1 /* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */
4 * Copyright (c) 2005 The NetBSD Foundation, Inc.
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
34 * Efficient memory file system supporting functions.
36 #include <sys/cdefs.h>
38 #include <sys/kernel.h>
39 #include <sys/param.h>
40 #include <sys/namei.h>
43 #include <sys/spinlock2.h>
45 #include <sys/systm.h>
46 #include <sys/vnode.h>
47 #include <sys/vmmeter.h>
49 #include <sys/mplock2.h>
52 #include <vm/vm_object.h>
53 #include <vm/vm_page.h>
54 #include <vm/vm_pager.h>
55 #include <vm/vm_extern.h>
57 #include <vfs/tmpfs/tmpfs.h>
58 #include <vfs/tmpfs/tmpfs_fifoops.h>
59 #include <vfs/tmpfs/tmpfs_vnops.h>
61 static ino_t t_ino = 2;
62 static struct spinlock ino_lock;
63 static ino_t tmpfs_fetch_ino(void);
65 /* --------------------------------------------------------------------- */
68 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
69 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
70 * using the credentials of the process 'p'.
72 * If the node type is set to 'VDIR', then the parent parameter must point
73 * to the parent directory of the node being created. It may only be NULL
74 * while allocating the root node.
76 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
77 * specifies the device the node represents.
79 * If the node type is set to 'VLNK', then the parameter target specifies
80 * the file name of the target file for the symbolic link that is being
83 * Note that new nodes are retrieved from the available list if it has
84 * items or, if it is empty, from the node pool as long as there is enough
85 * space to create them.
87 * Returns zero on success or an appropriate error code on failure.
90 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
91 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
92 char *target, int rmajor, int rminor, struct tmpfs_node **node)
94 struct tmpfs_node *nnode;
98 /* If the root directory of the 'tmp' file system is not yet
99 * allocated, this must be the request to do it. */
100 KKASSERT(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
102 KKASSERT(IFF(type == VLNK, target != NULL));
103 KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL));
105 if (tmp->tm_nodes_inuse > tmp->tm_nodes_max)
108 nnode = (struct tmpfs_node *)objcache_get(tmp->tm_node_pool, M_WAITOK);
110 /* Generic initialization. */
111 nnode->tn_type = type;
113 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime
115 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec
119 nnode->tn_mode = mode;
120 nnode->tn_id = tmpfs_fetch_ino();
121 nnode->tn_advlock.init_done = 0;
123 /* Type-specific initialization. */
124 switch (nnode->tn_type) {
127 rdev = makeudev(rmajor, rminor);
128 if (rdev == NOUDEV) {
131 nnode->tn_rdev = rdev;
135 TAILQ_INIT(&nnode->tn_dir.tn_dirhead);
136 KKASSERT(parent != nnode);
137 KKASSERT(IMPLIES(parent == NULL, tmp->tm_root == NULL));
138 nnode->tn_dir.tn_parent = parent;
139 nnode->tn_dir.tn_readdir_lastn = 0;
140 nnode->tn_dir.tn_readdir_lastp = NULL;
144 TMPFS_NODE_LOCK(parent);
146 TMPFS_NODE_UNLOCK(parent);
156 KKASSERT((strlen(target) +1) < MAXPATHLEN);
157 nnode->tn_size = strlen(target);
158 nnode->tn_link = kmalloc(nnode->tn_size + 1, M_TMPFSNAME,
160 bcopy(target, nnode->tn_link, nnode->tn_size);
161 nnode->tn_link[nnode->tn_size] = '\0';
165 nnode->tn_reg.tn_aobj =
166 swap_pager_alloc(NULL, 0, VM_PROT_DEFAULT, 0);
167 nnode->tn_reg.tn_aobj_pages = 0;
172 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
175 TMPFS_NODE_LOCK(nnode);
177 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
178 tmp->tm_nodes_inuse++;
180 TMPFS_NODE_UNLOCK(nnode);
186 /* --------------------------------------------------------------------- */
189 * Destroys the node pointed to by node from the file system 'tmp'.
190 * If the node does not belong to the given mount point, the results are
193 * If the node references a directory; no entries are allowed because
194 * their removal could need a recursive algorithm, something forbidden in
195 * kernel space. Furthermore, there is not need to provide such
196 * functionality (recursive removal) because the only primitives offered
197 * to the user are the removal of empty directories and the deletion of
200 * Note that nodes are not really deleted; in fact, when a node has been
201 * allocated, it cannot be deleted during the whole life of the file
202 * system. Instead, they are moved to the available list and remain there
206 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
211 TMPFS_ASSERT_ELOCKED(node);
212 KKASSERT(node->tn_vnode == NULL);
213 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
217 LIST_REMOVE(node, tn_entries);
218 tmp->tm_nodes_inuse--;
220 TMPFS_NODE_UNLOCK(node);
222 switch (node->tn_type) {
224 /* Do not do anything. VNON is provided to let the
225 * allocation routine clean itself easily by avoiding
226 * duplicating code in it. */
235 * The parent link can be NULL if this is the root
240 KKASSERT(node->tn_dir.tn_parent || node == tmp->tm_root);
241 if (node->tn_dir.tn_parent) {
242 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
243 node->tn_dir.tn_parent->tn_links--;
246 * If the parent directory has no more links and
247 * no vnode ref nothing is going to come along
248 * and clean it up unless we do it here.
250 if (node->tn_dir.tn_parent->tn_links == 0 &&
251 node->tn_dir.tn_parent->tn_vnode == NULL) {
252 tmpfs_free_node(tmp, node->tn_dir.tn_parent);
253 /* eats parent lock */
255 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
257 node->tn_dir.tn_parent = NULL;
261 * If the root node is being destroyed don't leave a
262 * dangling pointer in tmpfs_mount.
264 if (node == tmp->tm_root)
273 kfree(node->tn_link, M_TMPFSNAME);
274 node->tn_link = NULL;
279 if (node->tn_reg.tn_aobj != NULL)
280 vm_pager_deallocate(node->tn_reg.tn_aobj);
281 node->tn_reg.tn_aobj = NULL;
282 pages = node->tn_reg.tn_aobj_pages;
286 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
290 * Clean up fields for the next allocation. The objcache only ctors
293 tmpfs_node_ctor(node, NULL, 0);
294 objcache_put(tmp->tm_node_pool, node);
295 /* node is now invalid */
298 tmp->tm_pages_used -= pages;
302 /* --------------------------------------------------------------------- */
305 * Allocates a new directory entry for the node node with a name of name.
306 * The new directory entry is returned in *de.
308 * The link count of node is increased by one to reflect the new object
311 * Returns zero on success or an appropriate error code on failure.
314 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
315 const char *name, uint16_t len, struct tmpfs_dirent **de)
317 struct tmpfs_dirent *nde;
320 nde = (struct tmpfs_dirent *)objcache_get(tmp->tm_dirent_pool, M_WAITOK);
321 nde->td_name = kmalloc(len + 1, M_TMPFSNAME, M_WAITOK);
322 nde->td_namelen = len;
323 bcopy(name, nde->td_name, len);
324 nde->td_name[len] = '\0';
328 TMPFS_NODE_LOCK(node);
330 TMPFS_NODE_UNLOCK(node);
337 /* --------------------------------------------------------------------- */
340 * Frees a directory entry. It is the caller's responsibility to destroy
341 * the node referenced by it if needed.
343 * The link count of node is decreased by one to reflect the removal of an
344 * object that referenced it. This only happens if 'node_exists' is true;
345 * otherwise the function will not access the node referred to by the
346 * directory entry, as it may already have been released from the outside.
349 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
351 struct tmpfs_node *node;
355 TMPFS_NODE_LOCK(node);
356 TMPFS_ASSERT_ELOCKED(node);
357 KKASSERT(node->tn_links > 0);
359 TMPFS_NODE_UNLOCK(node);
361 kfree(de->td_name, M_TMPFSNAME);
365 objcache_put(tmp->tm_dirent_pool, de);
368 /* --------------------------------------------------------------------- */
371 * Allocates a new vnode for the node node or returns a new reference to
372 * an existing one if the node had already a vnode referencing it. The
373 * resulting locked vnode is returned in *vpp.
375 * Returns zero on success or an appropriate error code on failure.
378 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
386 * Interlocked extraction from node. This can race many things.
387 * We have to get a soft reference on the vnode while we hold
388 * the node locked, then acquire it properly and check for races.
390 TMPFS_NODE_LOCK(node);
391 if ((vp = node->tn_vnode) != NULL) {
392 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
393 vhold_interlocked(vp);
394 TMPFS_NODE_UNLOCK(node);
396 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) {
400 if (node->tn_vnode != vp) {
411 * This should never happen.
413 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) {
414 TMPFS_NODE_UNLOCK(node);
420 * Interlock against other calls to tmpfs_alloc_vp() trying to
421 * allocate and assign a vp to node.
423 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
424 node->tn_vpstate |= TMPFS_VNODE_WANT;
425 error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH,
426 "tmpfs_alloc_vp", 0);
427 TMPFS_NODE_UNLOCK(node);
432 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
433 TMPFS_NODE_UNLOCK(node);
436 * Allocate a new vnode (may block). The ALLOCATING flag should
437 * prevent a race against someone else assigning node->tn_vnode.
439 error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE);
443 KKASSERT(node->tn_vnode == NULL);
444 KKASSERT(vp != NULL);
446 vp->v_type = node->tn_type;
448 /* Type-specific initialization. */
449 switch (node->tn_type) {
457 vinitvmio(vp, node->tn_size, BMASK, -1);
462 vp->v_ops = &mp->mnt_vn_fifo_ops;
468 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
474 TMPFS_NODE_LOCK(node);
476 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
477 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
480 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
481 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
482 TMPFS_NODE_UNLOCK(node);
483 wakeup(&node->tn_vpstate);
485 TMPFS_NODE_UNLOCK(node);
491 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
493 TMPFS_NODE_LOCK(node);
494 KKASSERT(*vpp == node->tn_vnode);
495 TMPFS_NODE_UNLOCK(node);
501 /* --------------------------------------------------------------------- */
504 * Destroys the association between the vnode vp and the node it
508 tmpfs_free_vp(struct vnode *vp)
510 struct tmpfs_node *node;
512 node = VP_TO_TMPFS_NODE(vp);
514 TMPFS_NODE_LOCK(node);
515 KKASSERT(lockcount(TMPFS_NODE_MTX(node)) > 0);
516 node->tn_vnode = NULL;
517 TMPFS_NODE_UNLOCK(node);
521 /* --------------------------------------------------------------------- */
524 * Allocates a new file of type 'type' and adds it to the parent directory
525 * 'dvp'; this addition is done using the component name given in 'cnp'.
526 * The ownership of the new file is automatically assigned based on the
527 * credentials of the caller (through 'cnp'), the group is set based on
528 * the parent directory and the mode is determined from the 'vap' argument.
529 * If successful, *vpp holds a vnode to the newly created file and zero
530 * is returned. Otherwise *vpp is NULL and the function returns an
531 * appropriate error code.
534 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
535 struct namecache *ncp, struct ucred *cred, char *target)
538 struct tmpfs_dirent *de;
539 struct tmpfs_mount *tmp;
540 struct tmpfs_node *dnode;
541 struct tmpfs_node *node;
542 struct tmpfs_node *parent;
544 tmp = VFS_TO_TMPFS(dvp->v_mount);
545 dnode = VP_TO_TMPFS_DIR(dvp);
548 /* If the entry we are creating is a directory, we cannot overflow
549 * the number of links of its parent, because it will get a new
551 if (vap->va_type == VDIR) {
552 /* Ensure that we do not overflow the maximum number of links
553 * imposed by the system. */
554 KKASSERT(dnode->tn_links <= LINK_MAX);
555 if (dnode->tn_links == LINK_MAX) {
560 KKASSERT(parent != NULL);
564 /* Allocate a node that represents the new file. */
565 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid,
566 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rmajor, vap->va_rminor, &node);
569 TMPFS_NODE_LOCK(node);
571 /* Allocate a directory entry that points to the new file. */
572 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de);
574 tmpfs_free_node(tmp, node);
579 /* Allocate a vnode for the new file. */
580 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
582 tmpfs_free_dirent(tmp, de);
583 tmpfs_free_node(tmp, node);
588 /* Now that all required items are allocated, we can proceed to
589 * insert the new node into the directory, an operation that
591 tmpfs_dir_attach(dnode, de);
592 TMPFS_NODE_UNLOCK(node);
597 /* --------------------------------------------------------------------- */
600 * Attaches the directory entry de to the directory represented by vp.
601 * Note that this does not change the link count of the node pointed by
602 * the directory entry, as this is done by tmpfs_alloc_dirent.
605 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
607 TMPFS_NODE_LOCK(dnode);
608 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
610 TMPFS_ASSERT_ELOCKED(dnode);
611 dnode->tn_size += sizeof(struct tmpfs_dirent);
612 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
614 TMPFS_NODE_UNLOCK(dnode);
617 /* --------------------------------------------------------------------- */
620 * Detaches the directory entry de from the directory represented by vp.
621 * Note that this does not change the link count of the node pointed by
622 * the directory entry, as this is done by tmpfs_free_dirent.
625 tmpfs_dir_detach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
627 TMPFS_NODE_LOCK(dnode);
628 if (dnode->tn_dir.tn_readdir_lastp == de) {
629 dnode->tn_dir.tn_readdir_lastn = 0;
630 dnode->tn_dir.tn_readdir_lastp = NULL;
632 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
634 TMPFS_ASSERT_ELOCKED(dnode);
635 dnode->tn_size -= sizeof(struct tmpfs_dirent);
636 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
638 TMPFS_NODE_UNLOCK(dnode);
641 /* --------------------------------------------------------------------- */
644 * Looks for a directory entry in the directory represented by node.
645 * 'ncp' describes the name of the entry to look for. Note that the .
646 * and .. components are not allowed as they do not physically exist
647 * within directories.
649 * Returns a pointer to the entry when found, otherwise NULL.
651 struct tmpfs_dirent *
652 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
653 struct namecache *ncp)
655 struct tmpfs_dirent *de;
656 int len = ncp->nc_nlen;
658 TMPFS_VALIDATE_DIR(node);
660 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
661 if (f != NULL && de->td_node != f)
663 if (len == de->td_namelen) {
664 if (!memcmp(ncp->nc_name, de->td_name, len))
669 TMPFS_NODE_LOCK(node);
670 node->tn_status |= TMPFS_NODE_ACCESSED;
671 TMPFS_NODE_UNLOCK(node);
676 /* --------------------------------------------------------------------- */
679 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
680 * directory and returns it in the uio space. The function returns 0
681 * on success, -1 if there was not enough space in the uio structure to
682 * hold the directory entry or an appropriate error code if another
686 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
692 TMPFS_VALIDATE_DIR(node);
693 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
695 dent.d_ino = node->tn_id;
696 dent.d_type = DT_DIR;
698 dent.d_name[0] = '.';
699 dent.d_name[1] = '\0';
700 dirsize = _DIRENT_DIRSIZ(&dent);
702 if (dirsize > uio->uio_resid)
705 error = uiomove((caddr_t)&dent, dirsize, uio);
707 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
710 TMPFS_NODE_LOCK(node);
711 node->tn_status |= TMPFS_NODE_ACCESSED;
712 TMPFS_NODE_UNLOCK(node);
717 /* --------------------------------------------------------------------- */
720 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
721 * directory and returns it in the uio space. The function returns 0
722 * on success, -1 if there was not enough space in the uio structure to
723 * hold the directory entry or an appropriate error code if another
727 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
734 TMPFS_VALIDATE_DIR(node);
735 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
737 if (node->tn_dir.tn_parent) {
738 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
739 dent.d_ino = node->tn_dir.tn_parent->tn_id;
740 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
742 dent.d_ino = tmp->tm_root->tn_id;
745 dent.d_type = DT_DIR;
747 dent.d_name[0] = '.';
748 dent.d_name[1] = '.';
749 dent.d_name[2] = '\0';
750 dirsize = _DIRENT_DIRSIZ(&dent);
752 if (dirsize > uio->uio_resid)
755 error = uiomove((caddr_t)&dent, dirsize, uio);
757 struct tmpfs_dirent *de;
759 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
761 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
763 uio->uio_offset = tmpfs_dircookie(de);
767 TMPFS_NODE_LOCK(node);
768 node->tn_status |= TMPFS_NODE_ACCESSED;
769 TMPFS_NODE_UNLOCK(node);
774 /* --------------------------------------------------------------------- */
777 * Lookup a directory entry by its associated cookie.
779 struct tmpfs_dirent *
780 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
782 struct tmpfs_dirent *de;
784 if (cookie == node->tn_dir.tn_readdir_lastn &&
785 node->tn_dir.tn_readdir_lastp != NULL) {
786 return node->tn_dir.tn_readdir_lastp;
789 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
790 if (tmpfs_dircookie(de) == cookie) {
798 /* --------------------------------------------------------------------- */
801 * Helper function for tmpfs_readdir. Returns as much directory entries
802 * as can fit in the uio space. The read starts at uio->uio_offset.
803 * The function returns 0 on success, -1 if there was not enough space
804 * in the uio structure to hold the directory entry or an appropriate
805 * error code if another error happens.
808 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
812 struct tmpfs_dirent *de;
814 TMPFS_VALIDATE_DIR(node);
816 /* Locate the first directory entry we have to return. We have cached
817 * the last readdir in the node, so use those values if appropriate.
818 * Otherwise do a linear scan to find the requested entry. */
819 startcookie = uio->uio_offset;
820 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT);
821 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
822 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
825 de = tmpfs_dir_lookupbycookie(node, startcookie);
831 /* Read as much entries as possible; i.e., until we reach the end of
832 * the directory or we exhaust uio space. */
837 /* Create a dirent structure representing the current
838 * tmpfs_node and fill it. */
839 d.d_ino = de->td_node->tn_id;
840 switch (de->td_node->tn_type) {
870 panic("tmpfs_dir_getdents: type %p %d",
871 de->td_node, (int)de->td_node->tn_type);
873 d.d_namlen = de->td_namelen;
874 KKASSERT(de->td_namelen < sizeof(d.d_name));
875 bcopy(de->td_name, d.d_name, d.d_namlen);
876 d.d_name[d.d_namlen] = '\0';
877 reclen = _DIRENT_RECLEN(d.d_namlen);
879 /* Stop reading if the directory entry we are treating is
880 * bigger than the amount of data that can be returned. */
881 if (reclen > uio->uio_resid) {
886 /* Copy the new dirent structure into the output buffer and
887 * advance pointers. */
888 error = uiomove((caddr_t)&d, reclen, uio);
891 de = TAILQ_NEXT(de, td_entries);
892 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
894 /* Update the offset and cache. */
896 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
897 node->tn_dir.tn_readdir_lastn = 0;
898 node->tn_dir.tn_readdir_lastp = NULL;
900 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
901 node->tn_dir.tn_readdir_lastp = de;
903 node->tn_status |= TMPFS_NODE_ACCESSED;
908 /* --------------------------------------------------------------------- */
911 * Resizes the aobj associated to the regular file pointed to by vp to
912 * the size newsize. 'vp' must point to a vnode that represents a regular
913 * file. 'newsize' must be positive.
915 * pass trivial as 1 when buf content will be overwritten, otherwise set 0
918 * Returns zero on success or an appropriate error code on failure.
921 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial)
924 size_t newpages, oldpages;
925 struct tmpfs_mount *tmp;
926 struct tmpfs_node *node;
930 KKASSERT(vp->v_type == VREG);
931 KKASSERT(newsize >= 0);
934 node = VP_TO_TMPFS_NODE(vp);
935 tmp = VFS_TO_TMPFS(vp->v_mount);
937 /* Convert the old and new sizes to the number of pages needed to
938 * store them. It may happen that we do not need to do anything
939 * because the last allocated page can accommodate the change on
941 oldsize = node->tn_size;
942 oldpages = round_page(oldsize) / PAGE_SIZE;
943 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages);
944 newpages = round_page(newsize) / PAGE_SIZE;
946 if (newpages > oldpages &&
947 newpages - oldpages > TMPFS_PAGES_AVAIL(tmp)) {
953 tmp->tm_pages_used += (newpages - oldpages);
956 TMPFS_NODE_LOCK(node);
957 node->tn_reg.tn_aobj_pages = newpages;
958 node->tn_size = newsize;
959 TMPFS_NODE_UNLOCK(node);
962 * When adjusting the vnode filesize and its VM object we must
963 * also adjust our backing VM object (aobj). The blocksize
964 * used must match the block sized we use for the buffer cache.
966 * The backing VM object contains no VM pages, only swap
969 if (newsize < oldsize) {
974 error = nvtruncbuf(vp, newsize, BSIZE, -1);
975 aobj = node->tn_reg.tn_aobj;
978 nsize = vp->v_object->size;
981 swap_pager_freespace(aobj, nsize,
988 error = nvextendbuf(vp, oldsize, newsize, BSIZE, BSIZE,
990 aobj = node->tn_reg.tn_aobj;
992 aobj->size = vp->v_object->size;
999 /* --------------------------------------------------------------------- */
1002 * Change flags of the given vnode.
1003 * Caller should execute tmpfs_update on vp after a successful execution.
1004 * The vnode must be locked on entry and remain locked on exit.
1007 tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred)
1010 struct tmpfs_node *node;
1013 KKASSERT(vn_islocked(vp));
1015 node = VP_TO_TMPFS_NODE(vp);
1017 /* Disallow this operation if the file system is mounted read-only. */
1018 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1021 fmode = FFLAGS(node->tn_flags);
1023 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
1025 if (fmode & (FWRITE | O_TRUNC)) {
1026 if (vp->v_type == VDIR) {
1029 error = vn_writechk(vp, NULL);
1038 error = VOP_ACCESS(vp, mode, cred);
1043 * Unprivileged processes are not permitted to unset system
1044 * flags, or modify flags if any system flags are set.
1046 TMPFS_NODE_LOCK(node);
1047 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1050 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1051 error = securelevel_gt(cred, 0);
1055 /* Snapshot flag cannot be set or cleared */
1056 if (((flags & SF_SNAPSHOT) != 0 &&
1057 (node->tn_flags & SF_SNAPSHOT) == 0) ||
1058 ((flags & SF_SNAPSHOT) == 0 &&
1059 (node->tn_flags & SF_SNAPSHOT) != 0))
1062 node->tn_flags = flags;
1065 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1066 (flags & UF_SETTABLE) != flags)
1068 node->tn_flags &= SF_SETTABLE;
1069 node->tn_flags |= (flags & UF_SETTABLE);
1071 node->tn_status |= TMPFS_NODE_CHANGED;
1072 TMPFS_NODE_UNLOCK(node);
1074 KKASSERT(vn_islocked(vp));
1079 /* --------------------------------------------------------------------- */
1082 * Change access mode on the given vnode.
1083 * Caller should execute tmpfs_update on vp after a successful execution.
1084 * The vnode must be locked on entry and remain locked on exit.
1087 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred)
1090 struct tmpfs_node *node;
1093 KKASSERT(vn_islocked(vp));
1095 node = VP_TO_TMPFS_NODE(vp);
1097 /* Disallow this operation if the file system is mounted read-only. */
1098 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1101 /* Immutable or append-only files cannot be modified, either. */
1102 if (node->tn_flags & (IMMUTABLE | APPEND))
1105 fmode = FFLAGS(node->tn_flags);
1107 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
1109 if (fmode & (FWRITE | O_TRUNC)) {
1110 if (vp->v_type == VDIR) {
1113 error = vn_writechk(vp, NULL);
1122 error = VOP_ACCESS(vp, accmode, cred);
1128 * Privileged processes may set the sticky bit on non-directories,
1129 * as well as set the setgid bit on a file with a group that the
1130 * process is not a member of.
1132 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1133 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1136 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1137 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1143 TMPFS_NODE_LOCK(node);
1144 node->tn_mode &= ~ALLPERMS;
1145 node->tn_mode |= mode & ALLPERMS;
1147 node->tn_status |= TMPFS_NODE_CHANGED;
1148 TMPFS_NODE_UNLOCK(node);
1150 KKASSERT(vn_islocked(vp));
1155 /* --------------------------------------------------------------------- */
1158 * Change ownership of the given vnode. At least one of uid or gid must
1159 * be different than VNOVAL. If one is set to that value, the attribute
1161 * Caller should execute tmpfs_update on vp after a successful execution.
1162 * The vnode must be locked on entry and remain locked on exit.
1165 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
1168 struct tmpfs_node *node;
1173 KKASSERT(vn_islocked(vp));
1175 node = VP_TO_TMPFS_NODE(vp);
1177 /* Assign default values if they are unknown. */
1178 KKASSERT(uid != VNOVAL || gid != VNOVAL);
1183 KKASSERT(uid != VNOVAL && gid != VNOVAL);
1185 /* Disallow this operation if the file system is mounted read-only. */
1186 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1189 /* Immutable or append-only files cannot be modified, either. */
1190 if (node->tn_flags & (IMMUTABLE | APPEND))
1193 fmode = FFLAGS(node->tn_flags);
1195 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
1197 if (fmode & (FWRITE | O_TRUNC)) {
1198 if (vp->v_type == VDIR) {
1201 error = vn_writechk(vp, NULL);
1210 error = VOP_ACCESS(vp, mode, cred);
1216 * To change the owner of a file, or change the group of a file to a
1217 * group of which we are not a member, the caller must have
1220 if ((uid != node->tn_uid ||
1221 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1222 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1225 ogid = node->tn_gid;
1226 ouid = node->tn_uid;
1228 TMPFS_NODE_LOCK(node);
1232 node->tn_status |= TMPFS_NODE_CHANGED;
1234 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1235 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1236 node->tn_mode &= ~(S_ISUID | S_ISGID);
1238 TMPFS_NODE_UNLOCK(node);
1240 KKASSERT(vn_islocked(vp));
1245 /* --------------------------------------------------------------------- */
1248 * Change size of the given vnode.
1249 * Caller should execute tmpfs_update on vp after a successful execution.
1250 * The vnode must be locked on entry and remain locked on exit.
1253 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred)
1256 struct tmpfs_node *node;
1258 KKASSERT(vn_islocked(vp));
1260 node = VP_TO_TMPFS_NODE(vp);
1262 /* Decide whether this is a valid operation based on the file type. */
1264 switch (vp->v_type) {
1269 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1278 /* Allow modifications of special files even if in the file
1279 * system is mounted read-only (we are not modifying the
1280 * files themselves, but the objects they represent). */
1284 /* Anything else is unsupported. */
1288 /* Immutable or append-only files cannot be modified, either. */
1289 if (node->tn_flags & (IMMUTABLE | APPEND))
1292 error = tmpfs_truncate(vp, size);
1293 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1294 * for us, as will update tn_status; no need to do that here. */
1296 KKASSERT(vn_islocked(vp));
1301 /* --------------------------------------------------------------------- */
1304 * Change access and modification times of the given vnode.
1305 * Caller should execute tmpfs_update on vp after a successful execution.
1306 * The vnode must be locked on entry and remain locked on exit.
1309 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1310 int vaflags, struct ucred *cred)
1313 struct tmpfs_node *node;
1316 KKASSERT(vn_islocked(vp));
1318 node = VP_TO_TMPFS_NODE(vp);
1320 /* Disallow this operation if the file system is mounted read-only. */
1321 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1324 /* Immutable or append-only files cannot be modified, either. */
1325 if (node->tn_flags & (IMMUTABLE | APPEND))
1328 /* Determine if the user have proper privilege to update time. */
1329 fmode = FFLAGS(node->tn_flags);
1331 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
1333 if (fmode & (FWRITE | O_TRUNC)) {
1334 if (vp->v_type == VDIR) {
1337 error = vn_writechk(vp, NULL);
1347 if (vaflags & VA_UTIMES_NULL) {
1348 error = VOP_ACCESS(vp, mode, cred);
1350 error = VOP_ACCESS(vp, VWRITE, cred);
1352 error = VOP_ACCESS(vp, mode, cred);
1357 TMPFS_NODE_LOCK(node);
1358 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1359 node->tn_status |= TMPFS_NODE_ACCESSED;
1361 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1362 node->tn_status |= TMPFS_NODE_MODIFIED;
1364 TMPFS_NODE_UNLOCK(node);
1366 tmpfs_itimes(vp, atime, mtime);
1368 KKASSERT(vn_islocked(vp));
1373 /* --------------------------------------------------------------------- */
1374 /* Sync timestamps */
1376 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1377 const struct timespec *mod)
1379 struct tmpfs_node *node;
1380 struct timespec now;
1382 node = VP_TO_TMPFS_NODE(vp);
1384 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1385 TMPFS_NODE_CHANGED)) == 0)
1388 vfs_timestamp(&now);
1390 TMPFS_NODE_LOCK(node);
1391 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1394 node->tn_atime = acc->tv_sec;
1395 node->tn_atimensec = acc->tv_nsec;
1397 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1400 node->tn_mtime = mod->tv_sec;
1401 node->tn_mtimensec = mod->tv_nsec;
1403 if (node->tn_status & TMPFS_NODE_CHANGED) {
1404 node->tn_ctime = now.tv_sec;
1405 node->tn_ctimensec = now.tv_nsec;
1408 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1409 TMPFS_NODE_UNLOCK(node);
1412 /* --------------------------------------------------------------------- */
1415 tmpfs_update(struct vnode *vp)
1418 tmpfs_itimes(vp, NULL, NULL);
1421 /* --------------------------------------------------------------------- */
1424 tmpfs_truncate(struct vnode *vp, off_t length)
1427 struct tmpfs_node *node;
1429 node = VP_TO_TMPFS_NODE(vp);
1436 if (node->tn_size == length) {
1441 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1445 error = tmpfs_reg_resize(vp, length, 1);
1448 TMPFS_NODE_LOCK(node);
1449 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1450 TMPFS_NODE_UNLOCK(node);
1459 /* --------------------------------------------------------------------- */
1462 tmpfs_fetch_ino(void)
1466 spin_lock_wr(&ino_lock);
1468 spin_unlock_wr(&ino_lock);