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.
36 #include <sys/cdefs.h>
38 #include <sys/kernel.h>
39 #include <sys/param.h>
40 #include <sys/namei.h>
43 #include <sys/spinlock2.h>
45 #include <sys/systm.h>
46 #include <sys/vnode.h>
47 #include <sys/vmmeter.h>
49 #include <sys/mplock2.h>
52 #include <vm/vm_object.h>
53 #include <vm/vm_page.h>
54 #include <vm/vm_pager.h>
55 #include <vm/vm_extern.h>
57 #include <vfs/tmpfs/tmpfs.h>
58 #include <vfs/tmpfs/tmpfs_fifoops.h>
59 #include <vfs/tmpfs/tmpfs_vnops.h>
61 static ino_t t_ino = 2;
62 static struct spinlock ino_lock;
63 static ino_t tmpfs_fetch_ino(void);
65 /* --------------------------------------------------------------------- */
68 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
69 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
70 * using the credentials of the process 'p'.
72 * If the node type is set to 'VDIR', then the parent parameter must point
73 * to the parent directory of the node being created. It may only be NULL
74 * while allocating the root node.
76 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
77 * specifies the device the node represents.
79 * If the node type is set to 'VLNK', then the parameter target specifies
80 * the file name of the target file for the symbolic link that is being
83 * Note that new nodes are retrieved from the available list if it has
84 * items or, if it is empty, from the node pool as long as there is enough
85 * space to create them.
87 * Returns zero on success or an appropriate error code on failure.
90 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
91 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
92 char *target, int rmajor, int rminor, struct tmpfs_node **node)
94 struct tmpfs_node *nnode;
98 /* If the root directory of the 'tmp' file system is not yet
99 * allocated, this must be the request to do it. */
100 KKASSERT(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
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 = (struct tmpfs_node *)objcache_get(tmp->tm_node_pool, M_WAITOK);
110 /* Generic initialization. */
111 nnode->tn_type = type;
113 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime
115 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec
119 nnode->tn_mode = mode;
120 nnode->tn_id = tmpfs_fetch_ino();
121 nnode->tn_advlock.init_done = 0;
123 /* Type-specific initialization. */
124 switch (nnode->tn_type) {
127 rdev = makeudev(rmajor, rminor);
128 if (rdev == NOUDEV) {
131 nnode->tn_rdev = rdev;
135 TAILQ_INIT(&nnode->tn_dir.tn_dirhead);
136 KKASSERT(parent != nnode);
137 KKASSERT(IMPLIES(parent == NULL, tmp->tm_root == NULL));
138 nnode->tn_dir.tn_parent = parent;
139 nnode->tn_dir.tn_readdir_lastn = 0;
140 nnode->tn_dir.tn_readdir_lastp = NULL;
144 TMPFS_NODE_LOCK(parent);
146 TMPFS_NODE_UNLOCK(parent);
156 nnode->tn_size = strlen(target);
157 nnode->tn_link = kmalloc(nnode->tn_size + 1, M_TMPFSNAME,
159 bcopy(target, nnode->tn_link, nnode->tn_size);
160 nnode->tn_link[nnode->tn_size] = '\0';
164 nnode->tn_reg.tn_aobj =
165 swap_pager_alloc(NULL, 0, VM_PROT_DEFAULT, 0);
166 nnode->tn_reg.tn_aobj_pages = 0;
171 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
174 TMPFS_NODE_LOCK(nnode);
176 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
177 tmp->tm_nodes_inuse++;
179 TMPFS_NODE_UNLOCK(nnode);
185 /* --------------------------------------------------------------------- */
188 * Destroys the node pointed to by node from the file system 'tmp'.
189 * If the node does not belong to the given mount point, the results are
192 * If the node references a directory; no entries are allowed because
193 * their removal could need a recursive algorithm, something forbidden in
194 * kernel space. Furthermore, there is not need to provide such
195 * functionality (recursive removal) because the only primitives offered
196 * to the user are the removal of empty directories and the deletion of
199 * Note that nodes are not really deleted; in fact, when a node has been
200 * allocated, it cannot be deleted during the whole life of the file
201 * system. Instead, they are moved to the available list and remain there
205 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
207 vm_pindex_t pages = 0;
210 TMPFS_ASSERT_ELOCKED(node);
211 KKASSERT(node->tn_vnode == NULL);
212 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
216 LIST_REMOVE(node, tn_entries);
217 tmp->tm_nodes_inuse--;
219 TMPFS_NODE_UNLOCK(node);
221 switch (node->tn_type) {
223 /* Do not do anything. VNON is provided to let the
224 * allocation routine clean itself easily by avoiding
225 * duplicating code in it. */
234 * The parent link can be NULL if this is the root
239 KKASSERT(node->tn_dir.tn_parent || node == tmp->tm_root);
240 if (node->tn_dir.tn_parent) {
241 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
242 node->tn_dir.tn_parent->tn_links--;
245 * If the parent directory has no more links and
246 * no vnode ref nothing is going to come along
247 * and clean it up unless we do it here.
249 if (node->tn_dir.tn_parent->tn_links == 0 &&
250 node->tn_dir.tn_parent->tn_vnode == NULL) {
251 tmpfs_free_node(tmp, node->tn_dir.tn_parent);
252 /* eats parent lock */
254 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
256 node->tn_dir.tn_parent = NULL;
260 * If the root node is being destroyed don't leave a
261 * dangling pointer in tmpfs_mount.
263 if (node == tmp->tm_root)
272 kfree(node->tn_link, M_TMPFSNAME);
273 node->tn_link = NULL;
278 if (node->tn_reg.tn_aobj != NULL)
279 vm_object_deallocate(node->tn_reg.tn_aobj);
280 node->tn_reg.tn_aobj = NULL;
281 pages = node->tn_reg.tn_aobj_pages;
285 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
289 * Clean up fields for the next allocation. The objcache only ctors
292 tmpfs_node_ctor(node, NULL, 0);
293 objcache_put(tmp->tm_node_pool, node);
294 /* node is now invalid */
297 tmp->tm_pages_used -= pages;
301 /* --------------------------------------------------------------------- */
304 * Allocates a new directory entry for the node node with a name of name.
305 * The new directory entry is returned in *de.
307 * The link count of node is increased by one to reflect the new object
310 * Returns zero on success or an appropriate error code on failure.
313 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
314 const char *name, uint16_t len, struct tmpfs_dirent **de)
316 struct tmpfs_dirent *nde;
319 nde = (struct tmpfs_dirent *)objcache_get(tmp->tm_dirent_pool, M_WAITOK);
320 nde->td_name = kmalloc(len + 1, M_TMPFSNAME, M_WAITOK);
321 nde->td_namelen = len;
322 bcopy(name, nde->td_name, len);
323 nde->td_name[len] = '\0';
327 TMPFS_NODE_LOCK(node);
329 TMPFS_NODE_UNLOCK(node);
336 /* --------------------------------------------------------------------- */
339 * Frees a directory entry. It is the caller's responsibility to destroy
340 * the node referenced by it if needed.
342 * The link count of node is decreased by one to reflect the removal of an
343 * object that referenced it. This only happens if 'node_exists' is true;
344 * otherwise the function will not access the node referred to by the
345 * directory entry, as it may already have been released from the outside.
348 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
350 struct tmpfs_node *node;
354 TMPFS_NODE_LOCK(node);
355 TMPFS_ASSERT_ELOCKED(node);
356 KKASSERT(node->tn_links > 0);
358 TMPFS_NODE_UNLOCK(node);
360 kfree(de->td_name, M_TMPFSNAME);
364 objcache_put(tmp->tm_dirent_pool, de);
367 /* --------------------------------------------------------------------- */
370 * Allocates a new vnode for the node node or returns a new reference to
371 * an existing one if the node had already a vnode referencing it. The
372 * resulting locked vnode is returned in *vpp.
374 * Returns zero on success or an appropriate error code on failure.
377 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
385 * Interlocked extraction from node. This can race many things.
386 * We have to get a soft reference on the vnode while we hold
387 * the node locked, then acquire it properly and check for races.
389 TMPFS_NODE_LOCK(node);
390 if ((vp = node->tn_vnode) != NULL) {
391 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
392 vhold_interlocked(vp);
393 TMPFS_NODE_UNLOCK(node);
395 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) {
399 if (node->tn_vnode != vp) {
410 * This should never happen.
412 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) {
413 TMPFS_NODE_UNLOCK(node);
419 * Interlock against other calls to tmpfs_alloc_vp() trying to
420 * allocate and assign a vp to node.
422 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
423 node->tn_vpstate |= TMPFS_VNODE_WANT;
424 error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH,
425 "tmpfs_alloc_vp", 0);
426 TMPFS_NODE_UNLOCK(node);
431 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
432 TMPFS_NODE_UNLOCK(node);
435 * Allocate a new vnode (may block). The ALLOCATING flag should
436 * prevent a race against someone else assigning node->tn_vnode.
438 error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE);
442 KKASSERT(node->tn_vnode == NULL);
443 KKASSERT(vp != NULL);
445 vp->v_type = node->tn_type;
447 /* Type-specific initialization. */
448 switch (node->tn_type) {
456 vinitvmio(vp, node->tn_size, BMASK, -1);
461 vp->v_ops = &mp->mnt_vn_fifo_ops;
467 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
473 TMPFS_NODE_LOCK(node);
475 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
476 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
479 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
480 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
481 TMPFS_NODE_UNLOCK(node);
482 wakeup(&node->tn_vpstate);
484 TMPFS_NODE_UNLOCK(node);
490 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
492 TMPFS_NODE_LOCK(node);
493 KKASSERT(*vpp == node->tn_vnode);
494 TMPFS_NODE_UNLOCK(node);
500 /* --------------------------------------------------------------------- */
503 * Destroys the association between the vnode vp and the node it
507 tmpfs_free_vp(struct vnode *vp)
509 struct tmpfs_node *node;
511 node = VP_TO_TMPFS_NODE(vp);
513 TMPFS_NODE_LOCK(node);
514 KKASSERT(lockcount(TMPFS_NODE_MTX(node)) > 0);
515 node->tn_vnode = NULL;
516 TMPFS_NODE_UNLOCK(node);
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;
541 struct tmpfs_node *parent;
543 tmp = VFS_TO_TMPFS(dvp->v_mount);
544 dnode = VP_TO_TMPFS_DIR(dvp);
547 /* If the entry we are creating is a directory, we cannot overflow
548 * the number of links of its parent, because it will get a new
550 if (vap->va_type == VDIR) {
551 /* Ensure that we do not overflow the maximum number of links
552 * imposed by the system. */
553 KKASSERT(dnode->tn_links <= LINK_MAX);
554 if (dnode->tn_links == LINK_MAX) {
559 KKASSERT(parent != NULL);
563 /* Allocate a node that represents the new file. */
564 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid,
565 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rmajor, vap->va_rminor, &node);
568 TMPFS_NODE_LOCK(node);
570 /* Allocate a directory entry that points to the new file. */
571 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de);
573 tmpfs_free_node(tmp, node);
578 /* Allocate a vnode for the new file. */
579 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
581 tmpfs_free_dirent(tmp, de);
582 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
590 tmpfs_dir_attach(dnode, de);
591 TMPFS_NODE_UNLOCK(node);
596 /* --------------------------------------------------------------------- */
599 * Attaches the directory entry de to the directory represented by vp.
600 * Note that this does not change the link count of the node pointed by
601 * the directory entry, as this is done by tmpfs_alloc_dirent.
604 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
606 TMPFS_NODE_LOCK(dnode);
607 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
609 TMPFS_ASSERT_ELOCKED(dnode);
610 dnode->tn_size += sizeof(struct tmpfs_dirent);
611 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
613 TMPFS_NODE_UNLOCK(dnode);
616 /* --------------------------------------------------------------------- */
619 * Detaches the directory entry de from the directory represented by vp.
620 * Note that this does not change the link count of the node pointed by
621 * the directory entry, as this is done by tmpfs_free_dirent.
624 tmpfs_dir_detach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
626 TMPFS_NODE_LOCK(dnode);
627 if (dnode->tn_dir.tn_readdir_lastp == de) {
628 dnode->tn_dir.tn_readdir_lastn = 0;
629 dnode->tn_dir.tn_readdir_lastp = NULL;
631 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
633 TMPFS_ASSERT_ELOCKED(dnode);
634 dnode->tn_size -= sizeof(struct tmpfs_dirent);
635 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
637 TMPFS_NODE_UNLOCK(dnode);
640 /* --------------------------------------------------------------------- */
643 * Looks for a directory entry in the directory represented by node.
644 * 'ncp' describes the name of the entry to look for. Note that the .
645 * and .. components are not allowed as they do not physically exist
646 * within directories.
648 * Returns a pointer to the entry when found, otherwise NULL.
650 struct tmpfs_dirent *
651 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
652 struct namecache *ncp)
654 struct tmpfs_dirent *de;
655 int len = ncp->nc_nlen;
657 TMPFS_VALIDATE_DIR(node);
659 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
660 if (f != NULL && de->td_node != f)
662 if (len == de->td_namelen) {
663 if (!memcmp(ncp->nc_name, de->td_name, len))
668 TMPFS_NODE_LOCK(node);
669 node->tn_status |= TMPFS_NODE_ACCESSED;
670 TMPFS_NODE_UNLOCK(node);
675 /* --------------------------------------------------------------------- */
678 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
679 * directory and returns it in the uio space. The function returns 0
680 * on success, -1 if there was not enough space in the uio structure to
681 * hold the directory entry or an appropriate error code if another
685 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
691 TMPFS_VALIDATE_DIR(node);
692 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
694 dent.d_ino = node->tn_id;
695 dent.d_type = DT_DIR;
697 dent.d_name[0] = '.';
698 dent.d_name[1] = '\0';
699 dirsize = _DIRENT_DIRSIZ(&dent);
701 if (dirsize > uio->uio_resid)
704 error = uiomove((caddr_t)&dent, dirsize, uio);
706 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
709 TMPFS_NODE_LOCK(node);
710 node->tn_status |= TMPFS_NODE_ACCESSED;
711 TMPFS_NODE_UNLOCK(node);
716 /* --------------------------------------------------------------------- */
719 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
720 * directory and returns it in the uio space. The function returns 0
721 * on success, -1 if there was not enough space in the uio structure to
722 * hold the directory entry or an appropriate error code if another
726 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
733 TMPFS_VALIDATE_DIR(node);
734 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
736 if (node->tn_dir.tn_parent) {
737 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
738 dent.d_ino = node->tn_dir.tn_parent->tn_id;
739 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
741 dent.d_ino = tmp->tm_root->tn_id;
744 dent.d_type = DT_DIR;
746 dent.d_name[0] = '.';
747 dent.d_name[1] = '.';
748 dent.d_name[2] = '\0';
749 dirsize = _DIRENT_DIRSIZ(&dent);
751 if (dirsize > uio->uio_resid)
754 error = uiomove((caddr_t)&dent, dirsize, uio);
756 struct tmpfs_dirent *de;
758 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
760 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
762 uio->uio_offset = tmpfs_dircookie(de);
766 TMPFS_NODE_LOCK(node);
767 node->tn_status |= TMPFS_NODE_ACCESSED;
768 TMPFS_NODE_UNLOCK(node);
773 /* --------------------------------------------------------------------- */
776 * Lookup a directory entry by its associated cookie.
778 struct tmpfs_dirent *
779 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
781 struct tmpfs_dirent *de;
783 if (cookie == node->tn_dir.tn_readdir_lastn &&
784 node->tn_dir.tn_readdir_lastp != NULL) {
785 return node->tn_dir.tn_readdir_lastp;
788 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
789 if (tmpfs_dircookie(de) == cookie) {
797 /* --------------------------------------------------------------------- */
800 * Helper function for tmpfs_readdir. Returns as much directory entries
801 * as can fit in the uio space. The read starts at uio->uio_offset.
802 * The function returns 0 on success, -1 if there was not enough space
803 * in the uio structure to hold the directory entry or an appropriate
804 * error code if another error happens.
807 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
811 struct tmpfs_dirent *de;
813 TMPFS_VALIDATE_DIR(node);
815 /* Locate the first directory entry we have to return. We have cached
816 * the last readdir in the node, so use those values if appropriate.
817 * Otherwise do a linear scan to find the requested entry. */
818 startcookie = uio->uio_offset;
819 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT);
820 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
821 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
824 de = tmpfs_dir_lookupbycookie(node, startcookie);
830 /* Read as much entries as possible; i.e., until we reach the end of
831 * the directory or we exhaust uio space. */
836 /* Create a dirent structure representing the current
837 * tmpfs_node and fill it. */
838 d.d_ino = de->td_node->tn_id;
839 switch (de->td_node->tn_type) {
869 panic("tmpfs_dir_getdents: type %p %d",
870 de->td_node, (int)de->td_node->tn_type);
872 d.d_namlen = de->td_namelen;
873 KKASSERT(de->td_namelen < sizeof(d.d_name));
874 bcopy(de->td_name, d.d_name, d.d_namlen);
875 d.d_name[d.d_namlen] = '\0';
876 reclen = _DIRENT_RECLEN(d.d_namlen);
878 /* Stop reading if the directory entry we are treating is
879 * bigger than the amount of data that can be returned. */
880 if (reclen > uio->uio_resid) {
885 /* Copy the new dirent structure into the output buffer and
886 * advance pointers. */
887 error = uiomove((caddr_t)&d, reclen, uio);
890 de = TAILQ_NEXT(de, td_entries);
891 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
893 /* Update the offset and cache. */
895 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
896 node->tn_dir.tn_readdir_lastn = 0;
897 node->tn_dir.tn_readdir_lastp = NULL;
899 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
900 node->tn_dir.tn_readdir_lastp = de;
902 node->tn_status |= TMPFS_NODE_ACCESSED;
907 /* --------------------------------------------------------------------- */
910 * Resizes the aobj associated to the regular file pointed to by vp to
911 * the size newsize. 'vp' must point to a vnode that represents a regular
912 * file. 'newsize' must be positive.
914 * pass trivial as 1 when buf content will be overwritten, otherwise set 0
917 * Returns zero on success or an appropriate error code on failure.
920 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial)
923 vm_pindex_t newpages, oldpages;
924 struct tmpfs_mount *tmp;
925 struct tmpfs_node *node;
929 KKASSERT(vp->v_type == VREG);
930 KKASSERT(newsize >= 0);
933 node = VP_TO_TMPFS_NODE(vp);
934 tmp = VFS_TO_TMPFS(vp->v_mount);
936 /* Convert the old and new sizes to the number of pages needed to
937 * store them. It may happen that we do not need to do anything
938 * because the last allocated page can accommodate the change on
940 oldsize = node->tn_size;
941 oldpages = round_page64(oldsize) / PAGE_SIZE;
942 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages);
943 newpages = round_page64(newsize) / PAGE_SIZE;
945 if (newpages > oldpages &&
946 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) {
952 tmp->tm_pages_used += (newpages - oldpages);
955 TMPFS_NODE_LOCK(node);
956 node->tn_reg.tn_aobj_pages = newpages;
957 node->tn_size = newsize;
958 TMPFS_NODE_UNLOCK(node);
961 * When adjusting the vnode filesize and its VM object we must
962 * also adjust our backing VM object (aobj). The blocksize
963 * used must match the block sized we use for the buffer cache.
965 * The backing VM object contains no VM pages, only swap
968 if (newsize < oldsize) {
973 error = nvtruncbuf(vp, newsize, BSIZE, -1);
974 aobj = node->tn_reg.tn_aobj;
977 nsize = vp->v_object->size;
980 swap_pager_freespace(aobj, nsize,
987 error = nvextendbuf(vp, oldsize, newsize, BSIZE, BSIZE,
989 aobj = node->tn_reg.tn_aobj;
991 aobj->size = vp->v_object->size;
998 /* --------------------------------------------------------------------- */
1001 * Change flags of the given vnode.
1002 * Caller should execute tmpfs_update on vp after a successful execution.
1003 * The vnode must be locked on entry and remain locked on exit.
1006 tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred)
1009 struct tmpfs_node *node;
1012 KKASSERT(vn_islocked(vp));
1014 node = VP_TO_TMPFS_NODE(vp);
1016 /* Disallow this operation if the file system is mounted read-only. */
1017 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1020 fmode = FFLAGS(node->tn_flags);
1022 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
1024 if (fmode & (FWRITE | O_TRUNC)) {
1025 if (vp->v_type == VDIR) {
1028 error = vn_writechk(vp, NULL);
1037 error = VOP_ACCESS(vp, mode, cred);
1042 * Unprivileged processes are not permitted to unset system
1043 * flags, or modify flags if any system flags are set.
1045 * Silently enforce SF_NOCACHE on the root tmpfs vnode so
1046 * tmpfs data is not double-cached by swapcache.
1048 TMPFS_NODE_LOCK(node);
1049 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1052 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1053 error = securelevel_gt(cred, 0);
1057 /* Snapshot flag cannot be set or cleared */
1058 if (((flags & SF_SNAPSHOT) != 0 &&
1059 (node->tn_flags & SF_SNAPSHOT) == 0) ||
1060 ((flags & SF_SNAPSHOT) == 0 &&
1061 (node->tn_flags & SF_SNAPSHOT) != 0))
1064 if (vp->v_flag & VROOT)
1065 flags |= SF_NOCACHE;
1066 node->tn_flags = flags;
1069 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1070 (flags & UF_SETTABLE) != flags)
1072 node->tn_flags &= SF_SETTABLE;
1073 node->tn_flags |= (flags & UF_SETTABLE);
1075 node->tn_status |= TMPFS_NODE_CHANGED;
1076 TMPFS_NODE_UNLOCK(node);
1078 KKASSERT(vn_islocked(vp));
1083 /* --------------------------------------------------------------------- */
1086 * Change access mode on the given vnode.
1087 * Caller should execute tmpfs_update on vp after a successful execution.
1088 * The vnode must be locked on entry and remain locked on exit.
1091 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred)
1094 struct tmpfs_node *node;
1097 KKASSERT(vn_islocked(vp));
1099 node = VP_TO_TMPFS_NODE(vp);
1101 /* Disallow this operation if the file system is mounted read-only. */
1102 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1105 /* Immutable or append-only files cannot be modified, either. */
1106 if (node->tn_flags & (IMMUTABLE | APPEND))
1109 fmode = FFLAGS(node->tn_flags);
1111 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
1113 if (fmode & (FWRITE | O_TRUNC)) {
1114 if (vp->v_type == VDIR) {
1117 error = vn_writechk(vp, NULL);
1126 error = VOP_ACCESS(vp, accmode, cred);
1132 * Privileged processes may set the sticky bit on non-directories,
1133 * as well as set the setgid bit on a file with a group that the
1134 * process is not a member of.
1136 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1137 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1140 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1141 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1147 TMPFS_NODE_LOCK(node);
1148 node->tn_mode &= ~ALLPERMS;
1149 node->tn_mode |= mode & ALLPERMS;
1151 node->tn_status |= TMPFS_NODE_CHANGED;
1152 TMPFS_NODE_UNLOCK(node);
1154 KKASSERT(vn_islocked(vp));
1159 /* --------------------------------------------------------------------- */
1162 * Change ownership of the given vnode. At least one of uid or gid must
1163 * be different than VNOVAL. If one is set to that value, the attribute
1165 * Caller should execute tmpfs_update on vp after a successful execution.
1166 * The vnode must be locked on entry and remain locked on exit.
1169 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
1174 struct tmpfs_node *node;
1177 KKASSERT(vn_islocked(vp));
1178 node = VP_TO_TMPFS_NODE(vp);
1180 /* Disallow this operation if the file system is mounted read-only. */
1181 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1184 /* Immutable or append-only files cannot be modified, either. */
1185 if (node->tn_flags & (IMMUTABLE | APPEND))
1188 cur_uid = node->tn_uid;
1189 cur_gid = node->tn_gid;
1190 cur_mode = node->tn_mode;
1191 error = vop_helper_chown(vp, uid, gid, cred,
1192 &cur_uid, &cur_gid, &cur_mode);
1195 TMPFS_NODE_LOCK(node);
1196 if (cur_uid != node->tn_uid ||
1197 cur_gid != node->tn_gid ||
1198 cur_mode != node->tn_mode) {
1201 node->tn_mode = cur_mode;
1202 node->tn_status |= TMPFS_NODE_CHANGED;
1204 TMPFS_NODE_UNLOCK(node);
1210 /* --------------------------------------------------------------------- */
1213 * Change size of the given vnode.
1214 * Caller should execute tmpfs_update on vp after a successful execution.
1215 * The vnode must be locked on entry and remain locked on exit.
1218 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred)
1221 struct tmpfs_node *node;
1223 KKASSERT(vn_islocked(vp));
1225 node = VP_TO_TMPFS_NODE(vp);
1227 /* Decide whether this is a valid operation based on the file type. */
1229 switch (vp->v_type) {
1234 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1243 /* Allow modifications of special files even if in the file
1244 * system is mounted read-only (we are not modifying the
1245 * files themselves, but the objects they represent). */
1249 /* Anything else is unsupported. */
1253 /* Immutable or append-only files cannot be modified, either. */
1254 if (node->tn_flags & (IMMUTABLE | APPEND))
1257 error = tmpfs_truncate(vp, size);
1258 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1259 * for us, as will update tn_status; no need to do that here. */
1261 KKASSERT(vn_islocked(vp));
1266 /* --------------------------------------------------------------------- */
1269 * Change access and modification times of the given vnode.
1270 * Caller should execute tmpfs_update on vp after a successful execution.
1271 * The vnode must be locked on entry and remain locked on exit.
1274 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1275 int vaflags, struct ucred *cred)
1278 struct tmpfs_node *node;
1281 KKASSERT(vn_islocked(vp));
1283 node = VP_TO_TMPFS_NODE(vp);
1285 /* Disallow this operation if the file system is mounted read-only. */
1286 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1289 /* Immutable or append-only files cannot be modified, either. */
1290 if (node->tn_flags & (IMMUTABLE | APPEND))
1293 /* Determine if the user have proper privilege to update time. */
1294 fmode = FFLAGS(node->tn_flags);
1296 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
1298 if (fmode & (FWRITE | O_TRUNC)) {
1299 if (vp->v_type == VDIR) {
1302 error = vn_writechk(vp, NULL);
1312 if (vaflags & VA_UTIMES_NULL) {
1313 error = VOP_ACCESS(vp, mode, cred);
1315 error = VOP_ACCESS(vp, VWRITE, cred);
1317 error = VOP_ACCESS(vp, mode, cred);
1322 TMPFS_NODE_LOCK(node);
1323 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1324 node->tn_status |= TMPFS_NODE_ACCESSED;
1326 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1327 node->tn_status |= TMPFS_NODE_MODIFIED;
1329 TMPFS_NODE_UNLOCK(node);
1331 tmpfs_itimes(vp, atime, mtime);
1333 KKASSERT(vn_islocked(vp));
1338 /* --------------------------------------------------------------------- */
1339 /* Sync timestamps */
1341 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1342 const struct timespec *mod)
1344 struct tmpfs_node *node;
1345 struct timespec now;
1347 node = VP_TO_TMPFS_NODE(vp);
1349 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1350 TMPFS_NODE_CHANGED)) == 0)
1353 vfs_timestamp(&now);
1355 TMPFS_NODE_LOCK(node);
1356 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1359 node->tn_atime = acc->tv_sec;
1360 node->tn_atimensec = acc->tv_nsec;
1362 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1365 node->tn_mtime = mod->tv_sec;
1366 node->tn_mtimensec = mod->tv_nsec;
1368 if (node->tn_status & TMPFS_NODE_CHANGED) {
1369 node->tn_ctime = now.tv_sec;
1370 node->tn_ctimensec = now.tv_nsec;
1373 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1374 TMPFS_NODE_UNLOCK(node);
1377 /* --------------------------------------------------------------------- */
1380 tmpfs_update(struct vnode *vp)
1383 tmpfs_itimes(vp, NULL, NULL);
1386 /* --------------------------------------------------------------------- */
1389 tmpfs_truncate(struct vnode *vp, off_t length)
1392 struct tmpfs_node *node;
1394 node = VP_TO_TMPFS_NODE(vp);
1401 if (node->tn_size == length) {
1406 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1410 error = tmpfs_reg_resize(vp, length, 1);
1413 TMPFS_NODE_LOCK(node);
1414 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1415 TMPFS_NODE_UNLOCK(node);
1424 /* --------------------------------------------------------------------- */
1427 tmpfs_fetch_ino(void)
1431 spin_lock_wr(&ino_lock);
1433 spin_unlock_wr(&ino_lock);