1 /* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */
4 * Copyright (c) 2005 The NetBSD Foundation, Inc.
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
34 * Efficient memory file system supporting functions.
37 #include <sys/kernel.h>
38 #include <sys/param.h>
39 #include <sys/namei.h>
43 #include <sys/systm.h>
44 #include <sys/vnode.h>
45 #include <sys/vmmeter.h>
48 #include <vm/vm_object.h>
49 #include <vm/vm_page.h>
50 #include <vm/vm_pager.h>
51 #include <vm/vm_extern.h>
53 #include <vfs/tmpfs/tmpfs.h>
54 #include <vfs/tmpfs/tmpfs_vnops.h>
56 static ino_t tmpfs_fetch_ino(struct tmpfs_mount *);
58 static int tmpfs_dirtree_compare(struct tmpfs_dirent *a,
59 struct tmpfs_dirent *b);
60 RB_GENERATE(tmpfs_dirtree, tmpfs_dirent, rb_node, tmpfs_dirtree_compare);
62 static int tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a,
63 struct tmpfs_dirent *b);
64 RB_GENERATE(tmpfs_dirtree_cookie, tmpfs_dirent,
65 rb_cookienode, tmpfs_dirtree_compare_cookie);
68 /* --------------------------------------------------------------------- */
71 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
72 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
73 * using the credentials of the process 'p'.
75 * If the node type is set to 'VDIR', then the parent parameter must point
76 * to the parent directory of the node being created. It may only be NULL
77 * while allocating the root node.
79 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
80 * specifies the device the node represents.
82 * If the node type is set to 'VLNK', then the parameter target specifies
83 * the file name of the target file for the symbolic link that is being
86 * Note that new nodes are retrieved from the available list if it has
87 * items or, if it is empty, from the node pool as long as there is enough
88 * space to create them.
90 * Returns zero on success or an appropriate error code on failure.
93 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
94 uid_t uid, gid_t gid, mode_t mode,
95 char *target, int rmajor, int rminor,
96 struct tmpfs_node **node)
98 struct tmpfs_node *nnode;
102 KKASSERT(IFF(type == VLNK, target != NULL));
103 KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL));
105 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
108 nnode = objcache_get(tmp->tm_node_pool, M_WAITOK | M_NULLOK);
112 /* Generic initialization. */
113 nnode->tn_type = type;
115 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime
117 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec
121 nnode->tn_mode = mode;
122 nnode->tn_id = tmpfs_fetch_ino(tmp);
123 nnode->tn_advlock.init_done = 0;
124 KKASSERT(nnode->tn_links == 0);
126 /* Type-specific initialization. */
127 switch (nnode->tn_type) {
130 rdev = makeudev(rmajor, rminor);
131 if (rdev == NOUDEV) {
132 objcache_put(tmp->tm_node_pool, nnode);
135 nnode->tn_rdev = rdev;
139 RB_INIT(&nnode->tn_dir.tn_dirtree);
140 RB_INIT(&nnode->tn_dir.tn_cookietree);
150 nnode->tn_size = strlen(target);
151 nnode->tn_link = kmalloc(nnode->tn_size + 1, tmp->tm_name_zone,
152 M_WAITOK | M_NULLOK);
153 if (nnode->tn_link == NULL) {
154 objcache_put(tmp->tm_node_pool, nnode);
157 bcopy(target, nnode->tn_link, nnode->tn_size);
158 nnode->tn_link[nnode->tn_size] = '\0';
162 nnode->tn_reg.tn_aobj = swap_pager_alloc(NULL, 0,
164 nnode->tn_reg.tn_aobj_pages = 0;
166 vm_object_set_flag(nnode->tn_reg.tn_aobj, OBJ_NOPAGEIN);
170 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
173 TMPFS_NODE_LOCK(nnode);
175 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
176 tmp->tm_nodes_inuse++;
178 TMPFS_NODE_UNLOCK(nnode);
184 /* --------------------------------------------------------------------- */
187 * Destroys the node pointed to by node from the file system 'tmp'.
188 * If the node does not belong to the given mount point, the results are
191 * If the node references a directory; no entries are allowed because
192 * their removal could need a recursive algorithm, something forbidden in
193 * kernel space. Furthermore, there is not need to provide such
194 * functionality (recursive removal) because the only primitives offered
195 * to the user are the removal of empty directories and the deletion of
198 * Note that nodes are not really deleted; in fact, when a node has been
199 * allocated, it cannot be deleted during the whole life of the file
200 * system. Instead, they are moved to the available list and remain there
203 * A caller must have TMPFS_NODE_LOCK(node) and this function unlocks it.
206 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
208 vm_pindex_t pages = 0;
211 TMPFS_ASSERT_ELOCKED(node);
212 KKASSERT(node->tn_vnode == NULL);
213 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
217 LIST_REMOVE(node, tn_entries);
218 tmp->tm_nodes_inuse--;
220 TMPFS_NODE_UNLOCK(node); /* Caller has this lock */
222 switch (node->tn_type) {
224 /* Do not do anything. VNON is provided to let the
225 * allocation routine clean itself easily by avoiding
226 * duplicating code in it. */
235 * The parent link can be NULL if this is the root
236 * node or if it is a directory node that was rmdir'd.
238 * XXX what if node is a directory which still contains
239 * directory entries (e.g. due to a forced umount) ?
242 KKASSERT(node->tn_dir.tn_parent == NULL);
245 * If the root node is being destroyed don't leave a
246 * dangling pointer in tmpfs_mount.
248 if (node == tmp->tm_root)
257 kfree(node->tn_link, tmp->tm_name_zone);
258 node->tn_link = NULL;
263 if (node->tn_reg.tn_aobj != NULL)
264 vm_object_deallocate(node->tn_reg.tn_aobj);
265 node->tn_reg.tn_aobj = NULL;
266 pages = node->tn_reg.tn_aobj_pages;
270 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
274 * Clean up fields for the next allocation. The objcache only ctors
277 tmpfs_node_ctor(node, NULL, 0);
278 objcache_put(tmp->tm_node_pool, node);
279 /* node is now invalid */
282 atomic_add_long(&tmp->tm_pages_used, -(long)pages);
285 /* --------------------------------------------------------------------- */
288 * Allocates a new directory entry for the node node with a name of name.
289 * The new directory entry is returned in *de.
291 * The link count of node is increased by one to reflect the new object
294 * Returns zero on success or an appropriate error code on failure.
297 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
298 const char *name, uint16_t len, struct tmpfs_dirent **de)
300 struct tmpfs_dirent *nde;
302 nde = objcache_get(tmp->tm_dirent_pool, M_WAITOK);
303 nde->td_name = kmalloc(len + 1, tmp->tm_name_zone, M_WAITOK | M_NULLOK);
304 if (nde->td_name == NULL) {
305 objcache_put(tmp->tm_dirent_pool, nde);
309 nde->td_namelen = len;
310 bcopy(name, nde->td_name, len);
311 nde->td_name[len] = '\0';
315 TMPFS_NODE_LOCK(node);
317 TMPFS_NODE_UNLOCK(node);
324 /* --------------------------------------------------------------------- */
327 * Frees a directory entry. It is the caller's responsibility to destroy
328 * the node referenced by it if needed.
330 * The link count of node is decreased by one to reflect the removal of an
331 * object that referenced it. This only happens if 'node_exists' is true;
332 * otherwise the function will not access the node referred to by the
333 * directory entry, as it may already have been released from the outside.
336 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
338 struct tmpfs_node *node;
342 TMPFS_NODE_LOCK(node);
343 TMPFS_ASSERT_ELOCKED(node);
344 KKASSERT(node->tn_links > 0);
346 TMPFS_NODE_UNLOCK(node);
348 kfree(de->td_name, tmp->tm_name_zone);
352 objcache_put(tmp->tm_dirent_pool, de);
355 /* --------------------------------------------------------------------- */
358 * Allocates a new vnode for the node node or returns a new reference to
359 * an existing one if the node had already a vnode referencing it. The
360 * resulting locked vnode is returned in *vpp.
362 * Returns zero on success or an appropriate error code on failure.
364 * The caller must ensure that node cannot go away (usually by holding
365 * the related directory entry).
367 * If dnode is non-NULL this routine avoids deadlocking against it but
368 * can return EAGAIN. Caller must try again. The dnode lock will cycle
369 * in this case, it remains locked on return in all cases. dnode must
373 tmpfs_alloc_vp(struct mount *mp,
374 struct tmpfs_node *dnode, struct tmpfs_node *node, int lkflag,
382 * Interlocked extraction from node. This can race many things.
383 * We have to get a soft reference on the vnode while we hold
384 * the node locked, then acquire it properly and check for races.
386 TMPFS_NODE_LOCK(node);
387 if ((vp = node->tn_vnode) != NULL) {
388 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
390 TMPFS_NODE_UNLOCK(node);
394 * Special-case handling to avoid deadlocking against
395 * dnode. This case has been validated and occurs
396 * every so often during synth builds.
398 if (vget(vp, (lkflag & ~LK_RETRY) |
400 LK_EXCLUSIVE) != 0) {
401 TMPFS_NODE_UNLOCK(dnode);
402 if (vget(vp, (lkflag & ~LK_RETRY) |
404 LK_EXCLUSIVE) == 0) {
408 TMPFS_NODE_LOCK_SH(dnode);
416 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) {
421 if (node->tn_vnode != vp) {
432 * This should never happen.
434 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) {
435 TMPFS_NODE_UNLOCK(node);
441 * Interlock against other calls to tmpfs_alloc_vp() trying to
442 * allocate and assign a vp to node.
444 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
445 node->tn_vpstate |= TMPFS_VNODE_WANT;
446 error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH,
447 "tmpfs_alloc_vp", 0);
448 TMPFS_NODE_UNLOCK(node);
453 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
454 TMPFS_NODE_UNLOCK(node);
457 * Allocate a new vnode (may block). The ALLOCATING flag should
458 * prevent a race against someone else assigning node->tn_vnode.
460 error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE);
464 KKASSERT(node->tn_vnode == NULL);
465 KKASSERT(vp != NULL);
467 vp->v_type = node->tn_type;
469 /* Type-specific initialization. */
470 switch (node->tn_type) {
479 * VMIO is mandatory. Tmpfs also supports KVABIO
480 * for its tmpfs_strategy().
482 vsetflags(vp, VKVABIO);
483 vinitvmio(vp, node->tn_size, TMPFS_BLKSIZE, -1);
488 vp->v_ops = &mp->mnt_vn_fifo_ops;
494 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
499 TMPFS_NODE_LOCK(node);
501 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
502 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
505 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
506 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
507 TMPFS_NODE_UNLOCK(node);
508 wakeup(&node->tn_vpstate);
510 TMPFS_NODE_UNLOCK(node);
515 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
520 /* --------------------------------------------------------------------- */
523 * Allocates a new file of type 'type' and adds it to the parent directory
524 * 'dvp'; this addition is done using the component name given in 'cnp'.
525 * The ownership of the new file is automatically assigned based on the
526 * credentials of the caller (through 'cnp'), the group is set based on
527 * the parent directory and the mode is determined from the 'vap' argument.
528 * If successful, *vpp holds a vnode to the newly created file and zero
529 * is returned. Otherwise *vpp is NULL and the function returns an
530 * appropriate error code.
533 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
534 struct namecache *ncp, struct ucred *cred, char *target)
537 struct tmpfs_dirent *de;
538 struct tmpfs_mount *tmp;
539 struct tmpfs_node *dnode;
540 struct tmpfs_node *node;
542 tmp = VFS_TO_TMPFS(dvp->v_mount);
543 dnode = VP_TO_TMPFS_DIR(dvp);
547 * If the directory was removed but a process was CD'd into it,
548 * we do not allow any more file/dir creation within it. Otherwise
549 * we will lose track of it.
551 KKASSERT(dnode->tn_type == VDIR);
552 if (dnode != tmp->tm_root && dnode->tn_dir.tn_parent == NULL)
556 * Make sure the link count does not overflow.
558 if (vap->va_type == VDIR && dnode->tn_links >= LINK_MAX)
561 /* Allocate a node that represents the new file. */
562 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid,
563 dnode->tn_gid, vap->va_mode, target,
564 vap->va_rmajor, vap->va_rminor, &node);
567 TMPFS_NODE_LOCK(node);
569 /* Allocate a directory entry that points to the new file. */
570 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de);
572 tmpfs_free_node(tmp, node);
577 /* Allocate a vnode for the new file. */
578 error = tmpfs_alloc_vp(dvp->v_mount, NULL, node, LK_EXCLUSIVE, vpp);
580 tmpfs_free_dirent(tmp, de);
581 tmpfs_free_node(tmp, node);
587 * Now that all required items are allocated, we can proceed to
588 * insert the new node into the directory, an operation that
591 tmpfs_dir_attach(dnode, de);
592 TMPFS_NODE_UNLOCK(node);
597 /* --------------------------------------------------------------------- */
600 * Attaches the directory entry de to the directory represented by dnode.
601 * Note that this does not change the link count of the node pointed by
602 * the directory entry, as this is done by tmpfs_alloc_dirent.
605 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
607 struct tmpfs_node *node = de->td_node;
609 TMPFS_NODE_LOCK(dnode);
610 if (node && node->tn_type == VDIR) {
611 TMPFS_NODE_LOCK(node);
613 node->tn_status |= TMPFS_NODE_CHANGED;
614 node->tn_dir.tn_parent = dnode;
616 TMPFS_NODE_UNLOCK(node);
618 RB_INSERT(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de);
619 RB_INSERT(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de);
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 dnode.
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 struct tmpfs_node *node = de->td_node;
638 TMPFS_NODE_LOCK(dnode);
639 RB_REMOVE(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de);
640 RB_REMOVE(tmpfs_dirtree_cookie, &dnode->tn_dir.tn_cookietree, de);
641 dnode->tn_size -= sizeof(struct tmpfs_dirent);
642 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
644 TMPFS_NODE_UNLOCK(dnode);
647 * Clean out the tn_parent pointer immediately when removing a
650 * Removal of the parent linkage also cleans out the extra tn_links
651 * count we had on both node and dnode.
653 * node can be NULL (typ during a forced umount), in which case
654 * the mount code is dealing with the linkages from a linked list
657 if (node && node->tn_type == VDIR && node->tn_dir.tn_parent) {
658 TMPFS_NODE_LOCK(dnode);
659 TMPFS_NODE_LOCK(node);
660 KKASSERT(node->tn_dir.tn_parent == dnode);
663 node->tn_dir.tn_parent = NULL;
664 TMPFS_NODE_UNLOCK(node);
665 TMPFS_NODE_UNLOCK(dnode);
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 KKASSERT(f == NULL || f == de->td_node);
701 /* --------------------------------------------------------------------- */
704 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
705 * directory and returns it in the uio space. The function returns 0
706 * on success, -1 if there was not enough space in the uio structure to
707 * hold the directory entry or an appropriate error code if another
711 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
715 TMPFS_VALIDATE_DIR(node);
716 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
718 if (vop_write_dirent(&error, uio, node->tn_id, DT_DIR, 1, "."))
721 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
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,
741 TMPFS_VALIDATE_DIR(node);
742 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
744 if (node->tn_dir.tn_parent) {
745 TMPFS_NODE_LOCK(node);
746 if (node->tn_dir.tn_parent)
747 d_ino = node->tn_dir.tn_parent->tn_id;
749 d_ino = tmp->tm_root->tn_id;
750 TMPFS_NODE_UNLOCK(node);
752 d_ino = tmp->tm_root->tn_id;
755 if (vop_write_dirent(&error, uio, d_ino, DT_DIR, 2, ".."))
758 struct tmpfs_dirent *de;
759 de = RB_MIN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree);
761 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
763 uio->uio_offset = tmpfs_dircookie(de);
768 /* --------------------------------------------------------------------- */
771 * Lookup a directory entry by its associated cookie.
773 * Must be called with the directory node locked (shared ok)
775 struct lubycookie_info {
777 struct tmpfs_dirent *de;
781 lubycookie_cmp(struct tmpfs_dirent *de, void *arg)
783 struct lubycookie_info *info = arg;
784 off_t cookie = tmpfs_dircookie(de);
786 if (cookie < info->cookie)
788 if (cookie > info->cookie)
794 lubycookie_callback(struct tmpfs_dirent *de, void *arg)
796 struct lubycookie_info *info = arg;
798 if (tmpfs_dircookie(de) == info->cookie) {
805 struct tmpfs_dirent *
806 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
808 struct lubycookie_info info;
810 info.cookie = cookie;
812 RB_SCAN(tmpfs_dirtree_cookie, &node->tn_dir.tn_cookietree,
813 lubycookie_cmp, lubycookie_callback, &info);
817 /* --------------------------------------------------------------------- */
820 * Helper function for tmpfs_readdir. Returns as much directory entries
821 * as can fit in the uio space. The read starts at uio->uio_offset.
822 * The function returns 0 on success, -1 if there was not enough space
823 * in the uio structure to hold the directory entry or an appropriate
824 * error code if another error happens.
826 * Caller must hold the node locked (shared ok)
829 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
833 struct tmpfs_dirent *de;
835 TMPFS_VALIDATE_DIR(node);
838 * Locate the first directory entry we have to return. We have cached
839 * the last readdir in the node, so use those values if appropriate.
840 * Otherwise do a linear scan to find the requested entry.
842 startcookie = uio->uio_offset;
843 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT);
844 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
846 if (startcookie == TMPFS_DIRCOOKIE_EOF)
849 de = tmpfs_dir_lookupbycookie(node, startcookie);
854 * Read as much entries as possible; i.e., until we reach the end of
855 * the directory or we exhaust uio space.
861 /* Create a dirent structure representing the current
862 * tmpfs_node and fill it. */
863 d_ino = de->td_node->tn_id;
864 switch (de->td_node->tn_type) {
894 panic("tmpfs_dir_getdents: type %p %d",
895 de->td_node, (int)de->td_node->tn_type);
897 KKASSERT(de->td_namelen < 256); /* 255 + 1 */
899 if (vop_write_dirent(&error, uio, d_ino, d_type,
900 de->td_namelen, de->td_name)) {
906 de = RB_NEXT(tmpfs_dirtree_cookie,
907 node->tn_dir.tn_cookietree, de);
908 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
910 /* Update the offset and cache. */
912 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
914 uio->uio_offset = tmpfs_dircookie(de);
920 /* --------------------------------------------------------------------- */
923 * Resizes the aobj associated to the regular file pointed to by vp to
924 * the size newsize. 'vp' must point to a vnode that represents a regular
925 * file. 'newsize' must be positive.
927 * pass trivial as 1 when buf content will be overwritten, otherwise set 0
930 * Returns zero on success or an appropriate error code on failure.
932 * Caller must hold the node exclusively locked.
935 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial)
938 vm_pindex_t newpages, oldpages;
939 struct tmpfs_mount *tmp;
940 struct tmpfs_node *node;
944 KKASSERT(vp->v_type == VREG);
945 KKASSERT(newsize >= 0);
948 node = VP_TO_TMPFS_NODE(vp);
949 tmp = VFS_TO_TMPFS(vp->v_mount);
952 * Convert the old and new sizes to the number of pages needed to
953 * store them. It may happen that we do not need to do anything
954 * because the last allocated page can accommodate the change on
957 oldsize = node->tn_size;
958 oldpages = round_page64(oldsize) / PAGE_SIZE;
959 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages);
960 newpages = round_page64(newsize) / PAGE_SIZE;
962 if (newpages > oldpages &&
963 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) {
967 node->tn_reg.tn_aobj_pages = newpages;
968 node->tn_size = newsize;
970 if (newpages != oldpages)
971 atomic_add_long(&tmp->tm_pages_used, (newpages - oldpages));
974 * When adjusting the vnode filesize and its VM object we must
975 * also adjust our backing VM object (aobj). The blocksize
976 * used must match the block sized we use for the buffer cache.
978 * The backing VM object may contain VM pages as well as swap
979 * assignments if we previously renamed main object pages into
980 * it during deactivation.
982 if (newsize < oldsize) {
987 error = nvtruncbuf(vp, newsize, TMPFS_BLKSIZE, -1, 0);
988 aobj = node->tn_reg.tn_aobj;
991 nsize = vp->v_object->size;
994 swap_pager_freespace(aobj, nsize,
996 vm_object_page_remove(aobj, nsize, osize,
1003 error = nvextendbuf(vp, oldsize, newsize,
1004 TMPFS_BLKSIZE, TMPFS_BLKSIZE,
1006 aobj = node->tn_reg.tn_aobj;
1008 aobj->size = vp->v_object->size;
1015 /* --------------------------------------------------------------------- */
1018 * Change flags of the given vnode.
1019 * Caller should execute tmpfs_update on vp after a successful execution.
1020 * The vnode must be locked on entry and remain locked on exit.
1023 tmpfs_chflags(struct vnode *vp, u_long vaflags, struct ucred *cred)
1026 struct tmpfs_node *node;
1029 KKASSERT(vn_islocked(vp));
1031 node = VP_TO_TMPFS_NODE(vp);
1032 flags = node->tn_flags;
1034 /* Disallow this operation if the file system is mounted read-only. */
1035 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1037 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred);
1039 /* Actually change the flags on the node itself */
1041 TMPFS_NODE_LOCK(node);
1042 node->tn_flags = flags;
1043 node->tn_status |= TMPFS_NODE_CHANGED;
1044 TMPFS_NODE_UNLOCK(node);
1047 KKASSERT(vn_islocked(vp));
1052 /* --------------------------------------------------------------------- */
1055 * Change access mode on the given vnode.
1056 * Caller should execute tmpfs_update on vp after a successful execution.
1057 * The vnode must be locked on entry and remain locked on exit.
1060 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred)
1062 struct tmpfs_node *node;
1066 KKASSERT(vn_islocked(vp));
1068 node = VP_TO_TMPFS_NODE(vp);
1070 /* Disallow this operation if the file system is mounted read-only. */
1071 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1074 /* Immutable or append-only files cannot be modified, either. */
1075 if (node->tn_flags & (IMMUTABLE | APPEND))
1078 cur_mode = node->tn_mode;
1079 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid,
1083 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) {
1084 TMPFS_NODE_LOCK(node);
1085 node->tn_mode &= ~ALLPERMS;
1086 node->tn_mode |= cur_mode & ALLPERMS;
1088 node->tn_status |= TMPFS_NODE_CHANGED;
1089 TMPFS_NODE_UNLOCK(node);
1092 KKASSERT(vn_islocked(vp));
1097 /* --------------------------------------------------------------------- */
1100 * Change ownership of the given vnode. At least one of uid or gid must
1101 * be different than VNOVAL. If one is set to that value, the attribute
1103 * Caller should execute tmpfs_update on vp after a successful execution.
1104 * The vnode must be locked on entry and remain locked on exit.
1107 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
1112 struct tmpfs_node *node;
1115 KKASSERT(vn_islocked(vp));
1116 node = VP_TO_TMPFS_NODE(vp);
1118 /* Disallow this operation if the file system is mounted read-only. */
1119 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1122 /* Immutable or append-only files cannot be modified, either. */
1123 if (node->tn_flags & (IMMUTABLE | APPEND))
1126 cur_uid = node->tn_uid;
1127 cur_gid = node->tn_gid;
1128 cur_mode = node->tn_mode;
1129 error = vop_helper_chown(vp, uid, gid, cred,
1130 &cur_uid, &cur_gid, &cur_mode);
1133 TMPFS_NODE_LOCK(node);
1134 if (cur_uid != node->tn_uid ||
1135 cur_gid != node->tn_gid ||
1136 cur_mode != node->tn_mode) {
1137 node->tn_uid = cur_uid;
1138 node->tn_gid = cur_gid;
1139 node->tn_mode = cur_mode;
1140 node->tn_status |= TMPFS_NODE_CHANGED;
1142 TMPFS_NODE_UNLOCK(node);
1148 /* --------------------------------------------------------------------- */
1151 * Change size of the given vnode.
1152 * Caller should execute tmpfs_update on vp after a successful execution.
1153 * The vnode must be locked on entry and remain locked on exit.
1156 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred)
1159 struct tmpfs_node *node;
1161 KKASSERT(vn_islocked(vp));
1163 node = VP_TO_TMPFS_NODE(vp);
1165 /* Decide whether this is a valid operation based on the file type. */
1167 switch (vp->v_type) {
1172 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1181 /* Allow modifications of special files even if in the file
1182 * system is mounted read-only (we are not modifying the
1183 * files themselves, but the objects they represent). */
1187 /* Anything else is unsupported. */
1191 /* Immutable or append-only files cannot be modified, either. */
1192 if (node->tn_flags & (IMMUTABLE | APPEND))
1195 error = tmpfs_truncate(vp, size);
1196 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1197 * for us, as will update tn_status; no need to do that here. */
1199 KKASSERT(vn_islocked(vp));
1204 /* --------------------------------------------------------------------- */
1207 * Change access and modification times of the given vnode.
1208 * Caller should execute tmpfs_update on vp after a successful execution.
1209 * The vnode must be locked on entry and remain locked on exit.
1212 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1213 int vaflags, struct ucred *cred)
1215 struct tmpfs_node *node;
1217 KKASSERT(vn_islocked(vp));
1219 node = VP_TO_TMPFS_NODE(vp);
1221 /* Disallow this operation if the file system is mounted read-only. */
1222 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1225 /* Immutable or append-only files cannot be modified, either. */
1226 if (node->tn_flags & (IMMUTABLE | APPEND))
1229 TMPFS_NODE_LOCK(node);
1230 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1231 node->tn_status |= TMPFS_NODE_ACCESSED;
1233 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL) {
1234 node->tn_status |= TMPFS_NODE_MODIFIED;
1235 vclrflags(vp, VLASTWRITETS);
1238 TMPFS_NODE_UNLOCK(node);
1240 tmpfs_itimes(vp, atime, mtime);
1242 KKASSERT(vn_islocked(vp));
1247 /* --------------------------------------------------------------------- */
1248 /* Sync timestamps */
1250 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1251 const struct timespec *mod)
1253 struct tmpfs_node *node;
1254 struct timespec now;
1256 node = VP_TO_TMPFS_NODE(vp);
1258 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1259 TMPFS_NODE_CHANGED)) == 0)
1262 vfs_timestamp(&now);
1264 TMPFS_NODE_LOCK(node);
1265 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1268 node->tn_atime = acc->tv_sec;
1269 node->tn_atimensec = acc->tv_nsec;
1271 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1274 node->tn_mtime = mod->tv_sec;
1275 node->tn_mtimensec = mod->tv_nsec;
1277 if (node->tn_status & TMPFS_NODE_CHANGED) {
1278 node->tn_ctime = now.tv_sec;
1279 node->tn_ctimensec = now.tv_nsec;
1282 node->tn_status &= ~(TMPFS_NODE_ACCESSED |
1283 TMPFS_NODE_MODIFIED |
1284 TMPFS_NODE_CHANGED);
1285 TMPFS_NODE_UNLOCK(node);
1288 /* --------------------------------------------------------------------- */
1291 tmpfs_update(struct vnode *vp)
1293 tmpfs_itimes(vp, NULL, NULL);
1296 /* --------------------------------------------------------------------- */
1299 * Caller must hold an exclusive node lock.
1302 tmpfs_truncate(struct vnode *vp, off_t length)
1305 struct tmpfs_node *node;
1307 node = VP_TO_TMPFS_NODE(vp);
1314 if (node->tn_size == length) {
1319 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1323 error = tmpfs_reg_resize(vp, length, 1);
1326 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1334 /* --------------------------------------------------------------------- */
1337 tmpfs_fetch_ino(struct tmpfs_mount *tmp)
1341 ret = atomic_fetchadd_64(&tmp->tm_ino, 1);
1347 tmpfs_dirtree_compare(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1349 if (a->td_namelen > b->td_namelen)
1351 else if (a->td_namelen < b->td_namelen)
1354 return strncmp(a->td_name, b->td_name, a->td_namelen);
1358 tmpfs_dirtree_compare_cookie(struct tmpfs_dirent *a, struct tmpfs_dirent *b)