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 | CTLFLAG_MPSAFE, 0,
74 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
76 static uma_zone_t tmpfs_dirent_pool;
77 static uma_zone_t tmpfs_node_pool;
81 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
83 struct tmpfs_node *node;
91 node->tn_vnode = NULL;
97 tmpfs_node_dtor(void *mem, int size, void *arg)
99 struct tmpfs_node *node;
102 node->tn_type = VNON;
106 tmpfs_node_init(void *mem, int size, int flags)
108 struct tmpfs_node *node;
112 mtx_init(&node->tn_interlock, "tmpfsni", NULL, MTX_DEF);
113 node->tn_gen = arc4random();
118 tmpfs_node_fini(void *mem, int size)
120 struct tmpfs_node *node;
123 mtx_destroy(&node->tn_interlock);
127 tmpfs_subr_init(void)
129 tmpfs_dirent_pool = uma_zcreate("TMPFS dirent",
130 sizeof(struct tmpfs_dirent), NULL, NULL, NULL, NULL,
132 tmpfs_node_pool = uma_zcreate("TMPFS node",
133 sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor,
134 tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0);
135 VFS_SMR_ZONE_SET(tmpfs_node_pool);
139 tmpfs_subr_uninit(void)
141 uma_zdestroy(tmpfs_node_pool);
142 uma_zdestroy(tmpfs_dirent_pool);
146 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
151 pages = *(long *)arg1;
152 bytes = pages * PAGE_SIZE;
154 error = sysctl_handle_long(oidp, &bytes, 0, req);
155 if (error || !req->newptr)
158 pages = bytes / PAGE_SIZE;
159 if (pages < TMPFS_PAGES_MINRESERVED)
162 *(long *)arg1 = pages;
166 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved,
167 CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RW, &tmpfs_pages_reserved, 0,
168 sysctl_mem_reserved, "L",
169 "Amount of available memory and swap below which tmpfs growth stops");
171 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a,
172 struct tmpfs_dirent *b);
173 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
176 tmpfs_mem_avail(void)
180 avail = swap_pager_avail + vm_free_count() - tmpfs_pages_reserved;
181 if (__predict_false(avail < 0))
187 tmpfs_pages_used(struct tmpfs_mount *tmp)
189 const size_t node_size = sizeof(struct tmpfs_node) +
190 sizeof(struct tmpfs_dirent);
193 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
195 return (meta_pages + tmp->tm_pages_used);
199 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
201 if (tmpfs_mem_avail() < req_pages)
204 if (tmp->tm_pages_max != ULONG_MAX &&
205 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
212 tmpfs_ref_node(struct tmpfs_node *node)
215 TMPFS_NODE_LOCK(node);
216 tmpfs_ref_node_locked(node);
217 TMPFS_NODE_UNLOCK(node);
221 tmpfs_ref_node_locked(struct tmpfs_node *node)
224 TMPFS_NODE_ASSERT_LOCKED(node);
225 KASSERT(node->tn_refcount > 0, ("node %p zero refcount", node));
226 KASSERT(node->tn_refcount < UINT_MAX, ("node %p refcount %u", node,
232 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
233 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
234 * using the credentials of the process 'p'.
236 * If the node type is set to 'VDIR', then the parent parameter must point
237 * to the parent directory of the node being created. It may only be NULL
238 * while allocating the root node.
240 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
241 * specifies the device the node represents.
243 * If the node type is set to 'VLNK', then the parameter target specifies
244 * the file name of the target file for the symbolic link that is being
247 * Note that new nodes are retrieved from the available list if it has
248 * items or, if it is empty, from the node pool as long as there is enough
249 * space to create them.
251 * Returns zero on success or an appropriate error code on failure.
254 tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type,
255 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
256 const char *target, dev_t rdev, struct tmpfs_node **node)
258 struct tmpfs_node *nnode;
261 /* If the root directory of the 'tmp' file system is not yet
262 * allocated, this must be the request to do it. */
263 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
265 MPASS(IFF(type == VLNK, target != NULL));
266 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
268 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
270 if (tmpfs_pages_check_avail(tmp, 1) == 0)
273 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
275 * When a new tmpfs node is created for fully
276 * constructed mount point, there must be a parent
277 * node, which vnode is locked exclusively. As
278 * consequence, if the unmount is executing in
279 * parallel, vflush() cannot reclaim the parent vnode.
280 * Due to this, the check for MNTK_UNMOUNT flag is not
281 * racy: if we did not see MNTK_UNMOUNT flag, then tmp
282 * cannot be destroyed until node construction is
283 * finished and the parent vnode unlocked.
285 * Tmpfs does not need to instantiate new nodes during
290 if ((mp->mnt_kern_flag & MNT_RDONLY) != 0)
293 nnode = uma_zalloc_smr(tmpfs_node_pool, M_WAITOK);
295 /* Generic initialization. */
296 nnode->tn_type = type;
297 vfs_timestamp(&nnode->tn_atime);
298 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
302 nnode->tn_mode = mode;
303 nnode->tn_id = alloc_unr64(&tmp->tm_ino_unr);
304 nnode->tn_refcount = 1;
306 /* Type-specific initialization. */
307 switch (nnode->tn_type) {
310 nnode->tn_rdev = rdev;
314 RB_INIT(&nnode->tn_dir.tn_dirhead);
315 LIST_INIT(&nnode->tn_dir.tn_dupindex);
316 MPASS(parent != nnode);
317 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
318 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
319 nnode->tn_dir.tn_readdir_lastn = 0;
320 nnode->tn_dir.tn_readdir_lastp = NULL;
322 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
323 nnode->tn_dir.tn_parent->tn_links++;
324 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
333 MPASS(strlen(target) < MAXPATHLEN);
334 nnode->tn_size = strlen(target);
335 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
337 memcpy(nnode->tn_link, target, nnode->tn_size);
341 obj = nnode->tn_reg.tn_aobj =
342 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
343 NULL /* XXXKIB - tmpfs needs swap reservation */);
344 VM_OBJECT_WLOCK(obj);
345 /* OBJ_TMPFS is set together with the setting of vp->v_object */
346 vm_object_set_flag(obj, OBJ_TMPFS_NODE);
347 VM_OBJECT_WUNLOCK(obj);
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)
375 TMPFS_NODE_LOCK(node);
376 if (!tmpfs_free_node_locked(tmp, node, false)) {
377 TMPFS_NODE_UNLOCK(node);
383 tmpfs_free_node_locked(struct tmpfs_mount *tmp, struct tmpfs_node *node,
388 TMPFS_MP_ASSERT_LOCKED(tmp);
389 TMPFS_NODE_ASSERT_LOCKED(node);
390 KASSERT(node->tn_refcount > 0, ("node %p refcount zero", node));
393 if (node->tn_attached && (detach || node->tn_refcount == 0)) {
394 MPASS(tmp->tm_nodes_inuse > 0);
395 tmp->tm_nodes_inuse--;
396 LIST_REMOVE(node, tn_entries);
397 node->tn_attached = false;
399 if (node->tn_refcount > 0)
403 MPASS(node->tn_vnode == NULL);
404 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
406 TMPFS_NODE_UNLOCK(node);
409 switch (node->tn_type) {
422 free(node->tn_link, M_TMPFSNAME);
426 uobj = node->tn_reg.tn_aobj;
429 atomic_subtract_long(&tmp->tm_pages_used, uobj->size);
430 KASSERT((uobj->flags & OBJ_TMPFS) == 0,
431 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj));
432 vm_object_deallocate(uobj);
437 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
440 uma_zfree_smr(tmpfs_node_pool, node);
446 static __inline uint32_t
447 tmpfs_dirent_hash(const char *name, u_int len)
451 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK;
452 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP
455 if (hash < TMPFS_DIRCOOKIE_MIN)
456 hash += TMPFS_DIRCOOKIE_MIN;
461 static __inline off_t
462 tmpfs_dirent_cookie(struct tmpfs_dirent *de)
465 return (TMPFS_DIRCOOKIE_EOF);
467 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN);
469 return (de->td_cookie);
472 static __inline boolean_t
473 tmpfs_dirent_dup(struct tmpfs_dirent *de)
475 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0);
478 static __inline boolean_t
479 tmpfs_dirent_duphead(struct tmpfs_dirent *de)
481 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0);
485 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen)
487 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen);
488 memcpy(de->ud.td_name, name, namelen);
489 de->td_namelen = namelen;
493 * Allocates a new directory entry for the node node with a name of name.
494 * The new directory entry is returned in *de.
496 * The link count of node is increased by one to reflect the new object
499 * Returns zero on success or an appropriate error code on failure.
502 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
503 const char *name, u_int len, struct tmpfs_dirent **de)
505 struct tmpfs_dirent *nde;
507 nde = uma_zalloc(tmpfs_dirent_pool, M_WAITOK);
510 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
511 tmpfs_dirent_init(nde, name, len);
523 * Frees a directory entry. It is the caller's responsibility to destroy
524 * the node referenced by it if needed.
526 * The link count of node is decreased by one to reflect the removal of an
527 * object that referenced it. This only happens if 'node_exists' is true;
528 * otherwise the function will not access the node referred to by the
529 * directory entry, as it may already have been released from the outside.
532 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
534 struct tmpfs_node *node;
538 MPASS(node->tn_links > 0);
541 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL)
542 free(de->ud.td_name, M_TMPFSNAME);
543 uma_zfree(tmpfs_dirent_pool, de);
547 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj)
550 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject");
551 if (vp->v_type != VREG || obj == NULL)
554 VM_OBJECT_WLOCK(obj);
556 vm_object_clear_flag(obj, OBJ_TMPFS);
557 obj->un_pager.swp.swp_tmpfs = NULL;
558 if (vp->v_writecount < 0)
559 vp->v_writecount = 0;
561 VM_OBJECT_WUNLOCK(obj);
565 * Need to clear v_object for insmntque failure.
568 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg)
571 tmpfs_destroy_vobject(vp, vp->v_object);
574 vp->v_op = &dead_vnodeops;
580 * Allocates a new vnode for the node node or returns a new reference to
581 * an existing one if the node had already a vnode referencing it. The
582 * resulting locked vnode is returned in *vpp.
584 * Returns zero on success or an appropriate error code on failure.
587 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
591 struct tmpfs_mount *tm;
596 tm = VFS_TO_TMPFS(mp);
597 TMPFS_NODE_LOCK(node);
598 tmpfs_ref_node_locked(node);
600 TMPFS_NODE_ASSERT_LOCKED(node);
601 if ((vp = node->tn_vnode) != NULL) {
602 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
604 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) ||
606 (lkflag & LK_NOWAIT) != 0)) {
608 TMPFS_NODE_UNLOCK(node);
613 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),
622 TMPFS_NODE_UNLOCK(node);
623 error = vget(vp, lkflag | LK_INTERLOCK, curthread);
624 if (error == ENOENT) {
625 TMPFS_NODE_LOCK(node);
634 * Make sure the vnode is still there after
635 * getting the interlock to avoid racing a free.
637 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
639 TMPFS_NODE_LOCK(node);
646 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
647 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
648 TMPFS_NODE_UNLOCK(node);
655 * otherwise lock the vp list while we call getnewvnode
656 * since that can block.
658 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
659 node->tn_vpstate |= TMPFS_VNODE_WANT;
660 error = msleep((caddr_t) &node->tn_vpstate,
661 TMPFS_NODE_MTX(node), 0, "tmpfs_alloc_vp", 0);
666 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
668 TMPFS_NODE_UNLOCK(node);
670 /* Get a new vnode and associate it with our node. */
671 error = getnewvnode("tmpfs", mp, VFS_TO_TMPFS(mp)->tm_nonc ?
672 &tmpfs_vnodeop_nonc_entries : &tmpfs_vnodeop_entries, &vp);
677 /* lkflag is ignored, the lock is exclusive */
678 (void) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
681 vp->v_type = node->tn_type;
683 /* Type-specific initialization. */
684 switch (node->tn_type) {
694 vp->v_op = &tmpfs_fifoop_entries;
697 object = node->tn_reg.tn_aobj;
698 VM_OBJECT_WLOCK(object);
700 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
701 vp->v_object = object;
702 object->un_pager.swp.swp_tmpfs = vp;
703 vm_object_set_flag(object, OBJ_TMPFS);
705 VM_OBJECT_WUNLOCK(object);
708 MPASS(node->tn_dir.tn_parent != NULL);
709 if (node->tn_dir.tn_parent == node)
710 vp->v_vflag |= VV_ROOT;
714 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
716 if (vp->v_type != VFIFO)
719 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL);
724 TMPFS_NODE_LOCK(node);
726 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
727 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
730 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
731 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
732 TMPFS_NODE_UNLOCK(node);
733 wakeup((caddr_t) &node->tn_vpstate);
735 TMPFS_NODE_UNLOCK(node);
742 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
743 TMPFS_NODE_LOCK(node);
744 MPASS(*vpp == node->tn_vnode);
745 TMPFS_NODE_UNLOCK(node);
748 tmpfs_free_node(tm, node);
754 * Destroys the association between the vnode vp and the node it
758 tmpfs_free_vp(struct vnode *vp)
760 struct tmpfs_node *node;
762 node = VP_TO_TMPFS_NODE(vp);
764 TMPFS_NODE_ASSERT_LOCKED(node);
765 node->tn_vnode = NULL;
766 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
767 wakeup(&node->tn_vnode);
768 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
773 * Allocates a new file of type 'type' and adds it to the parent directory
774 * 'dvp'; this addition is done using the component name given in 'cnp'.
775 * The ownership of the new file is automatically assigned based on the
776 * credentials of the caller (through 'cnp'), the group is set based on
777 * the parent directory and the mode is determined from the 'vap' argument.
778 * If successful, *vpp holds a vnode to the newly created file and zero
779 * is returned. Otherwise *vpp is NULL and the function returns an
780 * appropriate error code.
783 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
784 struct componentname *cnp, const char *target)
787 struct tmpfs_dirent *de;
788 struct tmpfs_mount *tmp;
789 struct tmpfs_node *dnode;
790 struct tmpfs_node *node;
791 struct tmpfs_node *parent;
793 ASSERT_VOP_ELOCKED(dvp, "tmpfs_alloc_file");
794 MPASS(cnp->cn_flags & HASBUF);
796 tmp = VFS_TO_TMPFS(dvp->v_mount);
797 dnode = VP_TO_TMPFS_DIR(dvp);
800 /* If the entry we are creating is a directory, we cannot overflow
801 * the number of links of its parent, because it will get a new
803 if (vap->va_type == VDIR) {
804 /* Ensure that we do not overflow the maximum number of links
805 * imposed by the system. */
806 MPASS(dnode->tn_links <= TMPFS_LINK_MAX);
807 if (dnode->tn_links == TMPFS_LINK_MAX) {
812 MPASS(parent != NULL);
816 /* Allocate a node that represents the new file. */
817 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type,
818 cnp->cn_cred->cr_uid, dnode->tn_gid, vap->va_mode, parent,
819 target, vap->va_rdev, &node);
823 /* Allocate a directory entry that points to the new file. */
824 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
827 tmpfs_free_node(tmp, node);
831 /* Allocate a vnode for the new file. */
832 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
834 tmpfs_free_dirent(tmp, de);
835 tmpfs_free_node(tmp, node);
839 /* Now that all required items are allocated, we can proceed to
840 * insert the new node into the directory, an operation that
842 if (cnp->cn_flags & ISWHITEOUT)
843 tmpfs_dir_whiteout_remove(dvp, cnp);
844 tmpfs_dir_attach(dvp, de);
848 struct tmpfs_dirent *
849 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
851 struct tmpfs_dirent *de;
853 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
855 if (de != NULL && tmpfs_dirent_duphead(de))
856 de = LIST_FIRST(&de->ud.td_duphead);
857 dc->tdc_current = de;
859 return (dc->tdc_current);
862 struct tmpfs_dirent *
863 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
865 struct tmpfs_dirent *de;
867 MPASS(dc->tdc_tree != NULL);
868 if (tmpfs_dirent_dup(dc->tdc_current)) {
869 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
870 if (dc->tdc_current != NULL)
871 return (dc->tdc_current);
873 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
874 &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
875 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
876 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
877 MPASS(dc->tdc_current != NULL);
880 return (dc->tdc_current);
883 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
884 static struct tmpfs_dirent *
885 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
887 struct tmpfs_dirent *de, dekey;
889 dekey.td_hash = hash;
890 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
894 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
895 static struct tmpfs_dirent *
896 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
897 struct tmpfs_dir_cursor *dc)
899 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
900 struct tmpfs_dirent *de, dekey;
902 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
904 if (cookie == node->tn_dir.tn_readdir_lastn &&
905 (de = node->tn_dir.tn_readdir_lastp) != NULL) {
906 /* Protect against possible race, tn_readdir_last[pn]
907 * may be updated with only shared vnode lock held. */
908 if (cookie == tmpfs_dirent_cookie(de))
912 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
913 LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
914 uh.td_dup.index_entries) {
915 MPASS(tmpfs_dirent_dup(de));
916 if (de->td_cookie == cookie)
918 /* dupindex list is sorted. */
919 if (de->td_cookie < cookie) {
928 if ((cookie & TMPFS_DIRCOOKIE_MASK) != cookie) {
931 dekey.td_hash = cookie;
932 /* Recover if direntry for cookie was removed */
933 de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
936 dc->tdc_current = de;
937 if (de != NULL && tmpfs_dirent_duphead(de)) {
938 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
939 MPASS(dc->tdc_current != NULL);
941 return (dc->tdc_current);
945 dc->tdc_current = de;
946 if (de != NULL && tmpfs_dirent_dup(de))
947 dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
949 return (dc->tdc_current);
953 * Looks for a directory entry in the directory represented by node.
954 * 'cnp' describes the name of the entry to look for. Note that the .
955 * and .. components are not allowed as they do not physically exist
956 * within directories.
958 * Returns a pointer to the entry when found, otherwise NULL.
960 struct tmpfs_dirent *
961 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
962 struct componentname *cnp)
964 struct tmpfs_dir_duphead *duphead;
965 struct tmpfs_dirent *de;
968 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
969 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
970 cnp->cn_nameptr[1] == '.')));
971 TMPFS_VALIDATE_DIR(node);
973 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
974 de = tmpfs_dir_xlookup_hash(node, hash);
975 if (de != NULL && tmpfs_dirent_duphead(de)) {
976 duphead = &de->ud.td_duphead;
977 LIST_FOREACH(de, duphead, uh.td_dup.entries) {
978 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
982 } else if (de != NULL) {
983 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
987 if (de != NULL && f != NULL && de->td_node != f)
994 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
995 * list, allocate new cookie value.
998 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
999 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
1001 struct tmpfs_dir_duphead *dupindex;
1002 struct tmpfs_dirent *de, *pde;
1004 dupindex = &dnode->tn_dir.tn_dupindex;
1005 de = LIST_FIRST(dupindex);
1006 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
1008 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
1010 nde->td_cookie = de->td_cookie + 1;
1011 MPASS(tmpfs_dirent_dup(nde));
1012 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
1013 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1018 * Cookie numbers are near exhaustion. Scan dupindex list for unused
1019 * numbers. dupindex list is sorted in descending order. Keep it so
1020 * after inserting nde.
1024 de = LIST_NEXT(de, uh.td_dup.index_entries);
1025 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
1027 * Last element of the index doesn't have minimal cookie
1030 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
1031 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
1032 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1034 } else if (de == NULL) {
1036 * We are so lucky have 2^30 hash duplicates in single
1037 * directory :) Return largest possible cookie value.
1038 * It should be fine except possible issues with
1039 * VOP_READDIR restart.
1041 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
1042 LIST_INSERT_HEAD(dupindex, nde,
1043 uh.td_dup.index_entries);
1044 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1047 if (de->td_cookie + 1 == pde->td_cookie ||
1048 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
1049 continue; /* No hole or invalid cookie. */
1050 nde->td_cookie = de->td_cookie + 1;
1051 MPASS(tmpfs_dirent_dup(nde));
1052 MPASS(pde->td_cookie > nde->td_cookie);
1053 MPASS(nde->td_cookie > de->td_cookie);
1054 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
1055 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
1061 * Attaches the directory entry de to the directory represented by vp.
1062 * Note that this does not change the link count of the node pointed by
1063 * the directory entry, as this is done by tmpfs_alloc_dirent.
1066 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
1068 struct tmpfs_node *dnode;
1069 struct tmpfs_dirent *xde, *nde;
1071 ASSERT_VOP_ELOCKED(vp, __func__);
1072 MPASS(de->td_namelen > 0);
1073 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
1074 MPASS(de->td_cookie == de->td_hash);
1076 dnode = VP_TO_TMPFS_DIR(vp);
1077 dnode->tn_dir.tn_readdir_lastn = 0;
1078 dnode->tn_dir.tn_readdir_lastp = NULL;
1080 MPASS(!tmpfs_dirent_dup(de));
1081 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1082 if (xde != NULL && tmpfs_dirent_duphead(xde))
1083 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1084 else if (xde != NULL) {
1086 * Allocate new duphead. Swap xde with duphead to avoid
1087 * adding/removing elements with the same hash.
1089 MPASS(!tmpfs_dirent_dup(xde));
1090 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
1092 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
1093 memcpy(nde, xde, sizeof(*xde));
1094 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
1095 LIST_INIT(&xde->ud.td_duphead);
1096 xde->td_namelen = 0;
1097 xde->td_node = NULL;
1098 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
1099 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
1101 dnode->tn_size += sizeof(struct tmpfs_dirent);
1102 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1103 TMPFS_NODE_MODIFIED;
1108 * Detaches the directory entry de from the directory represented by vp.
1109 * Note that this does not change the link count of the node pointed by
1110 * the directory entry, as this is done by tmpfs_free_dirent.
1113 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
1115 struct tmpfs_mount *tmp;
1116 struct tmpfs_dir *head;
1117 struct tmpfs_node *dnode;
1118 struct tmpfs_dirent *xde;
1120 ASSERT_VOP_ELOCKED(vp, __func__);
1122 dnode = VP_TO_TMPFS_DIR(vp);
1123 head = &dnode->tn_dir.tn_dirhead;
1124 dnode->tn_dir.tn_readdir_lastn = 0;
1125 dnode->tn_dir.tn_readdir_lastp = NULL;
1127 if (tmpfs_dirent_dup(de)) {
1128 /* Remove duphead if de was last entry. */
1129 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1130 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1131 MPASS(tmpfs_dirent_duphead(xde));
1134 LIST_REMOVE(de, uh.td_dup.entries);
1135 LIST_REMOVE(de, uh.td_dup.index_entries);
1137 if (LIST_EMPTY(&xde->ud.td_duphead)) {
1138 RB_REMOVE(tmpfs_dir, head, xde);
1139 tmp = VFS_TO_TMPFS(vp->v_mount);
1140 MPASS(xde->td_node == NULL);
1141 tmpfs_free_dirent(tmp, xde);
1144 de->td_cookie = de->td_hash;
1146 RB_REMOVE(tmpfs_dir, head, de);
1148 dnode->tn_size -= sizeof(struct tmpfs_dirent);
1149 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1150 TMPFS_NODE_MODIFIED;
1155 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1157 struct tmpfs_dirent *de, *dde, *nde;
1159 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1160 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1161 /* Node may already be destroyed. */
1163 if (tmpfs_dirent_duphead(de)) {
1164 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1165 LIST_REMOVE(dde, uh.td_dup.entries);
1166 dde->td_node = NULL;
1167 tmpfs_free_dirent(tmp, dde);
1170 tmpfs_free_dirent(tmp, de);
1175 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
1176 * directory and returns it in the uio space. The function returns 0
1177 * on success, -1 if there was not enough space in the uio structure to
1178 * hold the directory entry or an appropriate error code if another
1182 tmpfs_dir_getdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node,
1188 TMPFS_VALIDATE_DIR(node);
1189 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1191 dent.d_fileno = node->tn_id;
1192 dent.d_type = DT_DIR;
1194 dent.d_name[0] = '.';
1195 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1196 dirent_terminate(&dent);
1198 if (dent.d_reclen > uio->uio_resid)
1199 error = EJUSTRETURN;
1201 error = uiomove(&dent, dent.d_reclen, uio);
1203 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1209 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
1210 * directory and returns it in the uio space. The function returns 0
1211 * on success, -1 if there was not enough space in the uio structure to
1212 * hold the directory entry or an appropriate error code if another
1216 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tm, struct tmpfs_node *node,
1219 struct tmpfs_node *parent;
1223 TMPFS_VALIDATE_DIR(node);
1224 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1227 * Return ENOENT if the current node is already removed.
1229 TMPFS_ASSERT_LOCKED(node);
1230 parent = node->tn_dir.tn_parent;
1234 TMPFS_NODE_LOCK(parent);
1235 dent.d_fileno = parent->tn_id;
1236 TMPFS_NODE_UNLOCK(parent);
1238 dent.d_type = DT_DIR;
1240 dent.d_name[0] = '.';
1241 dent.d_name[1] = '.';
1242 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1243 dirent_terminate(&dent);
1245 if (dent.d_reclen > uio->uio_resid)
1246 error = EJUSTRETURN;
1248 error = uiomove(&dent, dent.d_reclen, uio);
1250 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1256 * Helper function for tmpfs_readdir. Returns as much directory entries
1257 * as can fit in the uio space. The read starts at uio->uio_offset.
1258 * The function returns 0 on success, -1 if there was not enough space
1259 * in the uio structure to hold the directory entry or an appropriate
1260 * error code if another error happens.
1263 tmpfs_dir_getdents(struct tmpfs_mount *tm, struct tmpfs_node *node,
1264 struct uio *uio, int maxcookies, u_long *cookies, int *ncookies)
1266 struct tmpfs_dir_cursor dc;
1267 struct tmpfs_dirent *de;
1271 TMPFS_VALIDATE_DIR(node);
1276 * Lookup the node from the current offset. The starting offset of
1277 * 0 will lookup both '.' and '..', and then the first real entry,
1278 * or EOF if there are none. Then find all entries for the dir that
1279 * fit into the buffer. Once no more entries are found (de == NULL),
1280 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1283 switch (uio->uio_offset) {
1284 case TMPFS_DIRCOOKIE_DOT:
1285 error = tmpfs_dir_getdotdent(tm, node, uio);
1288 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
1289 if (cookies != NULL)
1290 cookies[(*ncookies)++] = off = uio->uio_offset;
1292 case TMPFS_DIRCOOKIE_DOTDOT:
1293 error = tmpfs_dir_getdotdotdent(tm, node, uio);
1296 de = tmpfs_dir_first(node, &dc);
1297 uio->uio_offset = tmpfs_dirent_cookie(de);
1298 if (cookies != NULL)
1299 cookies[(*ncookies)++] = off = uio->uio_offset;
1304 case TMPFS_DIRCOOKIE_EOF:
1307 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1310 if (cookies != NULL)
1311 off = tmpfs_dirent_cookie(de);
1314 /* Read as much entries as possible; i.e., until we reach the end of
1315 * the directory or we exhaust uio space. */
1319 /* Create a dirent structure representing the current
1320 * tmpfs_node and fill it. */
1321 if (de->td_node == NULL) {
1325 d.d_fileno = de->td_node->tn_id;
1326 switch (de->td_node->tn_type) {
1356 panic("tmpfs_dir_getdents: type %p %d",
1357 de->td_node, (int)de->td_node->tn_type);
1360 d.d_namlen = de->td_namelen;
1361 MPASS(de->td_namelen < sizeof(d.d_name));
1362 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1363 d.d_reclen = GENERIC_DIRSIZ(&d);
1364 dirent_terminate(&d);
1366 /* Stop reading if the directory entry we are treating is
1367 * bigger than the amount of data that can be returned. */
1368 if (d.d_reclen > uio->uio_resid) {
1369 error = EJUSTRETURN;
1373 /* Copy the new dirent structure into the output buffer and
1374 * advance pointers. */
1375 error = uiomove(&d, d.d_reclen, uio);
1377 de = tmpfs_dir_next(node, &dc);
1378 if (cookies != NULL) {
1379 off = tmpfs_dirent_cookie(de);
1380 MPASS(*ncookies < maxcookies);
1381 cookies[(*ncookies)++] = off;
1384 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
1386 /* Skip setting off when using cookies as it is already done above. */
1387 if (cookies == NULL)
1388 off = tmpfs_dirent_cookie(de);
1390 /* Update the offset and cache. */
1391 uio->uio_offset = off;
1392 node->tn_dir.tn_readdir_lastn = off;
1393 node->tn_dir.tn_readdir_lastp = de;
1395 tmpfs_set_status(tm, node, TMPFS_NODE_ACCESSED);
1400 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1402 struct tmpfs_dirent *de;
1405 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1406 cnp->cn_nameptr, cnp->cn_namelen, &de);
1409 tmpfs_dir_attach(dvp, de);
1414 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1416 struct tmpfs_dirent *de;
1418 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1419 MPASS(de != NULL && de->td_node == NULL);
1420 tmpfs_dir_detach(dvp, de);
1421 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1425 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1426 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
1427 * 'newsize' must be positive.
1429 * Returns zero on success or an appropriate error code on failure.
1432 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1434 struct tmpfs_mount *tmp;
1435 struct tmpfs_node *node;
1438 vm_pindex_t idx, newpages, oldpages;
1442 MPASS(vp->v_type == VREG);
1443 MPASS(newsize >= 0);
1445 node = VP_TO_TMPFS_NODE(vp);
1446 uobj = node->tn_reg.tn_aobj;
1447 tmp = VFS_TO_TMPFS(vp->v_mount);
1450 * Convert the old and new sizes to the number of pages needed to
1451 * store them. It may happen that we do not need to do anything
1452 * because the last allocated page can accommodate the change on
1455 oldsize = node->tn_size;
1456 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1457 MPASS(oldpages == uobj->size);
1458 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1460 if (__predict_true(newpages == oldpages && newsize >= oldsize)) {
1461 node->tn_size = newsize;
1465 if (newpages > oldpages &&
1466 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
1469 VM_OBJECT_WLOCK(uobj);
1470 if (newsize < oldsize) {
1472 * Zero the truncated part of the last page.
1474 base = newsize & PAGE_MASK;
1476 idx = OFF_TO_IDX(newsize);
1478 m = vm_page_grab(uobj, idx, VM_ALLOC_NOCREAT);
1480 MPASS(vm_page_all_valid(m));
1481 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1482 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL |
1486 vm_object_pip_add(uobj, 1);
1487 VM_OBJECT_WUNLOCK(uobj);
1488 rv = vm_pager_get_pages(uobj, &m, 1, NULL,
1490 VM_OBJECT_WLOCK(uobj);
1491 vm_object_pip_wakeup(uobj);
1492 if (rv == VM_PAGER_OK) {
1494 * Since the page was not resident,
1495 * and therefore not recently
1496 * accessed, immediately enqueue it
1497 * for asynchronous laundering. The
1498 * current operation is not regarded
1507 VM_OBJECT_WUNLOCK(uobj);
1513 pmap_zero_page_area(m, base, PAGE_SIZE - base);
1514 vm_page_set_dirty(m);
1520 * Release any swap space and free any whole pages.
1522 if (newpages < oldpages)
1523 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;
1628 ASSERT_VOP_ELOCKED(vp, "chmod");
1629 ASSERT_VOP_IN_SEQC(vp);
1631 node = VP_TO_TMPFS_NODE(vp);
1633 /* Disallow this operation if the file system is mounted read-only. */
1634 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1637 /* Immutable or append-only files cannot be modified, either. */
1638 if (node->tn_flags & (IMMUTABLE | APPEND))
1642 * To modify the permissions on a file, must possess VADMIN
1645 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1649 * Privileged processes may set the sticky bit on non-directories,
1650 * as well as set the setgid bit on a file with a group that the
1651 * process is not a member of.
1653 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1654 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE))
1657 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1658 error = priv_check_cred(cred, PRIV_VFS_SETGID);
1663 newmode = node->tn_mode & ~ALLPERMS;
1664 newmode |= mode & ALLPERMS;
1665 atomic_store_short(&node->tn_mode, newmode);
1667 node->tn_status |= TMPFS_NODE_CHANGED;
1669 ASSERT_VOP_ELOCKED(vp, "chmod2");
1675 * Change ownership of the given vnode. At least one of uid or gid must
1676 * be different than VNOVAL. If one is set to that value, the attribute
1678 * Caller should execute tmpfs_update on vp after a successful execution.
1679 * The vnode must be locked on entry and remain locked on exit.
1682 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1686 struct tmpfs_node *node;
1691 ASSERT_VOP_ELOCKED(vp, "chown");
1692 ASSERT_VOP_IN_SEQC(vp);
1694 node = VP_TO_TMPFS_NODE(vp);
1696 /* Assign default values if they are unknown. */
1697 MPASS(uid != VNOVAL || gid != VNOVAL);
1702 MPASS(uid != VNOVAL && gid != VNOVAL);
1704 /* Disallow this operation if the file system is mounted read-only. */
1705 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1708 /* Immutable or append-only files cannot be modified, either. */
1709 if (node->tn_flags & (IMMUTABLE | APPEND))
1713 * To modify the ownership of a file, must possess VADMIN for that
1716 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1720 * To change the owner of a file, or change the group of a file to a
1721 * group of which we are not a member, the caller must have
1724 if ((uid != node->tn_uid ||
1725 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1726 (error = priv_check_cred(cred, PRIV_VFS_CHOWN)))
1729 ogid = node->tn_gid;
1730 ouid = node->tn_uid;
1735 node->tn_status |= TMPFS_NODE_CHANGED;
1737 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1738 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID)) {
1739 newmode = node->tn_mode & ~(S_ISUID | S_ISGID);
1740 atomic_store_short(&node->tn_mode, newmode);
1744 ASSERT_VOP_ELOCKED(vp, "chown2");
1750 * Change size of the given vnode.
1751 * Caller should execute tmpfs_update on vp after a successful execution.
1752 * The vnode must be locked on entry and remain locked on exit.
1755 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1759 struct tmpfs_node *node;
1761 ASSERT_VOP_ELOCKED(vp, "chsize");
1763 node = VP_TO_TMPFS_NODE(vp);
1765 /* Decide whether this is a valid operation based on the file type. */
1767 switch (vp->v_type) {
1772 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1781 /* Allow modifications of special files even if in the file
1782 * system is mounted read-only (we are not modifying the
1783 * files themselves, but the objects they represent). */
1787 /* Anything else is unsupported. */
1791 /* Immutable or append-only files cannot be modified, either. */
1792 if (node->tn_flags & (IMMUTABLE | APPEND))
1795 error = tmpfs_truncate(vp, size);
1796 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1797 * for us, as will update tn_status; no need to do that here. */
1799 ASSERT_VOP_ELOCKED(vp, "chsize2");
1805 * Change access and modification times of the given vnode.
1806 * Caller should execute tmpfs_update on vp after a successful execution.
1807 * The vnode must be locked on entry and remain locked on exit.
1810 tmpfs_chtimes(struct vnode *vp, struct vattr *vap,
1811 struct ucred *cred, struct thread *l)
1814 struct tmpfs_node *node;
1816 ASSERT_VOP_ELOCKED(vp, "chtimes");
1818 node = VP_TO_TMPFS_NODE(vp);
1820 /* Disallow this operation if the file system is mounted read-only. */
1821 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1824 /* Immutable or append-only files cannot be modified, either. */
1825 if (node->tn_flags & (IMMUTABLE | APPEND))
1828 error = vn_utimes_perm(vp, vap, cred, l);
1832 if (vap->va_atime.tv_sec != VNOVAL)
1833 node->tn_status |= TMPFS_NODE_ACCESSED;
1835 if (vap->va_mtime.tv_sec != VNOVAL)
1836 node->tn_status |= TMPFS_NODE_MODIFIED;
1838 if (vap->va_birthtime.tv_sec != VNOVAL)
1839 node->tn_status |= TMPFS_NODE_MODIFIED;
1841 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime);
1843 if (vap->va_birthtime.tv_sec != VNOVAL)
1844 node->tn_birthtime = vap->va_birthtime;
1845 ASSERT_VOP_ELOCKED(vp, "chtimes2");
1851 tmpfs_set_status(struct tmpfs_mount *tm, struct tmpfs_node *node, int status)
1854 if ((node->tn_status & status) == status || tm->tm_ronly)
1856 TMPFS_NODE_LOCK(node);
1857 node->tn_status |= status;
1858 TMPFS_NODE_UNLOCK(node);
1861 /* Sync timestamps */
1863 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1864 const struct timespec *mod)
1866 struct tmpfs_node *node;
1867 struct timespec now;
1869 ASSERT_VOP_LOCKED(vp, "tmpfs_itimes");
1870 node = VP_TO_TMPFS_NODE(vp);
1872 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1873 TMPFS_NODE_CHANGED)) == 0)
1876 vfs_timestamp(&now);
1877 TMPFS_NODE_LOCK(node);
1878 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1881 node->tn_atime = *acc;
1883 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1886 node->tn_mtime = *mod;
1888 if (node->tn_status & TMPFS_NODE_CHANGED)
1889 node->tn_ctime = now;
1890 node->tn_status &= ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1891 TMPFS_NODE_CHANGED);
1892 TMPFS_NODE_UNLOCK(node);
1894 /* XXX: FIX? The entropy here is desirable, but the harvesting may be expensive */
1895 random_harvest_queue(node, sizeof(*node), RANDOM_FS_ATIME);
1899 tmpfs_truncate(struct vnode *vp, off_t length)
1902 struct tmpfs_node *node;
1904 node = VP_TO_TMPFS_NODE(vp);
1911 if (node->tn_size == length) {
1916 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1919 error = tmpfs_reg_resize(vp, length, FALSE);
1921 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1930 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1932 if (a->td_hash > b->td_hash)
1934 else if (a->td_hash < b->td_hash)
1939 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);