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 *);
58 static int tmpfs_dirtree_compare(struct tmpfs_dirent *a,
59 struct tmpfs_dirent *b);
61 RB_GENERATE(tmpfs_dirtree, tmpfs_dirent, rb_node, tmpfs_dirtree_compare);
64 /* --------------------------------------------------------------------- */
67 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
68 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
69 * using the credentials of the process 'p'.
71 * If the node type is set to 'VDIR', then the parent parameter must point
72 * to the parent directory of the node being created. It may only be NULL
73 * while allocating the root node.
75 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
76 * specifies the device the node represents.
78 * If the node type is set to 'VLNK', then the parameter target specifies
79 * the file name of the target file for the symbolic link that is being
82 * Note that new nodes are retrieved from the available list if it has
83 * items or, if it is empty, from the node pool as long as there is enough
84 * space to create them.
86 * Returns zero on success or an appropriate error code on failure.
89 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
90 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
91 char *target, int rmajor, int rminor, struct tmpfs_node **node)
93 struct tmpfs_node *nnode;
97 /* If the root directory of the 'tmp' file system is not yet
98 * allocated, this must be the request to do it. */
99 KKASSERT(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
101 KKASSERT(IFF(type == VLNK, target != NULL));
102 KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL));
104 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
107 nnode = objcache_get(tmp->tm_node_pool, M_WAITOK | M_NULLOK);
111 /* Generic initialization. */
112 nnode->tn_type = type;
114 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime
116 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec
120 nnode->tn_mode = mode;
121 nnode->tn_id = tmpfs_fetch_ino(tmp);
122 nnode->tn_advlock.init_done = 0;
124 /* Type-specific initialization. */
125 switch (nnode->tn_type) {
128 rdev = makeudev(rmajor, rminor);
129 if (rdev == NOUDEV) {
130 objcache_put(tmp->tm_node_pool, nnode);
133 nnode->tn_rdev = rdev;
137 RB_INIT(&nnode->tn_dir.tn_dirtree);
138 KKASSERT(parent != nnode);
139 KKASSERT(IMPLIES(parent == NULL, tmp->tm_root == NULL));
140 nnode->tn_dir.tn_parent = parent;
141 nnode->tn_dir.tn_readdir_lastn = 0;
142 nnode->tn_dir.tn_readdir_lastp = NULL;
146 TMPFS_NODE_LOCK(parent);
148 TMPFS_NODE_UNLOCK(parent);
158 nnode->tn_size = strlen(target);
159 nnode->tn_link = kmalloc(nnode->tn_size + 1, tmp->tm_name_zone,
160 M_WAITOK | M_NULLOK);
161 if (nnode->tn_link == NULL) {
162 objcache_put(tmp->tm_node_pool, nnode);
165 bcopy(target, nnode->tn_link, nnode->tn_size);
166 nnode->tn_link[nnode->tn_size] = '\0';
170 nnode->tn_reg.tn_aobj =
171 swap_pager_alloc(NULL, 0, VM_PROT_DEFAULT, 0);
172 nnode->tn_reg.tn_aobj_pages = 0;
177 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
180 TMPFS_NODE_LOCK(nnode);
182 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
183 tmp->tm_nodes_inuse++;
185 TMPFS_NODE_UNLOCK(nnode);
191 /* --------------------------------------------------------------------- */
194 * Destroys the node pointed to by node from the file system 'tmp'.
195 * If the node does not belong to the given mount point, the results are
198 * If the node references a directory; no entries are allowed because
199 * their removal could need a recursive algorithm, something forbidden in
200 * kernel space. Furthermore, there is not need to provide such
201 * functionality (recursive removal) because the only primitives offered
202 * to the user are the removal of empty directories and the deletion of
205 * Note that nodes are not really deleted; in fact, when a node has been
206 * allocated, it cannot be deleted during the whole life of the file
207 * system. Instead, they are moved to the available list and remain there
211 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
213 vm_pindex_t pages = 0;
216 TMPFS_ASSERT_ELOCKED(node);
217 KKASSERT(node->tn_vnode == NULL);
218 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
222 LIST_REMOVE(node, tn_entries);
223 tmp->tm_nodes_inuse--;
225 TMPFS_NODE_UNLOCK(node);
227 switch (node->tn_type) {
229 /* Do not do anything. VNON is provided to let the
230 * allocation routine clean itself easily by avoiding
231 * duplicating code in it. */
240 * The parent link can be NULL if this is the root
245 KKASSERT(node->tn_dir.tn_parent || node == tmp->tm_root);
246 if (node->tn_dir.tn_parent) {
247 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
248 node->tn_dir.tn_parent->tn_links--;
251 * If the parent directory has no more links and
252 * no vnode ref nothing is going to come along
253 * and clean it up unless we do it here.
255 if (node->tn_dir.tn_parent->tn_links == 0 &&
256 node->tn_dir.tn_parent->tn_vnode == NULL) {
257 tmpfs_free_node(tmp, node->tn_dir.tn_parent);
258 /* eats parent lock */
260 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
262 node->tn_dir.tn_parent = NULL;
266 * If the root node is being destroyed don't leave a
267 * dangling pointer in tmpfs_mount.
269 if (node == tmp->tm_root)
278 kfree(node->tn_link, tmp->tm_name_zone);
279 node->tn_link = NULL;
284 if (node->tn_reg.tn_aobj != NULL)
285 vm_object_deallocate(node->tn_reg.tn_aobj);
286 node->tn_reg.tn_aobj = NULL;
287 pages = node->tn_reg.tn_aobj_pages;
291 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
295 * Clean up fields for the next allocation. The objcache only ctors
298 tmpfs_node_ctor(node, NULL, 0);
299 objcache_put(tmp->tm_node_pool, node);
300 /* node is now invalid */
303 tmp->tm_pages_used -= pages;
307 /* --------------------------------------------------------------------- */
310 * Allocates a new directory entry for the node node with a name of name.
311 * The new directory entry is returned in *de.
313 * The link count of node is increased by one to reflect the new object
316 * Returns zero on success or an appropriate error code on failure.
319 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
320 const char *name, uint16_t len, struct tmpfs_dirent **de)
322 struct tmpfs_dirent *nde;
324 nde = objcache_get(tmp->tm_dirent_pool, M_WAITOK);
325 nde->td_name = kmalloc(len + 1, tmp->tm_name_zone, M_WAITOK | M_NULLOK);
326 if (nde->td_name == NULL) {
327 objcache_put(tmp->tm_dirent_pool, nde);
331 nde->td_namelen = len;
332 bcopy(name, nde->td_name, len);
333 nde->td_name[len] = '\0';
337 TMPFS_NODE_LOCK(node);
339 TMPFS_NODE_UNLOCK(node);
346 /* --------------------------------------------------------------------- */
349 * Frees a directory entry. It is the caller's responsibility to destroy
350 * the node referenced by it if needed.
352 * The link count of node is decreased by one to reflect the removal of an
353 * object that referenced it. This only happens if 'node_exists' is true;
354 * otherwise the function will not access the node referred to by the
355 * directory entry, as it may already have been released from the outside.
358 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
360 struct tmpfs_node *node;
364 TMPFS_NODE_LOCK(node);
365 TMPFS_ASSERT_ELOCKED(node);
366 KKASSERT(node->tn_links > 0);
368 TMPFS_NODE_UNLOCK(node);
370 kfree(de->td_name, tmp->tm_name_zone);
374 objcache_put(tmp->tm_dirent_pool, de);
377 /* --------------------------------------------------------------------- */
380 * Allocates a new vnode for the node node or returns a new reference to
381 * an existing one if the node had already a vnode referencing it. The
382 * resulting locked vnode is returned in *vpp.
384 * Returns zero on success or an appropriate error code on failure.
387 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
395 * Interlocked extraction from node. This can race many things.
396 * We have to get a soft reference on the vnode while we hold
397 * the node locked, then acquire it properly and check for races.
399 TMPFS_NODE_LOCK(node);
400 if ((vp = node->tn_vnode) != NULL) {
401 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
402 vhold_interlocked(vp);
403 TMPFS_NODE_UNLOCK(node);
405 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) {
409 if (node->tn_vnode != vp) {
420 * This should never happen.
422 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) {
423 TMPFS_NODE_UNLOCK(node);
429 * Interlock against other calls to tmpfs_alloc_vp() trying to
430 * allocate and assign a vp to node.
432 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
433 node->tn_vpstate |= TMPFS_VNODE_WANT;
434 error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH,
435 "tmpfs_alloc_vp", 0);
436 TMPFS_NODE_UNLOCK(node);
441 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
442 TMPFS_NODE_UNLOCK(node);
445 * Allocate a new vnode (may block). The ALLOCATING flag should
446 * prevent a race against someone else assigning node->tn_vnode.
448 error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE);
452 KKASSERT(node->tn_vnode == NULL);
453 KKASSERT(vp != NULL);
455 vp->v_type = node->tn_type;
457 /* Type-specific initialization. */
458 switch (node->tn_type) {
466 vinitvmio(vp, node->tn_size, BMASK, -1);
471 vp->v_ops = &mp->mnt_vn_fifo_ops;
477 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
483 TMPFS_NODE_LOCK(node);
485 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
486 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
489 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
490 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
491 TMPFS_NODE_UNLOCK(node);
492 wakeup(&node->tn_vpstate);
494 TMPFS_NODE_UNLOCK(node);
500 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
502 TMPFS_NODE_LOCK(node);
503 KKASSERT(*vpp == node->tn_vnode);
504 TMPFS_NODE_UNLOCK(node);
510 /* --------------------------------------------------------------------- */
513 * Destroys the association between the vnode vp and the node it
517 tmpfs_free_vp(struct vnode *vp)
519 struct tmpfs_node *node;
521 node = VP_TO_TMPFS_NODE(vp);
523 TMPFS_NODE_LOCK(node);
524 KKASSERT(lockcount(TMPFS_NODE_MTX(node)) > 0);
525 node->tn_vnode = NULL;
526 TMPFS_NODE_UNLOCK(node);
530 /* --------------------------------------------------------------------- */
533 * Allocates a new file of type 'type' and adds it to the parent directory
534 * 'dvp'; this addition is done using the component name given in 'cnp'.
535 * The ownership of the new file is automatically assigned based on the
536 * credentials of the caller (through 'cnp'), the group is set based on
537 * the parent directory and the mode is determined from the 'vap' argument.
538 * If successful, *vpp holds a vnode to the newly created file and zero
539 * is returned. Otherwise *vpp is NULL and the function returns an
540 * appropriate error code.
543 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
544 struct namecache *ncp, struct ucred *cred, char *target)
547 struct tmpfs_dirent *de;
548 struct tmpfs_mount *tmp;
549 struct tmpfs_node *dnode;
550 struct tmpfs_node *node;
551 struct tmpfs_node *parent;
553 tmp = VFS_TO_TMPFS(dvp->v_mount);
554 dnode = VP_TO_TMPFS_DIR(dvp);
557 /* If the entry we are creating is a directory, we cannot overflow
558 * the number of links of its parent, because it will get a new
560 if (vap->va_type == VDIR) {
561 /* Ensure that we do not overflow the maximum number of links
562 * imposed by the system. */
563 KKASSERT(dnode->tn_links <= LINK_MAX);
564 if (dnode->tn_links == LINK_MAX) {
569 KKASSERT(parent != NULL);
573 /* Allocate a node that represents the new file. */
574 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid,
575 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rmajor, vap->va_rminor, &node);
578 TMPFS_NODE_LOCK(node);
580 /* Allocate a directory entry that points to the new file. */
581 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de);
583 tmpfs_free_node(tmp, node);
588 /* Allocate a vnode for the new file. */
589 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
591 tmpfs_free_dirent(tmp, de);
592 tmpfs_free_node(tmp, node);
597 /* Now that all required items are allocated, we can proceed to
598 * insert the new node into the directory, an operation that
600 tmpfs_dir_attach(dnode, de);
601 TMPFS_NODE_UNLOCK(node);
606 /* --------------------------------------------------------------------- */
609 * Attaches the directory entry de to the directory represented by vp.
610 * Note that this does not change the link count of the node pointed by
611 * the directory entry, as this is done by tmpfs_alloc_dirent.
614 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
616 TMPFS_NODE_LOCK(dnode);
617 RB_INSERT(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de);
619 TMPFS_ASSERT_ELOCKED(dnode);
620 dnode->tn_size += sizeof(struct tmpfs_dirent);
621 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
623 TMPFS_NODE_UNLOCK(dnode);
626 /* --------------------------------------------------------------------- */
629 * Detaches the directory entry de from the directory represented by vp.
630 * Note that this does not change the link count of the node pointed by
631 * the directory entry, as this is done by tmpfs_free_dirent.
634 tmpfs_dir_detach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
636 TMPFS_NODE_LOCK(dnode);
637 if (dnode->tn_dir.tn_readdir_lastp == de) {
638 dnode->tn_dir.tn_readdir_lastn = 0;
639 dnode->tn_dir.tn_readdir_lastp = NULL;
641 RB_REMOVE(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de);
643 TMPFS_ASSERT_ELOCKED(dnode);
644 dnode->tn_size -= sizeof(struct tmpfs_dirent);
645 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
647 TMPFS_NODE_UNLOCK(dnode);
650 /* --------------------------------------------------------------------- */
653 * Looks for a directory entry in the directory represented by node.
654 * 'ncp' describes the name of the entry to look for. Note that the .
655 * and .. components are not allowed as they do not physically exist
656 * within directories.
658 * Returns a pointer to the entry when found, otherwise NULL.
660 struct tmpfs_dirent *
661 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
662 struct namecache *ncp)
664 struct tmpfs_dirent *de;
665 int len = ncp->nc_nlen;
666 struct tmpfs_dirent wanted;
668 wanted.td_namelen = len;
669 wanted.td_name = ncp->nc_name;
671 TMPFS_VALIDATE_DIR(node);
673 de = RB_FIND(tmpfs_dirtree, &node->tn_dir.tn_dirtree, &wanted);
675 KKASSERT(f == NULL || f == de->td_node);
677 TMPFS_NODE_LOCK(node);
678 node->tn_status |= TMPFS_NODE_ACCESSED;
679 TMPFS_NODE_UNLOCK(node);
684 /* --------------------------------------------------------------------- */
687 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
688 * directory and returns it in the uio space. The function returns 0
689 * on success, -1 if there was not enough space in the uio structure to
690 * hold the directory entry or an appropriate error code if another
694 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
700 TMPFS_VALIDATE_DIR(node);
701 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
703 dent.d_ino = node->tn_id;
704 dent.d_type = DT_DIR;
706 dent.d_name[0] = '.';
707 dent.d_name[1] = '\0';
708 dirsize = _DIRENT_DIRSIZ(&dent);
710 if (dirsize > uio->uio_resid)
713 error = uiomove((caddr_t)&dent, dirsize, uio);
715 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
718 TMPFS_NODE_LOCK(node);
719 node->tn_status |= TMPFS_NODE_ACCESSED;
720 TMPFS_NODE_UNLOCK(node);
725 /* --------------------------------------------------------------------- */
728 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
729 * directory and returns it in the uio space. The function returns 0
730 * on success, -1 if there was not enough space in the uio structure to
731 * hold the directory entry or an appropriate error code if another
735 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
742 TMPFS_VALIDATE_DIR(node);
743 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
745 if (node->tn_dir.tn_parent) {
746 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
747 dent.d_ino = node->tn_dir.tn_parent->tn_id;
748 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
750 dent.d_ino = tmp->tm_root->tn_id;
753 dent.d_type = DT_DIR;
755 dent.d_name[0] = '.';
756 dent.d_name[1] = '.';
757 dent.d_name[2] = '\0';
758 dirsize = _DIRENT_DIRSIZ(&dent);
760 if (dirsize > uio->uio_resid)
763 error = uiomove((caddr_t)&dent, dirsize, uio);
765 struct tmpfs_dirent *de;
767 de = RB_MIN(tmpfs_dirtree, &node->tn_dir.tn_dirtree);
769 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
771 uio->uio_offset = tmpfs_dircookie(de);
775 TMPFS_NODE_LOCK(node);
776 node->tn_status |= TMPFS_NODE_ACCESSED;
777 TMPFS_NODE_UNLOCK(node);
782 /* --------------------------------------------------------------------- */
785 * Lookup a directory entry by its associated cookie.
787 struct tmpfs_dirent *
788 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
790 struct tmpfs_dirent *de;
792 if (cookie == node->tn_dir.tn_readdir_lastn &&
793 node->tn_dir.tn_readdir_lastp != NULL) {
794 return node->tn_dir.tn_readdir_lastp;
797 RB_FOREACH(de, tmpfs_dirtree, &node->tn_dir.tn_dirtree) {
798 if (tmpfs_dircookie(de) == cookie) {
806 /* --------------------------------------------------------------------- */
809 * Helper function for tmpfs_readdir. Returns as much directory entries
810 * as can fit in the uio space. The read starts at uio->uio_offset.
811 * The function returns 0 on success, -1 if there was not enough space
812 * in the uio structure to hold the directory entry or an appropriate
813 * error code if another error happens.
816 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
820 struct tmpfs_dirent *de;
822 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. */
827 startcookie = uio->uio_offset;
828 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT);
829 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
830 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
833 de = tmpfs_dir_lookupbycookie(node, startcookie);
839 /* Read as much entries as possible; i.e., until we reach the end of
840 * the directory or we exhaust uio space. */
845 /* Create a dirent structure representing the current
846 * tmpfs_node and fill it. */
847 d.d_ino = de->td_node->tn_id;
848 switch (de->td_node->tn_type) {
878 panic("tmpfs_dir_getdents: type %p %d",
879 de->td_node, (int)de->td_node->tn_type);
881 d.d_namlen = de->td_namelen;
882 KKASSERT(de->td_namelen < sizeof(d.d_name));
883 bcopy(de->td_name, d.d_name, d.d_namlen);
884 d.d_name[d.d_namlen] = '\0';
885 reclen = _DIRENT_RECLEN(d.d_namlen);
887 /* Stop reading if the directory entry we are treating is
888 * bigger than the amount of data that can be returned. */
889 if (reclen > uio->uio_resid) {
894 /* Copy the new dirent structure into the output buffer and
895 * advance pointers. */
896 error = uiomove((caddr_t)&d, reclen, uio);
899 de = RB_NEXT(tmpfs_dirtree, node->tn_dir.tn_dirtree, de);
900 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
902 /* Update the offset and cache. */
904 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
905 node->tn_dir.tn_readdir_lastn = 0;
906 node->tn_dir.tn_readdir_lastp = NULL;
908 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
909 node->tn_dir.tn_readdir_lastp = de;
911 node->tn_status |= TMPFS_NODE_ACCESSED;
916 /* --------------------------------------------------------------------- */
919 * Resizes the aobj associated to the regular file pointed to by vp to
920 * the size newsize. 'vp' must point to a vnode that represents a regular
921 * file. 'newsize' must be positive.
923 * pass trivial as 1 when buf content will be overwritten, otherwise set 0
926 * Returns zero on success or an appropriate error code on failure.
929 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial)
932 vm_pindex_t newpages, oldpages;
933 struct tmpfs_mount *tmp;
934 struct tmpfs_node *node;
938 KKASSERT(vp->v_type == VREG);
939 KKASSERT(newsize >= 0);
942 node = VP_TO_TMPFS_NODE(vp);
943 tmp = VFS_TO_TMPFS(vp->v_mount);
945 /* Convert the old and new sizes to the number of pages needed to
946 * store them. It may happen that we do not need to do anything
947 * because the last allocated page can accommodate the change on
949 oldsize = node->tn_size;
950 oldpages = round_page64(oldsize) / PAGE_SIZE;
951 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages);
952 newpages = round_page64(newsize) / PAGE_SIZE;
954 if (newpages > oldpages &&
955 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) {
961 tmp->tm_pages_used += (newpages - oldpages);
964 TMPFS_NODE_LOCK(node);
965 node->tn_reg.tn_aobj_pages = newpages;
966 node->tn_size = newsize;
967 TMPFS_NODE_UNLOCK(node);
970 * When adjusting the vnode filesize and its VM object we must
971 * also adjust our backing VM object (aobj). The blocksize
972 * used must match the block sized we use for the buffer cache.
974 * The backing VM object contains no VM pages, only swap
977 if (newsize < oldsize) {
982 error = nvtruncbuf(vp, newsize, BSIZE, -1, 0);
983 aobj = node->tn_reg.tn_aobj;
986 nsize = vp->v_object->size;
989 swap_pager_freespace(aobj, nsize,
996 error = nvextendbuf(vp, oldsize, newsize, BSIZE, BSIZE,
998 aobj = node->tn_reg.tn_aobj;
1000 aobj->size = vp->v_object->size;
1007 /* --------------------------------------------------------------------- */
1010 * Change flags of the given vnode.
1011 * Caller should execute tmpfs_update on vp after a successful execution.
1012 * The vnode must be locked on entry and remain locked on exit.
1015 tmpfs_chflags(struct vnode *vp, int vaflags, struct ucred *cred)
1018 struct tmpfs_node *node;
1021 KKASSERT(vn_islocked(vp));
1023 node = VP_TO_TMPFS_NODE(vp);
1024 flags = node->tn_flags;
1026 /* Disallow this operation if the file system is mounted read-only. */
1027 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1029 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred);
1032 * Unprivileged processes are not permitted to unset system
1033 * flags, or modify flags if any system flags are set.
1035 * Silently enforce SF_NOCACHE on the root tmpfs vnode so
1036 * tmpfs data is not double-cached by swapcache.
1039 TMPFS_NODE_LOCK(node);
1040 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1041 if (vp->v_flag & VROOT)
1042 flags |= SF_NOCACHE;
1043 node->tn_flags = flags;
1045 if (node->tn_flags & (SF_NOUNLINK | SF_IMMUTABLE |
1047 (flags & UF_SETTABLE) != flags) {
1050 node->tn_flags &= SF_SETTABLE;
1051 node->tn_flags |= (flags & UF_SETTABLE);
1054 node->tn_status |= TMPFS_NODE_CHANGED;
1055 TMPFS_NODE_UNLOCK(node);
1058 KKASSERT(vn_islocked(vp));
1063 /* --------------------------------------------------------------------- */
1066 * Change access mode on the given vnode.
1067 * Caller should execute tmpfs_update on vp after a successful execution.
1068 * The vnode must be locked on entry and remain locked on exit.
1071 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred)
1073 struct tmpfs_node *node;
1077 KKASSERT(vn_islocked(vp));
1079 node = VP_TO_TMPFS_NODE(vp);
1081 /* Disallow this operation if the file system is mounted read-only. */
1082 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1085 /* Immutable or append-only files cannot be modified, either. */
1086 if (node->tn_flags & (IMMUTABLE | APPEND))
1089 cur_mode = node->tn_mode;
1090 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid,
1094 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) {
1095 TMPFS_NODE_LOCK(node);
1096 node->tn_mode &= ~ALLPERMS;
1097 node->tn_mode |= cur_mode & ALLPERMS;
1099 node->tn_status |= TMPFS_NODE_CHANGED;
1100 TMPFS_NODE_UNLOCK(node);
1103 KKASSERT(vn_islocked(vp));
1108 /* --------------------------------------------------------------------- */
1111 * Change ownership of the given vnode. At least one of uid or gid must
1112 * be different than VNOVAL. If one is set to that value, the attribute
1114 * Caller should execute tmpfs_update on vp after a successful execution.
1115 * The vnode must be locked on entry and remain locked on exit.
1118 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
1123 struct tmpfs_node *node;
1126 KKASSERT(vn_islocked(vp));
1127 node = VP_TO_TMPFS_NODE(vp);
1129 /* Disallow this operation if the file system is mounted read-only. */
1130 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1133 /* Immutable or append-only files cannot be modified, either. */
1134 if (node->tn_flags & (IMMUTABLE | APPEND))
1137 cur_uid = node->tn_uid;
1138 cur_gid = node->tn_gid;
1139 cur_mode = node->tn_mode;
1140 error = vop_helper_chown(vp, uid, gid, cred,
1141 &cur_uid, &cur_gid, &cur_mode);
1144 TMPFS_NODE_LOCK(node);
1145 if (cur_uid != node->tn_uid ||
1146 cur_gid != node->tn_gid ||
1147 cur_mode != node->tn_mode) {
1148 node->tn_uid = cur_uid;
1149 node->tn_gid = cur_gid;
1150 node->tn_mode = cur_mode;
1151 node->tn_status |= TMPFS_NODE_CHANGED;
1153 TMPFS_NODE_UNLOCK(node);
1159 /* --------------------------------------------------------------------- */
1162 * Change size of the given vnode.
1163 * Caller should execute tmpfs_update on vp after a successful execution.
1164 * The vnode must be locked on entry and remain locked on exit.
1167 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred)
1170 struct tmpfs_node *node;
1172 KKASSERT(vn_islocked(vp));
1174 node = VP_TO_TMPFS_NODE(vp);
1176 /* Decide whether this is a valid operation based on the file type. */
1178 switch (vp->v_type) {
1183 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1192 /* Allow modifications of special files even if in the file
1193 * system is mounted read-only (we are not modifying the
1194 * files themselves, but the objects they represent). */
1198 /* Anything else is unsupported. */
1202 /* Immutable or append-only files cannot be modified, either. */
1203 if (node->tn_flags & (IMMUTABLE | APPEND))
1206 error = tmpfs_truncate(vp, size);
1207 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1208 * for us, as will update tn_status; no need to do that here. */
1210 KKASSERT(vn_islocked(vp));
1215 /* --------------------------------------------------------------------- */
1218 * Change access and modification times of the given vnode.
1219 * Caller should execute tmpfs_update on vp after a successful execution.
1220 * The vnode must be locked on entry and remain locked on exit.
1223 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1224 int vaflags, struct ucred *cred)
1226 struct tmpfs_node *node;
1228 KKASSERT(vn_islocked(vp));
1230 node = VP_TO_TMPFS_NODE(vp);
1232 /* Disallow this operation if the file system is mounted read-only. */
1233 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1236 /* Immutable or append-only files cannot be modified, either. */
1237 if (node->tn_flags & (IMMUTABLE | APPEND))
1240 TMPFS_NODE_LOCK(node);
1241 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1242 node->tn_status |= TMPFS_NODE_ACCESSED;
1244 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1245 node->tn_status |= TMPFS_NODE_MODIFIED;
1247 TMPFS_NODE_UNLOCK(node);
1249 tmpfs_itimes(vp, atime, mtime);
1251 KKASSERT(vn_islocked(vp));
1256 /* --------------------------------------------------------------------- */
1257 /* Sync timestamps */
1259 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1260 const struct timespec *mod)
1262 struct tmpfs_node *node;
1263 struct timespec now;
1265 node = VP_TO_TMPFS_NODE(vp);
1267 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1268 TMPFS_NODE_CHANGED)) == 0)
1271 vfs_timestamp(&now);
1273 TMPFS_NODE_LOCK(node);
1274 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1277 node->tn_atime = acc->tv_sec;
1278 node->tn_atimensec = acc->tv_nsec;
1280 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1283 node->tn_mtime = mod->tv_sec;
1284 node->tn_mtimensec = mod->tv_nsec;
1286 if (node->tn_status & TMPFS_NODE_CHANGED) {
1287 node->tn_ctime = now.tv_sec;
1288 node->tn_ctimensec = now.tv_nsec;
1291 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1292 TMPFS_NODE_UNLOCK(node);
1295 /* --------------------------------------------------------------------- */
1298 tmpfs_update(struct vnode *vp)
1301 tmpfs_itimes(vp, NULL, NULL);
1304 /* --------------------------------------------------------------------- */
1307 tmpfs_truncate(struct vnode *vp, off_t length)
1310 struct tmpfs_node *node;
1312 node = VP_TO_TMPFS_NODE(vp);
1319 if (node->tn_size == length) {
1324 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1328 error = tmpfs_reg_resize(vp, length, 1);
1331 TMPFS_NODE_LOCK(node);
1332 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1333 TMPFS_NODE_UNLOCK(node);
1342 /* --------------------------------------------------------------------- */
1345 tmpfs_fetch_ino(struct tmpfs_mount *tmp)
1349 ret = tmp->tm_ino++;
1355 tmpfs_dirtree_compare(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1357 if (a->td_namelen > b->td_namelen)
1359 else if (a->td_namelen < b->td_namelen)
1362 return strncmp(a->td_name, b->td_name, a->td_namelen);