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/rwlock.h>
47 #include <sys/systm.h>
48 #include <sys/sysctl.h>
49 #include <sys/vnode.h>
50 #include <sys/vmmeter.h>
53 #include <vm/vm_param.h>
54 #include <vm/vm_object.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_pageout.h>
57 #include <vm/vm_pager.h>
58 #include <vm/vm_extern.h>
60 #include <fs/tmpfs/tmpfs.h>
61 #include <fs/tmpfs/tmpfs_fifoops.h>
62 #include <fs/tmpfs/tmpfs_vnops.h>
64 struct tmpfs_dir_cursor {
65 struct tmpfs_dirent *tdc_current;
66 struct tmpfs_dirent *tdc_tree;
69 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system");
71 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
74 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
79 pages = *(long *)arg1;
80 bytes = pages * PAGE_SIZE;
82 error = sysctl_handle_long(oidp, &bytes, 0, req);
83 if (error || !req->newptr)
86 pages = bytes / PAGE_SIZE;
87 if (pages < TMPFS_PAGES_MINRESERVED)
90 *(long *)arg1 = pages;
94 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, CTLTYPE_LONG|CTLFLAG_RW,
95 &tmpfs_pages_reserved, 0, sysctl_mem_reserved, "L",
96 "Amount of available memory and swap below which tmpfs growth stops");
98 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a,
99 struct tmpfs_dirent *b);
100 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
103 tmpfs_mem_avail(void)
107 avail = swap_pager_avail + vm_cnt.v_free_count + vm_cnt.v_cache_count -
108 tmpfs_pages_reserved;
109 if (__predict_false(avail < 0))
115 tmpfs_pages_used(struct tmpfs_mount *tmp)
117 const size_t node_size = sizeof(struct tmpfs_node) +
118 sizeof(struct tmpfs_dirent);
121 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
123 return (meta_pages + tmp->tm_pages_used);
127 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
129 if (tmpfs_mem_avail() < req_pages)
132 if (tmp->tm_pages_max != SIZE_MAX &&
133 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
139 /* --------------------------------------------------------------------- */
142 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
143 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
144 * using the credentials of the process 'p'.
146 * If the node type is set to 'VDIR', then the parent parameter must point
147 * to the parent directory of the node being created. It may only be NULL
148 * while allocating the root node.
150 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
151 * specifies the device the node represents.
153 * If the node type is set to 'VLNK', then the parameter target specifies
154 * the file name of the target file for the symbolic link that is being
157 * Note that new nodes are retrieved from the available list if it has
158 * items or, if it is empty, from the node pool as long as there is enough
159 * space to create them.
161 * Returns zero on success or an appropriate error code on failure.
164 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
165 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
166 char *target, dev_t rdev, struct tmpfs_node **node)
168 struct tmpfs_node *nnode;
171 /* If the root directory of the 'tmp' file system is not yet
172 * allocated, this must be the request to do it. */
173 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
175 MPASS(IFF(type == VLNK, target != NULL));
176 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
178 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
180 if (tmpfs_pages_check_avail(tmp, 1) == 0)
183 nnode = (struct tmpfs_node *)uma_zalloc_arg(
184 tmp->tm_node_pool, tmp, M_WAITOK);
186 /* Generic initialization. */
187 nnode->tn_type = type;
188 vfs_timestamp(&nnode->tn_atime);
189 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
193 nnode->tn_mode = mode;
194 nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
196 /* Type-specific initialization. */
197 switch (nnode->tn_type) {
200 nnode->tn_rdev = rdev;
204 RB_INIT(&nnode->tn_dir.tn_dirhead);
205 LIST_INIT(&nnode->tn_dir.tn_dupindex);
206 MPASS(parent != nnode);
207 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
208 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
209 nnode->tn_dir.tn_readdir_lastn = 0;
210 nnode->tn_dir.tn_readdir_lastp = NULL;
212 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
213 nnode->tn_dir.tn_parent->tn_links++;
214 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
223 MPASS(strlen(target) < MAXPATHLEN);
224 nnode->tn_size = strlen(target);
225 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
227 memcpy(nnode->tn_link, target, nnode->tn_size);
231 obj = nnode->tn_reg.tn_aobj =
232 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
233 NULL /* XXXKIB - tmpfs needs swap reservation */);
234 VM_OBJECT_WLOCK(obj);
235 /* OBJ_TMPFS is set together with the setting of vp->v_object */
236 vm_object_set_flag(obj, OBJ_NOSPLIT);
237 vm_object_clear_flag(obj, OBJ_ONEMAPPING);
238 VM_OBJECT_WUNLOCK(obj);
242 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
246 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
247 tmp->tm_nodes_inuse++;
254 /* --------------------------------------------------------------------- */
257 * Destroys the node pointed to by node from the file system 'tmp'.
258 * If the node does not belong to the given mount point, the results are
261 * If the node references a directory; no entries are allowed because
262 * their removal could need a recursive algorithm, something forbidden in
263 * kernel space. Furthermore, there is not need to provide such
264 * functionality (recursive removal) because the only primitives offered
265 * to the user are the removal of empty directories and the deletion of
268 * Note that nodes are not really deleted; in fact, when a node has been
269 * allocated, it cannot be deleted during the whole life of the file
270 * system. Instead, they are moved to the available list and remain there
274 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
279 TMPFS_NODE_LOCK(node);
280 MPASS(node->tn_vnode == NULL);
281 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
282 TMPFS_NODE_UNLOCK(node);
286 LIST_REMOVE(node, tn_entries);
287 tmp->tm_nodes_inuse--;
290 switch (node->tn_type) {
292 /* Do not do anything. VNON is provided to let the
293 * allocation routine clean itself easily by avoiding
294 * duplicating code in it. */
308 free(node->tn_link, M_TMPFSNAME);
312 uobj = node->tn_reg.tn_aobj;
315 tmp->tm_pages_used -= uobj->size;
317 KASSERT((uobj->flags & OBJ_TMPFS) == 0,
318 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj));
319 vm_object_deallocate(uobj);
324 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
327 free_unr(tmp->tm_ino_unr, node->tn_id);
328 uma_zfree(tmp->tm_node_pool, node);
331 /* --------------------------------------------------------------------- */
333 static __inline uint32_t
334 tmpfs_dirent_hash(const char *name, u_int len)
338 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK;
339 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP
342 if (hash < TMPFS_DIRCOOKIE_MIN)
343 hash += TMPFS_DIRCOOKIE_MIN;
348 static __inline off_t
349 tmpfs_dirent_cookie(struct tmpfs_dirent *de)
352 return (TMPFS_DIRCOOKIE_EOF);
354 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN);
356 return (de->td_cookie);
359 static __inline boolean_t
360 tmpfs_dirent_dup(struct tmpfs_dirent *de)
362 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0);
365 static __inline boolean_t
366 tmpfs_dirent_duphead(struct tmpfs_dirent *de)
368 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0);
372 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen)
374 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen);
375 memcpy(de->ud.td_name, name, namelen);
376 de->td_namelen = namelen;
380 * Allocates a new directory entry for the node node with a name of name.
381 * The new directory entry is returned in *de.
383 * The link count of node is increased by one to reflect the new object
386 * Returns zero on success or an appropriate error code on failure.
389 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
390 const char *name, u_int len, struct tmpfs_dirent **de)
392 struct tmpfs_dirent *nde;
394 nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK);
397 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
398 tmpfs_dirent_init(nde, name, len);
409 /* --------------------------------------------------------------------- */
412 * Frees a directory entry. It is the caller's responsibility to destroy
413 * the node referenced by it if needed.
415 * The link count of node is decreased by one to reflect the removal of an
416 * object that referenced it. This only happens if 'node_exists' is true;
417 * otherwise the function will not access the node referred to by the
418 * directory entry, as it may already have been released from the outside.
421 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
423 struct tmpfs_node *node;
427 MPASS(node->tn_links > 0);
430 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL)
431 free(de->ud.td_name, M_TMPFSNAME);
432 uma_zfree(tmp->tm_dirent_pool, de);
435 /* --------------------------------------------------------------------- */
438 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj)
441 if (vp->v_type != VREG || obj == NULL)
444 VM_OBJECT_WLOCK(obj);
446 vm_object_clear_flag(obj, OBJ_TMPFS);
447 obj->un_pager.swp.swp_tmpfs = NULL;
449 VM_OBJECT_WUNLOCK(obj);
453 * Need to clear v_object for insmntque failure.
456 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg)
459 tmpfs_destroy_vobject(vp, vp->v_object);
462 vp->v_op = &dead_vnodeops;
468 * Allocates a new vnode for the node node or returns a new reference to
469 * an existing one if the node had already a vnode referencing it. The
470 * resulting locked vnode is returned in *vpp.
472 * Returns zero on success or an appropriate error code on failure.
475 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
484 TMPFS_NODE_LOCK(node);
486 if ((vp = node->tn_vnode) != NULL) {
487 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
489 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) ||
490 ((vp->v_iflag & VI_DOOMED) != 0 &&
491 (lkflag & LK_NOWAIT) != 0)) {
493 TMPFS_NODE_UNLOCK(node);
498 if ((vp->v_iflag & VI_DOOMED) != 0) {
500 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM;
501 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) {
502 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node),
507 TMPFS_NODE_UNLOCK(node);
508 error = vget(vp, lkflag | LK_INTERLOCK, curthread);
517 * Make sure the vnode is still there after
518 * getting the interlock to avoid racing a free.
520 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
528 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
529 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
530 TMPFS_NODE_UNLOCK(node);
537 * otherwise lock the vp list while we call getnewvnode
538 * since that can block.
540 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
541 node->tn_vpstate |= TMPFS_VNODE_WANT;
542 error = msleep((caddr_t) &node->tn_vpstate,
543 TMPFS_NODE_MTX(node), PDROP | PCATCH,
544 "tmpfs_alloc_vp", 0);
550 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
552 TMPFS_NODE_UNLOCK(node);
554 /* Get a new vnode and associate it with our node. */
555 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp);
560 (void) vn_lock(vp, lkflag | LK_RETRY);
563 vp->v_type = node->tn_type;
565 /* Type-specific initialization. */
566 switch (node->tn_type) {
576 vp->v_op = &tmpfs_fifoop_entries;
579 object = node->tn_reg.tn_aobj;
580 VM_OBJECT_WLOCK(object);
582 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
583 vp->v_object = object;
584 object->un_pager.swp.swp_tmpfs = vp;
585 vm_object_set_flag(object, OBJ_TMPFS);
587 VM_OBJECT_WUNLOCK(object);
590 MPASS(node->tn_dir.tn_parent != NULL);
591 if (node->tn_dir.tn_parent == node)
592 vp->v_vflag |= VV_ROOT;
596 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
599 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL);
604 TMPFS_NODE_LOCK(node);
606 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
607 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
610 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
611 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
612 TMPFS_NODE_UNLOCK(node);
613 wakeup((caddr_t) &node->tn_vpstate);
615 TMPFS_NODE_UNLOCK(node);
622 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
623 TMPFS_NODE_LOCK(node);
624 MPASS(*vpp == node->tn_vnode);
625 TMPFS_NODE_UNLOCK(node);
632 /* --------------------------------------------------------------------- */
635 * Destroys the association between the vnode vp and the node it
639 tmpfs_free_vp(struct vnode *vp)
641 struct tmpfs_node *node;
643 node = VP_TO_TMPFS_NODE(vp);
645 mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED);
646 node->tn_vnode = NULL;
647 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
648 wakeup(&node->tn_vnode);
649 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
653 /* --------------------------------------------------------------------- */
656 * Allocates a new file of type 'type' and adds it to the parent directory
657 * 'dvp'; this addition is done using the component name given in 'cnp'.
658 * The ownership of the new file is automatically assigned based on the
659 * credentials of the caller (through 'cnp'), the group is set based on
660 * the parent directory and the mode is determined from the 'vap' argument.
661 * If successful, *vpp holds a vnode to the newly created file and zero
662 * is returned. Otherwise *vpp is NULL and the function returns an
663 * appropriate error code.
666 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
667 struct componentname *cnp, char *target)
670 struct tmpfs_dirent *de;
671 struct tmpfs_mount *tmp;
672 struct tmpfs_node *dnode;
673 struct tmpfs_node *node;
674 struct tmpfs_node *parent;
676 MPASS(VOP_ISLOCKED(dvp));
677 MPASS(cnp->cn_flags & HASBUF);
679 tmp = VFS_TO_TMPFS(dvp->v_mount);
680 dnode = VP_TO_TMPFS_DIR(dvp);
683 /* If the entry we are creating is a directory, we cannot overflow
684 * the number of links of its parent, because it will get a new
686 if (vap->va_type == VDIR) {
687 /* Ensure that we do not overflow the maximum number of links
688 * imposed by the system. */
689 MPASS(dnode->tn_links <= LINK_MAX);
690 if (dnode->tn_links == LINK_MAX) {
696 MPASS(parent != NULL);
700 /* Allocate a node that represents the new file. */
701 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid,
702 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node);
706 /* Allocate a directory entry that points to the new file. */
707 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
710 tmpfs_free_node(tmp, node);
714 /* Allocate a vnode for the new file. */
715 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
717 tmpfs_free_dirent(tmp, de);
718 tmpfs_free_node(tmp, node);
722 /* Now that all required items are allocated, we can proceed to
723 * insert the new node into the directory, an operation that
725 if (cnp->cn_flags & ISWHITEOUT)
726 tmpfs_dir_whiteout_remove(dvp, cnp);
727 tmpfs_dir_attach(dvp, de);
734 /* --------------------------------------------------------------------- */
736 static struct tmpfs_dirent *
737 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
739 struct tmpfs_dirent *de;
741 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
743 if (de != NULL && tmpfs_dirent_duphead(de))
744 de = LIST_FIRST(&de->ud.td_duphead);
745 dc->tdc_current = de;
747 return (dc->tdc_current);
750 static struct tmpfs_dirent *
751 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
753 struct tmpfs_dirent *de;
755 MPASS(dc->tdc_tree != NULL);
756 if (tmpfs_dirent_dup(dc->tdc_current)) {
757 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
758 if (dc->tdc_current != NULL)
759 return (dc->tdc_current);
761 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
762 &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
763 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
764 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
765 MPASS(dc->tdc_current != NULL);
768 return (dc->tdc_current);
771 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
772 static struct tmpfs_dirent *
773 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
775 struct tmpfs_dirent *de, dekey;
777 dekey.td_hash = hash;
778 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
782 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
783 static struct tmpfs_dirent *
784 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
785 struct tmpfs_dir_cursor *dc)
787 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
788 struct tmpfs_dirent *de, dekey;
790 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
792 if (cookie == node->tn_dir.tn_readdir_lastn &&
793 (de = node->tn_dir.tn_readdir_lastp) != NULL) {
794 /* Protect against possible race, tn_readdir_last[pn]
795 * may be updated with only shared vnode lock held. */
796 if (cookie == tmpfs_dirent_cookie(de))
800 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
801 LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
802 uh.td_dup.index_entries) {
803 MPASS(tmpfs_dirent_dup(de));
804 if (de->td_cookie == cookie)
806 /* dupindex list is sorted. */
807 if (de->td_cookie < cookie) {
816 MPASS((cookie & TMPFS_DIRCOOKIE_MASK) == cookie);
817 dekey.td_hash = cookie;
818 /* Recover if direntry for cookie was removed */
819 de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
821 dc->tdc_current = de;
822 if (de != NULL && tmpfs_dirent_duphead(de)) {
823 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
824 MPASS(dc->tdc_current != NULL);
826 return (dc->tdc_current);
830 dc->tdc_current = de;
831 if (de != NULL && tmpfs_dirent_dup(de))
832 dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
834 return (dc->tdc_current);
838 * Looks for a directory entry in the directory represented by node.
839 * 'cnp' describes the name of the entry to look for. Note that the .
840 * and .. components are not allowed as they do not physically exist
841 * within directories.
843 * Returns a pointer to the entry when found, otherwise NULL.
845 struct tmpfs_dirent *
846 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
847 struct componentname *cnp)
849 struct tmpfs_dir_duphead *duphead;
850 struct tmpfs_dirent *de;
853 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
854 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
855 cnp->cn_nameptr[1] == '.')));
856 TMPFS_VALIDATE_DIR(node);
858 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
859 de = tmpfs_dir_xlookup_hash(node, hash);
860 if (de != NULL && tmpfs_dirent_duphead(de)) {
861 duphead = &de->ud.td_duphead;
862 LIST_FOREACH(de, duphead, uh.td_dup.entries) {
863 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
867 } else if (de != NULL) {
868 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
872 if (de != NULL && f != NULL && de->td_node != f)
879 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
880 * list, allocate new cookie value.
883 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
884 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
886 struct tmpfs_dir_duphead *dupindex;
887 struct tmpfs_dirent *de, *pde;
889 dupindex = &dnode->tn_dir.tn_dupindex;
890 de = LIST_FIRST(dupindex);
891 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
893 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
895 nde->td_cookie = de->td_cookie + 1;
896 MPASS(tmpfs_dirent_dup(nde));
897 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
898 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
903 * Cookie numbers are near exhaustion. Scan dupindex list for unused
904 * numbers. dupindex list is sorted in descending order. Keep it so
905 * after inserting nde.
909 de = LIST_NEXT(de, uh.td_dup.index_entries);
910 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
912 * Last element of the index doesn't have minimal cookie
915 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
916 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
917 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
919 } else if (de == NULL) {
921 * We are so lucky have 2^30 hash duplicates in single
922 * directory :) Return largest possible cookie value.
923 * It should be fine except possible issues with
924 * VOP_READDIR restart.
926 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
927 LIST_INSERT_HEAD(dupindex, nde,
928 uh.td_dup.index_entries);
929 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
932 if (de->td_cookie + 1 == pde->td_cookie ||
933 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
934 continue; /* No hole or invalid cookie. */
935 nde->td_cookie = de->td_cookie + 1;
936 MPASS(tmpfs_dirent_dup(nde));
937 MPASS(pde->td_cookie > nde->td_cookie);
938 MPASS(nde->td_cookie > de->td_cookie);
939 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
940 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
946 * Attaches the directory entry de to the directory represented by vp.
947 * Note that this does not change the link count of the node pointed by
948 * the directory entry, as this is done by tmpfs_alloc_dirent.
951 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
953 struct tmpfs_node *dnode;
954 struct tmpfs_dirent *xde, *nde;
956 ASSERT_VOP_ELOCKED(vp, __func__);
957 MPASS(de->td_namelen > 0);
958 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
959 MPASS(de->td_cookie == de->td_hash);
961 dnode = VP_TO_TMPFS_DIR(vp);
962 dnode->tn_dir.tn_readdir_lastn = 0;
963 dnode->tn_dir.tn_readdir_lastp = NULL;
965 MPASS(!tmpfs_dirent_dup(de));
966 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
967 if (xde != NULL && tmpfs_dirent_duphead(xde))
968 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
969 else if (xde != NULL) {
971 * Allocate new duphead. Swap xde with duphead to avoid
972 * adding/removing elements with the same hash.
974 MPASS(!tmpfs_dirent_dup(xde));
975 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
977 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
978 memcpy(nde, xde, sizeof(*xde));
979 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
980 LIST_INIT(&xde->ud.td_duphead);
983 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
984 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
986 dnode->tn_size += sizeof(struct tmpfs_dirent);
987 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
991 /* --------------------------------------------------------------------- */
994 * Detaches the directory entry de from 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_free_dirent.
999 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
1001 struct tmpfs_mount *tmp;
1002 struct tmpfs_dir *head;
1003 struct tmpfs_node *dnode;
1004 struct tmpfs_dirent *xde;
1006 ASSERT_VOP_ELOCKED(vp, __func__);
1008 dnode = VP_TO_TMPFS_DIR(vp);
1009 head = &dnode->tn_dir.tn_dirhead;
1010 dnode->tn_dir.tn_readdir_lastn = 0;
1011 dnode->tn_dir.tn_readdir_lastp = NULL;
1013 if (tmpfs_dirent_dup(de)) {
1014 /* Remove duphead if de was last entry. */
1015 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1016 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1017 MPASS(tmpfs_dirent_duphead(xde));
1020 LIST_REMOVE(de, uh.td_dup.entries);
1021 LIST_REMOVE(de, uh.td_dup.index_entries);
1023 if (LIST_EMPTY(&xde->ud.td_duphead)) {
1024 RB_REMOVE(tmpfs_dir, head, xde);
1025 tmp = VFS_TO_TMPFS(vp->v_mount);
1026 MPASS(xde->td_node == NULL);
1027 tmpfs_free_dirent(tmp, xde);
1031 RB_REMOVE(tmpfs_dir, head, de);
1033 dnode->tn_size -= sizeof(struct tmpfs_dirent);
1034 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1035 TMPFS_NODE_MODIFIED;
1039 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1041 struct tmpfs_dirent *de, *dde, *nde;
1043 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1044 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1045 /* Node may already be destroyed. */
1047 if (tmpfs_dirent_duphead(de)) {
1048 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1049 LIST_REMOVE(dde, uh.td_dup.entries);
1050 dde->td_node = NULL;
1051 tmpfs_free_dirent(tmp, dde);
1054 tmpfs_free_dirent(tmp, de);
1058 /* --------------------------------------------------------------------- */
1061 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
1062 * directory and returns it in the uio space. The function returns 0
1063 * on success, -1 if there was not enough space in the uio structure to
1064 * hold the directory entry or an appropriate error code if another
1068 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
1073 TMPFS_VALIDATE_DIR(node);
1074 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1076 dent.d_fileno = node->tn_id;
1077 dent.d_type = DT_DIR;
1079 dent.d_name[0] = '.';
1080 dent.d_name[1] = '\0';
1081 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1083 if (dent.d_reclen > uio->uio_resid)
1084 error = EJUSTRETURN;
1086 error = uiomove(&dent, dent.d_reclen, uio);
1088 node->tn_status |= TMPFS_NODE_ACCESSED;
1093 /* --------------------------------------------------------------------- */
1096 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
1097 * directory and returns it in the uio space. The function returns 0
1098 * on success, -1 if there was not enough space in the uio structure to
1099 * hold the directory entry or an appropriate error code if another
1103 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
1108 TMPFS_VALIDATE_DIR(node);
1109 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1112 * Return ENOENT if the current node is already removed.
1114 TMPFS_ASSERT_LOCKED(node);
1115 if (node->tn_dir.tn_parent == NULL) {
1119 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
1120 dent.d_fileno = node->tn_dir.tn_parent->tn_id;
1121 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
1123 dent.d_type = DT_DIR;
1125 dent.d_name[0] = '.';
1126 dent.d_name[1] = '.';
1127 dent.d_name[2] = '\0';
1128 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1130 if (dent.d_reclen > uio->uio_resid)
1131 error = EJUSTRETURN;
1133 error = uiomove(&dent, dent.d_reclen, uio);
1135 node->tn_status |= TMPFS_NODE_ACCESSED;
1140 /* --------------------------------------------------------------------- */
1143 * Helper function for tmpfs_readdir. Returns as much directory entries
1144 * as can fit in the uio space. The read starts at uio->uio_offset.
1145 * The function returns 0 on success, -1 if there was not enough space
1146 * in the uio structure to hold the directory entry or an appropriate
1147 * error code if another error happens.
1150 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies,
1151 u_long *cookies, int *ncookies)
1153 struct tmpfs_dir_cursor dc;
1154 struct tmpfs_dirent *de;
1158 TMPFS_VALIDATE_DIR(node);
1163 * Lookup the node from the current offset. The starting offset of
1164 * 0 will lookup both '.' and '..', and then the first real entry,
1165 * or EOF if there are none. Then find all entries for the dir that
1166 * fit into the buffer. Once no more entries are found (de == NULL),
1167 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1170 switch (uio->uio_offset) {
1171 case TMPFS_DIRCOOKIE_DOT:
1172 error = tmpfs_dir_getdotdent(node, uio);
1175 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
1176 if (cookies != NULL)
1177 cookies[(*ncookies)++] = off = uio->uio_offset;
1179 case TMPFS_DIRCOOKIE_DOTDOT:
1180 error = tmpfs_dir_getdotdotdent(node, uio);
1183 de = tmpfs_dir_first(node, &dc);
1184 uio->uio_offset = tmpfs_dirent_cookie(de);
1185 if (cookies != NULL)
1186 cookies[(*ncookies)++] = off = uio->uio_offset;
1191 case TMPFS_DIRCOOKIE_EOF:
1194 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1197 if (cookies != NULL)
1198 off = tmpfs_dirent_cookie(de);
1201 /* Read as much entries as possible; i.e., until we reach the end of
1202 * the directory or we exhaust uio space. */
1206 /* Create a dirent structure representing the current
1207 * tmpfs_node and fill it. */
1208 if (de->td_node == NULL) {
1212 d.d_fileno = de->td_node->tn_id;
1213 switch (de->td_node->tn_type) {
1243 panic("tmpfs_dir_getdents: type %p %d",
1244 de->td_node, (int)de->td_node->tn_type);
1247 d.d_namlen = de->td_namelen;
1248 MPASS(de->td_namelen < sizeof(d.d_name));
1249 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1250 d.d_name[de->td_namelen] = '\0';
1251 d.d_reclen = GENERIC_DIRSIZ(&d);
1253 /* Stop reading if the directory entry we are treating is
1254 * bigger than the amount of data that can be returned. */
1255 if (d.d_reclen > uio->uio_resid) {
1256 error = EJUSTRETURN;
1260 /* Copy the new dirent structure into the output buffer and
1261 * advance pointers. */
1262 error = uiomove(&d, d.d_reclen, uio);
1264 de = tmpfs_dir_next(node, &dc);
1265 if (cookies != NULL) {
1266 off = tmpfs_dirent_cookie(de);
1267 MPASS(*ncookies < maxcookies);
1268 cookies[(*ncookies)++] = off;
1271 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
1273 /* Skip setting off when using cookies as it is already done above. */
1274 if (cookies == NULL)
1275 off = tmpfs_dirent_cookie(de);
1277 /* Update the offset and cache. */
1278 uio->uio_offset = off;
1279 node->tn_dir.tn_readdir_lastn = off;
1280 node->tn_dir.tn_readdir_lastp = de;
1282 node->tn_status |= TMPFS_NODE_ACCESSED;
1287 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1289 struct tmpfs_dirent *de;
1292 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1293 cnp->cn_nameptr, cnp->cn_namelen, &de);
1296 tmpfs_dir_attach(dvp, de);
1301 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1303 struct tmpfs_dirent *de;
1305 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1306 MPASS(de != NULL && de->td_node == NULL);
1307 tmpfs_dir_detach(dvp, de);
1308 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1311 /* --------------------------------------------------------------------- */
1314 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1315 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
1316 * 'newsize' must be positive.
1318 * Returns zero on success or an appropriate error code on failure.
1321 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1323 struct tmpfs_mount *tmp;
1324 struct tmpfs_node *node;
1327 vm_pindex_t idx, newpages, oldpages;
1331 MPASS(vp->v_type == VREG);
1332 MPASS(newsize >= 0);
1334 node = VP_TO_TMPFS_NODE(vp);
1335 uobj = node->tn_reg.tn_aobj;
1336 tmp = VFS_TO_TMPFS(vp->v_mount);
1339 * Convert the old and new sizes to the number of pages needed to
1340 * store them. It may happen that we do not need to do anything
1341 * because the last allocated page can accommodate the change on
1344 oldsize = node->tn_size;
1345 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1346 MPASS(oldpages == uobj->size);
1347 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1348 if (newpages > oldpages &&
1349 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
1352 VM_OBJECT_WLOCK(uobj);
1353 if (newsize < oldsize) {
1355 * Zero the truncated part of the last page.
1357 base = newsize & PAGE_MASK;
1359 idx = OFF_TO_IDX(newsize);
1361 m = vm_page_lookup(uobj, idx);
1363 if (vm_page_sleep_if_busy(m, "tmfssz"))
1365 MPASS(m->valid == VM_PAGE_BITS_ALL);
1366 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1367 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL);
1369 VM_OBJECT_WUNLOCK(uobj);
1371 VM_OBJECT_WLOCK(uobj);
1373 } else if (m->valid != VM_PAGE_BITS_ALL) {
1375 rv = vm_pager_get_pages(uobj, ma, 1, 0);
1376 m = vm_page_lookup(uobj, idx);
1378 /* A cached page was reactivated. */
1381 if (rv == VM_PAGER_OK) {
1382 vm_page_deactivate(m);
1391 VM_OBJECT_WUNLOCK(uobj);
1397 pmap_zero_page_area(m, base, PAGE_SIZE - base);
1399 vm_pager_page_unswapped(m);
1404 * Release any swap space and free any whole pages.
1406 if (newpages < oldpages) {
1407 swap_pager_freespace(uobj, newpages, oldpages -
1409 vm_object_page_remove(uobj, newpages, 0, 0);
1412 uobj->size = newpages;
1413 VM_OBJECT_WUNLOCK(uobj);
1416 tmp->tm_pages_used += (newpages - oldpages);
1419 node->tn_size = newsize;
1423 /* --------------------------------------------------------------------- */
1426 * Change flags of the given vnode.
1427 * Caller should execute tmpfs_update on vp after a successful execution.
1428 * The vnode must be locked on entry and remain locked on exit.
1431 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
1435 struct tmpfs_node *node;
1437 MPASS(VOP_ISLOCKED(vp));
1439 node = VP_TO_TMPFS_NODE(vp);
1441 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK |
1442 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP |
1443 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
1444 UF_SPARSE | UF_SYSTEM)) != 0)
1445 return (EOPNOTSUPP);
1447 /* Disallow this operation if the file system is mounted read-only. */
1448 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1452 * Callers may only modify the file flags on objects they
1453 * have VADMIN rights for.
1455 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1458 * Unprivileged processes are not permitted to unset system
1459 * flags, or modify flags if any system flags are set.
1461 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1462 if (node->tn_flags &
1463 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1464 error = securelevel_gt(cred, 0);
1469 if (node->tn_flags &
1470 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1471 ((flags ^ node->tn_flags) & SF_SETTABLE))
1474 node->tn_flags = flags;
1475 node->tn_status |= TMPFS_NODE_CHANGED;
1477 MPASS(VOP_ISLOCKED(vp));
1482 /* --------------------------------------------------------------------- */
1485 * Change access mode on the given vnode.
1486 * Caller should execute tmpfs_update on vp after a successful execution.
1487 * The vnode must be locked on entry and remain locked on exit.
1490 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1493 struct tmpfs_node *node;
1495 MPASS(VOP_ISLOCKED(vp));
1497 node = VP_TO_TMPFS_NODE(vp);
1499 /* Disallow this operation if the file system is mounted read-only. */
1500 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1503 /* Immutable or append-only files cannot be modified, either. */
1504 if (node->tn_flags & (IMMUTABLE | APPEND))
1508 * To modify the permissions on a file, must possess VADMIN
1511 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1515 * Privileged processes may set the sticky bit on non-directories,
1516 * as well as set the setgid bit on a file with a group that the
1517 * process is not a member of.
1519 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1520 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1523 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1524 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1530 node->tn_mode &= ~ALLPERMS;
1531 node->tn_mode |= mode & ALLPERMS;
1533 node->tn_status |= TMPFS_NODE_CHANGED;
1535 MPASS(VOP_ISLOCKED(vp));
1540 /* --------------------------------------------------------------------- */
1543 * Change ownership of the given vnode. At least one of uid or gid must
1544 * be different than VNOVAL. If one is set to that value, the attribute
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_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1554 struct tmpfs_node *node;
1558 MPASS(VOP_ISLOCKED(vp));
1560 node = VP_TO_TMPFS_NODE(vp);
1562 /* Assign default values if they are unknown. */
1563 MPASS(uid != VNOVAL || gid != VNOVAL);
1568 MPASS(uid != VNOVAL && gid != VNOVAL);
1570 /* Disallow this operation if the file system is mounted read-only. */
1571 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1574 /* Immutable or append-only files cannot be modified, either. */
1575 if (node->tn_flags & (IMMUTABLE | APPEND))
1579 * To modify the ownership of a file, must possess VADMIN for that
1582 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1586 * To change the owner of a file, or change the group of a file to a
1587 * group of which we are not a member, the caller must have
1590 if ((uid != node->tn_uid ||
1591 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1592 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1595 ogid = node->tn_gid;
1596 ouid = node->tn_uid;
1601 node->tn_status |= TMPFS_NODE_CHANGED;
1603 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1604 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1605 node->tn_mode &= ~(S_ISUID | S_ISGID);
1608 MPASS(VOP_ISLOCKED(vp));
1613 /* --------------------------------------------------------------------- */
1616 * Change size of the given vnode.
1617 * Caller should execute tmpfs_update on vp after a successful execution.
1618 * The vnode must be locked on entry and remain locked on exit.
1621 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1625 struct tmpfs_node *node;
1627 MPASS(VOP_ISLOCKED(vp));
1629 node = VP_TO_TMPFS_NODE(vp);
1631 /* Decide whether this is a valid operation based on the file type. */
1633 switch (vp->v_type) {
1638 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1647 /* Allow modifications of special files even if in the file
1648 * system is mounted read-only (we are not modifying the
1649 * files themselves, but the objects they represent). */
1653 /* Anything else is unsupported. */
1657 /* Immutable or append-only files cannot be modified, either. */
1658 if (node->tn_flags & (IMMUTABLE | APPEND))
1661 error = tmpfs_truncate(vp, size);
1662 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1663 * for us, as will update tn_status; no need to do that here. */
1665 MPASS(VOP_ISLOCKED(vp));
1670 /* --------------------------------------------------------------------- */
1673 * Change access and modification times of the given vnode.
1674 * Caller should execute tmpfs_update on vp after a successful execution.
1675 * The vnode must be locked on entry and remain locked on exit.
1678 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1679 struct timespec *birthtime, int vaflags, struct ucred *cred, struct thread *l)
1682 struct tmpfs_node *node;
1684 MPASS(VOP_ISLOCKED(vp));
1686 node = VP_TO_TMPFS_NODE(vp);
1688 /* Disallow this operation if the file system is mounted read-only. */
1689 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1692 /* Immutable or append-only files cannot be modified, either. */
1693 if (node->tn_flags & (IMMUTABLE | APPEND))
1696 /* Determine if the user have proper privilege to update time. */
1697 if (vaflags & VA_UTIMES_NULL) {
1698 error = VOP_ACCESS(vp, VADMIN, cred, l);
1700 error = VOP_ACCESS(vp, VWRITE, cred, l);
1702 error = VOP_ACCESS(vp, VADMIN, cred, l);
1706 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1707 node->tn_status |= TMPFS_NODE_ACCESSED;
1709 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1710 node->tn_status |= TMPFS_NODE_MODIFIED;
1712 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1713 node->tn_status |= TMPFS_NODE_MODIFIED;
1715 tmpfs_itimes(vp, atime, mtime);
1717 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1718 node->tn_birthtime = *birthtime;
1719 MPASS(VOP_ISLOCKED(vp));
1724 /* --------------------------------------------------------------------- */
1725 /* Sync timestamps */
1727 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1728 const struct timespec *mod)
1730 struct tmpfs_node *node;
1731 struct timespec now;
1733 node = VP_TO_TMPFS_NODE(vp);
1735 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1736 TMPFS_NODE_CHANGED)) == 0)
1739 vfs_timestamp(&now);
1740 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1743 node->tn_atime = *acc;
1745 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1748 node->tn_mtime = *mod;
1750 if (node->tn_status & TMPFS_NODE_CHANGED) {
1751 node->tn_ctime = now;
1754 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1757 /* --------------------------------------------------------------------- */
1760 tmpfs_update(struct vnode *vp)
1763 tmpfs_itimes(vp, NULL, NULL);
1766 /* --------------------------------------------------------------------- */
1769 tmpfs_truncate(struct vnode *vp, off_t length)
1772 struct tmpfs_node *node;
1774 node = VP_TO_TMPFS_NODE(vp);
1781 if (node->tn_size == length) {
1786 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1789 error = tmpfs_reg_resize(vp, length, FALSE);
1791 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1801 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1803 if (a->td_hash > b->td_hash)
1805 else if (a->td_hash < b->td_hash)
1810 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);