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();
122 /* Type-specific initialization. */
123 switch (nnode->tn_type) {
126 rdev = makeudev(rmajor, rminor);
127 if (rdev == NOUDEV) {
130 nnode->tn_rdev = rdev;
134 TAILQ_INIT(&nnode->tn_dir.tn_dirhead);
135 KKASSERT(parent != nnode);
136 KKASSERT(IMPLIES(parent == NULL, tmp->tm_root == NULL));
137 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
138 nnode->tn_dir.tn_readdir_lastn = 0;
139 nnode->tn_dir.tn_readdir_lastp = NULL;
142 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
143 nnode->tn_dir.tn_parent->tn_links++;
144 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
153 KKASSERT((strlen(target) +1) < MAXPATHLEN);
154 nnode->tn_size = strlen(target) +1;
155 nnode->tn_link = kmalloc(nnode->tn_size, M_TMPFSNAME,
157 bcopy(target, nnode->tn_link, nnode->tn_size);
158 nnode->tn_link[nnode->tn_size] = '\0';
162 nnode->tn_reg.tn_aobj =
163 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0);
164 nnode->tn_reg.tn_aobj_pages = 0;
169 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
172 TMPFS_NODE_LOCK(nnode);
173 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
175 tmp->tm_nodes_inuse++;
177 TMPFS_NODE_UNLOCK(nnode);
183 /* --------------------------------------------------------------------- */
186 * Destroys the node pointed to by node from the file system 'tmp'.
187 * If the node does not belong to the given mount point, the results are
190 * If the node references a directory; no entries are allowed because
191 * their removal could need a recursive algorithm, something forbidden in
192 * kernel space. Furthermore, there is not need to provide such
193 * functionality (recursive removal) because the only primitives offered
194 * to the user are the removal of empty directories and the deletion of
197 * Note that nodes are not really deleted; in fact, when a node has been
198 * allocated, it cannot be deleted during the whole life of the file
199 * system. Instead, they are moved to the available list and remain there
203 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
207 TMPFS_NODE_LOCK(node);
210 TMPFS_ASSERT_ELOCKED(node);
211 KKASSERT(node->tn_vnode == NULL);
212 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
215 LIST_REMOVE(node, tn_entries);
217 tmp->tm_nodes_inuse--;
220 switch (node->tn_type) {
222 /* Do not do anything. VNON is provided to let the
223 * allocation routine clean itself easily by avoiding
224 * duplicating code in it. */
234 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
235 node->tn_dir.tn_parent->tn_links--;
236 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
244 kfree(node->tn_link, M_TMPFSNAME);
248 if (node->tn_reg.tn_aobj != NULL)
249 vm_pager_deallocate(node->tn_reg.tn_aobj);
250 node->tn_reg.tn_aobj = NULL;
251 pages = node->tn_reg.tn_aobj_pages;
255 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
258 objcache_put(tmp->tm_node_pool, node);
261 tmp->tm_pages_used -= pages;
263 TMPFS_NODE_UNLOCK(node);
266 /* --------------------------------------------------------------------- */
269 * Allocates a new directory entry for the node node with a name of name.
270 * The new directory entry is returned in *de.
272 * The link count of node is increased by one to reflect the new object
275 * Returns zero on success or an appropriate error code on failure.
278 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
279 const char *name, uint16_t len, struct tmpfs_dirent **de)
281 struct tmpfs_dirent *nde;
284 nde = (struct tmpfs_dirent *)objcache_get(tmp->tm_dirent_pool, M_WAITOK);
285 nde->td_name = kmalloc(len, M_TMPFSNAME, M_WAITOK);
286 nde->td_namelen = len;
287 bcopy(name, nde->td_name, len);
288 nde->td_name[len] = '\0';
292 TMPFS_NODE_LOCK(node);
294 TMPFS_NODE_UNLOCK(node);
301 /* --------------------------------------------------------------------- */
304 * Frees a directory entry. It is the caller's responsibility to destroy
305 * the node referenced by it if needed.
307 * The link count of node is decreased by one to reflect the removal of an
308 * object that referenced it. This only happens if 'node_exists' is true;
309 * otherwise the function will not access the node referred to by the
310 * directory entry, as it may already have been released from the outside.
313 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de,
314 boolean_t node_exists)
317 struct tmpfs_node *node;
321 TMPFS_NODE_LOCK(node);
322 TMPFS_ASSERT_ELOCKED(node);
323 KKASSERT(node->tn_links > 0);
325 TMPFS_NODE_UNLOCK(node);
328 kfree(de->td_name, M_TMPFSNAME);
329 objcache_put(tmp->tm_dirent_pool, de);
332 /* --------------------------------------------------------------------- */
335 * Allocates a new vnode for the node node or returns a new reference to
336 * an existing one if the node had already a vnode referencing it. The
337 * resulting locked vnode is returned in *vpp.
339 * Returns zero on success or an appropriate error code on failure.
342 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
349 TMPFS_NODE_LOCK(node);
350 if ((vp = node->tn_vnode) != NULL) {
351 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
352 TMPFS_NODE_UNLOCK(node);
353 (void) vget(vp, lkflag | LK_EXCLUSIVE | LK_RETRY);
356 * Make sure the vnode is still there after
357 * getting the interlock to avoid racing a free.
359 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
367 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
368 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
369 TMPFS_NODE_UNLOCK(node);
376 * otherwise lock the vp list while we call getnewvnode
377 * since that can block.
379 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
380 node->tn_vpstate |= TMPFS_VNODE_WANT;
381 error = tsleep((caddr_t) &node->tn_vpstate,
382 PINTERLOCKED | PCATCH, "tmpfs_alloc_vp", 0);
383 TMPFS_NODE_UNLOCK(node);
389 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
391 TMPFS_NODE_UNLOCK(node);
393 /* Get a new vnode and associate it with our node. */
394 error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE);
397 KKASSERT(vp != NULL);
400 vp->v_type = node->tn_type;
402 /* Type-specific initialization. */
403 switch (node->tn_type) {
411 vinitvmio(vp, node->tn_size);
414 if (node->tn_size >= vp->v_mount->mnt_maxsymlinklen)
415 vinitvmio(vp, node->tn_size);
418 vp->v_ops = &mp->mnt_vn_fifo_ops;
421 vinitvmio(vp, node->tn_size);
422 KKASSERT(node->tn_dir.tn_parent != NULL);
426 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
432 TMPFS_NODE_LOCK(node);
434 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
435 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
438 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
439 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
440 TMPFS_NODE_UNLOCK(node);
441 wakeup((caddr_t) &node->tn_vpstate);
443 TMPFS_NODE_UNLOCK(node);
448 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
450 TMPFS_NODE_LOCK(node);
451 KKASSERT(*vpp == node->tn_vnode);
452 TMPFS_NODE_UNLOCK(node);
458 /* --------------------------------------------------------------------- */
461 * Destroys the association between the vnode vp and the node it
465 tmpfs_free_vp(struct vnode *vp)
467 struct tmpfs_node *node;
469 node = VP_TO_TMPFS_NODE(vp);
471 TMPFS_NODE_LOCK(node);
472 KKASSERT(lockcount(TMPFS_NODE_MTX(node)) > 0);
473 node->tn_vnode = NULL;
474 TMPFS_NODE_UNLOCK(node);
478 /* --------------------------------------------------------------------- */
481 * Allocates a new file of type 'type' and adds it to the parent directory
482 * 'dvp'; this addition is done using the component name given in 'cnp'.
483 * The ownership of the new file is automatically assigned based on the
484 * credentials of the caller (through 'cnp'), the group is set based on
485 * the parent directory and the mode is determined from the 'vap' argument.
486 * If successful, *vpp holds a vnode to the newly created file and zero
487 * is returned. Otherwise *vpp is NULL and the function returns an
488 * appropriate error code.
491 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
492 struct namecache *ncp, struct ucred *cred, char *target)
495 struct tmpfs_dirent *de;
496 struct tmpfs_mount *tmp;
497 struct tmpfs_node *dnode;
498 struct tmpfs_node *node;
499 struct tmpfs_node *parent;
501 tmp = VFS_TO_TMPFS(dvp->v_mount);
502 dnode = VP_TO_TMPFS_DIR(dvp);
505 /* If the entry we are creating is a directory, we cannot overflow
506 * the number of links of its parent, because it will get a new
508 if (vap->va_type == VDIR) {
509 /* Ensure that we do not overflow the maximum number of links
510 * imposed by the system. */
511 KKASSERT(dnode->tn_links <= LINK_MAX);
512 if (dnode->tn_links == LINK_MAX) {
517 KKASSERT(parent != NULL);
521 /* Allocate a node that represents the new file. */
522 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid,
523 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rmajor, vap->va_rminor, &node);
527 /* Allocate a directory entry that points to the new file. */
528 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen,
531 tmpfs_free_node(tmp, node);
535 /* Allocate a vnode for the new file. */
536 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
538 tmpfs_free_dirent(tmp, de, TRUE);
539 tmpfs_free_node(tmp, node);
543 /* Now that all required items are allocated, we can proceed to
544 * insert the new node into the directory, an operation that
546 tmpfs_dir_attach(dvp, de);
551 /* --------------------------------------------------------------------- */
554 * Attaches the directory entry de to the directory represented by vp.
555 * Note that this does not change the link count of the node pointed by
556 * the directory entry, as this is done by tmpfs_alloc_dirent.
559 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
561 struct tmpfs_node *dnode;
563 dnode = VP_TO_TMPFS_DIR(vp);
566 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
569 TMPFS_NODE_LOCK(dnode);
570 TMPFS_ASSERT_ELOCKED(dnode);
571 dnode->tn_size += sizeof(struct tmpfs_dirent);
572 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
574 TMPFS_NODE_UNLOCK(dnode);
577 /* --------------------------------------------------------------------- */
580 * Detaches the directory entry de from the directory represented by vp.
581 * Note that this does not change the link count of the node pointed by
582 * the directory entry, as this is done by tmpfs_free_dirent.
585 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
587 struct tmpfs_node *dnode;
589 dnode = VP_TO_TMPFS_DIR(vp);
593 if (dnode->tn_dir.tn_readdir_lastp == de) {
594 dnode->tn_dir.tn_readdir_lastn = 0;
595 dnode->tn_dir.tn_readdir_lastp = NULL;
598 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
601 TMPFS_NODE_LOCK(dnode);
602 TMPFS_ASSERT_ELOCKED(dnode);
603 dnode->tn_size -= sizeof(struct tmpfs_dirent);
604 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
606 TMPFS_NODE_UNLOCK(dnode);
609 /* --------------------------------------------------------------------- */
612 * Looks for a directory entry in the directory represented by node.
613 * 'ncp' describes the name of the entry to look for. Note that the .
614 * and .. components are not allowed as they do not physically exist
615 * within directories.
617 * Returns a pointer to the entry when found, otherwise NULL.
619 struct tmpfs_dirent *
620 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
621 struct namecache *ncp)
624 struct tmpfs_dirent *de;
625 int len = ncp->nc_nlen;
627 TMPFS_VALIDATE_DIR(node);
630 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
631 if (f != NULL && de->td_node != f)
633 if (len == de->td_namelen) {
634 if (!memcmp(ncp->nc_name, de->td_name, len)) {
641 TMPFS_NODE_LOCK(node);
642 node->tn_status |= TMPFS_NODE_ACCESSED;
643 TMPFS_NODE_UNLOCK(node);
645 return found ? de : NULL;
648 /* --------------------------------------------------------------------- */
651 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
652 * directory and returns it in the uio space. The function returns 0
653 * on success, -1 if there was not enough space in the uio structure to
654 * hold the directory entry or an appropriate error code if another
658 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
664 TMPFS_VALIDATE_DIR(node);
665 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
667 dent.d_ino = node->tn_id;
668 dent.d_type = DT_DIR;
670 dent.d_name[0] = '.';
671 dent.d_name[1] = '\0';
672 dirsize = _DIRENT_DIRSIZ(&dent);
674 if (dirsize > uio->uio_resid)
677 error = uiomove((caddr_t)&dent, dirsize, uio);
679 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
682 node->tn_status |= TMPFS_NODE_ACCESSED;
687 /* --------------------------------------------------------------------- */
690 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
691 * directory and returns it in the uio space. The function returns 0
692 * on success, -1 if there was not enough space in the uio structure to
693 * hold the directory entry or an appropriate error code if another
697 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
703 TMPFS_VALIDATE_DIR(node);
704 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
707 * Return ENOENT if the current node is already removed.
709 if (node->tn_dir.tn_parent == NULL) {
713 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
714 dent.d_ino = node->tn_dir.tn_parent->tn_id;
715 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
717 dent.d_type = DT_DIR;
719 dent.d_name[0] = '.';
720 dent.d_name[1] = '.';
721 dent.d_name[2] = '\0';
722 dirsize = _DIRENT_DIRSIZ(&dent);
724 if (dirsize > uio->uio_resid)
727 error = uiomove((caddr_t)&dent, dirsize, uio);
729 struct tmpfs_dirent *de;
731 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
733 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
735 uio->uio_offset = tmpfs_dircookie(de);
739 node->tn_status |= TMPFS_NODE_ACCESSED;
744 /* --------------------------------------------------------------------- */
747 * Lookup a directory entry by its associated cookie.
749 struct tmpfs_dirent *
750 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
752 struct tmpfs_dirent *de;
754 if (cookie == node->tn_dir.tn_readdir_lastn &&
755 node->tn_dir.tn_readdir_lastp != NULL) {
756 return node->tn_dir.tn_readdir_lastp;
759 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
760 if (tmpfs_dircookie(de) == cookie) {
768 /* --------------------------------------------------------------------- */
771 * Helper function for tmpfs_readdir. Returns as much directory entries
772 * as can fit in the uio space. The read starts at uio->uio_offset.
773 * The function returns 0 on success, -1 if there was not enough space
774 * in the uio structure to hold the directory entry or an appropriate
775 * error code if another error happens.
778 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
782 struct tmpfs_dirent *de;
784 TMPFS_VALIDATE_DIR(node);
786 /* Locate the first directory entry we have to return. We have cached
787 * the last readdir in the node, so use those values if appropriate.
788 * Otherwise do a linear scan to find the requested entry. */
789 startcookie = uio->uio_offset;
790 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT);
791 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
792 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
795 de = tmpfs_dir_lookupbycookie(node, startcookie);
801 /* Read as much entries as possible; i.e., until we reach the end of
802 * the directory or we exhaust uio space. */
807 /* Create a dirent structure representing the current
808 * tmpfs_node and fill it. */
809 d.d_ino = de->td_node->tn_id;
810 switch (de->td_node->tn_type) {
840 panic("tmpfs_dir_getdents: type %p %d",
841 de->td_node, (int)de->td_node->tn_type);
843 d.d_namlen = de->td_namelen;
844 KKASSERT(de->td_namelen < sizeof(d.d_name));
845 bcopy(de->td_name, d.d_name, d.d_namlen);
846 d.d_name[d.d_namlen] = '\0';
847 reclen = _DIRENT_RECLEN(d.d_namlen);
849 /* Stop reading if the directory entry we are treating is
850 * bigger than the amount of data that can be returned. */
851 if (reclen > uio->uio_resid) {
856 /* Copy the new dirent structure into the output buffer and
857 * advance pointers. */
858 error = uiomove((caddr_t)&d, reclen, uio);
861 de = TAILQ_NEXT(de, td_entries);
862 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
864 /* Update the offset and cache. */
866 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
867 node->tn_dir.tn_readdir_lastn = 0;
868 node->tn_dir.tn_readdir_lastp = NULL;
870 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
871 node->tn_dir.tn_readdir_lastp = de;
873 node->tn_status |= TMPFS_NODE_ACCESSED;
878 /* --------------------------------------------------------------------- */
881 * Resizes the aobj associated to the regular file pointed to by vp to
882 * the size newsize. 'vp' must point to a vnode that represents a regular
883 * file. 'newsize' must be positive.
885 * pass trivial as 1 when buf content will be overwritten, otherwise set 0
888 * Returns zero on success or an appropriate error code on failure.
891 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial)
894 size_t newpages, oldpages;
895 struct tmpfs_mount *tmp;
896 struct tmpfs_node *node;
898 int biosize = vp->v_mount->mnt_stat.f_iosize;
901 KKASSERT(vp->v_type == VREG);
902 KKASSERT(newsize >= 0);
905 node = VP_TO_TMPFS_NODE(vp);
906 tmp = VFS_TO_TMPFS(vp->v_mount);
908 /* Convert the old and new sizes to the number of pages needed to
909 * store them. It may happen that we do not need to do anything
910 * because the last allocated page can accommodate the change on
912 oldsize = node->tn_size;
913 oldpages = round_page(oldsize) / PAGE_SIZE;
914 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages);
915 newpages = round_page(newsize) / PAGE_SIZE;
917 if (newpages > oldpages &&
918 newpages - oldpages > TMPFS_PAGES_AVAIL(tmp)) {
924 tmp->tm_pages_used += (newpages - oldpages);
927 TMPFS_NODE_LOCK(node);
928 node->tn_reg.tn_aobj_pages = newpages;
929 node->tn_size = newsize;
930 TMPFS_NODE_UNLOCK(node);
932 if (newsize < oldsize)
933 error = nvtruncbuf(vp, newsize, biosize, -1);
935 error = nvextendbuf(vp, oldsize, newsize, biosize, biosize,
942 /* --------------------------------------------------------------------- */
945 * Change flags of the given vnode.
946 * Caller should execute tmpfs_update on vp after a successful execution.
947 * The vnode must be locked on entry and remain locked on exit.
950 tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred)
953 struct tmpfs_node *node;
956 KKASSERT(vn_islocked(vp));
958 node = VP_TO_TMPFS_NODE(vp);
960 /* Disallow this operation if the file system is mounted read-only. */
961 if (vp->v_mount->mnt_flag & MNT_RDONLY)
964 fmode = FFLAGS(node->tn_flags);
966 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
968 if (fmode & (FWRITE | O_TRUNC)) {
969 if (vp->v_type == VDIR) {
972 error = vn_writechk(vp, NULL);
981 error = VOP_ACCESS(vp, mode, cred);
986 * Unprivileged processes are not permitted to unset system
987 * flags, or modify flags if any system flags are set.
989 TMPFS_NODE_LOCK(node);
990 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
993 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
994 error = securelevel_gt(cred, 0);
998 /* Snapshot flag cannot be set or cleared */
999 if (((flags & SF_SNAPSHOT) != 0 &&
1000 (node->tn_flags & SF_SNAPSHOT) == 0) ||
1001 ((flags & SF_SNAPSHOT) == 0 &&
1002 (node->tn_flags & SF_SNAPSHOT) != 0))
1005 node->tn_flags = flags;
1008 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1009 (flags & UF_SETTABLE) != flags)
1011 node->tn_flags &= SF_SETTABLE;
1012 node->tn_flags |= (flags & UF_SETTABLE);
1014 node->tn_status |= TMPFS_NODE_CHANGED;
1015 TMPFS_NODE_UNLOCK(node);
1017 KKASSERT(vn_islocked(vp));
1022 /* --------------------------------------------------------------------- */
1025 * Change access mode on the given vnode.
1026 * Caller should execute tmpfs_update on vp after a successful execution.
1027 * The vnode must be locked on entry and remain locked on exit.
1030 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred)
1033 struct tmpfs_node *node;
1036 KKASSERT(vn_islocked(vp));
1038 node = VP_TO_TMPFS_NODE(vp);
1040 /* Disallow this operation if the file system is mounted read-only. */
1041 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1044 /* Immutable or append-only files cannot be modified, either. */
1045 if (node->tn_flags & (IMMUTABLE | APPEND))
1048 fmode = FFLAGS(node->tn_flags);
1050 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
1052 if (fmode & (FWRITE | O_TRUNC)) {
1053 if (vp->v_type == VDIR) {
1056 error = vn_writechk(vp, NULL);
1065 error = VOP_ACCESS(vp, accmode, cred);
1071 * Privileged processes may set the sticky bit on non-directories,
1072 * as well as set the setgid bit on a file with a group that the
1073 * process is not a member of.
1075 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1076 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1079 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1080 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1086 TMPFS_NODE_LOCK(node);
1087 node->tn_mode &= ~ALLPERMS;
1088 node->tn_mode |= mode & ALLPERMS;
1090 node->tn_status |= TMPFS_NODE_CHANGED;
1091 TMPFS_NODE_UNLOCK(node);
1093 KKASSERT(vn_islocked(vp));
1098 /* --------------------------------------------------------------------- */
1101 * Change ownership of the given vnode. At least one of uid or gid must
1102 * be different than VNOVAL. If one is set to that value, the attribute
1104 * Caller should execute tmpfs_update on vp after a successful execution.
1105 * The vnode must be locked on entry and remain locked on exit.
1108 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
1111 struct tmpfs_node *node;
1116 KKASSERT(vn_islocked(vp));
1118 node = VP_TO_TMPFS_NODE(vp);
1120 /* Assign default values if they are unknown. */
1121 KKASSERT(uid != VNOVAL || gid != VNOVAL);
1126 KKASSERT(uid != VNOVAL && gid != VNOVAL);
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 fmode = FFLAGS(node->tn_flags);
1138 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
1140 if (fmode & (FWRITE | O_TRUNC)) {
1141 if (vp->v_type == VDIR) {
1144 error = vn_writechk(vp, NULL);
1153 error = VOP_ACCESS(vp, mode, cred);
1159 * To change the owner of a file, or change the group of a file to a
1160 * group of which we are not a member, the caller must have
1163 if ((uid != node->tn_uid ||
1164 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1165 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1168 ogid = node->tn_gid;
1169 ouid = node->tn_uid;
1171 TMPFS_NODE_LOCK(node);
1175 node->tn_status |= TMPFS_NODE_CHANGED;
1177 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1178 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1179 node->tn_mode &= ~(S_ISUID | S_ISGID);
1181 TMPFS_NODE_UNLOCK(node);
1183 KKASSERT(vn_islocked(vp));
1188 /* --------------------------------------------------------------------- */
1191 * Change size of the given vnode.
1192 * Caller should execute tmpfs_update on vp after a successful execution.
1193 * The vnode must be locked on entry and remain locked on exit.
1196 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred)
1199 struct tmpfs_node *node;
1201 KKASSERT(vn_islocked(vp));
1203 node = VP_TO_TMPFS_NODE(vp);
1205 /* Decide whether this is a valid operation based on the file type. */
1207 switch (vp->v_type) {
1212 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1221 /* Allow modifications of special files even if in the file
1222 * system is mounted read-only (we are not modifying the
1223 * files themselves, but the objects they represent). */
1227 /* Anything else is unsupported. */
1231 /* Immutable or append-only files cannot be modified, either. */
1232 if (node->tn_flags & (IMMUTABLE | APPEND))
1235 error = tmpfs_truncate(vp, size);
1236 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1237 * for us, as will update tn_status; no need to do that here. */
1239 KKASSERT(vn_islocked(vp));
1244 /* --------------------------------------------------------------------- */
1247 * Change access and modification times of the given vnode.
1248 * Caller should execute tmpfs_update on vp after a successful execution.
1249 * The vnode must be locked on entry and remain locked on exit.
1252 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1253 int vaflags, struct ucred *cred)
1256 struct tmpfs_node *node;
1259 KKASSERT(vn_islocked(vp));
1261 node = VP_TO_TMPFS_NODE(vp);
1263 /* Disallow this operation if the file system is mounted read-only. */
1264 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1267 /* Immutable or append-only files cannot be modified, either. */
1268 if (node->tn_flags & (IMMUTABLE | APPEND))
1271 /* Determine if the user have proper privilege to update time. */
1272 fmode = FFLAGS(node->tn_flags);
1274 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
1276 if (fmode & (FWRITE | O_TRUNC)) {
1277 if (vp->v_type == VDIR) {
1280 error = vn_writechk(vp, NULL);
1290 if (vaflags & VA_UTIMES_NULL) {
1291 error = VOP_ACCESS(vp, mode, cred);
1293 error = VOP_ACCESS(vp, VWRITE, cred);
1295 error = VOP_ACCESS(vp, mode, cred);
1300 TMPFS_NODE_LOCK(node);
1301 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1302 node->tn_status |= TMPFS_NODE_ACCESSED;
1304 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1305 node->tn_status |= TMPFS_NODE_MODIFIED;
1307 TMPFS_NODE_UNLOCK(node);
1309 tmpfs_itimes(vp, atime, mtime);
1311 KKASSERT(vn_islocked(vp));
1316 /* --------------------------------------------------------------------- */
1317 /* Sync timestamps */
1319 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1320 const struct timespec *mod)
1322 struct tmpfs_node *node;
1323 struct timespec now;
1325 node = VP_TO_TMPFS_NODE(vp);
1327 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1328 TMPFS_NODE_CHANGED)) == 0)
1331 vfs_timestamp(&now);
1333 TMPFS_NODE_LOCK(node);
1334 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1337 node->tn_atime = acc->tv_sec;
1338 node->tn_atimensec = acc->tv_nsec;
1340 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1343 node->tn_mtime = mod->tv_sec;
1344 node->tn_mtimensec = mod->tv_nsec;
1346 if (node->tn_status & TMPFS_NODE_CHANGED) {
1347 node->tn_ctime = now.tv_sec;
1348 node->tn_ctimensec = now.tv_nsec;
1351 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1352 TMPFS_NODE_UNLOCK(node);
1355 /* --------------------------------------------------------------------- */
1358 tmpfs_update(struct vnode *vp)
1361 tmpfs_itimes(vp, NULL, NULL);
1364 /* --------------------------------------------------------------------- */
1367 tmpfs_truncate(struct vnode *vp, off_t length)
1370 struct tmpfs_node *node;
1372 node = VP_TO_TMPFS_NODE(vp);
1379 if (node->tn_size == length) {
1384 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1388 error = tmpfs_reg_resize(vp, length, 1);
1391 TMPFS_NODE_LOCK(node);
1392 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1393 TMPFS_NODE_UNLOCK(node);
1402 /* --------------------------------------------------------------------- */
1405 tmpfs_fetch_ino(void)
1409 spin_lock_wr(&ino_lock);
1411 spin_unlock_wr(&ino_lock);