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>
37 __FBSDID("$FreeBSD$");
39 #include <sys/param.h>
40 #include <sys/namei.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 <fs/tmpfs/tmpfs.h>
55 #include <fs/tmpfs/tmpfs_fifoops.h>
56 #include <fs/tmpfs/tmpfs_vnops.h>
58 /* --------------------------------------------------------------------- */
61 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
62 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
63 * using the credentials of the process 'p'.
65 * If the node type is set to 'VDIR', then the parent parameter must point
66 * to the parent directory of the node being created. It may only be NULL
67 * while allocating the root node.
69 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
70 * specifies the device the node represents.
72 * If the node type is set to 'VLNK', then the parameter target specifies
73 * the file name of the target file for the symbolic link that is being
76 * Note that new nodes are retrieved from the available list if it has
77 * items or, if it is empty, from the node pool as long as there is enough
78 * space to create them.
80 * Returns zero on success or an appropriate error code on failure.
83 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
84 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
85 char *target, dev_t rdev, struct thread *p, struct tmpfs_node **node)
87 struct tmpfs_node *nnode;
89 /* If the root directory of the 'tmp' file system is not yet
90 * allocated, this must be the request to do it. */
91 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
93 MPASS(IFF(type == VLNK, target != NULL));
94 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
96 if (tmp->tm_nodes_inuse > tmp->tm_nodes_max)
99 nnode = (struct tmpfs_node *)uma_zalloc_arg(
100 tmp->tm_node_pool, tmp, M_WAITOK);
102 /* Generic initialization. */
103 nnode->tn_type = type;
104 vfs_timestamp(&nnode->tn_atime);
105 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
109 nnode->tn_mode = mode;
110 nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
112 /* Type-specific initialization. */
113 switch (nnode->tn_type) {
116 nnode->tn_rdev = rdev;
120 TAILQ_INIT(&nnode->tn_dir.tn_dirhead);
121 MPASS(parent != nnode);
122 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
123 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
124 nnode->tn_dir.tn_readdir_lastn = 0;
125 nnode->tn_dir.tn_readdir_lastp = NULL;
127 nnode->tn_dir.tn_parent->tn_links++;
136 MPASS(strlen(target) < MAXPATHLEN);
137 nnode->tn_size = strlen(target);
138 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
140 memcpy(nnode->tn_link, target, nnode->tn_size);
144 nnode->tn_reg.tn_aobj =
145 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0);
146 nnode->tn_reg.tn_aobj_pages = 0;
150 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
154 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
155 tmp->tm_nodes_inuse++;
162 /* --------------------------------------------------------------------- */
165 * Destroys the node pointed to by node from the file system 'tmp'.
166 * If the node does not belong to the given mount point, the results are
169 * If the node references a directory; no entries are allowed because
170 * their removal could need a recursive algorithm, something forbidden in
171 * kernel space. Furthermore, there is not need to provide such
172 * functionality (recursive removal) because the only primitives offered
173 * to the user are the removal of empty directories and the deletion of
176 * Note that nodes are not really deleted; in fact, when a node has been
177 * allocated, it cannot be deleted during the whole life of the file
178 * system. Instead, they are moved to the available list and remain there
182 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
187 TMPFS_NODE_LOCK(node);
188 MPASS(node->tn_vnode == NULL);
189 TMPFS_NODE_UNLOCK(node);
193 LIST_REMOVE(node, tn_entries);
194 tmp->tm_nodes_inuse--;
197 switch (node->tn_type) {
199 /* Do not do anything. VNON is provided to let the
200 * allocation routine clean itself easily by avoiding
201 * duplicating code in it. */
215 free(node->tn_link, M_TMPFSNAME);
219 if (node->tn_reg.tn_aobj != NULL)
220 vm_object_deallocate(node->tn_reg.tn_aobj);
221 pages = node->tn_reg.tn_aobj_pages;
225 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
228 free_unr(tmp->tm_ino_unr, node->tn_id);
229 uma_zfree(tmp->tm_node_pool, node);
232 tmp->tm_pages_used -= pages;
236 /* --------------------------------------------------------------------- */
239 * Allocates a new directory entry for the node node with a name of name.
240 * The new directory entry is returned in *de.
242 * The link count of node is increased by one to reflect the new object
245 * Returns zero on success or an appropriate error code on failure.
248 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
249 const char *name, uint16_t len, struct tmpfs_dirent **de)
251 struct tmpfs_dirent *nde;
253 nde = (struct tmpfs_dirent *)uma_zalloc(
254 tmp->tm_dirent_pool, M_WAITOK);
255 nde->td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
256 nde->td_namelen = len;
257 memcpy(nde->td_name, name, len);
267 /* --------------------------------------------------------------------- */
270 * Frees a directory entry. It is the caller's responsibility to destroy
271 * the node referenced by it if needed.
273 * The link count of node is decreased by one to reflect the removal of an
274 * object that referenced it. This only happens if 'node_exists' is true;
275 * otherwise the function will not access the node referred to by the
276 * directory entry, as it may already have been released from the outside.
279 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de,
280 boolean_t node_exists)
283 struct tmpfs_node *node;
287 MPASS(node->tn_links > 0);
291 free(de->td_name, M_TMPFSNAME);
292 uma_zfree(tmp->tm_dirent_pool, de);
295 /* --------------------------------------------------------------------- */
298 * Allocates a new vnode for the node node or returns a new reference to
299 * an existing one if the node had already a vnode referencing it. The
300 * resulting locked vnode is returned in *vpp.
302 * Returns zero on success or an appropriate error code on failure.
305 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
306 struct vnode **vpp, struct thread *td)
312 TMPFS_NODE_LOCK(node);
313 if ((vp = node->tn_vnode) != NULL) {
315 TMPFS_NODE_UNLOCK(node);
317 (void) vget(vp, lkflag | LK_INTERLOCK | LK_RETRY, td);
321 * Make sure the vnode is still there after
322 * getting the interlock to avoid racing a free.
324 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
333 * otherwise lock the vp list while we call getnewvnode
334 * since that can block.
336 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
337 node->tn_vpstate |= TMPFS_VNODE_WANT;
338 error = msleep((caddr_t) &node->tn_vpstate,
339 TMPFS_NODE_MTX(node), PDROP | PCATCH,
340 "tmpfs_alloc_vp", 0);
346 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
348 TMPFS_NODE_UNLOCK(node);
350 /* Get a new vnode and associate it with our node. */
351 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp);
356 (void) vn_lock(vp, lkflag | LK_RETRY);
359 vp->v_type = node->tn_type;
361 /* Type-specific initialization. */
362 switch (node->tn_type) {
374 vp->v_op = &tmpfs_fifoop_entries;
377 if (node->tn_dir.tn_parent == node)
378 vp->v_vflag |= VV_ROOT;
382 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
385 vnode_pager_setsize(vp, node->tn_size);
386 error = insmntque(vp, mp);
391 TMPFS_NODE_LOCK(node);
393 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
394 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
397 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
398 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
399 TMPFS_NODE_UNLOCK(node);
400 wakeup((caddr_t) &node->tn_vpstate);
402 TMPFS_NODE_UNLOCK(node);
407 MPASS(IFF(error == 0, *vpp != NULL && VOP_ISLOCKED(*vpp)));
409 TMPFS_NODE_LOCK(node);
410 MPASS(*vpp == node->tn_vnode);
411 TMPFS_NODE_UNLOCK(node);
417 /* --------------------------------------------------------------------- */
420 * Destroys the association between the vnode vp and the node it
424 tmpfs_free_vp(struct vnode *vp)
426 struct tmpfs_node *node;
428 node = VP_TO_TMPFS_NODE(vp);
430 TMPFS_NODE_LOCK(node);
431 node->tn_vnode = NULL;
433 TMPFS_NODE_UNLOCK(node);
436 /* --------------------------------------------------------------------- */
439 * Allocates a new file of type 'type' and adds it to the parent directory
440 * 'dvp'; this addition is done using the component name given in 'cnp'.
441 * The ownership of the new file is automatically assigned based on the
442 * credentials of the caller (through 'cnp'), the group is set based on
443 * the parent directory and the mode is determined from the 'vap' argument.
444 * If successful, *vpp holds a vnode to the newly created file and zero
445 * is returned. Otherwise *vpp is NULL and the function returns an
446 * appropriate error code.
449 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
450 struct componentname *cnp, char *target)
453 struct tmpfs_dirent *de;
454 struct tmpfs_mount *tmp;
455 struct tmpfs_node *dnode;
456 struct tmpfs_node *node;
457 struct tmpfs_node *parent;
459 MPASS(VOP_ISLOCKED(dvp));
460 MPASS(cnp->cn_flags & HASBUF);
462 tmp = VFS_TO_TMPFS(dvp->v_mount);
463 dnode = VP_TO_TMPFS_DIR(dvp);
466 /* If the entry we are creating is a directory, we cannot overflow
467 * the number of links of its parent, because it will get a new
469 if (vap->va_type == VDIR) {
470 /* Ensure that we do not overflow the maximum number of links
471 * imposed by the system. */
472 MPASS(dnode->tn_links <= LINK_MAX);
473 if (dnode->tn_links == LINK_MAX) {
479 MPASS(parent != NULL);
483 /* Allocate a node that represents the new file. */
484 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid,
485 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev,
486 cnp->cn_thread, &node);
490 /* Allocate a directory entry that points to the new file. */
491 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
494 tmpfs_free_node(tmp, node);
498 /* Allocate a vnode for the new file. */
499 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp,
502 tmpfs_free_dirent(tmp, de, TRUE);
503 tmpfs_free_node(tmp, node);
507 /* Now that all required items are allocated, we can proceed to
508 * insert the new node into the directory, an operation that
510 tmpfs_dir_attach(dvp, de);
517 /* --------------------------------------------------------------------- */
520 * Attaches the directory entry de to the directory represented by vp.
521 * Note that this does not change the link count of the node pointed by
522 * the directory entry, as this is done by tmpfs_alloc_dirent.
525 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
527 struct tmpfs_node *dnode;
529 ASSERT_VOP_ELOCKED(vp, __func__);
530 dnode = VP_TO_TMPFS_DIR(vp);
531 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
532 dnode->tn_size += sizeof(struct tmpfs_dirent);
533 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
537 /* --------------------------------------------------------------------- */
540 * Detaches the directory entry de from the directory represented by vp.
541 * Note that this does not change the link count of the node pointed by
542 * the directory entry, as this is done by tmpfs_free_dirent.
545 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
547 struct tmpfs_node *dnode;
549 ASSERT_VOP_ELOCKED(vp, __func__);
550 dnode = VP_TO_TMPFS_DIR(vp);
552 if (dnode->tn_dir.tn_readdir_lastp == de) {
553 dnode->tn_dir.tn_readdir_lastn = 0;
554 dnode->tn_dir.tn_readdir_lastp = NULL;
557 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
558 dnode->tn_size -= sizeof(struct tmpfs_dirent);
559 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
563 /* --------------------------------------------------------------------- */
566 * Looks for a directory entry in the directory represented by node.
567 * 'cnp' describes the name of the entry to look for. Note that the .
568 * and .. components are not allowed as they do not physically exist
569 * within directories.
571 * Returns a pointer to the entry when found, otherwise NULL.
573 struct tmpfs_dirent *
574 tmpfs_dir_lookup(struct tmpfs_node *node, struct componentname *cnp)
577 struct tmpfs_dirent *de;
579 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
580 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
581 cnp->cn_nameptr[1] == '.')));
582 TMPFS_VALIDATE_DIR(node);
585 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
586 MPASS(cnp->cn_namelen < 0xffff);
587 if (de->td_namelen == (uint16_t)cnp->cn_namelen &&
588 bcmp(de->td_name, cnp->cn_nameptr, de->td_namelen) == 0) {
593 node->tn_status |= TMPFS_NODE_ACCESSED;
595 return found ? de : NULL;
598 struct tmpfs_dirent *
599 tmpfs_dir_search(struct tmpfs_node *node, struct tmpfs_node *f)
601 struct tmpfs_dirent *de;
603 TMPFS_VALIDATE_DIR(node);
604 node->tn_status |= TMPFS_NODE_ACCESSED;
605 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
606 if (de->td_node == f)
612 /* --------------------------------------------------------------------- */
615 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
616 * directory and returns it in the uio space. The function returns 0
617 * on success, -1 if there was not enough space in the uio structure to
618 * hold the directory entry or an appropriate error code if another
622 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
627 TMPFS_VALIDATE_DIR(node);
628 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
630 dent.d_fileno = node->tn_id;
631 dent.d_type = DT_DIR;
633 dent.d_name[0] = '.';
634 dent.d_name[1] = '\0';
635 dent.d_reclen = GENERIC_DIRSIZ(&dent);
637 if (dent.d_reclen > uio->uio_resid)
640 error = uiomove(&dent, dent.d_reclen, uio);
642 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
645 node->tn_status |= TMPFS_NODE_ACCESSED;
650 /* --------------------------------------------------------------------- */
653 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
654 * directory and returns it in the uio space. The function returns 0
655 * on success, -1 if there was not enough space in the uio structure to
656 * hold the directory entry or an appropriate error code if another
660 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
665 TMPFS_VALIDATE_DIR(node);
666 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
668 dent.d_fileno = node->tn_dir.tn_parent->tn_id;
669 dent.d_type = DT_DIR;
671 dent.d_name[0] = '.';
672 dent.d_name[1] = '.';
673 dent.d_name[2] = '\0';
674 dent.d_reclen = GENERIC_DIRSIZ(&dent);
676 if (dent.d_reclen > uio->uio_resid)
679 error = uiomove(&dent, dent.d_reclen, uio);
681 struct tmpfs_dirent *de;
683 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
685 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
687 uio->uio_offset = tmpfs_dircookie(de);
691 node->tn_status |= TMPFS_NODE_ACCESSED;
696 /* --------------------------------------------------------------------- */
699 * Lookup a directory entry by its associated cookie.
701 struct tmpfs_dirent *
702 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
704 struct tmpfs_dirent *de;
706 if (cookie == node->tn_dir.tn_readdir_lastn &&
707 node->tn_dir.tn_readdir_lastp != NULL) {
708 return node->tn_dir.tn_readdir_lastp;
711 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
712 if (tmpfs_dircookie(de) == cookie) {
720 /* --------------------------------------------------------------------- */
723 * Helper function for tmpfs_readdir. Returns as much directory entries
724 * as can fit in the uio space. The read starts at uio->uio_offset.
725 * The function returns 0 on success, -1 if there was not enough space
726 * in the uio structure to hold the directory entry or an appropriate
727 * error code if another error happens.
730 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
734 struct tmpfs_dirent *de;
736 TMPFS_VALIDATE_DIR(node);
738 /* Locate the first directory entry we have to return. We have cached
739 * the last readdir in the node, so use those values if appropriate.
740 * Otherwise do a linear scan to find the requested entry. */
741 startcookie = uio->uio_offset;
742 MPASS(startcookie != TMPFS_DIRCOOKIE_DOT);
743 MPASS(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
744 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
747 de = tmpfs_dir_lookupbycookie(node, startcookie);
753 /* Read as much entries as possible; i.e., until we reach the end of
754 * the directory or we exhaust uio space. */
758 /* Create a dirent structure representing the current
759 * tmpfs_node and fill it. */
760 d.d_fileno = de->td_node->tn_id;
761 switch (de->td_node->tn_type) {
791 panic("tmpfs_dir_getdents: type %p %d",
792 de->td_node, (int)de->td_node->tn_type);
794 d.d_namlen = de->td_namelen;
795 MPASS(de->td_namelen < sizeof(d.d_name));
796 (void)memcpy(d.d_name, de->td_name, de->td_namelen);
797 d.d_name[de->td_namelen] = '\0';
798 d.d_reclen = GENERIC_DIRSIZ(&d);
800 /* Stop reading if the directory entry we are treating is
801 * bigger than the amount of data that can be returned. */
802 if (d.d_reclen > uio->uio_resid) {
807 /* Copy the new dirent structure into the output buffer and
808 * advance pointers. */
809 error = uiomove(&d, d.d_reclen, uio);
812 de = TAILQ_NEXT(de, td_entries);
813 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
815 /* Update the offset and cache. */
817 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
818 node->tn_dir.tn_readdir_lastn = 0;
819 node->tn_dir.tn_readdir_lastp = NULL;
821 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
822 node->tn_dir.tn_readdir_lastp = de;
825 node->tn_status |= TMPFS_NODE_ACCESSED;
829 /* --------------------------------------------------------------------- */
832 * Resizes the aobj associated to the regular file pointed to by vp to
833 * the size newsize. 'vp' must point to a vnode that represents a regular
834 * file. 'newsize' must be positive.
836 * Returns zero on success or an appropriate error code on failure.
839 tmpfs_reg_resize(struct vnode *vp, off_t newsize)
842 size_t newpages, oldpages;
843 struct tmpfs_mount *tmp;
844 struct tmpfs_node *node;
847 MPASS(vp->v_type == VREG);
850 node = VP_TO_TMPFS_NODE(vp);
851 tmp = VFS_TO_TMPFS(vp->v_mount);
853 /* Convert the old and new sizes to the number of pages needed to
854 * store them. It may happen that we do not need to do anything
855 * because the last allocated page can accommodate the change on
857 oldsize = node->tn_size;
858 oldpages = round_page(oldsize) / PAGE_SIZE;
859 MPASS(oldpages == node->tn_reg.tn_aobj_pages);
860 newpages = round_page(newsize) / PAGE_SIZE;
862 if (newpages > oldpages &&
863 newpages - oldpages > TMPFS_PAGES_AVAIL(tmp)) {
868 node->tn_reg.tn_aobj_pages = newpages;
871 tmp->tm_pages_used += (newpages - oldpages);
874 node->tn_size = newsize;
875 vnode_pager_setsize(vp, newsize);
876 if (newsize < oldsize) {
877 size_t zerolen = round_page(newsize) - newsize;
878 vm_object_t uobj = node->tn_reg.tn_aobj;
882 * free "backing store"
884 VM_OBJECT_LOCK(uobj);
885 if (newpages < oldpages) {
886 swap_pager_freespace(uobj,
887 newpages, oldpages - newpages);
888 vm_object_page_remove(uobj,
889 OFF_TO_IDX(newsize + PAGE_MASK), 0, FALSE);
893 * zero out the truncated part of the last page.
897 m = vm_page_grab(uobj, OFF_TO_IDX(newsize),
898 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
899 pmap_zero_page_area(m, PAGE_SIZE - zerolen,
903 VM_OBJECT_UNLOCK(uobj);
913 /* --------------------------------------------------------------------- */
916 * Change flags of the given vnode.
917 * Caller should execute tmpfs_update on vp after a successful execution.
918 * The vnode must be locked on entry and remain locked on exit.
921 tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred, struct thread *p)
924 struct tmpfs_node *node;
926 MPASS(VOP_ISLOCKED(vp));
928 node = VP_TO_TMPFS_NODE(vp);
930 /* Disallow this operation if the file system is mounted read-only. */
931 if (vp->v_mount->mnt_flag & MNT_RDONLY)
935 * Callers may only modify the file flags on objects they
936 * have VADMIN rights for.
938 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
941 * Unprivileged processes are not permitted to unset system
942 * flags, or modify flags if any system flags are set.
944 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
946 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
947 error = securelevel_gt(cred, 0);
951 /* Snapshot flag cannot be set or cleared */
952 if (((flags & SF_SNAPSHOT) != 0 &&
953 (node->tn_flags & SF_SNAPSHOT) == 0) ||
954 ((flags & SF_SNAPSHOT) == 0 &&
955 (node->tn_flags & SF_SNAPSHOT) != 0))
957 node->tn_flags = flags;
960 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
961 (flags & UF_SETTABLE) != flags)
963 node->tn_flags &= SF_SETTABLE;
964 node->tn_flags |= (flags & UF_SETTABLE);
966 node->tn_status |= TMPFS_NODE_CHANGED;
968 MPASS(VOP_ISLOCKED(vp));
973 /* --------------------------------------------------------------------- */
976 * Change access mode on the given vnode.
977 * Caller should execute tmpfs_update on vp after a successful execution.
978 * The vnode must be locked on entry and remain locked on exit.
981 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
984 struct tmpfs_node *node;
986 MPASS(VOP_ISLOCKED(vp));
988 node = VP_TO_TMPFS_NODE(vp);
990 /* Disallow this operation if the file system is mounted read-only. */
991 if (vp->v_mount->mnt_flag & MNT_RDONLY)
994 /* Immutable or append-only files cannot be modified, either. */
995 if (node->tn_flags & (IMMUTABLE | APPEND))
999 * To modify the permissions on a file, must possess VADMIN
1002 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1006 * Privileged processes may set the sticky bit on non-directories,
1007 * as well as set the setgid bit on a file with a group that the
1008 * process is not a member of.
1010 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1011 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1014 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1015 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1021 node->tn_mode &= ~ALLPERMS;
1022 node->tn_mode |= mode & ALLPERMS;
1024 node->tn_status |= TMPFS_NODE_CHANGED;
1026 MPASS(VOP_ISLOCKED(vp));
1031 /* --------------------------------------------------------------------- */
1034 * Change ownership of the given vnode. At least one of uid or gid must
1035 * be different than VNOVAL. If one is set to that value, the attribute
1037 * Caller should execute tmpfs_update on vp after a successful execution.
1038 * The vnode must be locked on entry and remain locked on exit.
1041 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1045 struct tmpfs_node *node;
1049 MPASS(VOP_ISLOCKED(vp));
1051 node = VP_TO_TMPFS_NODE(vp);
1053 /* Assign default values if they are unknown. */
1054 MPASS(uid != VNOVAL || gid != VNOVAL);
1059 MPASS(uid != VNOVAL && gid != VNOVAL);
1061 /* Disallow this operation if the file system is mounted read-only. */
1062 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1065 /* Immutable or append-only files cannot be modified, either. */
1066 if (node->tn_flags & (IMMUTABLE | APPEND))
1070 * To modify the ownership of a file, must possess VADMIN for that
1073 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1077 * To change the owner of a file, or change the group of a file to a
1078 * group of which we are not a member, the caller must have
1081 if ((uid != node->tn_uid ||
1082 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1083 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1086 ogid = node->tn_gid;
1087 ouid = node->tn_uid;
1092 node->tn_status |= TMPFS_NODE_CHANGED;
1094 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1095 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1096 node->tn_mode &= ~(S_ISUID | S_ISGID);
1099 MPASS(VOP_ISLOCKED(vp));
1104 /* --------------------------------------------------------------------- */
1107 * Change size of the given vnode.
1108 * Caller should execute tmpfs_update on vp after a successful execution.
1109 * The vnode must be locked on entry and remain locked on exit.
1112 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1116 struct tmpfs_node *node;
1118 MPASS(VOP_ISLOCKED(vp));
1120 node = VP_TO_TMPFS_NODE(vp);
1122 /* Decide whether this is a valid operation based on the file type. */
1124 switch (vp->v_type) {
1129 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1138 /* Allow modifications of special files even if in the file
1139 * system is mounted read-only (we are not modifying the
1140 * files themselves, but the objects they represent). */
1144 /* Anything else is unsupported. */
1148 /* Immutable or append-only files cannot be modified, either. */
1149 if (node->tn_flags & (IMMUTABLE | APPEND))
1152 error = tmpfs_truncate(vp, size);
1153 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1154 * for us, as will update tn_status; no need to do that here. */
1156 MPASS(VOP_ISLOCKED(vp));
1161 /* --------------------------------------------------------------------- */
1164 * Change access and modification times of the given vnode.
1165 * Caller should execute tmpfs_update on vp after a successful execution.
1166 * The vnode must be locked on entry and remain locked on exit.
1169 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1170 struct timespec *birthtime, int vaflags, struct ucred *cred, struct thread *l)
1173 struct tmpfs_node *node;
1175 MPASS(VOP_ISLOCKED(vp));
1177 node = VP_TO_TMPFS_NODE(vp);
1179 /* Disallow this operation if the file system is mounted read-only. */
1180 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1183 /* Immutable or append-only files cannot be modified, either. */
1184 if (node->tn_flags & (IMMUTABLE | APPEND))
1187 /* Determine if the user have proper privilege to update time. */
1188 if (vaflags & VA_UTIMES_NULL) {
1189 error = VOP_ACCESS(vp, VADMIN, cred, l);
1191 error = VOP_ACCESS(vp, VWRITE, cred, l);
1193 error = VOP_ACCESS(vp, VADMIN, cred, l);
1197 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1198 node->tn_status |= TMPFS_NODE_ACCESSED;
1200 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1201 node->tn_status |= TMPFS_NODE_MODIFIED;
1203 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1204 node->tn_status |= TMPFS_NODE_MODIFIED;
1206 tmpfs_itimes(vp, atime, mtime);
1208 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1209 node->tn_birthtime = *birthtime;
1210 MPASS(VOP_ISLOCKED(vp));
1215 /* --------------------------------------------------------------------- */
1216 /* Sync timestamps */
1218 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1219 const struct timespec *mod)
1221 struct tmpfs_node *node;
1222 struct timespec now;
1224 node = VP_TO_TMPFS_NODE(vp);
1226 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1227 TMPFS_NODE_CHANGED)) == 0)
1230 vfs_timestamp(&now);
1231 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1234 node->tn_atime = *acc;
1236 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1239 node->tn_mtime = *mod;
1241 if (node->tn_status & TMPFS_NODE_CHANGED) {
1242 node->tn_ctime = now;
1245 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1248 /* --------------------------------------------------------------------- */
1251 tmpfs_update(struct vnode *vp)
1254 tmpfs_itimes(vp, NULL, NULL);
1257 /* --------------------------------------------------------------------- */
1260 tmpfs_truncate(struct vnode *vp, off_t length)
1264 struct tmpfs_node *node;
1266 node = VP_TO_TMPFS_NODE(vp);
1267 extended = length > node->tn_size;
1274 if (node->tn_size == length) {
1279 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1282 error = tmpfs_reg_resize(vp, length);
1284 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;