1 /* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */
4 * Copyright (c) 2005 The NetBSD Foundation, Inc.
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
34 * Efficient memory file system supporting functions.
37 #include <sys/kernel.h>
38 #include <sys/param.h>
39 #include <sys/namei.h>
43 #include <sys/systm.h>
44 #include <sys/vnode.h>
45 #include <sys/vmmeter.h>
48 #include <vm/vm_object.h>
49 #include <vm/vm_page.h>
50 #include <vm/vm_pager.h>
51 #include <vm/vm_extern.h>
53 #include <vfs/tmpfs/tmpfs.h>
54 #include <vfs/tmpfs/tmpfs_vnops.h>
56 static ino_t tmpfs_fetch_ino(struct tmpfs_mount *);
58 static int tmpfs_dirtree_compare(struct tmpfs_dirent *a,
59 struct tmpfs_dirent *b);
60 RB_GENERATE(tmpfs_dirtree, tmpfs_dirent, rb_node, tmpfs_dirtree_compare);
62 static int tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a,
63 struct tmpfs_dirent *b);
64 RB_GENERATE(tmpfs_dirtree_cookie, tmpfs_dirent,
65 rb_cookienode, tmpfs_dirtree_compare_cookie);
68 /* --------------------------------------------------------------------- */
71 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
72 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
73 * using the credentials of the process 'p'.
75 * If the node type is set to 'VDIR', then the parent parameter must point
76 * to the parent directory of the node being created. It may only be NULL
77 * while allocating the root node.
79 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
80 * specifies the device the node represents.
82 * If the node type is set to 'VLNK', then the parameter target specifies
83 * the file name of the target file for the symbolic link that is being
86 * Note that new nodes are retrieved from the available list if it has
87 * items or, if it is empty, from the node pool as long as there is enough
88 * space to create them.
90 * Returns zero on success or an appropriate error code on failure.
93 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
94 uid_t uid, gid_t gid, mode_t mode,
95 char *target, int rmajor, int rminor,
96 struct tmpfs_node **node)
98 struct tmpfs_node *nnode;
102 KKASSERT(IFF(type == VLNK, target != NULL));
103 KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL));
106 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max) {
112 nnode = objcache_get(tmp->tm_node_pool, M_WAITOK | M_NULLOK);
116 /* Generic initialization. */
117 nnode->tn_type = type;
119 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime
121 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec
125 nnode->tn_mode = mode;
126 nnode->tn_id = tmpfs_fetch_ino(tmp);
127 nnode->tn_advlock.init_done = 0;
128 KKASSERT(nnode->tn_links == 0);
130 /* Type-specific initialization. */
131 switch (nnode->tn_type) {
134 rdev = makeudev(rmajor, rminor);
135 if (rdev == NOUDEV) {
136 objcache_put(tmp->tm_node_pool, nnode);
139 nnode->tn_rdev = rdev;
143 RB_INIT(&nnode->tn_dir.tn_dirtree);
144 RB_INIT(&nnode->tn_dir.tn_cookietree);
154 nnode->tn_size = strlen(target);
155 nnode->tn_link = kmalloc(nnode->tn_size + 1, tmp->tm_name_zone,
156 M_WAITOK | M_NULLOK);
157 if (nnode->tn_link == NULL) {
158 objcache_put(tmp->tm_node_pool, nnode);
161 bcopy(target, nnode->tn_link, nnode->tn_size);
162 nnode->tn_link[nnode->tn_size] = '\0';
166 nnode->tn_reg.tn_aobj =
167 swap_pager_alloc(NULL, 0, VM_PROT_DEFAULT, 0);
168 nnode->tn_reg.tn_aobj_pages = 0;
173 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
176 TMPFS_NODE_LOCK(nnode);
178 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
179 tmp->tm_nodes_inuse++;
181 TMPFS_NODE_UNLOCK(nnode);
187 /* --------------------------------------------------------------------- */
190 * Destroys the node pointed to by node from the file system 'tmp'.
191 * If the node does not belong to the given mount point, the results are
194 * If the node references a directory; no entries are allowed because
195 * their removal could need a recursive algorithm, something forbidden in
196 * kernel space. Furthermore, there is not need to provide such
197 * functionality (recursive removal) because the only primitives offered
198 * to the user are the removal of empty directories and the deletion of
201 * Note that nodes are not really deleted; in fact, when a node has been
202 * allocated, it cannot be deleted during the whole life of the file
203 * system. Instead, they are moved to the available list and remain there
206 * A caller must have TMPFS_NODE_LOCK(node) and this function unlocks it.
209 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
211 vm_pindex_t pages = 0;
214 TMPFS_ASSERT_ELOCKED(node);
215 KKASSERT(node->tn_vnode == NULL);
216 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
220 LIST_REMOVE(node, tn_entries);
221 tmp->tm_nodes_inuse--;
223 TMPFS_NODE_UNLOCK(node); /* Caller has this lock */
225 switch (node->tn_type) {
227 /* Do not do anything. VNON is provided to let the
228 * allocation routine clean itself easily by avoiding
229 * duplicating code in it. */
238 * The parent link can be NULL if this is the root
239 * node or if it is a directory node that was rmdir'd.
241 * XXX what if node is a directory which still contains
242 * directory entries (e.g. due to a forced umount) ?
245 KKASSERT(node->tn_dir.tn_parent == NULL);
248 * If the root node is being destroyed don't leave a
249 * dangling pointer in tmpfs_mount.
251 if (node == tmp->tm_root)
260 kfree(node->tn_link, tmp->tm_name_zone);
261 node->tn_link = NULL;
266 if (node->tn_reg.tn_aobj != NULL)
267 vm_object_deallocate(node->tn_reg.tn_aobj);
268 node->tn_reg.tn_aobj = NULL;
269 pages = node->tn_reg.tn_aobj_pages;
273 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
277 * Clean up fields for the next allocation. The objcache only ctors
280 tmpfs_node_ctor(node, NULL, 0);
281 objcache_put(tmp->tm_node_pool, node);
282 /* node is now invalid */
285 tmp->tm_pages_used -= pages;
289 /* --------------------------------------------------------------------- */
292 * Allocates a new directory entry for the node node with a name of name.
293 * The new directory entry is returned in *de.
295 * The link count of node is increased by one to reflect the new object
298 * Returns zero on success or an appropriate error code on failure.
301 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
302 const char *name, uint16_t len, struct tmpfs_dirent **de)
304 struct tmpfs_dirent *nde;
306 nde = objcache_get(tmp->tm_dirent_pool, M_WAITOK);
307 nde->td_name = kmalloc(len + 1, tmp->tm_name_zone, M_WAITOK | M_NULLOK);
308 if (nde->td_name == NULL) {
309 objcache_put(tmp->tm_dirent_pool, nde);
313 nde->td_namelen = len;
314 bcopy(name, nde->td_name, len);
315 nde->td_name[len] = '\0';
319 TMPFS_NODE_LOCK(node);
321 TMPFS_NODE_UNLOCK(node);
328 /* --------------------------------------------------------------------- */
331 * Frees a directory entry. It is the caller's responsibility to destroy
332 * the node referenced by it if needed.
334 * The link count of node is decreased by one to reflect the removal of an
335 * object that referenced it. This only happens if 'node_exists' is true;
336 * otherwise the function will not access the node referred to by the
337 * directory entry, as it may already have been released from the outside.
340 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
342 struct tmpfs_node *node;
346 TMPFS_NODE_LOCK(node);
347 TMPFS_ASSERT_ELOCKED(node);
348 KKASSERT(node->tn_links > 0);
350 TMPFS_NODE_UNLOCK(node);
352 kfree(de->td_name, tmp->tm_name_zone);
356 objcache_put(tmp->tm_dirent_pool, de);
359 /* --------------------------------------------------------------------- */
362 * Allocates a new vnode for the node node or returns a new reference to
363 * an existing one if the node had already a vnode referencing it. The
364 * resulting locked vnode is returned in *vpp.
366 * Returns zero on success or an appropriate error code on failure.
369 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
377 * Interlocked extraction from node. This can race many things.
378 * We have to get a soft reference on the vnode while we hold
379 * the node locked, then acquire it properly and check for races.
381 TMPFS_NODE_LOCK(node);
382 if ((vp = node->tn_vnode) != NULL) {
383 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
385 TMPFS_NODE_UNLOCK(node);
387 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) {
391 if (node->tn_vnode != vp) {
402 * This should never happen.
404 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) {
405 TMPFS_NODE_UNLOCK(node);
411 * Interlock against other calls to tmpfs_alloc_vp() trying to
412 * allocate and assign a vp to node.
414 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
415 node->tn_vpstate |= TMPFS_VNODE_WANT;
416 error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH,
417 "tmpfs_alloc_vp", 0);
418 TMPFS_NODE_UNLOCK(node);
423 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
424 TMPFS_NODE_UNLOCK(node);
427 * Allocate a new vnode (may block). The ALLOCATING flag should
428 * prevent a race against someone else assigning node->tn_vnode.
430 error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE);
434 KKASSERT(node->tn_vnode == NULL);
435 KKASSERT(vp != NULL);
437 vp->v_type = node->tn_type;
439 /* Type-specific initialization. */
440 switch (node->tn_type) {
448 vinitvmio(vp, node->tn_size, TMPFS_BLKMASK, -1);
453 vp->v_ops = &mp->mnt_vn_fifo_ops;
459 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
464 TMPFS_NODE_LOCK(node);
466 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
467 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
470 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
471 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
472 TMPFS_NODE_UNLOCK(node);
473 wakeup(&node->tn_vpstate);
475 TMPFS_NODE_UNLOCK(node);
481 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
483 TMPFS_NODE_LOCK(node);
484 KKASSERT(*vpp == node->tn_vnode);
485 TMPFS_NODE_UNLOCK(node);
491 /* --------------------------------------------------------------------- */
494 * Destroys the association between the vnode vp and the node it
498 tmpfs_free_vp(struct vnode *vp)
500 struct tmpfs_node *node;
502 node = VP_TO_TMPFS_NODE(vp);
504 TMPFS_NODE_LOCK(node);
505 KKASSERT(lockcount(TMPFS_NODE_MTX(node)) > 0);
506 node->tn_vnode = NULL;
508 TMPFS_NODE_UNLOCK(node);
511 /* --------------------------------------------------------------------- */
514 * Allocates a new file of type 'type' and adds it to the parent directory
515 * 'dvp'; this addition is done using the component name given in 'cnp'.
516 * The ownership of the new file is automatically assigned based on the
517 * credentials of the caller (through 'cnp'), the group is set based on
518 * the parent directory and the mode is determined from the 'vap' argument.
519 * If successful, *vpp holds a vnode to the newly created file and zero
520 * is returned. Otherwise *vpp is NULL and the function returns an
521 * appropriate error code.
524 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
525 struct namecache *ncp, struct ucred *cred, char *target)
528 struct tmpfs_dirent *de;
529 struct tmpfs_mount *tmp;
530 struct tmpfs_node *dnode;
531 struct tmpfs_node *node;
533 tmp = VFS_TO_TMPFS(dvp->v_mount);
534 dnode = VP_TO_TMPFS_DIR(dvp);
538 * If the directory was removed but a process was CD'd into it,
539 * we do not allow any more file/dir creation within it. Otherwise
540 * we will lose track of it.
542 KKASSERT(dnode->tn_type == VDIR);
543 if (dnode != tmp->tm_root && dnode->tn_dir.tn_parent == NULL)
547 * Make sure the link count does not overflow.
549 if (vap->va_type == VDIR && dnode->tn_links >= LINK_MAX)
552 /* Allocate a node that represents the new file. */
553 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid,
554 dnode->tn_gid, vap->va_mode, target,
555 vap->va_rmajor, vap->va_rminor, &node);
558 TMPFS_NODE_LOCK(node);
560 /* Allocate a directory entry that points to the new file. */
561 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de);
563 tmpfs_free_node(tmp, node);
568 /* Allocate a vnode for the new file. */
569 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
571 tmpfs_free_dirent(tmp, de);
572 tmpfs_free_node(tmp, node);
578 * Now that all required items are allocated, we can proceed to
579 * insert the new node into the directory, an operation that
582 tmpfs_dir_attach(dnode, de);
583 TMPFS_NODE_UNLOCK(node);
588 /* --------------------------------------------------------------------- */
591 * Attaches the directory entry de to the directory represented by dnode.
592 * Note that this does not change the link count of the node pointed by
593 * the directory entry, as this is done by tmpfs_alloc_dirent.
596 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
598 struct tmpfs_node *node = de->td_node;
600 TMPFS_NODE_LOCK(dnode);
601 if (node && node->tn_type == VDIR) {
602 TMPFS_NODE_LOCK(node);
604 node->tn_status |= TMPFS_NODE_CHANGED;
605 node->tn_dir.tn_parent = dnode;
607 TMPFS_NODE_UNLOCK(node);
609 RB_INSERT(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de);
610 RB_INSERT(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de);
611 dnode->tn_size += sizeof(struct tmpfs_dirent);
612 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
614 TMPFS_NODE_UNLOCK(dnode);
617 /* --------------------------------------------------------------------- */
620 * Detaches the directory entry de from the directory represented by dnode.
621 * Note that this does not change the link count of the node pointed by
622 * the directory entry, as this is done by tmpfs_free_dirent.
625 tmpfs_dir_detach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
627 struct tmpfs_node *node = de->td_node;
629 TMPFS_NODE_LOCK(dnode);
630 RB_REMOVE(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de);
631 RB_REMOVE(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de);
632 dnode->tn_size -= sizeof(struct tmpfs_dirent);
633 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
635 TMPFS_NODE_UNLOCK(dnode);
638 * Clean out the tn_parent pointer immediately when removing a
641 * Removal of the parent linkage also cleans out the extra tn_links
642 * count we had on both node and dnode.
644 * node can be NULL (typ during a forced umount), in which case
645 * the mount code is dealing with the linkages from a linked list
648 if (node && node->tn_type == VDIR && node->tn_dir.tn_parent) {
649 TMPFS_NODE_LOCK(dnode);
650 TMPFS_NODE_LOCK(node);
651 KKASSERT(node->tn_dir.tn_parent == dnode);
654 node->tn_dir.tn_parent = NULL;
655 TMPFS_NODE_UNLOCK(node);
656 TMPFS_NODE_UNLOCK(dnode);
660 /* --------------------------------------------------------------------- */
663 * Looks for a directory entry in the directory represented by node.
664 * 'ncp' describes the name of the entry to look for. Note that the .
665 * and .. components are not allowed as they do not physically exist
666 * within directories.
668 * Returns a pointer to the entry when found, otherwise NULL.
670 * Caller must hold the node locked (shared ok)
672 struct tmpfs_dirent *
673 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
674 struct namecache *ncp)
676 struct tmpfs_dirent *de;
677 int len = ncp->nc_nlen;
678 struct tmpfs_dirent wanted;
680 wanted.td_namelen = len;
681 wanted.td_name = ncp->nc_name;
683 TMPFS_VALIDATE_DIR(node);
685 de = RB_FIND(tmpfs_dirtree, &node->tn_dir.tn_dirtree, &wanted);
687 KKASSERT(f == NULL || f == de->td_node);
692 /* --------------------------------------------------------------------- */
695 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
696 * directory and returns it in the uio space. The function returns 0
697 * on success, -1 if there was not enough space in the uio structure to
698 * hold the directory entry or an appropriate error code if another
702 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
708 TMPFS_VALIDATE_DIR(node);
709 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
711 dent.d_ino = node->tn_id;
712 dent.d_type = DT_DIR;
714 dent.d_name[0] = '.';
715 dent.d_name[1] = '\0';
716 dirsize = _DIRENT_DIRSIZ(&dent);
718 if (dirsize > uio->uio_resid)
721 error = uiomove((caddr_t)&dent, dirsize, uio);
723 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
728 /* --------------------------------------------------------------------- */
731 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
732 * directory and returns it in the uio space. The function returns 0
733 * on success, -1 if there was not enough space in the uio structure to
734 * hold the directory entry or an appropriate error code if another
738 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
745 TMPFS_VALIDATE_DIR(node);
746 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
748 if (node->tn_dir.tn_parent) {
749 TMPFS_NODE_LOCK(node);
750 if (node->tn_dir.tn_parent)
751 dent.d_ino = node->tn_dir.tn_parent->tn_id;
753 dent.d_ino = tmp->tm_root->tn_id;
754 TMPFS_NODE_UNLOCK(node);
756 dent.d_ino = tmp->tm_root->tn_id;
759 dent.d_type = DT_DIR;
761 dent.d_name[0] = '.';
762 dent.d_name[1] = '.';
763 dent.d_name[2] = '\0';
764 dirsize = _DIRENT_DIRSIZ(&dent);
766 if (dirsize > uio->uio_resid)
769 error = uiomove((caddr_t)&dent, dirsize, uio);
771 struct tmpfs_dirent *de;
773 de = RB_MIN(tmpfs_dirtree_cookie,
774 &node->tn_dir.tn_cookietree);
776 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
778 uio->uio_offset = tmpfs_dircookie(de);
784 /* --------------------------------------------------------------------- */
787 * Lookup a directory entry by its associated cookie.
789 * Must be called with the directory node locked (shared ok)
791 struct lubycookie_info {
793 struct tmpfs_dirent *de;
797 lubycookie_cmp(struct tmpfs_dirent *de, void *arg)
799 struct lubycookie_info *info = arg;
800 off_t cookie = tmpfs_dircookie(de);
802 if (cookie < info->cookie)
804 if (cookie > info->cookie)
810 lubycookie_callback(struct tmpfs_dirent *de, void *arg)
812 struct lubycookie_info *info = arg;
814 if (tmpfs_dircookie(de) == info->cookie) {
821 struct tmpfs_dirent *
822 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
824 struct lubycookie_info info;
826 info.cookie = cookie;
828 RB_SCAN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree,
829 lubycookie_cmp, lubycookie_callback, &info);
833 /* --------------------------------------------------------------------- */
836 * Helper function for tmpfs_readdir. Returns as much directory entries
837 * as can fit in the uio space. The read starts at uio->uio_offset.
838 * The function returns 0 on success, -1 if there was not enough space
839 * in the uio structure to hold the directory entry or an appropriate
840 * error code if another error happens.
842 * Caller must hold the node locked (shared ok)
845 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
849 struct tmpfs_dirent *de;
851 TMPFS_VALIDATE_DIR(node);
854 * Locate the first directory entry we have to return. We have cached
855 * the last readdir in the node, so use those values if appropriate.
856 * Otherwise do a linear scan to find the requested entry.
858 startcookie = uio->uio_offset;
859 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT);
860 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
862 if (startcookie == TMPFS_DIRCOOKIE_EOF)
865 de = tmpfs_dir_lookupbycookie(node, startcookie);
870 * Read as much entries as possible; i.e., until we reach the end of
871 * the directory or we exhaust uio space.
877 /* Create a dirent structure representing the current
878 * tmpfs_node and fill it. */
879 d.d_ino = de->td_node->tn_id;
880 switch (de->td_node->tn_type) {
910 panic("tmpfs_dir_getdents: type %p %d",
911 de->td_node, (int)de->td_node->tn_type);
913 d.d_namlen = de->td_namelen;
914 KKASSERT(de->td_namelen < sizeof(d.d_name));
915 bcopy(de->td_name, d.d_name, d.d_namlen);
916 d.d_name[d.d_namlen] = '\0';
917 reclen = _DIRENT_RECLEN(d.d_namlen);
919 /* Stop reading if the directory entry we are treating is
920 * bigger than the amount of data that can be returned. */
921 if (reclen > uio->uio_resid) {
926 /* Copy the new dirent structure into the output buffer and
927 * advance pointers. */
928 error = uiomove((caddr_t)&d, reclen, uio);
931 de = RB_NEXT(tmpfs_dirtree_cookie,
932 node->tn_dir.tn_cookietree, de);
933 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
935 /* Update the offset and cache. */
937 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
939 uio->uio_offset = tmpfs_dircookie(de);
945 /* --------------------------------------------------------------------- */
948 * Resizes the aobj associated to the regular file pointed to by vp to
949 * the size newsize. 'vp' must point to a vnode that represents a regular
950 * file. 'newsize' must be positive.
952 * pass trivial as 1 when buf content will be overwritten, otherwise set 0
955 * Returns zero on success or an appropriate error code on failure.
958 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial)
961 vm_pindex_t newpages, oldpages;
962 struct tmpfs_mount *tmp;
963 struct tmpfs_node *node;
967 KKASSERT(vp->v_type == VREG);
968 KKASSERT(newsize >= 0);
971 node = VP_TO_TMPFS_NODE(vp);
972 tmp = VFS_TO_TMPFS(vp->v_mount);
975 * Convert the old and new sizes to the number of pages needed to
976 * store them. It may happen that we do not need to do anything
977 * because the last allocated page can accommodate the change on
980 TMPFS_NODE_LOCK(node);
981 oldsize = node->tn_size;
982 oldpages = round_page64(oldsize) / PAGE_SIZE;
983 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages);
984 newpages = round_page64(newsize) / PAGE_SIZE;
986 if (newpages > oldpages &&
987 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) {
988 TMPFS_NODE_UNLOCK(node);
992 node->tn_reg.tn_aobj_pages = newpages;
993 node->tn_size = newsize;
994 TMPFS_NODE_UNLOCK(node);
997 tmp->tm_pages_used += (newpages - oldpages);
1001 * When adjusting the vnode filesize and its VM object we must
1002 * also adjust our backing VM object (aobj). The blocksize
1003 * used must match the block sized we use for the buffer cache.
1005 * The backing VM object contains no VM pages, only swap
1008 if (newsize < oldsize) {
1013 error = nvtruncbuf(vp, newsize, TMPFS_BLKSIZE, -1, 0);
1014 aobj = node->tn_reg.tn_aobj;
1017 nsize = vp->v_object->size;
1018 if (nsize < osize) {
1020 swap_pager_freespace(aobj, nsize,
1027 error = nvextendbuf(vp, oldsize, newsize,
1028 TMPFS_BLKSIZE, TMPFS_BLKSIZE,
1030 aobj = node->tn_reg.tn_aobj;
1032 aobj->size = vp->v_object->size;
1039 /* --------------------------------------------------------------------- */
1042 * Change flags of the given vnode.
1043 * Caller should execute tmpfs_update on vp after a successful execution.
1044 * The vnode must be locked on entry and remain locked on exit.
1047 tmpfs_chflags(struct vnode *vp, int vaflags, struct ucred *cred)
1050 struct tmpfs_node *node;
1053 KKASSERT(vn_islocked(vp));
1055 node = VP_TO_TMPFS_NODE(vp);
1056 flags = node->tn_flags;
1058 /* Disallow this operation if the file system is mounted read-only. */
1059 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1061 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred);
1063 /* Actually change the flags on the node itself */
1065 TMPFS_NODE_LOCK(node);
1066 node->tn_flags = flags;
1067 node->tn_status |= TMPFS_NODE_CHANGED;
1068 TMPFS_NODE_UNLOCK(node);
1071 KKASSERT(vn_islocked(vp));
1076 /* --------------------------------------------------------------------- */
1079 * Change access mode on the given vnode.
1080 * Caller should execute tmpfs_update on vp after a successful execution.
1081 * The vnode must be locked on entry and remain locked on exit.
1084 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred)
1086 struct tmpfs_node *node;
1090 KKASSERT(vn_islocked(vp));
1092 node = VP_TO_TMPFS_NODE(vp);
1094 /* Disallow this operation if the file system is mounted read-only. */
1095 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1098 /* Immutable or append-only files cannot be modified, either. */
1099 if (node->tn_flags & (IMMUTABLE | APPEND))
1102 cur_mode = node->tn_mode;
1103 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid,
1107 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) {
1108 TMPFS_NODE_LOCK(node);
1109 node->tn_mode &= ~ALLPERMS;
1110 node->tn_mode |= cur_mode & ALLPERMS;
1112 node->tn_status |= TMPFS_NODE_CHANGED;
1113 TMPFS_NODE_UNLOCK(node);
1116 KKASSERT(vn_islocked(vp));
1121 /* --------------------------------------------------------------------- */
1124 * Change ownership of the given vnode. At least one of uid or gid must
1125 * be different than VNOVAL. If one is set to that value, the attribute
1127 * Caller should execute tmpfs_update on vp after a successful execution.
1128 * The vnode must be locked on entry and remain locked on exit.
1131 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
1136 struct tmpfs_node *node;
1139 KKASSERT(vn_islocked(vp));
1140 node = VP_TO_TMPFS_NODE(vp);
1142 /* Disallow this operation if the file system is mounted read-only. */
1143 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1146 /* Immutable or append-only files cannot be modified, either. */
1147 if (node->tn_flags & (IMMUTABLE | APPEND))
1150 cur_uid = node->tn_uid;
1151 cur_gid = node->tn_gid;
1152 cur_mode = node->tn_mode;
1153 error = vop_helper_chown(vp, uid, gid, cred,
1154 &cur_uid, &cur_gid, &cur_mode);
1157 TMPFS_NODE_LOCK(node);
1158 if (cur_uid != node->tn_uid ||
1159 cur_gid != node->tn_gid ||
1160 cur_mode != node->tn_mode) {
1161 node->tn_uid = cur_uid;
1162 node->tn_gid = cur_gid;
1163 node->tn_mode = cur_mode;
1164 node->tn_status |= TMPFS_NODE_CHANGED;
1166 TMPFS_NODE_UNLOCK(node);
1172 /* --------------------------------------------------------------------- */
1175 * Change size of the given vnode.
1176 * Caller should execute tmpfs_update on vp after a successful execution.
1177 * The vnode must be locked on entry and remain locked on exit.
1180 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred)
1183 struct tmpfs_node *node;
1185 KKASSERT(vn_islocked(vp));
1187 node = VP_TO_TMPFS_NODE(vp);
1189 /* Decide whether this is a valid operation based on the file type. */
1191 switch (vp->v_type) {
1196 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1205 /* Allow modifications of special files even if in the file
1206 * system is mounted read-only (we are not modifying the
1207 * files themselves, but the objects they represent). */
1211 /* Anything else is unsupported. */
1215 /* Immutable or append-only files cannot be modified, either. */
1216 if (node->tn_flags & (IMMUTABLE | APPEND))
1219 error = tmpfs_truncate(vp, size);
1220 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1221 * for us, as will update tn_status; no need to do that here. */
1223 KKASSERT(vn_islocked(vp));
1228 /* --------------------------------------------------------------------- */
1231 * Change access and modification times of the given vnode.
1232 * Caller should execute tmpfs_update on vp after a successful execution.
1233 * The vnode must be locked on entry and remain locked on exit.
1236 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1237 int vaflags, struct ucred *cred)
1239 struct tmpfs_node *node;
1241 KKASSERT(vn_islocked(vp));
1243 node = VP_TO_TMPFS_NODE(vp);
1245 /* Disallow this operation if the file system is mounted read-only. */
1246 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1249 /* Immutable or append-only files cannot be modified, either. */
1250 if (node->tn_flags & (IMMUTABLE | APPEND))
1253 TMPFS_NODE_LOCK(node);
1254 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1255 node->tn_status |= TMPFS_NODE_ACCESSED;
1257 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1258 node->tn_status |= TMPFS_NODE_MODIFIED;
1260 TMPFS_NODE_UNLOCK(node);
1262 tmpfs_itimes(vp, atime, mtime);
1264 KKASSERT(vn_islocked(vp));
1269 /* --------------------------------------------------------------------- */
1270 /* Sync timestamps */
1272 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1273 const struct timespec *mod)
1275 struct tmpfs_node *node;
1276 struct timespec now;
1278 node = VP_TO_TMPFS_NODE(vp);
1280 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1281 TMPFS_NODE_CHANGED)) == 0)
1284 vfs_timestamp(&now);
1286 TMPFS_NODE_LOCK(node);
1287 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1290 node->tn_atime = acc->tv_sec;
1291 node->tn_atimensec = acc->tv_nsec;
1293 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1296 node->tn_mtime = mod->tv_sec;
1297 node->tn_mtimensec = mod->tv_nsec;
1299 if (node->tn_status & TMPFS_NODE_CHANGED) {
1300 node->tn_ctime = now.tv_sec;
1301 node->tn_ctimensec = now.tv_nsec;
1303 /* This is the only place that clears the tn_status */
1305 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1306 TMPFS_NODE_UNLOCK(node);
1309 /* --------------------------------------------------------------------- */
1312 tmpfs_update(struct vnode *vp)
1314 tmpfs_itimes(vp, NULL, NULL);
1317 /* --------------------------------------------------------------------- */
1320 tmpfs_truncate(struct vnode *vp, off_t length)
1323 struct tmpfs_node *node;
1325 node = VP_TO_TMPFS_NODE(vp);
1332 if (node->tn_size == length) {
1337 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1341 error = tmpfs_reg_resize(vp, length, 1);
1344 TMPFS_NODE_LOCK(node);
1345 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1346 TMPFS_NODE_UNLOCK(node);
1355 /* --------------------------------------------------------------------- */
1358 tmpfs_fetch_ino(struct tmpfs_mount *tmp)
1363 ret = tmp->tm_ino++;
1370 tmpfs_dirtree_compare(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1372 if (a->td_namelen > b->td_namelen)
1374 else if (a->td_namelen < b->td_namelen)
1377 return strncmp(a->td_name, b->td_name, a->td_namelen);
1381 tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, struct tmpfs_dirent *b)