1 /* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */
4 * Copyright (c) 2005 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 * Efficient memory file system supporting functions.
37 #include <sys/kernel.h>
38 #include <sys/param.h>
39 #include <sys/namei.h>
43 #include <sys/systm.h>
44 #include <sys/vnode.h>
45 #include <sys/vmmeter.h>
48 #include <vm/vm_object.h>
49 #include <vm/vm_page.h>
50 #include <vm/vm_pager.h>
51 #include <vm/vm_extern.h>
53 #include <vfs/tmpfs/tmpfs.h>
54 #include <vfs/tmpfs/tmpfs_vnops.h>
56 static ino_t tmpfs_fetch_ino(struct tmpfs_mount *);
58 static int tmpfs_dirtree_compare(struct tmpfs_dirent *a,
59 struct tmpfs_dirent *b);
60 RB_GENERATE(tmpfs_dirtree, tmpfs_dirent, rb_node, tmpfs_dirtree_compare);
62 static int tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a,
63 struct tmpfs_dirent *b);
64 RB_GENERATE(tmpfs_dirtree_cookie, tmpfs_dirent,
65 rb_cookienode, tmpfs_dirtree_compare_cookie);
68 /* --------------------------------------------------------------------- */
71 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
72 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
73 * using the credentials of the process 'p'.
75 * If the node type is set to 'VDIR', then the parent parameter must point
76 * to the parent directory of the node being created. It may only be NULL
77 * while allocating the root node.
79 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
80 * specifies the device the node represents.
82 * If the node type is set to 'VLNK', then the parameter target specifies
83 * the file name of the target file for the symbolic link that is being
86 * Note that new nodes are retrieved from the available list if it has
87 * items or, if it is empty, from the node pool as long as there is enough
88 * space to create them.
90 * Returns zero on success or an appropriate error code on failure.
93 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
94 uid_t uid, gid_t gid, mode_t mode,
95 char *target, int rmajor, int rminor,
96 struct tmpfs_node **node)
98 struct tmpfs_node *nnode;
102 KKASSERT(IFF(type == VLNK, target != NULL));
103 KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL));
105 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
108 nnode = objcache_get(tmp->tm_node_pool, M_WAITOK | M_NULLOK);
112 /* Generic initialization. */
113 nnode->tn_type = type;
115 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime
117 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec
121 nnode->tn_mode = mode;
122 nnode->tn_id = tmpfs_fetch_ino(tmp);
123 nnode->tn_advlock.init_done = 0;
124 KKASSERT(nnode->tn_links == 0);
126 /* Type-specific initialization. */
127 switch (nnode->tn_type) {
130 rdev = makeudev(rmajor, rminor);
131 if (rdev == NOUDEV) {
132 objcache_put(tmp->tm_node_pool, nnode);
135 nnode->tn_rdev = rdev;
139 RB_INIT(&nnode->tn_dir.tn_dirtree);
140 RB_INIT(&nnode->tn_dir.tn_cookietree);
150 nnode->tn_size = strlen(target);
151 nnode->tn_link = kmalloc(nnode->tn_size + 1, tmp->tm_name_zone,
152 M_WAITOK | M_NULLOK);
153 if (nnode->tn_link == NULL) {
154 objcache_put(tmp->tm_node_pool, nnode);
157 bcopy(target, nnode->tn_link, nnode->tn_size);
158 nnode->tn_link[nnode->tn_size] = '\0';
162 nnode->tn_reg.tn_aobj =
163 swap_pager_alloc(NULL, 0, VM_PROT_DEFAULT, 0);
164 nnode->tn_reg.tn_aobj_pages = 0;
169 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
172 TMPFS_NODE_LOCK(nnode);
174 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
175 tmp->tm_nodes_inuse++;
177 TMPFS_NODE_UNLOCK(nnode);
183 /* --------------------------------------------------------------------- */
186 * Destroys the node pointed to by node from the file system 'tmp'.
187 * If the node does not belong to the given mount point, the results are
190 * If the node references a directory; no entries are allowed because
191 * their removal could need a recursive algorithm, something forbidden in
192 * kernel space. Furthermore, there is not need to provide such
193 * functionality (recursive removal) because the only primitives offered
194 * to the user are the removal of empty directories and the deletion of
197 * Note that nodes are not really deleted; in fact, when a node has been
198 * allocated, it cannot be deleted during the whole life of the file
199 * system. Instead, they are moved to the available list and remain there
202 * A caller must have TMPFS_NODE_LOCK(node) and this function unlocks it.
205 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
207 vm_pindex_t pages = 0;
210 TMPFS_ASSERT_ELOCKED(node);
211 KKASSERT(node->tn_vnode == NULL);
212 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
216 LIST_REMOVE(node, tn_entries);
217 tmp->tm_nodes_inuse--;
219 TMPFS_NODE_UNLOCK(node); /* Caller has this lock */
221 switch (node->tn_type) {
223 /* Do not do anything. VNON is provided to let the
224 * allocation routine clean itself easily by avoiding
225 * duplicating code in it. */
234 * The parent link can be NULL if this is the root
235 * node or if it is a directory node that was rmdir'd.
237 * XXX what if node is a directory which still contains
238 * directory entries (e.g. due to a forced umount) ?
241 KKASSERT(node->tn_dir.tn_parent == NULL);
244 * If the root node is being destroyed don't leave a
245 * dangling pointer in tmpfs_mount.
247 if (node == tmp->tm_root)
256 kfree(node->tn_link, tmp->tm_name_zone);
257 node->tn_link = NULL;
262 if (node->tn_reg.tn_aobj != NULL)
263 vm_object_deallocate(node->tn_reg.tn_aobj);
264 node->tn_reg.tn_aobj = NULL;
265 pages = node->tn_reg.tn_aobj_pages;
269 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
273 * Clean up fields for the next allocation. The objcache only ctors
276 tmpfs_node_ctor(node, NULL, 0);
277 objcache_put(tmp->tm_node_pool, node);
278 /* node is now invalid */
281 atomic_add_long(&tmp->tm_pages_used, -(long)pages);
284 /* --------------------------------------------------------------------- */
287 * Allocates a new directory entry for the node node with a name of name.
288 * The new directory entry is returned in *de.
290 * The link count of node is increased by one to reflect the new object
293 * Returns zero on success or an appropriate error code on failure.
296 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
297 const char *name, uint16_t len, struct tmpfs_dirent **de)
299 struct tmpfs_dirent *nde;
301 nde = objcache_get(tmp->tm_dirent_pool, M_WAITOK);
302 nde->td_name = kmalloc(len + 1, tmp->tm_name_zone, M_WAITOK | M_NULLOK);
303 if (nde->td_name == NULL) {
304 objcache_put(tmp->tm_dirent_pool, nde);
308 nde->td_namelen = len;
309 bcopy(name, nde->td_name, len);
310 nde->td_name[len] = '\0';
314 TMPFS_NODE_LOCK(node);
316 TMPFS_NODE_UNLOCK(node);
323 /* --------------------------------------------------------------------- */
326 * Frees a directory entry. It is the caller's responsibility to destroy
327 * the node referenced by it if needed.
329 * The link count of node is decreased by one to reflect the removal of an
330 * object that referenced it. This only happens if 'node_exists' is true;
331 * otherwise the function will not access the node referred to by the
332 * directory entry, as it may already have been released from the outside.
335 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
337 struct tmpfs_node *node;
341 TMPFS_NODE_LOCK(node);
342 TMPFS_ASSERT_ELOCKED(node);
343 KKASSERT(node->tn_links > 0);
345 TMPFS_NODE_UNLOCK(node);
347 kfree(de->td_name, tmp->tm_name_zone);
351 objcache_put(tmp->tm_dirent_pool, de);
354 /* --------------------------------------------------------------------- */
357 * Allocates a new vnode for the node node or returns a new reference to
358 * an existing one if the node had already a vnode referencing it. The
359 * resulting locked vnode is returned in *vpp.
361 * Returns zero on success or an appropriate error code on failure.
364 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
372 * Interlocked extraction from node. This can race many things.
373 * We have to get a soft reference on the vnode while we hold
374 * the node locked, then acquire it properly and check for races.
376 TMPFS_NODE_LOCK(node);
377 if ((vp = node->tn_vnode) != NULL) {
378 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
380 TMPFS_NODE_UNLOCK(node);
382 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) {
386 if (node->tn_vnode != vp) {
397 * This should never happen.
399 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) {
400 TMPFS_NODE_UNLOCK(node);
406 * Interlock against other calls to tmpfs_alloc_vp() trying to
407 * allocate and assign a vp to node.
409 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
410 node->tn_vpstate |= TMPFS_VNODE_WANT;
411 error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH,
412 "tmpfs_alloc_vp", 0);
413 TMPFS_NODE_UNLOCK(node);
418 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
419 TMPFS_NODE_UNLOCK(node);
422 * Allocate a new vnode (may block). The ALLOCATING flag should
423 * prevent a race against someone else assigning node->tn_vnode.
425 error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE);
429 KKASSERT(node->tn_vnode == NULL);
430 KKASSERT(vp != NULL);
432 vp->v_type = node->tn_type;
434 /* Type-specific initialization. */
435 switch (node->tn_type) {
443 vinitvmio(vp, node->tn_size, TMPFS_BLKSIZE, -1);
448 vp->v_ops = &mp->mnt_vn_fifo_ops;
454 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
459 TMPFS_NODE_LOCK(node);
461 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
462 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
465 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
466 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
467 TMPFS_NODE_UNLOCK(node);
468 wakeup(&node->tn_vpstate);
470 TMPFS_NODE_UNLOCK(node);
476 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
478 TMPFS_NODE_LOCK(node);
479 KKASSERT(*vpp == node->tn_vnode);
480 TMPFS_NODE_UNLOCK(node);
486 /* --------------------------------------------------------------------- */
489 * Destroys the association between the vnode vp and the node it
493 tmpfs_free_vp(struct vnode *vp)
495 struct tmpfs_node *node;
497 node = VP_TO_TMPFS_NODE(vp);
499 TMPFS_NODE_LOCK(node);
500 KKASSERT(lockcount(TMPFS_NODE_MTX(node)) > 0);
501 node->tn_vnode = NULL;
503 TMPFS_NODE_UNLOCK(node);
506 /* --------------------------------------------------------------------- */
509 * Allocates a new file of type 'type' and adds it to the parent directory
510 * 'dvp'; this addition is done using the component name given in 'cnp'.
511 * The ownership of the new file is automatically assigned based on the
512 * credentials of the caller (through 'cnp'), the group is set based on
513 * the parent directory and the mode is determined from the 'vap' argument.
514 * If successful, *vpp holds a vnode to the newly created file and zero
515 * is returned. Otherwise *vpp is NULL and the function returns an
516 * appropriate error code.
519 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
520 struct namecache *ncp, struct ucred *cred, char *target)
523 struct tmpfs_dirent *de;
524 struct tmpfs_mount *tmp;
525 struct tmpfs_node *dnode;
526 struct tmpfs_node *node;
528 tmp = VFS_TO_TMPFS(dvp->v_mount);
529 dnode = VP_TO_TMPFS_DIR(dvp);
533 * If the directory was removed but a process was CD'd into it,
534 * we do not allow any more file/dir creation within it. Otherwise
535 * we will lose track of it.
537 KKASSERT(dnode->tn_type == VDIR);
538 if (dnode != tmp->tm_root && dnode->tn_dir.tn_parent == NULL)
542 * Make sure the link count does not overflow.
544 if (vap->va_type == VDIR && dnode->tn_links >= LINK_MAX)
547 /* Allocate a node that represents the new file. */
548 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid,
549 dnode->tn_gid, vap->va_mode, target,
550 vap->va_rmajor, vap->va_rminor, &node);
553 TMPFS_NODE_LOCK(node);
555 /* Allocate a directory entry that points to the new file. */
556 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de);
558 tmpfs_free_node(tmp, node);
563 /* Allocate a vnode for the new file. */
564 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
566 tmpfs_free_dirent(tmp, de);
567 tmpfs_free_node(tmp, node);
573 * Now that all required items are allocated, we can proceed to
574 * insert the new node into the directory, an operation that
577 tmpfs_dir_attach(dnode, de);
578 TMPFS_NODE_UNLOCK(node);
583 /* --------------------------------------------------------------------- */
586 * Attaches the directory entry de to the directory represented by dnode.
587 * Note that this does not change the link count of the node pointed by
588 * the directory entry, as this is done by tmpfs_alloc_dirent.
591 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
593 struct tmpfs_node *node = de->td_node;
595 TMPFS_NODE_LOCK(dnode);
596 if (node && node->tn_type == VDIR) {
597 TMPFS_NODE_LOCK(node);
599 node->tn_status |= TMPFS_NODE_CHANGED;
600 node->tn_dir.tn_parent = dnode;
602 TMPFS_NODE_UNLOCK(node);
604 RB_INSERT(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de);
605 RB_INSERT(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de);
606 dnode->tn_size += sizeof(struct tmpfs_dirent);
607 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
609 TMPFS_NODE_UNLOCK(dnode);
612 /* --------------------------------------------------------------------- */
615 * Detaches the directory entry de from the directory represented by dnode.
616 * Note that this does not change the link count of the node pointed by
617 * the directory entry, as this is done by tmpfs_free_dirent.
620 tmpfs_dir_detach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
622 struct tmpfs_node *node = de->td_node;
624 TMPFS_NODE_LOCK(dnode);
625 RB_REMOVE(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de);
626 RB_REMOVE(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de);
627 dnode->tn_size -= sizeof(struct tmpfs_dirent);
628 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
630 TMPFS_NODE_UNLOCK(dnode);
633 * Clean out the tn_parent pointer immediately when removing a
636 * Removal of the parent linkage also cleans out the extra tn_links
637 * count we had on both node and dnode.
639 * node can be NULL (typ during a forced umount), in which case
640 * the mount code is dealing with the linkages from a linked list
643 if (node && node->tn_type == VDIR && node->tn_dir.tn_parent) {
644 TMPFS_NODE_LOCK(dnode);
645 TMPFS_NODE_LOCK(node);
646 KKASSERT(node->tn_dir.tn_parent == dnode);
649 node->tn_dir.tn_parent = NULL;
650 TMPFS_NODE_UNLOCK(node);
651 TMPFS_NODE_UNLOCK(dnode);
655 /* --------------------------------------------------------------------- */
658 * Looks for a directory entry in the directory represented by node.
659 * 'ncp' describes the name of the entry to look for. Note that the .
660 * and .. components are not allowed as they do not physically exist
661 * within directories.
663 * Returns a pointer to the entry when found, otherwise NULL.
665 * Caller must hold the node locked (shared ok)
667 struct tmpfs_dirent *
668 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
669 struct namecache *ncp)
671 struct tmpfs_dirent *de;
672 int len = ncp->nc_nlen;
673 struct tmpfs_dirent wanted;
675 wanted.td_namelen = len;
676 wanted.td_name = ncp->nc_name;
678 TMPFS_VALIDATE_DIR(node);
680 de = RB_FIND(tmpfs_dirtree, &node->tn_dir.tn_dirtree, &wanted);
682 KKASSERT(f == NULL || f == de->td_node);
687 /* --------------------------------------------------------------------- */
690 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
691 * directory and returns it in the uio space. The function returns 0
692 * on success, -1 if there was not enough space in the uio structure to
693 * hold the directory entry or an appropriate error code if another
697 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
701 TMPFS_VALIDATE_DIR(node);
702 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
704 if (vop_write_dirent(&error, uio, node->tn_id, DT_DIR, 1, "."))
707 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
711 /* --------------------------------------------------------------------- */
714 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
715 * directory and returns it in the uio space. The function returns 0
716 * on success, -1 if there was not enough space in the uio structure to
717 * hold the directory entry or an appropriate error code if another
721 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
727 TMPFS_VALIDATE_DIR(node);
728 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
730 if (node->tn_dir.tn_parent) {
731 TMPFS_NODE_LOCK(node);
732 if (node->tn_dir.tn_parent)
733 d_ino = node->tn_dir.tn_parent->tn_id;
735 d_ino = tmp->tm_root->tn_id;
736 TMPFS_NODE_UNLOCK(node);
738 d_ino = tmp->tm_root->tn_id;
741 if (vop_write_dirent(&error, uio, d_ino, DT_DIR, 2, ".."))
744 struct tmpfs_dirent *de;
745 de = RB_MIN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree);
747 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
749 uio->uio_offset = tmpfs_dircookie(de);
754 /* --------------------------------------------------------------------- */
757 * Lookup a directory entry by its associated cookie.
759 * Must be called with the directory node locked (shared ok)
761 struct lubycookie_info {
763 struct tmpfs_dirent *de;
767 lubycookie_cmp(struct tmpfs_dirent *de, void *arg)
769 struct lubycookie_info *info = arg;
770 off_t cookie = tmpfs_dircookie(de);
772 if (cookie < info->cookie)
774 if (cookie > info->cookie)
780 lubycookie_callback(struct tmpfs_dirent *de, void *arg)
782 struct lubycookie_info *info = arg;
784 if (tmpfs_dircookie(de) == info->cookie) {
791 struct tmpfs_dirent *
792 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
794 struct lubycookie_info info;
796 info.cookie = cookie;
798 RB_SCAN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree,
799 lubycookie_cmp, lubycookie_callback, &info);
803 /* --------------------------------------------------------------------- */
806 * Helper function for tmpfs_readdir. Returns as much directory entries
807 * as can fit in the uio space. The read starts at uio->uio_offset.
808 * The function returns 0 on success, -1 if there was not enough space
809 * in the uio structure to hold the directory entry or an appropriate
810 * error code if another error happens.
812 * Caller must hold the node locked (shared ok)
815 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
819 struct tmpfs_dirent *de;
821 TMPFS_VALIDATE_DIR(node);
824 * Locate the first directory entry we have to return. We have cached
825 * the last readdir in the node, so use those values if appropriate.
826 * Otherwise do a linear scan to find the requested entry.
828 startcookie = uio->uio_offset;
829 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT);
830 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
832 if (startcookie == TMPFS_DIRCOOKIE_EOF)
835 de = tmpfs_dir_lookupbycookie(node, startcookie);
840 * Read as much entries as possible; i.e., until we reach the end of
841 * the directory or we exhaust uio space.
847 /* Create a dirent structure representing the current
848 * tmpfs_node and fill it. */
849 d_ino = de->td_node->tn_id;
850 switch (de->td_node->tn_type) {
880 panic("tmpfs_dir_getdents: type %p %d",
881 de->td_node, (int)de->td_node->tn_type);
883 KKASSERT(de->td_namelen < 256); /* 255 + 1 */
885 if (vop_write_dirent(&error, uio, d_ino, d_type,
886 de->td_namelen, de->td_name)) {
892 de = RB_NEXT(tmpfs_dirtree_cookie,
893 node->tn_dir.tn_cookietree, de);
894 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
896 /* Update the offset and cache. */
898 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
900 uio->uio_offset = tmpfs_dircookie(de);
906 /* --------------------------------------------------------------------- */
909 * Resizes the aobj associated to the regular file pointed to by vp to
910 * the size newsize. 'vp' must point to a vnode that represents a regular
911 * file. 'newsize' must be positive.
913 * pass trivial as 1 when buf content will be overwritten, otherwise set 0
916 * Returns zero on success or an appropriate error code on failure.
918 * Caller must hold the node exclusively locked.
921 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial)
924 vm_pindex_t newpages, oldpages;
925 struct tmpfs_mount *tmp;
926 struct tmpfs_node *node;
930 KKASSERT(vp->v_type == VREG);
931 KKASSERT(newsize >= 0);
934 node = VP_TO_TMPFS_NODE(vp);
935 tmp = VFS_TO_TMPFS(vp->v_mount);
938 * Convert the old and new sizes to the number of pages needed to
939 * store them. It may happen that we do not need to do anything
940 * because the last allocated page can accommodate the change on
943 oldsize = node->tn_size;
944 oldpages = round_page64(oldsize) / PAGE_SIZE;
945 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages);
946 newpages = round_page64(newsize) / PAGE_SIZE;
948 if (newpages > oldpages &&
949 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) {
953 node->tn_reg.tn_aobj_pages = newpages;
954 node->tn_size = newsize;
956 if (newpages != oldpages)
957 atomic_add_long(&tmp->tm_pages_used, (newpages - oldpages));
960 * When adjusting the vnode filesize and its VM object we must
961 * also adjust our backing VM object (aobj). The blocksize
962 * used must match the block sized we use for the buffer cache.
964 * The backing VM object contains no VM pages, only swap
967 if (newsize < oldsize) {
972 error = nvtruncbuf(vp, newsize, TMPFS_BLKSIZE, -1, 0);
973 aobj = node->tn_reg.tn_aobj;
976 nsize = vp->v_object->size;
979 swap_pager_freespace(aobj, nsize,
986 error = nvextendbuf(vp, oldsize, newsize,
987 TMPFS_BLKSIZE, TMPFS_BLKSIZE,
989 aobj = node->tn_reg.tn_aobj;
991 aobj->size = vp->v_object->size;
998 /* --------------------------------------------------------------------- */
1001 * Change flags of the given vnode.
1002 * Caller should execute tmpfs_update on vp after a successful execution.
1003 * The vnode must be locked on entry and remain locked on exit.
1006 tmpfs_chflags(struct vnode *vp, int vaflags, struct ucred *cred)
1009 struct tmpfs_node *node;
1012 KKASSERT(vn_islocked(vp));
1014 node = VP_TO_TMPFS_NODE(vp);
1015 flags = node->tn_flags;
1017 /* Disallow this operation if the file system is mounted read-only. */
1018 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1020 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred);
1022 /* Actually change the flags on the node itself */
1024 TMPFS_NODE_LOCK(node);
1025 node->tn_flags = flags;
1026 node->tn_status |= TMPFS_NODE_CHANGED;
1027 TMPFS_NODE_UNLOCK(node);
1030 KKASSERT(vn_islocked(vp));
1035 /* --------------------------------------------------------------------- */
1038 * Change access mode on the given vnode.
1039 * Caller should execute tmpfs_update on vp after a successful execution.
1040 * The vnode must be locked on entry and remain locked on exit.
1043 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred)
1045 struct tmpfs_node *node;
1049 KKASSERT(vn_islocked(vp));
1051 node = VP_TO_TMPFS_NODE(vp);
1053 /* Disallow this operation if the file system is mounted read-only. */
1054 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1057 /* Immutable or append-only files cannot be modified, either. */
1058 if (node->tn_flags & (IMMUTABLE | APPEND))
1061 cur_mode = node->tn_mode;
1062 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid,
1066 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) {
1067 TMPFS_NODE_LOCK(node);
1068 node->tn_mode &= ~ALLPERMS;
1069 node->tn_mode |= cur_mode & ALLPERMS;
1071 node->tn_status |= TMPFS_NODE_CHANGED;
1072 TMPFS_NODE_UNLOCK(node);
1075 KKASSERT(vn_islocked(vp));
1080 /* --------------------------------------------------------------------- */
1083 * Change ownership of the given vnode. At least one of uid or gid must
1084 * be different than VNOVAL. If one is set to that value, the attribute
1086 * Caller should execute tmpfs_update on vp after a successful execution.
1087 * The vnode must be locked on entry and remain locked on exit.
1090 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
1095 struct tmpfs_node *node;
1098 KKASSERT(vn_islocked(vp));
1099 node = VP_TO_TMPFS_NODE(vp);
1101 /* Disallow this operation if the file system is mounted read-only. */
1102 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1105 /* Immutable or append-only files cannot be modified, either. */
1106 if (node->tn_flags & (IMMUTABLE | APPEND))
1109 cur_uid = node->tn_uid;
1110 cur_gid = node->tn_gid;
1111 cur_mode = node->tn_mode;
1112 error = vop_helper_chown(vp, uid, gid, cred,
1113 &cur_uid, &cur_gid, &cur_mode);
1116 TMPFS_NODE_LOCK(node);
1117 if (cur_uid != node->tn_uid ||
1118 cur_gid != node->tn_gid ||
1119 cur_mode != node->tn_mode) {
1120 node->tn_uid = cur_uid;
1121 node->tn_gid = cur_gid;
1122 node->tn_mode = cur_mode;
1123 node->tn_status |= TMPFS_NODE_CHANGED;
1125 TMPFS_NODE_UNLOCK(node);
1131 /* --------------------------------------------------------------------- */
1134 * Change size of the given vnode.
1135 * Caller should execute tmpfs_update on vp after a successful execution.
1136 * The vnode must be locked on entry and remain locked on exit.
1139 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred)
1142 struct tmpfs_node *node;
1144 KKASSERT(vn_islocked(vp));
1146 node = VP_TO_TMPFS_NODE(vp);
1148 /* Decide whether this is a valid operation based on the file type. */
1150 switch (vp->v_type) {
1155 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1164 /* Allow modifications of special files even if in the file
1165 * system is mounted read-only (we are not modifying the
1166 * files themselves, but the objects they represent). */
1170 /* Anything else is unsupported. */
1174 /* Immutable or append-only files cannot be modified, either. */
1175 if (node->tn_flags & (IMMUTABLE | APPEND))
1178 error = tmpfs_truncate(vp, size);
1179 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1180 * for us, as will update tn_status; no need to do that here. */
1182 KKASSERT(vn_islocked(vp));
1187 /* --------------------------------------------------------------------- */
1190 * Change access and modification times of the given vnode.
1191 * Caller should execute tmpfs_update on vp after a successful execution.
1192 * The vnode must be locked on entry and remain locked on exit.
1195 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1196 int vaflags, struct ucred *cred)
1198 struct tmpfs_node *node;
1200 KKASSERT(vn_islocked(vp));
1202 node = VP_TO_TMPFS_NODE(vp);
1204 /* Disallow this operation if the file system is mounted read-only. */
1205 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1208 /* Immutable or append-only files cannot be modified, either. */
1209 if (node->tn_flags & (IMMUTABLE | APPEND))
1212 TMPFS_NODE_LOCK(node);
1213 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1214 node->tn_status |= TMPFS_NODE_ACCESSED;
1216 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1217 node->tn_status |= TMPFS_NODE_MODIFIED;
1219 TMPFS_NODE_UNLOCK(node);
1221 tmpfs_itimes(vp, atime, mtime);
1223 KKASSERT(vn_islocked(vp));
1228 /* --------------------------------------------------------------------- */
1229 /* Sync timestamps */
1231 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1232 const struct timespec *mod)
1234 struct tmpfs_node *node;
1235 struct timespec now;
1237 node = VP_TO_TMPFS_NODE(vp);
1239 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1240 TMPFS_NODE_CHANGED)) == 0)
1243 vfs_timestamp(&now);
1245 TMPFS_NODE_LOCK(node);
1246 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1249 node->tn_atime = acc->tv_sec;
1250 node->tn_atimensec = acc->tv_nsec;
1252 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1255 node->tn_mtime = mod->tv_sec;
1256 node->tn_mtimensec = mod->tv_nsec;
1258 if (node->tn_status & TMPFS_NODE_CHANGED) {
1259 node->tn_ctime = now.tv_sec;
1260 node->tn_ctimensec = now.tv_nsec;
1263 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1264 TMPFS_NODE_UNLOCK(node);
1267 /* --------------------------------------------------------------------- */
1270 tmpfs_update(struct vnode *vp)
1272 tmpfs_itimes(vp, NULL, NULL);
1275 /* --------------------------------------------------------------------- */
1278 * Caller must hold an exclusive node lock.
1281 tmpfs_truncate(struct vnode *vp, off_t length)
1284 struct tmpfs_node *node;
1286 node = VP_TO_TMPFS_NODE(vp);
1293 if (node->tn_size == length) {
1298 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1302 error = tmpfs_reg_resize(vp, length, 1);
1305 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1313 /* --------------------------------------------------------------------- */
1316 tmpfs_fetch_ino(struct tmpfs_mount *tmp)
1321 ret = tmp->tm_ino++;
1328 tmpfs_dirtree_compare(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1330 if (a->td_namelen > b->td_namelen)
1332 else if (a->td_namelen < b->td_namelen)
1335 return strncmp(a->td_name, b->td_name, a->td_namelen);
1339 tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, struct tmpfs_dirent *b)