1 /* $NetBSD: tmpfs_subr.c,v 1.35 2007/07/09 21:10:50 ad Exp $ */
4 * Copyright (c) 2005 The NetBSD Foundation, Inc.
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30 * POSSIBILITY OF SUCH DAMAGE.
34 * Efficient memory file system supporting functions.
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
39 #include <sys/param.h>
40 #include <sys/fnv_hash.h>
42 #include <sys/namei.h>
45 #include <sys/rwlock.h>
47 #include <sys/systm.h>
48 #include <sys/sysctl.h>
49 #include <sys/vnode.h>
50 #include <sys/vmmeter.h>
53 #include <vm/vm_param.h>
54 #include <vm/vm_object.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_pageout.h>
57 #include <vm/vm_pager.h>
58 #include <vm/vm_extern.h>
60 #include <fs/tmpfs/tmpfs.h>
61 #include <fs/tmpfs/tmpfs_fifoops.h>
62 #include <fs/tmpfs/tmpfs_vnops.h>
64 struct tmpfs_dir_cursor {
65 struct tmpfs_dirent *tdc_current;
66 struct tmpfs_dirent *tdc_tree;
69 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system");
71 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
74 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
79 pages = *(long *)arg1;
80 bytes = pages * PAGE_SIZE;
82 error = sysctl_handle_long(oidp, &bytes, 0, req);
83 if (error || !req->newptr)
86 pages = bytes / PAGE_SIZE;
87 if (pages < TMPFS_PAGES_MINRESERVED)
90 *(long *)arg1 = pages;
94 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, CTLTYPE_LONG|CTLFLAG_RW,
95 &tmpfs_pages_reserved, 0, sysctl_mem_reserved, "L",
96 "Amount of available memory and swap below which tmpfs growth stops");
98 static __inline int tmpfs_dirtree_cmp(struct tmpfs_dirent *a,
99 struct tmpfs_dirent *b);
100 RB_PROTOTYPE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);
103 tmpfs_mem_avail(void)
107 avail = swap_pager_avail + vm_cnt.v_free_count + vm_cnt.v_cache_count -
108 tmpfs_pages_reserved;
109 if (__predict_false(avail < 0))
115 tmpfs_pages_used(struct tmpfs_mount *tmp)
117 const size_t node_size = sizeof(struct tmpfs_node) +
118 sizeof(struct tmpfs_dirent);
121 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
123 return (meta_pages + tmp->tm_pages_used);
127 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
129 if (tmpfs_mem_avail() < req_pages)
132 if (tmp->tm_pages_max != SIZE_MAX &&
133 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
140 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
141 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
142 * using the credentials of the process 'p'.
144 * If the node type is set to 'VDIR', then the parent parameter must point
145 * to the parent directory of the node being created. It may only be NULL
146 * while allocating the root node.
148 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
149 * specifies the device the node represents.
151 * If the node type is set to 'VLNK', then the parameter target specifies
152 * the file name of the target file for the symbolic link that is being
155 * Note that new nodes are retrieved from the available list if it has
156 * items or, if it is empty, from the node pool as long as there is enough
157 * space to create them.
159 * Returns zero on success or an appropriate error code on failure.
162 tmpfs_alloc_node(struct mount *mp, struct tmpfs_mount *tmp, enum vtype type,
163 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
164 char *target, dev_t rdev, struct tmpfs_node **node)
166 struct tmpfs_node *nnode;
169 /* If the root directory of the 'tmp' file system is not yet
170 * allocated, this must be the request to do it. */
171 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
172 KASSERT(tmp->tm_root == NULL || mp->mnt_writeopcount > 0,
173 ("creating node not under vn_start_write"));
175 MPASS(IFF(type == VLNK, target != NULL));
176 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
178 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
180 if (tmpfs_pages_check_avail(tmp, 1) == 0)
183 if ((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) {
185 * When a new tmpfs node is created for fully
186 * constructed mount point, there must be a parent
187 * node, which vnode is locked exclusively. As
188 * consequence, if the unmount is executing in
189 * parallel, vflush() cannot reclaim the parent vnode.
190 * Due to this, the check for MNTK_UNMOUNT flag is not
191 * racy: if we did not see MNTK_UNMOUNT flag, then tmp
192 * cannot be destroyed until node construction is
193 * finished and the parent vnode unlocked.
195 * Tmpfs does not need to instantiate new nodes during
201 nnode = (struct tmpfs_node *)uma_zalloc_arg(
202 tmp->tm_node_pool, tmp, M_WAITOK);
204 /* Generic initialization. */
205 nnode->tn_type = type;
206 vfs_timestamp(&nnode->tn_atime);
207 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
211 nnode->tn_mode = mode;
212 nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
214 /* Type-specific initialization. */
215 switch (nnode->tn_type) {
218 nnode->tn_rdev = rdev;
222 RB_INIT(&nnode->tn_dir.tn_dirhead);
223 LIST_INIT(&nnode->tn_dir.tn_dupindex);
224 MPASS(parent != nnode);
225 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
226 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
227 nnode->tn_dir.tn_readdir_lastn = 0;
228 nnode->tn_dir.tn_readdir_lastp = NULL;
230 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
231 nnode->tn_dir.tn_parent->tn_links++;
232 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
241 MPASS(strlen(target) < MAXPATHLEN);
242 nnode->tn_size = strlen(target);
243 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
245 memcpy(nnode->tn_link, target, nnode->tn_size);
249 obj = nnode->tn_reg.tn_aobj =
250 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
251 NULL /* XXXKIB - tmpfs needs swap reservation */);
252 VM_OBJECT_WLOCK(obj);
253 /* OBJ_TMPFS is set together with the setting of vp->v_object */
254 vm_object_set_flag(obj, OBJ_NOSPLIT | OBJ_TMPFS_NODE);
255 vm_object_clear_flag(obj, OBJ_ONEMAPPING);
256 VM_OBJECT_WUNLOCK(obj);
260 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
264 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
265 tmp->tm_nodes_inuse++;
273 * Destroys the node pointed to by node from the file system 'tmp'.
274 * If the node does not belong to the given mount point, the results are
277 * If the node references a directory; no entries are allowed because
278 * their removal could need a recursive algorithm, something forbidden in
279 * kernel space. Furthermore, there is not need to provide such
280 * functionality (recursive removal) because the only primitives offered
281 * to the user are the removal of empty directories and the deletion of
284 * Note that nodes are not really deleted; in fact, when a node has been
285 * allocated, it cannot be deleted during the whole life of the file
286 * system. Instead, they are moved to the available list and remain there
290 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
295 TMPFS_NODE_LOCK(node);
296 MPASS(node->tn_vnode == NULL);
297 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
298 TMPFS_NODE_UNLOCK(node);
302 LIST_REMOVE(node, tn_entries);
303 tmp->tm_nodes_inuse--;
306 switch (node->tn_type) {
308 /* Do not do anything. VNON is provided to let the
309 * allocation routine clean itself easily by avoiding
310 * duplicating code in it. */
324 free(node->tn_link, M_TMPFSNAME);
328 uobj = node->tn_reg.tn_aobj;
331 tmp->tm_pages_used -= uobj->size;
333 KASSERT((uobj->flags & OBJ_TMPFS) == 0,
334 ("leaked OBJ_TMPFS node %p vm_obj %p", node, uobj));
335 vm_object_deallocate(uobj);
340 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
343 free_unr(tmp->tm_ino_unr, node->tn_id);
344 uma_zfree(tmp->tm_node_pool, node);
347 static __inline uint32_t
348 tmpfs_dirent_hash(const char *name, u_int len)
352 hash = fnv_32_buf(name, len, FNV1_32_INIT + len) & TMPFS_DIRCOOKIE_MASK;
353 #ifdef TMPFS_DEBUG_DIRCOOKIE_DUP
356 if (hash < TMPFS_DIRCOOKIE_MIN)
357 hash += TMPFS_DIRCOOKIE_MIN;
362 static __inline off_t
363 tmpfs_dirent_cookie(struct tmpfs_dirent *de)
366 return (TMPFS_DIRCOOKIE_EOF);
368 MPASS(de->td_cookie >= TMPFS_DIRCOOKIE_MIN);
370 return (de->td_cookie);
373 static __inline boolean_t
374 tmpfs_dirent_dup(struct tmpfs_dirent *de)
376 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUP) != 0);
379 static __inline boolean_t
380 tmpfs_dirent_duphead(struct tmpfs_dirent *de)
382 return ((de->td_cookie & TMPFS_DIRCOOKIE_DUPHEAD) != 0);
386 tmpfs_dirent_init(struct tmpfs_dirent *de, const char *name, u_int namelen)
388 de->td_hash = de->td_cookie = tmpfs_dirent_hash(name, namelen);
389 memcpy(de->ud.td_name, name, namelen);
390 de->td_namelen = namelen;
394 * Allocates a new directory entry for the node node with a name of name.
395 * The new directory entry is returned in *de.
397 * The link count of node is increased by one to reflect the new object
400 * Returns zero on success or an appropriate error code on failure.
403 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
404 const char *name, u_int len, struct tmpfs_dirent **de)
406 struct tmpfs_dirent *nde;
408 nde = uma_zalloc(tmp->tm_dirent_pool, M_WAITOK);
411 nde->ud.td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
412 tmpfs_dirent_init(nde, name, len);
424 * Frees a directory entry. It is the caller's responsibility to destroy
425 * the node referenced by it if needed.
427 * The link count of node is decreased by one to reflect the removal of an
428 * object that referenced it. This only happens if 'node_exists' is true;
429 * otherwise the function will not access the node referred to by the
430 * directory entry, as it may already have been released from the outside.
433 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
435 struct tmpfs_node *node;
439 MPASS(node->tn_links > 0);
442 if (!tmpfs_dirent_duphead(de) && de->ud.td_name != NULL)
443 free(de->ud.td_name, M_TMPFSNAME);
444 uma_zfree(tmp->tm_dirent_pool, de);
448 tmpfs_destroy_vobject(struct vnode *vp, vm_object_t obj)
451 ASSERT_VOP_ELOCKED(vp, "tmpfs_destroy_vobject");
452 if (vp->v_type != VREG || obj == NULL)
455 VM_OBJECT_WLOCK(obj);
457 vm_object_clear_flag(obj, OBJ_TMPFS);
458 obj->un_pager.swp.swp_tmpfs = NULL;
460 VM_OBJECT_WUNLOCK(obj);
464 * Need to clear v_object for insmntque failure.
467 tmpfs_insmntque_dtr(struct vnode *vp, void *dtr_arg)
470 tmpfs_destroy_vobject(vp, vp->v_object);
473 vp->v_op = &dead_vnodeops;
479 * Allocates a new vnode for the node node or returns a new reference to
480 * an existing one if the node had already a vnode referencing it. The
481 * resulting locked vnode is returned in *vpp.
483 * Returns zero on success or an appropriate error code on failure.
486 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
495 TMPFS_NODE_LOCK(node);
497 if ((vp = node->tn_vnode) != NULL) {
498 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
500 if ((node->tn_type == VDIR && node->tn_dir.tn_parent == NULL) ||
501 ((vp->v_iflag & VI_DOOMED) != 0 &&
502 (lkflag & LK_NOWAIT) != 0)) {
504 TMPFS_NODE_UNLOCK(node);
509 if ((vp->v_iflag & VI_DOOMED) != 0) {
511 node->tn_vpstate |= TMPFS_VNODE_WRECLAIM;
512 while ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0) {
513 msleep(&node->tn_vnode, TMPFS_NODE_MTX(node),
518 TMPFS_NODE_UNLOCK(node);
519 error = vget(vp, lkflag | LK_INTERLOCK, curthread);
528 * Make sure the vnode is still there after
529 * getting the interlock to avoid racing a free.
531 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
539 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
540 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
541 TMPFS_NODE_UNLOCK(node);
548 * otherwise lock the vp list while we call getnewvnode
549 * since that can block.
551 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
552 node->tn_vpstate |= TMPFS_VNODE_WANT;
553 error = msleep((caddr_t) &node->tn_vpstate,
554 TMPFS_NODE_MTX(node), PDROP | PCATCH,
555 "tmpfs_alloc_vp", 0);
561 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
563 TMPFS_NODE_UNLOCK(node);
565 /* Get a new vnode and associate it with our node. */
566 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp);
571 /* lkflag is ignored, the lock is exclusive */
572 (void) vn_lock(vp, lkflag | LK_RETRY);
575 vp->v_type = node->tn_type;
577 /* Type-specific initialization. */
578 switch (node->tn_type) {
588 vp->v_op = &tmpfs_fifoop_entries;
591 object = node->tn_reg.tn_aobj;
592 VM_OBJECT_WLOCK(object);
594 KASSERT(vp->v_object == NULL, ("Not NULL v_object in tmpfs"));
595 vp->v_object = object;
596 object->un_pager.swp.swp_tmpfs = vp;
597 vm_object_set_flag(object, OBJ_TMPFS);
599 VM_OBJECT_WUNLOCK(object);
602 MPASS(node->tn_dir.tn_parent != NULL);
603 if (node->tn_dir.tn_parent == node)
604 vp->v_vflag |= VV_ROOT;
608 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
610 if (vp->v_type != VFIFO)
613 error = insmntque1(vp, mp, tmpfs_insmntque_dtr, NULL);
618 TMPFS_NODE_LOCK(node);
620 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
621 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
624 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
625 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
626 TMPFS_NODE_UNLOCK(node);
627 wakeup((caddr_t) &node->tn_vpstate);
629 TMPFS_NODE_UNLOCK(node);
636 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
637 TMPFS_NODE_LOCK(node);
638 MPASS(*vpp == node->tn_vnode);
639 TMPFS_NODE_UNLOCK(node);
647 * Destroys the association between the vnode vp and the node it
651 tmpfs_free_vp(struct vnode *vp)
653 struct tmpfs_node *node;
655 node = VP_TO_TMPFS_NODE(vp);
657 TMPFS_NODE_ASSERT_LOCKED(node);
658 node->tn_vnode = NULL;
659 if ((node->tn_vpstate & TMPFS_VNODE_WRECLAIM) != 0)
660 wakeup(&node->tn_vnode);
661 node->tn_vpstate &= ~TMPFS_VNODE_WRECLAIM;
666 * Allocates a new file of type 'type' and adds it to the parent directory
667 * 'dvp'; this addition is done using the component name given in 'cnp'.
668 * The ownership of the new file is automatically assigned based on the
669 * credentials of the caller (through 'cnp'), the group is set based on
670 * the parent directory and the mode is determined from the 'vap' argument.
671 * If successful, *vpp holds a vnode to the newly created file and zero
672 * is returned. Otherwise *vpp is NULL and the function returns an
673 * appropriate error code.
676 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
677 struct componentname *cnp, char *target)
680 struct tmpfs_dirent *de;
681 struct tmpfs_mount *tmp;
682 struct tmpfs_node *dnode;
683 struct tmpfs_node *node;
684 struct tmpfs_node *parent;
686 MPASS(VOP_ISLOCKED(dvp));
687 MPASS(cnp->cn_flags & HASBUF);
689 tmp = VFS_TO_TMPFS(dvp->v_mount);
690 dnode = VP_TO_TMPFS_DIR(dvp);
693 /* If the entry we are creating is a directory, we cannot overflow
694 * the number of links of its parent, because it will get a new
696 if (vap->va_type == VDIR) {
697 /* Ensure that we do not overflow the maximum number of links
698 * imposed by the system. */
699 MPASS(dnode->tn_links <= LINK_MAX);
700 if (dnode->tn_links == LINK_MAX) {
705 MPASS(parent != NULL);
709 /* Allocate a node that represents the new file. */
710 error = tmpfs_alloc_node(dvp->v_mount, tmp, vap->va_type,
711 cnp->cn_cred->cr_uid,
712 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node);
716 /* Allocate a directory entry that points to the new file. */
717 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
720 tmpfs_free_node(tmp, node);
724 /* Allocate a vnode for the new file. */
725 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
727 tmpfs_free_dirent(tmp, de);
728 tmpfs_free_node(tmp, node);
732 /* Now that all required items are allocated, we can proceed to
733 * insert the new node into the directory, an operation that
735 if (cnp->cn_flags & ISWHITEOUT)
736 tmpfs_dir_whiteout_remove(dvp, cnp);
737 tmpfs_dir_attach(dvp, de);
741 static struct tmpfs_dirent *
742 tmpfs_dir_first(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
744 struct tmpfs_dirent *de;
746 de = RB_MIN(tmpfs_dir, &dnode->tn_dir.tn_dirhead);
748 if (de != NULL && tmpfs_dirent_duphead(de))
749 de = LIST_FIRST(&de->ud.td_duphead);
750 dc->tdc_current = de;
752 return (dc->tdc_current);
755 static struct tmpfs_dirent *
756 tmpfs_dir_next(struct tmpfs_node *dnode, struct tmpfs_dir_cursor *dc)
758 struct tmpfs_dirent *de;
760 MPASS(dc->tdc_tree != NULL);
761 if (tmpfs_dirent_dup(dc->tdc_current)) {
762 dc->tdc_current = LIST_NEXT(dc->tdc_current, uh.td_dup.entries);
763 if (dc->tdc_current != NULL)
764 return (dc->tdc_current);
766 dc->tdc_tree = dc->tdc_current = RB_NEXT(tmpfs_dir,
767 &dnode->tn_dir.tn_dirhead, dc->tdc_tree);
768 if ((de = dc->tdc_current) != NULL && tmpfs_dirent_duphead(de)) {
769 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
770 MPASS(dc->tdc_current != NULL);
773 return (dc->tdc_current);
776 /* Lookup directory entry in RB-Tree. Function may return duphead entry. */
777 static struct tmpfs_dirent *
778 tmpfs_dir_xlookup_hash(struct tmpfs_node *dnode, uint32_t hash)
780 struct tmpfs_dirent *de, dekey;
782 dekey.td_hash = hash;
783 de = RB_FIND(tmpfs_dir, &dnode->tn_dir.tn_dirhead, &dekey);
787 /* Lookup directory entry by cookie, initialize directory cursor accordingly. */
788 static struct tmpfs_dirent *
789 tmpfs_dir_lookup_cookie(struct tmpfs_node *node, off_t cookie,
790 struct tmpfs_dir_cursor *dc)
792 struct tmpfs_dir *dirhead = &node->tn_dir.tn_dirhead;
793 struct tmpfs_dirent *de, dekey;
795 MPASS(cookie >= TMPFS_DIRCOOKIE_MIN);
797 if (cookie == node->tn_dir.tn_readdir_lastn &&
798 (de = node->tn_dir.tn_readdir_lastp) != NULL) {
799 /* Protect against possible race, tn_readdir_last[pn]
800 * may be updated with only shared vnode lock held. */
801 if (cookie == tmpfs_dirent_cookie(de))
805 if ((cookie & TMPFS_DIRCOOKIE_DUP) != 0) {
806 LIST_FOREACH(de, &node->tn_dir.tn_dupindex,
807 uh.td_dup.index_entries) {
808 MPASS(tmpfs_dirent_dup(de));
809 if (de->td_cookie == cookie)
811 /* dupindex list is sorted. */
812 if (de->td_cookie < cookie) {
821 MPASS((cookie & TMPFS_DIRCOOKIE_MASK) == cookie);
822 dekey.td_hash = cookie;
823 /* Recover if direntry for cookie was removed */
824 de = RB_NFIND(tmpfs_dir, dirhead, &dekey);
826 dc->tdc_current = de;
827 if (de != NULL && tmpfs_dirent_duphead(de)) {
828 dc->tdc_current = LIST_FIRST(&de->ud.td_duphead);
829 MPASS(dc->tdc_current != NULL);
831 return (dc->tdc_current);
835 dc->tdc_current = de;
836 if (de != NULL && tmpfs_dirent_dup(de))
837 dc->tdc_tree = tmpfs_dir_xlookup_hash(node,
839 return (dc->tdc_current);
843 * Looks for a directory entry in the directory represented by node.
844 * 'cnp' describes the name of the entry to look for. Note that the .
845 * and .. components are not allowed as they do not physically exist
846 * within directories.
848 * Returns a pointer to the entry when found, otherwise NULL.
850 struct tmpfs_dirent *
851 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
852 struct componentname *cnp)
854 struct tmpfs_dir_duphead *duphead;
855 struct tmpfs_dirent *de;
858 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
859 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
860 cnp->cn_nameptr[1] == '.')));
861 TMPFS_VALIDATE_DIR(node);
863 hash = tmpfs_dirent_hash(cnp->cn_nameptr, cnp->cn_namelen);
864 de = tmpfs_dir_xlookup_hash(node, hash);
865 if (de != NULL && tmpfs_dirent_duphead(de)) {
866 duphead = &de->ud.td_duphead;
867 LIST_FOREACH(de, duphead, uh.td_dup.entries) {
868 if (TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
872 } else if (de != NULL) {
873 if (!TMPFS_DIRENT_MATCHES(de, cnp->cn_nameptr,
877 if (de != NULL && f != NULL && de->td_node != f)
884 * Attach duplicate-cookie directory entry nde to dnode and insert to dupindex
885 * list, allocate new cookie value.
888 tmpfs_dir_attach_dup(struct tmpfs_node *dnode,
889 struct tmpfs_dir_duphead *duphead, struct tmpfs_dirent *nde)
891 struct tmpfs_dir_duphead *dupindex;
892 struct tmpfs_dirent *de, *pde;
894 dupindex = &dnode->tn_dir.tn_dupindex;
895 de = LIST_FIRST(dupindex);
896 if (de == NULL || de->td_cookie < TMPFS_DIRCOOKIE_DUP_MAX) {
898 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
900 nde->td_cookie = de->td_cookie + 1;
901 MPASS(tmpfs_dirent_dup(nde));
902 LIST_INSERT_HEAD(dupindex, nde, uh.td_dup.index_entries);
903 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
908 * Cookie numbers are near exhaustion. Scan dupindex list for unused
909 * numbers. dupindex list is sorted in descending order. Keep it so
910 * after inserting nde.
914 de = LIST_NEXT(de, uh.td_dup.index_entries);
915 if (de == NULL && pde->td_cookie != TMPFS_DIRCOOKIE_DUP_MIN) {
917 * Last element of the index doesn't have minimal cookie
920 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MIN;
921 LIST_INSERT_AFTER(pde, nde, uh.td_dup.index_entries);
922 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
924 } else if (de == NULL) {
926 * We are so lucky have 2^30 hash duplicates in single
927 * directory :) Return largest possible cookie value.
928 * It should be fine except possible issues with
929 * VOP_READDIR restart.
931 nde->td_cookie = TMPFS_DIRCOOKIE_DUP_MAX;
932 LIST_INSERT_HEAD(dupindex, nde,
933 uh.td_dup.index_entries);
934 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
937 if (de->td_cookie + 1 == pde->td_cookie ||
938 de->td_cookie >= TMPFS_DIRCOOKIE_DUP_MAX)
939 continue; /* No hole or invalid cookie. */
940 nde->td_cookie = de->td_cookie + 1;
941 MPASS(tmpfs_dirent_dup(nde));
942 MPASS(pde->td_cookie > nde->td_cookie);
943 MPASS(nde->td_cookie > de->td_cookie);
944 LIST_INSERT_BEFORE(de, nde, uh.td_dup.index_entries);
945 LIST_INSERT_HEAD(duphead, nde, uh.td_dup.entries);
951 * Attaches the directory entry de to the directory represented by vp.
952 * Note that this does not change the link count of the node pointed by
953 * the directory entry, as this is done by tmpfs_alloc_dirent.
956 tmpfs_dir_attach(struct vnode *vp, struct tmpfs_dirent *de)
958 struct tmpfs_node *dnode;
959 struct tmpfs_dirent *xde, *nde;
961 ASSERT_VOP_ELOCKED(vp, __func__);
962 MPASS(de->td_namelen > 0);
963 MPASS(de->td_hash >= TMPFS_DIRCOOKIE_MIN);
964 MPASS(de->td_cookie == de->td_hash);
966 dnode = VP_TO_TMPFS_DIR(vp);
967 dnode->tn_dir.tn_readdir_lastn = 0;
968 dnode->tn_dir.tn_readdir_lastp = NULL;
970 MPASS(!tmpfs_dirent_dup(de));
971 xde = RB_INSERT(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
972 if (xde != NULL && tmpfs_dirent_duphead(xde))
973 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
974 else if (xde != NULL) {
976 * Allocate new duphead. Swap xde with duphead to avoid
977 * adding/removing elements with the same hash.
979 MPASS(!tmpfs_dirent_dup(xde));
980 tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), NULL, NULL, 0,
982 /* *nde = *xde; XXX gcc 4.2.1 may generate invalid code. */
983 memcpy(nde, xde, sizeof(*xde));
984 xde->td_cookie |= TMPFS_DIRCOOKIE_DUPHEAD;
985 LIST_INIT(&xde->ud.td_duphead);
988 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, nde);
989 tmpfs_dir_attach_dup(dnode, &xde->ud.td_duphead, de);
991 dnode->tn_size += sizeof(struct tmpfs_dirent);
992 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
997 * Detaches the directory entry de from the directory represented by vp.
998 * Note that this does not change the link count of the node pointed by
999 * the directory entry, as this is done by tmpfs_free_dirent.
1002 tmpfs_dir_detach(struct vnode *vp, struct tmpfs_dirent *de)
1004 struct tmpfs_mount *tmp;
1005 struct tmpfs_dir *head;
1006 struct tmpfs_node *dnode;
1007 struct tmpfs_dirent *xde;
1009 ASSERT_VOP_ELOCKED(vp, __func__);
1011 dnode = VP_TO_TMPFS_DIR(vp);
1012 head = &dnode->tn_dir.tn_dirhead;
1013 dnode->tn_dir.tn_readdir_lastn = 0;
1014 dnode->tn_dir.tn_readdir_lastp = NULL;
1016 if (tmpfs_dirent_dup(de)) {
1017 /* Remove duphead if de was last entry. */
1018 if (LIST_NEXT(de, uh.td_dup.entries) == NULL) {
1019 xde = tmpfs_dir_xlookup_hash(dnode, de->td_hash);
1020 MPASS(tmpfs_dirent_duphead(xde));
1023 LIST_REMOVE(de, uh.td_dup.entries);
1024 LIST_REMOVE(de, uh.td_dup.index_entries);
1026 if (LIST_EMPTY(&xde->ud.td_duphead)) {
1027 RB_REMOVE(tmpfs_dir, head, xde);
1028 tmp = VFS_TO_TMPFS(vp->v_mount);
1029 MPASS(xde->td_node == NULL);
1030 tmpfs_free_dirent(tmp, xde);
1034 RB_REMOVE(tmpfs_dir, head, de);
1036 dnode->tn_size -= sizeof(struct tmpfs_dirent);
1037 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
1038 TMPFS_NODE_MODIFIED;
1042 tmpfs_dir_destroy(struct tmpfs_mount *tmp, struct tmpfs_node *dnode)
1044 struct tmpfs_dirent *de, *dde, *nde;
1046 RB_FOREACH_SAFE(de, tmpfs_dir, &dnode->tn_dir.tn_dirhead, nde) {
1047 RB_REMOVE(tmpfs_dir, &dnode->tn_dir.tn_dirhead, de);
1048 /* Node may already be destroyed. */
1050 if (tmpfs_dirent_duphead(de)) {
1051 while ((dde = LIST_FIRST(&de->ud.td_duphead)) != NULL) {
1052 LIST_REMOVE(dde, uh.td_dup.entries);
1053 dde->td_node = NULL;
1054 tmpfs_free_dirent(tmp, dde);
1057 tmpfs_free_dirent(tmp, de);
1062 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
1063 * directory and returns it in the uio space. The function returns 0
1064 * on success, -1 if there was not enough space in the uio structure to
1065 * hold the directory entry or an appropriate error code if another
1069 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
1074 TMPFS_VALIDATE_DIR(node);
1075 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
1077 dent.d_fileno = node->tn_id;
1078 dent.d_type = DT_DIR;
1080 dent.d_name[0] = '.';
1081 dent.d_name[1] = '\0';
1082 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1084 if (dent.d_reclen > uio->uio_resid)
1085 error = EJUSTRETURN;
1087 error = uiomove(&dent, dent.d_reclen, uio);
1089 node->tn_status |= TMPFS_NODE_ACCESSED;
1095 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
1096 * directory and returns it in the uio space. The function returns 0
1097 * on success, -1 if there was not enough space in the uio structure to
1098 * hold the directory entry or an appropriate error code if another
1102 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
1107 TMPFS_VALIDATE_DIR(node);
1108 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
1111 * Return ENOENT if the current node is already removed.
1113 TMPFS_ASSERT_LOCKED(node);
1114 if (node->tn_dir.tn_parent == NULL) {
1118 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
1119 dent.d_fileno = node->tn_dir.tn_parent->tn_id;
1120 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
1122 dent.d_type = DT_DIR;
1124 dent.d_name[0] = '.';
1125 dent.d_name[1] = '.';
1126 dent.d_name[2] = '\0';
1127 dent.d_reclen = GENERIC_DIRSIZ(&dent);
1129 if (dent.d_reclen > uio->uio_resid)
1130 error = EJUSTRETURN;
1132 error = uiomove(&dent, dent.d_reclen, uio);
1134 node->tn_status |= TMPFS_NODE_ACCESSED;
1140 * Helper function for tmpfs_readdir. Returns as much directory entries
1141 * as can fit in the uio space. The read starts at uio->uio_offset.
1142 * The function returns 0 on success, -1 if there was not enough space
1143 * in the uio structure to hold the directory entry or an appropriate
1144 * error code if another error happens.
1147 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, int maxcookies,
1148 u_long *cookies, int *ncookies)
1150 struct tmpfs_dir_cursor dc;
1151 struct tmpfs_dirent *de;
1155 TMPFS_VALIDATE_DIR(node);
1160 * Lookup the node from the current offset. The starting offset of
1161 * 0 will lookup both '.' and '..', and then the first real entry,
1162 * or EOF if there are none. Then find all entries for the dir that
1163 * fit into the buffer. Once no more entries are found (de == NULL),
1164 * the offset is set to TMPFS_DIRCOOKIE_EOF, which will cause the next
1167 switch (uio->uio_offset) {
1168 case TMPFS_DIRCOOKIE_DOT:
1169 error = tmpfs_dir_getdotdent(node, uio);
1172 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
1173 if (cookies != NULL)
1174 cookies[(*ncookies)++] = off = uio->uio_offset;
1176 case TMPFS_DIRCOOKIE_DOTDOT:
1177 error = tmpfs_dir_getdotdotdent(node, uio);
1180 de = tmpfs_dir_first(node, &dc);
1181 uio->uio_offset = tmpfs_dirent_cookie(de);
1182 if (cookies != NULL)
1183 cookies[(*ncookies)++] = off = uio->uio_offset;
1188 case TMPFS_DIRCOOKIE_EOF:
1191 de = tmpfs_dir_lookup_cookie(node, uio->uio_offset, &dc);
1194 if (cookies != NULL)
1195 off = tmpfs_dirent_cookie(de);
1198 /* Read as much entries as possible; i.e., until we reach the end of
1199 * the directory or we exhaust uio space. */
1203 /* Create a dirent structure representing the current
1204 * tmpfs_node and fill it. */
1205 if (de->td_node == NULL) {
1209 d.d_fileno = de->td_node->tn_id;
1210 switch (de->td_node->tn_type) {
1240 panic("tmpfs_dir_getdents: type %p %d",
1241 de->td_node, (int)de->td_node->tn_type);
1244 d.d_namlen = de->td_namelen;
1245 MPASS(de->td_namelen < sizeof(d.d_name));
1246 (void)memcpy(d.d_name, de->ud.td_name, de->td_namelen);
1247 d.d_name[de->td_namelen] = '\0';
1248 d.d_reclen = GENERIC_DIRSIZ(&d);
1250 /* Stop reading if the directory entry we are treating is
1251 * bigger than the amount of data that can be returned. */
1252 if (d.d_reclen > uio->uio_resid) {
1253 error = EJUSTRETURN;
1257 /* Copy the new dirent structure into the output buffer and
1258 * advance pointers. */
1259 error = uiomove(&d, d.d_reclen, uio);
1261 de = tmpfs_dir_next(node, &dc);
1262 if (cookies != NULL) {
1263 off = tmpfs_dirent_cookie(de);
1264 MPASS(*ncookies < maxcookies);
1265 cookies[(*ncookies)++] = off;
1268 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
1270 /* Skip setting off when using cookies as it is already done above. */
1271 if (cookies == NULL)
1272 off = tmpfs_dirent_cookie(de);
1274 /* Update the offset and cache. */
1275 uio->uio_offset = off;
1276 node->tn_dir.tn_readdir_lastn = off;
1277 node->tn_dir.tn_readdir_lastp = de;
1279 node->tn_status |= TMPFS_NODE_ACCESSED;
1284 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
1286 struct tmpfs_dirent *de;
1289 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
1290 cnp->cn_nameptr, cnp->cn_namelen, &de);
1293 tmpfs_dir_attach(dvp, de);
1298 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
1300 struct tmpfs_dirent *de;
1302 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
1303 MPASS(de != NULL && de->td_node == NULL);
1304 tmpfs_dir_detach(dvp, de);
1305 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de);
1309 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
1310 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
1311 * 'newsize' must be positive.
1313 * Returns zero on success or an appropriate error code on failure.
1316 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
1318 struct tmpfs_mount *tmp;
1319 struct tmpfs_node *node;
1322 vm_pindex_t idx, newpages, oldpages;
1326 MPASS(vp->v_type == VREG);
1327 MPASS(newsize >= 0);
1329 node = VP_TO_TMPFS_NODE(vp);
1330 uobj = node->tn_reg.tn_aobj;
1331 tmp = VFS_TO_TMPFS(vp->v_mount);
1334 * Convert the old and new sizes to the number of pages needed to
1335 * store them. It may happen that we do not need to do anything
1336 * because the last allocated page can accommodate the change on
1339 oldsize = node->tn_size;
1340 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
1341 MPASS(oldpages == uobj->size);
1342 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
1343 if (newpages > oldpages &&
1344 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
1347 VM_OBJECT_WLOCK(uobj);
1348 if (newsize < oldsize) {
1350 * Zero the truncated part of the last page.
1352 base = newsize & PAGE_MASK;
1354 idx = OFF_TO_IDX(newsize);
1356 m = vm_page_lookup(uobj, idx);
1358 if (vm_page_sleep_if_busy(m, "tmfssz"))
1360 MPASS(m->valid == VM_PAGE_BITS_ALL);
1361 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1362 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL);
1364 VM_OBJECT_WUNLOCK(uobj);
1366 VM_OBJECT_WLOCK(uobj);
1368 } else if (m->valid != VM_PAGE_BITS_ALL) {
1370 rv = vm_pager_get_pages(uobj, ma, 1, 0);
1371 m = vm_page_lookup(uobj, idx);
1373 /* A cached page was reactivated. */
1376 if (rv == VM_PAGER_OK) {
1377 vm_page_deactivate(m);
1386 VM_OBJECT_WUNLOCK(uobj);
1392 pmap_zero_page_area(m, base, PAGE_SIZE - base);
1394 vm_pager_page_unswapped(m);
1399 * Release any swap space and free any whole pages.
1401 if (newpages < oldpages) {
1402 swap_pager_freespace(uobj, newpages, oldpages -
1404 vm_object_page_remove(uobj, newpages, 0, 0);
1407 uobj->size = newpages;
1408 VM_OBJECT_WUNLOCK(uobj);
1411 tmp->tm_pages_used += (newpages - oldpages);
1414 node->tn_size = newsize;
1419 * Change flags of the given vnode.
1420 * Caller should execute tmpfs_update on vp after a successful execution.
1421 * The vnode must be locked on entry and remain locked on exit.
1424 tmpfs_chflags(struct vnode *vp, u_long flags, struct ucred *cred,
1428 struct tmpfs_node *node;
1430 MPASS(VOP_ISLOCKED(vp));
1432 node = VP_TO_TMPFS_NODE(vp);
1434 if ((flags & ~(SF_APPEND | SF_ARCHIVED | SF_IMMUTABLE | SF_NOUNLINK |
1435 UF_APPEND | UF_ARCHIVE | UF_HIDDEN | UF_IMMUTABLE | UF_NODUMP |
1436 UF_NOUNLINK | UF_OFFLINE | UF_OPAQUE | UF_READONLY | UF_REPARSE |
1437 UF_SPARSE | UF_SYSTEM)) != 0)
1438 return (EOPNOTSUPP);
1440 /* Disallow this operation if the file system is mounted read-only. */
1441 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1445 * Callers may only modify the file flags on objects they
1446 * have VADMIN rights for.
1448 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1451 * Unprivileged processes are not permitted to unset system
1452 * flags, or modify flags if any system flags are set.
1454 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1455 if (node->tn_flags &
1456 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1457 error = securelevel_gt(cred, 0);
1462 if (node->tn_flags &
1463 (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1464 ((flags ^ node->tn_flags) & SF_SETTABLE))
1467 node->tn_flags = flags;
1468 node->tn_status |= TMPFS_NODE_CHANGED;
1470 MPASS(VOP_ISLOCKED(vp));
1476 * Change access mode on the given vnode.
1477 * Caller should execute tmpfs_update on vp after a successful execution.
1478 * The vnode must be locked on entry and remain locked on exit.
1481 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1484 struct tmpfs_node *node;
1486 MPASS(VOP_ISLOCKED(vp));
1488 node = VP_TO_TMPFS_NODE(vp);
1490 /* Disallow this operation if the file system is mounted read-only. */
1491 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1494 /* Immutable or append-only files cannot be modified, either. */
1495 if (node->tn_flags & (IMMUTABLE | APPEND))
1499 * To modify the permissions on a file, must possess VADMIN
1502 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1506 * Privileged processes may set the sticky bit on non-directories,
1507 * as well as set the setgid bit on a file with a group that the
1508 * process is not a member of.
1510 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1511 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1514 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1515 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1521 node->tn_mode &= ~ALLPERMS;
1522 node->tn_mode |= mode & ALLPERMS;
1524 node->tn_status |= TMPFS_NODE_CHANGED;
1526 MPASS(VOP_ISLOCKED(vp));
1532 * Change ownership of the given vnode. At least one of uid or gid must
1533 * be different than VNOVAL. If one is set to that value, the attribute
1535 * Caller should execute tmpfs_update on vp after a successful execution.
1536 * The vnode must be locked on entry and remain locked on exit.
1539 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1543 struct tmpfs_node *node;
1547 MPASS(VOP_ISLOCKED(vp));
1549 node = VP_TO_TMPFS_NODE(vp);
1551 /* Assign default values if they are unknown. */
1552 MPASS(uid != VNOVAL || gid != VNOVAL);
1557 MPASS(uid != VNOVAL && gid != VNOVAL);
1559 /* Disallow this operation if the file system is mounted read-only. */
1560 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1563 /* Immutable or append-only files cannot be modified, either. */
1564 if (node->tn_flags & (IMMUTABLE | APPEND))
1568 * To modify the ownership of a file, must possess VADMIN for that
1571 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1575 * To change the owner of a file, or change the group of a file to a
1576 * group of which we are not a member, the caller must have
1579 if ((uid != node->tn_uid ||
1580 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1581 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1584 ogid = node->tn_gid;
1585 ouid = node->tn_uid;
1590 node->tn_status |= TMPFS_NODE_CHANGED;
1592 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1593 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1594 node->tn_mode &= ~(S_ISUID | S_ISGID);
1597 MPASS(VOP_ISLOCKED(vp));
1603 * Change size of the given vnode.
1604 * Caller should execute tmpfs_update on vp after a successful execution.
1605 * The vnode must be locked on entry and remain locked on exit.
1608 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1612 struct tmpfs_node *node;
1614 MPASS(VOP_ISLOCKED(vp));
1616 node = VP_TO_TMPFS_NODE(vp);
1618 /* Decide whether this is a valid operation based on the file type. */
1620 switch (vp->v_type) {
1625 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1634 /* Allow modifications of special files even if in the file
1635 * system is mounted read-only (we are not modifying the
1636 * files themselves, but the objects they represent). */
1640 /* Anything else is unsupported. */
1644 /* Immutable or append-only files cannot be modified, either. */
1645 if (node->tn_flags & (IMMUTABLE | APPEND))
1648 error = tmpfs_truncate(vp, size);
1649 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1650 * for us, as will update tn_status; no need to do that here. */
1652 MPASS(VOP_ISLOCKED(vp));
1658 * Change access and modification times of the given vnode.
1659 * Caller should execute tmpfs_update on vp after a successful execution.
1660 * The vnode must be locked on entry and remain locked on exit.
1663 tmpfs_chtimes(struct vnode *vp, struct vattr *vap,
1664 struct ucred *cred, struct thread *l)
1667 struct tmpfs_node *node;
1669 MPASS(VOP_ISLOCKED(vp));
1671 node = VP_TO_TMPFS_NODE(vp);
1673 /* Disallow this operation if the file system is mounted read-only. */
1674 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1677 /* Immutable or append-only files cannot be modified, either. */
1678 if (node->tn_flags & (IMMUTABLE | APPEND))
1681 error = vn_utimes_perm(vp, vap, cred, l);
1685 if (vap->va_atime.tv_sec != VNOVAL && vap->va_atime.tv_nsec != VNOVAL)
1686 node->tn_status |= TMPFS_NODE_ACCESSED;
1688 if (vap->va_mtime.tv_sec != VNOVAL && vap->va_mtime.tv_nsec != VNOVAL)
1689 node->tn_status |= TMPFS_NODE_MODIFIED;
1691 if (vap->va_birthtime.tv_nsec != VNOVAL &&
1692 vap->va_birthtime.tv_nsec != VNOVAL)
1693 node->tn_status |= TMPFS_NODE_MODIFIED;
1695 tmpfs_itimes(vp, &vap->va_atime, &vap->va_mtime);
1697 if (vap->va_birthtime.tv_nsec != VNOVAL &&
1698 vap->va_birthtime.tv_nsec != VNOVAL)
1699 node->tn_birthtime = vap->va_birthtime;
1700 MPASS(VOP_ISLOCKED(vp));
1705 /* Sync timestamps */
1707 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1708 const struct timespec *mod)
1710 struct tmpfs_node *node;
1711 struct timespec now;
1713 node = VP_TO_TMPFS_NODE(vp);
1715 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1716 TMPFS_NODE_CHANGED)) == 0)
1719 vfs_timestamp(&now);
1720 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1723 node->tn_atime = *acc;
1725 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1728 node->tn_mtime = *mod;
1730 if (node->tn_status & TMPFS_NODE_CHANGED) {
1731 node->tn_ctime = now;
1734 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1738 tmpfs_update(struct vnode *vp)
1741 tmpfs_itimes(vp, NULL, NULL);
1745 tmpfs_truncate(struct vnode *vp, off_t length)
1748 struct tmpfs_node *node;
1750 node = VP_TO_TMPFS_NODE(vp);
1757 if (node->tn_size == length) {
1762 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1765 error = tmpfs_reg_resize(vp, length, FALSE);
1767 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1777 tmpfs_dirtree_cmp(struct tmpfs_dirent *a, struct tmpfs_dirent *b)
1779 if (a->td_hash > b->td_hash)
1781 else if (a->td_hash < b->td_hash)
1786 RB_GENERATE_STATIC(tmpfs_dir, tmpfs_dirent, uh.td_entries, tmpfs_dirtree_cmp);