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/fnv_hash.h>
42 #include <sys/namei.h>
45 #include <sys/random.h>
46 #include <sys/rwlock.h>
48 #include <sys/systm.h>
49 #include <sys/sysctl.h>
50 #include <sys/vnode.h>
51 #include <sys/vmmeter.h>
54 #include <vm/vm_param.h>
55 #include <vm/vm_object.h>
56 #include <vm/vm_page.h>
57 #include <vm/vm_pageout.h>
58 #include <vm/vm_pager.h>
59 #include <vm/vm_extern.h>
61 #include <fs/tmpfs/tmpfs.h>
62 #include <fs/tmpfs/tmpfs_fifoops.h>
63 #include <fs/tmpfs/tmpfs_vnops.h>
65 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system");
67 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
70 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
75 pages = *(long *)arg1;
76 bytes = pages * PAGE_SIZE;
78 error = sysctl_handle_long(oidp, &bytes, 0, req);
79 if (error || !req->newptr)
82 pages = bytes / PAGE_SIZE;
83 if (pages < TMPFS_PAGES_MINRESERVED)
86 *(long *)arg1 = pages;
90 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, CTLTYPE_LONG|CTLFLAG_RW,
91 &tmpfs_pages_reserved, 0, sysctl_mem_reserved, "L",
92 "Amount of available memory and swap below which tmpfs growth stops");
94 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a,
95 struct tmpfs_dirent *b);
96 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
103 avail = swap_pager_avail + vm_cnt.v_free_count - tmpfs_pages_reserved;
104 if (__predict_false(avail < 0))
110 tmpfs_pages_used(struct tmpfs_mount *tmp)
112 const size_t node_size = sizeof(struct tmpfs_node) +
113 sizeof(struct tmpfs_dirent);
116 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
118 return (meta_pages + tmp->tm_pages_used);
122 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
124 if (tmpfs_mem_avail() < req_pages)
127 if (tmp->tm_pages_max != ULONG_MAX &&
128 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
135 tmpfs_ref_node(struct tmpfs_node *node)
138 TMPFS_NODE_LOCK(node);
139 tmpfs_ref_node_locked(node);
140 TMPFS_NODE_UNLOCK(node);
144 tmpfs_ref_node_locked(struct tmpfs_node *node)
147 TMPFS_NODE_ASSERT_LOCKED(node);
148 KASSERT(node->tn_refcount > 0, ("node %p zero refcount", node));
149 KASSERT(node->tn_refcount < UINT_MAX, ("node %p refcount %u", node,
155 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
156 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
157 * using the credentials of the process 'p'.
159 * If the node type is set to 'VDIR', then the parent parameter must point
160 * to the parent directory of the node being created. It may only be NULL
161 * while allocating the root node.
163 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
164 * specifies the device the node represents.
166 * If the node type is set to 'VLNK', then the parameter target specifies
167 * the file name of the target file for the symbolic link that is being
170 * Note that new nodes are retrieved from the available list if it has
171 * items or, if it is empty, from the node pool as long as there is enough
172 * space to create them.
174 * Returns zero on success or an appropriate error code on failure.
177 tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type,
178 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
179 char *target, dev_t rdev, struct tmpfs_node **node)
181 struct tmpfs_node *nnode;
184 /* If the root directory of the 'tmp' file system is not yet
185 * allocated, this must be the request to do it. */
186 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
187 KASSERT(tmp->tm_root == NULL || mp->mnt_writeopcount > 0,
188 ("creating node not under vn_start_write"));
190 MPASS(IFF(type == VLNK, target != NULL));
191 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
193 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
195 if (tmpfs_pages_check_avail(tmp, 1) == 0)
198 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
200 * When a new tmpfs node is created for fully
201 * constructed mount point, there must be a parent
202 * node, which vnode is locked exclusively. As
203 * consequence, if the unmount is executing in
204 * parallel, vflush() cannot reclaim the parent vnode.
205 * Due to this, the check for MNTK_UNMOUNT flag is not
206 * racy: if we did not see MNTK_UNMOUNT flag, then tmp
207 * cannot be destroyed until node construction is
208 * finished and the parent vnode unlocked.
210 * Tmpfs does not need to instantiate new nodes during
216 nnode = (struct tmpfs_node *)uma_zalloc_arg(tmp->tm_node_pool, tmp,
219 /* Generic initialization. */
220 nnode->tn_type = type;
221 vfs_timestamp(&nnode->tn_atime);
222 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
226 nnode->tn_mode = mode;
227 nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
228 nnode->tn_refcount = 1;
230 /* Type-specific initialization. */
231 switch (nnode->tn_type) {
234 nnode->tn_rdev = rdev;
238 RB_INIT(&nnode->tn_dir.tn_dirhead);
239 LIST_INIT(&nnode->tn_dir.tn_dupindex);
240 MPASS(parent != nnode);
241 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
242 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
243 nnode->tn_dir.tn_readdir_lastn = 0;
244 nnode->tn_dir.tn_readdir_lastp = NULL;
246 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
247 nnode->tn_dir.tn_parent->tn_links++;
248 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
257 MPASS(strlen(target) < MAXPATHLEN);
258 nnode->tn_size = strlen(target);
259 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
261 memcpy(nnode->tn_link, target, nnode->tn_size);
265 obj = nnode->tn_reg.tn_aobj =
266 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
267 NULL /* XXXKIB - tmpfs needs swap reservation */);
268 VM_OBJECT_WLOCK(obj);
269 /* OBJ_TMPFS is set together with the setting of vp->v_object */
270 vm_object_set_flag(obj, OBJ_NOSPLIT | OBJ_TMPFS_NODE);
271 vm_object_clear_flag(obj, OBJ_ONEMAPPING);
272 VM_OBJECT_WUNLOCK(obj);
276 panic("tmpfs_alloc_node: type %p %d", nnode,
277 (int)nnode->tn_type);
281 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
282 nnode->tn_attached = true;
283 tmp->tm_nodes_inuse++;
292 * Destroys the node pointed to by node from the file system 'tmp'.
293 * If the node references a directory, no entries are allowed.
296 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
300 TMPFS_NODE_LOCK(node);
301 if (!tmpfs_free_node_locked(tmp, node, false)) {
302 TMPFS_NODE_UNLOCK(node);
308 tmpfs_free_node_locked(struct tmpfs_mount *tmp, struct tmpfs_node *node,
313 TMPFS_MP_ASSERT_LOCKED(tmp);
314 TMPFS_NODE_ASSERT_LOCKED(node);
315 KASSERT(node->tn_refcount > 0, ("node %p refcount zero", node));
318 if (node->tn_attached && (detach || node->tn_refcount == 0)) {
319 MPASS(tmp->tm_nodes_inuse > 0);
320 tmp->tm_nodes_inuse--;
321 LIST_REMOVE(node, tn_entries);
322 node->tn_attached = false;
324 if (node->tn_refcount > 0)
328 MPASS(node->tn_vnode == NULL);
329 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
331 TMPFS_NODE_UNLOCK(node);
334 switch (node->tn_type) {
347 free(node->tn_link, M_TMPFSNAME);
351 uobj = node->tn_reg.tn_aobj;
354 atomic_subtract_long(&tmp->tm_pages_used, uobj->size);
355 KASSERT((uobj->flags & OBJ_TMPFS) == 0,
356 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj));
357 vm_object_deallocate(uobj);
362 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
366 * If we are unmounting there is no need for going through the overhead
367 * of freeing the inodes from the unr individually, so free them all in
371 free_unr(tmp->tm_ino_unr, node->tn_id);
372 uma_zfree(tmp->tm_node_pool, node);
378 static __inline uint32_t
379 tmpfs_dirent_hash(const char *name, u_int len)
383 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK;
384 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP
387 if (hash < TMPFS_DIRCOOKIE_MIN)
388 hash += TMPFS_DIRCOOKIE_MIN;
393 static __inline off_t
394 tmpfs_dirent_cookie(struct tmpfs_dirent *de)
397 return (TMPFS_DIRCOOKIE_EOF);
399 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN);
401 return (de->td_cookie);
404 static __inline boolean_t
405 tmpfs_dirent_dup(struct tmpfs_dirent *de)
407 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0);
410 static __inline boolean_t
411 tmpfs_dirent_duphead(struct tmpfs_dirent *de)
413 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0);
417 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen)
419 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen);
420 memcpy(de->ud.td_name, name, namelen);
421 de->td_namelen = namelen;
425 * Allocates a new directory entry for the node node with a name of name.
426 * The new directory entry is returned in *de.
428 * The link count of node is increased by one to reflect the new object
431 * Returns zero on success or an appropriate error code on failure.
434 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
435 const char *name, u_int len, struct tmpfs_dirent **de)
437 struct tmpfs_dirent *nde;
439 nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK);
442 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
443 tmpfs_dirent_init(nde, name, len);
455 * Frees a directory entry. It is the caller's responsibility to destroy
456 * the node referenced by it if needed.
458 * The link count of node is decreased by one to reflect the removal of an
459 * object that referenced it. This only happens if 'node_exists' is true;
460 * otherwise the function will not access the node referred to by the
461 * directory entry, as it may already have been released from the outside.
464 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
466 struct tmpfs_node *node;
470 MPASS(node->tn_links > 0);
473 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL)
474 free(de->ud.td_name, M_TMPFSNAME);
475 uma_zfree(tmp->tm_dirent_pool, de);
479 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj)
482 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject");
483 if (vp->v_type != VREG || obj == NULL)
486 VM_OBJECT_WLOCK(obj);
488 vm_object_clear_flag(obj, OBJ_TMPFS);
489 obj->un_pager.swp.swp_tmpfs = NULL;
491 VM_OBJECT_WUNLOCK(obj);
495 * Need to clear v_object for insmntque failure.
498 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg)
501 tmpfs_destroy_vobject(vp, vp->v_object);
504 vp->v_op = &dead_vnodeops;
510 * Allocates a new vnode for the node node or returns a new reference to
511 * an existing one if the node had already a vnode referencing it. The
512 * resulting locked vnode is returned in *vpp.
514 * Returns zero on success or an appropriate error code on failure.
517 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
521 struct tmpfs_mount *tm;
526 tm = VFS_TO_TMPFS(mp);
527 TMPFS_NODE_LOCK(node);
528 tmpfs_ref_node_locked(node);
530 TMPFS_NODE_ASSERT_LOCKED(node);
531 if ((vp = node->tn_vnode) != NULL) {
532 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
534 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) ||
535 ((vp->v_iflag & VI_DOOMED) != 0 &&
536 (lkflag & LK_NOWAIT) != 0)) {
538 TMPFS_NODE_UNLOCK(node);
543 if ((vp->v_iflag & VI_DOOMED) != 0) {
545 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM;
546 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) {
547 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node),
552 TMPFS_NODE_UNLOCK(node);
553 error = vget(vp, lkflag | LK_INTERLOCK, curthread);
554 if (error == ENOENT) {
555 TMPFS_NODE_LOCK(node);
564 * Make sure the vnode is still there after
565 * getting the interlock to avoid racing a free.
567 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
569 TMPFS_NODE_LOCK(node);
576 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
577 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
578 TMPFS_NODE_UNLOCK(node);
585 * otherwise lock the vp list while we call getnewvnode
586 * since that can block.
588 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
589 node->tn_vpstate |= TMPFS_VNODE_WANT;
590 error = msleep((caddr_t) &node->tn_vpstate,
591 TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0);
596 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
598 TMPFS_NODE_UNLOCK(node);
600 /* Get a new vnode and associate it with our node. */
601 error = getnewvnode("tmpfs", mp, VFS_TO_TMPFS(mp)->tm_nonc ?
602 &tmpfs_vnodeop_nonc_entries : &tmpfs_vnodeop_entries, &vp);
607 /* lkflag is ignored, the lock is exclusive */
608 (void) vn_lock(vp, lkflag | LK_RETRY);
611 vp->v_type = node->tn_type;
613 /* Type-specific initialization. */
614 switch (node->tn_type) {
624 vp->v_op = &tmpfs_fifoop_entries;
627 object = node->tn_reg.tn_aobj;
628 VM_OBJECT_WLOCK(object);
630 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
631 vp->v_object = object;
632 object->un_pager.swp.swp_tmpfs = vp;
633 vm_object_set_flag(object, OBJ_TMPFS);
635 VM_OBJECT_WUNLOCK(object);
638 MPASS(node->tn_dir.tn_parent != NULL);
639 if (node->tn_dir.tn_parent == node)
640 vp->v_vflag |= VV_ROOT;
644 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
646 if (vp->v_type != VFIFO)
649 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL);
654 TMPFS_NODE_LOCK(node);
656 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
657 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
660 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
661 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
662 TMPFS_NODE_UNLOCK(node);
663 wakeup((caddr_t) &node->tn_vpstate);
665 TMPFS_NODE_UNLOCK(node);
672 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
673 TMPFS_NODE_LOCK(node);
674 MPASS(*vpp == node->tn_vnode);
675 TMPFS_NODE_UNLOCK(node);
678 tmpfs_free_node(tm, node);
684 * Destroys the association between the vnode vp and the node it
688 tmpfs_free_vp(struct vnode *vp)
690 struct tmpfs_node *node;
692 node = VP_TO_TMPFS_NODE(vp);
694 TMPFS_NODE_ASSERT_LOCKED(node);
695 node->tn_vnode = NULL;
696 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
697 wakeup(&node->tn_vnode);
698 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
703 * Allocates a new file of type 'type' and adds it to the parent directory
704 * 'dvp'; this addition is done using the component name given in 'cnp'.
705 * The ownership of the new file is automatically assigned based on the
706 * credentials of the caller (through 'cnp'), the group is set based on
707 * the parent directory and the mode is determined from the 'vap' argument.
708 * If successful, *vpp holds a vnode to the newly created file and zero
709 * is returned. Otherwise *vpp is NULL and the function returns an
710 * appropriate error code.
713 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
714 struct componentname *cnp, char *target)
717 struct tmpfs_dirent *de;
718 struct tmpfs_mount *tmp;
719 struct tmpfs_node *dnode;
720 struct tmpfs_node *node;
721 struct tmpfs_node *parent;
723 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file");
724 MPASS(cnp->cn_flags & HASBUF);
726 tmp = VFS_TO_TMPFS(dvp->v_mount);
727 dnode = VP_TO_TMPFS_DIR(dvp);
730 /* If the entry we are creating is a directory, we cannot overflow
731 * the number of links of its parent, because it will get a new
733 if (vap->va_type == VDIR) {
734 /* Ensure that we do not overflow the maximum number of links
735 * imposed by the system. */
736 MPASS(dnode->tn_links <= LINK_MAX);
737 if (dnode->tn_links == LINK_MAX) {
742 MPASS(parent != NULL);
746 /* Allocate a node that represents the new file. */
747 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type,
748 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent,
749 target, vap->va_rdev, &node);
753 /* Allocate a directory entry that points to the new file. */
754 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
757 tmpfs_free_node(tmp, node);
761 /* Allocate a vnode for the new file. */
762 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
764 tmpfs_free_dirent(tmp, de);
765 tmpfs_free_node(tmp, node);
769 /* Now that all required items are allocated, we can proceed to
770 * insert the new node into the directory, an operation that
772 if (cnp->cn_flags & ISWHITEOUT)
773 tmpfs_dir_whiteout_remove(dvp, cnp);
774 tmpfs_dir_attach(dvp, de);
778 struct tmpfs_dirent *
779 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
781 struct tmpfs_dirent *de;
783 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
785 if (de != NULL && tmpfs_dirent_duphead(de))
786 de = LIST_FIRST(&de->ud.td_duphead);
787 dc->tdc_current = de;
789 return (dc->tdc_current);
792 struct tmpfs_dirent *
793 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
795 struct tmpfs_dirent *de;
797 MPASS(dc->tdc_tree != NULL);
798 if (tmpfs_dirent_dup(dc->tdc_current)) {
799 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
800 if (dc->tdc_current != NULL)
801 return (dc->tdc_current);
803 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
804 &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
805 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
806 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
807 MPASS(dc->tdc_current != NULL);
810 return (dc->tdc_current);
813 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
814 static struct tmpfs_dirent *
815 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
817 struct tmpfs_dirent *de, dekey;
819 dekey.td_hash = hash;
820 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
824 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
825 static struct tmpfs_dirent *
826 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
827 struct tmpfs_dir_cursor *dc)
829 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
830 struct tmpfs_dirent *de, dekey;
832 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
834 if (cookie == node->tn_dir.tn_readdir_lastn &&
835 (de = node->tn_dir.tn_readdir_lastp) != NULL) {
836 /* Protect against possible race, tn_readdir_last[pn]
837 * may be updated with only shared vnode lock held. */
838 if (cookie == tmpfs_dirent_cookie(de))
842 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
843 LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
844 uh.td_dup.index_entries) {
845 MPASS(tmpfs_dirent_dup(de));
846 if (de->td_cookie == cookie)
848 /* dupindex list is sorted. */
849 if (de->td_cookie < cookie) {
858 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) {
861 dekey.td_hash = cookie;
862 /* Recover if direntry for cookie was removed */
863 de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
866 dc->tdc_current = de;
867 if (de != NULL && tmpfs_dirent_duphead(de)) {
868 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
869 MPASS(dc->tdc_current != NULL);
871 return (dc->tdc_current);
875 dc->tdc_current = de;
876 if (de != NULL && tmpfs_dirent_dup(de))
877 dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
879 return (dc->tdc_current);
883 * Looks for a directory entry in the directory represented by node.
884 * 'cnp' describes the name of the entry to look for. Note that the .
885 * and .. components are not allowed as they do not physically exist
886 * within directories.
888 * Returns a pointer to the entry when found, otherwise NULL.
890 struct tmpfs_dirent *
891 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
892 struct componentname *cnp)
894 struct tmpfs_dir_duphead *duphead;
895 struct tmpfs_dirent *de;
898 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
899 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
900 cnp->cn_nameptr[1] == '.')));
901 TMPFS_VALIDATE_DIR(node);
903 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
904 de = tmpfs_dir_xlookup_hash(node, hash);
905 if (de != NULL && tmpfs_dirent_duphead(de)) {
906 duphead = &de->ud.td_duphead;
907 LIST_FOREACH(de, duphead, uh.td_dup.entries) {
908 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
912 } else if (de != NULL) {
913 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
917 if (de != NULL && f != NULL && de->td_node != f)
924 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
925 * list, allocate new cookie value.
928 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
929 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
931 struct tmpfs_dir_duphead *dupindex;
932 struct tmpfs_dirent *de, *pde;
934 dupindex = &dnode->tn_dir.tn_dupindex;
935 de = LIST_FIRST(dupindex);
936 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
938 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
940 nde->td_cookie = de->td_cookie + 1;
941 MPASS(tmpfs_dirent_dup(nde));
942 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
943 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
948 * Cookie numbers are near exhaustion. Scan dupindex list for unused
949 * numbers. dupindex list is sorted in descending order. Keep it so
950 * after inserting nde.
954 de = LIST_NEXT(de, uh.td_dup.index_entries);
955 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
957 * Last element of the index doesn't have minimal cookie
960 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
961 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
962 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
964 } else if (de == NULL) {
966 * We are so lucky have 2^30 hash duplicates in single
967 * directory :) Return largest possible cookie value.
968 * It should be fine except possible issues with
969 * VOP_READDIR restart.
971 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
972 LIST_INSERT_HEAD(dupindex, nde,
973 uh.td_dup.index_entries);
974 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
977 if (de->td_cookie + 1 == pde->td_cookie ||
978 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
979 continue; /* No hole or invalid cookie. */
980 nde->td_cookie = de->td_cookie + 1;
981 MPASS(tmpfs_dirent_dup(nde));
982 MPASS(pde->td_cookie > nde->td_cookie);
983 MPASS(nde->td_cookie > de->td_cookie);
984 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
985 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
991 * Attaches the directory entry de to the directory represented by vp.
992 * Note that this does not change the link count of the node pointed by
993 * the directory entry, as this is done by tmpfs_alloc_dirent.
996 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
998 struct tmpfs_node *dnode;
999 struct tmpfs_dirent *xde, *nde;
1001 ASSERT_VOP_ELOCKED(vp, __func__);
1002 MPASS(de->td_namelen > 0);
1003 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
1004 MPASS(de->td_cookie == de->td_hash);
1006 dnode = VP_TO_TMPFS_DIR(vp);
1007 dnode->tn_dir.tn_readdir_lastn = 0;
1008 dnode->tn_dir.tn_readdir_lastp = NULL;
1010 MPASS(!tmpfs_dirent_dup(de));
1011 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1012 if (xde != NULL && tmpfs_dirent_duphead(xde))
1013 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1014 else if (xde != NULL) {
1016 * Allocate new duphead. Swap xde with duphead to avoid
1017 * adding/removing elements with the same hash.
1019 MPASS(!tmpfs_dirent_dup(xde));
1020 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
1022 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
1023 memcpy(nde, xde, sizeof(*xde));
1024 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
1025 LIST_INIT(&xde->ud.td_duphead);
1026 xde->td_namelen = 0;
1027 xde->td_node = NULL;
1028 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
1029 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1031 dnode->tn_size += sizeof(struct tmpfs_dirent);
1032 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1033 TMPFS_NODE_MODIFIED;
1038 * Detaches the directory entry de from the directory represented by vp.
1039 * Note that this does not change the link count of the node pointed by
1040 * the directory entry, as this is done by tmpfs_free_dirent.
1043 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
1045 struct tmpfs_mount *tmp;
1046 struct tmpfs_dir *head;
1047 struct tmpfs_node *dnode;
1048 struct tmpfs_dirent *xde;
1050 ASSERT_VOP_ELOCKED(vp, __func__);
1052 dnode = VP_TO_TMPFS_DIR(vp);
1053 head = &dnode->tn_dir.tn_dirhead;
1054 dnode->tn_dir.tn_readdir_lastn = 0;
1055 dnode->tn_dir.tn_readdir_lastp = NULL;
1057 if (tmpfs_dirent_dup(de)) {
1058 /* Remove duphead if de was last entry. */
1059 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1060 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1061 MPASS(tmpfs_dirent_duphead(xde));
1064 LIST_REMOVE(de, uh.td_dup.entries);
1065 LIST_REMOVE(de, uh.td_dup.index_entries);
1067 if (LIST_EMPTY(&xde->ud.td_duphead)) {
1068 RB_REMOVE(tmpfs_dir, head, xde);
1069 tmp = VFS_TO_TMPFS(vp->v_mount);
1070 MPASS(xde->td_node == NULL);
1071 tmpfs_free_dirent(tmp, xde);
1074 de->td_cookie = de->td_hash;
1076 RB_REMOVE(tmpfs_dir, head, de);
1078 dnode->tn_size -= sizeof(struct tmpfs_dirent);
1079 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1080 TMPFS_NODE_MODIFIED;
1085 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1087 struct tmpfs_dirent *de, *dde, *nde;
1089 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1090 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1091 /* Node may already be destroyed. */
1093 if (tmpfs_dirent_duphead(de)) {
1094 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1095 LIST_REMOVE(dde, uh.td_dup.entries);
1096 dde->td_node = NULL;
1097 tmpfs_free_dirent(tmp, dde);
1100 tmpfs_free_dirent(tmp, de);
1105 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
1106 * directory and returns it in the uio space. The function returns 0
1107 * on success, -1 if there was not enough space in the uio structure to
1108 * hold the directory entry or an appropriate error code if another
1112 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
1117 TMPFS_VALIDATE_DIR(node);
1118 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1120 dent.d_fileno = node->tn_id;
1121 dent.d_type = DT_DIR;
1123 dent.d_name[0] = '.';
1124 dent.d_name[1] = '\0';
1125 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1127 if (dent.d_reclen > uio->uio_resid)
1128 error = EJUSTRETURN;
1130 error = uiomove(&dent, dent.d_reclen, uio);
1132 tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
1138 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
1139 * directory and returns it in the uio space. The function returns 0
1140 * on success, -1 if there was not enough space in the uio structure to
1141 * hold the directory entry or an appropriate error code if another
1145 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
1150 TMPFS_VALIDATE_DIR(node);
1151 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1154 * Return ENOENT if the current node is already removed.
1156 TMPFS_ASSERT_LOCKED(node);
1157 if (node->tn_dir.tn_parent == NULL)
1160 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
1161 dent.d_fileno = node->tn_dir.tn_parent->tn_id;
1162 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
1164 dent.d_type = DT_DIR;
1166 dent.d_name[0] = '.';
1167 dent.d_name[1] = '.';
1168 dent.d_name[2] = '\0';
1169 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1171 if (dent.d_reclen > uio->uio_resid)
1172 error = EJUSTRETURN;
1174 error = uiomove(&dent, dent.d_reclen, uio);
1176 tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
1182 * Helper function for tmpfs_readdir. Returns as much directory entries
1183 * as can fit in the uio space. The read starts at uio->uio_offset.
1184 * The function returns 0 on success, -1 if there was not enough space
1185 * in the uio structure to hold the directory entry or an appropriate
1186 * error code if another error happens.
1189 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies,
1190 u_long *cookies, int *ncookies)
1192 struct tmpfs_dir_cursor dc;
1193 struct tmpfs_dirent *de;
1197 TMPFS_VALIDATE_DIR(node);
1202 * Lookup the node from the current offset. The starting offset of
1203 * 0 will lookup both '.' and '..', and then the first real entry,
1204 * or EOF if there are none. Then find all entries for the dir that
1205 * fit into the buffer. Once no more entries are found (de == NULL),
1206 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1209 switch (uio->uio_offset) {
1210 case TMPFS_DIRCOOKIE_DOT:
1211 error = tmpfs_dir_getdotdent(node, uio);
1214 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
1215 if (cookies != NULL)
1216 cookies[(*ncookies)++] = off = uio->uio_offset;
1218 case TMPFS_DIRCOOKIE_DOTDOT:
1219 error = tmpfs_dir_getdotdotdent(node, uio);
1222 de = tmpfs_dir_first(node, &dc);
1223 uio->uio_offset = tmpfs_dirent_cookie(de);
1224 if (cookies != NULL)
1225 cookies[(*ncookies)++] = off = uio->uio_offset;
1230 case TMPFS_DIRCOOKIE_EOF:
1233 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1236 if (cookies != NULL)
1237 off = tmpfs_dirent_cookie(de);
1240 /* Read as much entries as possible; i.e., until we reach the end of
1241 * the directory or we exhaust uio space. */
1245 /* Create a dirent structure representing the current
1246 * tmpfs_node and fill it. */
1247 if (de->td_node == NULL) {
1251 d.d_fileno = de->td_node->tn_id;
1252 switch (de->td_node->tn_type) {
1282 panic("tmpfs_dir_getdents: type %p %d",
1283 de->td_node, (int)de->td_node->tn_type);
1286 d.d_namlen = de->td_namelen;
1287 MPASS(de->td_namelen < sizeof(d.d_name));
1288 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1289 d.d_name[de->td_namelen] = '\0';
1290 d.d_reclen = GENERIC_DIRSIZ(&d);
1292 /* Stop reading if the directory entry we are treating is
1293 * bigger than the amount of data that can be returned. */
1294 if (d.d_reclen > uio->uio_resid) {
1295 error = EJUSTRETURN;
1299 /* Copy the new dirent structure into the output buffer and
1300 * advance pointers. */
1301 error = uiomove(&d, d.d_reclen, uio);
1303 de = tmpfs_dir_next(node, &dc);
1304 if (cookies != NULL) {
1305 off = tmpfs_dirent_cookie(de);
1306 MPASS(*ncookies < maxcookies);
1307 cookies[(*ncookies)++] = off;
1310 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
1312 /* Skip setting off when using cookies as it is already done above. */
1313 if (cookies == NULL)
1314 off = tmpfs_dirent_cookie(de);
1316 /* Update the offset and cache. */
1317 uio->uio_offset = off;
1318 node->tn_dir.tn_readdir_lastn = off;
1319 node->tn_dir.tn_readdir_lastp = de;
1321 tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
1326 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1328 struct tmpfs_dirent *de;
1331 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1332 cnp->cn_nameptr, cnp->cn_namelen, &de);
1335 tmpfs_dir_attach(dvp, de);
1340 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1342 struct tmpfs_dirent *de;
1344 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1345 MPASS(de != NULL && de->td_node == NULL);
1346 tmpfs_dir_detach(dvp, de);
1347 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1351 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1352 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
1353 * 'newsize' must be positive.
1355 * Returns zero on success or an appropriate error code on failure.
1358 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1360 struct tmpfs_mount *tmp;
1361 struct tmpfs_node *node;
1364 vm_pindex_t idx, newpages, oldpages;
1368 MPASS(vp->v_type == VREG);
1369 MPASS(newsize >= 0);
1371 node = VP_TO_TMPFS_NODE(vp);
1372 uobj = node->tn_reg.tn_aobj;
1373 tmp = VFS_TO_TMPFS(vp->v_mount);
1376 * Convert the old and new sizes to the number of pages needed to
1377 * store them. It may happen that we do not need to do anything
1378 * because the last allocated page can accommodate the change on
1381 oldsize = node->tn_size;
1382 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1383 MPASS(oldpages == uobj->size);
1384 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1386 if (__predict_true(newpages == oldpages && newsize >= oldsize)) {
1387 node->tn_size = newsize;
1391 if (newpages > oldpages &&
1392 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
1395 VM_OBJECT_WLOCK(uobj);
1396 if (newsize < oldsize) {
1398 * Zero the truncated part of the last page.
1400 base = newsize & PAGE_MASK;
1402 idx = OFF_TO_IDX(newsize);
1404 m = vm_page_lookup(uobj, idx);
1406 if (vm_page_sleep_if_busy(m, "tmfssz"))
1408 MPASS(m->valid == VM_PAGE_BITS_ALL);
1409 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1410 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL);
1412 VM_OBJECT_WUNLOCK(uobj);
1414 VM_OBJECT_WLOCK(uobj);
1417 rv = vm_pager_get_pages(uobj, &m, 1, NULL,
1420 if (rv == VM_PAGER_OK) {
1422 * Since the page was not resident,
1423 * and therefore not recently
1424 * accessed, immediately enqueue it
1425 * for asynchronous laundering. The
1426 * current operation is not regarded
1438 VM_OBJECT_WUNLOCK(uobj);
1444 pmap_zero_page_area(m, base, PAGE_SIZE - base);
1446 vm_pager_page_unswapped(m);
1451 * Release any swap space and free any whole pages.
1453 if (newpages < oldpages) {
1454 swap_pager_freespace(uobj, newpages, oldpages -
1456 vm_object_page_remove(uobj, newpages, 0, 0);
1459 uobj->size = newpages;
1460 VM_OBJECT_WUNLOCK(uobj);
1462 atomic_add_long(&tmp->tm_pages_used, newpages - oldpages);
1464 node->tn_size = newsize;
1469 tmpfs_check_mtime(struct vnode *vp)
1471 struct tmpfs_node *node;
1472 struct vm_object *obj;
1474 ASSERT_VOP_ELOCKED(vp, "check_mtime");
1475 if (vp->v_type != VREG)
1478 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
1479 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
1481 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) {
1482 VM_OBJECT_WLOCK(obj);
1483 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) {
1484 obj->flags &= ~OBJ_TMPFS_DIRTY;
1485 node = VP_TO_TMPFS_NODE(vp);
1486 node->tn_status |= TMPFS_NODE_MODIFIED |
1489 VM_OBJECT_WUNLOCK(obj);
1494 * Change flags of the given vnode.
1495 * Caller should execute tmpfs_update on vp after a successful execution.
1496 * The vnode must be locked on entry and remain locked on exit.
1499 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
1503 struct tmpfs_node *node;
1505 ASSERT_VOP_ELOCKED(vp, "chflags");
1507 node = VP_TO_TMPFS_NODE(vp);
1509 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK |
1510 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP |
1511 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
1512 UF_SPARSE | UF_SYSTEM)) != 0)
1513 return (EOPNOTSUPP);
1515 /* Disallow this operation if the file system is mounted read-only. */
1516 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1520 * Callers may only modify the file flags on objects they
1521 * have VADMIN rights for.
1523 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1526 * Unprivileged processes are not permitted to unset system
1527 * flags, or modify flags if any system flags are set.
1529 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1530 if (node->tn_flags &
1531 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1532 error = securelevel_gt(cred, 0);
1537 if (node->tn_flags &
1538 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1539 ((flags ^ node->tn_flags) & SF_SETTABLE))
1542 node->tn_flags = flags;
1543 node->tn_status |= TMPFS_NODE_CHANGED;
1545 ASSERT_VOP_ELOCKED(vp, "chflags2");
1551 * Change access mode on the given vnode.
1552 * Caller should execute tmpfs_update on vp after a successful execution.
1553 * The vnode must be locked on entry and remain locked on exit.
1556 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1559 struct tmpfs_node *node;
1561 ASSERT_VOP_ELOCKED(vp, "chmod");
1563 node = VP_TO_TMPFS_NODE(vp);
1565 /* Disallow this operation if the file system is mounted read-only. */
1566 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1569 /* Immutable or append-only files cannot be modified, either. */
1570 if (node->tn_flags & (IMMUTABLE | APPEND))
1574 * To modify the permissions on a file, must possess VADMIN
1577 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1581 * Privileged processes may set the sticky bit on non-directories,
1582 * as well as set the setgid bit on a file with a group that the
1583 * process is not a member of.
1585 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1586 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1589 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1590 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1596 node->tn_mode &= ~ALLPERMS;
1597 node->tn_mode |= mode & ALLPERMS;
1599 node->tn_status |= TMPFS_NODE_CHANGED;
1601 ASSERT_VOP_ELOCKED(vp, "chmod2");
1607 * Change ownership of the given vnode. At least one of uid or gid must
1608 * be different than VNOVAL. If one is set to that value, the attribute
1610 * Caller should execute tmpfs_update on vp after a successful execution.
1611 * The vnode must be locked on entry and remain locked on exit.
1614 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1618 struct tmpfs_node *node;
1622 ASSERT_VOP_ELOCKED(vp, "chown");
1624 node = VP_TO_TMPFS_NODE(vp);
1626 /* Assign default values if they are unknown. */
1627 MPASS(uid != VNOVAL || gid != VNOVAL);
1632 MPASS(uid != VNOVAL && gid != VNOVAL);
1634 /* Disallow this operation if the file system is mounted read-only. */
1635 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1638 /* Immutable or append-only files cannot be modified, either. */
1639 if (node->tn_flags & (IMMUTABLE | APPEND))
1643 * To modify the ownership of a file, must possess VADMIN for that
1646 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1650 * To change the owner of a file, or change the group of a file to a
1651 * group of which we are not a member, the caller must have
1654 if ((uid != node->tn_uid ||
1655 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1656 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1659 ogid = node->tn_gid;
1660 ouid = node->tn_uid;
1665 node->tn_status |= TMPFS_NODE_CHANGED;
1667 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1668 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1669 node->tn_mode &= ~(S_ISUID | S_ISGID);
1672 ASSERT_VOP_ELOCKED(vp, "chown2");
1678 * Change size of the given vnode.
1679 * Caller should execute tmpfs_update on vp after a successful execution.
1680 * The vnode must be locked on entry and remain locked on exit.
1683 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1687 struct tmpfs_node *node;
1689 ASSERT_VOP_ELOCKED(vp, "chsize");
1691 node = VP_TO_TMPFS_NODE(vp);
1693 /* Decide whether this is a valid operation based on the file type. */
1695 switch (vp->v_type) {
1700 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1709 /* Allow modifications of special files even if in the file
1710 * system is mounted read-only (we are not modifying the
1711 * files themselves, but the objects they represent). */
1715 /* Anything else is unsupported. */
1719 /* Immutable or append-only files cannot be modified, either. */
1720 if (node->tn_flags & (IMMUTABLE | APPEND))
1723 error = tmpfs_truncate(vp, size);
1724 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1725 * for us, as will update tn_status; no need to do that here. */
1727 ASSERT_VOP_ELOCKED(vp, "chsize2");
1733 * Change access and modification times of the given vnode.
1734 * Caller should execute tmpfs_update on vp after a successful execution.
1735 * The vnode must be locked on entry and remain locked on exit.
1738 tmpfs_chtimes(struct vnode *vp, struct vattr *vap,
1739 struct ucred *cred, struct thread *l)
1742 struct tmpfs_node *node;
1744 ASSERT_VOP_ELOCKED(vp, "chtimes");
1746 node = VP_TO_TMPFS_NODE(vp);
1748 /* Disallow this operation if the file system is mounted read-only. */
1749 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1752 /* Immutable or append-only files cannot be modified, either. */
1753 if (node->tn_flags & (IMMUTABLE | APPEND))
1756 error = vn_utimes_perm(vp, vap, cred, l);
1760 if (vap->va_atime.tv_sec != VNOVAL)
1761 node->tn_status |= TMPFS_NODE_ACCESSED;
1763 if (vap->va_mtime.tv_sec != VNOVAL)
1764 node->tn_status |= TMPFS_NODE_MODIFIED;
1766 if (vap->va_birthtime.tv_sec != VNOVAL)
1767 node->tn_status |= TMPFS_NODE_MODIFIED;
1769 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime);
1771 if (vap->va_birthtime.tv_sec != VNOVAL)
1772 node->tn_birthtime = vap->va_birthtime;
1773 ASSERT_VOP_ELOCKED(vp, "chtimes2");
1779 tmpfs_set_status(struct tmpfs_node *node, int status)
1782 if ((node->tn_status & status) == status)
1784 TMPFS_NODE_LOCK(node);
1785 node->tn_status |= status;
1786 TMPFS_NODE_UNLOCK(node);
1789 /* Sync timestamps */
1791 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1792 const struct timespec *mod)
1794 struct tmpfs_node *node;
1795 struct timespec now;
1797 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes");
1798 node = VP_TO_TMPFS_NODE(vp);
1800 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1801 TMPFS_NODE_CHANGED)) == 0)
1804 vfs_timestamp(&now);
1805 TMPFS_NODE_LOCK(node);
1806 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1809 node->tn_atime = *acc;
1811 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1814 node->tn_mtime = *mod;
1816 if (node->tn_status & TMPFS_NODE_CHANGED)
1817 node->tn_ctime = now;
1818 node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1819 TMPFS_NODE_CHANGED);
1820 TMPFS_NODE_UNLOCK(node);
1822 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
1823 random_harvest_queue(node, sizeof(*node), 1, RANDOM_FS_ATIME);
1827 tmpfs_update(struct vnode *vp)
1830 tmpfs_itimes(vp, NULL, NULL);
1834 tmpfs_truncate(struct vnode *vp, off_t length)
1837 struct tmpfs_node *node;
1839 node = VP_TO_TMPFS_NODE(vp);
1846 if (node->tn_size == length) {
1851 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1854 error = tmpfs_reg_resize(vp, length, FALSE);
1856 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1865 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1867 if (a->td_hash > b->td_hash)
1869 else if (a->td_hash < b->td_hash)
1874 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);