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>
38 #include <sys/kernel.h>
39 #include <sys/param.h>
40 #include <sys/namei.h>
43 #include <sys/spinlock2.h>
45 #include <sys/systm.h>
46 #include <sys/vnode.h>
47 #include <sys/vmmeter.h>
49 #include <sys/mplock2.h>
52 #include <vm/vm_object.h>
53 #include <vm/vm_page.h>
54 #include <vm/vm_pager.h>
55 #include <vm/vm_extern.h>
57 #include <vfs/tmpfs/tmpfs.h>
58 #include <vfs/tmpfs/tmpfs_fifoops.h>
59 #include <vfs/tmpfs/tmpfs_vnops.h>
61 static ino_t t_ino = 2;
62 static struct spinlock ino_lock;
63 static ino_t tmpfs_fetch_ino(void);
65 /* --------------------------------------------------------------------- */
68 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
69 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
70 * using the credentials of the process 'p'.
72 * If the node type is set to 'VDIR', then the parent parameter must point
73 * to the parent directory of the node being created. It may only be NULL
74 * while allocating the root node.
76 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
77 * specifies the device the node represents.
79 * If the node type is set to 'VLNK', then the parameter target specifies
80 * the file name of the target file for the symbolic link that is being
83 * Note that new nodes are retrieved from the available list if it has
84 * items or, if it is empty, from the node pool as long as there is enough
85 * space to create them.
87 * Returns zero on success or an appropriate error code on failure.
90 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
91 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
92 char *target, int rmajor, int rminor, struct tmpfs_node **node)
94 struct tmpfs_node *nnode;
98 /* If the root directory of the 'tmp' file system is not yet
99 * allocated, this must be the request to do it. */
100 KKASSERT(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
102 KKASSERT(IFF(type == VLNK, target != NULL));
103 KKASSERT(IFF(type == VBLK || type == VCHR, rmajor != VNOVAL));
105 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
108 nnode = (struct tmpfs_node *)objcache_get(tmp->tm_node_pool, M_WAITOK);
110 /* Generic initialization. */
111 nnode->tn_type = type;
113 nnode->tn_ctime = nnode->tn_mtime = nnode->tn_atime
115 nnode->tn_ctimensec = nnode->tn_mtimensec = nnode->tn_atimensec
119 nnode->tn_mode = mode;
120 nnode->tn_id = tmpfs_fetch_ino();
121 nnode->tn_advlock.init_done = 0;
123 /* Type-specific initialization. */
124 switch (nnode->tn_type) {
127 rdev = makeudev(rmajor, rminor);
128 if (rdev == NOUDEV) {
129 objcache_put(tmp->tm_node_pool, nnode);
132 nnode->tn_rdev = rdev;
136 TAILQ_INIT(&nnode->tn_dir.tn_dirhead);
137 KKASSERT(parent != nnode);
138 KKASSERT(IMPLIES(parent == NULL, tmp->tm_root == NULL));
139 nnode->tn_dir.tn_parent = parent;
140 nnode->tn_dir.tn_readdir_lastn = 0;
141 nnode->tn_dir.tn_readdir_lastp = NULL;
145 TMPFS_NODE_LOCK(parent);
147 TMPFS_NODE_UNLOCK(parent);
157 nnode->tn_size = strlen(target);
158 nnode->tn_link = kmalloc(nnode->tn_size + 1, M_TMPFSNAME,
159 M_WAITOK | M_NULLOK);
160 if (nnode->tn_link == NULL) {
161 objcache_put(tmp->tm_node_pool, nnode);
164 bcopy(target, nnode->tn_link, nnode->tn_size);
165 nnode->tn_link[nnode->tn_size] = '\0';
169 nnode->tn_reg.tn_aobj =
170 swap_pager_alloc(NULL, 0, VM_PROT_DEFAULT, 0);
171 nnode->tn_reg.tn_aobj_pages = 0;
176 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
179 TMPFS_NODE_LOCK(nnode);
181 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
182 tmp->tm_nodes_inuse++;
184 TMPFS_NODE_UNLOCK(nnode);
190 /* --------------------------------------------------------------------- */
193 * Destroys the node pointed to by node from the file system 'tmp'.
194 * If the node does not belong to the given mount point, the results are
197 * If the node references a directory; no entries are allowed because
198 * their removal could need a recursive algorithm, something forbidden in
199 * kernel space. Furthermore, there is not need to provide such
200 * functionality (recursive removal) because the only primitives offered
201 * to the user are the removal of empty directories and the deletion of
204 * Note that nodes are not really deleted; in fact, when a node has been
205 * allocated, it cannot be deleted during the whole life of the file
206 * system. Instead, they are moved to the available list and remain there
210 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
212 vm_pindex_t pages = 0;
215 TMPFS_ASSERT_ELOCKED(node);
216 KKASSERT(node->tn_vnode == NULL);
217 KKASSERT((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
221 LIST_REMOVE(node, tn_entries);
222 tmp->tm_nodes_inuse--;
224 TMPFS_NODE_UNLOCK(node);
226 switch (node->tn_type) {
228 /* Do not do anything. VNON is provided to let the
229 * allocation routine clean itself easily by avoiding
230 * duplicating code in it. */
239 * The parent link can be NULL if this is the root
244 KKASSERT(node->tn_dir.tn_parent || node == tmp->tm_root);
245 if (node->tn_dir.tn_parent) {
246 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
247 node->tn_dir.tn_parent->tn_links--;
250 * If the parent directory has no more links and
251 * no vnode ref nothing is going to come along
252 * and clean it up unless we do it here.
254 if (node->tn_dir.tn_parent->tn_links == 0 &&
255 node->tn_dir.tn_parent->tn_vnode == NULL) {
256 tmpfs_free_node(tmp, node->tn_dir.tn_parent);
257 /* eats parent lock */
259 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
261 node->tn_dir.tn_parent = NULL;
265 * If the root node is being destroyed don't leave a
266 * dangling pointer in tmpfs_mount.
268 if (node == tmp->tm_root)
277 kfree(node->tn_link, M_TMPFSNAME);
278 node->tn_link = NULL;
283 if (node->tn_reg.tn_aobj != NULL)
284 vm_object_deallocate(node->tn_reg.tn_aobj);
285 node->tn_reg.tn_aobj = NULL;
286 pages = node->tn_reg.tn_aobj_pages;
290 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
294 * Clean up fields for the next allocation. The objcache only ctors
297 tmpfs_node_ctor(node, NULL, 0);
298 objcache_put(tmp->tm_node_pool, node);
299 /* node is now invalid */
302 tmp->tm_pages_used -= pages;
306 /* --------------------------------------------------------------------- */
309 * Allocates a new directory entry for the node node with a name of name.
310 * The new directory entry is returned in *de.
312 * The link count of node is increased by one to reflect the new object
315 * Returns zero on success or an appropriate error code on failure.
318 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
319 const char *name, uint16_t len, struct tmpfs_dirent **de)
321 struct tmpfs_dirent *nde;
323 nde = objcache_get(tmp->tm_dirent_pool, M_WAITOK);
324 nde->td_name = kmalloc(len + 1, M_TMPFSNAME, M_WAITOK | M_NULLOK);
325 if (nde->td_name == NULL) {
326 objcache_put(tmp->tm_dirent_pool, nde);
330 nde->td_namelen = len;
331 bcopy(name, nde->td_name, len);
332 nde->td_name[len] = '\0';
336 TMPFS_NODE_LOCK(node);
338 TMPFS_NODE_UNLOCK(node);
345 /* --------------------------------------------------------------------- */
348 * Frees a directory entry. It is the caller's responsibility to destroy
349 * the node referenced by it if needed.
351 * The link count of node is decreased by one to reflect the removal of an
352 * object that referenced it. This only happens if 'node_exists' is true;
353 * otherwise the function will not access the node referred to by the
354 * directory entry, as it may already have been released from the outside.
357 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de)
359 struct tmpfs_node *node;
363 TMPFS_NODE_LOCK(node);
364 TMPFS_ASSERT_ELOCKED(node);
365 KKASSERT(node->tn_links > 0);
367 TMPFS_NODE_UNLOCK(node);
369 kfree(de->td_name, M_TMPFSNAME);
373 objcache_put(tmp->tm_dirent_pool, de);
376 /* --------------------------------------------------------------------- */
379 * Allocates a new vnode for the node node or returns a new reference to
380 * an existing one if the node had already a vnode referencing it. The
381 * resulting locked vnode is returned in *vpp.
383 * Returns zero on success or an appropriate error code on failure.
386 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
394 * Interlocked extraction from node. This can race many things.
395 * We have to get a soft reference on the vnode while we hold
396 * the node locked, then acquire it properly and check for races.
398 TMPFS_NODE_LOCK(node);
399 if ((vp = node->tn_vnode) != NULL) {
400 KKASSERT((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
401 vhold_interlocked(vp);
402 TMPFS_NODE_UNLOCK(node);
404 if (vget(vp, lkflag | LK_EXCLUSIVE) != 0) {
408 if (node->tn_vnode != vp) {
419 * This should never happen.
421 if (node->tn_vpstate & TMPFS_VNODE_DOOMED) {
422 TMPFS_NODE_UNLOCK(node);
428 * Interlock against other calls to tmpfs_alloc_vp() trying to
429 * allocate and assign a vp to node.
431 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
432 node->tn_vpstate |= TMPFS_VNODE_WANT;
433 error = tsleep(&node->tn_vpstate, PINTERLOCKED | PCATCH,
434 "tmpfs_alloc_vp", 0);
435 TMPFS_NODE_UNLOCK(node);
440 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
441 TMPFS_NODE_UNLOCK(node);
444 * Allocate a new vnode (may block). The ALLOCATING flag should
445 * prevent a race against someone else assigning node->tn_vnode.
447 error = getnewvnode(VT_TMPFS, mp, &vp, VLKTIMEOUT, LK_CANRECURSE);
451 KKASSERT(node->tn_vnode == NULL);
452 KKASSERT(vp != NULL);
454 vp->v_type = node->tn_type;
456 /* Type-specific initialization. */
457 switch (node->tn_type) {
465 vinitvmio(vp, node->tn_size, BMASK, -1);
470 vp->v_ops = &mp->mnt_vn_fifo_ops;
476 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
482 TMPFS_NODE_LOCK(node);
484 KKASSERT(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
485 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
488 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
489 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
490 TMPFS_NODE_UNLOCK(node);
491 wakeup(&node->tn_vpstate);
493 TMPFS_NODE_UNLOCK(node);
499 KKASSERT(IFF(error == 0, *vpp != NULL && vn_islocked(*vpp)));
501 TMPFS_NODE_LOCK(node);
502 KKASSERT(*vpp == node->tn_vnode);
503 TMPFS_NODE_UNLOCK(node);
509 /* --------------------------------------------------------------------- */
512 * Destroys the association between the vnode vp and the node it
516 tmpfs_free_vp(struct vnode *vp)
518 struct tmpfs_node *node;
520 node = VP_TO_TMPFS_NODE(vp);
522 TMPFS_NODE_LOCK(node);
523 KKASSERT(lockcount(TMPFS_NODE_MTX(node)) > 0);
524 node->tn_vnode = NULL;
525 TMPFS_NODE_UNLOCK(node);
529 /* --------------------------------------------------------------------- */
532 * Allocates a new file of type 'type' and adds it to the parent directory
533 * 'dvp'; this addition is done using the component name given in 'cnp'.
534 * The ownership of the new file is automatically assigned based on the
535 * credentials of the caller (through 'cnp'), the group is set based on
536 * the parent directory and the mode is determined from the 'vap' argument.
537 * If successful, *vpp holds a vnode to the newly created file and zero
538 * is returned. Otherwise *vpp is NULL and the function returns an
539 * appropriate error code.
542 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
543 struct namecache *ncp, struct ucred *cred, char *target)
546 struct tmpfs_dirent *de;
547 struct tmpfs_mount *tmp;
548 struct tmpfs_node *dnode;
549 struct tmpfs_node *node;
550 struct tmpfs_node *parent;
552 tmp = VFS_TO_TMPFS(dvp->v_mount);
553 dnode = VP_TO_TMPFS_DIR(dvp);
556 /* If the entry we are creating is a directory, we cannot overflow
557 * the number of links of its parent, because it will get a new
559 if (vap->va_type == VDIR) {
560 /* Ensure that we do not overflow the maximum number of links
561 * imposed by the system. */
562 KKASSERT(dnode->tn_links <= LINK_MAX);
563 if (dnode->tn_links == LINK_MAX) {
568 KKASSERT(parent != NULL);
572 /* Allocate a node that represents the new file. */
573 error = tmpfs_alloc_node(tmp, vap->va_type, cred->cr_uid,
574 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rmajor, vap->va_rminor, &node);
577 TMPFS_NODE_LOCK(node);
579 /* Allocate a directory entry that points to the new file. */
580 error = tmpfs_alloc_dirent(tmp, node, ncp->nc_name, ncp->nc_nlen, &de);
582 tmpfs_free_node(tmp, node);
587 /* Allocate a vnode for the new file. */
588 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
590 tmpfs_free_dirent(tmp, de);
591 tmpfs_free_node(tmp, node);
596 /* Now that all required items are allocated, we can proceed to
597 * insert the new node into the directory, an operation that
599 tmpfs_dir_attach(dnode, de);
600 TMPFS_NODE_UNLOCK(node);
605 /* --------------------------------------------------------------------- */
608 * Attaches the directory entry de to the directory represented by vp.
609 * Note that this does not change the link count of the node pointed by
610 * the directory entry, as this is done by tmpfs_alloc_dirent.
613 tmpfs_dir_attach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
615 TMPFS_NODE_LOCK(dnode);
616 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
618 TMPFS_ASSERT_ELOCKED(dnode);
619 dnode->tn_size += sizeof(struct tmpfs_dirent);
620 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
622 TMPFS_NODE_UNLOCK(dnode);
625 /* --------------------------------------------------------------------- */
628 * Detaches the directory entry de from the directory represented by vp.
629 * Note that this does not change the link count of the node pointed by
630 * the directory entry, as this is done by tmpfs_free_dirent.
633 tmpfs_dir_detach(struct tmpfs_node *dnode, struct tmpfs_dirent *de)
635 TMPFS_NODE_LOCK(dnode);
636 if (dnode->tn_dir.tn_readdir_lastp == de) {
637 dnode->tn_dir.tn_readdir_lastn = 0;
638 dnode->tn_dir.tn_readdir_lastp = NULL;
640 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
642 TMPFS_ASSERT_ELOCKED(dnode);
643 dnode->tn_size -= sizeof(struct tmpfs_dirent);
644 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED |
646 TMPFS_NODE_UNLOCK(dnode);
649 /* --------------------------------------------------------------------- */
652 * Looks for a directory entry in the directory represented by node.
653 * 'ncp' describes the name of the entry to look for. Note that the .
654 * and .. components are not allowed as they do not physically exist
655 * within directories.
657 * Returns a pointer to the entry when found, otherwise NULL.
659 struct tmpfs_dirent *
660 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
661 struct namecache *ncp)
663 struct tmpfs_dirent *de;
664 int len = ncp->nc_nlen;
666 TMPFS_VALIDATE_DIR(node);
668 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
669 if (f != NULL && de->td_node != f)
671 if (len == de->td_namelen) {
672 if (!memcmp(ncp->nc_name, de->td_name, len))
677 TMPFS_NODE_LOCK(node);
678 node->tn_status |= TMPFS_NODE_ACCESSED;
679 TMPFS_NODE_UNLOCK(node);
684 /* --------------------------------------------------------------------- */
687 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
688 * directory and returns it in the uio space. The function returns 0
689 * on success, -1 if there was not enough space in the uio structure to
690 * hold the directory entry or an appropriate error code if another
694 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
700 TMPFS_VALIDATE_DIR(node);
701 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
703 dent.d_ino = node->tn_id;
704 dent.d_type = DT_DIR;
706 dent.d_name[0] = '.';
707 dent.d_name[1] = '\0';
708 dirsize = _DIRENT_DIRSIZ(&dent);
710 if (dirsize > uio->uio_resid)
713 error = uiomove((caddr_t)&dent, dirsize, uio);
715 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
718 TMPFS_NODE_LOCK(node);
719 node->tn_status |= TMPFS_NODE_ACCESSED;
720 TMPFS_NODE_UNLOCK(node);
725 /* --------------------------------------------------------------------- */
728 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
729 * directory and returns it in the uio space. The function returns 0
730 * on success, -1 if there was not enough space in the uio structure to
731 * hold the directory entry or an appropriate error code if another
735 tmpfs_dir_getdotdotdent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
742 TMPFS_VALIDATE_DIR(node);
743 KKASSERT(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
745 if (node->tn_dir.tn_parent) {
746 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
747 dent.d_ino = node->tn_dir.tn_parent->tn_id;
748 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
750 dent.d_ino = tmp->tm_root->tn_id;
753 dent.d_type = DT_DIR;
755 dent.d_name[0] = '.';
756 dent.d_name[1] = '.';
757 dent.d_name[2] = '\0';
758 dirsize = _DIRENT_DIRSIZ(&dent);
760 if (dirsize > uio->uio_resid)
763 error = uiomove((caddr_t)&dent, dirsize, uio);
765 struct tmpfs_dirent *de;
767 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
769 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
771 uio->uio_offset = tmpfs_dircookie(de);
775 TMPFS_NODE_LOCK(node);
776 node->tn_status |= TMPFS_NODE_ACCESSED;
777 TMPFS_NODE_UNLOCK(node);
782 /* --------------------------------------------------------------------- */
785 * Lookup a directory entry by its associated cookie.
787 struct tmpfs_dirent *
788 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
790 struct tmpfs_dirent *de;
792 if (cookie == node->tn_dir.tn_readdir_lastn &&
793 node->tn_dir.tn_readdir_lastp != NULL) {
794 return node->tn_dir.tn_readdir_lastp;
797 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
798 if (tmpfs_dircookie(de) == cookie) {
806 /* --------------------------------------------------------------------- */
809 * Helper function for tmpfs_readdir. Returns as much directory entries
810 * as can fit in the uio space. The read starts at uio->uio_offset.
811 * The function returns 0 on success, -1 if there was not enough space
812 * in the uio structure to hold the directory entry or an appropriate
813 * error code if another error happens.
816 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
820 struct tmpfs_dirent *de;
822 TMPFS_VALIDATE_DIR(node);
824 /* Locate the first directory entry we have to return. We have cached
825 * the last readdir in the node, so use those values if appropriate.
826 * Otherwise do a linear scan to find the requested entry. */
827 startcookie = uio->uio_offset;
828 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOT);
829 KKASSERT(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
830 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
833 de = tmpfs_dir_lookupbycookie(node, startcookie);
839 /* Read as much entries as possible; i.e., until we reach the end of
840 * the directory or we exhaust uio space. */
845 /* Create a dirent structure representing the current
846 * tmpfs_node and fill it. */
847 d.d_ino = de->td_node->tn_id;
848 switch (de->td_node->tn_type) {
878 panic("tmpfs_dir_getdents: type %p %d",
879 de->td_node, (int)de->td_node->tn_type);
881 d.d_namlen = de->td_namelen;
882 KKASSERT(de->td_namelen < sizeof(d.d_name));
883 bcopy(de->td_name, d.d_name, d.d_namlen);
884 d.d_name[d.d_namlen] = '\0';
885 reclen = _DIRENT_RECLEN(d.d_namlen);
887 /* Stop reading if the directory entry we are treating is
888 * bigger than the amount of data that can be returned. */
889 if (reclen > uio->uio_resid) {
894 /* Copy the new dirent structure into the output buffer and
895 * advance pointers. */
896 error = uiomove((caddr_t)&d, reclen, uio);
899 de = TAILQ_NEXT(de, td_entries);
900 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
902 /* Update the offset and cache. */
904 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
905 node->tn_dir.tn_readdir_lastn = 0;
906 node->tn_dir.tn_readdir_lastp = NULL;
908 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
909 node->tn_dir.tn_readdir_lastp = de;
911 node->tn_status |= TMPFS_NODE_ACCESSED;
916 /* --------------------------------------------------------------------- */
919 * Resizes the aobj associated to the regular file pointed to by vp to
920 * the size newsize. 'vp' must point to a vnode that represents a regular
921 * file. 'newsize' must be positive.
923 * pass trivial as 1 when buf content will be overwritten, otherwise set 0
926 * Returns zero on success or an appropriate error code on failure.
929 tmpfs_reg_resize(struct vnode *vp, off_t newsize, int trivial)
932 vm_pindex_t newpages, oldpages;
933 struct tmpfs_mount *tmp;
934 struct tmpfs_node *node;
938 KKASSERT(vp->v_type == VREG);
939 KKASSERT(newsize >= 0);
942 node = VP_TO_TMPFS_NODE(vp);
943 tmp = VFS_TO_TMPFS(vp->v_mount);
945 /* Convert the old and new sizes to the number of pages needed to
946 * store them. It may happen that we do not need to do anything
947 * because the last allocated page can accommodate the change on
949 oldsize = node->tn_size;
950 oldpages = round_page64(oldsize) / PAGE_SIZE;
951 KKASSERT(oldpages == node->tn_reg.tn_aobj_pages);
952 newpages = round_page64(newsize) / PAGE_SIZE;
954 if (newpages > oldpages &&
955 tmp->tm_pages_used + newpages - oldpages > tmp->tm_pages_max) {
961 tmp->tm_pages_used += (newpages - oldpages);
964 TMPFS_NODE_LOCK(node);
965 node->tn_reg.tn_aobj_pages = newpages;
966 node->tn_size = newsize;
967 TMPFS_NODE_UNLOCK(node);
970 * When adjusting the vnode filesize and its VM object we must
971 * also adjust our backing VM object (aobj). The blocksize
972 * used must match the block sized we use for the buffer cache.
974 * The backing VM object contains no VM pages, only swap
977 if (newsize < oldsize) {
982 error = nvtruncbuf(vp, newsize, BSIZE, -1);
983 aobj = node->tn_reg.tn_aobj;
986 nsize = vp->v_object->size;
989 swap_pager_freespace(aobj, nsize,
996 error = nvextendbuf(vp, oldsize, newsize, BSIZE, BSIZE,
998 aobj = node->tn_reg.tn_aobj;
1000 aobj->size = vp->v_object->size;
1007 /* --------------------------------------------------------------------- */
1010 * Change flags of the given vnode.
1011 * Caller should execute tmpfs_update on vp after a successful execution.
1012 * The vnode must be locked on entry and remain locked on exit.
1015 tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred)
1018 struct tmpfs_node *node;
1021 KKASSERT(vn_islocked(vp));
1023 node = VP_TO_TMPFS_NODE(vp);
1025 /* Disallow this operation if the file system is mounted read-only. */
1026 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1029 fmode = FFLAGS(node->tn_flags);
1031 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
1033 if (fmode & (FWRITE | O_TRUNC)) {
1034 if (vp->v_type == VDIR) {
1037 error = vn_writechk(vp, NULL);
1046 error = VOP_ACCESS(vp, mode, cred);
1051 * Unprivileged processes are not permitted to unset system
1052 * flags, or modify flags if any system flags are set.
1054 * Silently enforce SF_NOCACHE on the root tmpfs vnode so
1055 * tmpfs data is not double-cached by swapcache.
1057 TMPFS_NODE_LOCK(node);
1058 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1061 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1062 error = securelevel_gt(cred, 0);
1066 /* Snapshot flag cannot be set or cleared */
1067 if (((flags & SF_SNAPSHOT) != 0 &&
1068 (node->tn_flags & SF_SNAPSHOT) == 0) ||
1069 ((flags & SF_SNAPSHOT) == 0 &&
1070 (node->tn_flags & SF_SNAPSHOT) != 0))
1073 if (vp->v_flag & VROOT)
1074 flags |= SF_NOCACHE;
1075 node->tn_flags = flags;
1078 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1079 (flags & UF_SETTABLE) != flags)
1081 node->tn_flags &= SF_SETTABLE;
1082 node->tn_flags |= (flags & UF_SETTABLE);
1084 node->tn_status |= TMPFS_NODE_CHANGED;
1085 TMPFS_NODE_UNLOCK(node);
1087 KKASSERT(vn_islocked(vp));
1092 /* --------------------------------------------------------------------- */
1095 * Change access mode on the given vnode.
1096 * Caller should execute tmpfs_update on vp after a successful execution.
1097 * The vnode must be locked on entry and remain locked on exit.
1100 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred)
1103 struct tmpfs_node *node;
1106 KKASSERT(vn_islocked(vp));
1108 node = VP_TO_TMPFS_NODE(vp);
1110 /* Disallow this operation if the file system is mounted read-only. */
1111 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1114 /* Immutable or append-only files cannot be modified, either. */
1115 if (node->tn_flags & (IMMUTABLE | APPEND))
1118 fmode = FFLAGS(node->tn_flags);
1120 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
1122 if (fmode & (FWRITE | O_TRUNC)) {
1123 if (vp->v_type == VDIR) {
1126 error = vn_writechk(vp, NULL);
1135 error = VOP_ACCESS(vp, accmode, cred);
1141 * Privileged processes may set the sticky bit on non-directories,
1142 * as well as set the setgid bit on a file with a group that the
1143 * process is not a member of.
1145 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1146 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1149 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1150 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1156 TMPFS_NODE_LOCK(node);
1157 node->tn_mode &= ~ALLPERMS;
1158 node->tn_mode |= mode & ALLPERMS;
1160 node->tn_status |= TMPFS_NODE_CHANGED;
1161 TMPFS_NODE_UNLOCK(node);
1163 KKASSERT(vn_islocked(vp));
1168 /* --------------------------------------------------------------------- */
1171 * Change ownership of the given vnode. At least one of uid or gid must
1172 * be different than VNOVAL. If one is set to that value, the attribute
1174 * Caller should execute tmpfs_update on vp after a successful execution.
1175 * The vnode must be locked on entry and remain locked on exit.
1178 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred)
1183 struct tmpfs_node *node;
1186 KKASSERT(vn_islocked(vp));
1187 node = VP_TO_TMPFS_NODE(vp);
1189 /* Disallow this operation if the file system is mounted read-only. */
1190 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1193 /* Immutable or append-only files cannot be modified, either. */
1194 if (node->tn_flags & (IMMUTABLE | APPEND))
1197 cur_uid = node->tn_uid;
1198 cur_gid = node->tn_gid;
1199 cur_mode = node->tn_mode;
1200 error = vop_helper_chown(vp, uid, gid, cred,
1201 &cur_uid, &cur_gid, &cur_mode);
1204 TMPFS_NODE_LOCK(node);
1205 if (cur_uid != node->tn_uid ||
1206 cur_gid != node->tn_gid ||
1207 cur_mode != node->tn_mode) {
1210 node->tn_mode = cur_mode;
1211 node->tn_status |= TMPFS_NODE_CHANGED;
1213 TMPFS_NODE_UNLOCK(node);
1219 /* --------------------------------------------------------------------- */
1222 * Change size of the given vnode.
1223 * Caller should execute tmpfs_update on vp after a successful execution.
1224 * The vnode must be locked on entry and remain locked on exit.
1227 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred)
1230 struct tmpfs_node *node;
1232 KKASSERT(vn_islocked(vp));
1234 node = VP_TO_TMPFS_NODE(vp);
1236 /* Decide whether this is a valid operation based on the file type. */
1238 switch (vp->v_type) {
1243 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1252 /* Allow modifications of special files even if in the file
1253 * system is mounted read-only (we are not modifying the
1254 * files themselves, but the objects they represent). */
1258 /* Anything else is unsupported. */
1262 /* Immutable or append-only files cannot be modified, either. */
1263 if (node->tn_flags & (IMMUTABLE | APPEND))
1266 error = tmpfs_truncate(vp, size);
1267 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1268 * for us, as will update tn_status; no need to do that here. */
1270 KKASSERT(vn_islocked(vp));
1275 /* --------------------------------------------------------------------- */
1278 * Change access and modification times of the given vnode.
1279 * Caller should execute tmpfs_update on vp after a successful execution.
1280 * The vnode must be locked on entry and remain locked on exit.
1283 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1284 int vaflags, struct ucred *cred)
1287 struct tmpfs_node *node;
1290 KKASSERT(vn_islocked(vp));
1292 node = VP_TO_TMPFS_NODE(vp);
1294 /* Disallow this operation if the file system is mounted read-only. */
1295 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1298 /* Immutable or append-only files cannot be modified, either. */
1299 if (node->tn_flags & (IMMUTABLE | APPEND))
1302 /* Determine if the user have proper privilege to update time. */
1303 fmode = FFLAGS(node->tn_flags);
1305 if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT))
1307 if (fmode & (FWRITE | O_TRUNC)) {
1308 if (vp->v_type == VDIR) {
1311 error = vn_writechk(vp, NULL);
1321 if (vaflags & VA_UTIMES_NULL) {
1322 error = VOP_ACCESS(vp, mode, cred);
1324 error = VOP_ACCESS(vp, VWRITE, cred);
1326 error = VOP_ACCESS(vp, mode, cred);
1331 TMPFS_NODE_LOCK(node);
1332 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1333 node->tn_status |= TMPFS_NODE_ACCESSED;
1335 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1336 node->tn_status |= TMPFS_NODE_MODIFIED;
1338 TMPFS_NODE_UNLOCK(node);
1340 tmpfs_itimes(vp, atime, mtime);
1342 KKASSERT(vn_islocked(vp));
1347 /* --------------------------------------------------------------------- */
1348 /* Sync timestamps */
1350 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1351 const struct timespec *mod)
1353 struct tmpfs_node *node;
1354 struct timespec now;
1356 node = VP_TO_TMPFS_NODE(vp);
1358 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1359 TMPFS_NODE_CHANGED)) == 0)
1362 vfs_timestamp(&now);
1364 TMPFS_NODE_LOCK(node);
1365 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1368 node->tn_atime = acc->tv_sec;
1369 node->tn_atimensec = acc->tv_nsec;
1371 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1374 node->tn_mtime = mod->tv_sec;
1375 node->tn_mtimensec = mod->tv_nsec;
1377 if (node->tn_status & TMPFS_NODE_CHANGED) {
1378 node->tn_ctime = now.tv_sec;
1379 node->tn_ctimensec = now.tv_nsec;
1382 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1383 TMPFS_NODE_UNLOCK(node);
1386 /* --------------------------------------------------------------------- */
1389 tmpfs_update(struct vnode *vp)
1392 tmpfs_itimes(vp, NULL, NULL);
1395 /* --------------------------------------------------------------------- */
1398 tmpfs_truncate(struct vnode *vp, off_t length)
1401 struct tmpfs_node *node;
1403 node = VP_TO_TMPFS_NODE(vp);
1410 if (node->tn_size == length) {
1415 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1419 error = tmpfs_reg_resize(vp, length, 1);
1422 TMPFS_NODE_LOCK(node);
1423 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;
1424 TMPFS_NODE_UNLOCK(node);
1433 /* --------------------------------------------------------------------- */
1436 tmpfs_fetch_ino(void)
1440 spin_lock(&ino_lock);
1442 spin_unlock(&ino_lock);