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>
55 #include <sys/sysctl.h>
56 #include <sys/vnode.h>
57 #include <sys/vmmeter.h>
60 #include <vm/vm_param.h>
61 #include <vm/vm_object.h>
62 #include <vm/vm_page.h>
63 #include <vm/vm_pageout.h>
64 #include <vm/vm_pager.h>
65 #include <vm/vm_extern.h>
66 #include <vm/swap_pager.h>
68 #include <fs/tmpfs/tmpfs.h>
69 #include <fs/tmpfs/tmpfs_fifoops.h>
70 #include <fs/tmpfs/tmpfs_vnops.h>
72 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
75 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
77 static uma_zone_t tmpfs_dirent_pool;
78 static uma_zone_t tmpfs_node_pool;
82 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
84 struct tmpfs_node *node;
92 node->tn_vnode = NULL;
98 tmpfs_node_dtor(void *mem, int size, void *arg)
100 struct tmpfs_node *node;
103 node->tn_type = VNON;
107 tmpfs_node_init(void *mem, int size, int flags)
109 struct tmpfs_node *node;
113 mtx_init(&node->tn_interlock, "tmpfsni", NULL, MTX_DEF);
114 node->tn_gen = arc4random();
119 tmpfs_node_fini(void *mem, int size)
121 struct tmpfs_node *node;
124 mtx_destroy(&node->tn_interlock);
128 tmpfs_subr_init(void)
130 tmpfs_dirent_pool = uma_zcreate("TMPFS dirent",
131 sizeof(struct tmpfs_dirent), NULL, NULL, NULL, NULL,
133 tmpfs_node_pool = uma_zcreate("TMPFS node",
134 sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor,
135 tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0);
136 VFS_SMR_ZONE_SET(tmpfs_node_pool);
140 tmpfs_subr_uninit(void)
142 uma_zdestroy(tmpfs_node_pool);
143 uma_zdestroy(tmpfs_dirent_pool);
147 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
152 pages = *(long *)arg1;
153 bytes = pages * PAGE_SIZE;
155 error = sysctl_handle_long(oidp, &bytes, 0, req);
156 if (error || !req->newptr)
159 pages = bytes / PAGE_SIZE;
160 if (pages < TMPFS_PAGES_MINRESERVED)
163 *(long *)arg1 = pages;
167 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved,
168 CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RW, &tmpfs_pages_reserved, 0,
169 sysctl_mem_reserved, "L",
170 "Amount of available memory and swap below which tmpfs growth stops");
172 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a,
173 struct tmpfs_dirent *b);
174 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
177 tmpfs_mem_avail(void)
181 avail = swap_pager_avail + vm_free_count() - tmpfs_pages_reserved;
182 if (__predict_false(avail < 0))
188 tmpfs_pages_used(struct tmpfs_mount *tmp)
190 const size_t node_size = sizeof(struct tmpfs_node) +
191 sizeof(struct tmpfs_dirent);
194 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
196 return (meta_pages + tmp->tm_pages_used);
200 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
202 if (tmpfs_mem_avail() < req_pages)
205 if (tmp->tm_pages_max != ULONG_MAX &&
206 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
213 tmpfs_ref_node(struct tmpfs_node *node)
220 refcount_acquire(&node->tn_refcount);
222 KASSERT(old > 0, ("node %p zero refcount", node));
227 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
228 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
229 * using the credentials of the process 'p'.
231 * If the node type is set to 'VDIR', then the parent parameter must point
232 * to the parent directory of the node being created. It may only be NULL
233 * while allocating the root node.
235 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
236 * specifies the device the node represents.
238 * If the node type is set to 'VLNK', then the parameter target specifies
239 * the file name of the target file for the symbolic link that is being
242 * Note that new nodes are retrieved from the available list if it has
243 * items or, if it is empty, from the node pool as long as there is enough
244 * space to create them.
246 * Returns zero on success or an appropriate error code on failure.
249 tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type,
250 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
251 const char *target, dev_t rdev, struct tmpfs_node **node)
253 struct tmpfs_node *nnode;
256 /* If the root directory of the 'tmp' file system is not yet
257 * allocated, this must be the request to do it. */
258 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
260 MPASS(IFF(type == VLNK, target != NULL));
261 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
263 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
265 if (tmpfs_pages_check_avail(tmp, 1) == 0)
268 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
270 * When a new tmpfs node is created for fully
271 * constructed mount point, there must be a parent
272 * node, which vnode is locked exclusively. As
273 * consequence, if the unmount is executing in
274 * parallel, vflush() cannot reclaim the parent vnode.
275 * Due to this, the check for MNTK_UNMOUNT flag is not
276 * racy: if we did not see MNTK_UNMOUNT flag, then tmp
277 * cannot be destroyed until node construction is
278 * finished and the parent vnode unlocked.
280 * Tmpfs does not need to instantiate new nodes during
285 if ((mp->mnt_kern_flag & MNT_RDONLY) != 0)
288 nnode = uma_zalloc_smr(tmpfs_node_pool, M_WAITOK);
290 /* Generic initialization. */
291 nnode->tn_type = type;
292 vfs_timestamp(&nnode->tn_atime);
293 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
297 nnode->tn_mode = mode;
298 nnode->tn_id = alloc_unr64(&tmp->tm_ino_unr);
299 nnode->tn_refcount = 1;
301 /* Type-specific initialization. */
302 switch (nnode->tn_type) {
305 nnode->tn_rdev = rdev;
309 RB_INIT(&nnode->tn_dir.tn_dirhead);
310 LIST_INIT(&nnode->tn_dir.tn_dupindex);
311 MPASS(parent != nnode);
312 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
313 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
314 nnode->tn_dir.tn_readdir_lastn = 0;
315 nnode->tn_dir.tn_readdir_lastp = NULL;
317 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
318 nnode->tn_dir.tn_parent->tn_links++;
319 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
328 MPASS(strlen(target) < MAXPATHLEN);
329 nnode->tn_size = strlen(target);
330 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
332 memcpy(nnode->tn_link, target, nnode->tn_size);
336 obj = nnode->tn_reg.tn_aobj =
337 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
338 NULL /* XXXKIB - tmpfs needs swap reservation */);
339 VM_OBJECT_WLOCK(obj);
340 /* OBJ_TMPFS is set together with the setting of vp->v_object */
341 vm_object_set_flag(obj, OBJ_TMPFS_NODE);
342 VM_OBJECT_WUNLOCK(obj);
346 panic("tmpfs_alloc_node: type %p %d", nnode,
347 (int)nnode->tn_type);
351 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
352 nnode->tn_attached = true;
353 tmp->tm_nodes_inuse++;
362 * Destroys the node pointed to by node from the file system 'tmp'.
363 * If the node references a directory, no entries are allowed.
366 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
368 if (refcount_release_if_not_last(&node->tn_refcount))
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,
386 TMPFS_MP_ASSERT_LOCKED(tmp);
387 TMPFS_NODE_ASSERT_LOCKED(node);
389 last = refcount_release(&node->tn_refcount);
390 if (node->tn_attached && (detach || last)) {
391 MPASS(tmp->tm_nodes_inuse > 0);
392 tmp->tm_nodes_inuse--;
393 LIST_REMOVE(node, tn_entries);
394 node->tn_attached = false;
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_smr(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,
589 struct tmpfs_mount *tm;
594 tm = VFS_TO_TMPFS(mp);
595 TMPFS_NODE_LOCK(node);
596 tmpfs_ref_node(node);
598 TMPFS_NODE_ASSERT_LOCKED(node);
599 if ((vp = node->tn_vnode) != NULL) {
600 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)) {
604 TMPFS_NODE_UNLOCK(node);
609 if (VN_IS_DOOMED(vp)) {
610 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM;
611 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) {
612 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node),
618 TMPFS_NODE_UNLOCK(node);
619 error = vget_finish(vp, lkflag, vs);
620 if (error == ENOENT) {
621 TMPFS_NODE_LOCK(node);
630 * Make sure the vnode is still there after
631 * getting the interlock to avoid racing a free.
633 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
635 TMPFS_NODE_LOCK(node);
642 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
643 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
644 TMPFS_NODE_UNLOCK(node);
651 * otherwise lock the vp list while we call getnewvnode
652 * since that can block.
654 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
655 node->tn_vpstate |= TMPFS_VNODE_WANT;
656 error = msleep((caddr_t) &node->tn_vpstate,
657 TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0);
662 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
664 TMPFS_NODE_UNLOCK(node);
666 /* Get a new vnode and associate it with our node. */
667 error = getnewvnode("tmpfs", mp, VFS_TO_TMPFS(mp)->tm_nonc ?
668 &tmpfs_vnodeop_nonc_entries : &tmpfs_vnodeop_entries, &vp);
673 /* lkflag is ignored, the lock is exclusive */
674 (void) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
677 vp->v_type = node->tn_type;
679 /* Type-specific initialization. */
680 switch (node->tn_type) {
690 vp->v_op = &tmpfs_fifoop_entries;
693 object = node->tn_reg.tn_aobj;
694 VM_OBJECT_WLOCK(object);
696 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
697 vp->v_object = object;
698 object->un_pager.swp.swp_tmpfs = vp;
699 vm_object_set_flag(object, OBJ_TMPFS);
701 VM_OBJECT_WUNLOCK(object);
704 MPASS(node->tn_dir.tn_parent != NULL);
705 if (node->tn_dir.tn_parent == node)
706 vp->v_vflag |= VV_ROOT;
710 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
712 if (vp->v_type != VFIFO)
715 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL);
720 TMPFS_NODE_LOCK(node);
722 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
723 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
726 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
727 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
728 TMPFS_NODE_UNLOCK(node);
729 wakeup((caddr_t) &node->tn_vpstate);
731 TMPFS_NODE_UNLOCK(node);
738 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
739 TMPFS_NODE_LOCK(node);
740 MPASS(*vpp == node->tn_vnode);
741 TMPFS_NODE_UNLOCK(node);
744 tmpfs_free_node(tm, node);
750 * Destroys the association between the vnode vp and the node it
754 tmpfs_free_vp(struct vnode *vp)
756 struct tmpfs_node *node;
758 node = VP_TO_TMPFS_NODE(vp);
760 TMPFS_NODE_ASSERT_LOCKED(node);
761 node->tn_vnode = NULL;
762 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
763 wakeup(&node->tn_vnode);
764 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
769 * Allocates a new file of type 'type' and adds it to the parent directory
770 * 'dvp'; this addition is done using the component name given in 'cnp'.
771 * The ownership of the new file is automatically assigned based on the
772 * credentials of the caller (through 'cnp'), the group is set based on
773 * the parent directory and the mode is determined from the 'vap' argument.
774 * If successful, *vpp holds a vnode to the newly created file and zero
775 * is returned. Otherwise *vpp is NULL and the function returns an
776 * appropriate error code.
779 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
780 struct componentname *cnp, const char *target)
783 struct tmpfs_dirent *de;
784 struct tmpfs_mount *tmp;
785 struct tmpfs_node *dnode;
786 struct tmpfs_node *node;
787 struct tmpfs_node *parent;
789 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file");
790 MPASS(cnp->cn_flags & HASBUF);
792 tmp = VFS_TO_TMPFS(dvp->v_mount);
793 dnode = VP_TO_TMPFS_DIR(dvp);
796 /* If the entry we are creating is a directory, we cannot overflow
797 * the number of links of its parent, because it will get a new
799 if (vap->va_type == VDIR) {
800 /* Ensure that we do not overflow the maximum number of links
801 * imposed by the system. */
802 MPASS(dnode->tn_links <= TMPFS_LINK_MAX);
803 if (dnode->tn_links == TMPFS_LINK_MAX) {
808 MPASS(parent != NULL);
812 /* Allocate a node that represents the new file. */
813 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type,
814 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent,
815 target, vap->va_rdev, &node);
819 /* Allocate a directory entry that points to the new file. */
820 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
823 tmpfs_free_node(tmp, node);
827 /* Allocate a vnode for the new file. */
828 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
830 tmpfs_free_dirent(tmp, de);
831 tmpfs_free_node(tmp, node);
835 /* Now that all required items are allocated, we can proceed to
836 * insert the new node into the directory, an operation that
838 if (cnp->cn_flags & ISWHITEOUT)
839 tmpfs_dir_whiteout_remove(dvp, cnp);
840 tmpfs_dir_attach(dvp, de);
844 struct tmpfs_dirent *
845 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
847 struct tmpfs_dirent *de;
849 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
851 if (de != NULL && tmpfs_dirent_duphead(de))
852 de = LIST_FIRST(&de->ud.td_duphead);
853 dc->tdc_current = de;
855 return (dc->tdc_current);
858 struct tmpfs_dirent *
859 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
861 struct tmpfs_dirent *de;
863 MPASS(dc->tdc_tree != NULL);
864 if (tmpfs_dirent_dup(dc->tdc_current)) {
865 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
866 if (dc->tdc_current != NULL)
867 return (dc->tdc_current);
869 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
870 &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
871 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
872 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
873 MPASS(dc->tdc_current != NULL);
876 return (dc->tdc_current);
879 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
880 static struct tmpfs_dirent *
881 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
883 struct tmpfs_dirent *de, dekey;
885 dekey.td_hash = hash;
886 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
890 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
891 static struct tmpfs_dirent *
892 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
893 struct tmpfs_dir_cursor *dc)
895 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
896 struct tmpfs_dirent *de, dekey;
898 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
900 if (cookie == node->tn_dir.tn_readdir_lastn &&
901 (de = node->tn_dir.tn_readdir_lastp) != NULL) {
902 /* Protect against possible race, tn_readdir_last[pn]
903 * may be updated with only shared vnode lock held. */
904 if (cookie == tmpfs_dirent_cookie(de))
908 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
909 LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
910 uh.td_dup.index_entries) {
911 MPASS(tmpfs_dirent_dup(de));
912 if (de->td_cookie == cookie)
914 /* dupindex list is sorted. */
915 if (de->td_cookie < cookie) {
924 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) {
927 dekey.td_hash = cookie;
928 /* Recover if direntry for cookie was removed */
929 de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
932 dc->tdc_current = de;
933 if (de != NULL && tmpfs_dirent_duphead(de)) {
934 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
935 MPASS(dc->tdc_current != NULL);
937 return (dc->tdc_current);
941 dc->tdc_current = de;
942 if (de != NULL && tmpfs_dirent_dup(de))
943 dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
945 return (dc->tdc_current);
949 * Looks for a directory entry in the directory represented by node.
950 * 'cnp' describes the name of the entry to look for. Note that the .
951 * and .. components are not allowed as they do not physically exist
952 * within directories.
954 * Returns a pointer to the entry when found, otherwise NULL.
956 struct tmpfs_dirent *
957 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
958 struct componentname *cnp)
960 struct tmpfs_dir_duphead *duphead;
961 struct tmpfs_dirent *de;
964 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
965 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
966 cnp->cn_nameptr[1] == '.')));
967 TMPFS_VALIDATE_DIR(node);
969 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
970 de = tmpfs_dir_xlookup_hash(node, hash);
971 if (de != NULL && tmpfs_dirent_duphead(de)) {
972 duphead = &de->ud.td_duphead;
973 LIST_FOREACH(de, duphead, uh.td_dup.entries) {
974 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
978 } else if (de != NULL) {
979 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
983 if (de != NULL && f != NULL && de->td_node != f)
990 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
991 * list, allocate new cookie value.
994 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
995 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
997 struct tmpfs_dir_duphead *dupindex;
998 struct tmpfs_dirent *de, *pde;
1000 dupindex = &dnode->tn_dir.tn_dupindex;
1001 de = LIST_FIRST(dupindex);
1002 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
1004 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
1006 nde->td_cookie = de->td_cookie + 1;
1007 MPASS(tmpfs_dirent_dup(nde));
1008 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
1009 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1014 * Cookie numbers are near exhaustion. Scan dupindex list for unused
1015 * numbers. dupindex list is sorted in descending order. Keep it so
1016 * after inserting nde.
1020 de = LIST_NEXT(de, uh.td_dup.index_entries);
1021 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
1023 * Last element of the index doesn't have minimal cookie
1026 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
1027 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
1028 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1030 } else if (de == NULL) {
1032 * We are so lucky have 2^30 hash duplicates in single
1033 * directory :) Return largest possible cookie value.
1034 * It should be fine except possible issues with
1035 * VOP_READDIR restart.
1037 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
1038 LIST_INSERT_HEAD(dupindex, nde,
1039 uh.td_dup.index_entries);
1040 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1043 if (de->td_cookie + 1 == pde->td_cookie ||
1044 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
1045 continue; /* No hole or invalid cookie. */
1046 nde->td_cookie = de->td_cookie + 1;
1047 MPASS(tmpfs_dirent_dup(nde));
1048 MPASS(pde->td_cookie > nde->td_cookie);
1049 MPASS(nde->td_cookie > de->td_cookie);
1050 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
1051 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1057 * Attaches the directory entry de to the directory represented by vp.
1058 * Note that this does not change the link count of the node pointed by
1059 * the directory entry, as this is done by tmpfs_alloc_dirent.
1062 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
1064 struct tmpfs_node *dnode;
1065 struct tmpfs_dirent *xde, *nde;
1067 ASSERT_VOP_ELOCKED(vp, __func__);
1068 MPASS(de->td_namelen > 0);
1069 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
1070 MPASS(de->td_cookie == de->td_hash);
1072 dnode = VP_TO_TMPFS_DIR(vp);
1073 dnode->tn_dir.tn_readdir_lastn = 0;
1074 dnode->tn_dir.tn_readdir_lastp = NULL;
1076 MPASS(!tmpfs_dirent_dup(de));
1077 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1078 if (xde != NULL && tmpfs_dirent_duphead(xde))
1079 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1080 else if (xde != NULL) {
1082 * Allocate new duphead. Swap xde with duphead to avoid
1083 * adding/removing elements with the same hash.
1085 MPASS(!tmpfs_dirent_dup(xde));
1086 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
1088 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
1089 memcpy(nde, xde, sizeof(*xde));
1090 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
1091 LIST_INIT(&xde->ud.td_duphead);
1092 xde->td_namelen = 0;
1093 xde->td_node = NULL;
1094 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
1095 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1097 dnode->tn_size += sizeof(struct tmpfs_dirent);
1098 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1099 TMPFS_NODE_MODIFIED;
1104 * Detaches the directory entry de from the directory represented by vp.
1105 * Note that this does not change the link count of the node pointed by
1106 * the directory entry, as this is done by tmpfs_free_dirent.
1109 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
1111 struct tmpfs_mount *tmp;
1112 struct tmpfs_dir *head;
1113 struct tmpfs_node *dnode;
1114 struct tmpfs_dirent *xde;
1116 ASSERT_VOP_ELOCKED(vp, __func__);
1118 dnode = VP_TO_TMPFS_DIR(vp);
1119 head = &dnode->tn_dir.tn_dirhead;
1120 dnode->tn_dir.tn_readdir_lastn = 0;
1121 dnode->tn_dir.tn_readdir_lastp = NULL;
1123 if (tmpfs_dirent_dup(de)) {
1124 /* Remove duphead if de was last entry. */
1125 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1126 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1127 MPASS(tmpfs_dirent_duphead(xde));
1130 LIST_REMOVE(de, uh.td_dup.entries);
1131 LIST_REMOVE(de, uh.td_dup.index_entries);
1133 if (LIST_EMPTY(&xde->ud.td_duphead)) {
1134 RB_REMOVE(tmpfs_dir, head, xde);
1135 tmp = VFS_TO_TMPFS(vp->v_mount);
1136 MPASS(xde->td_node == NULL);
1137 tmpfs_free_dirent(tmp, xde);
1140 de->td_cookie = de->td_hash;
1142 RB_REMOVE(tmpfs_dir, head, de);
1144 dnode->tn_size -= sizeof(struct tmpfs_dirent);
1145 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1146 TMPFS_NODE_MODIFIED;
1151 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1153 struct tmpfs_dirent *de, *dde, *nde;
1155 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1156 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1157 /* Node may already be destroyed. */
1159 if (tmpfs_dirent_duphead(de)) {
1160 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1161 LIST_REMOVE(dde, uh.td_dup.entries);
1162 dde->td_node = NULL;
1163 tmpfs_free_dirent(tmp, dde);
1166 tmpfs_free_dirent(tmp, de);
1171 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
1172 * directory and returns it in the uio space. The function returns 0
1173 * on success, -1 if there was not enough space in the uio structure to
1174 * hold the directory entry or an appropriate error code if another
1178 tmpfs_dir_getdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node,
1184 TMPFS_VALIDATE_DIR(node);
1185 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1187 dent.d_fileno = node->tn_id;
1188 dent.d_type = DT_DIR;
1190 dent.d_name[0] = '.';
1191 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1192 dirent_terminate(&dent);
1194 if (dent.d_reclen > uio->uio_resid)
1195 error = EJUSTRETURN;
1197 error = uiomove(&dent, dent.d_reclen, uio);
1199 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1205 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
1206 * directory and returns it in the uio space. The function returns 0
1207 * on success, -1 if there was not enough space in the uio structure to
1208 * hold the directory entry or an appropriate error code if another
1212 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node,
1215 struct tmpfs_node *parent;
1219 TMPFS_VALIDATE_DIR(node);
1220 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1223 * Return ENOENT if the current node is already removed.
1225 TMPFS_ASSERT_LOCKED(node);
1226 parent = node->tn_dir.tn_parent;
1230 TMPFS_NODE_LOCK(parent);
1231 dent.d_fileno = parent->tn_id;
1232 TMPFS_NODE_UNLOCK(parent);
1234 dent.d_type = DT_DIR;
1236 dent.d_name[0] = '.';
1237 dent.d_name[1] = '.';
1238 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1239 dirent_terminate(&dent);
1241 if (dent.d_reclen > uio->uio_resid)
1242 error = EJUSTRETURN;
1244 error = uiomove(&dent, dent.d_reclen, uio);
1246 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1252 * Helper function for tmpfs_readdir. Returns as much directory entries
1253 * as can fit in the uio space. The read starts at uio->uio_offset.
1254 * The function returns 0 on success, -1 if there was not enough space
1255 * in the uio structure to hold the directory entry or an appropriate
1256 * error code if another error happens.
1259 tmpfs_dir_getdents(struct tmpfs_mount *tm, struct tmpfs_node *node,
1260 struct uio *uio, int maxcookies, u_long *cookies, int *ncookies)
1262 struct tmpfs_dir_cursor dc;
1263 struct tmpfs_dirent *de;
1267 TMPFS_VALIDATE_DIR(node);
1272 * Lookup the node from the current offset. The starting offset of
1273 * 0 will lookup both '.' and '..', and then the first real entry,
1274 * or EOF if there are none. Then find all entries for the dir that
1275 * fit into the buffer. Once no more entries are found (de == NULL),
1276 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1279 switch (uio->uio_offset) {
1280 case TMPFS_DIRCOOKIE_DOT:
1281 error = tmpfs_dir_getdotdent(tm, node, uio);
1284 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
1285 if (cookies != NULL)
1286 cookies[(*ncookies)++] = off = uio->uio_offset;
1288 case TMPFS_DIRCOOKIE_DOTDOT:
1289 error = tmpfs_dir_getdotdotdent(tm, node, uio);
1292 de = tmpfs_dir_first(node, &dc);
1293 uio->uio_offset = tmpfs_dirent_cookie(de);
1294 if (cookies != NULL)
1295 cookies[(*ncookies)++] = off = uio->uio_offset;
1300 case TMPFS_DIRCOOKIE_EOF:
1303 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1306 if (cookies != NULL)
1307 off = tmpfs_dirent_cookie(de);
1310 /* Read as much entries as possible; i.e., until we reach the end of
1311 * the directory or we exhaust uio space. */
1315 /* Create a dirent structure representing the current
1316 * tmpfs_node and fill it. */
1317 if (de->td_node == NULL) {
1321 d.d_fileno = de->td_node->tn_id;
1322 switch (de->td_node->tn_type) {
1352 panic("tmpfs_dir_getdents: type %p %d",
1353 de->td_node, (int)de->td_node->tn_type);
1356 d.d_namlen = de->td_namelen;
1357 MPASS(de->td_namelen < sizeof(d.d_name));
1358 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1359 d.d_reclen = GENERIC_DIRSIZ(&d);
1360 dirent_terminate(&d);
1362 /* Stop reading if the directory entry we are treating is
1363 * bigger than the amount of data that can be returned. */
1364 if (d.d_reclen > uio->uio_resid) {
1365 error = EJUSTRETURN;
1369 /* Copy the new dirent structure into the output buffer and
1370 * advance pointers. */
1371 error = uiomove(&d, d.d_reclen, uio);
1373 de = tmpfs_dir_next(node, &dc);
1374 if (cookies != NULL) {
1375 off = tmpfs_dirent_cookie(de);
1376 MPASS(*ncookies < maxcookies);
1377 cookies[(*ncookies)++] = off;
1380 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
1382 /* Skip setting off when using cookies as it is already done above. */
1383 if (cookies == NULL)
1384 off = tmpfs_dirent_cookie(de);
1386 /* Update the offset and cache. */
1387 uio->uio_offset = off;
1388 node->tn_dir.tn_readdir_lastn = off;
1389 node->tn_dir.tn_readdir_lastp = de;
1391 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1396 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1398 struct tmpfs_dirent *de;
1401 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1402 cnp->cn_nameptr, cnp->cn_namelen, &de);
1405 tmpfs_dir_attach(dvp, de);
1410 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1412 struct tmpfs_dirent *de;
1414 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1415 MPASS(de != NULL && de->td_node == NULL);
1416 tmpfs_dir_detach(dvp, de);
1417 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1421 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1422 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
1423 * 'newsize' must be positive.
1425 * Returns zero on success or an appropriate error code on failure.
1428 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1430 struct tmpfs_mount *tmp;
1431 struct tmpfs_node *node;
1434 vm_pindex_t idx, newpages, oldpages;
1438 MPASS(vp->v_type == VREG);
1439 MPASS(newsize >= 0);
1441 node = VP_TO_TMPFS_NODE(vp);
1442 uobj = node->tn_reg.tn_aobj;
1443 tmp = VFS_TO_TMPFS(vp->v_mount);
1446 * Convert the old and new sizes to the number of pages needed to
1447 * store them. It may happen that we do not need to do anything
1448 * because the last allocated page can accommodate the change on
1451 oldsize = node->tn_size;
1452 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1453 MPASS(oldpages == uobj->size);
1454 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1456 if (__predict_true(newpages == oldpages && newsize >= oldsize)) {
1457 node->tn_size = newsize;
1461 if (newpages > oldpages &&
1462 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
1465 VM_OBJECT_WLOCK(uobj);
1466 if (newsize < oldsize) {
1468 * Zero the truncated part of the last page.
1470 base = newsize & PAGE_MASK;
1472 idx = OFF_TO_IDX(newsize);
1474 m = vm_page_grab(uobj, idx, VM_ALLOC_NOCREAT);
1476 MPASS(vm_page_all_valid(m));
1477 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1478 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL |
1482 vm_object_pip_add(uobj, 1);
1483 VM_OBJECT_WUNLOCK(uobj);
1484 rv = vm_pager_get_pages(uobj, &m, 1, NULL,
1486 VM_OBJECT_WLOCK(uobj);
1487 vm_object_pip_wakeup(uobj);
1488 if (rv == VM_PAGER_OK) {
1490 * Since the page was not resident,
1491 * and therefore not recently
1492 * accessed, immediately enqueue it
1493 * for asynchronous laundering. The
1494 * current operation is not regarded
1503 VM_OBJECT_WUNLOCK(uobj);
1509 pmap_zero_page_area(m, base, PAGE_SIZE - base);
1510 vm_page_set_dirty(m);
1516 * Release any swap space and free any whole pages.
1518 if (newpages < oldpages)
1519 vm_object_page_remove(uobj, newpages, 0, 0);
1521 uobj->size = newpages;
1522 VM_OBJECT_WUNLOCK(uobj);
1524 atomic_add_long(&tmp->tm_pages_used, newpages - oldpages);
1526 node->tn_size = newsize;
1531 tmpfs_check_mtime(struct vnode *vp)
1533 struct tmpfs_node *node;
1534 struct vm_object *obj;
1536 ASSERT_VOP_ELOCKED(vp, "check_mtime");
1537 if (vp->v_type != VREG)
1540 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
1541 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
1543 if (obj->generation != obj->cleangeneration) {
1544 VM_OBJECT_WLOCK(obj);
1545 if (obj->generation != obj->cleangeneration) {
1546 obj->cleangeneration = obj->generation;
1547 node = VP_TO_TMPFS_NODE(vp);
1548 node->tn_status |= TMPFS_NODE_MODIFIED |
1551 VM_OBJECT_WUNLOCK(obj);
1556 * Change flags of the given vnode.
1557 * Caller should execute tmpfs_update on vp after a successful execution.
1558 * The vnode must be locked on entry and remain locked on exit.
1561 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
1565 struct tmpfs_node *node;
1567 ASSERT_VOP_ELOCKED(vp, "chflags");
1569 node = VP_TO_TMPFS_NODE(vp);
1571 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK |
1572 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP |
1573 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
1574 UF_SPARSE | UF_SYSTEM)) != 0)
1575 return (EOPNOTSUPP);
1577 /* Disallow this operation if the file system is mounted read-only. */
1578 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1582 * Callers may only modify the file flags on objects they
1583 * have VADMIN rights for.
1585 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1588 * Unprivileged processes are not permitted to unset system
1589 * flags, or modify flags if any system flags are set.
1591 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS)) {
1592 if (node->tn_flags &
1593 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1594 error = securelevel_gt(cred, 0);
1599 if (node->tn_flags &
1600 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1601 ((flags ^ node->tn_flags) & SF_SETTABLE))
1604 node->tn_flags = flags;
1605 node->tn_status |= TMPFS_NODE_CHANGED;
1607 ASSERT_VOP_ELOCKED(vp, "chflags2");
1613 * Change access mode on the given vnode.
1614 * Caller should execute tmpfs_update on vp after a successful execution.
1615 * The vnode must be locked on entry and remain locked on exit.
1618 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1621 struct tmpfs_node *node;
1624 ASSERT_VOP_ELOCKED(vp, "chmod");
1625 ASSERT_VOP_IN_SEQC(vp);
1627 node = VP_TO_TMPFS_NODE(vp);
1629 /* Disallow this operation if the file system is mounted read-only. */
1630 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1633 /* Immutable or append-only files cannot be modified, either. */
1634 if (node->tn_flags & (IMMUTABLE | APPEND))
1638 * To modify the permissions on a file, must possess VADMIN
1641 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1645 * Privileged processes may set the sticky bit on non-directories,
1646 * as well as set the setgid bit on a file with a group that the
1647 * process is not a member of.
1649 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1650 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE))
1653 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1654 error = priv_check_cred(cred, PRIV_VFS_SETGID);
1659 newmode = node->tn_mode & ~ALLPERMS;
1660 newmode |= mode & ALLPERMS;
1661 atomic_store_short(&node->tn_mode, newmode);
1663 node->tn_status |= TMPFS_NODE_CHANGED;
1665 ASSERT_VOP_ELOCKED(vp, "chmod2");
1671 * Change ownership of the given vnode. At least one of uid or gid must
1672 * be different than VNOVAL. If one is set to that value, the attribute
1674 * Caller should execute tmpfs_update on vp after a successful execution.
1675 * The vnode must be locked on entry and remain locked on exit.
1678 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1682 struct tmpfs_node *node;
1687 ASSERT_VOP_ELOCKED(vp, "chown");
1688 ASSERT_VOP_IN_SEQC(vp);
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 newmode = node->tn_mode & ~(S_ISUID | S_ISGID);
1736 atomic_store_short(&node->tn_mode, newmode);
1740 ASSERT_VOP_ELOCKED(vp, "chown2");
1746 * Change size of the given vnode.
1747 * Caller should execute tmpfs_update on vp after a successful execution.
1748 * The vnode must be locked on entry and remain locked on exit.
1751 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1755 struct tmpfs_node *node;
1757 ASSERT_VOP_ELOCKED(vp, "chsize");
1759 node = VP_TO_TMPFS_NODE(vp);
1761 /* Decide whether this is a valid operation based on the file type. */
1763 switch (vp->v_type) {
1768 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1777 /* Allow modifications of special files even if in the file
1778 * system is mounted read-only (we are not modifying the
1779 * files themselves, but the objects they represent). */
1783 /* Anything else is unsupported. */
1787 /* Immutable or append-only files cannot be modified, either. */
1788 if (node->tn_flags & (IMMUTABLE | APPEND))
1791 error = tmpfs_truncate(vp, size);
1792 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1793 * for us, as will update tn_status; no need to do that here. */
1795 ASSERT_VOP_ELOCKED(vp, "chsize2");
1801 * Change access and modification times of the given vnode.
1802 * Caller should execute tmpfs_update on vp after a successful execution.
1803 * The vnode must be locked on entry and remain locked on exit.
1806 tmpfs_chtimes(struct vnode *vp, struct vattr *vap,
1807 struct ucred *cred, struct thread *l)
1810 struct tmpfs_node *node;
1812 ASSERT_VOP_ELOCKED(vp, "chtimes");
1814 node = VP_TO_TMPFS_NODE(vp);
1816 /* Disallow this operation if the file system is mounted read-only. */
1817 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1820 /* Immutable or append-only files cannot be modified, either. */
1821 if (node->tn_flags & (IMMUTABLE | APPEND))
1824 error = vn_utimes_perm(vp, vap, cred, l);
1828 if (vap->va_atime.tv_sec != VNOVAL)
1829 node->tn_status |= TMPFS_NODE_ACCESSED;
1831 if (vap->va_mtime.tv_sec != VNOVAL)
1832 node->tn_status |= TMPFS_NODE_MODIFIED;
1834 if (vap->va_birthtime.tv_sec != VNOVAL)
1835 node->tn_status |= TMPFS_NODE_MODIFIED;
1837 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime);
1839 if (vap->va_birthtime.tv_sec != VNOVAL)
1840 node->tn_birthtime = vap->va_birthtime;
1841 ASSERT_VOP_ELOCKED(vp, "chtimes2");
1847 tmpfs_set_status(struct tmpfs_mount *tm, struct tmpfs_node *node, int status)
1850 if ((node->tn_status & status) == status || tm->tm_ronly)
1852 TMPFS_NODE_LOCK(node);
1853 node->tn_status |= status;
1854 TMPFS_NODE_UNLOCK(node);
1857 /* Sync timestamps */
1859 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1860 const struct timespec *mod)
1862 struct tmpfs_node *node;
1863 struct timespec now;
1865 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes");
1866 node = VP_TO_TMPFS_NODE(vp);
1868 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1869 TMPFS_NODE_CHANGED)) == 0)
1872 vfs_timestamp(&now);
1873 TMPFS_NODE_LOCK(node);
1874 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1877 node->tn_atime = *acc;
1879 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1882 node->tn_mtime = *mod;
1884 if (node->tn_status & TMPFS_NODE_CHANGED)
1885 node->tn_ctime = now;
1886 node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1887 TMPFS_NODE_CHANGED);
1888 TMPFS_NODE_UNLOCK(node);
1890 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
1891 random_harvest_queue(node, sizeof(*node), RANDOM_FS_ATIME);
1895 tmpfs_truncate(struct vnode *vp, off_t length)
1898 struct tmpfs_node *node;
1900 node = VP_TO_TMPFS_NODE(vp);
1907 if (node->tn_size == length) {
1912 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1915 error = tmpfs_reg_resize(vp, length, FALSE);
1917 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1926 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1928 if (a->td_hash > b->td_hash)
1930 else if (a->td_hash < b->td_hash)
1935 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);