1 /* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */
4 * SPDX-License-Identifier: BSD-2-Clause-NetBSD
6 * Copyright (c) 2005 The NetBSD Foundation, Inc.
9 * This code is derived from software contributed to The NetBSD Foundation
10 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
22 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
24 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
25 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
26 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGE.
36 * Efficient memory file system supporting functions.
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/dirent.h>
44 #include <sys/fnv_hash.h>
46 #include <sys/limits.h>
47 #include <sys/mount.h>
48 #include <sys/namei.h>
51 #include <sys/random.h>
52 #include <sys/refcount.h>
53 #include <sys/rwlock.h>
56 #include <sys/sysctl.h>
57 #include <sys/vnode.h>
58 #include <sys/vmmeter.h>
61 #include <vm/vm_param.h>
62 #include <vm/vm_object.h>
63 #include <vm/vm_page.h>
64 #include <vm/vm_pageout.h>
65 #include <vm/vm_pager.h>
66 #include <vm/vm_extern.h>
67 #include <vm/swap_pager.h>
69 #include <fs/tmpfs/tmpfs.h>
70 #include <fs/tmpfs/tmpfs_fifoops.h>
71 #include <fs/tmpfs/tmpfs_vnops.h>
73 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
76 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
78 static uma_zone_t tmpfs_dirent_pool;
79 static uma_zone_t tmpfs_node_pool;
83 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
85 struct tmpfs_node *node;
91 node->tn_accessed = false;
94 node->tn_vnode = NULL;
100 tmpfs_node_dtor(void *mem, int size, void *arg)
102 struct tmpfs_node *node;
105 node->tn_type = VNON;
109 tmpfs_node_init(void *mem, int size, int flags)
111 struct tmpfs_node *node;
115 mtx_init(&node->tn_interlock, "tmpfsni", NULL, MTX_DEF);
116 node->tn_gen = arc4random();
121 tmpfs_node_fini(void *mem, int size)
123 struct tmpfs_node *node;
126 mtx_destroy(&node->tn_interlock);
130 tmpfs_subr_init(void)
132 tmpfs_dirent_pool = uma_zcreate("TMPFS dirent",
133 sizeof(struct tmpfs_dirent), NULL, NULL, NULL, NULL,
135 tmpfs_node_pool = uma_zcreate("TMPFS node",
136 sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor,
137 tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0);
138 VFS_SMR_ZONE_SET(tmpfs_node_pool);
142 tmpfs_subr_uninit(void)
144 uma_zdestroy(tmpfs_node_pool);
145 uma_zdestroy(tmpfs_dirent_pool);
149 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
154 pages = *(long *)arg1;
155 bytes = pages * PAGE_SIZE;
157 error = sysctl_handle_long(oidp, &bytes, 0, req);
158 if (error || !req->newptr)
161 pages = bytes / PAGE_SIZE;
162 if (pages < TMPFS_PAGES_MINRESERVED)
165 *(long *)arg1 = pages;
169 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved,
170 CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RW, &tmpfs_pages_reserved, 0,
171 sysctl_mem_reserved, "L",
172 "Amount of available memory and swap below which tmpfs growth stops");
174 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a,
175 struct tmpfs_dirent *b);
176 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
179 tmpfs_mem_avail(void)
184 avail = swap_pager_avail + vm_free_count();
185 reserved = atomic_load_long(&tmpfs_pages_reserved);
186 if (__predict_false(avail < reserved))
188 return (avail - reserved);
192 tmpfs_pages_used(struct tmpfs_mount *tmp)
194 const size_t node_size = sizeof(struct tmpfs_node) +
195 sizeof(struct tmpfs_dirent);
198 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
200 return (meta_pages + tmp->tm_pages_used);
204 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
206 if (tmpfs_mem_avail() < req_pages)
209 if (tmp->tm_pages_max != ULONG_MAX &&
210 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
217 tmpfs_ref_node(struct tmpfs_node *node)
224 refcount_acquire(&node->tn_refcount);
226 KASSERT(old > 0, ("node %p zero refcount", node));
231 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
232 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
233 * using the credentials of the process 'p'.
235 * If the node type is set to 'VDIR', then the parent parameter must point
236 * to the parent directory of the node being created. It may only be NULL
237 * while allocating the root node.
239 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
240 * specifies the device the node represents.
242 * If the node type is set to 'VLNK', then the parameter target specifies
243 * the file name of the target file for the symbolic link that is being
246 * Note that new nodes are retrieved from the available list if it has
247 * items or, if it is empty, from the node pool as long as there is enough
248 * space to create them.
250 * Returns zero on success or an appropriate error code on failure.
253 tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type,
254 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
255 const char *target, dev_t rdev, struct tmpfs_node **node)
257 struct tmpfs_node *nnode;
260 /* If the root directory of the 'tmp' file system is not yet
261 * allocated, this must be the request to do it. */
262 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
264 MPASS(IFF(type == VLNK, target != NULL));
265 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
267 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
269 if (tmpfs_pages_check_avail(tmp, 1) == 0)
272 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
274 * When a new tmpfs node is created for fully
275 * constructed mount point, there must be a parent
276 * node, which vnode is locked exclusively. As
277 * consequence, if the unmount is executing in
278 * parallel, vflush() cannot reclaim the parent vnode.
279 * Due to this, the check for MNTK_UNMOUNT flag is not
280 * racy: if we did not see MNTK_UNMOUNT flag, then tmp
281 * cannot be destroyed until node construction is
282 * finished and the parent vnode unlocked.
284 * Tmpfs does not need to instantiate new nodes during
289 if ((mp->mnt_kern_flag & MNT_RDONLY) != 0)
292 nnode = uma_zalloc_smr(tmpfs_node_pool, M_WAITOK);
294 /* Generic initialization. */
295 nnode->tn_type = type;
296 vfs_timestamp(&nnode->tn_atime);
297 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
301 nnode->tn_mode = mode;
302 nnode->tn_id = alloc_unr64(&tmp->tm_ino_unr);
303 nnode->tn_refcount = 1;
305 /* Type-specific initialization. */
306 switch (nnode->tn_type) {
309 nnode->tn_rdev = rdev;
313 RB_INIT(&nnode->tn_dir.tn_dirhead);
314 LIST_INIT(&nnode->tn_dir.tn_dupindex);
315 MPASS(parent != nnode);
316 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
317 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
318 nnode->tn_dir.tn_readdir_lastn = 0;
319 nnode->tn_dir.tn_readdir_lastp = NULL;
321 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
322 nnode->tn_dir.tn_parent->tn_links++;
323 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
332 MPASS(strlen(target) < MAXPATHLEN);
333 nnode->tn_size = strlen(target);
334 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
336 memcpy(nnode->tn_link, target, nnode->tn_size);
340 obj = nnode->tn_reg.tn_aobj =
341 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
342 NULL /* XXXKIB - tmpfs needs swap reservation */);
343 VM_OBJECT_WLOCK(obj);
344 /* OBJ_TMPFS is set together with the setting of vp->v_object */
345 vm_object_set_flag(obj, OBJ_TMPFS_NODE);
346 VM_OBJECT_WUNLOCK(obj);
347 nnode->tn_reg.tn_tmp = tmp;
351 panic("tmpfs_alloc_node: type %p %d", nnode,
352 (int)nnode->tn_type);
356 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
357 nnode->tn_attached = true;
358 tmp->tm_nodes_inuse++;
367 * Destroys the node pointed to by node from the file system 'tmp'.
368 * If the node references a directory, no entries are allowed.
371 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
373 if (refcount_release_if_not_last(&node->tn_refcount))
377 TMPFS_NODE_LOCK(node);
378 if (!tmpfs_free_node_locked(tmp, node, false)) {
379 TMPFS_NODE_UNLOCK(node);
385 tmpfs_free_node_locked(struct tmpfs_mount *tmp, struct tmpfs_node *node,
391 TMPFS_MP_ASSERT_LOCKED(tmp);
392 TMPFS_NODE_ASSERT_LOCKED(node);
394 last = refcount_release(&node->tn_refcount);
395 if (node->tn_attached && (detach || last)) {
396 MPASS(tmp->tm_nodes_inuse > 0);
397 tmp->tm_nodes_inuse--;
398 LIST_REMOVE(node, tn_entries);
399 node->tn_attached = false;
405 MPASS(node->tn_vnode == NULL);
406 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
408 TMPFS_NODE_UNLOCK(node);
411 switch (node->tn_type) {
424 free(node->tn_link, M_TMPFSNAME);
428 uobj = node->tn_reg.tn_aobj;
431 atomic_subtract_long(&tmp->tm_pages_used, uobj->size);
432 KASSERT((uobj->flags & OBJ_TMPFS) == 0,
433 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj));
434 vm_object_deallocate(uobj);
439 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
442 uma_zfree_smr(tmpfs_node_pool, node);
448 static __inline uint32_t
449 tmpfs_dirent_hash(const char *name, u_int len)
453 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK;
454 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP
457 if (hash < TMPFS_DIRCOOKIE_MIN)
458 hash += TMPFS_DIRCOOKIE_MIN;
463 static __inline off_t
464 tmpfs_dirent_cookie(struct tmpfs_dirent *de)
467 return (TMPFS_DIRCOOKIE_EOF);
469 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN);
471 return (de->td_cookie);
474 static __inline boolean_t
475 tmpfs_dirent_dup(struct tmpfs_dirent *de)
477 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0);
480 static __inline boolean_t
481 tmpfs_dirent_duphead(struct tmpfs_dirent *de)
483 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0);
487 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen)
489 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen);
490 memcpy(de->ud.td_name, name, namelen);
491 de->td_namelen = namelen;
495 * Allocates a new directory entry for the node node with a name of name.
496 * The new directory entry is returned in *de.
498 * The link count of node is increased by one to reflect the new object
501 * Returns zero on success or an appropriate error code on failure.
504 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
505 const char *name, u_int len, struct tmpfs_dirent **de)
507 struct tmpfs_dirent *nde;
509 nde = uma_zalloc(tmpfs_dirent_pool, M_WAITOK);
512 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
513 tmpfs_dirent_init(nde, name, len);
525 * Frees a directory entry. It is the caller's responsibility to destroy
526 * the node referenced by it if needed.
528 * The link count of node is decreased by one to reflect the removal of an
529 * object that referenced it. This only happens if 'node_exists' is true;
530 * otherwise the function will not access the node referred to by the
531 * directory entry, as it may already have been released from the outside.
534 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
536 struct tmpfs_node *node;
540 MPASS(node->tn_links > 0);
543 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL)
544 free(de->ud.td_name, M_TMPFSNAME);
545 uma_zfree(tmpfs_dirent_pool, de);
549 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj)
552 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject");
553 if (vp->v_type != VREG || obj == NULL)
556 VM_OBJECT_WLOCK(obj);
558 vm_object_clear_flag(obj, OBJ_TMPFS);
559 obj->un_pager.swp.swp_tmpfs = NULL;
560 if (vp->v_writecount < 0)
561 vp->v_writecount = 0;
563 VM_OBJECT_WUNLOCK(obj);
567 * Need to clear v_object for insmntque failure.
570 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg)
573 tmpfs_destroy_vobject(vp, vp->v_object);
576 vp->v_op = &dead_vnodeops;
582 * Allocates a new vnode for the node node or returns a new reference to
583 * an existing one if the node had already a vnode referencing it. The
584 * resulting locked vnode is returned in *vpp.
586 * Returns zero on success or an appropriate error code on failure.
589 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
594 struct tmpfs_mount *tm;
599 tm = VFS_TO_TMPFS(mp);
600 TMPFS_NODE_LOCK(node);
601 tmpfs_ref_node(node);
603 TMPFS_NODE_ASSERT_LOCKED(node);
604 if ((vp = node->tn_vnode) != NULL) {
605 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
606 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) ||
608 (lkflag & LK_NOWAIT) != 0)) {
609 TMPFS_NODE_UNLOCK(node);
614 if (VN_IS_DOOMED(vp)) {
615 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM;
616 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) {
617 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node),
623 TMPFS_NODE_UNLOCK(node);
624 error = vget_finish(vp, lkflag, vs);
625 if (error == ENOENT) {
626 TMPFS_NODE_LOCK(node);
635 * Make sure the vnode is still there after
636 * getting the interlock to avoid racing a free.
638 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
640 TMPFS_NODE_LOCK(node);
647 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
648 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
649 TMPFS_NODE_UNLOCK(node);
656 * otherwise lock the vp list while we call getnewvnode
657 * since that can block.
659 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
660 node->tn_vpstate |= TMPFS_VNODE_WANT;
661 error = msleep((caddr_t) &node->tn_vpstate,
662 TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0);
667 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
669 TMPFS_NODE_UNLOCK(node);
671 /* Get a new vnode and associate it with our node. */
672 error = getnewvnode("tmpfs", mp, VFS_TO_TMPFS(mp)->tm_nonc ?
673 &tmpfs_vnodeop_nonc_entries : &tmpfs_vnodeop_entries, &vp);
678 /* lkflag is ignored, the lock is exclusive */
679 (void) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
682 vp->v_type = node->tn_type;
684 /* Type-specific initialization. */
685 switch (node->tn_type) {
695 vp->v_op = &tmpfs_fifoop_entries;
698 object = node->tn_reg.tn_aobj;
699 VM_OBJECT_WLOCK(object);
701 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
702 vp->v_object = object;
703 object->un_pager.swp.swp_tmpfs = vp;
704 vm_object_set_flag(object, OBJ_TMPFS);
705 vp->v_irflag |= VIRF_PGREAD;
707 VM_OBJECT_WUNLOCK(object);
710 MPASS(node->tn_dir.tn_parent != NULL);
711 if (node->tn_dir.tn_parent == node)
712 vp->v_vflag |= VV_ROOT;
716 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
718 if (vp->v_type != VFIFO)
721 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL);
726 TMPFS_NODE_LOCK(node);
728 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
729 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
732 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
733 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
734 TMPFS_NODE_UNLOCK(node);
735 wakeup((caddr_t) &node->tn_vpstate);
737 TMPFS_NODE_UNLOCK(node);
744 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
745 TMPFS_NODE_LOCK(node);
746 MPASS(*vpp == node->tn_vnode);
747 TMPFS_NODE_UNLOCK(node);
750 tmpfs_free_node(tm, node);
756 * Destroys the association between the vnode vp and the node it
760 tmpfs_free_vp(struct vnode *vp)
762 struct tmpfs_node *node;
764 node = VP_TO_TMPFS_NODE(vp);
766 TMPFS_NODE_ASSERT_LOCKED(node);
767 node->tn_vnode = NULL;
768 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
769 wakeup(&node->tn_vnode);
770 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
775 * Allocates a new file of type 'type' and adds it to the parent directory
776 * 'dvp'; this addition is done using the component name given in 'cnp'.
777 * The ownership of the new file is automatically assigned based on the
778 * credentials of the caller (through 'cnp'), the group is set based on
779 * the parent directory and the mode is determined from the 'vap' argument.
780 * If successful, *vpp holds a vnode to the newly created file and zero
781 * is returned. Otherwise *vpp is NULL and the function returns an
782 * appropriate error code.
785 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
786 struct componentname *cnp, const char *target)
789 struct tmpfs_dirent *de;
790 struct tmpfs_mount *tmp;
791 struct tmpfs_node *dnode;
792 struct tmpfs_node *node;
793 struct tmpfs_node *parent;
795 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file");
796 MPASS(cnp->cn_flags & HASBUF);
798 tmp = VFS_TO_TMPFS(dvp->v_mount);
799 dnode = VP_TO_TMPFS_DIR(dvp);
802 /* If the entry we are creating is a directory, we cannot overflow
803 * the number of links of its parent, because it will get a new
805 if (vap->va_type == VDIR) {
806 /* Ensure that we do not overflow the maximum number of links
807 * imposed by the system. */
808 MPASS(dnode->tn_links <= TMPFS_LINK_MAX);
809 if (dnode->tn_links == TMPFS_LINK_MAX) {
814 MPASS(parent != NULL);
818 /* Allocate a node that represents the new file. */
819 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type,
820 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent,
821 target, vap->va_rdev, &node);
825 /* Allocate a directory entry that points to the new file. */
826 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
829 tmpfs_free_node(tmp, node);
833 /* Allocate a vnode for the new file. */
834 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
836 tmpfs_free_dirent(tmp, de);
837 tmpfs_free_node(tmp, node);
841 /* Now that all required items are allocated, we can proceed to
842 * insert the new node into the directory, an operation that
844 if (cnp->cn_flags & ISWHITEOUT)
845 tmpfs_dir_whiteout_remove(dvp, cnp);
846 tmpfs_dir_attach(dvp, de);
850 struct tmpfs_dirent *
851 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
853 struct tmpfs_dirent *de;
855 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
857 if (de != NULL && tmpfs_dirent_duphead(de))
858 de = LIST_FIRST(&de->ud.td_duphead);
859 dc->tdc_current = de;
861 return (dc->tdc_current);
864 struct tmpfs_dirent *
865 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
867 struct tmpfs_dirent *de;
869 MPASS(dc->tdc_tree != NULL);
870 if (tmpfs_dirent_dup(dc->tdc_current)) {
871 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
872 if (dc->tdc_current != NULL)
873 return (dc->tdc_current);
875 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
876 &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
877 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
878 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
879 MPASS(dc->tdc_current != NULL);
882 return (dc->tdc_current);
885 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
886 static struct tmpfs_dirent *
887 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
889 struct tmpfs_dirent *de, dekey;
891 dekey.td_hash = hash;
892 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
896 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
897 static struct tmpfs_dirent *
898 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
899 struct tmpfs_dir_cursor *dc)
901 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
902 struct tmpfs_dirent *de, dekey;
904 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
906 if (cookie == node->tn_dir.tn_readdir_lastn &&
907 (de = node->tn_dir.tn_readdir_lastp) != NULL) {
908 /* Protect against possible race, tn_readdir_last[pn]
909 * may be updated with only shared vnode lock held. */
910 if (cookie == tmpfs_dirent_cookie(de))
914 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
915 LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
916 uh.td_dup.index_entries) {
917 MPASS(tmpfs_dirent_dup(de));
918 if (de->td_cookie == cookie)
920 /* dupindex list is sorted. */
921 if (de->td_cookie < cookie) {
930 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) {
933 dekey.td_hash = cookie;
934 /* Recover if direntry for cookie was removed */
935 de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
938 dc->tdc_current = de;
939 if (de != NULL && tmpfs_dirent_duphead(de)) {
940 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
941 MPASS(dc->tdc_current != NULL);
943 return (dc->tdc_current);
947 dc->tdc_current = de;
948 if (de != NULL && tmpfs_dirent_dup(de))
949 dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
951 return (dc->tdc_current);
955 * Looks for a directory entry in the directory represented by node.
956 * 'cnp' describes the name of the entry to look for. Note that the .
957 * and .. components are not allowed as they do not physically exist
958 * within directories.
960 * Returns a pointer to the entry when found, otherwise NULL.
962 struct tmpfs_dirent *
963 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
964 struct componentname *cnp)
966 struct tmpfs_dir_duphead *duphead;
967 struct tmpfs_dirent *de;
970 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
971 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
972 cnp->cn_nameptr[1] == '.')));
973 TMPFS_VALIDATE_DIR(node);
975 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
976 de = tmpfs_dir_xlookup_hash(node, hash);
977 if (de != NULL && tmpfs_dirent_duphead(de)) {
978 duphead = &de->ud.td_duphead;
979 LIST_FOREACH(de, duphead, uh.td_dup.entries) {
980 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
984 } else if (de != NULL) {
985 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
989 if (de != NULL && f != NULL && de->td_node != f)
996 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
997 * list, allocate new cookie value.
1000 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
1001 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
1003 struct tmpfs_dir_duphead *dupindex;
1004 struct tmpfs_dirent *de, *pde;
1006 dupindex = &dnode->tn_dir.tn_dupindex;
1007 de = LIST_FIRST(dupindex);
1008 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
1010 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
1012 nde->td_cookie = de->td_cookie + 1;
1013 MPASS(tmpfs_dirent_dup(nde));
1014 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
1015 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1020 * Cookie numbers are near exhaustion. Scan dupindex list for unused
1021 * numbers. dupindex list is sorted in descending order. Keep it so
1022 * after inserting nde.
1026 de = LIST_NEXT(de, uh.td_dup.index_entries);
1027 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
1029 * Last element of the index doesn't have minimal cookie
1032 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
1033 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
1034 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1036 } else if (de == NULL) {
1038 * We are so lucky have 2^30 hash duplicates in single
1039 * directory :) Return largest possible cookie value.
1040 * It should be fine except possible issues with
1041 * VOP_READDIR restart.
1043 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
1044 LIST_INSERT_HEAD(dupindex, nde,
1045 uh.td_dup.index_entries);
1046 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1049 if (de->td_cookie + 1 == pde->td_cookie ||
1050 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
1051 continue; /* No hole or invalid cookie. */
1052 nde->td_cookie = de->td_cookie + 1;
1053 MPASS(tmpfs_dirent_dup(nde));
1054 MPASS(pde->td_cookie > nde->td_cookie);
1055 MPASS(nde->td_cookie > de->td_cookie);
1056 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
1057 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1063 * Attaches the directory entry de to the directory represented by vp.
1064 * Note that this does not change the link count of the node pointed by
1065 * the directory entry, as this is done by tmpfs_alloc_dirent.
1068 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
1070 struct tmpfs_node *dnode;
1071 struct tmpfs_dirent *xde, *nde;
1073 ASSERT_VOP_ELOCKED(vp, __func__);
1074 MPASS(de->td_namelen > 0);
1075 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
1076 MPASS(de->td_cookie == de->td_hash);
1078 dnode = VP_TO_TMPFS_DIR(vp);
1079 dnode->tn_dir.tn_readdir_lastn = 0;
1080 dnode->tn_dir.tn_readdir_lastp = NULL;
1082 MPASS(!tmpfs_dirent_dup(de));
1083 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1084 if (xde != NULL && tmpfs_dirent_duphead(xde))
1085 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1086 else if (xde != NULL) {
1088 * Allocate new duphead. Swap xde with duphead to avoid
1089 * adding/removing elements with the same hash.
1091 MPASS(!tmpfs_dirent_dup(xde));
1092 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
1094 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
1095 memcpy(nde, xde, sizeof(*xde));
1096 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
1097 LIST_INIT(&xde->ud.td_duphead);
1098 xde->td_namelen = 0;
1099 xde->td_node = NULL;
1100 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
1101 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1103 dnode->tn_size += sizeof(struct tmpfs_dirent);
1104 dnode->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1105 dnode->tn_accessed = true;
1110 * Detaches the directory entry de from the directory represented by vp.
1111 * Note that this does not change the link count of the node pointed by
1112 * the directory entry, as this is done by tmpfs_free_dirent.
1115 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
1117 struct tmpfs_mount *tmp;
1118 struct tmpfs_dir *head;
1119 struct tmpfs_node *dnode;
1120 struct tmpfs_dirent *xde;
1122 ASSERT_VOP_ELOCKED(vp, __func__);
1124 dnode = VP_TO_TMPFS_DIR(vp);
1125 head = &dnode->tn_dir.tn_dirhead;
1126 dnode->tn_dir.tn_readdir_lastn = 0;
1127 dnode->tn_dir.tn_readdir_lastp = NULL;
1129 if (tmpfs_dirent_dup(de)) {
1130 /* Remove duphead if de was last entry. */
1131 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1132 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1133 MPASS(tmpfs_dirent_duphead(xde));
1136 LIST_REMOVE(de, uh.td_dup.entries);
1137 LIST_REMOVE(de, uh.td_dup.index_entries);
1139 if (LIST_EMPTY(&xde->ud.td_duphead)) {
1140 RB_REMOVE(tmpfs_dir, head, xde);
1141 tmp = VFS_TO_TMPFS(vp->v_mount);
1142 MPASS(xde->td_node == NULL);
1143 tmpfs_free_dirent(tmp, xde);
1146 de->td_cookie = de->td_hash;
1148 RB_REMOVE(tmpfs_dir, head, de);
1150 dnode->tn_size -= sizeof(struct tmpfs_dirent);
1151 dnode->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1152 dnode->tn_accessed = true;
1157 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1159 struct tmpfs_dirent *de, *dde, *nde;
1161 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1162 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1163 /* Node may already be destroyed. */
1165 if (tmpfs_dirent_duphead(de)) {
1166 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1167 LIST_REMOVE(dde, uh.td_dup.entries);
1168 dde->td_node = NULL;
1169 tmpfs_free_dirent(tmp, dde);
1172 tmpfs_free_dirent(tmp, de);
1177 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
1178 * directory and returns it in the uio space. The function returns 0
1179 * on success, -1 if there was not enough space in the uio structure to
1180 * hold the directory entry or an appropriate error code if another
1184 tmpfs_dir_getdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node,
1190 TMPFS_VALIDATE_DIR(node);
1191 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1193 dent.d_fileno = node->tn_id;
1194 dent.d_type = DT_DIR;
1196 dent.d_name[0] = '.';
1197 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1198 dirent_terminate(&dent);
1200 if (dent.d_reclen > uio->uio_resid)
1201 error = EJUSTRETURN;
1203 error = uiomove(&dent, dent.d_reclen, uio);
1205 tmpfs_set_accessed(tm, node);
1211 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
1212 * directory and returns it in the uio space. The function returns 0
1213 * on success, -1 if there was not enough space in the uio structure to
1214 * hold the directory entry or an appropriate error code if another
1218 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node,
1221 struct tmpfs_node *parent;
1225 TMPFS_VALIDATE_DIR(node);
1226 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1229 * Return ENOENT if the current node is already removed.
1231 TMPFS_ASSERT_LOCKED(node);
1232 parent = node->tn_dir.tn_parent;
1236 TMPFS_NODE_LOCK(parent);
1237 dent.d_fileno = parent->tn_id;
1238 TMPFS_NODE_UNLOCK(parent);
1240 dent.d_type = DT_DIR;
1242 dent.d_name[0] = '.';
1243 dent.d_name[1] = '.';
1244 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1245 dirent_terminate(&dent);
1247 if (dent.d_reclen > uio->uio_resid)
1248 error = EJUSTRETURN;
1250 error = uiomove(&dent, dent.d_reclen, uio);
1252 tmpfs_set_accessed(tm, node);
1258 * Helper function for tmpfs_readdir. Returns as much directory entries
1259 * as can fit in the uio space. The read starts at uio->uio_offset.
1260 * The function returns 0 on success, -1 if there was not enough space
1261 * in the uio structure to hold the directory entry or an appropriate
1262 * error code if another error happens.
1265 tmpfs_dir_getdents(struct tmpfs_mount *tm, struct tmpfs_node *node,
1266 struct uio *uio, int maxcookies, u_long *cookies, int *ncookies)
1268 struct tmpfs_dir_cursor dc;
1269 struct tmpfs_dirent *de;
1273 TMPFS_VALIDATE_DIR(node);
1278 * Lookup the node from the current offset. The starting offset of
1279 * 0 will lookup both '.' and '..', and then the first real entry,
1280 * or EOF if there are none. Then find all entries for the dir that
1281 * fit into the buffer. Once no more entries are found (de == NULL),
1282 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1285 switch (uio->uio_offset) {
1286 case TMPFS_DIRCOOKIE_DOT:
1287 error = tmpfs_dir_getdotdent(tm, node, uio);
1290 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
1291 if (cookies != NULL)
1292 cookies[(*ncookies)++] = off = uio->uio_offset;
1294 case TMPFS_DIRCOOKIE_DOTDOT:
1295 error = tmpfs_dir_getdotdotdent(tm, node, uio);
1298 de = tmpfs_dir_first(node, &dc);
1299 uio->uio_offset = tmpfs_dirent_cookie(de);
1300 if (cookies != NULL)
1301 cookies[(*ncookies)++] = off = uio->uio_offset;
1306 case TMPFS_DIRCOOKIE_EOF:
1309 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1312 if (cookies != NULL)
1313 off = tmpfs_dirent_cookie(de);
1316 /* Read as much entries as possible; i.e., until we reach the end of
1317 * the directory or we exhaust uio space. */
1321 /* Create a dirent structure representing the current
1322 * tmpfs_node and fill it. */
1323 if (de->td_node == NULL) {
1327 d.d_fileno = de->td_node->tn_id;
1328 switch (de->td_node->tn_type) {
1358 panic("tmpfs_dir_getdents: type %p %d",
1359 de->td_node, (int)de->td_node->tn_type);
1362 d.d_namlen = de->td_namelen;
1363 MPASS(de->td_namelen < sizeof(d.d_name));
1364 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1365 d.d_reclen = GENERIC_DIRSIZ(&d);
1366 dirent_terminate(&d);
1368 /* Stop reading if the directory entry we are treating is
1369 * bigger than the amount of data that can be returned. */
1370 if (d.d_reclen > uio->uio_resid) {
1371 error = EJUSTRETURN;
1375 /* Copy the new dirent structure into the output buffer and
1376 * advance pointers. */
1377 error = uiomove(&d, d.d_reclen, uio);
1379 de = tmpfs_dir_next(node, &dc);
1380 if (cookies != NULL) {
1381 off = tmpfs_dirent_cookie(de);
1382 MPASS(*ncookies < maxcookies);
1383 cookies[(*ncookies)++] = off;
1386 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
1388 /* Skip setting off when using cookies as it is already done above. */
1389 if (cookies == NULL)
1390 off = tmpfs_dirent_cookie(de);
1392 /* Update the offset and cache. */
1393 uio->uio_offset = off;
1394 node->tn_dir.tn_readdir_lastn = off;
1395 node->tn_dir.tn_readdir_lastp = de;
1397 tmpfs_set_accessed(tm, node);
1402 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1404 struct tmpfs_dirent *de;
1407 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1408 cnp->cn_nameptr, cnp->cn_namelen, &de);
1411 tmpfs_dir_attach(dvp, de);
1416 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1418 struct tmpfs_dirent *de;
1420 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1421 MPASS(de != NULL && de->td_node == NULL);
1422 tmpfs_dir_detach(dvp, de);
1423 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1427 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1428 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
1429 * 'newsize' must be positive.
1431 * Returns zero on success or an appropriate error code on failure.
1434 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1436 struct tmpfs_mount *tmp;
1437 struct tmpfs_node *node;
1440 vm_pindex_t idx, newpages, oldpages;
1444 MPASS(vp->v_type == VREG);
1445 MPASS(newsize >= 0);
1447 node = VP_TO_TMPFS_NODE(vp);
1448 uobj = node->tn_reg.tn_aobj;
1449 tmp = VFS_TO_TMPFS(vp->v_mount);
1452 * Convert the old and new sizes to the number of pages needed to
1453 * store them. It may happen that we do not need to do anything
1454 * because the last allocated page can accommodate the change on
1457 oldsize = node->tn_size;
1458 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1459 MPASS(oldpages == uobj->size);
1460 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1462 if (__predict_true(newpages == oldpages && newsize >= oldsize)) {
1463 node->tn_size = newsize;
1467 if (newpages > oldpages &&
1468 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
1471 VM_OBJECT_WLOCK(uobj);
1472 if (newsize < oldsize) {
1474 * Zero the truncated part of the last page.
1476 base = newsize & PAGE_MASK;
1478 idx = OFF_TO_IDX(newsize);
1480 m = vm_page_grab(uobj, idx, VM_ALLOC_NOCREAT);
1482 MPASS(vm_page_all_valid(m));
1483 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1484 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL |
1488 vm_object_pip_add(uobj, 1);
1489 VM_OBJECT_WUNLOCK(uobj);
1490 rv = vm_pager_get_pages(uobj, &m, 1, NULL,
1492 VM_OBJECT_WLOCK(uobj);
1493 vm_object_pip_wakeup(uobj);
1494 if (rv == VM_PAGER_OK) {
1496 * Since the page was not resident,
1497 * and therefore not recently
1498 * accessed, immediately enqueue it
1499 * for asynchronous laundering. The
1500 * current operation is not regarded
1509 VM_OBJECT_WUNLOCK(uobj);
1515 pmap_zero_page_area(m, base, PAGE_SIZE - base);
1516 vm_page_set_dirty(m);
1522 * Release any swap space and free any whole pages.
1524 if (newpages < oldpages)
1525 vm_object_page_remove(uobj, newpages, 0, 0);
1527 uobj->size = newpages;
1528 VM_OBJECT_WUNLOCK(uobj);
1530 atomic_add_long(&tmp->tm_pages_used, newpages - oldpages);
1532 node->tn_size = newsize;
1537 tmpfs_check_mtime(struct vnode *vp)
1539 struct tmpfs_node *node;
1540 struct vm_object *obj;
1542 ASSERT_VOP_ELOCKED(vp, "check_mtime");
1543 if (vp->v_type != VREG)
1546 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
1547 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
1549 if (obj->generation != obj->cleangeneration) {
1550 VM_OBJECT_WLOCK(obj);
1551 if (obj->generation != obj->cleangeneration) {
1552 obj->cleangeneration = obj->generation;
1553 node = VP_TO_TMPFS_NODE(vp);
1554 node->tn_status |= TMPFS_NODE_MODIFIED |
1557 VM_OBJECT_WUNLOCK(obj);
1562 * Change flags of the given vnode.
1563 * Caller should execute tmpfs_update on vp after a successful execution.
1564 * The vnode must be locked on entry and remain locked on exit.
1567 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
1571 struct tmpfs_node *node;
1573 ASSERT_VOP_ELOCKED(vp, "chflags");
1575 node = VP_TO_TMPFS_NODE(vp);
1577 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK |
1578 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP |
1579 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
1580 UF_SPARSE | UF_SYSTEM)) != 0)
1581 return (EOPNOTSUPP);
1583 /* Disallow this operation if the file system is mounted read-only. */
1584 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1588 * Callers may only modify the file flags on objects they
1589 * have VADMIN rights for.
1591 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1594 * Unprivileged processes are not permitted to unset system
1595 * flags, or modify flags if any system flags are set.
1597 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) {
1598 if (node->tn_flags &
1599 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1600 error = securelevel_gt(cred, 0);
1605 if (node->tn_flags &
1606 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1607 ((flags ^ node->tn_flags) & SF_SETTABLE))
1610 node->tn_flags = flags;
1611 node->tn_status |= TMPFS_NODE_CHANGED;
1613 ASSERT_VOP_ELOCKED(vp, "chflags2");
1619 * Change access mode on the given vnode.
1620 * Caller should execute tmpfs_update on vp after a successful execution.
1621 * The vnode must be locked on entry and remain locked on exit.
1624 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1627 struct tmpfs_node *node;
1630 ASSERT_VOP_ELOCKED(vp, "chmod");
1631 ASSERT_VOP_IN_SEQC(vp);
1633 node = VP_TO_TMPFS_NODE(vp);
1635 /* Disallow this operation if the file system is mounted read-only. */
1636 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1639 /* Immutable or append-only files cannot be modified, either. */
1640 if (node->tn_flags & (IMMUTABLE | APPEND))
1644 * To modify the permissions on a file, must possess VADMIN
1647 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1651 * Privileged processes may set the sticky bit on non-directories,
1652 * as well as set the setgid bit on a file with a group that the
1653 * process is not a member of.
1655 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1656 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE))
1659 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1660 error = priv_check_cred(cred, PRIV_VFS_SETGID);
1665 newmode = node->tn_mode & ~ALLPERMS;
1666 newmode |= mode & ALLPERMS;
1667 atomic_store_short(&node->tn_mode, newmode);
1669 node->tn_status |= TMPFS_NODE_CHANGED;
1671 ASSERT_VOP_ELOCKED(vp, "chmod2");
1677 * Change ownership of the given vnode. At least one of uid or gid must
1678 * be different than VNOVAL. If one is set to that value, the attribute
1680 * Caller should execute tmpfs_update on vp after a successful execution.
1681 * The vnode must be locked on entry and remain locked on exit.
1684 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1688 struct tmpfs_node *node;
1693 ASSERT_VOP_ELOCKED(vp, "chown");
1694 ASSERT_VOP_IN_SEQC(vp);
1696 node = VP_TO_TMPFS_NODE(vp);
1698 /* Assign default values if they are unknown. */
1699 MPASS(uid != VNOVAL || gid != VNOVAL);
1704 MPASS(uid != VNOVAL && gid != VNOVAL);
1706 /* Disallow this operation if the file system is mounted read-only. */
1707 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1710 /* Immutable or append-only files cannot be modified, either. */
1711 if (node->tn_flags & (IMMUTABLE | APPEND))
1715 * To modify the ownership of a file, must possess VADMIN for that
1718 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1722 * To change the owner of a file, or change the group of a file to a
1723 * group of which we are not a member, the caller must have
1726 if ((uid != node->tn_uid ||
1727 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1728 (error = priv_check_cred(cred, PRIV_VFS_CHOWN)))
1731 ogid = node->tn_gid;
1732 ouid = node->tn_uid;
1737 node->tn_status |= TMPFS_NODE_CHANGED;
1739 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1740 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) {
1741 newmode = node->tn_mode & ~(S_ISUID | S_ISGID);
1742 atomic_store_short(&node->tn_mode, newmode);
1746 ASSERT_VOP_ELOCKED(vp, "chown2");
1752 * Change size of the given vnode.
1753 * Caller should execute tmpfs_update on vp after a successful execution.
1754 * The vnode must be locked on entry and remain locked on exit.
1757 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1761 struct tmpfs_node *node;
1763 ASSERT_VOP_ELOCKED(vp, "chsize");
1765 node = VP_TO_TMPFS_NODE(vp);
1767 /* Decide whether this is a valid operation based on the file type. */
1769 switch (vp->v_type) {
1774 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1784 * Allow modifications of special files even if in the file
1785 * system is mounted read-only (we are not modifying the
1786 * files themselves, but the objects they represent).
1791 /* Anything else is unsupported. */
1792 return (EOPNOTSUPP);
1795 /* Immutable or append-only files cannot be modified, either. */
1796 if (node->tn_flags & (IMMUTABLE | APPEND))
1799 error = tmpfs_truncate(vp, size);
1801 * tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1802 * for us, as will update tn_status; no need to do that here.
1805 ASSERT_VOP_ELOCKED(vp, "chsize2");
1811 * Change access and modification times of the given vnode.
1812 * Caller should execute tmpfs_update on vp after a successful execution.
1813 * The vnode must be locked on entry and remain locked on exit.
1816 tmpfs_chtimes(struct vnode *vp, struct vattr *vap,
1817 struct ucred *cred, struct thread *l)
1820 struct tmpfs_node *node;
1822 ASSERT_VOP_ELOCKED(vp, "chtimes");
1824 node = VP_TO_TMPFS_NODE(vp);
1826 /* Disallow this operation if the file system is mounted read-only. */
1827 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1830 /* Immutable or append-only files cannot be modified, either. */
1831 if (node->tn_flags & (IMMUTABLE | APPEND))
1834 error = vn_utimes_perm(vp, vap, cred, l);
1838 if (vap->va_atime.tv_sec != VNOVAL)
1839 node->tn_accessed = true;
1841 if (vap->va_mtime.tv_sec != VNOVAL)
1842 node->tn_status |= TMPFS_NODE_MODIFIED;
1844 if (vap->va_birthtime.tv_sec != VNOVAL)
1845 node->tn_status |= TMPFS_NODE_MODIFIED;
1847 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime);
1849 if (vap->va_birthtime.tv_sec != VNOVAL)
1850 node->tn_birthtime = vap->va_birthtime;
1851 ASSERT_VOP_ELOCKED(vp, "chtimes2");
1857 tmpfs_set_status(struct tmpfs_mount *tm, struct tmpfs_node *node, int status)
1860 if ((node->tn_status & status) == status || tm->tm_ronly)
1862 TMPFS_NODE_LOCK(node);
1863 node->tn_status |= status;
1864 TMPFS_NODE_UNLOCK(node);
1868 tmpfs_set_accessed(struct tmpfs_mount *tm, struct tmpfs_node *node)
1870 if (node->tn_accessed || tm->tm_ronly)
1872 atomic_store_8(&node->tn_accessed, true);
1875 /* Sync timestamps */
1877 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1878 const struct timespec *mod)
1880 struct tmpfs_node *node;
1881 struct timespec now;
1883 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes");
1884 node = VP_TO_TMPFS_NODE(vp);
1886 if (!node->tn_accessed &&
1887 (node->tn_status & (TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED)) == 0)
1890 vfs_timestamp(&now);
1891 TMPFS_NODE_LOCK(node);
1892 if (node->tn_accessed) {
1895 node->tn_atime = *acc;
1897 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1900 node->tn_mtime = *mod;
1902 if (node->tn_status & TMPFS_NODE_CHANGED)
1903 node->tn_ctime = now;
1904 node->tn_status &= ~(TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1905 node->tn_accessed = false;
1906 TMPFS_NODE_UNLOCK(node);
1908 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
1909 random_harvest_queue(node, sizeof(*node), RANDOM_FS_ATIME);
1913 tmpfs_truncate(struct vnode *vp, off_t length)
1916 struct tmpfs_node *node;
1918 node = VP_TO_TMPFS_NODE(vp);
1925 if (node->tn_size == length) {
1930 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1933 error = tmpfs_reg_resize(vp, length, FALSE);
1935 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1944 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1946 if (a->td_hash > b->td_hash)
1948 else if (a->td_hash < b->td_hash)
1953 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);