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>
42 #include <sys/spinlock2.h>
44 #include <sys/systm.h>
45 #include <sys/vnode.h>
46 #include <sys/vmmeter.h>
49 #include <vm/vm_object.h>
50 #include <vm/vm_page.h>
51 #include <vm/vm_pager.h>
52 #include <vm/vm_extern.h>
54 #include <vfs/tmpfs/tmpfs.h>
55 #include <vfs/tmpfs/tmpfs_vnops.h>
57 static ino_t tmpfs_fetch_ino(struct tmpfs_mount *);
59 static int tmpfs_dirtree_compare(struct tmpfs_dirent *a,
60 struct tmpfs_dirent *b);
61 RB_GENERATE(tmpfs_dirtree, tmpfs_dirent, rb_node, tmpfs_dirtree_compare);
63 static int tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a,
64 struct tmpfs_dirent *b);
65 RB_GENERATE(tmpfs_dirtree_cookie, tmpfs_dirent,
66 rb_cookienode, tmpfs_dirtree_compare_cookie);
69 /* --------------------------------------------------------------------- */
72 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
73 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
74 * using the credentials of the process 'p'.
76 * If the node type is set to 'VDIR', then the parent parameter must point
77 * to the parent directory of the node being created. It may only be NULL
78 * while allocating the root node.
80 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
81 * specifies the device the node represents.
83 * If the node type is set to 'VLNK', then the parameter target specifies
84 * the file name of the target file for the symbolic link that is being
87 * Note that new nodes are retrieved from the available list if it has
88 * items or, if it is empty, from the node pool as long as there is enough
89 * space to create them.
91 * Returns zero on success or an appropriate error code on failure.
94 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
95 uid_t uid, gid_t gid, mode_t mode,
96 char *target, int rmajor, int rminor,
97 struct tmpfs_node **node)
99 struct tmpfs_node *nnode;
103 KKASSERT(IFF(type == VLNK, target != NULL));
104 KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL));
106 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
109 nnode = objcache_get(tmp->tm_node_pool, M_WAITOK | M_NULLOK);
113 /* Generic initialization. */
114 nnode->tn_type = type;
116 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime
118 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec
122 nnode->tn_mode = mode;
123 nnode->tn_id = tmpfs_fetch_ino(tmp);
124 nnode->tn_advlock.init_done = 0;
125 KKASSERT(nnode->tn_links == 0);
127 /* Type-specific initialization. */
128 switch (nnode->tn_type) {
131 rdev = makeudev(rmajor, rminor);
132 if (rdev == NOUDEV) {
133 objcache_put(tmp->tm_node_pool, nnode);
136 nnode->tn_rdev = rdev;
140 RB_INIT(&nnode->tn_dir.tn_dirtree);
141 RB_INIT(&nnode->tn_dir.tn_cookietree);
151 nnode->tn_size = strlen(target);
152 nnode->tn_link = kmalloc(nnode->tn_size + 1, tmp->tm_name_zone,
153 M_WAITOK | M_NULLOK);
154 if (nnode->tn_link == NULL) {
155 objcache_put(tmp->tm_node_pool, nnode);
158 bcopy(target, nnode->tn_link, nnode->tn_size);
159 nnode->tn_link[nnode->tn_size] = '\0';
163 nnode->tn_reg.tn_aobj =
164 swap_pager_alloc(NULL, 0, VM_PROT_DEFAULT, 0);
165 nnode->tn_reg.tn_aobj_pages = 0;
170 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
173 TMPFS_NODE_LOCK(nnode);
175 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
176 tmp->tm_nodes_inuse++;
178 TMPFS_NODE_UNLOCK(nnode);
184 /* --------------------------------------------------------------------- */
187 * Destroys the node pointed to by node from the file system 'tmp'.
188 * If the node does not belong to the given mount point, the results are
191 * If the node references a directory; no entries are allowed because
192 * their removal could need a recursive algorithm, something forbidden in
193 * kernel space. Furthermore, there is not need to provide such
194 * functionality (recursive removal) because the only primitives offered
195 * to the user are the removal of empty directories and the deletion of
198 * Note that nodes are not really deleted; in fact, when a node has been
199 * allocated, it cannot be deleted during the whole life of the file
200 * system. Instead, they are moved to the available list and remain there
204 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
206 vm_pindex_t pages = 0;
209 TMPFS_ASSERT_ELOCKED(node);
210 KKASSERT(node->tn_vnode == NULL);
211 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
215 LIST_REMOVE(node, tn_entries);
216 tmp->tm_nodes_inuse--;
218 TMPFS_NODE_UNLOCK(node);
220 switch (node->tn_type) {
222 /* Do not do anything. VNON is provided to let the
223 * allocation routine clean itself easily by avoiding
224 * duplicating code in it. */
233 * The parent link can be NULL if this is the root
234 * node or if it is a directory node that was rmdir'd.
236 * XXX what if node is a directory which still contains
237 * directory entries (e.g. due to a forced umount) ?
240 KKASSERT(node->tn_dir.tn_parent == NULL);
243 * If the root node is being destroyed don't leave a
244 * dangling pointer in tmpfs_mount.
246 if (node == tmp->tm_root)
255 kfree(node->tn_link, tmp->tm_name_zone);
256 node->tn_link = NULL;
261 if (node->tn_reg.tn_aobj != NULL)
262 vm_object_deallocate(node->tn_reg.tn_aobj);
263 node->tn_reg.tn_aobj = NULL;
264 pages = node->tn_reg.tn_aobj_pages;
268 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
272 * Clean up fields for the next allocation. The objcache only ctors
275 tmpfs_node_ctor(node, NULL, 0);
276 objcache_put(tmp->tm_node_pool, node);
277 /* node is now invalid */
280 tmp->tm_pages_used -= 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_BLKMASK, -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 vp.
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 vp.
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)
703 TMPFS_VALIDATE_DIR(node);
704 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
706 dent.d_ino = node->tn_id;
707 dent.d_type = DT_DIR;
709 dent.d_name[0] = '.';
710 dent.d_name[1] = '\0';
711 dirsize = _DIRENT_DIRSIZ(&dent);
713 if (dirsize > uio->uio_resid)
716 error = uiomove((caddr_t)&dent, dirsize, uio);
718 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
723 /* --------------------------------------------------------------------- */
726 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
727 * directory and returns it in the uio space. The function returns 0
728 * on success, -1 if there was not enough space in the uio structure to
729 * hold the directory entry or an appropriate error code if another
733 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
740 TMPFS_VALIDATE_DIR(node);
741 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
743 if (node->tn_dir.tn_parent) {
744 TMPFS_NODE_LOCK(node);
745 if (node->tn_dir.tn_parent)
746 dent.d_ino = node->tn_dir.tn_parent->tn_id;
748 dent.d_ino = tmp->tm_root->tn_id;
749 TMPFS_NODE_UNLOCK(node);
751 dent.d_ino = tmp->tm_root->tn_id;
754 dent.d_type = DT_DIR;
756 dent.d_name[0] = '.';
757 dent.d_name[1] = '.';
758 dent.d_name[2] = '\0';
759 dirsize = _DIRENT_DIRSIZ(&dent);
761 if (dirsize > uio->uio_resid)
764 error = uiomove((caddr_t)&dent, dirsize, uio);
766 struct tmpfs_dirent *de;
768 de = RB_MIN(tmpfs_dirtree_cookie,
769 &node->tn_dir.tn_cookietree);
771 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
773 uio->uio_offset = tmpfs_dircookie(de);
779 /* --------------------------------------------------------------------- */
782 * Lookup a directory entry by its associated cookie.
784 * Must be called with the directory node locked (shared ok)
786 struct lubycookie_info {
788 struct tmpfs_dirent *de;
792 lubycookie_cmp(struct tmpfs_dirent *de, void *arg)
794 struct lubycookie_info *info = arg;
795 off_t cookie = tmpfs_dircookie(de);
797 if (cookie < info->cookie)
799 if (cookie > info->cookie)
805 lubycookie_callback(struct tmpfs_dirent *de, void *arg)
807 struct lubycookie_info *info = arg;
809 if (tmpfs_dircookie(de) == info->cookie) {
816 struct tmpfs_dirent *
817 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
819 struct lubycookie_info info;
821 info.cookie = cookie;
823 RB_SCAN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree,
824 lubycookie_cmp, lubycookie_callback, &info);
828 /* --------------------------------------------------------------------- */
831 * Helper function for tmpfs_readdir. Returns as much directory entries
832 * as can fit in the uio space. The read starts at uio->uio_offset.
833 * The function returns 0 on success, -1 if there was not enough space
834 * in the uio structure to hold the directory entry or an appropriate
835 * error code if another error happens.
837 * Caller must hold the node locked (shared ok)
840 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
844 struct tmpfs_dirent *de;
846 TMPFS_VALIDATE_DIR(node);
849 * Locate the first directory entry we have to return. We have cached
850 * the last readdir in the node, so use those values if appropriate.
851 * Otherwise do a linear scan to find the requested entry.
853 startcookie = uio->uio_offset;
854 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT);
855 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
857 if (startcookie == TMPFS_DIRCOOKIE_EOF)
860 de = tmpfs_dir_lookupbycookie(node, startcookie);
865 * Read as much entries as possible; i.e., until we reach the end of
866 * the directory or we exhaust uio space.
872 /* Create a dirent structure representing the current
873 * tmpfs_node and fill it. */
874 d.d_ino = de->td_node->tn_id;
875 switch (de->td_node->tn_type) {
905 panic("tmpfs_dir_getdents: type %p %d",
906 de->td_node, (int)de->td_node->tn_type);
908 d.d_namlen = de->td_namelen;
909 KKASSERT(de->td_namelen < sizeof(d.d_name));
910 bcopy(de->td_name, d.d_name, d.d_namlen);
911 d.d_name[d.d_namlen] = '\0';
912 reclen = _DIRENT_RECLEN(d.d_namlen);
914 /* Stop reading if the directory entry we are treating is
915 * bigger than the amount of data that can be returned. */
916 if (reclen > uio->uio_resid) {
921 /* Copy the new dirent structure into the output buffer and
922 * advance pointers. */
923 error = uiomove((caddr_t)&d, reclen, uio);
926 de = RB_NEXT(tmpfs_dirtree_cookie,
927 node->tn_dir.tn_cookietree, de);
928 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
930 /* Update the offset and cache. */
932 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
934 uio->uio_offset = tmpfs_dircookie(de);
940 /* --------------------------------------------------------------------- */
943 * Resizes the aobj associated to the regular file pointed to by vp to
944 * the size newsize. 'vp' must point to a vnode that represents a regular
945 * file. 'newsize' must be positive.
947 * pass trivial as 1 when buf content will be overwritten, otherwise set 0
950 * Returns zero on success or an appropriate error code on failure.
953 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial)
956 vm_pindex_t newpages, oldpages;
957 struct tmpfs_mount *tmp;
958 struct tmpfs_node *node;
962 KKASSERT(vp->v_type == VREG);
963 KKASSERT(newsize >= 0);
966 node = VP_TO_TMPFS_NODE(vp);
967 tmp = VFS_TO_TMPFS(vp->v_mount);
970 * Convert the old and new sizes to the number of pages needed to
971 * store them. It may happen that we do not need to do anything
972 * because the last allocated page can accommodate the change on
975 TMPFS_NODE_LOCK(node);
976 oldsize = node->tn_size;
977 oldpages = round_page64(oldsize) / PAGE_SIZE;
978 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages);
979 newpages = round_page64(newsize) / PAGE_SIZE;
981 if (newpages > oldpages &&
982 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) {
983 TMPFS_NODE_UNLOCK(node);
987 node->tn_reg.tn_aobj_pages = newpages;
988 node->tn_size = newsize;
989 TMPFS_NODE_UNLOCK(node);
992 tmp->tm_pages_used += (newpages - oldpages);
996 * When adjusting the vnode filesize and its VM object we must
997 * also adjust our backing VM object (aobj). The blocksize
998 * used must match the block sized we use for the buffer cache.
1000 * The backing VM object contains no VM pages, only swap
1003 if (newsize < oldsize) {
1008 error = nvtruncbuf(vp, newsize, TMPFS_BLKSIZE, -1, 0);
1009 aobj = node->tn_reg.tn_aobj;
1012 nsize = vp->v_object->size;
1013 if (nsize < osize) {
1015 swap_pager_freespace(aobj, nsize,
1022 error = nvextendbuf(vp, oldsize, newsize,
1023 TMPFS_BLKSIZE, TMPFS_BLKSIZE,
1025 aobj = node->tn_reg.tn_aobj;
1027 aobj->size = vp->v_object->size;
1034 /* --------------------------------------------------------------------- */
1037 * Change flags of the given vnode.
1038 * Caller should execute tmpfs_update on vp after a successful execution.
1039 * The vnode must be locked on entry and remain locked on exit.
1042 tmpfs_chflags(struct vnode *vp, int vaflags, struct ucred *cred)
1045 struct tmpfs_node *node;
1048 KKASSERT(vn_islocked(vp));
1050 node = VP_TO_TMPFS_NODE(vp);
1051 flags = node->tn_flags;
1053 /* Disallow this operation if the file system is mounted read-only. */
1054 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1056 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred);
1058 /* Actually change the flags on the node itself */
1060 TMPFS_NODE_LOCK(node);
1061 node->tn_flags = flags;
1062 node->tn_status |= TMPFS_NODE_CHANGED;
1063 TMPFS_NODE_UNLOCK(node);
1066 KKASSERT(vn_islocked(vp));
1071 /* --------------------------------------------------------------------- */
1074 * Change access mode on the given vnode.
1075 * Caller should execute tmpfs_update on vp after a successful execution.
1076 * The vnode must be locked on entry and remain locked on exit.
1079 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred)
1081 struct tmpfs_node *node;
1085 KKASSERT(vn_islocked(vp));
1087 node = VP_TO_TMPFS_NODE(vp);
1089 /* Disallow this operation if the file system is mounted read-only. */
1090 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1093 /* Immutable or append-only files cannot be modified, either. */
1094 if (node->tn_flags & (IMMUTABLE | APPEND))
1097 cur_mode = node->tn_mode;
1098 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid,
1102 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) {
1103 TMPFS_NODE_LOCK(node);
1104 node->tn_mode &= ~ALLPERMS;
1105 node->tn_mode |= cur_mode & ALLPERMS;
1107 node->tn_status |= TMPFS_NODE_CHANGED;
1108 TMPFS_NODE_UNLOCK(node);
1111 KKASSERT(vn_islocked(vp));
1116 /* --------------------------------------------------------------------- */
1119 * Change ownership of the given vnode. At least one of uid or gid must
1120 * be different than VNOVAL. If one is set to that value, the attribute
1122 * Caller should execute tmpfs_update on vp after a successful execution.
1123 * The vnode must be locked on entry and remain locked on exit.
1126 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
1131 struct tmpfs_node *node;
1134 KKASSERT(vn_islocked(vp));
1135 node = VP_TO_TMPFS_NODE(vp);
1137 /* Disallow this operation if the file system is mounted read-only. */
1138 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1141 /* Immutable or append-only files cannot be modified, either. */
1142 if (node->tn_flags & (IMMUTABLE | APPEND))
1145 cur_uid = node->tn_uid;
1146 cur_gid = node->tn_gid;
1147 cur_mode = node->tn_mode;
1148 error = vop_helper_chown(vp, uid, gid, cred,
1149 &cur_uid, &cur_gid, &cur_mode);
1152 TMPFS_NODE_LOCK(node);
1153 if (cur_uid != node->tn_uid ||
1154 cur_gid != node->tn_gid ||
1155 cur_mode != node->tn_mode) {
1156 node->tn_uid = cur_uid;
1157 node->tn_gid = cur_gid;
1158 node->tn_mode = cur_mode;
1159 node->tn_status |= TMPFS_NODE_CHANGED;
1161 TMPFS_NODE_UNLOCK(node);
1167 /* --------------------------------------------------------------------- */
1170 * Change size of the given vnode.
1171 * Caller should execute tmpfs_update on vp after a successful execution.
1172 * The vnode must be locked on entry and remain locked on exit.
1175 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred)
1178 struct tmpfs_node *node;
1180 KKASSERT(vn_islocked(vp));
1182 node = VP_TO_TMPFS_NODE(vp);
1184 /* Decide whether this is a valid operation based on the file type. */
1186 switch (vp->v_type) {
1191 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1200 /* Allow modifications of special files even if in the file
1201 * system is mounted read-only (we are not modifying the
1202 * files themselves, but the objects they represent). */
1206 /* Anything else is unsupported. */
1210 /* Immutable or append-only files cannot be modified, either. */
1211 if (node->tn_flags & (IMMUTABLE | APPEND))
1214 error = tmpfs_truncate(vp, size);
1215 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1216 * for us, as will update tn_status; no need to do that here. */
1218 KKASSERT(vn_islocked(vp));
1223 /* --------------------------------------------------------------------- */
1226 * Change access and modification times of the given vnode.
1227 * Caller should execute tmpfs_update on vp after a successful execution.
1228 * The vnode must be locked on entry and remain locked on exit.
1231 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1232 int vaflags, struct ucred *cred)
1234 struct tmpfs_node *node;
1236 KKASSERT(vn_islocked(vp));
1238 node = VP_TO_TMPFS_NODE(vp);
1240 /* Disallow this operation if the file system is mounted read-only. */
1241 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1244 /* Immutable or append-only files cannot be modified, either. */
1245 if (node->tn_flags & (IMMUTABLE | APPEND))
1248 TMPFS_NODE_LOCK(node);
1249 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1250 node->tn_status |= TMPFS_NODE_ACCESSED;
1252 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1253 node->tn_status |= TMPFS_NODE_MODIFIED;
1255 TMPFS_NODE_UNLOCK(node);
1257 tmpfs_itimes(vp, atime, mtime);
1259 KKASSERT(vn_islocked(vp));
1264 /* --------------------------------------------------------------------- */
1265 /* Sync timestamps */
1267 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1268 const struct timespec *mod)
1270 struct tmpfs_node *node;
1271 struct timespec now;
1273 node = VP_TO_TMPFS_NODE(vp);
1275 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1276 TMPFS_NODE_CHANGED)) == 0)
1279 vfs_timestamp(&now);
1281 TMPFS_NODE_LOCK(node);
1282 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1285 node->tn_atime = acc->tv_sec;
1286 node->tn_atimensec = acc->tv_nsec;
1288 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1291 node->tn_mtime = mod->tv_sec;
1292 node->tn_mtimensec = mod->tv_nsec;
1294 if (node->tn_status & TMPFS_NODE_CHANGED) {
1295 node->tn_ctime = now.tv_sec;
1296 node->tn_ctimensec = now.tv_nsec;
1299 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1300 TMPFS_NODE_UNLOCK(node);
1303 /* --------------------------------------------------------------------- */
1306 tmpfs_update(struct vnode *vp)
1308 tmpfs_itimes(vp, NULL, NULL);
1311 /* --------------------------------------------------------------------- */
1314 tmpfs_truncate(struct vnode *vp, off_t length)
1317 struct tmpfs_node *node;
1319 node = VP_TO_TMPFS_NODE(vp);
1326 if (node->tn_size == length) {
1331 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1335 error = tmpfs_reg_resize(vp, length, 1);
1338 TMPFS_NODE_LOCK(node);
1339 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1340 TMPFS_NODE_UNLOCK(node);
1349 /* --------------------------------------------------------------------- */
1352 tmpfs_fetch_ino(struct tmpfs_mount *tmp)
1356 ret = tmp->tm_ino++;
1362 tmpfs_dirtree_compare(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1364 if (a->td_namelen > b->td_namelen)
1366 else if (a->td_namelen < b->td_namelen)
1369 return strncmp(a->td_name, b->td_name, a->td_namelen);
1373 tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, struct tmpfs_dirent *b)