1 /* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */
4 * SPDX-License-Identifier: BSD-2-Clause-NetBSD
6 * Copyright (c) 2005 The NetBSD Foundation, Inc.
9 * This code is derived from software contributed to The NetBSD Foundation
10 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
22 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
24 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
25 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
26 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGE.
36 * Efficient memory file system supporting functions.
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/dirent.h>
44 #include <sys/fnv_hash.h>
46 #include <sys/limits.h>
47 #include <sys/mount.h>
48 #include <sys/namei.h>
51 #include <sys/random.h>
52 #include <sys/rwlock.h>
54 #include <sys/sysctl.h>
55 #include <sys/vnode.h>
56 #include <sys/vmmeter.h>
59 #include <vm/vm_param.h>
60 #include <vm/vm_object.h>
61 #include <vm/vm_page.h>
62 #include <vm/vm_pageout.h>
63 #include <vm/vm_pager.h>
64 #include <vm/vm_extern.h>
65 #include <vm/swap_pager.h>
67 #include <fs/tmpfs/tmpfs.h>
68 #include <fs/tmpfs/tmpfs_fifoops.h>
69 #include <fs/tmpfs/tmpfs_vnops.h>
71 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system");
73 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
76 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
81 pages = *(long *)arg1;
82 bytes = pages * PAGE_SIZE;
84 error = sysctl_handle_long(oidp, &bytes, 0, req);
85 if (error || !req->newptr)
88 pages = bytes / PAGE_SIZE;
89 if (pages < TMPFS_PAGES_MINRESERVED)
92 *(long *)arg1 = pages;
96 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, CTLTYPE_LONG|CTLFLAG_RW,
97 &tmpfs_pages_reserved, 0, sysctl_mem_reserved, "L",
98 "Amount of available memory and swap below which tmpfs growth stops");
100 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a,
101 struct tmpfs_dirent *b);
102 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
105 tmpfs_mem_avail(void)
109 avail = swap_pager_avail + vm_free_count() - tmpfs_pages_reserved;
110 if (__predict_false(avail < 0))
116 tmpfs_pages_used(struct tmpfs_mount *tmp)
118 const size_t node_size = sizeof(struct tmpfs_node) +
119 sizeof(struct tmpfs_dirent);
122 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
124 return (meta_pages + tmp->tm_pages_used);
128 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
130 if (tmpfs_mem_avail() < req_pages)
133 if (tmp->tm_pages_max != ULONG_MAX &&
134 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
141 tmpfs_ref_node(struct tmpfs_node *node)
144 TMPFS_NODE_LOCK(node);
145 tmpfs_ref_node_locked(node);
146 TMPFS_NODE_UNLOCK(node);
150 tmpfs_ref_node_locked(struct tmpfs_node *node)
153 TMPFS_NODE_ASSERT_LOCKED(node);
154 KASSERT(node->tn_refcount > 0, ("node %p zero refcount", node));
155 KASSERT(node->tn_refcount < UINT_MAX, ("node %p refcount %u", node,
161 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
162 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
163 * using the credentials of the process 'p'.
165 * If the node type is set to 'VDIR', then the parent parameter must point
166 * to the parent directory of the node being created. It may only be NULL
167 * while allocating the root node.
169 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
170 * specifies the device the node represents.
172 * If the node type is set to 'VLNK', then the parameter target specifies
173 * the file name of the target file for the symbolic link that is being
176 * Note that new nodes are retrieved from the available list if it has
177 * items or, if it is empty, from the node pool as long as there is enough
178 * space to create them.
180 * Returns zero on success or an appropriate error code on failure.
183 tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type,
184 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
185 char *target, dev_t rdev, struct tmpfs_node **node)
187 struct tmpfs_node *nnode;
190 /* If the root directory of the 'tmp' file system is not yet
191 * allocated, this must be the request to do it. */
192 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
193 KASSERT(tmp->tm_root == NULL || mp->mnt_writeopcount > 0,
194 ("creating node not under vn_start_write"));
196 MPASS(IFF(type == VLNK, target != NULL));
197 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
199 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
201 if (tmpfs_pages_check_avail(tmp, 1) == 0)
204 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
206 * When a new tmpfs node is created for fully
207 * constructed mount point, there must be a parent
208 * node, which vnode is locked exclusively. As
209 * consequence, if the unmount is executing in
210 * parallel, vflush() cannot reclaim the parent vnode.
211 * Due to this, the check for MNTK_UNMOUNT flag is not
212 * racy: if we did not see MNTK_UNMOUNT flag, then tmp
213 * cannot be destroyed until node construction is
214 * finished and the parent vnode unlocked.
216 * Tmpfs does not need to instantiate new nodes during
221 if ((mp->mnt_kern_flag & MNT_RDONLY) != 0)
224 nnode = (struct tmpfs_node *)uma_zalloc_arg(tmp->tm_node_pool, tmp,
227 /* Generic initialization. */
228 nnode->tn_type = type;
229 vfs_timestamp(&nnode->tn_atime);
230 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
234 nnode->tn_mode = mode;
235 nnode->tn_id = alloc_unr64(&tmp->tm_ino_unr);
236 nnode->tn_refcount = 1;
238 /* Type-specific initialization. */
239 switch (nnode->tn_type) {
242 nnode->tn_rdev = rdev;
246 RB_INIT(&nnode->tn_dir.tn_dirhead);
247 LIST_INIT(&nnode->tn_dir.tn_dupindex);
248 MPASS(parent != nnode);
249 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
250 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
251 nnode->tn_dir.tn_readdir_lastn = 0;
252 nnode->tn_dir.tn_readdir_lastp = NULL;
254 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
255 nnode->tn_dir.tn_parent->tn_links++;
256 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
265 MPASS(strlen(target) < MAXPATHLEN);
266 nnode->tn_size = strlen(target);
267 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
269 memcpy(nnode->tn_link, target, nnode->tn_size);
273 obj = nnode->tn_reg.tn_aobj =
274 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
275 NULL /* XXXKIB - tmpfs needs swap reservation */);
276 VM_OBJECT_WLOCK(obj);
277 /* OBJ_TMPFS is set together with the setting of vp->v_object */
278 vm_object_set_flag(obj, OBJ_NOSPLIT | OBJ_TMPFS_NODE);
279 vm_object_clear_flag(obj, OBJ_ONEMAPPING);
280 VM_OBJECT_WUNLOCK(obj);
284 panic("tmpfs_alloc_node: type %p %d", nnode,
285 (int)nnode->tn_type);
289 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
290 nnode->tn_attached = true;
291 tmp->tm_nodes_inuse++;
300 * Destroys the node pointed to by node from the file system 'tmp'.
301 * If the node references a directory, no entries are allowed.
304 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
308 TMPFS_NODE_LOCK(node);
309 if (!tmpfs_free_node_locked(tmp, node, false)) {
310 TMPFS_NODE_UNLOCK(node);
316 tmpfs_free_node_locked(struct tmpfs_mount *tmp, struct tmpfs_node *node,
321 TMPFS_MP_ASSERT_LOCKED(tmp);
322 TMPFS_NODE_ASSERT_LOCKED(node);
323 KASSERT(node->tn_refcount > 0, ("node %p refcount zero", node));
326 if (node->tn_attached && (detach || node->tn_refcount == 0)) {
327 MPASS(tmp->tm_nodes_inuse > 0);
328 tmp->tm_nodes_inuse--;
329 LIST_REMOVE(node, tn_entries);
330 node->tn_attached = false;
332 if (node->tn_refcount > 0)
336 MPASS(node->tn_vnode == NULL);
337 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
339 TMPFS_NODE_UNLOCK(node);
342 switch (node->tn_type) {
355 free(node->tn_link, M_TMPFSNAME);
359 uobj = node->tn_reg.tn_aobj;
362 atomic_subtract_long(&tmp->tm_pages_used, uobj->size);
363 KASSERT((uobj->flags & OBJ_TMPFS) == 0,
364 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj));
365 vm_object_deallocate(uobj);
370 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
373 uma_zfree(tmp->tm_node_pool, node);
379 static __inline uint32_t
380 tmpfs_dirent_hash(const char *name, u_int len)
384 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK;
385 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP
388 if (hash < TMPFS_DIRCOOKIE_MIN)
389 hash += TMPFS_DIRCOOKIE_MIN;
394 static __inline off_t
395 tmpfs_dirent_cookie(struct tmpfs_dirent *de)
398 return (TMPFS_DIRCOOKIE_EOF);
400 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN);
402 return (de->td_cookie);
405 static __inline boolean_t
406 tmpfs_dirent_dup(struct tmpfs_dirent *de)
408 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0);
411 static __inline boolean_t
412 tmpfs_dirent_duphead(struct tmpfs_dirent *de)
414 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0);
418 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen)
420 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen);
421 memcpy(de->ud.td_name, name, namelen);
422 de->td_namelen = namelen;
426 * Allocates a new directory entry for the node node with a name of name.
427 * The new directory entry is returned in *de.
429 * The link count of node is increased by one to reflect the new object
432 * Returns zero on success or an appropriate error code on failure.
435 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
436 const char *name, u_int len, struct tmpfs_dirent **de)
438 struct tmpfs_dirent *nde;
440 nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK);
443 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
444 tmpfs_dirent_init(nde, name, len);
456 * Frees a directory entry. It is the caller's responsibility to destroy
457 * the node referenced by it if needed.
459 * The link count of node is decreased by one to reflect the removal of an
460 * object that referenced it. This only happens if 'node_exists' is true;
461 * otherwise the function will not access the node referred to by the
462 * directory entry, as it may already have been released from the outside.
465 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
467 struct tmpfs_node *node;
471 MPASS(node->tn_links > 0);
474 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL)
475 free(de->ud.td_name, M_TMPFSNAME);
476 uma_zfree(tmp->tm_dirent_pool, de);
480 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj)
483 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject");
484 if (vp->v_type != VREG || obj == NULL)
487 VM_OBJECT_WLOCK(obj);
489 vm_object_clear_flag(obj, OBJ_TMPFS);
490 obj->un_pager.swp.swp_tmpfs = NULL;
491 if (vp->v_writecount < 0)
492 vp->v_writecount = 0;
494 VM_OBJECT_WUNLOCK(obj);
498 * Need to clear v_object for insmntque failure.
501 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg)
504 tmpfs_destroy_vobject(vp, vp->v_object);
507 vp->v_op = &dead_vnodeops;
513 * Allocates a new vnode for the node node or returns a new reference to
514 * an existing one if the node had already a vnode referencing it. The
515 * resulting locked vnode is returned in *vpp.
517 * Returns zero on success or an appropriate error code on failure.
520 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
524 struct tmpfs_mount *tm;
529 tm = VFS_TO_TMPFS(mp);
530 TMPFS_NODE_LOCK(node);
531 tmpfs_ref_node_locked(node);
533 TMPFS_NODE_ASSERT_LOCKED(node);
534 if ((vp = node->tn_vnode) != NULL) {
535 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
537 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) ||
538 ((vp->v_iflag & VI_DOOMED) != 0 &&
539 (lkflag & LK_NOWAIT) != 0)) {
541 TMPFS_NODE_UNLOCK(node);
546 if ((vp->v_iflag & VI_DOOMED) != 0) {
548 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM;
549 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) {
550 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node),
555 TMPFS_NODE_UNLOCK(node);
556 error = vget(vp, lkflag | LK_INTERLOCK, curthread);
557 if (error == ENOENT) {
558 TMPFS_NODE_LOCK(node);
567 * Make sure the vnode is still there after
568 * getting the interlock to avoid racing a free.
570 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
572 TMPFS_NODE_LOCK(node);
579 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
580 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
581 TMPFS_NODE_UNLOCK(node);
588 * otherwise lock the vp list while we call getnewvnode
589 * since that can block.
591 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
592 node->tn_vpstate |= TMPFS_VNODE_WANT;
593 error = msleep((caddr_t) &node->tn_vpstate,
594 TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0);
599 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
601 TMPFS_NODE_UNLOCK(node);
603 /* Get a new vnode and associate it with our node. */
604 error = getnewvnode("tmpfs", mp, VFS_TO_TMPFS(mp)->tm_nonc ?
605 &tmpfs_vnodeop_nonc_entries : &tmpfs_vnodeop_entries, &vp);
610 /* lkflag is ignored, the lock is exclusive */
611 (void) vn_lock(vp, lkflag | LK_RETRY);
614 vp->v_type = node->tn_type;
616 /* Type-specific initialization. */
617 switch (node->tn_type) {
627 vp->v_op = &tmpfs_fifoop_entries;
630 object = node->tn_reg.tn_aobj;
631 VM_OBJECT_WLOCK(object);
633 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
634 vp->v_object = object;
635 object->un_pager.swp.swp_tmpfs = vp;
636 vm_object_set_flag(object, OBJ_TMPFS);
638 VM_OBJECT_WUNLOCK(object);
641 MPASS(node->tn_dir.tn_parent != NULL);
642 if (node->tn_dir.tn_parent == node)
643 vp->v_vflag |= VV_ROOT;
647 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
649 if (vp->v_type != VFIFO)
652 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL);
657 TMPFS_NODE_LOCK(node);
659 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
660 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
663 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
664 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
665 TMPFS_NODE_UNLOCK(node);
666 wakeup((caddr_t) &node->tn_vpstate);
668 TMPFS_NODE_UNLOCK(node);
675 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
676 TMPFS_NODE_LOCK(node);
677 MPASS(*vpp == node->tn_vnode);
678 TMPFS_NODE_UNLOCK(node);
681 tmpfs_free_node(tm, node);
687 * Destroys the association between the vnode vp and the node it
691 tmpfs_free_vp(struct vnode *vp)
693 struct tmpfs_node *node;
695 node = VP_TO_TMPFS_NODE(vp);
697 TMPFS_NODE_ASSERT_LOCKED(node);
698 node->tn_vnode = NULL;
699 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
700 wakeup(&node->tn_vnode);
701 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
706 * Allocates a new file of type 'type' and adds it to the parent directory
707 * 'dvp'; this addition is done using the component name given in 'cnp'.
708 * The ownership of the new file is automatically assigned based on the
709 * credentials of the caller (through 'cnp'), the group is set based on
710 * the parent directory and the mode is determined from the 'vap' argument.
711 * If successful, *vpp holds a vnode to the newly created file and zero
712 * is returned. Otherwise *vpp is NULL and the function returns an
713 * appropriate error code.
716 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
717 struct componentname *cnp, char *target)
720 struct tmpfs_dirent *de;
721 struct tmpfs_mount *tmp;
722 struct tmpfs_node *dnode;
723 struct tmpfs_node *node;
724 struct tmpfs_node *parent;
726 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file");
727 MPASS(cnp->cn_flags & HASBUF);
729 tmp = VFS_TO_TMPFS(dvp->v_mount);
730 dnode = VP_TO_TMPFS_DIR(dvp);
733 /* If the entry we are creating is a directory, we cannot overflow
734 * the number of links of its parent, because it will get a new
736 if (vap->va_type == VDIR) {
737 /* Ensure that we do not overflow the maximum number of links
738 * imposed by the system. */
739 MPASS(dnode->tn_links <= TMPFS_LINK_MAX);
740 if (dnode->tn_links == TMPFS_LINK_MAX) {
745 MPASS(parent != NULL);
749 /* Allocate a node that represents the new file. */
750 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type,
751 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent,
752 target, vap->va_rdev, &node);
756 /* Allocate a directory entry that points to the new file. */
757 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
760 tmpfs_free_node(tmp, node);
764 /* Allocate a vnode for the new file. */
765 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
767 tmpfs_free_dirent(tmp, de);
768 tmpfs_free_node(tmp, node);
772 /* Now that all required items are allocated, we can proceed to
773 * insert the new node into the directory, an operation that
775 if (cnp->cn_flags & ISWHITEOUT)
776 tmpfs_dir_whiteout_remove(dvp, cnp);
777 tmpfs_dir_attach(dvp, de);
781 struct tmpfs_dirent *
782 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
784 struct tmpfs_dirent *de;
786 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
788 if (de != NULL && tmpfs_dirent_duphead(de))
789 de = LIST_FIRST(&de->ud.td_duphead);
790 dc->tdc_current = de;
792 return (dc->tdc_current);
795 struct tmpfs_dirent *
796 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
798 struct tmpfs_dirent *de;
800 MPASS(dc->tdc_tree != NULL);
801 if (tmpfs_dirent_dup(dc->tdc_current)) {
802 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
803 if (dc->tdc_current != NULL)
804 return (dc->tdc_current);
806 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
807 &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
808 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
809 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
810 MPASS(dc->tdc_current != NULL);
813 return (dc->tdc_current);
816 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
817 static struct tmpfs_dirent *
818 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
820 struct tmpfs_dirent *de, dekey;
822 dekey.td_hash = hash;
823 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
827 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
828 static struct tmpfs_dirent *
829 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
830 struct tmpfs_dir_cursor *dc)
832 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
833 struct tmpfs_dirent *de, dekey;
835 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
837 if (cookie == node->tn_dir.tn_readdir_lastn &&
838 (de = node->tn_dir.tn_readdir_lastp) != NULL) {
839 /* Protect against possible race, tn_readdir_last[pn]
840 * may be updated with only shared vnode lock held. */
841 if (cookie == tmpfs_dirent_cookie(de))
845 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
846 LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
847 uh.td_dup.index_entries) {
848 MPASS(tmpfs_dirent_dup(de));
849 if (de->td_cookie == cookie)
851 /* dupindex list is sorted. */
852 if (de->td_cookie < cookie) {
861 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) {
864 dekey.td_hash = cookie;
865 /* Recover if direntry for cookie was removed */
866 de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
869 dc->tdc_current = de;
870 if (de != NULL && tmpfs_dirent_duphead(de)) {
871 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
872 MPASS(dc->tdc_current != NULL);
874 return (dc->tdc_current);
878 dc->tdc_current = de;
879 if (de != NULL && tmpfs_dirent_dup(de))
880 dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
882 return (dc->tdc_current);
886 * Looks for a directory entry in the directory represented by node.
887 * 'cnp' describes the name of the entry to look for. Note that the .
888 * and .. components are not allowed as they do not physically exist
889 * within directories.
891 * Returns a pointer to the entry when found, otherwise NULL.
893 struct tmpfs_dirent *
894 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
895 struct componentname *cnp)
897 struct tmpfs_dir_duphead *duphead;
898 struct tmpfs_dirent *de;
901 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
902 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
903 cnp->cn_nameptr[1] == '.')));
904 TMPFS_VALIDATE_DIR(node);
906 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
907 de = tmpfs_dir_xlookup_hash(node, hash);
908 if (de != NULL && tmpfs_dirent_duphead(de)) {
909 duphead = &de->ud.td_duphead;
910 LIST_FOREACH(de, duphead, uh.td_dup.entries) {
911 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
915 } else if (de != NULL) {
916 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
920 if (de != NULL && f != NULL && de->td_node != f)
927 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
928 * list, allocate new cookie value.
931 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
932 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
934 struct tmpfs_dir_duphead *dupindex;
935 struct tmpfs_dirent *de, *pde;
937 dupindex = &dnode->tn_dir.tn_dupindex;
938 de = LIST_FIRST(dupindex);
939 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
941 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
943 nde->td_cookie = de->td_cookie + 1;
944 MPASS(tmpfs_dirent_dup(nde));
945 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
946 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
951 * Cookie numbers are near exhaustion. Scan dupindex list for unused
952 * numbers. dupindex list is sorted in descending order. Keep it so
953 * after inserting nde.
957 de = LIST_NEXT(de, uh.td_dup.index_entries);
958 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
960 * Last element of the index doesn't have minimal cookie
963 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
964 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
965 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
967 } else if (de == NULL) {
969 * We are so lucky have 2^30 hash duplicates in single
970 * directory :) Return largest possible cookie value.
971 * It should be fine except possible issues with
972 * VOP_READDIR restart.
974 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
975 LIST_INSERT_HEAD(dupindex, nde,
976 uh.td_dup.index_entries);
977 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
980 if (de->td_cookie + 1 == pde->td_cookie ||
981 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
982 continue; /* No hole or invalid cookie. */
983 nde->td_cookie = de->td_cookie + 1;
984 MPASS(tmpfs_dirent_dup(nde));
985 MPASS(pde->td_cookie > nde->td_cookie);
986 MPASS(nde->td_cookie > de->td_cookie);
987 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
988 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
994 * Attaches the directory entry de to the directory represented by vp.
995 * Note that this does not change the link count of the node pointed by
996 * the directory entry, as this is done by tmpfs_alloc_dirent.
999 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
1001 struct tmpfs_node *dnode;
1002 struct tmpfs_dirent *xde, *nde;
1004 ASSERT_VOP_ELOCKED(vp, __func__);
1005 MPASS(de->td_namelen > 0);
1006 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
1007 MPASS(de->td_cookie == de->td_hash);
1009 dnode = VP_TO_TMPFS_DIR(vp);
1010 dnode->tn_dir.tn_readdir_lastn = 0;
1011 dnode->tn_dir.tn_readdir_lastp = NULL;
1013 MPASS(!tmpfs_dirent_dup(de));
1014 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1015 if (xde != NULL && tmpfs_dirent_duphead(xde))
1016 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1017 else if (xde != NULL) {
1019 * Allocate new duphead. Swap xde with duphead to avoid
1020 * adding/removing elements with the same hash.
1022 MPASS(!tmpfs_dirent_dup(xde));
1023 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
1025 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
1026 memcpy(nde, xde, sizeof(*xde));
1027 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
1028 LIST_INIT(&xde->ud.td_duphead);
1029 xde->td_namelen = 0;
1030 xde->td_node = NULL;
1031 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
1032 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1034 dnode->tn_size += sizeof(struct tmpfs_dirent);
1035 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1036 TMPFS_NODE_MODIFIED;
1041 * Detaches the directory entry de from the directory represented by vp.
1042 * Note that this does not change the link count of the node pointed by
1043 * the directory entry, as this is done by tmpfs_free_dirent.
1046 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
1048 struct tmpfs_mount *tmp;
1049 struct tmpfs_dir *head;
1050 struct tmpfs_node *dnode;
1051 struct tmpfs_dirent *xde;
1053 ASSERT_VOP_ELOCKED(vp, __func__);
1055 dnode = VP_TO_TMPFS_DIR(vp);
1056 head = &dnode->tn_dir.tn_dirhead;
1057 dnode->tn_dir.tn_readdir_lastn = 0;
1058 dnode->tn_dir.tn_readdir_lastp = NULL;
1060 if (tmpfs_dirent_dup(de)) {
1061 /* Remove duphead if de was last entry. */
1062 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1063 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1064 MPASS(tmpfs_dirent_duphead(xde));
1067 LIST_REMOVE(de, uh.td_dup.entries);
1068 LIST_REMOVE(de, uh.td_dup.index_entries);
1070 if (LIST_EMPTY(&xde->ud.td_duphead)) {
1071 RB_REMOVE(tmpfs_dir, head, xde);
1072 tmp = VFS_TO_TMPFS(vp->v_mount);
1073 MPASS(xde->td_node == NULL);
1074 tmpfs_free_dirent(tmp, xde);
1077 de->td_cookie = de->td_hash;
1079 RB_REMOVE(tmpfs_dir, head, de);
1081 dnode->tn_size -= sizeof(struct tmpfs_dirent);
1082 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1083 TMPFS_NODE_MODIFIED;
1088 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1090 struct tmpfs_dirent *de, *dde, *nde;
1092 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1093 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1094 /* Node may already be destroyed. */
1096 if (tmpfs_dirent_duphead(de)) {
1097 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1098 LIST_REMOVE(dde, uh.td_dup.entries);
1099 dde->td_node = NULL;
1100 tmpfs_free_dirent(tmp, dde);
1103 tmpfs_free_dirent(tmp, de);
1108 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
1109 * directory and returns it in the uio space. The function returns 0
1110 * on success, -1 if there was not enough space in the uio structure to
1111 * hold the directory entry or an appropriate error code if another
1115 tmpfs_dir_getdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node,
1121 TMPFS_VALIDATE_DIR(node);
1122 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1124 dent.d_fileno = node->tn_id;
1125 dent.d_off = TMPFS_DIRCOOKIE_DOTDOT;
1126 dent.d_type = DT_DIR;
1128 dent.d_name[0] = '.';
1129 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1130 dirent_terminate(&dent);
1132 if (dent.d_reclen > uio->uio_resid)
1133 error = EJUSTRETURN;
1135 error = uiomove(&dent, dent.d_reclen, uio);
1137 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1143 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
1144 * directory and returns it in the uio space. The function returns 0
1145 * on success, -1 if there was not enough space in the uio structure to
1146 * hold the directory entry or an appropriate error code if another
1150 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node,
1151 struct uio *uio, off_t next)
1153 struct tmpfs_node *parent;
1157 TMPFS_VALIDATE_DIR(node);
1158 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1161 * Return ENOENT if the current node is already removed.
1163 TMPFS_ASSERT_LOCKED(node);
1164 parent = node->tn_dir.tn_parent;
1168 TMPFS_NODE_LOCK(parent);
1169 dent.d_fileno = parent->tn_id;
1170 TMPFS_NODE_UNLOCK(parent);
1173 dent.d_type = DT_DIR;
1175 dent.d_name[0] = '.';
1176 dent.d_name[1] = '.';
1177 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1178 dirent_terminate(&dent);
1180 if (dent.d_reclen > uio->uio_resid)
1181 error = EJUSTRETURN;
1183 error = uiomove(&dent, dent.d_reclen, uio);
1185 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1191 * Helper function for tmpfs_readdir. Returns as much directory entries
1192 * as can fit in the uio space. The read starts at uio->uio_offset.
1193 * The function returns 0 on success, -1 if there was not enough space
1194 * in the uio structure to hold the directory entry or an appropriate
1195 * error code if another error happens.
1198 tmpfs_dir_getdents(struct tmpfs_mount *tm, struct tmpfs_node *node,
1199 struct uio *uio, int maxcookies, u_long *cookies, int *ncookies)
1201 struct tmpfs_dir_cursor dc;
1202 struct tmpfs_dirent *de, *nde;
1206 TMPFS_VALIDATE_DIR(node);
1211 * Lookup the node from the current offset. The starting offset of
1212 * 0 will lookup both '.' and '..', and then the first real entry,
1213 * or EOF if there are none. Then find all entries for the dir that
1214 * fit into the buffer. Once no more entries are found (de == NULL),
1215 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1218 switch (uio->uio_offset) {
1219 case TMPFS_DIRCOOKIE_DOT:
1220 error = tmpfs_dir_getdotdent(tm, node, uio);
1223 uio->uio_offset = off = TMPFS_DIRCOOKIE_DOTDOT;
1224 if (cookies != NULL)
1225 cookies[(*ncookies)++] = off;
1227 case TMPFS_DIRCOOKIE_DOTDOT:
1228 de = tmpfs_dir_first(node, &dc);
1229 off = tmpfs_dirent_cookie(de);
1230 error = tmpfs_dir_getdotdotdent(tm, node, uio, off);
1233 uio->uio_offset = off;
1234 if (cookies != NULL)
1235 cookies[(*ncookies)++] = off;
1240 case TMPFS_DIRCOOKIE_EOF:
1243 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1246 if (cookies != NULL)
1247 off = tmpfs_dirent_cookie(de);
1251 * Read as much entries as possible; i.e., until we reach the end of the
1252 * directory or we exhaust uio space.
1258 * Create a dirent structure representing the current tmpfs_node
1261 if (de->td_node == NULL) {
1265 d.d_fileno = de->td_node->tn_id;
1266 switch (de->td_node->tn_type) {
1296 panic("tmpfs_dir_getdents: type %p %d",
1297 de->td_node, (int)de->td_node->tn_type);
1300 d.d_namlen = de->td_namelen;
1301 MPASS(de->td_namelen < sizeof(d.d_name));
1302 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1303 d.d_reclen = GENERIC_DIRSIZ(&d);
1306 * Stop reading if the directory entry we are treating is bigger
1307 * than the amount of data that can be returned.
1309 if (d.d_reclen > uio->uio_resid) {
1310 error = EJUSTRETURN;
1314 nde = tmpfs_dir_next(node, &dc);
1315 d.d_off = tmpfs_dirent_cookie(nde);
1316 dirent_terminate(&d);
1319 * Copy the new dirent structure into the output buffer and
1322 error = uiomove(&d, d.d_reclen, uio);
1325 if (cookies != NULL) {
1326 off = tmpfs_dirent_cookie(de);
1327 MPASS(*ncookies < maxcookies);
1328 cookies[(*ncookies)++] = off;
1331 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
1333 /* Skip setting off when using cookies as it is already done above. */
1334 if (cookies == NULL)
1335 off = tmpfs_dirent_cookie(de);
1337 /* Update the offset and cache. */
1338 uio->uio_offset = off;
1339 node->tn_dir.tn_readdir_lastn = off;
1340 node->tn_dir.tn_readdir_lastp = de;
1342 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1347 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1349 struct tmpfs_dirent *de;
1352 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1353 cnp->cn_nameptr, cnp->cn_namelen, &de);
1356 tmpfs_dir_attach(dvp, de);
1361 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1363 struct tmpfs_dirent *de;
1365 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1366 MPASS(de != NULL && de->td_node == NULL);
1367 tmpfs_dir_detach(dvp, de);
1368 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1372 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1373 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
1374 * 'newsize' must be positive.
1376 * Returns zero on success or an appropriate error code on failure.
1379 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1381 struct tmpfs_mount *tmp;
1382 struct tmpfs_node *node;
1385 vm_pindex_t idx, newpages, oldpages;
1389 MPASS(vp->v_type == VREG);
1390 MPASS(newsize >= 0);
1392 node = VP_TO_TMPFS_NODE(vp);
1393 uobj = node->tn_reg.tn_aobj;
1394 tmp = VFS_TO_TMPFS(vp->v_mount);
1397 * Convert the old and new sizes to the number of pages needed to
1398 * store them. It may happen that we do not need to do anything
1399 * because the last allocated page can accommodate the change on
1402 oldsize = node->tn_size;
1403 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1404 MPASS(oldpages == uobj->size);
1405 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1407 if (__predict_true(newpages == oldpages && newsize >= oldsize)) {
1408 node->tn_size = newsize;
1412 if (newpages > oldpages &&
1413 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
1416 VM_OBJECT_WLOCK(uobj);
1417 if (newsize < oldsize) {
1419 * Zero the truncated part of the last page.
1421 base = newsize & PAGE_MASK;
1423 idx = OFF_TO_IDX(newsize);
1425 m = vm_page_lookup(uobj, idx);
1427 if (vm_page_sleep_if_busy(m, "tmfssz"))
1429 MPASS(m->valid == VM_PAGE_BITS_ALL);
1430 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1431 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL |
1435 rv = vm_pager_get_pages(uobj, &m, 1, NULL,
1438 if (rv == VM_PAGER_OK) {
1440 * Since the page was not resident,
1441 * and therefore not recently
1442 * accessed, immediately enqueue it
1443 * for asynchronous laundering. The
1444 * current operation is not regarded
1456 VM_OBJECT_WUNLOCK(uobj);
1462 pmap_zero_page_area(m, base, PAGE_SIZE - base);
1464 vm_pager_page_unswapped(m);
1469 * Release any swap space and free any whole pages.
1471 if (newpages < oldpages) {
1472 swap_pager_freespace(uobj, newpages, oldpages -
1474 vm_object_page_remove(uobj, newpages, 0, 0);
1477 uobj->size = newpages;
1478 VM_OBJECT_WUNLOCK(uobj);
1480 atomic_add_long(&tmp->tm_pages_used, newpages - oldpages);
1482 node->tn_size = newsize;
1487 tmpfs_check_mtime(struct vnode *vp)
1489 struct tmpfs_node *node;
1490 struct vm_object *obj;
1492 ASSERT_VOP_ELOCKED(vp, "check_mtime");
1493 if (vp->v_type != VREG)
1496 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
1497 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
1499 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) {
1500 VM_OBJECT_WLOCK(obj);
1501 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) {
1502 obj->flags &= ~OBJ_TMPFS_DIRTY;
1503 node = VP_TO_TMPFS_NODE(vp);
1504 node->tn_status |= TMPFS_NODE_MODIFIED |
1507 VM_OBJECT_WUNLOCK(obj);
1512 * Change flags of the given vnode.
1513 * Caller should execute tmpfs_update on vp after a successful execution.
1514 * The vnode must be locked on entry and remain locked on exit.
1517 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
1521 struct tmpfs_node *node;
1523 ASSERT_VOP_ELOCKED(vp, "chflags");
1525 node = VP_TO_TMPFS_NODE(vp);
1527 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK |
1528 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP |
1529 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
1530 UF_SPARSE | UF_SYSTEM)) != 0)
1531 return (EOPNOTSUPP);
1533 /* Disallow this operation if the file system is mounted read-only. */
1534 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1538 * Callers may only modify the file flags on objects they
1539 * have VADMIN rights for.
1541 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1544 * Unprivileged processes are not permitted to unset system
1545 * flags, or modify flags if any system flags are set.
1547 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1548 if (node->tn_flags &
1549 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1550 error = securelevel_gt(cred, 0);
1555 if (node->tn_flags &
1556 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1557 ((flags ^ node->tn_flags) & SF_SETTABLE))
1560 node->tn_flags = flags;
1561 node->tn_status |= TMPFS_NODE_CHANGED;
1563 ASSERT_VOP_ELOCKED(vp, "chflags2");
1569 * Change access mode on the given vnode.
1570 * Caller should execute tmpfs_update on vp after a successful execution.
1571 * The vnode must be locked on entry and remain locked on exit.
1574 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1577 struct tmpfs_node *node;
1579 ASSERT_VOP_ELOCKED(vp, "chmod");
1581 node = VP_TO_TMPFS_NODE(vp);
1583 /* Disallow this operation if the file system is mounted read-only. */
1584 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1587 /* Immutable or append-only files cannot be modified, either. */
1588 if (node->tn_flags & (IMMUTABLE | APPEND))
1592 * To modify the permissions on a file, must possess VADMIN
1595 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1599 * Privileged processes may set the sticky bit on non-directories,
1600 * as well as set the setgid bit on a file with a group that the
1601 * process is not a member of.
1603 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1604 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1607 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1608 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1614 node->tn_mode &= ~ALLPERMS;
1615 node->tn_mode |= mode & ALLPERMS;
1617 node->tn_status |= TMPFS_NODE_CHANGED;
1619 ASSERT_VOP_ELOCKED(vp, "chmod2");
1625 * Change ownership of the given vnode. At least one of uid or gid must
1626 * be different than VNOVAL. If one is set to that value, the attribute
1628 * Caller should execute tmpfs_update on vp after a successful execution.
1629 * The vnode must be locked on entry and remain locked on exit.
1632 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1636 struct tmpfs_node *node;
1640 ASSERT_VOP_ELOCKED(vp, "chown");
1642 node = VP_TO_TMPFS_NODE(vp);
1644 /* Assign default values if they are unknown. */
1645 MPASS(uid != VNOVAL || gid != VNOVAL);
1650 MPASS(uid != VNOVAL && gid != VNOVAL);
1652 /* Disallow this operation if the file system is mounted read-only. */
1653 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1656 /* Immutable or append-only files cannot be modified, either. */
1657 if (node->tn_flags & (IMMUTABLE | APPEND))
1661 * To modify the ownership of a file, must possess VADMIN for that
1664 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1668 * To change the owner of a file, or change the group of a file to a
1669 * group of which we are not a member, the caller must have
1672 if ((uid != node->tn_uid ||
1673 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1674 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1677 ogid = node->tn_gid;
1678 ouid = node->tn_uid;
1683 node->tn_status |= TMPFS_NODE_CHANGED;
1685 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1686 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1687 node->tn_mode &= ~(S_ISUID | S_ISGID);
1690 ASSERT_VOP_ELOCKED(vp, "chown2");
1696 * Change size of the given vnode.
1697 * Caller should execute tmpfs_update on vp after a successful execution.
1698 * The vnode must be locked on entry and remain locked on exit.
1701 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1705 struct tmpfs_node *node;
1707 ASSERT_VOP_ELOCKED(vp, "chsize");
1709 node = VP_TO_TMPFS_NODE(vp);
1711 /* Decide whether this is a valid operation based on the file type. */
1713 switch (vp->v_type) {
1718 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1727 /* Allow modifications of special files even if in the file
1728 * system is mounted read-only (we are not modifying the
1729 * files themselves, but the objects they represent). */
1733 /* Anything else is unsupported. */
1737 /* Immutable or append-only files cannot be modified, either. */
1738 if (node->tn_flags & (IMMUTABLE | APPEND))
1741 error = tmpfs_truncate(vp, size);
1742 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1743 * for us, as will update tn_status; no need to do that here. */
1745 ASSERT_VOP_ELOCKED(vp, "chsize2");
1751 * Change access and modification times of the given vnode.
1752 * Caller should execute tmpfs_update on vp after a successful execution.
1753 * The vnode must be locked on entry and remain locked on exit.
1756 tmpfs_chtimes(struct vnode *vp, struct vattr *vap,
1757 struct ucred *cred, struct thread *l)
1760 struct tmpfs_node *node;
1762 ASSERT_VOP_ELOCKED(vp, "chtimes");
1764 node = VP_TO_TMPFS_NODE(vp);
1766 /* Disallow this operation if the file system is mounted read-only. */
1767 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1770 /* Immutable or append-only files cannot be modified, either. */
1771 if (node->tn_flags & (IMMUTABLE | APPEND))
1774 error = vn_utimes_perm(vp, vap, cred, l);
1778 if (vap->va_atime.tv_sec != VNOVAL)
1779 node->tn_status |= TMPFS_NODE_ACCESSED;
1781 if (vap->va_mtime.tv_sec != VNOVAL)
1782 node->tn_status |= TMPFS_NODE_MODIFIED;
1784 if (vap->va_birthtime.tv_sec != VNOVAL)
1785 node->tn_status |= TMPFS_NODE_MODIFIED;
1787 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime);
1789 if (vap->va_birthtime.tv_sec != VNOVAL)
1790 node->tn_birthtime = vap->va_birthtime;
1791 ASSERT_VOP_ELOCKED(vp, "chtimes2");
1797 tmpfs_set_status(struct tmpfs_mount *tm, struct tmpfs_node *node, int status)
1800 if ((node->tn_status & status) == status || tm->tm_ronly)
1802 TMPFS_NODE_LOCK(node);
1803 node->tn_status |= status;
1804 TMPFS_NODE_UNLOCK(node);
1807 /* Sync timestamps */
1809 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1810 const struct timespec *mod)
1812 struct tmpfs_node *node;
1813 struct timespec now;
1815 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes");
1816 node = VP_TO_TMPFS_NODE(vp);
1818 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1819 TMPFS_NODE_CHANGED)) == 0)
1822 vfs_timestamp(&now);
1823 TMPFS_NODE_LOCK(node);
1824 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1827 node->tn_atime = *acc;
1829 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1832 node->tn_mtime = *mod;
1834 if (node->tn_status & TMPFS_NODE_CHANGED)
1835 node->tn_ctime = now;
1836 node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1837 TMPFS_NODE_CHANGED);
1838 TMPFS_NODE_UNLOCK(node);
1840 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
1841 random_harvest_queue(node, sizeof(*node), RANDOM_FS_ATIME);
1845 tmpfs_update(struct vnode *vp)
1848 tmpfs_itimes(vp, NULL, NULL);
1852 tmpfs_truncate(struct vnode *vp, off_t length)
1855 struct tmpfs_node *node;
1857 node = VP_TO_TMPFS_NODE(vp);
1864 if (node->tn_size == length) {
1869 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1872 error = tmpfs_reg_resize(vp, length, FALSE);
1874 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1883 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1885 if (a->td_hash > b->td_hash)
1887 else if (a->td_hash < b->td_hash)
1892 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);