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 = 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();
123 nnode->tn_advlock.init_done = 0;
125 /* Type-specific initialization. */
126 switch (nnode->tn_type) {
129 rdev = makeudev(rmajor, rminor);
130 if (rdev == NOUDEV) {
131 objcache_put(tmp->tm_node_pool, nnode);
134 nnode->tn_rdev = rdev;
138 TAILQ_INIT(&nnode->tn_dir.tn_dirhead);
139 KKASSERT(parent != nnode);
140 KKASSERT(IMPLIES(parent == NULL, tmp->tm_root == NULL));
141 nnode->tn_dir.tn_parent = parent;
142 nnode->tn_dir.tn_readdir_lastn = 0;
143 nnode->tn_dir.tn_readdir_lastp = NULL;
147 TMPFS_NODE_LOCK(parent);
149 TMPFS_NODE_UNLOCK(parent);
159 nnode->tn_size = strlen(target);
160 nnode->tn_link = kmalloc(nnode->tn_size + 1, tmp->tm_name_zone,
161 M_WAITOK | M_NULLOK);
162 if (nnode->tn_link == NULL) {
163 objcache_put(tmp->tm_node_pool, nnode);
166 bcopy(target, nnode->tn_link, nnode->tn_size);
167 nnode->tn_link[nnode->tn_size] = '\0';
171 nnode->tn_reg.tn_aobj =
172 swap_pager_alloc(NULL, 0, VM_PROT_DEFAULT, 0);
173 nnode->tn_reg.tn_aobj_pages = 0;
178 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
181 TMPFS_NODE_LOCK(nnode);
183 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
184 tmp->tm_nodes_inuse++;
186 TMPFS_NODE_UNLOCK(nnode);
192 /* --------------------------------------------------------------------- */
195 * Destroys the node pointed to by node from the file system 'tmp'.
196 * If the node does not belong to the given mount point, the results are
199 * If the node references a directory; no entries are allowed because
200 * their removal could need a recursive algorithm, something forbidden in
201 * kernel space. Furthermore, there is not need to provide such
202 * functionality (recursive removal) because the only primitives offered
203 * to the user are the removal of empty directories and the deletion of
206 * Note that nodes are not really deleted; in fact, when a node has been
207 * allocated, it cannot be deleted during the whole life of the file
208 * system. Instead, they are moved to the available list and remain there
212 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
214 vm_pindex_t pages = 0;
217 TMPFS_ASSERT_ELOCKED(node);
218 KKASSERT(node->tn_vnode == NULL);
219 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
223 LIST_REMOVE(node, tn_entries);
224 tmp->tm_nodes_inuse--;
226 TMPFS_NODE_UNLOCK(node);
228 switch (node->tn_type) {
230 /* Do not do anything. VNON is provided to let the
231 * allocation routine clean itself easily by avoiding
232 * duplicating code in it. */
241 * The parent link can be NULL if this is the root
246 KKASSERT(node->tn_dir.tn_parent || node == tmp->tm_root);
247 if (node->tn_dir.tn_parent) {
248 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
249 node->tn_dir.tn_parent->tn_links--;
252 * If the parent directory has no more links and
253 * no vnode ref nothing is going to come along
254 * and clean it up unless we do it here.
256 if (node->tn_dir.tn_parent->tn_links == 0 &&
257 node->tn_dir.tn_parent->tn_vnode == NULL) {
258 tmpfs_free_node(tmp, node->tn_dir.tn_parent);
259 /* eats parent lock */
261 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
263 node->tn_dir.tn_parent = NULL;
267 * If the root node is being destroyed don't leave a
268 * dangling pointer in tmpfs_mount.
270 if (node == tmp->tm_root)
279 kfree(node->tn_link, tmp->tm_name_zone);
280 node->tn_link = NULL;
285 if (node->tn_reg.tn_aobj != NULL)
286 vm_object_deallocate(node->tn_reg.tn_aobj);
287 node->tn_reg.tn_aobj = NULL;
288 pages = node->tn_reg.tn_aobj_pages;
292 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
296 * Clean up fields for the next allocation. The objcache only ctors
299 tmpfs_node_ctor(node, NULL, 0);
300 objcache_put(tmp->tm_node_pool, node);
301 /* node is now invalid */
304 tmp->tm_pages_used -= pages;
308 /* --------------------------------------------------------------------- */
311 * Allocates a new directory entry for the node node with a name of name.
312 * The new directory entry is returned in *de.
314 * The link count of node is increased by one to reflect the new object
317 * Returns zero on success or an appropriate error code on failure.
320 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
321 const char *name, uint16_t len, struct tmpfs_dirent **de)
323 struct tmpfs_dirent *nde;
325 nde = objcache_get(tmp->tm_dirent_pool, M_WAITOK);
326 nde->td_name = kmalloc(len + 1, tmp->tm_name_zone, M_WAITOK | M_NULLOK);
327 if (nde->td_name == NULL) {
328 objcache_put(tmp->tm_dirent_pool, nde);
332 nde->td_namelen = len;
333 bcopy(name, nde->td_name, len);
334 nde->td_name[len] = '\0';
338 TMPFS_NODE_LOCK(node);
340 TMPFS_NODE_UNLOCK(node);
347 /* --------------------------------------------------------------------- */
350 * Frees a directory entry. It is the caller's responsibility to destroy
351 * the node referenced by it if needed.
353 * The link count of node is decreased by one to reflect the removal of an
354 * object that referenced it. This only happens if 'node_exists' is true;
355 * otherwise the function will not access the node referred to by the
356 * directory entry, as it may already have been released from the outside.
359 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
361 struct tmpfs_node *node;
365 TMPFS_NODE_LOCK(node);
366 TMPFS_ASSERT_ELOCKED(node);
367 KKASSERT(node->tn_links > 0);
369 TMPFS_NODE_UNLOCK(node);
371 kfree(de->td_name, tmp->tm_name_zone);
375 objcache_put(tmp->tm_dirent_pool, de);
378 /* --------------------------------------------------------------------- */
381 * Allocates a new vnode for the node node or returns a new reference to
382 * an existing one if the node had already a vnode referencing it. The
383 * resulting locked vnode is returned in *vpp.
385 * Returns zero on success or an appropriate error code on failure.
388 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
396 * Interlocked extraction from node. This can race many things.
397 * We have to get a soft reference on the vnode while we hold
398 * the node locked, then acquire it properly and check for races.
400 TMPFS_NODE_LOCK(node);
401 if ((vp = node->tn_vnode) != NULL) {
402 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
403 vhold_interlocked(vp);
404 TMPFS_NODE_UNLOCK(node);
406 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) {
410 if (node->tn_vnode != vp) {
421 * This should never happen.
423 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) {
424 TMPFS_NODE_UNLOCK(node);
430 * Interlock against other calls to tmpfs_alloc_vp() trying to
431 * allocate and assign a vp to node.
433 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
434 node->tn_vpstate |= TMPFS_VNODE_WANT;
435 error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH,
436 "tmpfs_alloc_vp", 0);
437 TMPFS_NODE_UNLOCK(node);
442 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
443 TMPFS_NODE_UNLOCK(node);
446 * Allocate a new vnode (may block). The ALLOCATING flag should
447 * prevent a race against someone else assigning node->tn_vnode.
449 error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE);
453 KKASSERT(node->tn_vnode == NULL);
454 KKASSERT(vp != NULL);
456 vp->v_type = node->tn_type;
458 /* Type-specific initialization. */
459 switch (node->tn_type) {
467 vinitvmio(vp, node->tn_size, BMASK, -1);
472 vp->v_ops = &mp->mnt_vn_fifo_ops;
478 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
484 TMPFS_NODE_LOCK(node);
486 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
487 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
490 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
491 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
492 TMPFS_NODE_UNLOCK(node);
493 wakeup(&node->tn_vpstate);
495 TMPFS_NODE_UNLOCK(node);
501 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
503 TMPFS_NODE_LOCK(node);
504 KKASSERT(*vpp == node->tn_vnode);
505 TMPFS_NODE_UNLOCK(node);
511 /* --------------------------------------------------------------------- */
514 * Destroys the association between the vnode vp and the node it
518 tmpfs_free_vp(struct vnode *vp)
520 struct tmpfs_node *node;
522 node = VP_TO_TMPFS_NODE(vp);
524 TMPFS_NODE_LOCK(node);
525 KKASSERT(lockcount(TMPFS_NODE_MTX(node)) > 0);
526 node->tn_vnode = NULL;
527 TMPFS_NODE_UNLOCK(node);
531 /* --------------------------------------------------------------------- */
534 * Allocates a new file of type 'type' and adds it to the parent directory
535 * 'dvp'; this addition is done using the component name given in 'cnp'.
536 * The ownership of the new file is automatically assigned based on the
537 * credentials of the caller (through 'cnp'), the group is set based on
538 * the parent directory and the mode is determined from the 'vap' argument.
539 * If successful, *vpp holds a vnode to the newly created file and zero
540 * is returned. Otherwise *vpp is NULL and the function returns an
541 * appropriate error code.
544 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
545 struct namecache *ncp, struct ucred *cred, char *target)
548 struct tmpfs_dirent *de;
549 struct tmpfs_mount *tmp;
550 struct tmpfs_node *dnode;
551 struct tmpfs_node *node;
552 struct tmpfs_node *parent;
554 tmp = VFS_TO_TMPFS(dvp->v_mount);
555 dnode = VP_TO_TMPFS_DIR(dvp);
558 /* If the entry we are creating is a directory, we cannot overflow
559 * the number of links of its parent, because it will get a new
561 if (vap->va_type == VDIR) {
562 /* Ensure that we do not overflow the maximum number of links
563 * imposed by the system. */
564 KKASSERT(dnode->tn_links <= LINK_MAX);
565 if (dnode->tn_links == LINK_MAX) {
570 KKASSERT(parent != NULL);
574 /* Allocate a node that represents the new file. */
575 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid,
576 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rmajor, vap->va_rminor, &node);
579 TMPFS_NODE_LOCK(node);
581 /* Allocate a directory entry that points to the new file. */
582 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de);
584 tmpfs_free_node(tmp, node);
589 /* Allocate a vnode for the new file. */
590 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
592 tmpfs_free_dirent(tmp, de);
593 tmpfs_free_node(tmp, node);
598 /* Now that all required items are allocated, we can proceed to
599 * insert the new node into the directory, an operation that
601 tmpfs_dir_attach(dnode, de);
602 TMPFS_NODE_UNLOCK(node);
607 /* --------------------------------------------------------------------- */
610 * Attaches the directory entry de to the directory represented by vp.
611 * Note that this does not change the link count of the node pointed by
612 * the directory entry, as this is done by tmpfs_alloc_dirent.
615 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
617 TMPFS_NODE_LOCK(dnode);
618 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
620 TMPFS_ASSERT_ELOCKED(dnode);
621 dnode->tn_size += sizeof(struct tmpfs_dirent);
622 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
624 TMPFS_NODE_UNLOCK(dnode);
627 /* --------------------------------------------------------------------- */
630 * Detaches the directory entry de from the directory represented by vp.
631 * Note that this does not change the link count of the node pointed by
632 * the directory entry, as this is done by tmpfs_free_dirent.
635 tmpfs_dir_detach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
637 TMPFS_NODE_LOCK(dnode);
638 if (dnode->tn_dir.tn_readdir_lastp == de) {
639 dnode->tn_dir.tn_readdir_lastn = 0;
640 dnode->tn_dir.tn_readdir_lastp = NULL;
642 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
644 TMPFS_ASSERT_ELOCKED(dnode);
645 dnode->tn_size -= sizeof(struct tmpfs_dirent);
646 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
648 TMPFS_NODE_UNLOCK(dnode);
651 /* --------------------------------------------------------------------- */
654 * Looks for a directory entry in the directory represented by node.
655 * 'ncp' describes the name of the entry to look for. Note that the .
656 * and .. components are not allowed as they do not physically exist
657 * within directories.
659 * Returns a pointer to the entry when found, otherwise NULL.
661 struct tmpfs_dirent *
662 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
663 struct namecache *ncp)
665 struct tmpfs_dirent *de;
666 int len = ncp->nc_nlen;
668 TMPFS_VALIDATE_DIR(node);
670 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
671 if (f != NULL && de->td_node != f)
673 if (len == de->td_namelen) {
674 if (!memcmp(ncp->nc_name, de->td_name, len))
679 TMPFS_NODE_LOCK(node);
680 node->tn_status |= TMPFS_NODE_ACCESSED;
681 TMPFS_NODE_UNLOCK(node);
686 /* --------------------------------------------------------------------- */
689 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
690 * directory and returns it in the uio space. The function returns 0
691 * on success, -1 if there was not enough space in the uio structure to
692 * hold the directory entry or an appropriate error code if another
696 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
702 TMPFS_VALIDATE_DIR(node);
703 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
705 dent.d_ino = node->tn_id;
706 dent.d_type = DT_DIR;
708 dent.d_name[0] = '.';
709 dent.d_name[1] = '\0';
710 dirsize = _DIRENT_DIRSIZ(&dent);
712 if (dirsize > uio->uio_resid)
715 error = uiomove((caddr_t)&dent, dirsize, uio);
717 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
720 TMPFS_NODE_LOCK(node);
721 node->tn_status |= TMPFS_NODE_ACCESSED;
722 TMPFS_NODE_UNLOCK(node);
727 /* --------------------------------------------------------------------- */
730 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
731 * directory and returns it in the uio space. The function returns 0
732 * on success, -1 if there was not enough space in the uio structure to
733 * hold the directory entry or an appropriate error code if another
737 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
744 TMPFS_VALIDATE_DIR(node);
745 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
747 if (node->tn_dir.tn_parent) {
748 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
749 dent.d_ino = node->tn_dir.tn_parent->tn_id;
750 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
752 dent.d_ino = tmp->tm_root->tn_id;
755 dent.d_type = DT_DIR;
757 dent.d_name[0] = '.';
758 dent.d_name[1] = '.';
759 dent.d_name[2] = '\0';
760 dirsize = _DIRENT_DIRSIZ(&dent);
762 if (dirsize > uio->uio_resid)
765 error = uiomove((caddr_t)&dent, dirsize, uio);
767 struct tmpfs_dirent *de;
769 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
771 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
773 uio->uio_offset = tmpfs_dircookie(de);
777 TMPFS_NODE_LOCK(node);
778 node->tn_status |= TMPFS_NODE_ACCESSED;
779 TMPFS_NODE_UNLOCK(node);
784 /* --------------------------------------------------------------------- */
787 * Lookup a directory entry by its associated cookie.
789 struct tmpfs_dirent *
790 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
792 struct tmpfs_dirent *de;
794 if (cookie == node->tn_dir.tn_readdir_lastn &&
795 node->tn_dir.tn_readdir_lastp != NULL) {
796 return node->tn_dir.tn_readdir_lastp;
799 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
800 if (tmpfs_dircookie(de) == cookie) {
808 /* --------------------------------------------------------------------- */
811 * Helper function for tmpfs_readdir. Returns as much directory entries
812 * as can fit in the uio space. The read starts at uio->uio_offset.
813 * The function returns 0 on success, -1 if there was not enough space
814 * in the uio structure to hold the directory entry or an appropriate
815 * error code if another error happens.
818 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
822 struct tmpfs_dirent *de;
824 TMPFS_VALIDATE_DIR(node);
826 /* Locate the first directory entry we have to return. We have cached
827 * the last readdir in the node, so use those values if appropriate.
828 * Otherwise do a linear scan to find the requested entry. */
829 startcookie = uio->uio_offset;
830 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT);
831 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
832 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
835 de = tmpfs_dir_lookupbycookie(node, startcookie);
841 /* Read as much entries as possible; i.e., until we reach the end of
842 * the directory or we exhaust uio space. */
847 /* Create a dirent structure representing the current
848 * tmpfs_node and fill it. */
849 d.d_ino = de->td_node->tn_id;
850 switch (de->td_node->tn_type) {
880 panic("tmpfs_dir_getdents: type %p %d",
881 de->td_node, (int)de->td_node->tn_type);
883 d.d_namlen = de->td_namelen;
884 KKASSERT(de->td_namelen < sizeof(d.d_name));
885 bcopy(de->td_name, d.d_name, d.d_namlen);
886 d.d_name[d.d_namlen] = '\0';
887 reclen = _DIRENT_RECLEN(d.d_namlen);
889 /* Stop reading if the directory entry we are treating is
890 * bigger than the amount of data that can be returned. */
891 if (reclen > uio->uio_resid) {
896 /* Copy the new dirent structure into the output buffer and
897 * advance pointers. */
898 error = uiomove((caddr_t)&d, reclen, uio);
901 de = TAILQ_NEXT(de, td_entries);
902 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
904 /* Update the offset and cache. */
906 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
907 node->tn_dir.tn_readdir_lastn = 0;
908 node->tn_dir.tn_readdir_lastp = NULL;
910 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
911 node->tn_dir.tn_readdir_lastp = de;
913 node->tn_status |= TMPFS_NODE_ACCESSED;
918 /* --------------------------------------------------------------------- */
921 * Resizes the aobj associated to the regular file pointed to by vp to
922 * the size newsize. 'vp' must point to a vnode that represents a regular
923 * file. 'newsize' must be positive.
925 * pass trivial as 1 when buf content will be overwritten, otherwise set 0
928 * Returns zero on success or an appropriate error code on failure.
931 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial)
934 vm_pindex_t newpages, oldpages;
935 struct tmpfs_mount *tmp;
936 struct tmpfs_node *node;
940 KKASSERT(vp->v_type == VREG);
941 KKASSERT(newsize >= 0);
944 node = VP_TO_TMPFS_NODE(vp);
945 tmp = VFS_TO_TMPFS(vp->v_mount);
947 /* Convert the old and new sizes to the number of pages needed to
948 * store them. It may happen that we do not need to do anything
949 * because the last allocated page can accommodate the change on
951 oldsize = node->tn_size;
952 oldpages = round_page64(oldsize) / PAGE_SIZE;
953 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages);
954 newpages = round_page64(newsize) / PAGE_SIZE;
956 if (newpages > oldpages &&
957 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) {
963 tmp->tm_pages_used += (newpages - oldpages);
966 TMPFS_NODE_LOCK(node);
967 node->tn_reg.tn_aobj_pages = newpages;
968 node->tn_size = newsize;
969 TMPFS_NODE_UNLOCK(node);
972 * When adjusting the vnode filesize and its VM object we must
973 * also adjust our backing VM object (aobj). The blocksize
974 * used must match the block sized we use for the buffer cache.
976 * The backing VM object contains no VM pages, only swap
979 if (newsize < oldsize) {
984 error = nvtruncbuf(vp, newsize, BSIZE, -1);
985 aobj = node->tn_reg.tn_aobj;
988 nsize = vp->v_object->size;
991 swap_pager_freespace(aobj, nsize,
998 error = nvextendbuf(vp, oldsize, newsize, BSIZE, BSIZE,
1000 aobj = node->tn_reg.tn_aobj;
1002 aobj->size = vp->v_object->size;
1009 /* --------------------------------------------------------------------- */
1012 * Change flags of the given vnode.
1013 * Caller should execute tmpfs_update on vp after a successful execution.
1014 * The vnode must be locked on entry and remain locked on exit.
1017 tmpfs_chflags(struct vnode *vp, int vaflags, struct ucred *cred)
1020 struct tmpfs_node *node;
1023 KKASSERT(vn_islocked(vp));
1025 node = VP_TO_TMPFS_NODE(vp);
1026 flags = node->tn_flags;
1028 /* Disallow this operation if the file system is mounted read-only. */
1029 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1031 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred);
1034 * Unprivileged processes are not permitted to unset system
1035 * flags, or modify flags if any system flags are set.
1037 * Silently enforce SF_NOCACHE on the root tmpfs vnode so
1038 * tmpfs data is not double-cached by swapcache.
1041 TMPFS_NODE_LOCK(node);
1042 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1043 if (vp->v_flag & VROOT)
1044 flags |= SF_NOCACHE;
1045 node->tn_flags = flags;
1047 if (node->tn_flags & (SF_NOUNLINK | SF_IMMUTABLE |
1049 (flags & UF_SETTABLE) != flags) {
1052 node->tn_flags &= SF_SETTABLE;
1053 node->tn_flags |= (flags & UF_SETTABLE);
1056 node->tn_status |= TMPFS_NODE_CHANGED;
1057 TMPFS_NODE_UNLOCK(node);
1060 KKASSERT(vn_islocked(vp));
1065 /* --------------------------------------------------------------------- */
1068 * Change access mode on the given vnode.
1069 * Caller should execute tmpfs_update on vp after a successful execution.
1070 * The vnode must be locked on entry and remain locked on exit.
1073 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred)
1075 struct tmpfs_node *node;
1079 KKASSERT(vn_islocked(vp));
1081 node = VP_TO_TMPFS_NODE(vp);
1083 /* Disallow this operation if the file system is mounted read-only. */
1084 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1087 /* Immutable or append-only files cannot be modified, either. */
1088 if (node->tn_flags & (IMMUTABLE | APPEND))
1091 cur_mode = node->tn_mode;
1092 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid,
1096 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) {
1097 TMPFS_NODE_LOCK(node);
1098 node->tn_mode &= ~ALLPERMS;
1099 node->tn_mode |= cur_mode & ALLPERMS;
1101 node->tn_status |= TMPFS_NODE_CHANGED;
1102 TMPFS_NODE_UNLOCK(node);
1105 KKASSERT(vn_islocked(vp));
1110 /* --------------------------------------------------------------------- */
1113 * Change ownership of the given vnode. At least one of uid or gid must
1114 * be different than VNOVAL. If one is set to that value, the attribute
1116 * Caller should execute tmpfs_update on vp after a successful execution.
1117 * The vnode must be locked on entry and remain locked on exit.
1120 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
1125 struct tmpfs_node *node;
1128 KKASSERT(vn_islocked(vp));
1129 node = VP_TO_TMPFS_NODE(vp);
1131 /* Disallow this operation if the file system is mounted read-only. */
1132 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1135 /* Immutable or append-only files cannot be modified, either. */
1136 if (node->tn_flags & (IMMUTABLE | APPEND))
1139 cur_uid = node->tn_uid;
1140 cur_gid = node->tn_gid;
1141 cur_mode = node->tn_mode;
1142 error = vop_helper_chown(vp, uid, gid, cred,
1143 &cur_uid, &cur_gid, &cur_mode);
1146 TMPFS_NODE_LOCK(node);
1147 if (cur_uid != node->tn_uid ||
1148 cur_gid != node->tn_gid ||
1149 cur_mode != node->tn_mode) {
1152 node->tn_mode = cur_mode;
1153 node->tn_status |= TMPFS_NODE_CHANGED;
1155 TMPFS_NODE_UNLOCK(node);
1161 /* --------------------------------------------------------------------- */
1164 * Change size of the given vnode.
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_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred)
1172 struct tmpfs_node *node;
1174 KKASSERT(vn_islocked(vp));
1176 node = VP_TO_TMPFS_NODE(vp);
1178 /* Decide whether this is a valid operation based on the file type. */
1180 switch (vp->v_type) {
1185 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1194 /* Allow modifications of special files even if in the file
1195 * system is mounted read-only (we are not modifying the
1196 * files themselves, but the objects they represent). */
1200 /* Anything else is unsupported. */
1204 /* Immutable or append-only files cannot be modified, either. */
1205 if (node->tn_flags & (IMMUTABLE | APPEND))
1208 error = tmpfs_truncate(vp, size);
1209 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1210 * for us, as will update tn_status; no need to do that here. */
1212 KKASSERT(vn_islocked(vp));
1217 /* --------------------------------------------------------------------- */
1220 * Change access and modification times of the given vnode.
1221 * Caller should execute tmpfs_update on vp after a successful execution.
1222 * The vnode must be locked on entry and remain locked on exit.
1225 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1226 int vaflags, struct ucred *cred)
1228 struct tmpfs_node *node;
1230 KKASSERT(vn_islocked(vp));
1232 node = VP_TO_TMPFS_NODE(vp);
1234 /* Disallow this operation if the file system is mounted read-only. */
1235 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1238 /* Immutable or append-only files cannot be modified, either. */
1239 if (node->tn_flags & (IMMUTABLE | APPEND))
1242 TMPFS_NODE_LOCK(node);
1243 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1244 node->tn_status |= TMPFS_NODE_ACCESSED;
1246 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1247 node->tn_status |= TMPFS_NODE_MODIFIED;
1249 TMPFS_NODE_UNLOCK(node);
1251 tmpfs_itimes(vp, atime, mtime);
1253 KKASSERT(vn_islocked(vp));
1258 /* --------------------------------------------------------------------- */
1259 /* Sync timestamps */
1261 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1262 const struct timespec *mod)
1264 struct tmpfs_node *node;
1265 struct timespec now;
1267 node = VP_TO_TMPFS_NODE(vp);
1269 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1270 TMPFS_NODE_CHANGED)) == 0)
1273 vfs_timestamp(&now);
1275 TMPFS_NODE_LOCK(node);
1276 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1279 node->tn_atime = acc->tv_sec;
1280 node->tn_atimensec = acc->tv_nsec;
1282 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1285 node->tn_mtime = mod->tv_sec;
1286 node->tn_mtimensec = mod->tv_nsec;
1288 if (node->tn_status & TMPFS_NODE_CHANGED) {
1289 node->tn_ctime = now.tv_sec;
1290 node->tn_ctimensec = now.tv_nsec;
1293 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1294 TMPFS_NODE_UNLOCK(node);
1297 /* --------------------------------------------------------------------- */
1300 tmpfs_update(struct vnode *vp)
1303 tmpfs_itimes(vp, NULL, NULL);
1306 /* --------------------------------------------------------------------- */
1309 tmpfs_truncate(struct vnode *vp, off_t length)
1312 struct tmpfs_node *node;
1314 node = VP_TO_TMPFS_NODE(vp);
1321 if (node->tn_size == length) {
1326 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1330 error = tmpfs_reg_resize(vp, length, 1);
1333 TMPFS_NODE_LOCK(node);
1334 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1335 TMPFS_NODE_UNLOCK(node);
1344 /* --------------------------------------------------------------------- */
1347 tmpfs_fetch_ino(void)
1351 spin_lock(&ino_lock);
1353 spin_unlock(&ino_lock);