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_vnops.h>
57 static ino_t tmpfs_fetch_ino(struct tmpfs_mount *);
58 static int tmpfs_dirtree_compare(struct tmpfs_dirent *a,
59 struct tmpfs_dirent *b);
61 RB_GENERATE(tmpfs_dirtree, tmpfs_dirent, rb_node, tmpfs_dirtree_compare);
64 /* --------------------------------------------------------------------- */
67 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
68 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
69 * using the credentials of the process 'p'.
71 * If the node type is set to 'VDIR', then the parent parameter must point
72 * to the parent directory of the node being created. It may only be NULL
73 * while allocating the root node.
75 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
76 * specifies the device the node represents.
78 * If the node type is set to 'VLNK', then the parameter target specifies
79 * the file name of the target file for the symbolic link that is being
82 * Note that new nodes are retrieved from the available list if it has
83 * items or, if it is empty, from the node pool as long as there is enough
84 * space to create them.
86 * Returns zero on success or an appropriate error code on failure.
89 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
90 uid_t uid, gid_t gid, mode_t mode,
91 char *target, int rmajor, int rminor,
92 struct tmpfs_node **node)
94 struct tmpfs_node *nnode;
98 KKASSERT(IFF(type == VLNK, target != NULL));
99 KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL));
101 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
104 nnode = objcache_get(tmp->tm_node_pool, M_WAITOK | M_NULLOK);
108 /* Generic initialization. */
109 nnode->tn_type = type;
111 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime
113 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec
117 nnode->tn_mode = mode;
118 nnode->tn_id = tmpfs_fetch_ino(tmp);
119 nnode->tn_advlock.init_done = 0;
120 KKASSERT(nnode->tn_links == 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 RB_INIT(&nnode->tn_dir.tn_dirtree);
136 nnode->tn_dir.tn_readdir_lastn = 0;
137 nnode->tn_dir.tn_readdir_lastp = NULL;
147 nnode->tn_size = strlen(target);
148 nnode->tn_link = kmalloc(nnode->tn_size + 1, tmp->tm_name_zone,
149 M_WAITOK | M_NULLOK);
150 if (nnode->tn_link == NULL) {
151 objcache_put(tmp->tm_node_pool, nnode);
154 bcopy(target, nnode->tn_link, nnode->tn_size);
155 nnode->tn_link[nnode->tn_size] = '\0';
159 nnode->tn_reg.tn_aobj =
160 swap_pager_alloc(NULL, 0, VM_PROT_DEFAULT, 0);
161 nnode->tn_reg.tn_aobj_pages = 0;
166 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
169 TMPFS_NODE_LOCK(nnode);
171 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
172 tmp->tm_nodes_inuse++;
174 TMPFS_NODE_UNLOCK(nnode);
180 /* --------------------------------------------------------------------- */
183 * Destroys the node pointed to by node from the file system 'tmp'.
184 * If the node does not belong to the given mount point, the results are
187 * If the node references a directory; no entries are allowed because
188 * their removal could need a recursive algorithm, something forbidden in
189 * kernel space. Furthermore, there is not need to provide such
190 * functionality (recursive removal) because the only primitives offered
191 * to the user are the removal of empty directories and the deletion of
194 * Note that nodes are not really deleted; in fact, when a node has been
195 * allocated, it cannot be deleted during the whole life of the file
196 * system. Instead, they are moved to the available list and remain there
200 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
202 vm_pindex_t pages = 0;
205 TMPFS_ASSERT_ELOCKED(node);
206 KKASSERT(node->tn_vnode == NULL);
207 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
211 LIST_REMOVE(node, tn_entries);
212 tmp->tm_nodes_inuse--;
214 TMPFS_NODE_UNLOCK(node);
216 switch (node->tn_type) {
218 /* Do not do anything. VNON is provided to let the
219 * allocation routine clean itself easily by avoiding
220 * duplicating code in it. */
229 * The parent link can be NULL if this is the root
230 * node or if it is a directory node that was rmdir'd.
232 * XXX what if node is a directory which still contains
233 * directory entries (e.g. due to a forced umount) ?
236 KKASSERT(node->tn_dir.tn_parent == NULL);
239 * If the root node is being destroyed don't leave a
240 * dangling pointer in tmpfs_mount.
242 if (node == tmp->tm_root)
251 kfree(node->tn_link, tmp->tm_name_zone);
252 node->tn_link = NULL;
257 if (node->tn_reg.tn_aobj != NULL)
258 vm_object_deallocate(node->tn_reg.tn_aobj);
259 node->tn_reg.tn_aobj = NULL;
260 pages = node->tn_reg.tn_aobj_pages;
264 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
268 * Clean up fields for the next allocation. The objcache only ctors
271 tmpfs_node_ctor(node, NULL, 0);
272 objcache_put(tmp->tm_node_pool, node);
273 /* node is now invalid */
276 tmp->tm_pages_used -= pages;
280 /* --------------------------------------------------------------------- */
283 * Allocates a new directory entry for the node node with a name of name.
284 * The new directory entry is returned in *de.
286 * The link count of node is increased by one to reflect the new object
289 * Returns zero on success or an appropriate error code on failure.
292 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
293 const char *name, uint16_t len, struct tmpfs_dirent **de)
295 struct tmpfs_dirent *nde;
297 nde = objcache_get(tmp->tm_dirent_pool, M_WAITOK);
298 nde->td_name = kmalloc(len + 1, tmp->tm_name_zone, M_WAITOK | M_NULLOK);
299 if (nde->td_name == NULL) {
300 objcache_put(tmp->tm_dirent_pool, nde);
304 nde->td_namelen = len;
305 bcopy(name, nde->td_name, len);
306 nde->td_name[len] = '\0';
310 TMPFS_NODE_LOCK(node);
311 ++node->tn_links; /* also requires mnt_token protection */
312 TMPFS_NODE_UNLOCK(node);
319 /* --------------------------------------------------------------------- */
322 * Frees a directory entry. It is the caller's responsibility to destroy
323 * the node referenced by it if needed.
325 * The link count of node is decreased by one to reflect the removal of an
326 * object that referenced it. This only happens if 'node_exists' is true;
327 * otherwise the function will not access the node referred to by the
328 * directory entry, as it may already have been released from the outside.
331 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
333 struct tmpfs_node *node;
337 TMPFS_NODE_LOCK(node);
338 TMPFS_ASSERT_ELOCKED(node);
339 KKASSERT(node->tn_links > 0);
340 node->tn_links--; /* also requires mnt_token protection */
341 TMPFS_NODE_UNLOCK(node);
343 kfree(de->td_name, tmp->tm_name_zone);
347 objcache_put(tmp->tm_dirent_pool, de);
350 /* --------------------------------------------------------------------- */
353 * Allocates a new vnode for the node node or returns a new reference to
354 * an existing one if the node had already a vnode referencing it. The
355 * resulting locked vnode is returned in *vpp.
357 * Returns zero on success or an appropriate error code on failure.
360 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
368 * Interlocked extraction from node. This can race many things.
369 * We have to get a soft reference on the vnode while we hold
370 * the node locked, then acquire it properly and check for races.
372 TMPFS_NODE_LOCK(node);
373 if ((vp = node->tn_vnode) != NULL) {
374 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
375 vhold_interlocked(vp);
376 TMPFS_NODE_UNLOCK(node);
378 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) {
382 if (node->tn_vnode != vp) {
393 * This should never happen.
395 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) {
396 TMPFS_NODE_UNLOCK(node);
402 * Interlock against other calls to tmpfs_alloc_vp() trying to
403 * allocate and assign a vp to node.
405 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
406 node->tn_vpstate |= TMPFS_VNODE_WANT;
407 error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH,
408 "tmpfs_alloc_vp", 0);
409 TMPFS_NODE_UNLOCK(node);
414 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
415 TMPFS_NODE_UNLOCK(node);
418 * Allocate a new vnode (may block). The ALLOCATING flag should
419 * prevent a race against someone else assigning node->tn_vnode.
421 error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE);
425 KKASSERT(node->tn_vnode == NULL);
426 KKASSERT(vp != NULL);
428 vp->v_type = node->tn_type;
430 /* Type-specific initialization. */
431 switch (node->tn_type) {
439 vinitvmio(vp, node->tn_size, BMASK, -1);
444 vp->v_ops = &mp->mnt_vn_fifo_ops;
450 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
456 TMPFS_NODE_LOCK(node);
458 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
459 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
462 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
463 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
464 TMPFS_NODE_UNLOCK(node);
465 wakeup(&node->tn_vpstate);
467 TMPFS_NODE_UNLOCK(node);
473 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
475 TMPFS_NODE_LOCK(node);
476 KKASSERT(*vpp == node->tn_vnode);
477 TMPFS_NODE_UNLOCK(node);
483 /* --------------------------------------------------------------------- */
486 * Destroys the association between the vnode vp and the node it
490 tmpfs_free_vp(struct vnode *vp)
492 struct tmpfs_node *node;
494 node = VP_TO_TMPFS_NODE(vp);
496 TMPFS_NODE_LOCK(node);
497 KKASSERT(lockcount(TMPFS_NODE_MTX(node)) > 0);
498 node->tn_vnode = NULL;
500 TMPFS_NODE_UNLOCK(node);
503 /* --------------------------------------------------------------------- */
506 * Allocates a new file of type 'type' and adds it to the parent directory
507 * 'dvp'; this addition is done using the component name given in 'cnp'.
508 * The ownership of the new file is automatically assigned based on the
509 * credentials of the caller (through 'cnp'), the group is set based on
510 * the parent directory and the mode is determined from the 'vap' argument.
511 * If successful, *vpp holds a vnode to the newly created file and zero
512 * is returned. Otherwise *vpp is NULL and the function returns an
513 * appropriate error code.
516 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
517 struct namecache *ncp, struct ucred *cred, char *target)
520 struct tmpfs_dirent *de;
521 struct tmpfs_mount *tmp;
522 struct tmpfs_node *dnode;
523 struct tmpfs_node *node;
525 tmp = VFS_TO_TMPFS(dvp->v_mount);
526 dnode = VP_TO_TMPFS_DIR(dvp);
530 * If the directory was removed but a process was CD'd into it,
531 * we do not allow any more file/dir creation within it. Otherwise
532 * we will lose track of it.
534 KKASSERT(dnode->tn_type == VDIR);
535 if (dnode != tmp->tm_root && dnode->tn_dir.tn_parent == NULL)
539 * Make sure the link count does not overflow.
541 if (vap->va_type == VDIR && dnode->tn_links >= LINK_MAX)
544 /* Allocate a node that represents the new file. */
545 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid,
546 dnode->tn_gid, vap->va_mode, target,
547 vap->va_rmajor, vap->va_rminor, &node);
550 TMPFS_NODE_LOCK(node);
552 /* Allocate a directory entry that points to the new file. */
553 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de);
555 tmpfs_free_node(tmp, node);
560 /* Allocate a vnode for the new file. */
561 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
563 tmpfs_free_dirent(tmp, de);
564 tmpfs_free_node(tmp, node);
570 * Now that all required items are allocated, we can proceed to
571 * insert the new node into the directory, an operation that
574 tmpfs_dir_attach(dnode, de);
575 TMPFS_NODE_UNLOCK(node);
580 /* --------------------------------------------------------------------- */
583 * Attaches the directory entry de to the directory represented by vp.
584 * Note that this does not change the link count of the node pointed by
585 * the directory entry, as this is done by tmpfs_alloc_dirent.
588 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
590 struct tmpfs_node *node = de->td_node;
592 TMPFS_NODE_LOCK(dnode);
593 if (node && node->tn_type == VDIR) {
594 TMPFS_NODE_LOCK(node);
595 ++node->tn_links; /* also requires mnt_token protection */
596 node->tn_status |= TMPFS_NODE_CHANGED;
597 node->tn_dir.tn_parent = dnode;
598 ++dnode->tn_links; /* also requires mnt_token protection */
599 TMPFS_NODE_UNLOCK(node);
601 RB_INSERT(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de);
602 dnode->tn_size += sizeof(struct tmpfs_dirent);
603 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
605 TMPFS_NODE_UNLOCK(dnode);
608 /* --------------------------------------------------------------------- */
611 * Detaches the directory entry de from the directory represented by vp.
612 * Note that this does not change the link count of the node pointed by
613 * the directory entry, as this is done by tmpfs_free_dirent.
616 tmpfs_dir_detach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
618 struct tmpfs_node *node = de->td_node;
620 TMPFS_NODE_LOCK(dnode);
621 if (dnode->tn_dir.tn_readdir_lastp == de) {
622 dnode->tn_dir.tn_readdir_lastn = 0;
623 dnode->tn_dir.tn_readdir_lastp = NULL;
625 RB_REMOVE(tmpfs_dirtree, &dnode->tn_dir.tn_dirtree, de);
626 dnode->tn_size -= sizeof(struct tmpfs_dirent);
627 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
629 TMPFS_NODE_UNLOCK(dnode);
632 * Clean out the tn_parent pointer immediately when removing a
635 * Removal of the parent linkage also cleans out the extra tn_links
636 * count we had on both node and dnode.
638 * node can be NULL (typ during a forced umount), in which case
639 * the mount code is dealing with the linkages from a linked list
642 if (node && node->tn_type == VDIR && node->tn_dir.tn_parent) {
643 TMPFS_NODE_LOCK(dnode);
644 TMPFS_NODE_LOCK(node);
645 KKASSERT(node->tn_dir.tn_parent == dnode);
646 dnode->tn_links--; /* also requires mnt_token protection */
647 node->tn_links--; /* also requires mnt_token protection */
648 node->tn_dir.tn_parent = NULL;
649 TMPFS_NODE_UNLOCK(node);
650 TMPFS_NODE_UNLOCK(dnode);
654 /* --------------------------------------------------------------------- */
657 * Looks for a directory entry in the directory represented by node.
658 * 'ncp' describes the name of the entry to look for. Note that the .
659 * and .. components are not allowed as they do not physically exist
660 * within directories.
662 * Returns a pointer to the entry when found, otherwise NULL.
664 struct tmpfs_dirent *
665 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
666 struct namecache *ncp)
668 struct tmpfs_dirent *de;
669 int len = ncp->nc_nlen;
670 struct tmpfs_dirent wanted;
672 wanted.td_namelen = len;
673 wanted.td_name = ncp->nc_name;
675 TMPFS_VALIDATE_DIR(node);
677 de = RB_FIND(tmpfs_dirtree, &node->tn_dir.tn_dirtree, &wanted);
679 KKASSERT(f == NULL || f == de->td_node);
681 TMPFS_NODE_LOCK(node);
682 node->tn_status |= TMPFS_NODE_ACCESSED;
683 TMPFS_NODE_UNLOCK(node);
688 /* --------------------------------------------------------------------- */
691 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
692 * directory and returns it in the uio space. The function returns 0
693 * on success, -1 if there was not enough space in the uio structure to
694 * hold the directory entry or an appropriate error code if another
698 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
704 TMPFS_VALIDATE_DIR(node);
705 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
707 dent.d_ino = node->tn_id;
708 dent.d_type = DT_DIR;
710 dent.d_name[0] = '.';
711 dent.d_name[1] = '\0';
712 dirsize = _DIRENT_DIRSIZ(&dent);
714 if (dirsize > uio->uio_resid)
717 error = uiomove((caddr_t)&dent, dirsize, uio);
719 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
722 TMPFS_NODE_LOCK(node);
723 node->tn_status |= TMPFS_NODE_ACCESSED;
724 TMPFS_NODE_UNLOCK(node);
729 /* --------------------------------------------------------------------- */
732 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
733 * directory and returns it in the uio space. The function returns 0
734 * on success, -1 if there was not enough space in the uio structure to
735 * hold the directory entry or an appropriate error code if another
739 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
746 TMPFS_VALIDATE_DIR(node);
747 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
749 if (node->tn_dir.tn_parent) {
750 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
751 dent.d_ino = node->tn_dir.tn_parent->tn_id;
752 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
754 dent.d_ino = tmp->tm_root->tn_id;
757 dent.d_type = DT_DIR;
759 dent.d_name[0] = '.';
760 dent.d_name[1] = '.';
761 dent.d_name[2] = '\0';
762 dirsize = _DIRENT_DIRSIZ(&dent);
764 if (dirsize > uio->uio_resid)
767 error = uiomove((caddr_t)&dent, dirsize, uio);
769 struct tmpfs_dirent *de;
771 de = RB_MIN(tmpfs_dirtree, &node->tn_dir.tn_dirtree);
773 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
775 uio->uio_offset = tmpfs_dircookie(de);
779 TMPFS_NODE_LOCK(node);
780 node->tn_status |= TMPFS_NODE_ACCESSED;
781 TMPFS_NODE_UNLOCK(node);
786 /* --------------------------------------------------------------------- */
789 * Lookup a directory entry by its associated cookie.
791 struct tmpfs_dirent *
792 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
794 struct tmpfs_dirent *de;
796 if (cookie == node->tn_dir.tn_readdir_lastn &&
797 node->tn_dir.tn_readdir_lastp != NULL) {
798 return node->tn_dir.tn_readdir_lastp;
801 RB_FOREACH(de, tmpfs_dirtree, &node->tn_dir.tn_dirtree) {
802 if (tmpfs_dircookie(de) == cookie) {
810 /* --------------------------------------------------------------------- */
813 * Helper function for tmpfs_readdir. Returns as much directory entries
814 * as can fit in the uio space. The read starts at uio->uio_offset.
815 * The function returns 0 on success, -1 if there was not enough space
816 * in the uio structure to hold the directory entry or an appropriate
817 * error code if another error happens.
820 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
824 struct tmpfs_dirent *de;
826 TMPFS_VALIDATE_DIR(node);
828 /* Locate the first directory entry we have to return. We have cached
829 * the last readdir in the node, so use those values if appropriate.
830 * Otherwise do a linear scan to find the requested entry. */
831 startcookie = uio->uio_offset;
832 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT);
833 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
834 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
837 de = tmpfs_dir_lookupbycookie(node, startcookie);
843 /* Read as much entries as possible; i.e., until we reach the end of
844 * the directory or we exhaust uio space. */
849 /* Create a dirent structure representing the current
850 * tmpfs_node and fill it. */
851 d.d_ino = de->td_node->tn_id;
852 switch (de->td_node->tn_type) {
882 panic("tmpfs_dir_getdents: type %p %d",
883 de->td_node, (int)de->td_node->tn_type);
885 d.d_namlen = de->td_namelen;
886 KKASSERT(de->td_namelen < sizeof(d.d_name));
887 bcopy(de->td_name, d.d_name, d.d_namlen);
888 d.d_name[d.d_namlen] = '\0';
889 reclen = _DIRENT_RECLEN(d.d_namlen);
891 /* Stop reading if the directory entry we are treating is
892 * bigger than the amount of data that can be returned. */
893 if (reclen > uio->uio_resid) {
898 /* Copy the new dirent structure into the output buffer and
899 * advance pointers. */
900 error = uiomove((caddr_t)&d, reclen, uio);
903 de = RB_NEXT(tmpfs_dirtree, node->tn_dir.tn_dirtree, de);
904 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
906 /* Update the offset and cache. */
908 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
909 node->tn_dir.tn_readdir_lastn = 0;
910 node->tn_dir.tn_readdir_lastp = NULL;
912 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
913 node->tn_dir.tn_readdir_lastp = de;
915 node->tn_status |= TMPFS_NODE_ACCESSED;
920 /* --------------------------------------------------------------------- */
923 * Resizes the aobj associated to the regular file pointed to by vp to
924 * the size newsize. 'vp' must point to a vnode that represents a regular
925 * file. 'newsize' must be positive.
927 * pass trivial as 1 when buf content will be overwritten, otherwise set 0
930 * Returns zero on success or an appropriate error code on failure.
933 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial)
936 vm_pindex_t newpages, oldpages;
937 struct tmpfs_mount *tmp;
938 struct tmpfs_node *node;
942 KKASSERT(vp->v_type == VREG);
943 KKASSERT(newsize >= 0);
946 node = VP_TO_TMPFS_NODE(vp);
947 tmp = VFS_TO_TMPFS(vp->v_mount);
949 /* Convert the old and new sizes to the number of pages needed to
950 * store them. It may happen that we do not need to do anything
951 * because the last allocated page can accommodate the change on
953 oldsize = node->tn_size;
954 oldpages = round_page64(oldsize) / PAGE_SIZE;
955 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages);
956 newpages = round_page64(newsize) / PAGE_SIZE;
958 if (newpages > oldpages &&
959 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) {
965 tmp->tm_pages_used += (newpages - oldpages);
968 TMPFS_NODE_LOCK(node);
969 node->tn_reg.tn_aobj_pages = newpages;
970 node->tn_size = newsize;
971 TMPFS_NODE_UNLOCK(node);
974 * When adjusting the vnode filesize and its VM object we must
975 * also adjust our backing VM object (aobj). The blocksize
976 * used must match the block sized we use for the buffer cache.
978 * The backing VM object contains no VM pages, only swap
981 if (newsize < oldsize) {
986 error = nvtruncbuf(vp, newsize, BSIZE, -1, 0);
987 aobj = node->tn_reg.tn_aobj;
990 nsize = vp->v_object->size;
993 swap_pager_freespace(aobj, nsize,
1000 error = nvextendbuf(vp, oldsize, newsize, BSIZE, BSIZE,
1002 aobj = node->tn_reg.tn_aobj;
1004 aobj->size = vp->v_object->size;
1011 /* --------------------------------------------------------------------- */
1014 * Change flags of the given vnode.
1015 * Caller should execute tmpfs_update on vp after a successful execution.
1016 * The vnode must be locked on entry and remain locked on exit.
1019 tmpfs_chflags(struct vnode *vp, int vaflags, struct ucred *cred)
1022 struct tmpfs_node *node;
1025 KKASSERT(vn_islocked(vp));
1027 node = VP_TO_TMPFS_NODE(vp);
1028 flags = node->tn_flags;
1030 /* Disallow this operation if the file system is mounted read-only. */
1031 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1033 error = vop_helper_setattr_flags(&flags, vaflags, node->tn_uid, cred);
1036 * Unprivileged processes are not permitted to unset system
1037 * flags, or modify flags if any system flags are set.
1039 * Silently enforce SF_NOCACHE on the root tmpfs vnode so
1040 * tmpfs data is not double-cached by swapcache.
1043 TMPFS_NODE_LOCK(node);
1044 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1045 if (vp->v_flag & VROOT)
1046 flags |= SF_NOCACHE;
1047 node->tn_flags = flags;
1049 if (node->tn_flags & (SF_NOUNLINK | SF_IMMUTABLE |
1051 (flags & UF_SETTABLE) != flags) {
1054 node->tn_flags &= SF_SETTABLE;
1055 node->tn_flags |= (flags & UF_SETTABLE);
1058 node->tn_status |= TMPFS_NODE_CHANGED;
1059 TMPFS_NODE_UNLOCK(node);
1062 KKASSERT(vn_islocked(vp));
1067 /* --------------------------------------------------------------------- */
1070 * Change access mode on the given vnode.
1071 * Caller should execute tmpfs_update on vp after a successful execution.
1072 * The vnode must be locked on entry and remain locked on exit.
1075 tmpfs_chmod(struct vnode *vp, mode_t vamode, struct ucred *cred)
1077 struct tmpfs_node *node;
1081 KKASSERT(vn_islocked(vp));
1083 node = VP_TO_TMPFS_NODE(vp);
1085 /* Disallow this operation if the file system is mounted read-only. */
1086 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1089 /* Immutable or append-only files cannot be modified, either. */
1090 if (node->tn_flags & (IMMUTABLE | APPEND))
1093 cur_mode = node->tn_mode;
1094 error = vop_helper_chmod(vp, vamode, cred, node->tn_uid, node->tn_gid,
1098 (node->tn_mode & ALLPERMS) != (cur_mode & ALLPERMS)) {
1099 TMPFS_NODE_LOCK(node);
1100 node->tn_mode &= ~ALLPERMS;
1101 node->tn_mode |= cur_mode & ALLPERMS;
1103 node->tn_status |= TMPFS_NODE_CHANGED;
1104 TMPFS_NODE_UNLOCK(node);
1107 KKASSERT(vn_islocked(vp));
1112 /* --------------------------------------------------------------------- */
1115 * Change ownership of the given vnode. At least one of uid or gid must
1116 * be different than VNOVAL. If one is set to that value, the attribute
1118 * Caller should execute tmpfs_update on vp after a successful execution.
1119 * The vnode must be locked on entry and remain locked on exit.
1122 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
1127 struct tmpfs_node *node;
1130 KKASSERT(vn_islocked(vp));
1131 node = VP_TO_TMPFS_NODE(vp);
1133 /* Disallow this operation if the file system is mounted read-only. */
1134 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1137 /* Immutable or append-only files cannot be modified, either. */
1138 if (node->tn_flags & (IMMUTABLE | APPEND))
1141 cur_uid = node->tn_uid;
1142 cur_gid = node->tn_gid;
1143 cur_mode = node->tn_mode;
1144 error = vop_helper_chown(vp, uid, gid, cred,
1145 &cur_uid, &cur_gid, &cur_mode);
1148 TMPFS_NODE_LOCK(node);
1149 if (cur_uid != node->tn_uid ||
1150 cur_gid != node->tn_gid ||
1151 cur_mode != node->tn_mode) {
1152 node->tn_uid = cur_uid;
1153 node->tn_gid = cur_gid;
1154 node->tn_mode = cur_mode;
1155 node->tn_status |= TMPFS_NODE_CHANGED;
1157 TMPFS_NODE_UNLOCK(node);
1163 /* --------------------------------------------------------------------- */
1166 * Change size of the given vnode.
1167 * Caller should execute tmpfs_update on vp after a successful execution.
1168 * The vnode must be locked on entry and remain locked on exit.
1171 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred)
1174 struct tmpfs_node *node;
1176 KKASSERT(vn_islocked(vp));
1178 node = VP_TO_TMPFS_NODE(vp);
1180 /* Decide whether this is a valid operation based on the file type. */
1182 switch (vp->v_type) {
1187 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1196 /* Allow modifications of special files even if in the file
1197 * system is mounted read-only (we are not modifying the
1198 * files themselves, but the objects they represent). */
1202 /* Anything else is unsupported. */
1206 /* Immutable or append-only files cannot be modified, either. */
1207 if (node->tn_flags & (IMMUTABLE | APPEND))
1210 error = tmpfs_truncate(vp, size);
1211 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1212 * for us, as will update tn_status; no need to do that here. */
1214 KKASSERT(vn_islocked(vp));
1219 /* --------------------------------------------------------------------- */
1222 * Change access and modification times of the given vnode.
1223 * Caller should execute tmpfs_update on vp after a successful execution.
1224 * The vnode must be locked on entry and remain locked on exit.
1227 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1228 int vaflags, struct ucred *cred)
1230 struct tmpfs_node *node;
1232 KKASSERT(vn_islocked(vp));
1234 node = VP_TO_TMPFS_NODE(vp);
1236 /* Disallow this operation if the file system is mounted read-only. */
1237 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1240 /* Immutable or append-only files cannot be modified, either. */
1241 if (node->tn_flags & (IMMUTABLE | APPEND))
1244 TMPFS_NODE_LOCK(node);
1245 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1246 node->tn_status |= TMPFS_NODE_ACCESSED;
1248 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1249 node->tn_status |= TMPFS_NODE_MODIFIED;
1251 TMPFS_NODE_UNLOCK(node);
1253 tmpfs_itimes(vp, atime, mtime);
1255 KKASSERT(vn_islocked(vp));
1260 /* --------------------------------------------------------------------- */
1261 /* Sync timestamps */
1263 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1264 const struct timespec *mod)
1266 struct tmpfs_node *node;
1267 struct timespec now;
1269 node = VP_TO_TMPFS_NODE(vp);
1271 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1272 TMPFS_NODE_CHANGED)) == 0)
1275 vfs_timestamp(&now);
1277 TMPFS_NODE_LOCK(node);
1278 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1281 node->tn_atime = acc->tv_sec;
1282 node->tn_atimensec = acc->tv_nsec;
1284 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1287 node->tn_mtime = mod->tv_sec;
1288 node->tn_mtimensec = mod->tv_nsec;
1290 if (node->tn_status & TMPFS_NODE_CHANGED) {
1291 node->tn_ctime = now.tv_sec;
1292 node->tn_ctimensec = now.tv_nsec;
1295 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1296 TMPFS_NODE_UNLOCK(node);
1299 /* --------------------------------------------------------------------- */
1302 tmpfs_update(struct vnode *vp)
1305 tmpfs_itimes(vp, NULL, NULL);
1308 /* --------------------------------------------------------------------- */
1311 tmpfs_truncate(struct vnode *vp, off_t length)
1314 struct tmpfs_node *node;
1316 node = VP_TO_TMPFS_NODE(vp);
1323 if (node->tn_size == length) {
1328 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1332 error = tmpfs_reg_resize(vp, length, 1);
1335 TMPFS_NODE_LOCK(node);
1336 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1337 TMPFS_NODE_UNLOCK(node);
1346 /* --------------------------------------------------------------------- */
1349 tmpfs_fetch_ino(struct tmpfs_mount *tmp)
1353 ret = tmp->tm_ino++;
1359 tmpfs_dirtree_compare(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1361 if (a->td_namelen > b->td_namelen)
1363 else if (a->td_namelen < b->td_namelen)
1366 return strncmp(a->td_name, b->td_name, a->td_namelen);