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 + cnt.v_free_count + 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 != ULONG_MAX &&
133 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
140 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
141 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
142 * using the credentials of the process 'p'.
144 * If the node type is set to 'VDIR', then the parent parameter must point
145 * to the parent directory of the node being created. It may only be NULL
146 * while allocating the root node.
148 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
149 * specifies the device the node represents.
151 * If the node type is set to 'VLNK', then the parameter target specifies
152 * the file name of the target file for the symbolic link that is being
155 * Note that new nodes are retrieved from the available list if it has
156 * items or, if it is empty, from the node pool as long as there is enough
157 * space to create them.
159 * Returns zero on success or an appropriate error code on failure.
162 tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type,
163 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
164 char *target, dev_t rdev, struct tmpfs_node **node)
166 struct tmpfs_node *nnode;
169 /* If the root directory of the 'tmp' file system is not yet
170 * allocated, this must be the request to do it. */
171 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
172 KASSERT(tmp->tm_root == NULL || mp->mnt_writeopcount > 0,
173 ("creating node not under vn_start_write"));
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 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
185 * When a new tmpfs node is created for fully
186 * constructed mount point, there must be a parent
187 * node, which vnode is locked exclusively. As
188 * consequence, if the unmount is executing in
189 * parallel, vflush() cannot reclaim the parent vnode.
190 * Due to this, the check for MNTK_UNMOUNT flag is not
191 * racy: if we did not see MNTK_UNMOUNT flag, then tmp
192 * cannot be destroyed until node construction is
193 * finished and the parent vnode unlocked.
195 * Tmpfs does not need to instantiate new nodes during
201 nnode = (struct tmpfs_node *)uma_zalloc_arg(tmp->tm_node_pool, tmp,
204 /* Generic initialization. */
205 nnode->tn_type = type;
206 vfs_timestamp(&nnode->tn_atime);
207 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
211 nnode->tn_mode = mode;
212 nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
214 /* Type-specific initialization. */
215 switch (nnode->tn_type) {
218 nnode->tn_rdev = rdev;
222 RB_INIT(&nnode->tn_dir.tn_dirhead);
223 LIST_INIT(&nnode->tn_dir.tn_dupindex);
224 MPASS(parent != nnode);
225 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
226 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
227 nnode->tn_dir.tn_readdir_lastn = 0;
228 nnode->tn_dir.tn_readdir_lastp = NULL;
230 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
231 nnode->tn_dir.tn_parent->tn_links++;
232 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
241 MPASS(strlen(target) < MAXPATHLEN);
242 nnode->tn_size = strlen(target);
243 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
245 memcpy(nnode->tn_link, target, nnode->tn_size);
249 obj = nnode->tn_reg.tn_aobj =
250 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
251 NULL /* XXXKIB - tmpfs needs swap reservation */);
252 VM_OBJECT_WLOCK(obj);
253 /* OBJ_TMPFS is set together with the setting of vp->v_object */
254 vm_object_set_flag(obj, OBJ_NOSPLIT | OBJ_TMPFS_NODE);
255 vm_object_clear_flag(obj, OBJ_ONEMAPPING);
256 VM_OBJECT_WUNLOCK(obj);
260 panic("tmpfs_alloc_node: type %p %d", nnode,
261 (int)nnode->tn_type);
265 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
266 tmp->tm_nodes_inuse++;
274 * Destroys the node pointed to by node from the file system 'tmp'.
275 * If the node references a directory, no entries are allowed.
278 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
283 TMPFS_NODE_LOCK(node);
284 MPASS(node->tn_vnode == NULL);
285 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
286 TMPFS_NODE_UNLOCK(node);
290 LIST_REMOVE(node, tn_entries);
291 tmp->tm_nodes_inuse--;
294 switch (node->tn_type) {
296 /* Do not do anything. VNON is provided to let the
297 * allocation routine clean itself easily by avoiding
298 * duplicating code in it. */
312 free(node->tn_link, M_TMPFSNAME);
316 uobj = node->tn_reg.tn_aobj;
318 atomic_subtract_long(&tmp->tm_pages_used, uobj->size);
319 KASSERT((uobj->flags & OBJ_TMPFS) == 0,
320 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj));
321 vm_object_deallocate(uobj);
326 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
329 free_unr(tmp->tm_ino_unr, node->tn_id);
330 uma_zfree(tmp->tm_node_pool, node);
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);
410 * Frees a directory entry. It is the caller's responsibility to destroy
411 * the node referenced by it if needed.
413 * The link count of node is decreased by one to reflect the removal of an
414 * object that referenced it. This only happens if 'node_exists' is true;
415 * otherwise the function will not access the node referred to by the
416 * directory entry, as it may already have been released from the outside.
419 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
421 struct tmpfs_node *node;
425 MPASS(node->tn_links > 0);
428 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL)
429 free(de->ud.td_name, M_TMPFSNAME);
430 uma_zfree(tmp->tm_dirent_pool, de);
434 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj)
437 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject");
438 if (vp->v_type != VREG || obj == NULL)
441 VM_OBJECT_WLOCK(obj);
443 vm_object_clear_flag(obj, OBJ_TMPFS);
444 obj->un_pager.swp.swp_tmpfs = NULL;
446 VM_OBJECT_WUNLOCK(obj);
450 * Need to clear v_object for insmntque failure.
453 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg)
456 tmpfs_destroy_vobject(vp, vp->v_object);
459 vp->v_op = &dead_vnodeops;
465 * Allocates a new vnode for the node node or returns a new reference to
466 * an existing one if the node had already a vnode referencing it. The
467 * resulting locked vnode is returned in *vpp.
469 * Returns zero on success or an appropriate error code on failure.
472 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
481 TMPFS_NODE_LOCK(node);
483 if ((vp = node->tn_vnode) != NULL) {
484 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
486 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) ||
487 ((vp->v_iflag & VI_DOOMED) != 0 &&
488 (lkflag & LK_NOWAIT) != 0)) {
490 TMPFS_NODE_UNLOCK(node);
495 if ((vp->v_iflag & VI_DOOMED) != 0) {
497 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM;
498 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) {
499 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node),
504 TMPFS_NODE_UNLOCK(node);
505 error = vget(vp, lkflag | LK_INTERLOCK, curthread);
514 * Make sure the vnode is still there after
515 * getting the interlock to avoid racing a free.
517 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
525 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
526 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
527 TMPFS_NODE_UNLOCK(node);
534 * otherwise lock the vp list while we call getnewvnode
535 * since that can block.
537 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
538 node->tn_vpstate |= TMPFS_VNODE_WANT;
539 error = msleep((caddr_t) &node->tn_vpstate,
540 TMPFS_NODE_MTX(node), PDROP | PCATCH,
541 "tmpfs_alloc_vp", 0);
547 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
549 TMPFS_NODE_UNLOCK(node);
551 /* Get a new vnode and associate it with our node. */
552 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp);
557 /* lkflag is ignored, the lock is exclusive */
558 (void) vn_lock(vp, lkflag | LK_RETRY);
561 vp->v_type = node->tn_type;
563 /* Type-specific initialization. */
564 switch (node->tn_type) {
574 vp->v_op = &tmpfs_fifoop_entries;
577 object = node->tn_reg.tn_aobj;
578 VM_OBJECT_WLOCK(object);
580 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
581 vp->v_object = object;
582 object->un_pager.swp.swp_tmpfs = vp;
583 vm_object_set_flag(object, OBJ_TMPFS);
585 VM_OBJECT_WUNLOCK(object);
588 MPASS(node->tn_dir.tn_parent != NULL);
589 if (node->tn_dir.tn_parent == node)
590 vp->v_vflag |= VV_ROOT;
594 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
596 if (vp->v_type != VFIFO)
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);
633 * Destroys the association between the vnode vp and the node it
637 tmpfs_free_vp(struct vnode *vp)
639 struct tmpfs_node *node;
641 node = VP_TO_TMPFS_NODE(vp);
643 TMPFS_NODE_ASSERT_LOCKED(node);
644 node->tn_vnode = NULL;
645 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
646 wakeup(&node->tn_vnode);
647 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
652 * Allocates a new file of type 'type' and adds it to the parent directory
653 * 'dvp'; this addition is done using the component name given in 'cnp'.
654 * The ownership of the new file is automatically assigned based on the
655 * credentials of the caller (through 'cnp'), the group is set based on
656 * the parent directory and the mode is determined from the 'vap' argument.
657 * If successful, *vpp holds a vnode to the newly created file and zero
658 * is returned. Otherwise *vpp is NULL and the function returns an
659 * appropriate error code.
662 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
663 struct componentname *cnp, char *target)
666 struct tmpfs_dirent *de;
667 struct tmpfs_mount *tmp;
668 struct tmpfs_node *dnode;
669 struct tmpfs_node *node;
670 struct tmpfs_node *parent;
672 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file");
673 MPASS(cnp->cn_flags & HASBUF);
675 tmp = VFS_TO_TMPFS(dvp->v_mount);
676 dnode = VP_TO_TMPFS_DIR(dvp);
679 /* If the entry we are creating is a directory, we cannot overflow
680 * the number of links of its parent, because it will get a new
682 if (vap->va_type == VDIR) {
683 /* Ensure that we do not overflow the maximum number of links
684 * imposed by the system. */
685 MPASS(dnode->tn_links <= LINK_MAX);
686 if (dnode->tn_links == LINK_MAX) {
691 MPASS(parent != NULL);
695 /* Allocate a node that represents the new file. */
696 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type,
697 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent,
698 target, vap->va_rdev, &node);
702 /* Allocate a directory entry that points to the new file. */
703 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
706 tmpfs_free_node(tmp, node);
710 /* Allocate a vnode for the new file. */
711 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
713 tmpfs_free_dirent(tmp, de);
714 tmpfs_free_node(tmp, node);
718 /* Now that all required items are allocated, we can proceed to
719 * insert the new node into the directory, an operation that
721 if (cnp->cn_flags & ISWHITEOUT)
722 tmpfs_dir_whiteout_remove(dvp, cnp);
723 tmpfs_dir_attach(dvp, de);
727 static struct tmpfs_dirent *
728 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
730 struct tmpfs_dirent *de;
732 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
734 if (de != NULL && tmpfs_dirent_duphead(de))
735 de = LIST_FIRST(&de->ud.td_duphead);
736 dc->tdc_current = de;
738 return (dc->tdc_current);
741 static struct tmpfs_dirent *
742 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
744 struct tmpfs_dirent *de;
746 MPASS(dc->tdc_tree != NULL);
747 if (tmpfs_dirent_dup(dc->tdc_current)) {
748 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
749 if (dc->tdc_current != NULL)
750 return (dc->tdc_current);
752 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
753 &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
754 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
755 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
756 MPASS(dc->tdc_current != NULL);
759 return (dc->tdc_current);
762 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
763 static struct tmpfs_dirent *
764 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
766 struct tmpfs_dirent *de, dekey;
768 dekey.td_hash = hash;
769 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
773 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
774 static struct tmpfs_dirent *
775 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
776 struct tmpfs_dir_cursor *dc)
778 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
779 struct tmpfs_dirent *de, dekey;
781 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
783 if (cookie == node->tn_dir.tn_readdir_lastn &&
784 (de = node->tn_dir.tn_readdir_lastp) != NULL) {
785 /* Protect against possible race, tn_readdir_last[pn]
786 * may be updated with only shared vnode lock held. */
787 if (cookie == tmpfs_dirent_cookie(de))
791 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
792 LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
793 uh.td_dup.index_entries) {
794 MPASS(tmpfs_dirent_dup(de));
795 if (de->td_cookie == cookie)
797 /* dupindex list is sorted. */
798 if (de->td_cookie < cookie) {
807 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) {
810 dekey.td_hash = cookie;
811 /* Recover if direntry for cookie was removed */
812 de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
815 dc->tdc_current = de;
816 if (de != NULL && tmpfs_dirent_duphead(de)) {
817 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
818 MPASS(dc->tdc_current != NULL);
820 return (dc->tdc_current);
824 dc->tdc_current = de;
825 if (de != NULL && tmpfs_dirent_dup(de))
826 dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
828 return (dc->tdc_current);
832 * Looks for a directory entry in the directory represented by node.
833 * 'cnp' describes the name of the entry to look for. Note that the .
834 * and .. components are not allowed as they do not physically exist
835 * within directories.
837 * Returns a pointer to the entry when found, otherwise NULL.
839 struct tmpfs_dirent *
840 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
841 struct componentname *cnp)
843 struct tmpfs_dir_duphead *duphead;
844 struct tmpfs_dirent *de;
847 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
848 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
849 cnp->cn_nameptr[1] == '.')));
850 TMPFS_VALIDATE_DIR(node);
852 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
853 de = tmpfs_dir_xlookup_hash(node, hash);
854 if (de != NULL && tmpfs_dirent_duphead(de)) {
855 duphead = &de->ud.td_duphead;
856 LIST_FOREACH(de, duphead, uh.td_dup.entries) {
857 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
861 } else if (de != NULL) {
862 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
866 if (de != NULL && f != NULL && de->td_node != f)
873 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
874 * list, allocate new cookie value.
877 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
878 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
880 struct tmpfs_dir_duphead *dupindex;
881 struct tmpfs_dirent *de, *pde;
883 dupindex = &dnode->tn_dir.tn_dupindex;
884 de = LIST_FIRST(dupindex);
885 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
887 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
889 nde->td_cookie = de->td_cookie + 1;
890 MPASS(tmpfs_dirent_dup(nde));
891 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
892 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
897 * Cookie numbers are near exhaustion. Scan dupindex list for unused
898 * numbers. dupindex list is sorted in descending order. Keep it so
899 * after inserting nde.
903 de = LIST_NEXT(de, uh.td_dup.index_entries);
904 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
906 * Last element of the index doesn't have minimal cookie
909 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
910 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
911 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
913 } else if (de == NULL) {
915 * We are so lucky have 2^30 hash duplicates in single
916 * directory :) Return largest possible cookie value.
917 * It should be fine except possible issues with
918 * VOP_READDIR restart.
920 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
921 LIST_INSERT_HEAD(dupindex, nde,
922 uh.td_dup.index_entries);
923 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
926 if (de->td_cookie + 1 == pde->td_cookie ||
927 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
928 continue; /* No hole or invalid cookie. */
929 nde->td_cookie = de->td_cookie + 1;
930 MPASS(tmpfs_dirent_dup(nde));
931 MPASS(pde->td_cookie > nde->td_cookie);
932 MPASS(nde->td_cookie > de->td_cookie);
933 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
934 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
940 * Attaches the directory entry de to the directory represented by vp.
941 * Note that this does not change the link count of the node pointed by
942 * the directory entry, as this is done by tmpfs_alloc_dirent.
945 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
947 struct tmpfs_node *dnode;
948 struct tmpfs_dirent *xde, *nde;
950 ASSERT_VOP_ELOCKED(vp, __func__);
951 MPASS(de->td_namelen > 0);
952 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
953 MPASS(de->td_cookie == de->td_hash);
955 dnode = VP_TO_TMPFS_DIR(vp);
956 dnode->tn_dir.tn_readdir_lastn = 0;
957 dnode->tn_dir.tn_readdir_lastp = NULL;
959 MPASS(!tmpfs_dirent_dup(de));
960 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
961 if (xde != NULL && tmpfs_dirent_duphead(xde))
962 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
963 else if (xde != NULL) {
965 * Allocate new duphead. Swap xde with duphead to avoid
966 * adding/removing elements with the same hash.
968 MPASS(!tmpfs_dirent_dup(xde));
969 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
971 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
972 memcpy(nde, xde, sizeof(*xde));
973 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
974 LIST_INIT(&xde->ud.td_duphead);
977 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
978 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
980 dnode->tn_size += sizeof(struct tmpfs_dirent);
981 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
987 * Detaches the directory entry de from the directory represented by vp.
988 * Note that this does not change the link count of the node pointed by
989 * the directory entry, as this is done by tmpfs_free_dirent.
992 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
994 struct tmpfs_mount *tmp;
995 struct tmpfs_dir *head;
996 struct tmpfs_node *dnode;
997 struct tmpfs_dirent *xde;
999 ASSERT_VOP_ELOCKED(vp, __func__);
1001 dnode = VP_TO_TMPFS_DIR(vp);
1002 head = &dnode->tn_dir.tn_dirhead;
1003 dnode->tn_dir.tn_readdir_lastn = 0;
1004 dnode->tn_dir.tn_readdir_lastp = NULL;
1006 if (tmpfs_dirent_dup(de)) {
1007 /* Remove duphead if de was last entry. */
1008 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1009 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1010 MPASS(tmpfs_dirent_duphead(xde));
1013 LIST_REMOVE(de, uh.td_dup.entries);
1014 LIST_REMOVE(de, uh.td_dup.index_entries);
1016 if (LIST_EMPTY(&xde->ud.td_duphead)) {
1017 RB_REMOVE(tmpfs_dir, head, xde);
1018 tmp = VFS_TO_TMPFS(vp->v_mount);
1019 MPASS(xde->td_node == NULL);
1020 tmpfs_free_dirent(tmp, xde);
1023 de->td_cookie = de->td_hash;
1025 RB_REMOVE(tmpfs_dir, head, de);
1027 dnode->tn_size -= sizeof(struct tmpfs_dirent);
1028 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1029 TMPFS_NODE_MODIFIED;
1034 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1036 struct tmpfs_dirent *de, *dde, *nde;
1038 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1039 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1040 /* Node may already be destroyed. */
1042 if (tmpfs_dirent_duphead(de)) {
1043 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1044 LIST_REMOVE(dde, uh.td_dup.entries);
1045 dde->td_node = NULL;
1046 tmpfs_free_dirent(tmp, dde);
1049 tmpfs_free_dirent(tmp, de);
1054 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
1055 * directory and returns it in the uio space. The function returns 0
1056 * on success, -1 if there was not enough space in the uio structure to
1057 * hold the directory entry or an appropriate error code if another
1061 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
1066 TMPFS_VALIDATE_DIR(node);
1067 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1069 dent.d_fileno = node->tn_id;
1070 dent.d_type = DT_DIR;
1072 dent.d_name[0] = '.';
1073 dent.d_name[1] = '\0';
1074 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1076 if (dent.d_reclen > uio->uio_resid)
1077 error = EJUSTRETURN;
1079 error = uiomove(&dent, dent.d_reclen, uio);
1081 tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
1087 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
1088 * directory and returns it in the uio space. The function returns 0
1089 * on success, -1 if there was not enough space in the uio structure to
1090 * hold the directory entry or an appropriate error code if another
1094 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
1099 TMPFS_VALIDATE_DIR(node);
1100 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1103 * Return ENOENT if the current node is already removed.
1105 TMPFS_ASSERT_LOCKED(node);
1106 if (node->tn_dir.tn_parent == NULL)
1109 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
1110 dent.d_fileno = node->tn_dir.tn_parent->tn_id;
1111 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
1113 dent.d_type = DT_DIR;
1115 dent.d_name[0] = '.';
1116 dent.d_name[1] = '.';
1117 dent.d_name[2] = '\0';
1118 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1120 if (dent.d_reclen > uio->uio_resid)
1121 error = EJUSTRETURN;
1123 error = uiomove(&dent, dent.d_reclen, uio);
1125 tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
1131 * Helper function for tmpfs_readdir. Returns as much directory entries
1132 * as can fit in the uio space. The read starts at uio->uio_offset.
1133 * The function returns 0 on success, -1 if there was not enough space
1134 * in the uio structure to hold the directory entry or an appropriate
1135 * error code if another error happens.
1138 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies,
1139 u_long *cookies, int *ncookies)
1141 struct tmpfs_dir_cursor dc;
1142 struct tmpfs_dirent *de;
1146 TMPFS_VALIDATE_DIR(node);
1151 * Lookup the node from the current offset. The starting offset of
1152 * 0 will lookup both '.' and '..', and then the first real entry,
1153 * or EOF if there are none. Then find all entries for the dir that
1154 * fit into the buffer. Once no more entries are found (de == NULL),
1155 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1158 switch (uio->uio_offset) {
1159 case TMPFS_DIRCOOKIE_DOT:
1160 error = tmpfs_dir_getdotdent(node, uio);
1163 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
1164 if (cookies != NULL)
1165 cookies[(*ncookies)++] = off = uio->uio_offset;
1167 case TMPFS_DIRCOOKIE_DOTDOT:
1168 error = tmpfs_dir_getdotdotdent(node, uio);
1171 de = tmpfs_dir_first(node, &dc);
1172 uio->uio_offset = tmpfs_dirent_cookie(de);
1173 if (cookies != NULL)
1174 cookies[(*ncookies)++] = off = uio->uio_offset;
1179 case TMPFS_DIRCOOKIE_EOF:
1182 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1185 if (cookies != NULL)
1186 off = tmpfs_dirent_cookie(de);
1189 /* Read as much entries as possible; i.e., until we reach the end of
1190 * the directory or we exhaust uio space. */
1194 /* Create a dirent structure representing the current
1195 * tmpfs_node and fill it. */
1196 if (de->td_node == NULL) {
1200 d.d_fileno = de->td_node->tn_id;
1201 switch (de->td_node->tn_type) {
1231 panic("tmpfs_dir_getdents: type %p %d",
1232 de->td_node, (int)de->td_node->tn_type);
1235 d.d_namlen = de->td_namelen;
1236 MPASS(de->td_namelen < sizeof(d.d_name));
1237 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1238 d.d_name[de->td_namelen] = '\0';
1239 d.d_reclen = GENERIC_DIRSIZ(&d);
1241 /* Stop reading if the directory entry we are treating is
1242 * bigger than the amount of data that can be returned. */
1243 if (d.d_reclen > uio->uio_resid) {
1244 error = EJUSTRETURN;
1248 /* Copy the new dirent structure into the output buffer and
1249 * advance pointers. */
1250 error = uiomove(&d, d.d_reclen, uio);
1252 de = tmpfs_dir_next(node, &dc);
1253 if (cookies != NULL) {
1254 off = tmpfs_dirent_cookie(de);
1255 MPASS(*ncookies < maxcookies);
1256 cookies[(*ncookies)++] = off;
1259 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
1261 /* Skip setting off when using cookies as it is already done above. */
1262 if (cookies == NULL)
1263 off = tmpfs_dirent_cookie(de);
1265 /* Update the offset and cache. */
1266 uio->uio_offset = off;
1267 node->tn_dir.tn_readdir_lastn = off;
1268 node->tn_dir.tn_readdir_lastp = de;
1270 tmpfs_set_status(node, TMPFS_NODE_ACCESSED);
1275 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1277 struct tmpfs_dirent *de;
1280 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1281 cnp->cn_nameptr, cnp->cn_namelen, &de);
1284 tmpfs_dir_attach(dvp, de);
1289 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1291 struct tmpfs_dirent *de;
1293 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1294 MPASS(de != NULL && de->td_node == NULL);
1295 tmpfs_dir_detach(dvp, de);
1296 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1300 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1301 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
1302 * 'newsize' must be positive.
1304 * Returns zero on success or an appropriate error code on failure.
1307 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1309 struct tmpfs_mount *tmp;
1310 struct tmpfs_node *node;
1313 vm_pindex_t idx, newpages, oldpages;
1317 MPASS(vp->v_type == VREG);
1318 MPASS(newsize >= 0);
1320 node = VP_TO_TMPFS_NODE(vp);
1321 uobj = node->tn_reg.tn_aobj;
1322 tmp = VFS_TO_TMPFS(vp->v_mount);
1325 * Convert the old and new sizes to the number of pages needed to
1326 * store them. It may happen that we do not need to do anything
1327 * because the last allocated page can accommodate the change on
1330 oldsize = node->tn_size;
1331 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1332 MPASS(oldpages == uobj->size);
1333 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1334 if (newpages > oldpages &&
1335 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
1338 VM_OBJECT_WLOCK(uobj);
1339 if (newsize < oldsize) {
1341 * Zero the truncated part of the last page.
1343 base = newsize & PAGE_MASK;
1345 idx = OFF_TO_IDX(newsize);
1347 m = vm_page_lookup(uobj, idx);
1349 if (vm_page_sleep_if_busy(m, "tmfssz"))
1351 MPASS(m->valid == VM_PAGE_BITS_ALL);
1352 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1353 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL);
1355 VM_OBJECT_WUNLOCK(uobj);
1357 VM_OBJECT_WLOCK(uobj);
1359 } else if (m->valid != VM_PAGE_BITS_ALL) {
1361 rv = vm_pager_get_pages(uobj, ma, 1, 0);
1362 m = vm_page_lookup(uobj, idx);
1364 /* A cached page was reactivated. */
1367 if (rv == VM_PAGER_OK) {
1368 vm_page_deactivate(m);
1377 VM_OBJECT_WUNLOCK(uobj);
1383 pmap_zero_page_area(m, base, PAGE_SIZE - base);
1385 vm_pager_page_unswapped(m);
1390 * Release any swap space and free any whole pages.
1392 if (newpages < oldpages) {
1393 swap_pager_freespace(uobj, newpages, oldpages -
1395 vm_object_page_remove(uobj, newpages, 0, 0);
1398 uobj->size = newpages;
1399 VM_OBJECT_WUNLOCK(uobj);
1401 atomic_add_long(&tmp->tm_pages_used, newpages - oldpages);
1403 node->tn_size = newsize;
1408 tmpfs_check_mtime(struct vnode *vp)
1410 struct tmpfs_node *node;
1411 struct vm_object *obj;
1413 ASSERT_VOP_ELOCKED(vp, "check_mtime");
1414 if (vp->v_type != VREG)
1417 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
1418 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
1420 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) {
1421 VM_OBJECT_WLOCK(obj);
1422 if ((obj->flags & OBJ_TMPFS_DIRTY) != 0) {
1423 obj->flags &= ~OBJ_TMPFS_DIRTY;
1424 node = VP_TO_TMPFS_NODE(vp);
1425 node->tn_status |= TMPFS_NODE_MODIFIED |
1428 VM_OBJECT_WUNLOCK(obj);
1433 * Change flags of the given vnode.
1434 * Caller should execute tmpfs_update on vp after a successful execution.
1435 * The vnode must be locked on entry and remain locked on exit.
1438 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
1442 struct tmpfs_node *node;
1444 ASSERT_VOP_ELOCKED(vp, "chflags");
1446 node = VP_TO_TMPFS_NODE(vp);
1448 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK |
1449 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP |
1450 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
1451 UF_SPARSE | UF_SYSTEM)) != 0)
1452 return (EOPNOTSUPP);
1454 /* Disallow this operation if the file system is mounted read-only. */
1455 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1459 * Callers may only modify the file flags on objects they
1460 * have VADMIN rights for.
1462 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1465 * Unprivileged processes are not permitted to unset system
1466 * flags, or modify flags if any system flags are set.
1468 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1469 if (node->tn_flags &
1470 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1471 error = securelevel_gt(cred, 0);
1476 if (node->tn_flags &
1477 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1478 ((flags ^ node->tn_flags) & SF_SETTABLE))
1481 node->tn_flags = flags;
1482 node->tn_status |= TMPFS_NODE_CHANGED;
1484 ASSERT_VOP_ELOCKED(vp, "chflags2");
1490 * Change access mode on the given vnode.
1491 * Caller should execute tmpfs_update on vp after a successful execution.
1492 * The vnode must be locked on entry and remain locked on exit.
1495 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1498 struct tmpfs_node *node;
1500 ASSERT_VOP_ELOCKED(vp, "chmod");
1502 node = VP_TO_TMPFS_NODE(vp);
1504 /* Disallow this operation if the file system is mounted read-only. */
1505 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1508 /* Immutable or append-only files cannot be modified, either. */
1509 if (node->tn_flags & (IMMUTABLE | APPEND))
1513 * To modify the permissions on a file, must possess VADMIN
1516 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1520 * Privileged processes may set the sticky bit on non-directories,
1521 * as well as set the setgid bit on a file with a group that the
1522 * process is not a member of.
1524 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1525 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1528 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1529 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1535 node->tn_mode &= ~ALLPERMS;
1536 node->tn_mode |= mode & ALLPERMS;
1538 node->tn_status |= TMPFS_NODE_CHANGED;
1540 ASSERT_VOP_ELOCKED(vp, "chmod2");
1546 * Change ownership of the given vnode. At least one of uid or gid must
1547 * be different than VNOVAL. If one is set to that value, the attribute
1549 * Caller should execute tmpfs_update on vp after a successful execution.
1550 * The vnode must be locked on entry and remain locked on exit.
1553 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1557 struct tmpfs_node *node;
1561 ASSERT_VOP_ELOCKED(vp, "chown");
1563 node = VP_TO_TMPFS_NODE(vp);
1565 /* Assign default values if they are unknown. */
1566 MPASS(uid != VNOVAL || gid != VNOVAL);
1571 MPASS(uid != VNOVAL && gid != VNOVAL);
1573 /* Disallow this operation if the file system is mounted read-only. */
1574 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1577 /* Immutable or append-only files cannot be modified, either. */
1578 if (node->tn_flags & (IMMUTABLE | APPEND))
1582 * To modify the ownership of a file, must possess VADMIN for that
1585 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1589 * To change the owner of a file, or change the group of a file to a
1590 * group of which we are not a member, the caller must have
1593 if ((uid != node->tn_uid ||
1594 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1595 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1598 ogid = node->tn_gid;
1599 ouid = node->tn_uid;
1604 node->tn_status |= TMPFS_NODE_CHANGED;
1606 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1607 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1608 node->tn_mode &= ~(S_ISUID | S_ISGID);
1611 ASSERT_VOP_ELOCKED(vp, "chown2");
1617 * Change size of the given vnode.
1618 * Caller should execute tmpfs_update on vp after a successful execution.
1619 * The vnode must be locked on entry and remain locked on exit.
1622 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1626 struct tmpfs_node *node;
1628 ASSERT_VOP_ELOCKED(vp, "chsize");
1630 node = VP_TO_TMPFS_NODE(vp);
1632 /* Decide whether this is a valid operation based on the file type. */
1634 switch (vp->v_type) {
1639 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1648 /* Allow modifications of special files even if in the file
1649 * system is mounted read-only (we are not modifying the
1650 * files themselves, but the objects they represent). */
1654 /* Anything else is unsupported. */
1658 /* Immutable or append-only files cannot be modified, either. */
1659 if (node->tn_flags & (IMMUTABLE | APPEND))
1662 error = tmpfs_truncate(vp, size);
1663 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1664 * for us, as will update tn_status; no need to do that here. */
1666 ASSERT_VOP_ELOCKED(vp, "chsize2");
1672 * Change access and modification times 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_chtimes(struct vnode *vp, struct vattr *vap,
1678 struct ucred *cred, struct thread *l)
1681 struct tmpfs_node *node;
1683 ASSERT_VOP_ELOCKED(vp, "chtimes");
1685 node = VP_TO_TMPFS_NODE(vp);
1687 /* Disallow this operation if the file system is mounted read-only. */
1688 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1691 /* Immutable or append-only files cannot be modified, either. */
1692 if (node->tn_flags & (IMMUTABLE | APPEND))
1695 error = vn_utimes_perm(vp, vap, cred, l);
1699 if (vap->va_atime.tv_sec != VNOVAL && vap->va_atime.tv_nsec != VNOVAL)
1700 node->tn_status |= TMPFS_NODE_ACCESSED;
1702 if (vap->va_mtime.tv_sec != VNOVAL && vap->va_mtime.tv_nsec != VNOVAL)
1703 node->tn_status |= TMPFS_NODE_MODIFIED;
1705 if (vap->va_birthtime.tv_nsec != VNOVAL &&
1706 vap->va_birthtime.tv_nsec != VNOVAL)
1707 node->tn_status |= TMPFS_NODE_MODIFIED;
1709 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime);
1711 if (vap->va_birthtime.tv_nsec != VNOVAL &&
1712 vap->va_birthtime.tv_nsec != VNOVAL)
1713 node->tn_birthtime = vap->va_birthtime;
1714 ASSERT_VOP_ELOCKED(vp, "chtimes2");
1720 tmpfs_set_status(struct tmpfs_node *node, int status)
1723 if ((node->tn_status & status) == status)
1725 TMPFS_NODE_LOCK(node);
1726 node->tn_status |= status;
1727 TMPFS_NODE_UNLOCK(node);
1730 /* Sync timestamps */
1732 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1733 const struct timespec *mod)
1735 struct tmpfs_node *node;
1736 struct timespec now;
1738 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes");
1739 node = VP_TO_TMPFS_NODE(vp);
1741 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1742 TMPFS_NODE_CHANGED)) == 0)
1745 vfs_timestamp(&now);
1746 TMPFS_NODE_LOCK(node);
1747 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1750 node->tn_atime = *acc;
1752 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1755 node->tn_mtime = *mod;
1757 if (node->tn_status & TMPFS_NODE_CHANGED)
1758 node->tn_ctime = now;
1759 node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1760 TMPFS_NODE_CHANGED);
1761 TMPFS_NODE_UNLOCK(node);
1766 tmpfs_update(struct vnode *vp)
1769 tmpfs_itimes(vp, NULL, NULL);
1773 tmpfs_truncate(struct vnode *vp, off_t length)
1776 struct tmpfs_node *node;
1778 node = VP_TO_TMPFS_NODE(vp);
1785 if (node->tn_size == length) {
1790 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1793 error = tmpfs_reg_resize(vp, length, FALSE);
1795 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1804 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1806 if (a->td_hash > b->td_hash)
1808 else if (a->td_hash < b->td_hash)
1813 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);