1 /* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */
4 * SPDX-License-Identifier: BSD-2-Clause-NetBSD
6 * Copyright (c) 2005 The NetBSD Foundation, Inc.
9 * This code is derived from software contributed to The NetBSD Foundation
10 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
22 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
24 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
25 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
26 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGE.
36 * Efficient memory file system supporting functions.
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/dirent.h>
44 #include <sys/fnv_hash.h>
46 #include <sys/limits.h>
47 #include <sys/mount.h>
48 #include <sys/namei.h>
51 #include <sys/random.h>
52 #include <sys/rwlock.h>
54 #include <sys/sysctl.h>
55 #include <sys/vnode.h>
56 #include <sys/vmmeter.h>
59 #include <vm/vm_param.h>
60 #include <vm/vm_object.h>
61 #include <vm/vm_page.h>
62 #include <vm/vm_pageout.h>
63 #include <vm/vm_pager.h>
64 #include <vm/vm_extern.h>
65 #include <vm/swap_pager.h>
67 #include <fs/tmpfs/tmpfs.h>
68 #include <fs/tmpfs/tmpfs_fifoops.h>
69 #include <fs/tmpfs/tmpfs_vnops.h>
71 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system");
73 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
75 static uma_zone_t tmpfs_dirent_pool;
76 static uma_zone_t tmpfs_node_pool;
79 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
81 struct tmpfs_node *node;
89 node->tn_vnode = NULL;
95 tmpfs_node_dtor(void *mem, int size, void *arg)
97 struct tmpfs_node *node;
100 node->tn_type = VNON;
104 tmpfs_node_init(void *mem, int size, int flags)
106 struct tmpfs_node *node;
110 mtx_init(&node->tn_interlock, "tmpfsni", NULL, MTX_DEF);
111 node->tn_gen = arc4random();
116 tmpfs_node_fini(void *mem, int size)
118 struct tmpfs_node *node;
121 mtx_destroy(&node->tn_interlock);
125 tmpfs_subr_init(void)
127 tmpfs_dirent_pool = uma_zcreate("TMPFS dirent",
128 sizeof(struct tmpfs_dirent), NULL, NULL, NULL, NULL,
130 tmpfs_node_pool = uma_zcreate("TMPFS node",
131 sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor,
132 tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0);
136 tmpfs_subr_uninit(void)
138 uma_zdestroy(tmpfs_node_pool);
139 uma_zdestroy(tmpfs_dirent_pool);
143 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
148 pages = *(long *)arg1;
149 bytes = pages * PAGE_SIZE;
151 error = sysctl_handle_long(oidp, &bytes, 0, req);
152 if (error || !req->newptr)
155 pages = bytes / PAGE_SIZE;
156 if (pages < TMPFS_PAGES_MINRESERVED)
159 *(long *)arg1 = pages;
163 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved,
164 CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RW, &tmpfs_pages_reserved, 0,
165 sysctl_mem_reserved, "L",
166 "Amount of available memory and swap below which tmpfs growth stops");
168 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a,
169 struct tmpfs_dirent *b);
170 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
173 tmpfs_mem_avail(void)
177 avail = swap_pager_avail + vm_free_count() - tmpfs_pages_reserved;
178 if (__predict_false(avail < 0))
184 tmpfs_pages_used(struct tmpfs_mount *tmp)
186 const size_t node_size = sizeof(struct tmpfs_node) +
187 sizeof(struct tmpfs_dirent);
190 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
192 return (meta_pages + tmp->tm_pages_used);
196 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
198 if (tmpfs_mem_avail() < req_pages)
201 if (tmp->tm_pages_max != ULONG_MAX &&
202 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
209 tmpfs_ref_node(struct tmpfs_node *node)
212 TMPFS_NODE_LOCK(node);
213 tmpfs_ref_node_locked(node);
214 TMPFS_NODE_UNLOCK(node);
218 tmpfs_ref_node_locked(struct tmpfs_node *node)
221 TMPFS_NODE_ASSERT_LOCKED(node);
222 KASSERT(node->tn_refcount > 0, ("node %p zero refcount", node));
223 KASSERT(node->tn_refcount < UINT_MAX, ("node %p refcount %u", node,
229 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
230 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
231 * using the credentials of the process 'p'.
233 * If the node type is set to 'VDIR', then the parent parameter must point
234 * to the parent directory of the node being created. It may only be NULL
235 * while allocating the root node.
237 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
238 * specifies the device the node represents.
240 * If the node type is set to 'VLNK', then the parameter target specifies
241 * the file name of the target file for the symbolic link that is being
244 * Note that new nodes are retrieved from the available list if it has
245 * items or, if it is empty, from the node pool as long as there is enough
246 * space to create them.
248 * Returns zero on success or an appropriate error code on failure.
251 tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type,
252 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
253 const char *target, dev_t rdev, struct tmpfs_node **node)
255 struct tmpfs_node *nnode;
258 /* If the root directory of the 'tmp' file system is not yet
259 * allocated, this must be the request to do it. */
260 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
262 MPASS(IFF(type == VLNK, target != NULL));
263 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
265 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
267 if (tmpfs_pages_check_avail(tmp, 1) == 0)
270 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
272 * When a new tmpfs node is created for fully
273 * constructed mount point, there must be a parent
274 * node, which vnode is locked exclusively. As
275 * consequence, if the unmount is executing in
276 * parallel, vflush() cannot reclaim the parent vnode.
277 * Due to this, the check for MNTK_UNMOUNT flag is not
278 * racy: if we did not see MNTK_UNMOUNT flag, then tmp
279 * cannot be destroyed until node construction is
280 * finished and the parent vnode unlocked.
282 * Tmpfs does not need to instantiate new nodes during
287 if ((mp->mnt_kern_flag & MNT_RDONLY) != 0)
290 nnode = uma_zalloc_arg(tmpfs_node_pool, tmp, M_WAITOK);
292 /* Generic initialization. */
293 nnode->tn_type = type;
294 vfs_timestamp(&nnode->tn_atime);
295 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
299 nnode->tn_mode = mode;
300 nnode->tn_id = alloc_unr64(&tmp->tm_ino_unr);
301 nnode->tn_refcount = 1;
303 /* Type-specific initialization. */
304 switch (nnode->tn_type) {
307 nnode->tn_rdev = rdev;
311 RB_INIT(&nnode->tn_dir.tn_dirhead);
312 LIST_INIT(&nnode->tn_dir.tn_dupindex);
313 MPASS(parent != nnode);
314 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
315 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
316 nnode->tn_dir.tn_readdir_lastn = 0;
317 nnode->tn_dir.tn_readdir_lastp = NULL;
319 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
320 nnode->tn_dir.tn_parent->tn_links++;
321 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
330 MPASS(strlen(target) < MAXPATHLEN);
331 nnode->tn_size = strlen(target);
332 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
334 memcpy(nnode->tn_link, target, nnode->tn_size);
338 obj = nnode->tn_reg.tn_aobj =
339 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
340 NULL /* XXXKIB - tmpfs needs swap reservation */);
341 VM_OBJECT_WLOCK(obj);
342 /* OBJ_TMPFS is set together with the setting of vp->v_object */
343 vm_object_set_flag(obj, OBJ_TMPFS_NODE);
344 VM_OBJECT_WUNLOCK(obj);
348 panic("tmpfs_alloc_node: type %p %d", nnode,
349 (int)nnode->tn_type);
353 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
354 nnode->tn_attached = true;
355 tmp->tm_nodes_inuse++;
364 * Destroys the node pointed to by node from the file system 'tmp'.
365 * If the node references a directory, no entries are allowed.
368 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
372 TMPFS_NODE_LOCK(node);
373 if (!tmpfs_free_node_locked(tmp, node, false)) {
374 TMPFS_NODE_UNLOCK(node);
380 tmpfs_free_node_locked(struct tmpfs_mount *tmp, struct tmpfs_node *node,
385 TMPFS_MP_ASSERT_LOCKED(tmp);
386 TMPFS_NODE_ASSERT_LOCKED(node);
387 KASSERT(node->tn_refcount > 0, ("node %p refcount zero", node));
390 if (node->tn_attached && (detach || node->tn_refcount == 0)) {
391 MPASS(tmp->tm_nodes_inuse > 0);
392 tmp->tm_nodes_inuse--;
393 LIST_REMOVE(node, tn_entries);
394 node->tn_attached = false;
396 if (node->tn_refcount > 0)
400 MPASS(node->tn_vnode == NULL);
401 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
403 TMPFS_NODE_UNLOCK(node);
406 switch (node->tn_type) {
419 free(node->tn_link, M_TMPFSNAME);
423 uobj = node->tn_reg.tn_aobj;
426 atomic_subtract_long(&tmp->tm_pages_used, uobj->size);
427 KASSERT((uobj->flags & OBJ_TMPFS) == 0,
428 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj));
429 vm_object_deallocate(uobj);
434 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
437 uma_zfree(tmpfs_node_pool, node);
443 static __inline uint32_t
444 tmpfs_dirent_hash(const char *name, u_int len)
448 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK;
449 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP
452 if (hash < TMPFS_DIRCOOKIE_MIN)
453 hash += TMPFS_DIRCOOKIE_MIN;
458 static __inline off_t
459 tmpfs_dirent_cookie(struct tmpfs_dirent *de)
462 return (TMPFS_DIRCOOKIE_EOF);
464 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN);
466 return (de->td_cookie);
469 static __inline boolean_t
470 tmpfs_dirent_dup(struct tmpfs_dirent *de)
472 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0);
475 static __inline boolean_t
476 tmpfs_dirent_duphead(struct tmpfs_dirent *de)
478 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0);
482 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen)
484 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen);
485 memcpy(de->ud.td_name, name, namelen);
486 de->td_namelen = namelen;
490 * Allocates a new directory entry for the node node with a name of name.
491 * The new directory entry is returned in *de.
493 * The link count of node is increased by one to reflect the new object
496 * Returns zero on success or an appropriate error code on failure.
499 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
500 const char *name, u_int len, struct tmpfs_dirent **de)
502 struct tmpfs_dirent *nde;
504 nde = uma_zalloc(tmpfs_dirent_pool, M_WAITOK);
507 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
508 tmpfs_dirent_init(nde, name, len);
520 * Frees a directory entry. It is the caller's responsibility to destroy
521 * the node referenced by it if needed.
523 * The link count of node is decreased by one to reflect the removal of an
524 * object that referenced it. This only happens if 'node_exists' is true;
525 * otherwise the function will not access the node referred to by the
526 * directory entry, as it may already have been released from the outside.
529 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
531 struct tmpfs_node *node;
535 MPASS(node->tn_links > 0);
538 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL)
539 free(de->ud.td_name, M_TMPFSNAME);
540 uma_zfree(tmpfs_dirent_pool, de);
544 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj)
547 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject");
548 if (vp->v_type != VREG || obj == NULL)
551 VM_OBJECT_WLOCK(obj);
553 vm_object_clear_flag(obj, OBJ_TMPFS);
554 obj->un_pager.swp.swp_tmpfs = NULL;
555 if (vp->v_writecount < 0)
556 vp->v_writecount = 0;
558 VM_OBJECT_WUNLOCK(obj);
562 * Need to clear v_object for insmntque failure.
565 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg)
568 tmpfs_destroy_vobject(vp, vp->v_object);
571 vp->v_op = &dead_vnodeops;
577 * Allocates a new vnode for the node node or returns a new reference to
578 * an existing one if the node had already a vnode referencing it. The
579 * resulting locked vnode is returned in *vpp.
581 * Returns zero on success or an appropriate error code on failure.
584 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
588 struct tmpfs_mount *tm;
593 tm = VFS_TO_TMPFS(mp);
594 TMPFS_NODE_LOCK(node);
595 tmpfs_ref_node_locked(node);
597 TMPFS_NODE_ASSERT_LOCKED(node);
598 if ((vp = node->tn_vnode) != NULL) {
599 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
601 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) ||
603 (lkflag & LK_NOWAIT) != 0)) {
605 TMPFS_NODE_UNLOCK(node);
610 if (VN_IS_DOOMED(vp)) {
612 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM;
613 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) {
614 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node),
619 TMPFS_NODE_UNLOCK(node);
620 error = vget(vp, lkflag | LK_INTERLOCK, curthread);
621 if (error == ENOENT) {
622 TMPFS_NODE_LOCK(node);
631 * Make sure the vnode is still there after
632 * getting the interlock to avoid racing a free.
634 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
636 TMPFS_NODE_LOCK(node);
643 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
644 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
645 TMPFS_NODE_UNLOCK(node);
652 * otherwise lock the vp list while we call getnewvnode
653 * since that can block.
655 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
656 node->tn_vpstate |= TMPFS_VNODE_WANT;
657 error = msleep((caddr_t) &node->tn_vpstate,
658 TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0);
663 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
665 TMPFS_NODE_UNLOCK(node);
667 /* Get a new vnode and associate it with our node. */
668 error = getnewvnode("tmpfs", mp, VFS_TO_TMPFS(mp)->tm_nonc ?
669 &tmpfs_vnodeop_nonc_entries : &tmpfs_vnodeop_entries, &vp);
674 /* lkflag is ignored, the lock is exclusive */
675 (void) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
678 vp->v_type = node->tn_type;
680 /* Type-specific initialization. */
681 switch (node->tn_type) {
691 vp->v_op = &tmpfs_fifoop_entries;
694 object = node->tn_reg.tn_aobj;
695 VM_OBJECT_WLOCK(object);
697 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
698 vp->v_object = object;
699 object->un_pager.swp.swp_tmpfs = vp;
700 vm_object_set_flag(object, OBJ_TMPFS);
702 VM_OBJECT_WUNLOCK(object);
705 MPASS(node->tn_dir.tn_parent != NULL);
706 if (node->tn_dir.tn_parent == node)
707 vp->v_vflag |= VV_ROOT;
711 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
713 if (vp->v_type != VFIFO)
716 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL);
721 TMPFS_NODE_LOCK(node);
723 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
724 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
727 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
728 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
729 TMPFS_NODE_UNLOCK(node);
730 wakeup((caddr_t) &node->tn_vpstate);
732 TMPFS_NODE_UNLOCK(node);
739 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
740 TMPFS_NODE_LOCK(node);
741 MPASS(*vpp == node->tn_vnode);
742 TMPFS_NODE_UNLOCK(node);
745 tmpfs_free_node(tm, node);
751 * Destroys the association between the vnode vp and the node it
755 tmpfs_free_vp(struct vnode *vp)
757 struct tmpfs_node *node;
759 node = VP_TO_TMPFS_NODE(vp);
761 TMPFS_NODE_ASSERT_LOCKED(node);
762 node->tn_vnode = NULL;
763 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
764 wakeup(&node->tn_vnode);
765 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
770 * Allocates a new file of type 'type' and adds it to the parent directory
771 * 'dvp'; this addition is done using the component name given in 'cnp'.
772 * The ownership of the new file is automatically assigned based on the
773 * credentials of the caller (through 'cnp'), the group is set based on
774 * the parent directory and the mode is determined from the 'vap' argument.
775 * If successful, *vpp holds a vnode to the newly created file and zero
776 * is returned. Otherwise *vpp is NULL and the function returns an
777 * appropriate error code.
780 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
781 struct componentname *cnp, const char *target)
784 struct tmpfs_dirent *de;
785 struct tmpfs_mount *tmp;
786 struct tmpfs_node *dnode;
787 struct tmpfs_node *node;
788 struct tmpfs_node *parent;
790 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file");
791 MPASS(cnp->cn_flags & HASBUF);
793 tmp = VFS_TO_TMPFS(dvp->v_mount);
794 dnode = VP_TO_TMPFS_DIR(dvp);
797 /* If the entry we are creating is a directory, we cannot overflow
798 * the number of links of its parent, because it will get a new
800 if (vap->va_type == VDIR) {
801 /* Ensure that we do not overflow the maximum number of links
802 * imposed by the system. */
803 MPASS(dnode->tn_links <= TMPFS_LINK_MAX);
804 if (dnode->tn_links == TMPFS_LINK_MAX) {
809 MPASS(parent != NULL);
813 /* Allocate a node that represents the new file. */
814 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type,
815 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent,
816 target, vap->va_rdev, &node);
820 /* Allocate a directory entry that points to the new file. */
821 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
824 tmpfs_free_node(tmp, node);
828 /* Allocate a vnode for the new file. */
829 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
831 tmpfs_free_dirent(tmp, de);
832 tmpfs_free_node(tmp, node);
836 /* Now that all required items are allocated, we can proceed to
837 * insert the new node into the directory, an operation that
839 if (cnp->cn_flags & ISWHITEOUT)
840 tmpfs_dir_whiteout_remove(dvp, cnp);
841 tmpfs_dir_attach(dvp, de);
845 struct tmpfs_dirent *
846 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
848 struct tmpfs_dirent *de;
850 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
852 if (de != NULL && tmpfs_dirent_duphead(de))
853 de = LIST_FIRST(&de->ud.td_duphead);
854 dc->tdc_current = de;
856 return (dc->tdc_current);
859 struct tmpfs_dirent *
860 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
862 struct tmpfs_dirent *de;
864 MPASS(dc->tdc_tree != NULL);
865 if (tmpfs_dirent_dup(dc->tdc_current)) {
866 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
867 if (dc->tdc_current != NULL)
868 return (dc->tdc_current);
870 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
871 &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
872 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
873 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
874 MPASS(dc->tdc_current != NULL);
877 return (dc->tdc_current);
880 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
881 static struct tmpfs_dirent *
882 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
884 struct tmpfs_dirent *de, dekey;
886 dekey.td_hash = hash;
887 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
891 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
892 static struct tmpfs_dirent *
893 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
894 struct tmpfs_dir_cursor *dc)
896 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
897 struct tmpfs_dirent *de, dekey;
899 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
901 if (cookie == node->tn_dir.tn_readdir_lastn &&
902 (de = node->tn_dir.tn_readdir_lastp) != NULL) {
903 /* Protect against possible race, tn_readdir_last[pn]
904 * may be updated with only shared vnode lock held. */
905 if (cookie == tmpfs_dirent_cookie(de))
909 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
910 LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
911 uh.td_dup.index_entries) {
912 MPASS(tmpfs_dirent_dup(de));
913 if (de->td_cookie == cookie)
915 /* dupindex list is sorted. */
916 if (de->td_cookie < cookie) {
925 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) {
928 dekey.td_hash = cookie;
929 /* Recover if direntry for cookie was removed */
930 de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
933 dc->tdc_current = de;
934 if (de != NULL && tmpfs_dirent_duphead(de)) {
935 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
936 MPASS(dc->tdc_current != NULL);
938 return (dc->tdc_current);
942 dc->tdc_current = de;
943 if (de != NULL && tmpfs_dirent_dup(de))
944 dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
946 return (dc->tdc_current);
950 * Looks for a directory entry in the directory represented by node.
951 * 'cnp' describes the name of the entry to look for. Note that the .
952 * and .. components are not allowed as they do not physically exist
953 * within directories.
955 * Returns a pointer to the entry when found, otherwise NULL.
957 struct tmpfs_dirent *
958 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
959 struct componentname *cnp)
961 struct tmpfs_dir_duphead *duphead;
962 struct tmpfs_dirent *de;
965 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
966 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
967 cnp->cn_nameptr[1] == '.')));
968 TMPFS_VALIDATE_DIR(node);
970 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
971 de = tmpfs_dir_xlookup_hash(node, hash);
972 if (de != NULL && tmpfs_dirent_duphead(de)) {
973 duphead = &de->ud.td_duphead;
974 LIST_FOREACH(de, duphead, uh.td_dup.entries) {
975 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
979 } else if (de != NULL) {
980 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
984 if (de != NULL && f != NULL && de->td_node != f)
991 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
992 * list, allocate new cookie value.
995 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
996 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
998 struct tmpfs_dir_duphead *dupindex;
999 struct tmpfs_dirent *de, *pde;
1001 dupindex = &dnode->tn_dir.tn_dupindex;
1002 de = LIST_FIRST(dupindex);
1003 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
1005 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
1007 nde->td_cookie = de->td_cookie + 1;
1008 MPASS(tmpfs_dirent_dup(nde));
1009 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
1010 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1015 * Cookie numbers are near exhaustion. Scan dupindex list for unused
1016 * numbers. dupindex list is sorted in descending order. Keep it so
1017 * after inserting nde.
1021 de = LIST_NEXT(de, uh.td_dup.index_entries);
1022 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
1024 * Last element of the index doesn't have minimal cookie
1027 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
1028 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
1029 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1031 } else if (de == NULL) {
1033 * We are so lucky have 2^30 hash duplicates in single
1034 * directory :) Return largest possible cookie value.
1035 * It should be fine except possible issues with
1036 * VOP_READDIR restart.
1038 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
1039 LIST_INSERT_HEAD(dupindex, nde,
1040 uh.td_dup.index_entries);
1041 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1044 if (de->td_cookie + 1 == pde->td_cookie ||
1045 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
1046 continue; /* No hole or invalid cookie. */
1047 nde->td_cookie = de->td_cookie + 1;
1048 MPASS(tmpfs_dirent_dup(nde));
1049 MPASS(pde->td_cookie > nde->td_cookie);
1050 MPASS(nde->td_cookie > de->td_cookie);
1051 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
1052 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1058 * Attaches the directory entry de to the directory represented by vp.
1059 * Note that this does not change the link count of the node pointed by
1060 * the directory entry, as this is done by tmpfs_alloc_dirent.
1063 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
1065 struct tmpfs_node *dnode;
1066 struct tmpfs_dirent *xde, *nde;
1068 ASSERT_VOP_ELOCKED(vp, __func__);
1069 MPASS(de->td_namelen > 0);
1070 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
1071 MPASS(de->td_cookie == de->td_hash);
1073 dnode = VP_TO_TMPFS_DIR(vp);
1074 dnode->tn_dir.tn_readdir_lastn = 0;
1075 dnode->tn_dir.tn_readdir_lastp = NULL;
1077 MPASS(!tmpfs_dirent_dup(de));
1078 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1079 if (xde != NULL && tmpfs_dirent_duphead(xde))
1080 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1081 else if (xde != NULL) {
1083 * Allocate new duphead. Swap xde with duphead to avoid
1084 * adding/removing elements with the same hash.
1086 MPASS(!tmpfs_dirent_dup(xde));
1087 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
1089 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
1090 memcpy(nde, xde, sizeof(*xde));
1091 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
1092 LIST_INIT(&xde->ud.td_duphead);
1093 xde->td_namelen = 0;
1094 xde->td_node = NULL;
1095 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
1096 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1098 dnode->tn_size += sizeof(struct tmpfs_dirent);
1099 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1100 TMPFS_NODE_MODIFIED;
1105 * Detaches the directory entry de from the directory represented by vp.
1106 * Note that this does not change the link count of the node pointed by
1107 * the directory entry, as this is done by tmpfs_free_dirent.
1110 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
1112 struct tmpfs_mount *tmp;
1113 struct tmpfs_dir *head;
1114 struct tmpfs_node *dnode;
1115 struct tmpfs_dirent *xde;
1117 ASSERT_VOP_ELOCKED(vp, __func__);
1119 dnode = VP_TO_TMPFS_DIR(vp);
1120 head = &dnode->tn_dir.tn_dirhead;
1121 dnode->tn_dir.tn_readdir_lastn = 0;
1122 dnode->tn_dir.tn_readdir_lastp = NULL;
1124 if (tmpfs_dirent_dup(de)) {
1125 /* Remove duphead if de was last entry. */
1126 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1127 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1128 MPASS(tmpfs_dirent_duphead(xde));
1131 LIST_REMOVE(de, uh.td_dup.entries);
1132 LIST_REMOVE(de, uh.td_dup.index_entries);
1134 if (LIST_EMPTY(&xde->ud.td_duphead)) {
1135 RB_REMOVE(tmpfs_dir, head, xde);
1136 tmp = VFS_TO_TMPFS(vp->v_mount);
1137 MPASS(xde->td_node == NULL);
1138 tmpfs_free_dirent(tmp, xde);
1141 de->td_cookie = de->td_hash;
1143 RB_REMOVE(tmpfs_dir, head, de);
1145 dnode->tn_size -= sizeof(struct tmpfs_dirent);
1146 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1147 TMPFS_NODE_MODIFIED;
1152 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1154 struct tmpfs_dirent *de, *dde, *nde;
1156 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1157 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1158 /* Node may already be destroyed. */
1160 if (tmpfs_dirent_duphead(de)) {
1161 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1162 LIST_REMOVE(dde, uh.td_dup.entries);
1163 dde->td_node = NULL;
1164 tmpfs_free_dirent(tmp, dde);
1167 tmpfs_free_dirent(tmp, de);
1172 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
1173 * directory and returns it in the uio space. The function returns 0
1174 * on success, -1 if there was not enough space in the uio structure to
1175 * hold the directory entry or an appropriate error code if another
1179 tmpfs_dir_getdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node,
1185 TMPFS_VALIDATE_DIR(node);
1186 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1188 dent.d_fileno = node->tn_id;
1189 dent.d_type = DT_DIR;
1191 dent.d_name[0] = '.';
1192 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1193 dirent_terminate(&dent);
1195 if (dent.d_reclen > uio->uio_resid)
1196 error = EJUSTRETURN;
1198 error = uiomove(&dent, dent.d_reclen, uio);
1200 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1206 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
1207 * directory and returns it in the uio space. The function returns 0
1208 * on success, -1 if there was not enough space in the uio structure to
1209 * hold the directory entry or an appropriate error code if another
1213 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node,
1216 struct tmpfs_node *parent;
1220 TMPFS_VALIDATE_DIR(node);
1221 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1224 * Return ENOENT if the current node is already removed.
1226 TMPFS_ASSERT_LOCKED(node);
1227 parent = node->tn_dir.tn_parent;
1231 TMPFS_NODE_LOCK(parent);
1232 dent.d_fileno = parent->tn_id;
1233 TMPFS_NODE_UNLOCK(parent);
1235 dent.d_type = DT_DIR;
1237 dent.d_name[0] = '.';
1238 dent.d_name[1] = '.';
1239 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1240 dirent_terminate(&dent);
1242 if (dent.d_reclen > uio->uio_resid)
1243 error = EJUSTRETURN;
1245 error = uiomove(&dent, dent.d_reclen, uio);
1247 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1253 * Helper function for tmpfs_readdir. Returns as much directory entries
1254 * as can fit in the uio space. The read starts at uio->uio_offset.
1255 * The function returns 0 on success, -1 if there was not enough space
1256 * in the uio structure to hold the directory entry or an appropriate
1257 * error code if another error happens.
1260 tmpfs_dir_getdents(struct tmpfs_mount *tm, struct tmpfs_node *node,
1261 struct uio *uio, int maxcookies, u_long *cookies, int *ncookies)
1263 struct tmpfs_dir_cursor dc;
1264 struct tmpfs_dirent *de;
1268 TMPFS_VALIDATE_DIR(node);
1273 * Lookup the node from the current offset. The starting offset of
1274 * 0 will lookup both '.' and '..', and then the first real entry,
1275 * or EOF if there are none. Then find all entries for the dir that
1276 * fit into the buffer. Once no more entries are found (de == NULL),
1277 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1280 switch (uio->uio_offset) {
1281 case TMPFS_DIRCOOKIE_DOT:
1282 error = tmpfs_dir_getdotdent(tm, node, uio);
1285 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
1286 if (cookies != NULL)
1287 cookies[(*ncookies)++] = off = uio->uio_offset;
1289 case TMPFS_DIRCOOKIE_DOTDOT:
1290 error = tmpfs_dir_getdotdotdent(tm, node, uio);
1293 de = tmpfs_dir_first(node, &dc);
1294 uio->uio_offset = tmpfs_dirent_cookie(de);
1295 if (cookies != NULL)
1296 cookies[(*ncookies)++] = off = uio->uio_offset;
1301 case TMPFS_DIRCOOKIE_EOF:
1304 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1307 if (cookies != NULL)
1308 off = tmpfs_dirent_cookie(de);
1311 /* Read as much entries as possible; i.e., until we reach the end of
1312 * the directory or we exhaust uio space. */
1316 /* Create a dirent structure representing the current
1317 * tmpfs_node and fill it. */
1318 if (de->td_node == NULL) {
1322 d.d_fileno = de->td_node->tn_id;
1323 switch (de->td_node->tn_type) {
1353 panic("tmpfs_dir_getdents: type %p %d",
1354 de->td_node, (int)de->td_node->tn_type);
1357 d.d_namlen = de->td_namelen;
1358 MPASS(de->td_namelen < sizeof(d.d_name));
1359 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1360 d.d_reclen = GENERIC_DIRSIZ(&d);
1361 dirent_terminate(&d);
1363 /* Stop reading if the directory entry we are treating is
1364 * bigger than the amount of data that can be returned. */
1365 if (d.d_reclen > uio->uio_resid) {
1366 error = EJUSTRETURN;
1370 /* Copy the new dirent structure into the output buffer and
1371 * advance pointers. */
1372 error = uiomove(&d, d.d_reclen, uio);
1374 de = tmpfs_dir_next(node, &dc);
1375 if (cookies != NULL) {
1376 off = tmpfs_dirent_cookie(de);
1377 MPASS(*ncookies < maxcookies);
1378 cookies[(*ncookies)++] = off;
1381 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
1383 /* Skip setting off when using cookies as it is already done above. */
1384 if (cookies == NULL)
1385 off = tmpfs_dirent_cookie(de);
1387 /* Update the offset and cache. */
1388 uio->uio_offset = off;
1389 node->tn_dir.tn_readdir_lastn = off;
1390 node->tn_dir.tn_readdir_lastp = de;
1392 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1397 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1399 struct tmpfs_dirent *de;
1402 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1403 cnp->cn_nameptr, cnp->cn_namelen, &de);
1406 tmpfs_dir_attach(dvp, de);
1411 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1413 struct tmpfs_dirent *de;
1415 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1416 MPASS(de != NULL && de->td_node == NULL);
1417 tmpfs_dir_detach(dvp, de);
1418 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1422 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1423 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
1424 * 'newsize' must be positive.
1426 * Returns zero on success or an appropriate error code on failure.
1429 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1431 struct tmpfs_mount *tmp;
1432 struct tmpfs_node *node;
1435 vm_pindex_t idx, newpages, oldpages;
1439 MPASS(vp->v_type == VREG);
1440 MPASS(newsize >= 0);
1442 node = VP_TO_TMPFS_NODE(vp);
1443 uobj = node->tn_reg.tn_aobj;
1444 tmp = VFS_TO_TMPFS(vp->v_mount);
1447 * Convert the old and new sizes to the number of pages needed to
1448 * store them. It may happen that we do not need to do anything
1449 * because the last allocated page can accommodate the change on
1452 oldsize = node->tn_size;
1453 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1454 MPASS(oldpages == uobj->size);
1455 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1457 if (__predict_true(newpages == oldpages && newsize >= oldsize)) {
1458 node->tn_size = newsize;
1462 if (newpages > oldpages &&
1463 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
1466 VM_OBJECT_WLOCK(uobj);
1467 if (newsize < oldsize) {
1469 * Zero the truncated part of the last page.
1471 base = newsize & PAGE_MASK;
1473 idx = OFF_TO_IDX(newsize);
1475 m = vm_page_grab(uobj, idx, VM_ALLOC_NOCREAT);
1477 MPASS(vm_page_all_valid(m));
1478 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1479 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL |
1483 vm_object_pip_add(uobj, 1);
1484 VM_OBJECT_WUNLOCK(uobj);
1485 rv = vm_pager_get_pages(uobj, &m, 1, NULL,
1487 VM_OBJECT_WLOCK(uobj);
1488 vm_object_pip_wakeup(uobj);
1489 if (rv == VM_PAGER_OK) {
1491 * Since the page was not resident,
1492 * and therefore not recently
1493 * accessed, immediately enqueue it
1494 * for asynchronous laundering. The
1495 * current operation is not regarded
1504 VM_OBJECT_WUNLOCK(uobj);
1510 pmap_zero_page_area(m, base, PAGE_SIZE - base);
1511 vm_page_set_dirty(m);
1517 * Release any swap space and free any whole pages.
1519 if (newpages < oldpages) {
1520 swap_pager_freespace(uobj, newpages, oldpages -
1522 vm_object_page_remove(uobj, newpages, 0, 0);
1525 uobj->size = newpages;
1526 VM_OBJECT_WUNLOCK(uobj);
1528 atomic_add_long(&tmp->tm_pages_used, newpages - oldpages);
1530 node->tn_size = newsize;
1535 tmpfs_check_mtime(struct vnode *vp)
1537 struct tmpfs_node *node;
1538 struct vm_object *obj;
1540 ASSERT_VOP_ELOCKED(vp, "check_mtime");
1541 if (vp->v_type != VREG)
1544 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
1545 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
1547 if (obj->generation != obj->cleangeneration) {
1548 VM_OBJECT_WLOCK(obj);
1549 if (obj->generation != obj->cleangeneration) {
1550 obj->cleangeneration = obj->generation;
1551 node = VP_TO_TMPFS_NODE(vp);
1552 node->tn_status |= TMPFS_NODE_MODIFIED |
1555 VM_OBJECT_WUNLOCK(obj);
1560 * Change flags of the given vnode.
1561 * Caller should execute tmpfs_update on vp after a successful execution.
1562 * The vnode must be locked on entry and remain locked on exit.
1565 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
1569 struct tmpfs_node *node;
1571 ASSERT_VOP_ELOCKED(vp, "chflags");
1573 node = VP_TO_TMPFS_NODE(vp);
1575 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK |
1576 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP |
1577 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
1578 UF_SPARSE | UF_SYSTEM)) != 0)
1579 return (EOPNOTSUPP);
1581 /* Disallow this operation if the file system is mounted read-only. */
1582 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1586 * Callers may only modify the file flags on objects they
1587 * have VADMIN rights for.
1589 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1592 * Unprivileged processes are not permitted to unset system
1593 * flags, or modify flags if any system flags are set.
1595 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) {
1596 if (node->tn_flags &
1597 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1598 error = securelevel_gt(cred, 0);
1603 if (node->tn_flags &
1604 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1605 ((flags ^ node->tn_flags) & SF_SETTABLE))
1608 node->tn_flags = flags;
1609 node->tn_status |= TMPFS_NODE_CHANGED;
1611 ASSERT_VOP_ELOCKED(vp, "chflags2");
1617 * Change access mode on 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_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1625 struct tmpfs_node *node;
1627 ASSERT_VOP_ELOCKED(vp, "chmod");
1629 node = VP_TO_TMPFS_NODE(vp);
1631 /* Disallow this operation if the file system is mounted read-only. */
1632 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1635 /* Immutable or append-only files cannot be modified, either. */
1636 if (node->tn_flags & (IMMUTABLE | APPEND))
1640 * To modify the permissions on a file, must possess VADMIN
1643 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1647 * Privileged processes may set the sticky bit on non-directories,
1648 * as well as set the setgid bit on a file with a group that the
1649 * process is not a member of.
1651 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1652 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE))
1655 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1656 error = priv_check_cred(cred, PRIV_VFS_SETGID);
1662 node->tn_mode &= ~ALLPERMS;
1663 node->tn_mode |= mode & ALLPERMS;
1665 node->tn_status |= TMPFS_NODE_CHANGED;
1667 ASSERT_VOP_ELOCKED(vp, "chmod2");
1673 * Change ownership of the given vnode. At least one of uid or gid must
1674 * be different than VNOVAL. If one is set to that value, the attribute
1676 * Caller should execute tmpfs_update on vp after a successful execution.
1677 * The vnode must be locked on entry and remain locked on exit.
1680 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1684 struct tmpfs_node *node;
1688 ASSERT_VOP_ELOCKED(vp, "chown");
1690 node = VP_TO_TMPFS_NODE(vp);
1692 /* Assign default values if they are unknown. */
1693 MPASS(uid != VNOVAL || gid != VNOVAL);
1698 MPASS(uid != VNOVAL && gid != VNOVAL);
1700 /* Disallow this operation if the file system is mounted read-only. */
1701 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1704 /* Immutable or append-only files cannot be modified, either. */
1705 if (node->tn_flags & (IMMUTABLE | APPEND))
1709 * To modify the ownership of a file, must possess VADMIN for that
1712 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1716 * To change the owner of a file, or change the group of a file to a
1717 * group of which we are not a member, the caller must have
1720 if ((uid != node->tn_uid ||
1721 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1722 (error = priv_check_cred(cred, PRIV_VFS_CHOWN)))
1725 ogid = node->tn_gid;
1726 ouid = node->tn_uid;
1731 node->tn_status |= TMPFS_NODE_CHANGED;
1733 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1734 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID))
1735 node->tn_mode &= ~(S_ISUID | S_ISGID);
1738 ASSERT_VOP_ELOCKED(vp, "chown2");
1744 * Change size of the given vnode.
1745 * Caller should execute tmpfs_update on vp after a successful execution.
1746 * The vnode must be locked on entry and remain locked on exit.
1749 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1753 struct tmpfs_node *node;
1755 ASSERT_VOP_ELOCKED(vp, "chsize");
1757 node = VP_TO_TMPFS_NODE(vp);
1759 /* Decide whether this is a valid operation based on the file type. */
1761 switch (vp->v_type) {
1766 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1775 /* Allow modifications of special files even if in the file
1776 * system is mounted read-only (we are not modifying the
1777 * files themselves, but the objects they represent). */
1781 /* Anything else is unsupported. */
1785 /* Immutable or append-only files cannot be modified, either. */
1786 if (node->tn_flags & (IMMUTABLE | APPEND))
1789 error = tmpfs_truncate(vp, size);
1790 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1791 * for us, as will update tn_status; no need to do that here. */
1793 ASSERT_VOP_ELOCKED(vp, "chsize2");
1799 * Change access and modification times of the given vnode.
1800 * Caller should execute tmpfs_update on vp after a successful execution.
1801 * The vnode must be locked on entry and remain locked on exit.
1804 tmpfs_chtimes(struct vnode *vp, struct vattr *vap,
1805 struct ucred *cred, struct thread *l)
1808 struct tmpfs_node *node;
1810 ASSERT_VOP_ELOCKED(vp, "chtimes");
1812 node = VP_TO_TMPFS_NODE(vp);
1814 /* Disallow this operation if the file system is mounted read-only. */
1815 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1818 /* Immutable or append-only files cannot be modified, either. */
1819 if (node->tn_flags & (IMMUTABLE | APPEND))
1822 error = vn_utimes_perm(vp, vap, cred, l);
1826 if (vap->va_atime.tv_sec != VNOVAL)
1827 node->tn_status |= TMPFS_NODE_ACCESSED;
1829 if (vap->va_mtime.tv_sec != VNOVAL)
1830 node->tn_status |= TMPFS_NODE_MODIFIED;
1832 if (vap->va_birthtime.tv_sec != VNOVAL)
1833 node->tn_status |= TMPFS_NODE_MODIFIED;
1835 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime);
1837 if (vap->va_birthtime.tv_sec != VNOVAL)
1838 node->tn_birthtime = vap->va_birthtime;
1839 ASSERT_VOP_ELOCKED(vp, "chtimes2");
1845 tmpfs_set_status(struct tmpfs_mount *tm, struct tmpfs_node *node, int status)
1848 if ((node->tn_status & status) == status || tm->tm_ronly)
1850 TMPFS_NODE_LOCK(node);
1851 node->tn_status |= status;
1852 TMPFS_NODE_UNLOCK(node);
1855 /* Sync timestamps */
1857 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1858 const struct timespec *mod)
1860 struct tmpfs_node *node;
1861 struct timespec now;
1863 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes");
1864 node = VP_TO_TMPFS_NODE(vp);
1866 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1867 TMPFS_NODE_CHANGED)) == 0)
1870 vfs_timestamp(&now);
1871 TMPFS_NODE_LOCK(node);
1872 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1875 node->tn_atime = *acc;
1877 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1880 node->tn_mtime = *mod;
1882 if (node->tn_status & TMPFS_NODE_CHANGED)
1883 node->tn_ctime = now;
1884 node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1885 TMPFS_NODE_CHANGED);
1886 TMPFS_NODE_UNLOCK(node);
1888 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
1889 random_harvest_queue(node, sizeof(*node), RANDOM_FS_ATIME);
1893 tmpfs_update(struct vnode *vp)
1896 tmpfs_itimes(vp, NULL, NULL);
1900 tmpfs_truncate(struct vnode *vp, off_t length)
1903 struct tmpfs_node *node;
1905 node = VP_TO_TMPFS_NODE(vp);
1912 if (node->tn_size == length) {
1917 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1920 error = tmpfs_reg_resize(vp, length, FALSE);
1922 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1931 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1933 if (a->td_hash > b->td_hash)
1935 else if (a->td_hash < b->td_hash)
1940 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);