantiating
1516 * vnode for directories which cannot point to our
1519 error = tnp->tn_vnode == NULL ? tmpfs_vptocnp_dir(tn, tnp,
1523 TMPFS_NODE_UNLOCK(tnp);
1525 error = tmpfs_vptocnp_fill(vp, tn, tnp, buf, buflen,
1528 tmpfs_free_node(tm, tnp);
1531 if ((vp->v_iflag & VI_DOOMED) != 0) {
1532 tmpfs_free_node(tm, tnp);
1536 TMPFS_NODE_LOCK(tnp);
1538 if (tmpfs_free_node_locked(tm, tnp, false)) {
1541 KASSERT(tnp->tn_refcount > 0,
1542 ("node %p refcount zero", tnp));
1543 tnp1 = LIST_NEXT(tnp, tn_entries);
1544 TMPFS_NODE_UNLOCK(tnp);
1552 * Vnode operations vector used for files stored in a tmpfs file system.
1554 struct vop_vector tmpfs_vnodeop_entries = {
1555 .vop_default = &default_vnodeops,
1556 .vop_lookup = vfs_cache_lookup,
1557 .vop_cachedlookup = tmpfs_cached_lookup,
1558 .vop_create = tmpfs_create,
1559 .vop_mknod = tmpfs_mknod,
1560 .vop_open = tmpfs_open,
1561 .vop_close = tmpfs_close,
1562 .vop_access = tmpfs_access,
1563 .vop_getattr = tmpfs_getattr,
1564 .vop_setattr = tmpfs_setattr,
1565 .vop_read = tmpfs_read,
1566 .vop_write = tmpfs_write,
1567 .vop_fsync = tmpfs_fsync,
1568 .vop_remove = tmpfs_remove,
1569 .vop_link = tmpfs_link,
1570 .vop_rename = tmpfs_rename,
1571 .vop_mkdir = tmpfs_mkdir,
1572 .vop_rmdir = tmpfs_rmdir,
1573 .vop_symlink = tmpfs_symlink,
1574 .vop_readdir = tmpfs_readdir,
1575 .vop_readlink = tmpfs_readlink,
1576 .vop_inactive = tmpfs_inactive,
1577 .vop_reclaim = tmpfs_reclaim,
1578 .vop_print = tmpfs_print,
1579 .vop_pathconf = tmpfs_pathconf,
1580 .vop_vptofh = tmpfs_vptofh,
1581 .vop_whiteout = tmpfs_whiteout,
1582 .vop_bmap = VOP_EOPNOTSUPP,
1583 .vop_vptocnp = tmpfs_vptocnp,
1587 * Same vector for mounts which do not use namecache.
1589 struct vop_vector tmpfs_vnodeop_nonc_entries = {
1590 .vop_default = &tmpfs_vnodeop_entries,
1591 .vop_lookup = tmpfs_lookup,