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);
365 free_unr(tmp->tm_ino_unr, node->tn_id);
366 uma_zfree(tmp->tm_node_pool, node);
372 static __inline uint32_t
373 tmpfs_dirent_hash(const char *name, u_int len)
377 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK;
378 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP
381 if (hash < TMPFS_DIRCOOKIE_MIN)
382 hash += TMPFS_DIRCOOKIE_MIN;
387 static __inline off_t
388 tmpfs_dirent_cookie(struct tmpfs_dirent *de)
391 return (TMPFS_DIRCOOKIE_EOF);
393 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN);
395 return (de->td_cookie);
398 static __inline boolean_t
399 tmpfs_dirent_dup(struct tmpfs_dirent *de)
401 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0);
404 static __inline boolean_t
405 tmpfs_dirent_duphead(struct tmpfs_dirent *de)
407 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0);
411 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen)
413 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen);
414 memcpy(de->ud.td_name, name, namelen);
415 de->td_namelen = namelen;
419 * Allocates a new directory entry for the node node with a name of name.
420 * The new directory entry is returned in *de.
422 * The link count of node is increased by one to reflect the new object
425 * Returns zero on success or an appropriate error code on failure.
428 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
429 const char *name, u_int len, struct tmpfs_dirent **de)
431 struct tmpfs_dirent *nde;
433 nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK);
436 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
437 tmpfs_dirent_init(nde, name, len);
449 * Frees a directory entry. It is the caller's responsibility to destroy
450 * the node referenced by it if needed.
452 * The link count of node is decreased by one to reflect the removal of an
453 * object that referenced it. This only happens if 'node_exists' is true;
454 * otherwise the function will not access the node referred to by the
455 * directory entry, as it may already have been released from the outside.
458 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
460 struct tmpfs_node *node;
464 MPASS(node->tn_links > 0);
467 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL)
468 free(de->ud.td_name, M_TMPFSNAME);
469 uma_zfree(tmp->tm_dirent_pool, de);
473 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj)
476 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject");
477 if (vp->v_type != VREG || obj == NULL)
480 VM_OBJECT_WLOCK(obj);
482 vm_object_clear_flag(obj, OBJ_TMPFS);
483 obj->un_pager.swp.swp_tmpfs = NULL;
485 VM_OBJECT_WUNLOCK(obj);
489 * Need to clear v_object for insmntque failure.
492 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg)
495 tmpfs_destroy_vobject(vp, vp->v_object);
498 vp->v_op = &dead_vnodeops;
504 * Allocates a new vnode for the node node or returns a new reference to
505 * an existing one if the node had already a vnode referencing it. The
506 * resulting locked vnode is returned in *vpp.
508 * Returns zero on success or an appropriate error code on failure.
511 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
515 struct tmpfs_mount *tm;
520 tm = VFS_TO_TMPFS(mp);
521 TMPFS_NODE_LOCK(node);
522 tmpfs_ref_node_locked(node);
524 TMPFS_NODE_ASSERT_LOCKED(node);
525 if ((vp = node->tn_vnode) != NULL) {
526 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
528 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) ||
529 ((vp->v_iflag & VI_DOOMED) != 0 &&
530 (lkflag & LK_NOWAIT) != 0)) {
532 TMPFS_NODE_UNLOCK(node);
537 if ((vp->v_iflag & VI_DOOMED) != 0) {
539 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM;
540 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) {
541 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node),
546 TMPFS_NODE_UNLOCK(node);
547 error = vget(vp, lkflag | LK_INTERLOCK, curthread);
548 if (error == ENOENT) {
549 TMPFS_NODE_LOCK(node);
558 * Make sure the vnode is still there after
559 * getting the interlock to avoid racing a free.
561 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
563 TMPFS_NODE_LOCK(node);
570 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
571 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
572 TMPFS_NODE_UNLOCK(node);
579 * otherwise lock the vp list while we call getnewvnode
580 * since that can block.
582 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
583 node->tn_vpstate |= TMPFS_VNODE_WANT;
584 error = msleep((caddr_t) &node->tn_vpstate,
585 TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0);
590 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
592 TMPFS_NODE_UNLOCK(node);
594 /* Get a new vnode and associate it with our node. */
595 error = getnewvnode("tmpfs", mp, VFS_TO_TMPFS(mp)->tm_nonc ?
596 &tmpfs_vnodeop_nonc_entries : &tmpfs_vnodeop_entries, &vp);
601 /* lkflag is ignored, the lock is exclusive */
602 (void) vn_lock(vp, lkflag | LK_RETRY);
605 vp->v_type = node->tn_type;
607 /* Type-specific initialization. */
608 switch (node->tn_type) {
618 vp->v_op = &tmpfs_fifoop_entries;
621 object = node->tn_reg.tn_aobj;
622 VM_OBJECT_WLOCK(object);
624 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
625 vp->v_object = object;
626 object->un_pager.swp.swp_tmpfs = vp;
627 vm_object_set_flag(object, OBJ_TMPFS);
629 VM_OBJECT_WUNLOCK(object);
632 MPASS(node->tn_dir.tn_parent != NULL);
633 if (node->tn_dir.tn_parent == node)
634 vp->v_vflag |= VV_ROOT;
638 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
640 if (vp->v_type != VFIFO)
643 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL);
648 TMPFS_NODE_LOCK(node);
650 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
651 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
654 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
655 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
656 TMPFS_NODE_UNLOCK(node);
657 wakeup((caddr_t) &node->tn_vpstate);
659 TMPFS_NODE_UNLOCK(node);
666 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
667 TMPFS_NODE_LOCK(node);
668 MPASS(*vpp == node->tn_vnode);
669 TMPFS_NODE_UNLOCK(node);
672 tmpfs_free_node(tm, node);
678 * Destroys the association between the vnode vp and the node it
682 tmpfs_free_vp(struct vnode *vp)
684 struct tmpfs_node *node;
686 node = VP_TO_TMPFS_NODE(vp);
688 TMPFS_NODE_ASSERT_LOCKED(node);
689 node->tn_vnode = NULL;
690 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
691 wakeup(&node->tn_vnode);
692 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
697 * Allocates a new file of type 'type' and adds it to the parent directory
698 * 'dvp'; this addition is done using the component name given in 'cnp'.
699 * The ownership of the new file is automatically assigned based on the
700 * credentials of the caller (through 'cnp'), the group is set based on
701 * the parent directory and the mode is determined from the 'vap' argument.
702 * If successful, *vpp holds a vnode to the newly created file and zero
703 * is returned. Otherwise *vpp is NULL and the function returns an
704 * appropriate error code.
707 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
708 struct componentname *cnp, char *target)
711 struct tmpfs_dirent *de;
712 struct tmpfs_mount *tmp;
713 struct tmpfs_node *dnode;
714 struct tmpfs_node *node;
715 struct tmpfs_node *parent;
717 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file");
718 MPASS(cnp->cn_flags & HASBUF);
720 tmp = VFS_TO_TMPFS(dvp->v_mount);
721 dnode = VP_TO_TMPFS_DIR(dvp);
724 /* If the entry we are creating is a directory, we cannot overflow
725 * the number of links of its parent, because it will get a new
727 if (vap->va_type == VDIR) {
728 /* Ensure that we do not overflow the maximum number of links
729 * imposed by the system. */
730 MPASS(dnode->tn_links <= LINK_MAX);
731 if (dnode->tn_links == LINK_MAX) {
736 MPASS(parent != NULL);
740 /* Allocate a node that represents the new file. */
741 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type,
742 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent,
743 target, vap->va_rdev, &node);
747 /* Allocate a directory entry that points to the new file. */
748 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
751 tmpfs_free_node(tmp, node);
755 /* Allocate a vnode for the new file. */
756 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
758 tmpfs_free_dirent(tmp, de);
759 tmpfs_free_node(tmp, node);
763 /* Now that all required items are allocated, we can proceed to
764 * insert the new node into the directory, an operation that
766 if (cnp->cn_flags & ISWHITEOUT)
767 tmpfs_dir_whiteout_remove(dvp, cnp);
768 tmpfs_dir_attach(dvp, de);
772 struct tmpfs_dirent *
773 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
775 struct tmpfs_dirent *de;
777 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
779 if (de != NULL && tmpfs_dirent_duphead(de))
780 de = LIST_FIRST(&de->ud.td_duphead);
781 dc->tdc_current = de;
783 return (dc->tdc_current);
786 struct tmpfs_dirent *
787 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
789 struct tmpfs_dirent *de;
791 MPASS(dc->tdc_tree != NULL);
792 if (tmpfs_dirent_dup(dc->tdc_current)) {
793 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
794 if (dc->tdc_current != NULL)
795 return (dc->tdc_current);
797 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
798 &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
799 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
800 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
801 MPASS(dc->tdc_current != NULL);
804 return (dc->tdc_current);
807 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
808 static struct tmpfs_dirent *
809 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
811 struct tmpfs_dirent *de, dekey;
813 dekey.td_hash = hash;
814 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
818 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
819 static struct tmpfs_dirent *
820 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
821 struct tmpfs_dir_cursor *dc)
823 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
824 struct tmpfs_dirent *de, dekey;
826 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
828 if (cookie == node->tn_dir.tn_readdir_lastn &&
829 (de = node->tn_dir.tn_readdir_lastp) != NULL) {
830 /* Protect against possible race, tn_readdir_last[pn]
831 * may be updated with only shared vnode lock held. */
832 if (cookie == tmpfs_dirent_cookie(de))
836 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
837 LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
838 uh.td_dup.index_entries) {
839 MPASS(tmpfs_dirent_dup(de));
840 if (de->td_cookie == cookie)
842 /* dupindex list is sorted. */
843 if (de->td_cookie < cookie) {
852 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) {
855 dekey.td_hash = cookie;
856 /* Recover if direntry for cookie was removed */
857 de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
860 dc->tdc_current = de;
861 if (de != NULL && tmpfs_dirent_duphead(de)) {
862 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
863 MPASS(dc->tdc_current != NULL);
865 return (dc->tdc_current);
869 dc->tdc_current = de;
870 if (de != NULL && tmpfs_dirent_dup(de))
871 dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
873 return (dc->tdc_current);
877 * Looks for a directory entry in the directory represented by node.
878 * 'cnp' describes the name of the entry to look for. Note that the .
879 * and .. components are not allowed as they do not physically exist
880 * within directories.
882 * Returns a pointer to the entry when found, otherwise NULL.
884 struct tmpfs_dirent *
885 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
886 struct componentname *cnp)
888 struct tmpfs_dir_duphead *duphead;
889 struct tmpfs_dirent *de;
892 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
893 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
894 cnp->cn_nameptr[1] == '.')));
895 TMPFS_VALIDATE_DIR(node);
897 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
898 de = tmpfs_dir_xlookup_hash(node, hash);
899 if (de != NULL && tmpfs_dirent_duphead(de)) {
900 duphead = &de->ud.td_duphead;
901 LIST_FOREACH(de, duphead, uh.td_dup.entries) {
902 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
906 } else if (de != NULL) {
907 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
911 if (de != NULL && f != NULL && de->td_node != f)
918 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
919 * list, allocate new cookie value.
922 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
923 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
925 struct tmpfs_dir_duphead *dupindex;
926 struct tmpfs_dirent *de, *pde;
928 dupindex = &dnode->tn_dir.tn_dupindex;
929 de = LIST_FIRST(dupindex);
930 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
932 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
934 nde->td_cookie = de->td_cookie + 1;
935 MPASS(tmpfs_dirent_dup(nde));
936 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
937 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
942 * Cookie numbers are near exhaustion. Scan dupindex list for unused
943 * numbers. dupindex list is sorted in descending order. Keep it so
944 * after inserting nde.
948 de = LIST_NEXT(de, uh.td_dup.index_entries);
949 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
951 * Last element of the index doesn't have minimal cookie
954 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
955 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
956 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
958 } else if (de == NULL) {
960 * We are so lucky have 2^30 hash duplicates in single
961 * directory :) Return largest possible cookie value.
962 * It should be fine except possible issues with
963 * VOP_READDIR restart.
965 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
966 LIST_INSERT_HEAD(dupindex, nde,
967 uh.td_dup.index_entries);
968 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
971 if (de->td_cookie + 1 == pde->td_cookie ||
972 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
973 continue; /* No hole or invalid cookie. */
974 nde->td_cookie = de->td_cookie + 1;
975 MPASS(tmpfs_dirent_dup(nde));
976 MPASS(pde->td_cookie > nde->td_cookie);
977 MPASS(nde->td_cookie > de->td_cookie);
978 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
979 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
985 * Attaches the directory entry de to the directory represented by vp.
986 * Note that this does not change the link count of the node pointed by
987 * the directory entry, as this is done by tmpfs_alloc_dirent.
990 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
992 struct tmpfs_node *dnode;
993 struct tmpfs_dirent *xde, *nde;
995 ASSERT_VOP_ELOCKED(vp, __func__);
996 MPASS(de->td_namelen > 0);
997 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
998 MPASS(de->td_cookie == de->td_hash);
1000 dnode = VP_TO_TMPFS_DIR(vp);
1001 dnode->tn_dir.tn_readdir_lastn = 0;
1002 dnode->tn_dir.tn_readdir_lastp = NULL;
1004 MPASS(!tmpfs_dirent_dup(de));
1005 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1006 if (xde != NULL && tmpfs_dirent_duphead(xde))
1007 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1008 else if (xde != NULL) {
1010 * Allocate new duphead. Swap xde with duphead to avoid
1011 * adding/removing elements with the same hash.
1013 MPASS(!tmpfs_dirent_dup(xde));
1014 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
1016 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
1017 memcpy(nde, xde, sizeof(*xde));
1018 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
1019 LIST_INIT(&xde->ud.td_duphead);
1020 xde->td_namelen = 0;
1021 xde->td_node = NULL;
1022 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
1023 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1025 dnode->tn_size += sizeof(struct tmpfs_dirent);
1026 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1027 TMPFS_NODE_MODIFIED;
1032 * Detaches the directory entry de from the directory represented by vp.
1033 * Note that this does not change the link count of the node pointed by
1034 * the directory entry, as this is done by tmpfs_free_dirent.
1037 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
1039 struct tmpfs_mount *tmp;
1040 struct tmpfs_dir *head;
1041 struct tmpfs_node *dnode;
1042 struct tmpfs_dirent *xde;
1044 ASSERT_VOP_ELOCKED(vp, __func__);
1046 dnode = VP_TO_TMPFS_DIR(vp);
1047 head = &dnode->tn_dir.tn_dirhead;
1048 dnode->tn_dir.tn_readdir_lastn = 0;
1049 dnode->tn_dir.tn_readdir_lastp = NULL;
1051 if (tmpfs_dirent_dup(de)) {
1052 /* Remove duphead if de was last entry. */
1053 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1054 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1055 MPASS(tmpfs_dirent_duphead(xde));
1058 LIST_REMOVE(de, uh.td_dup.entries);
1059 LIST_REMOVE(de, uh.td_dup.index_entries);
1061 if (LIST_EMPTY(&xde->ud.td_duphead)) {
1062 RB_REMOVE(tmpfs_dir, head, xde);
1063 tmp = VFS_TO_TMPFS(vp->v_mount);
1064 MPASS(xde->td_node == NULL);
1065 tmpfs_free_dirent(tmp, xde);
1068 de->td_cookie = de->td_hash;
1070 RB_REMOVE(tmpfs_dir, head, de);
1072 dnode->tn_size -= sizeof(struct tmpfs_dirent);
1073 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1074 TMPFS_NODE_MODIFIED;
1079 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1081 struct tmpfs_dirent *de, *dde, *nde;
1083 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1084 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1085 /* Node may already be destroyed. */
1087 if (tmpfs_dirent_duphead(de)) {
1088 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1089 LIST_REMOVE(dde, uh.td_dup.entries);
1090 dde->td_node = NULL;
1091 tmpfs_free_dirent(tmp, dde);
1094 tmpfs_free_dirent(tmp, de);
1099 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
1100 * directory and returns it in the uio space. The function returns 0
1101 * on success, -1 if there was not enough space in the uio structure to
1102 * hold the directory entry or an appropriate error code if another
1106 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
1111 TMPFS_VALIDATE_DIR(node);
1112 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1114 dent.d_fileno = node->tn_id;
1115 dent.d_type = DT_DIR;
1117 dent.d_name[0] = '.';
1118 dent.d_name[1] = '\0';
1119 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1121 if (dent.d_reclen > uio->uio_resid)
1122 error = EJUSTRETURN;
1124 error = uiomove(&dent, dent.d_reclen, uio);
1126 tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
1132 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
1133 * directory and returns it in the uio space. The function returns 0
1134 * on success, -1 if there was not enough space in the uio structure to
1135 * hold the directory entry or an appropriate error code if another
1139 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
1144 TMPFS_VALIDATE_DIR(node);
1145 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1148 * Return ENOENT if the current node is already removed.
1150 TMPFS_ASSERT_LOCKED(node);
1151 if (node->tn_dir.tn_parent == NULL)
1154 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
1155 dent.d_fileno = node->tn_dir.tn_parent->tn_id;
1156 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
1158 dent.d_type = DT_DIR;
1160 dent.d_name[0] = '.';
1161 dent.d_name[1] = '.';
1162 dent.d_name[2] = '\0';
1163 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1165 if (dent.d_reclen > uio->uio_resid)
1166 error = EJUSTRETURN;
1168 error = uiomove(&dent, dent.d_reclen, uio);
1170 tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
1176 * Helper function for tmpfs_readdir. Returns as much directory entries
1177 * as can fit in the uio space. The read starts at uio->uio_offset.
1178 * The function returns 0 on success, -1 if there was not enough space
1179 * in the uio structure to hold the directory entry or an appropriate
1180 * error code if another error happens.
1183 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies,
1184 u_long *cookies, int *ncookies)
1186 struct tmpfs_dir_cursor dc;
1187 struct tmpfs_dirent *de;
1191 TMPFS_VALIDATE_DIR(node);
1196 * Lookup the node from the current offset. The starting offset of
1197 * 0 will lookup both '.' and '..', and then the first real entry,
1198 * or EOF if there are none. Then find all entries for the dir that
1199 * fit into the buffer. Once no more entries are found (de == NULL),
1200 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1203 switch (uio->uio_offset) {
1204 case TMPFS_DIRCOOKIE_DOT:
1205 error = tmpfs_dir_getdotdent(node, uio);
1208 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
1209 if (cookies != NULL)
1210 cookies[(*ncookies)++] = off = uio->uio_offset;
1212 case TMPFS_DIRCOOKIE_DOTDOT:
1213 error = tmpfs_dir_getdotdotdent(node, uio);
1216 de = tmpfs_dir_first(node, &dc);
1217 uio->uio_offset = tmpfs_dirent_cookie(de);
1218 if (cookies != NULL)
1219 cookies[(*ncookies)++] = off = uio->uio_offset;
1224 case TMPFS_DIRCOOKIE_EOF:
1227 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1230 if (cookies != NULL)
1231 off = tmpfs_dirent_cookie(de);
1234 /* Read as much entries as possible; i.e., until we reach the end of
1235 * the directory or we exhaust uio space. */
1239 /* Create a dirent structure representing the current
1240 * tmpfs_node and fill it. */
1241 if (de->td_node == NULL) {
1245 d.d_fileno = de->td_node->tn_id;
1246 switch (de->td_node->tn_type) {
1276 panic("tmpfs_dir_getdents: type %p %d",
1277 de->td_node, (int)de->td_node->tn_type);
1280 d.d_namlen = de->td_namelen;
1281 MPASS(de->td_namelen < sizeof(d.d_name));
1282 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1283 d.d_name[de->td_namelen] = '\0';
1284 d.d_reclen = GENERIC_DIRSIZ(&d);
1286 /* Stop reading if the directory entry we are treating is
1287 * bigger than the amount of data that can be returned. */
1288 if (d.d_reclen > uio->uio_resid) {
1289 error = EJUSTRETURN;
1293 /* Copy the new dirent structure into the output buffer and
1294 * advance pointers. */
1295 error = uiomove(&d, d.d_reclen, uio);
1297 de = tmpfs_dir_next(node, &dc);
1298 if (cookies != NULL) {
1299 off = tmpfs_dirent_cookie(de);
1300 MPASS(*ncookies < maxcookies);
1301 cookies[(*ncookies)++] = off;
1304 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
1306 /* Skip setting off when using cookies as it is already done above. */
1307 if (cookies == NULL)
1308 off = tmpfs_dirent_cookie(de);
1310 /* Update the offset and cache. */
1311 uio->uio_offset = off;
1312 node->tn_dir.tn_readdir_lastn = off;
1313 node->tn_dir.tn_readdir_lastp = de;
1315 tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
1320 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1322 struct tmpfs_dirent *de;
1325 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1326 cnp->cn_nameptr, cnp->cn_namelen, &de);
1329 tmpfs_dir_attach(dvp, de);
1334 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1336 struct tmpfs_dirent *de;
1338 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1339 MPASS(de != NULL && de->td_node == NULL);
1340 tmpfs_dir_detach(dvp, de);
1341 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1345 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1346 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
1347 * 'newsize' must be positive.
1349 * Returns zero on success or an appropriate error code on failure.
1352 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1354 struct tmpfs_mount *tmp;
1355 struct tmpfs_node *node;
1358 vm_pindex_t idx, newpages, oldpages;
1362 MPASS(vp->v_type == VREG);
1363 MPASS(newsize >= 0);
1365 node = VP_TO_TMPFS_NODE(vp);
1366 uobj = node->tn_reg.tn_aobj;
1367 tmp = VFS_TO_TMPFS(vp->v_mount);
1370 * Convert the old and new sizes to the number of pages needed to
1371 * store them. It may happen that we do not need to do anything
1372 * because the last allocated page can accommodate the change on
1375 oldsize = node->tn_size;
1376 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1377 MPASS(oldpages == uobj->size);
1378 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1380 if (__predict_true(newpages == oldpages && newsize >= oldsize)) {
1381 node->tn_size = newsize;
1385 if (newpages > oldpages &&
1386 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
1389 VM_OBJECT_WLOCK(uobj);
1390 if (newsize < oldsize) {
1392 * Zero the truncated part of the last page.
1394 base = newsize & PAGE_MASK;
1396 idx = OFF_TO_IDX(newsize);
1398 m = vm_page_lookup(uobj, idx);
1400 if (vm_page_sleep_if_busy(m, "tmfssz"))
1402 MPASS(m->valid == VM_PAGE_BITS_ALL);
1403 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1404 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL);
1406 VM_OBJECT_WUNLOCK(uobj);
1408 VM_OBJECT_WLOCK(uobj);
1411 rv = vm_pager_get_pages(uobj, &m, 1, NULL,
1414 if (rv == VM_PAGER_OK) {
1416 * Since the page was not resident,
1417 * and therefore not recently
1418 * accessed, immediately enqueue it
1419 * for asynchronous laundering. The
1420 * current operation is not regarded
1432 VM_OBJECT_WUNLOCK(uobj);
1438 pmap_zero_page_area(m, base, PAGE_SIZE - base);
1440 vm_pager_page_unswapped(m);
1445 * Release any swap space and free any whole pages.
1447 if (newpages < oldpages) {
1448 swap_pager_freespace(uobj, newpages, oldpages -
1450 vm_object_page_remove(uobj, newpages, 0, 0);
1453 uobj->size = newpages;
1454 VM_OBJECT_WUNLOCK(uobj);
1456 atomic_add_long(&tmp->tm_pages_used, newpages - oldpages);
1458 node->tn_size = newsize;
1463 tmpfs_check_mtime(struct vnode *vp)
1465 struct tmpfs_node *node;
1466 struct vm_object *obj;
1468 ASSERT_VOP_ELOCKED(vp, "check_mtime");
1469 if (vp->v_type != VREG)
1472 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
1473 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
1475 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) {
1476 VM_OBJECT_WLOCK(obj);
1477 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) {
1478 obj->flags &= ~OBJ_TMPFS_DIRTY;
1479 node = VP_TO_TMPFS_NODE(vp);
1480 node->tn_status |= TMPFS_NODE_MODIFIED |
1483 VM_OBJECT_WUNLOCK(obj);
1488 * Change flags of the given vnode.
1489 * Caller should execute tmpfs_update on vp after a successful execution.
1490 * The vnode must be locked on entry and remain locked on exit.
1493 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
1497 struct tmpfs_node *node;
1499 ASSERT_VOP_ELOCKED(vp, "chflags");
1501 node = VP_TO_TMPFS_NODE(vp);
1503 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK |
1504 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP |
1505 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
1506 UF_SPARSE | UF_SYSTEM)) != 0)
1507 return (EOPNOTSUPP);
1509 /* Disallow this operation if the file system is mounted read-only. */
1510 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1514 * Callers may only modify the file flags on objects they
1515 * have VADMIN rights for.
1517 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1520 * Unprivileged processes are not permitted to unset system
1521 * flags, or modify flags if any system flags are set.
1523 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1524 if (node->tn_flags &
1525 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1526 error = securelevel_gt(cred, 0);
1531 if (node->tn_flags &
1532 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1533 ((flags ^ node->tn_flags) & SF_SETTABLE))
1536 node->tn_flags = flags;
1537 node->tn_status |= TMPFS_NODE_CHANGED;
1539 ASSERT_VOP_ELOCKED(vp, "chflags2");
1545 * Change access mode on the given vnode.
1546 * Caller should execute tmpfs_update on vp after a successful execution.
1547 * The vnode must be locked on entry and remain locked on exit.
1550 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1553 struct tmpfs_node *node;
1555 ASSERT_VOP_ELOCKED(vp, "chmod");
1557 node = VP_TO_TMPFS_NODE(vp);
1559 /* Disallow this operation if the file system is mounted read-only. */
1560 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1563 /* Immutable or append-only files cannot be modified, either. */
1564 if (node->tn_flags & (IMMUTABLE | APPEND))
1568 * To modify the permissions on a file, must possess VADMIN
1571 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1575 * Privileged processes may set the sticky bit on non-directories,
1576 * as well as set the setgid bit on a file with a group that the
1577 * process is not a member of.
1579 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1580 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1583 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1584 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1590 node->tn_mode &= ~ALLPERMS;
1591 node->tn_mode |= mode & ALLPERMS;
1593 node->tn_status |= TMPFS_NODE_CHANGED;
1595 ASSERT_VOP_ELOCKED(vp, "chmod2");
1601 * Change ownership of the given vnode. At least one of uid or gid must
1602 * be different than VNOVAL. If one is set to that value, the attribute
1604 * Caller should execute tmpfs_update on vp after a successful execution.
1605 * The vnode must be locked on entry and remain locked on exit.
1608 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1612 struct tmpfs_node *node;
1616 ASSERT_VOP_ELOCKED(vp, "chown");
1618 node = VP_TO_TMPFS_NODE(vp);
1620 /* Assign default values if they are unknown. */
1621 MPASS(uid != VNOVAL || gid != VNOVAL);
1626 MPASS(uid != VNOVAL && gid != VNOVAL);
1628 /* Disallow this operation if the file system is mounted read-only. */
1629 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1632 /* Immutable or append-only files cannot be modified, either. */
1633 if (node->tn_flags & (IMMUTABLE | APPEND))
1637 * To modify the ownership of a file, must possess VADMIN for that
1640 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1644 * To change the owner of a file, or change the group of a file to a
1645 * group of which we are not a member, the caller must have
1648 if ((uid != node->tn_uid ||
1649 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1650 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1653 ogid = node->tn_gid;
1654 ouid = node->tn_uid;
1659 node->tn_status |= TMPFS_NODE_CHANGED;
1661 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1662 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1663 node->tn_mode &= ~(S_ISUID | S_ISGID);
1666 ASSERT_VOP_ELOCKED(vp, "chown2");
1672 * Change size of the given vnode.
1673 * Caller should execute tmpfs_update on vp after a successful execution.
1674 * The vnode must be locked on entry and remain locked on exit.
1677 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1681 struct tmpfs_node *node;
1683 ASSERT_VOP_ELOCKED(vp, "chsize");
1685 node = VP_TO_TMPFS_NODE(vp);
1687 /* Decide whether this is a valid operation based on the file type. */
1689 switch (vp->v_type) {
1694 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1703 /* Allow modifications of special files even if in the file
1704 * system is mounted read-only (we are not modifying the
1705 * files themselves, but the objects they represent). */
1709 /* Anything else is unsupported. */
1713 /* Immutable or append-only files cannot be modified, either. */
1714 if (node->tn_flags & (IMMUTABLE | APPEND))
1717 error = tmpfs_truncate(vp, size);
1718 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1719 * for us, as will update tn_status; no need to do that here. */
1721 ASSERT_VOP_ELOCKED(vp, "chsize2");
1727 * Change access and modification times of the given vnode.
1728 * Caller should execute tmpfs_update on vp after a successful execution.
1729 * The vnode must be locked on entry and remain locked on exit.
1732 tmpfs_chtimes(struct vnode *vp, struct vattr *vap,
1733 struct ucred *cred, struct thread *l)
1736 struct tmpfs_node *node;
1738 ASSERT_VOP_ELOCKED(vp, "chtimes");
1740 node = VP_TO_TMPFS_NODE(vp);
1742 /* Disallow this operation if the file system is mounted read-only. */
1743 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1746 /* Immutable or append-only files cannot be modified, either. */
1747 if (node->tn_flags & (IMMUTABLE | APPEND))
1750 error = vn_utimes_perm(vp, vap, cred, l);
1754 if (vap->va_atime.tv_sec != VNOVAL)
1755 node->tn_status |= TMPFS_NODE_ACCESSED;
1757 if (vap->va_mtime.tv_sec != VNOVAL)
1758 node->tn_status |= TMPFS_NODE_MODIFIED;
1760 if (vap->va_birthtime.tv_sec != VNOVAL)
1761 node->tn_status |= TMPFS_NODE_MODIFIED;
1763 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime);
1765 if (vap->va_birthtime.tv_sec != VNOVAL)
1766 node->tn_birthtime = vap->va_birthtime;
1767 ASSERT_VOP_ELOCKED(vp, "chtimes2");
1773 tmpfs_set_status(struct tmpfs_node *node, int status)
1776 if ((node->tn_status & status) == status)
1778 TMPFS_NODE_LOCK(node);
1779 node->tn_status |= status;
1780 TMPFS_NODE_UNLOCK(node);
1783 /* Sync timestamps */
1785 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1786 const struct timespec *mod)
1788 struct tmpfs_node *node;
1789 struct timespec now;
1791 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes");
1792 node = VP_TO_TMPFS_NODE(vp);
1794 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1795 TMPFS_NODE_CHANGED)) == 0)
1798 vfs_timestamp(&now);
1799 TMPFS_NODE_LOCK(node);
1800 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1803 node->tn_atime = *acc;
1805 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1808 node->tn_mtime = *mod;
1810 if (node->tn_status & TMPFS_NODE_CHANGED)
1811 node->tn_ctime = now;
1812 node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1813 TMPFS_NODE_CHANGED);
1814 TMPFS_NODE_UNLOCK(node);
1816 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
1817 random_harvest_queue(node, sizeof(*node), 1, RANDOM_FS_ATIME);
1821 tmpfs_update(struct vnode *vp)
1824 tmpfs_itimes(vp, NULL, NULL);
1828 tmpfs_truncate(struct vnode *vp, off_t length)
1831 struct tmpfs_node *node;
1833 node = VP_TO_TMPFS_NODE(vp);
1840 if (node->tn_size == length) {
1845 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1848 error = tmpfs_reg_resize(vp, length, FALSE);
1850 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1859 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1861 if (a->td_hash > b->td_hash)
1863 else if (a->td_hash < b->td_hash)
1868 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);