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>
42 #include <sys/spinlock2.h>
44 #include <sys/systm.h>
45 #include <sys/vnode.h>
46 #include <sys/vmmeter.h>
49 #include <vm/vm_object.h>
50 #include <vm/vm_page.h>
51 #include <vm/vm_pager.h>
52 #include <vm/vm_extern.h>
54 #include <vfs/tmpfs/tmpfs.h>
55 #include <vfs/tmpfs/tmpfs_fifoops.h>
56 #include <vfs/tmpfs/tmpfs_vnops.h>
58 static ino_t t_ino = 2;
59 static struct spinlock ino_lock;
60 static ino_t tmpfs_fetch_ino(void);
62 /* --------------------------------------------------------------------- */
65 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
66 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
67 * using the credentials of the process 'p'.
69 * If the node type is set to 'VDIR', then the parent parameter must point
70 * to the parent directory of the node being created. It may only be NULL
71 * while allocating the root node.
73 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
74 * specifies the device the node represents.
76 * If the node type is set to 'VLNK', then the parameter target specifies
77 * the file name of the target file for the symbolic link that is being
80 * Note that new nodes are retrieved from the available list if it has
81 * items or, if it is empty, from the node pool as long as there is enough
82 * space to create them.
84 * Returns zero on success or an appropriate error code on failure.
87 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
88 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
89 char *target, int rmajor, int rminor, struct tmpfs_node **node)
91 struct tmpfs_node *nnode;
95 /* If the root directory of the 'tmp' file system is not yet
96 * allocated, this must be the request to do it. */
97 KKASSERT(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
99 KKASSERT(IFF(type == VLNK, target != NULL));
100 KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL));
102 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
105 nnode = objcache_get(tmp->tm_node_pool, M_WAITOK | M_NULLOK);
109 /* Generic initialization. */
110 nnode->tn_type = type;
112 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime
114 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec
118 nnode->tn_mode = mode;
119 nnode->tn_id = tmpfs_fetch_ino();
120 nnode->tn_advlock.init_done = 0;
122 /* Type-specific initialization. */
123 switch (nnode->tn_type) {
126 rdev = makeudev(rmajor, rminor);
127 if (rdev == NOUDEV) {
128 objcache_put(tmp->tm_node_pool, nnode);
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, tmp->tm_name_zone,
158 M_WAITOK | M_NULLOK);
159 if (nnode->tn_link == NULL) {
160 objcache_put(tmp->tm_node_pool, nnode);
163 bcopy(target, nnode->tn_link, nnode->tn_size);
164 nnode->tn_link[nnode->tn_size] = '\0';
168 nnode->tn_reg.tn_aobj =
169 swap_pager_alloc(NULL, 0, VM_PROT_DEFAULT, 0);
170 nnode->tn_reg.tn_aobj_pages = 0;
175 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
178 TMPFS_NODE_LOCK(nnode);
180 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
181 tmp->tm_nodes_inuse++;
183 TMPFS_NODE_UNLOCK(nnode);
189 /* --------------------------------------------------------------------- */
192 * Destroys the node pointed to by node from the file system 'tmp'.
193 * If the node does not belong to the given mount point, the results are
196 * If the node references a directory; no entries are allowed because
197 * their removal could need a recursive algorithm, something forbidden in
198 * kernel space. Furthermore, there is not need to provide such
199 * functionality (recursive removal) because the only primitives offered
200 * to the user are the removal of empty directories and the deletion of
203 * Note that nodes are not really deleted; in fact, when a node has been
204 * allocated, it cannot be deleted during the whole life of the file
205 * system. Instead, they are moved to the available list and remain there
209 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
211 vm_pindex_t pages = 0;
214 TMPFS_ASSERT_ELOCKED(node);
215 KKASSERT(node->tn_vnode == NULL);
216 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
220 LIST_REMOVE(node, tn_entries);
221 tmp->tm_nodes_inuse--;
223 TMPFS_NODE_UNLOCK(node);
225 switch (node->tn_type) {
227 /* Do not do anything. VNON is provided to let the
228 * allocation routine clean itself easily by avoiding
229 * duplicating code in it. */
238 * The parent link can be NULL if this is the root
243 KKASSERT(node->tn_dir.tn_parent || node == tmp->tm_root);
244 if (node->tn_dir.tn_parent) {
245 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
246 node->tn_dir.tn_parent->tn_links--;
249 * If the parent directory has no more links and
250 * no vnode ref nothing is going to come along
251 * and clean it up unless we do it here.
253 if (node->tn_dir.tn_parent->tn_links == 0 &&
254 node->tn_dir.tn_parent->tn_vnode == NULL) {
255 tmpfs_free_node(tmp, node->tn_dir.tn_parent);
256 /* eats parent lock */
258 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
260 node->tn_dir.tn_parent = NULL;
264 * If the root node is being destroyed don't leave a
265 * dangling pointer in tmpfs_mount.
267 if (node == tmp->tm_root)
276 kfree(node->tn_link, tmp->tm_name_zone);
277 node->tn_link = NULL;
282 if (node->tn_reg.tn_aobj != NULL)
283 vm_object_deallocate(node->tn_reg.tn_aobj);
284 node->tn_reg.tn_aobj = NULL;
285 pages = node->tn_reg.tn_aobj_pages;
289 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
293 * Clean up fields for the next allocation. The objcache only ctors
296 tmpfs_node_ctor(node, NULL, 0);
297 objcache_put(tmp->tm_node_pool, node);
298 /* node is now invalid */
301 tmp->tm_pages_used -= pages;
305 /* --------------------------------------------------------------------- */
308 * Allocates a new directory entry for the node node with a name of name.
309 * The new directory entry is returned in *de.
311 * The link count of node is increased by one to reflect the new object
314 * Returns zero on success or an appropriate error code on failure.
317 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
318 const char *name, uint16_t len, struct tmpfs_dirent **de)
320 struct tmpfs_dirent *nde;
322 nde = objcache_get(tmp->tm_dirent_pool, M_WAITOK);
323 nde->td_name = kmalloc(len + 1, tmp->tm_name_zone, M_WAITOK | M_NULLOK);
324 if (nde->td_name == NULL) {
325 objcache_put(tmp->tm_dirent_pool, nde);
329 nde->td_namelen = len;
330 bcopy(name, nde->td_name, len);
331 nde->td_name[len] = '\0';
335 TMPFS_NODE_LOCK(node);
337 TMPFS_NODE_UNLOCK(node);
344 /* --------------------------------------------------------------------- */
347 * Frees a directory entry. It is the caller's responsibility to destroy
348 * the node referenced by it if needed.
350 * The link count of node is decreased by one to reflect the removal of an
351 * object that referenced it. This only happens if 'node_exists' is true;
352 * otherwise the function will not access the node referred to by the
353 * directory entry, as it may already have been released from the outside.
356 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
358 struct tmpfs_node *node;
362 TMPFS_NODE_LOCK(node);
363 TMPFS_ASSERT_ELOCKED(node);
364 KKASSERT(node->tn_links > 0);
366 TMPFS_NODE_UNLOCK(node);
368 kfree(de->td_name, tmp->tm_name_zone);
372 objcache_put(tmp->tm_dirent_pool, de);
375 /* --------------------------------------------------------------------- */
378 * Allocates a new vnode for the node node or returns a new reference to
379 * an existing one if the node had already a vnode referencing it. The
380 * resulting locked vnode is returned in *vpp.
382 * Returns zero on success or an appropriate error code on failure.
385 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
393 * Interlocked extraction from node. This can race many things.
394 * We have to get a soft reference on the vnode while we hold
395 * the node locked, then acquire it properly and check for races.
397 TMPFS_NODE_LOCK(node);
398 if ((vp = node->tn_vnode) != NULL) {
399 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
400 vhold_interlocked(vp);
401 TMPFS_NODE_UNLOCK(node);
403 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) {
407 if (node->tn_vnode != vp) {
418 * This should never happen.
420 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) {
421 TMPFS_NODE_UNLOCK(node);
427 * Interlock against other calls to tmpfs_alloc_vp() trying to
428 * allocate and assign a vp to node.
430 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
431 node->tn_vpstate |= TMPFS_VNODE_WANT;
432 error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH,
433 "tmpfs_alloc_vp", 0);
434 TMPFS_NODE_UNLOCK(node);
439 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
440 TMPFS_NODE_UNLOCK(node);
443 * Allocate a new vnode (may block). The ALLOCATING flag should
444 * prevent a race against someone else assigning node->tn_vnode.
446 error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE);
450 KKASSERT(node->tn_vnode == NULL);
451 KKASSERT(vp != NULL);
453 vp->v_type = node->tn_type;
455 /* Type-specific initialization. */
456 switch (node->tn_type) {
464 vinitvmio(vp, node->tn_size, BMASK, -1);
469 vp->v_ops = &mp->mnt_vn_fifo_ops;
475 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
481 TMPFS_NODE_LOCK(node);
483 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
484 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
487 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
488 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
489 TMPFS_NODE_UNLOCK(node);
490 wakeup(&node->tn_vpstate);
492 TMPFS_NODE_UNLOCK(node);
498 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
500 TMPFS_NODE_LOCK(node);
501 KKASSERT(*vpp == node->tn_vnode);
502 TMPFS_NODE_UNLOCK(node);
508 /* --------------------------------------------------------------------- */
511 * Destroys the association between the vnode vp and the node it
515 tmpfs_free_vp(struct vnode *vp)
517 struct tmpfs_node *node;
519 node = VP_TO_TMPFS_NODE(vp);
521 TMPFS_NODE_LOCK(node);
522 KKASSERT(lockcount(TMPFS_NODE_MTX(node)) > 0);
523 node->tn_vnode = NULL;
524 TMPFS_NODE_UNLOCK(node);
528 /* --------------------------------------------------------------------- */
531 * Allocates a new file of type 'type' and adds it to the parent directory
532 * 'dvp'; this addition is done using the component name given in 'cnp'.
533 * The ownership of the new file is automatically assigned based on the
534 * credentials of the caller (through 'cnp'), the group is set based on
535 * the parent directory and the mode is determined from the 'vap' argument.
536 * If successful, *vpp holds a vnode to the newly created file and zero
537 * is returned. Otherwise *vpp is NULL and the function returns an
538 * appropriate error code.
541 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
542 struct namecache *ncp, struct ucred *cred, char *target)
545 struct tmpfs_dirent *de;
546 struct tmpfs_mount *tmp;
547 struct tmpfs_node *dnode;
548 struct tmpfs_node *node;
549 struct tmpfs_node *parent;
551 tmp = VFS_TO_TMPFS(dvp->v_mount);
552 dnode = VP_TO_TMPFS_DIR(dvp);
555 /* If the entry we are creating is a directory, we cannot overflow
556 * the number of links of its parent, because it will get a new
558 if (vap->va_type == VDIR) {
559 /* Ensure that we do not overflow the maximum number of links
560 * imposed by the system. */
561 KKASSERT(dnode->tn_links <= LINK_MAX);
562 if (dnode->tn_links == LINK_MAX) {
567 KKASSERT(parent != NULL);
571 /* Allocate a node that represents the new file. */
572 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid,
573 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rmajor, vap->va_rminor, &node);
576 TMPFS_NODE_LOCK(node);
578 /* Allocate a directory entry that points to the new file. */
579 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de);
581 tmpfs_free_node(tmp, node);
586 /* Allocate a vnode for the new file. */
587 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
589 tmpfs_free_dirent(tmp, de);
590 tmpfs_free_node(tmp, node);
595 /* Now that all required items are allocated, we can proceed to
596 * insert the new node into the directory, an operation that
598 tmpfs_dir_attach(dnode, de);
599 TMPFS_NODE_UNLOCK(node);
604 /* --------------------------------------------------------------------- */
607 * Attaches the directory entry de to the directory represented by vp.
608 * Note that this does not change the link count of the node pointed by
609 * the directory entry, as this is done by tmpfs_alloc_dirent.
612 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
614 TMPFS_NODE_LOCK(dnode);
615 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
617 TMPFS_ASSERT_ELOCKED(dnode);
618 dnode->tn_size += sizeof(struct tmpfs_dirent);
619 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
621 TMPFS_NODE_UNLOCK(dnode);
624 /* --------------------------------------------------------------------- */
627 * Detaches the directory entry de from the directory represented by vp.
628 * Note that this does not change the link count of the node pointed by
629 * the directory entry, as this is done by tmpfs_free_dirent.
632 tmpfs_dir_detach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
634 TMPFS_NODE_LOCK(dnode);
635 if (dnode->tn_dir.tn_readdir_lastp == de) {
636 dnode->tn_dir.tn_readdir_lastn = 0;
637 dnode->tn_dir.tn_readdir_lastp = NULL;
639 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
641 TMPFS_ASSERT_ELOCKED(dnode);
642 dnode->tn_size -= sizeof(struct tmpfs_dirent);
643 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
645 TMPFS_NODE_UNLOCK(dnode);
648 /* --------------------------------------------------------------------- */
651 * Looks for a directory entry in the directory represented by node.
652 * 'ncp' describes the name of the entry to look for. Note that the .
653 * and .. components are not allowed as they do not physically exist
654 * within directories.
656 * Returns a pointer to the entry when found, otherwise NULL.
658 struct tmpfs_dirent *
659 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
660 struct namecache *ncp)
662 struct tmpfs_dirent *de;
663 int len = ncp->nc_nlen;
665 TMPFS_VALIDATE_DIR(node);
667 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
668 if (f != NULL && de->td_node != f)
670 if (len == de->td_namelen) {
671 if (!memcmp(ncp->nc_name, de->td_name, len))
676 TMPFS_NODE_LOCK(node);
677 node->tn_status |= TMPFS_NODE_ACCESSED;
678 TMPFS_NODE_UNLOCK(node);
683 /* --------------------------------------------------------------------- */
686 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
687 * directory and returns it in the uio space. The function returns 0
688 * on success, -1 if there was not enough space in the uio structure to
689 * hold the directory entry or an appropriate error code if another
693 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
699 TMPFS_VALIDATE_DIR(node);
700 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
702 dent.d_ino = node->tn_id;
703 dent.d_type = DT_DIR;
705 dent.d_name[0] = '.';
706 dent.d_name[1] = '\0';
707 dirsize = _DIRENT_DIRSIZ(&dent);
709 if (dirsize > uio->uio_resid)
712 error = uiomove((caddr_t)&dent, dirsize, uio);
714 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
717 TMPFS_NODE_LOCK(node);
718 node->tn_status |= TMPFS_NODE_ACCESSED;
719 TMPFS_NODE_UNLOCK(node);
724 /* --------------------------------------------------------------------- */
727 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
728 * directory and returns it in the uio space. The function returns 0
729 * on success, -1 if there was not enough space in the uio structure to
730 * hold the directory entry or an appropriate error code if another
734 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->tn_dir.tn_parent);
746 dent.d_ino = node->tn_dir.tn_parent->tn_id;
747 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
749 dent.d_ino = tmp->tm_root->tn_id;
752 dent.d_type = DT_DIR;
754 dent.d_name[0] = '.';
755 dent.d_name[1] = '.';
756 dent.d_name[2] = '\0';
757 dirsize = _DIRENT_DIRSIZ(&dent);
759 if (dirsize > uio->uio_resid)
762 error = uiomove((caddr_t)&dent, dirsize, uio);
764 struct tmpfs_dirent *de;
766 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
768 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
770 uio->uio_offset = tmpfs_dircookie(de);
774 TMPFS_NODE_LOCK(node);
775 node->tn_status |= TMPFS_NODE_ACCESSED;
776 TMPFS_NODE_UNLOCK(node);
781 /* --------------------------------------------------------------------- */
784 * Lookup a directory entry by its associated cookie.
786 struct tmpfs_dirent *
787 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
789 struct tmpfs_dirent *de;
791 if (cookie == node->tn_dir.tn_readdir_lastn &&
792 node->tn_dir.tn_readdir_lastp != NULL) {
793 return node->tn_dir.tn_readdir_lastp;
796 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
797 if (tmpfs_dircookie(de) == cookie) {
805 /* --------------------------------------------------------------------- */
808 * Helper function for tmpfs_readdir. Returns as much directory entries
809 * as can fit in the uio space. The read starts at uio->uio_offset.
810 * The function returns 0 on success, -1 if there was not enough space
811 * in the uio structure to hold the directory entry or an appropriate
812 * error code if another error happens.
815 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
819 struct tmpfs_dirent *de;
821 TMPFS_VALIDATE_DIR(node);
823 /* Locate the first directory entry we have to return. We have cached
824 * the last readdir in the node, so use those values if appropriate.
825 * Otherwise do a linear scan to find the requested entry. */
826 startcookie = uio->uio_offset;
827 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT);
828 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
829 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
832 de = tmpfs_dir_lookupbycookie(node, startcookie);
838 /* Read as much entries as possible; i.e., until we reach the end of
839 * the directory or we exhaust uio space. */
844 /* Create a dirent structure representing the current
845 * tmpfs_node and fill it. */
846 d.d_ino = de->td_node->tn_id;
847 switch (de->td_node->tn_type) {
877 panic("tmpfs_dir_getdents: type %p %d",
878 de->td_node, (int)de->td_node->tn_type);
880 d.d_namlen = de->td_namelen;
881 KKASSERT(de->td_namelen < sizeof(d.d_name));
882 bcopy(de->td_name, d.d_name, d.d_namlen);
883 d.d_name[d.d_namlen] = '\0';
884 reclen = _DIRENT_RECLEN(d.d_namlen);
886 /* Stop reading if the directory entry we are treating is
887 * bigger than the amount of data that can be returned. */
888 if (reclen > uio->uio_resid) {
893 /* Copy the new dirent structure into the output buffer and
894 * advance pointers. */
895 error = uiomove((caddr_t)&d, reclen, uio);
898 de = TAILQ_NEXT(de, td_entries);
899 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
901 /* Update the offset and cache. */
903 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
904 node->tn_dir.tn_readdir_lastn = 0;
905 node->tn_dir.tn_readdir_lastp = NULL;
907 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
908 node->tn_dir.tn_readdir_lastp = de;
910 node->tn_status |= TMPFS_NODE_ACCESSED;
915 /* --------------------------------------------------------------------- */
918 * Resizes the aobj associated to the regular file pointed to by vp to
919 * the size newsize. 'vp' must point to a vnode that represents a regular
920 * file. 'newsize' must be positive.
922 * pass trivial as 1 when buf content will be overwritten, otherwise set 0
925 * Returns zero on success or an appropriate error code on failure.
928 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial)
931 vm_pindex_t newpages, oldpages;
932 struct tmpfs_mount *tmp;
933 struct tmpfs_node *node;
937 KKASSERT(vp->v_type == VREG);
938 KKASSERT(newsize >= 0);
941 node = VP_TO_TMPFS_NODE(vp);
942 tmp = VFS_TO_TMPFS(vp->v_mount);
944 /* Convert the old and new sizes to the number of pages needed to
945 * store them. It may happen that we do not need to do anything
946 * because the last allocated page can accommodate the change on
948 oldsize = node->tn_size;
949 oldpages = round_page64(oldsize) / PAGE_SIZE;
950 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages);
951 newpages = round_page64(newsize) / PAGE_SIZE;
953 if (newpages > oldpages &&
954 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) {
960 tmp->tm_pages_used += (newpages - oldpages);
963 TMPFS_NODE_LOCK(node);
964 node->tn_reg.tn_aobj_pages = newpages;
965 node->tn_size = newsize;
966 TMPFS_NODE_UNLOCK(node);
969 * When adjusting the vnode filesize and its VM object we must
970 * also adjust our backing VM object (aobj). The blocksize
971 * used must match the block sized we use for the buffer cache.
973 * The backing VM object contains no VM pages, only swap
976 if (newsize < oldsize) {
981 error = nvtruncbuf(vp, newsize, BSIZE, -1);
982 aobj = node->tn_reg.tn_aobj;
985 nsize = vp->v_object->size;
988 swap_pager_freespace(aobj, nsize,
995 error = nvextendbuf(vp, oldsize, newsize, BSIZE, BSIZE,
997 aobj = node->tn_reg.tn_aobj;
999 aobj->size = vp->v_object->size;
1006 /* --------------------------------------------------------------------- */
1009 * Change flags of the given vnode.
1010 * Caller should execute tmpfs_update on vp after a successful execution.
1011 * The vnode must be locked on entry and remain locked on exit.
1014 tmpfs_chflags(struct vnode *vp, int vaflags, struct ucred *cred)
1017 struct tmpfs_node *node;
1020 KKASSERT(vn_islocked(vp));
1022 node = VP_TO_TMPFS_NODE(vp);
1023 flags = node->tn_flags;
1025 /* Disallow this operation if the file system is mounted read-only. */
1026 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1028 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred);
1031 * Unprivileged processes are not permitted to unset system
1032 * flags, or modify flags if any system flags are set.
1034 * Silently enforce SF_NOCACHE on the root tmpfs vnode so
1035 * tmpfs data is not double-cached by swapcache.
1038 TMPFS_NODE_LOCK(node);
1039 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1040 if (vp->v_flag & VROOT)
1041 flags |= SF_NOCACHE;
1042 node->tn_flags = flags;
1044 if (node->tn_flags & (SF_NOUNLINK | SF_IMMUTABLE |
1046 (flags & UF_SETTABLE) != flags) {
1049 node->tn_flags &= SF_SETTABLE;
1050 node->tn_flags |= (flags & UF_SETTABLE);
1053 node->tn_status |= TMPFS_NODE_CHANGED;
1054 TMPFS_NODE_UNLOCK(node);
1057 KKASSERT(vn_islocked(vp));
1062 /* --------------------------------------------------------------------- */
1065 * Change access mode on the given vnode.
1066 * Caller should execute tmpfs_update on vp after a successful execution.
1067 * The vnode must be locked on entry and remain locked on exit.
1070 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred)
1072 struct tmpfs_node *node;
1076 KKASSERT(vn_islocked(vp));
1078 node = VP_TO_TMPFS_NODE(vp);
1080 /* Disallow this operation if the file system is mounted read-only. */
1081 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1084 /* Immutable or append-only files cannot be modified, either. */
1085 if (node->tn_flags & (IMMUTABLE | APPEND))
1088 cur_mode = node->tn_mode;
1089 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid,
1093 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) {
1094 TMPFS_NODE_LOCK(node);
1095 node->tn_mode &= ~ALLPERMS;
1096 node->tn_mode |= cur_mode & ALLPERMS;
1098 node->tn_status |= TMPFS_NODE_CHANGED;
1099 TMPFS_NODE_UNLOCK(node);
1102 KKASSERT(vn_islocked(vp));
1107 /* --------------------------------------------------------------------- */
1110 * Change ownership of the given vnode. At least one of uid or gid must
1111 * be different than VNOVAL. If one is set to that value, the attribute
1113 * Caller should execute tmpfs_update on vp after a successful execution.
1114 * The vnode must be locked on entry and remain locked on exit.
1117 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
1122 struct tmpfs_node *node;
1125 KKASSERT(vn_islocked(vp));
1126 node = VP_TO_TMPFS_NODE(vp);
1128 /* Disallow this operation if the file system is mounted read-only. */
1129 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1132 /* Immutable or append-only files cannot be modified, either. */
1133 if (node->tn_flags & (IMMUTABLE | APPEND))
1136 cur_uid = node->tn_uid;
1137 cur_gid = node->tn_gid;
1138 cur_mode = node->tn_mode;
1139 error = vop_helper_chown(vp, uid, gid, cred,
1140 &cur_uid, &cur_gid, &cur_mode);
1143 TMPFS_NODE_LOCK(node);
1144 if (cur_uid != node->tn_uid ||
1145 cur_gid != node->tn_gid ||
1146 cur_mode != node->tn_mode) {
1147 node->tn_uid = cur_uid;
1148 node->tn_gid = cur_gid;
1149 node->tn_mode = cur_mode;
1150 node->tn_status |= TMPFS_NODE_CHANGED;
1152 TMPFS_NODE_UNLOCK(node);
1158 /* --------------------------------------------------------------------- */
1161 * Change size of the given vnode.
1162 * Caller should execute tmpfs_update on vp after a successful execution.
1163 * The vnode must be locked on entry and remain locked on exit.
1166 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred)
1169 struct tmpfs_node *node;
1171 KKASSERT(vn_islocked(vp));
1173 node = VP_TO_TMPFS_NODE(vp);
1175 /* Decide whether this is a valid operation based on the file type. */
1177 switch (vp->v_type) {
1182 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1191 /* Allow modifications of special files even if in the file
1192 * system is mounted read-only (we are not modifying the
1193 * files themselves, but the objects they represent). */
1197 /* Anything else is unsupported. */
1201 /* Immutable or append-only files cannot be modified, either. */
1202 if (node->tn_flags & (IMMUTABLE | APPEND))
1205 error = tmpfs_truncate(vp, size);
1206 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1207 * for us, as will update tn_status; no need to do that here. */
1209 KKASSERT(vn_islocked(vp));
1214 /* --------------------------------------------------------------------- */
1217 * Change access and modification times of the given vnode.
1218 * Caller should execute tmpfs_update on vp after a successful execution.
1219 * The vnode must be locked on entry and remain locked on exit.
1222 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1223 int vaflags, struct ucred *cred)
1225 struct tmpfs_node *node;
1227 KKASSERT(vn_islocked(vp));
1229 node = VP_TO_TMPFS_NODE(vp);
1231 /* Disallow this operation if the file system is mounted read-only. */
1232 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1235 /* Immutable or append-only files cannot be modified, either. */
1236 if (node->tn_flags & (IMMUTABLE | APPEND))
1239 TMPFS_NODE_LOCK(node);
1240 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1241 node->tn_status |= TMPFS_NODE_ACCESSED;
1243 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1244 node->tn_status |= TMPFS_NODE_MODIFIED;
1246 TMPFS_NODE_UNLOCK(node);
1248 tmpfs_itimes(vp, atime, mtime);
1250 KKASSERT(vn_islocked(vp));
1255 /* --------------------------------------------------------------------- */
1256 /* Sync timestamps */
1258 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1259 const struct timespec *mod)
1261 struct tmpfs_node *node;
1262 struct timespec now;
1264 node = VP_TO_TMPFS_NODE(vp);
1266 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1267 TMPFS_NODE_CHANGED)) == 0)
1270 vfs_timestamp(&now);
1272 TMPFS_NODE_LOCK(node);
1273 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1276 node->tn_atime = acc->tv_sec;
1277 node->tn_atimensec = acc->tv_nsec;
1279 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1282 node->tn_mtime = mod->tv_sec;
1283 node->tn_mtimensec = mod->tv_nsec;
1285 if (node->tn_status & TMPFS_NODE_CHANGED) {
1286 node->tn_ctime = now.tv_sec;
1287 node->tn_ctimensec = now.tv_nsec;
1290 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1291 TMPFS_NODE_UNLOCK(node);
1294 /* --------------------------------------------------------------------- */
1297 tmpfs_update(struct vnode *vp)
1300 tmpfs_itimes(vp, NULL, NULL);
1303 /* --------------------------------------------------------------------- */
1306 tmpfs_truncate(struct vnode *vp, off_t length)
1309 struct tmpfs_node *node;
1311 node = VP_TO_TMPFS_NODE(vp);
1318 if (node->tn_size == length) {
1323 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1327 error = tmpfs_reg_resize(vp, length, 1);
1330 TMPFS_NODE_LOCK(node);
1331 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1332 TMPFS_NODE_UNLOCK(node);
1341 /* --------------------------------------------------------------------- */
1344 tmpfs_fetch_ino(void)
1348 spin_lock(&ino_lock);
1350 spin_unlock(&ino_lock);