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>
42 #include <sys/systm.h>
43 #include <sys/vnode.h>
44 #include <sys/vmmeter.h>
47 #include <vm/vm_object.h>
48 #include <vm/vm_page.h>
49 #include <vm/vm_pager.h>
50 #include <vm/vm_extern.h>
51 #include <vm/vm_pageout.h>
52 #include <vm/vm_page2.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);
142 nnode->tn_dir.tn_parent = NULL;
152 nnode->tn_size = strlen(target);
153 nnode->tn_link = kmalloc(nnode->tn_size + 1, tmp->tm_name_zone,
154 M_WAITOK | M_NULLOK);
155 if (nnode->tn_link == NULL) {
156 objcache_put(tmp->tm_node_pool, nnode);
159 bcopy(target, nnode->tn_link, nnode->tn_size);
160 nnode->tn_link[nnode->tn_size] = '\0';
164 nnode->tn_reg.tn_aobj = swap_pager_alloc(NULL, 0,
166 nnode->tn_reg.tn_aobj_pages = 0;
168 vm_object_set_flag(nnode->tn_reg.tn_aobj, OBJ_NOPAGEIN);
172 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
175 TMPFS_NODE_LOCK(nnode);
177 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
178 tmp->tm_nodes_inuse++;
180 TMPFS_NODE_UNLOCK(nnode);
186 /* --------------------------------------------------------------------- */
189 * Destroys the node pointed to by node from the file system 'tmp'.
190 * If the node does not belong to the given mount point, the results are
193 * If the node references a directory; no entries are allowed because
194 * their removal could need a recursive algorithm, something forbidden in
195 * kernel space. Furthermore, there is not need to provide such
196 * functionality (recursive removal) because the only primitives offered
197 * to the user are the removal of empty directories and the deletion of
200 * Note that nodes are not really deleted; in fact, when a node has been
201 * allocated, it cannot be deleted during the whole life of the file
202 * system. Instead, they are moved to the available list and remain there
205 * A caller must have TMPFS_NODE_LOCK(node) and this function unlocks it.
208 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
210 vm_pindex_t pages = 0;
213 TMPFS_ASSERT_ELOCKED(node);
214 KKASSERT(node->tn_vnode == NULL);
218 LIST_REMOVE(node, tn_entries);
219 tmp->tm_nodes_inuse--;
221 TMPFS_NODE_UNLOCK(node); /* Caller has this lock */
223 switch (node->tn_type) {
225 /* Do not do anything. VNON is provided to let the
226 * allocation routine clean itself easily by avoiding
227 * duplicating code in it. */
236 * The parent link can be NULL if this is the root
237 * node or if it is a directory node that was rmdir'd.
239 * XXX what if node is a directory which still contains
240 * directory entries (e.g. due to a forced umount) ?
243 KKASSERT(node->tn_dir.tn_parent == NULL);
246 * If the root node is being destroyed don't leave a
247 * dangling pointer in tmpfs_mount.
249 if (node == tmp->tm_root)
258 kfree(node->tn_link, tmp->tm_name_zone);
259 node->tn_link = NULL;
264 if (node->tn_reg.tn_aobj != NULL)
265 vm_object_deallocate(node->tn_reg.tn_aobj);
266 node->tn_reg.tn_aobj = NULL;
267 pages = node->tn_reg.tn_aobj_pages;
271 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
275 * Clean up fields for the next allocation. The objcache only ctors
278 tmpfs_node_ctor(node, NULL, 0);
279 objcache_put(tmp->tm_node_pool, node);
280 /* node is now invalid */
283 atomic_add_long(&tmp->tm_pages_used, -(long)pages);
286 /* --------------------------------------------------------------------- */
289 * Allocates a new directory entry for the node node with a name of name.
290 * The new directory entry is returned in *de.
292 * The link count of node is increased by one to reflect the new object
295 * Returns zero on success or an appropriate error code on failure.
298 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
299 const char *name, uint16_t len, struct tmpfs_dirent **de)
301 struct tmpfs_dirent *nde;
303 nde = objcache_get(tmp->tm_dirent_pool, M_WAITOK);
304 nde->td_name = kmalloc(len + 1, tmp->tm_name_zone, M_WAITOK | M_NULLOK);
305 if (nde->td_name == NULL) {
306 objcache_put(tmp->tm_dirent_pool, nde);
310 nde->td_namelen = len;
311 bcopy(name, nde->td_name, len);
312 nde->td_name[len] = '\0';
316 atomic_add_int(&node->tn_links, 1);
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 KKASSERT(node->tn_links > 0);
342 atomic_add_int(&node->tn_links, -1);
344 kfree(de->td_name, tmp->tm_name_zone);
348 objcache_put(tmp->tm_dirent_pool, de);
351 /* --------------------------------------------------------------------- */
354 * Allocates a new vnode for the node node or returns a new reference to
355 * an existing one if the node had already a vnode referencing it. The
356 * resulting locked vnode is returned in *vpp.
358 * Returns zero on success or an appropriate error code on failure.
360 * The caller must ensure that node cannot go away (usually by holding
361 * the related directory entry).
363 * If dnode is non-NULL this routine avoids deadlocking against it but
364 * can return EAGAIN. Caller must try again. The dnode lock will cycle
365 * in this case, it remains locked on return in all cases. dnode must
369 tmpfs_alloc_vp(struct mount *mp,
370 struct tmpfs_node *dnode, struct tmpfs_node *node, int lkflag,
378 if (node->tn_vnode == NULL) {
379 error = getnewvnode(VT_TMPFS, mp, &vp,
380 VLKTIMEOUT, LK_CANRECURSE);
386 * Interlocked extraction from node. This can race many things.
387 * We have to get a soft reference on the vnode while we hold
388 * the node locked, then acquire it properly and check for races.
390 TMPFS_NODE_LOCK(node);
391 if (node->tn_vnode) {
398 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
400 TMPFS_NODE_UNLOCK(node);
404 * Special-case handling to avoid deadlocking against
405 * dnode. This case has been validated and occurs
406 * every so often during synth builds.
408 if (vget(vp, (lkflag & ~LK_RETRY) |
410 LK_EXCLUSIVE) != 0) {
411 TMPFS_NODE_UNLOCK(dnode);
412 if (vget(vp, (lkflag & ~LK_RETRY) |
414 LK_EXCLUSIVE) == 0) {
418 TMPFS_NODE_LOCK_SH(dnode);
426 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) {
431 if (node->tn_vnode != vp) {
441 * We need to assign node->tn_vnode. If vp is NULL, loop up to
442 * allocate the vp. This can happen due to SMP races.
445 TMPFS_NODE_UNLOCK(node);
450 * This should never happen.
452 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) {
453 TMPFS_NODE_UNLOCK(node);
460 KKASSERT(node->tn_vnode == NULL);
461 KKASSERT(vp != NULL);
463 vp->v_type = node->tn_type;
465 /* Type-specific initialization. */
466 switch (node->tn_type) {
475 * VMIO is mandatory. Tmpfs also supports KVABIO
476 * for its tmpfs_strategy().
478 vsetflags(vp, VKVABIO);
479 vinitvmio(vp, node->tn_size, node->tn_blksize, -1);
484 vp->v_ops = &mp->mnt_vn_fifo_ops;
490 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
494 TMPFS_NODE_UNLOCK(node);
499 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
504 /* --------------------------------------------------------------------- */
507 * Allocates a new file of type 'type' and adds it to the parent directory
508 * 'dvp'; this addition is done using the component name given in 'cnp'.
509 * The ownership of the new file is automatically assigned based on the
510 * credentials of the caller (through 'cnp'), the group is set based on
511 * the parent directory and the mode is determined from the 'vap' argument.
512 * If successful, *vpp holds a vnode to the newly created file and zero
513 * is returned. Otherwise *vpp is NULL and the function returns an
514 * appropriate error code.
517 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
518 struct namecache *ncp, struct ucred *cred, char *target)
521 struct tmpfs_dirent *de;
522 struct tmpfs_mount *tmp;
523 struct tmpfs_node *dnode;
524 struct tmpfs_node *node;
526 tmp = VFS_TO_TMPFS(dvp->v_mount);
527 dnode = VP_TO_TMPFS_DIR(dvp);
530 TMPFS_NODE_LOCK(dnode);
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) {
539 TMPFS_NODE_UNLOCK(dnode);
544 * Make sure the link count does not overflow.
546 if (vap->va_type == VDIR && dnode->tn_links >= LINK_MAX) {
547 TMPFS_NODE_UNLOCK(dnode);
551 /* Allocate a node that represents the new file. */
552 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid,
553 dnode->tn_gid, vap->va_mode, target,
554 vap->va_rmajor, vap->va_rminor, &node);
556 TMPFS_NODE_UNLOCK(dnode);
559 TMPFS_NODE_LOCK(node);
561 /* Allocate a directory entry that points to the new file. */
562 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de);
564 TMPFS_NODE_UNLOCK(dnode);
565 tmpfs_free_node(tmp, node);
570 /* Allocate a vnode for the new file. */
571 error = tmpfs_alloc_vp(dvp->v_mount, NULL, node, LK_EXCLUSIVE, vpp);
573 TMPFS_NODE_UNLOCK(dnode);
574 tmpfs_free_dirent(tmp, de);
575 tmpfs_free_node(tmp, node);
581 * Now that all required items are allocated, we can proceed to
582 * insert the new node into the directory, an operation that
585 tmpfs_dir_attach_locked(dnode, de);
586 TMPFS_NODE_UNLOCK(dnode);
587 TMPFS_NODE_UNLOCK(node);
592 /* --------------------------------------------------------------------- */
595 * Attaches the directory entry de to the directory represented by dnode.
596 * Note that this does not change the link count of the node pointed by
597 * the directory entry, as this is done by tmpfs_alloc_dirent.
599 * dnode must be locked.
602 tmpfs_dir_attach_locked(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
604 struct tmpfs_node *node = de->td_node;
605 struct tmpfs_dirent *de2;
607 if (node && node->tn_type == VDIR) {
608 TMPFS_NODE_LOCK(node);
609 atomic_add_int(&node->tn_links, 1);
610 node->tn_status |= TMPFS_NODE_CHANGED;
611 node->tn_dir.tn_parent = dnode;
612 atomic_add_int(&dnode->tn_links, 1);
613 TMPFS_NODE_UNLOCK(node);
615 de2 = RB_INSERT(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de);
617 ("tmpfs_dir_attach_lockedA: duplicate insertion of %p, has %p\n",
619 de2 = RB_INSERT(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de);
621 ("tmpfs_dir_attach_lockedB: duplicate insertion of %p, has %p\n",
623 dnode->tn_size += sizeof(struct tmpfs_dirent);
624 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
628 /* --------------------------------------------------------------------- */
631 * Detaches the directory entry de from the directory represented by dnode.
632 * Note that this does not change the link count of the node pointed by
633 * the directory entry, as this is done by tmpfs_free_dirent.
635 * dnode must be locked.
638 tmpfs_dir_detach_locked(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
640 struct tmpfs_node *node = de->td_node;
642 RB_REMOVE(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de);
643 RB_REMOVE(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de);
644 dnode->tn_size -= sizeof(struct tmpfs_dirent);
645 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
649 * Clean out the tn_parent pointer immediately when removing a
652 * Removal of the parent linkage also cleans out the extra tn_links
653 * count we had on both node and dnode.
655 * node can be NULL (typ during a forced umount), in which case
656 * the mount code is dealing with the linkages from a linked list
659 if (node && node->tn_type == VDIR && node->tn_dir.tn_parent) {
660 TMPFS_NODE_LOCK(node);
661 KKASSERT(node->tn_dir.tn_parent == dnode);
662 atomic_add_int(&dnode->tn_links, -1);
663 atomic_add_int(&node->tn_links, -1);
664 node->tn_dir.tn_parent = NULL;
665 TMPFS_NODE_UNLOCK(node);
669 /* --------------------------------------------------------------------- */
672 * Looks for a directory entry in the directory represented by node.
673 * 'ncp' describes the name of the entry to look for. Note that the .
674 * and .. components are not allowed as they do not physically exist
675 * within directories.
677 * Returns a pointer to the entry when found, otherwise NULL.
679 * Caller must hold the node locked (shared ok)
681 struct tmpfs_dirent *
682 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
683 struct namecache *ncp)
685 struct tmpfs_dirent *de;
686 int len = ncp->nc_nlen;
687 struct tmpfs_dirent wanted;
689 wanted.td_namelen = len;
690 wanted.td_name = ncp->nc_name;
692 TMPFS_VALIDATE_DIR(node);
694 de = RB_FIND(tmpfs_dirtree, &node->tn_dir.tn_dirtree, &wanted);
696 KASSERT((f == NULL || de == NULL || f == de->td_node),
697 ("tmpfs_dir_lookup: Incorrect node %p %p %p",
698 f, de, (de ? de->td_node : NULL)));
703 /* --------------------------------------------------------------------- */
706 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
707 * directory and returns it in the uio space. The function returns 0
708 * on success, -1 if there was not enough space in the uio structure to
709 * hold the directory entry or an appropriate error code if another
713 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
717 TMPFS_VALIDATE_DIR(node);
718 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
720 if (vop_write_dirent(&error, uio, node->tn_id, DT_DIR, 1, "."))
723 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
727 /* --------------------------------------------------------------------- */
730 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
731 * directory and returns it in the uio space. The function returns 0
732 * on success, -1 if there was not enough space in the uio structure to
733 * hold the directory entry or an appropriate error code if another
737 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
743 TMPFS_VALIDATE_DIR(node);
744 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
746 if (node->tn_dir.tn_parent) {
747 TMPFS_NODE_LOCK(node);
748 if (node->tn_dir.tn_parent)
749 d_ino = node->tn_dir.tn_parent->tn_id;
751 d_ino = tmp->tm_root->tn_id;
752 TMPFS_NODE_UNLOCK(node);
754 d_ino = tmp->tm_root->tn_id;
757 if (vop_write_dirent(&error, uio, d_ino, DT_DIR, 2, ".."))
760 struct tmpfs_dirent *de;
761 de = RB_MIN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree);
763 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
765 uio->uio_offset = tmpfs_dircookie(de);
770 /* --------------------------------------------------------------------- */
773 * Lookup a directory entry by its associated cookie.
775 * Must be called with the directory node locked (shared ok)
777 struct lubycookie_info {
779 struct tmpfs_dirent *de;
783 lubycookie_cmp(struct tmpfs_dirent *de, void *arg)
785 struct lubycookie_info *info = arg;
786 off_t cookie = tmpfs_dircookie(de);
788 if (cookie < info->cookie)
790 if (cookie > info->cookie)
796 lubycookie_callback(struct tmpfs_dirent *de, void *arg)
798 struct lubycookie_info *info = arg;
800 if (tmpfs_dircookie(de) == info->cookie) {
807 struct tmpfs_dirent *
808 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
810 struct lubycookie_info info;
812 info.cookie = cookie;
814 RB_SCAN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree,
815 lubycookie_cmp, lubycookie_callback, &info);
819 /* --------------------------------------------------------------------- */
822 * Helper function for tmpfs_readdir. Returns as much directory entries
823 * as can fit in the uio space. The read starts at uio->uio_offset.
824 * The function returns 0 on success, -1 if there was not enough space
825 * in the uio structure to hold the directory entry or an appropriate
826 * error code if another error happens.
828 * Caller must hold the node locked (shared ok)
831 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
835 struct tmpfs_dirent *de;
837 TMPFS_VALIDATE_DIR(node);
840 * Locate the first directory entry we have to return. We have cached
841 * the last readdir in the node, so use those values if appropriate.
842 * Otherwise do a linear scan to find the requested entry.
844 startcookie = uio->uio_offset;
845 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT);
846 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
848 if (startcookie == TMPFS_DIRCOOKIE_EOF)
851 de = tmpfs_dir_lookupbycookie(node, startcookie);
856 * Read as much entries as possible; i.e., until we reach the end of
857 * the directory or we exhaust uio space.
863 /* Create a dirent structure representing the current
864 * tmpfs_node and fill it. */
865 d_ino = de->td_node->tn_id;
866 switch (de->td_node->tn_type) {
896 panic("tmpfs_dir_getdents: type %p %d",
897 de->td_node, (int)de->td_node->tn_type);
899 KKASSERT(de->td_namelen < 256); /* 255 + 1 */
901 if (vop_write_dirent(&error, uio, d_ino, d_type,
902 de->td_namelen, de->td_name)) {
908 de = RB_NEXT(tmpfs_dirtree_cookie,
909 node->tn_dir.tn_cookietree, de);
910 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
912 /* Update the offset and cache. */
914 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
916 uio->uio_offset = tmpfs_dircookie(de);
922 /* --------------------------------------------------------------------- */
925 * Resizes the aobj associated to the regular file pointed to by vp to
926 * the size newsize. 'vp' must point to a vnode that represents a regular
927 * file. 'newsize' must be positive.
929 * pass NVEXTF_TRIVIAL when buf content will be overwritten, otherwise set 0
932 * Returns zero on success or an appropriate error code on failure.
934 * Caller must hold the node exclusively locked.
937 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial)
940 vm_pindex_t newpages, oldpages;
941 struct tmpfs_mount *tmp;
942 struct tmpfs_node *node;
947 KKASSERT(vp->v_type == VREG);
948 KKASSERT(newsize >= 0);
951 node = VP_TO_TMPFS_NODE(vp);
952 tmp = VFS_TO_TMPFS(vp->v_mount);
955 * Convert the old and new sizes to the number of pages needed to
956 * store them. It may happen that we do not need to do anything
957 * because the last allocated page can accommodate the change on
960 oldsize = node->tn_size;
961 oldpages = round_page64(oldsize) / PAGE_SIZE;
962 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages);
963 newpages = round_page64(newsize) / PAGE_SIZE;
965 if (newpages > oldpages &&
966 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) {
970 node->tn_reg.tn_aobj_pages = newpages;
971 node->tn_size = newsize;
973 if (newpages != oldpages)
974 atomic_add_long(&tmp->tm_pages_used, (newpages - oldpages));
977 * nvextflags to pass along for bdwrite() vs buwrite()
979 if (vm_pages_needed || vm_paging_needed(0) ||
980 tmpfs_bufcache_mode >= 2) {
983 nvextflags = NVEXTF_BUWRITE;
988 * When adjusting the vnode filesize and its VM object we must
989 * also adjust our backing VM object (aobj). The blocksize
990 * used must match the block sized we use for the buffer cache.
992 * The backing VM object may contain VM pages as well as swap
993 * assignments if we previously renamed main object pages into
994 * it during deactivation.
996 * To make things easier tmpfs uses a blksize in multiples of
997 * PAGE_SIZE, and will only increase the blksize as a small file
998 * increases in size. Once a file has exceeded TMPFS_BLKSIZE (16KB),
999 * the blksize is maxed out. Truncating the file does not reduce
1002 if (newsize < oldsize) {
1007 error = nvtruncbuf(vp, newsize, node->tn_blksize,
1009 aobj = node->tn_reg.tn_aobj;
1012 nsize = vp->v_object->size;
1013 if (nsize < osize) {
1015 swap_pager_freespace(aobj, nsize,
1017 vm_object_page_remove(aobj, nsize, osize,
1026 * The first (and only the first) buffer in the file is resized
1027 * in multiples of PAGE_SIZE, up to TMPFS_BLKSIZE.
1029 nblksize = node->tn_blksize;
1030 while (nblksize < TMPFS_BLKSIZE &&
1031 nblksize < newsize) {
1032 nblksize += PAGE_SIZE;
1036 nvextflags |= NVEXTF_TRIVIAL;
1038 error = nvextendbuf(vp, oldsize, newsize,
1039 node->tn_blksize, nblksize,
1040 -1, -1, nvextflags);
1041 node->tn_blksize = nblksize;
1042 aobj = node->tn_reg.tn_aobj;
1044 aobj->size = vp->v_object->size;
1051 /* --------------------------------------------------------------------- */
1054 * Change flags of the given vnode.
1055 * Caller should execute tmpfs_update on vp after a successful execution.
1056 * The vnode must be locked on entry and remain locked on exit.
1059 tmpfs_chflags(struct vnode *vp, u_long vaflags, struct ucred *cred)
1062 struct tmpfs_node *node;
1065 KKASSERT(vn_islocked(vp));
1067 node = VP_TO_TMPFS_NODE(vp);
1068 flags = node->tn_flags;
1070 /* Disallow this operation if the file system is mounted read-only. */
1071 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1073 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred);
1075 /* Actually change the flags on the node itself */
1077 TMPFS_NODE_LOCK(node);
1078 node->tn_flags = flags;
1079 node->tn_status |= TMPFS_NODE_CHANGED;
1080 TMPFS_NODE_UNLOCK(node);
1083 KKASSERT(vn_islocked(vp));
1088 /* --------------------------------------------------------------------- */
1091 * Change access mode on the given vnode.
1092 * Caller should execute tmpfs_update on vp after a successful execution.
1093 * The vnode must be locked on entry and remain locked on exit.
1096 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred)
1098 struct tmpfs_node *node;
1102 KKASSERT(vn_islocked(vp));
1104 node = VP_TO_TMPFS_NODE(vp);
1106 /* Disallow this operation if the file system is mounted read-only. */
1107 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1110 /* Immutable or append-only files cannot be modified, either. */
1111 if (node->tn_flags & (IMMUTABLE | APPEND))
1114 cur_mode = node->tn_mode;
1115 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid,
1119 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) {
1120 TMPFS_NODE_LOCK(node);
1121 node->tn_mode &= ~ALLPERMS;
1122 node->tn_mode |= cur_mode & ALLPERMS;
1124 node->tn_status |= TMPFS_NODE_CHANGED;
1125 TMPFS_NODE_UNLOCK(node);
1128 KKASSERT(vn_islocked(vp));
1133 /* --------------------------------------------------------------------- */
1136 * Change ownership of the given vnode. At least one of uid or gid must
1137 * be different than VNOVAL. If one is set to that value, the attribute
1139 * Caller should execute tmpfs_update on vp after a successful execution.
1140 * The vnode must be locked on entry and remain locked on exit.
1143 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
1148 struct tmpfs_node *node;
1151 KKASSERT(vn_islocked(vp));
1152 node = VP_TO_TMPFS_NODE(vp);
1154 /* Disallow this operation if the file system is mounted read-only. */
1155 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1158 /* Immutable or append-only files cannot be modified, either. */
1159 if (node->tn_flags & (IMMUTABLE | APPEND))
1162 cur_uid = node->tn_uid;
1163 cur_gid = node->tn_gid;
1164 cur_mode = node->tn_mode;
1165 error = vop_helper_chown(vp, uid, gid, cred,
1166 &cur_uid, &cur_gid, &cur_mode);
1169 TMPFS_NODE_LOCK(node);
1170 if (cur_uid != node->tn_uid ||
1171 cur_gid != node->tn_gid ||
1172 cur_mode != node->tn_mode) {
1173 node->tn_uid = cur_uid;
1174 node->tn_gid = cur_gid;
1175 node->tn_mode = cur_mode;
1176 node->tn_status |= TMPFS_NODE_CHANGED;
1178 TMPFS_NODE_UNLOCK(node);
1184 /* --------------------------------------------------------------------- */
1187 * Change size of the given vnode.
1188 * Caller should execute tmpfs_update on vp after a successful execution.
1189 * The vnode must be locked on entry and remain locked on exit.
1192 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred)
1195 struct tmpfs_node *node;
1197 KKASSERT(vn_islocked(vp));
1199 node = VP_TO_TMPFS_NODE(vp);
1201 /* Decide whether this is a valid operation based on the file type. */
1203 switch (vp->v_type) {
1208 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1217 /* Allow modifications of special files even if in the file
1218 * system is mounted read-only (we are not modifying the
1219 * files themselves, but the objects they represent). */
1223 /* Anything else is unsupported. */
1227 /* Immutable or append-only files cannot be modified, either. */
1228 if (node->tn_flags & (IMMUTABLE | APPEND))
1231 error = tmpfs_truncate(vp, size);
1232 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1233 * for us, as will update tn_status; no need to do that here. */
1235 KKASSERT(vn_islocked(vp));
1240 /* --------------------------------------------------------------------- */
1243 * Change access and modification times of the given vnode.
1244 * Caller should execute tmpfs_update on vp after a successful execution.
1245 * The vnode must be locked on entry and remain locked on exit.
1248 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1249 int vaflags, struct ucred *cred)
1251 struct tmpfs_node *node;
1253 KKASSERT(vn_islocked(vp));
1255 node = VP_TO_TMPFS_NODE(vp);
1257 /* Disallow this operation if the file system is mounted read-only. */
1258 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1261 /* Immutable or append-only files cannot be modified, either. */
1262 if (node->tn_flags & (IMMUTABLE | APPEND))
1265 TMPFS_NODE_LOCK(node);
1266 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1267 node->tn_status |= TMPFS_NODE_ACCESSED;
1269 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL) {
1270 node->tn_status |= TMPFS_NODE_MODIFIED;
1271 vclrflags(vp, VLASTWRITETS);
1274 TMPFS_NODE_UNLOCK(node);
1276 tmpfs_itimes(vp, atime, mtime);
1278 KKASSERT(vn_islocked(vp));
1283 /* --------------------------------------------------------------------- */
1284 /* Sync timestamps */
1286 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1287 const struct timespec *mod)
1289 struct tmpfs_node *node;
1290 struct timespec now;
1292 node = VP_TO_TMPFS_NODE(vp);
1294 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1295 TMPFS_NODE_CHANGED)) == 0) {
1299 vfs_timestamp(&now);
1301 TMPFS_NODE_LOCK(node);
1302 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1305 node->tn_atime = acc->tv_sec;
1306 node->tn_atimensec = acc->tv_nsec;
1308 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1311 node->tn_mtime = mod->tv_sec;
1312 node->tn_mtimensec = mod->tv_nsec;
1314 if (node->tn_status & TMPFS_NODE_CHANGED) {
1315 node->tn_ctime = now.tv_sec;
1316 node->tn_ctimensec = now.tv_nsec;
1319 node->tn_status &= ~(TMPFS_NODE_ACCESSED |
1320 TMPFS_NODE_MODIFIED |
1321 TMPFS_NODE_CHANGED);
1322 TMPFS_NODE_UNLOCK(node);
1325 /* --------------------------------------------------------------------- */
1328 tmpfs_update(struct vnode *vp)
1330 tmpfs_itimes(vp, NULL, NULL);
1333 /* --------------------------------------------------------------------- */
1336 * Caller must hold an exclusive node lock.
1339 tmpfs_truncate(struct vnode *vp, off_t length)
1342 struct tmpfs_node *node;
1344 node = VP_TO_TMPFS_NODE(vp);
1351 if (node->tn_size == length) {
1356 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1360 error = tmpfs_reg_resize(vp, length, 1);
1363 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1371 /* --------------------------------------------------------------------- */
1374 tmpfs_fetch_ino(struct tmpfs_mount *tmp)
1378 ret = atomic_fetchadd_64(&tmp->tm_ino, 1);
1384 tmpfs_dirtree_compare(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1386 if (a->td_namelen > b->td_namelen)
1388 else if (a->td_namelen < b->td_namelen)
1391 return strncmp(a->td_name, b->td_name, a->td_namelen);
1395 tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1405 * Lock for rename. The namecache entries for the related terminal files
1406 * are already locked but the directories are not. A directory lock order
1407 * reversal is possible so use a deterministic order.
1409 * Generally order path parent-to-child or using a simple pointer comparison.
1410 * Probably not perfect but it should catch most of the cases.
1412 * Underlying files must be locked after the related directory.
1415 tmpfs_lock4(struct tmpfs_node *node1, struct tmpfs_node *node2,
1416 struct tmpfs_node *node3, struct tmpfs_node *node4)
1418 if (node1->tn_dir.tn_parent != node2 &&
1419 (node1 < node2 || node2->tn_dir.tn_parent == node1)) {
1420 TMPFS_NODE_LOCK(node1); /* fdir */
1421 TMPFS_NODE_LOCK(node3); /* ffile */
1422 TMPFS_NODE_LOCK(node2); /* tdir */
1424 TMPFS_NODE_LOCK(node4); /* tfile */
1426 TMPFS_NODE_LOCK(node2); /* tdir */
1428 TMPFS_NODE_LOCK(node4); /* tfile */
1429 TMPFS_NODE_LOCK(node1); /* fdir */
1430 TMPFS_NODE_LOCK(node3); /* ffile */
1435 tmpfs_unlock4(struct tmpfs_node *node1, struct tmpfs_node *node2,
1436 struct tmpfs_node *node3, struct tmpfs_node *node4)
1439 TMPFS_NODE_UNLOCK(node4);
1440 TMPFS_NODE_UNLOCK(node2);
1441 TMPFS_NODE_UNLOCK(node3);
1442 TMPFS_NODE_UNLOCK(node1);