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/namei.h>
44 #include <sys/systm.h>
45 #include <sys/sysctl.h>
46 #include <sys/vnode.h>
47 #include <sys/vmmeter.h>
50 #include <vm/vm_object.h>
51 #include <vm/vm_page.h>
52 #include <vm/vm_pageout.h>
53 #include <vm/vm_pager.h>
54 #include <vm/vm_extern.h>
56 #include <fs/tmpfs/tmpfs.h>
57 #include <fs/tmpfs/tmpfs_fifoops.h>
58 #include <fs/tmpfs/tmpfs_vnops.h>
60 SYSCTL_NODE(_vfs, OID_AUTO, tmpfs, CTLFLAG_RW, 0, "tmpfs file system");
62 static long tmpfs_pages_reserved = TMPFS_PAGES_MINRESERVED;
65 sysctl_mem_reserved(SYSCTL_HANDLER_ARGS)
70 pages = *(long *)arg1;
71 bytes = pages * PAGE_SIZE;
73 error = sysctl_handle_long(oidp, &bytes, 0, req);
74 if (error || !req->newptr)
77 pages = bytes / PAGE_SIZE;
78 if (pages < TMPFS_PAGES_MINRESERVED)
81 *(long *)arg1 = pages;
85 SYSCTL_PROC(_vfs_tmpfs, OID_AUTO, memory_reserved, CTLTYPE_LONG|CTLFLAG_RW,
86 &tmpfs_pages_reserved, 0, sysctl_mem_reserved, "L",
87 "Amount of available memory and swap below which tmpfs growth stops");
94 avail = swap_pager_avail + cnt.v_free_count + cnt.v_cache_count -
96 if (__predict_false(avail < 0))
102 tmpfs_pages_used(struct tmpfs_mount *tmp)
104 const size_t node_size = sizeof(struct tmpfs_node) +
105 sizeof(struct tmpfs_dirent);
108 meta_pages = howmany((uintmax_t)tmp->tm_nodes_inuse * node_size,
110 return (meta_pages + tmp->tm_pages_used);
114 tmpfs_pages_check_avail(struct tmpfs_mount *tmp, size_t req_pages)
116 if (tmpfs_mem_avail() < req_pages)
119 if (tmp->tm_pages_max != SIZE_MAX &&
120 tmp->tm_pages_max < req_pages + tmpfs_pages_used(tmp))
126 /* --------------------------------------------------------------------- */
129 * Allocates a new node of type 'type' inside the 'tmp' mount point, with
130 * its owner set to 'uid', its group to 'gid' and its mode set to 'mode',
131 * using the credentials of the process 'p'.
133 * If the node type is set to 'VDIR', then the parent parameter must point
134 * to the parent directory of the node being created. It may only be NULL
135 * while allocating the root node.
137 * If the node type is set to 'VBLK' or 'VCHR', then the rdev parameter
138 * specifies the device the node represents.
140 * If the node type is set to 'VLNK', then the parameter target specifies
141 * the file name of the target file for the symbolic link that is being
144 * Note that new nodes are retrieved from the available list if it has
145 * items or, if it is empty, from the node pool as long as there is enough
146 * space to create them.
148 * Returns zero on success or an appropriate error code on failure.
151 tmpfs_alloc_node(struct tmpfs_mount *tmp, enum vtype type,
152 uid_t uid, gid_t gid, mode_t mode, struct tmpfs_node *parent,
153 char *target, dev_t rdev, struct tmpfs_node **node)
155 struct tmpfs_node *nnode;
157 /* If the root directory of the 'tmp' file system is not yet
158 * allocated, this must be the request to do it. */
159 MPASS(IMPLIES(tmp->tm_root == NULL, parent == NULL && type == VDIR));
161 MPASS(IFF(type == VLNK, target != NULL));
162 MPASS(IFF(type == VBLK || type == VCHR, rdev != VNOVAL));
164 if (tmp->tm_nodes_inuse >= tmp->tm_nodes_max)
166 if (tmpfs_pages_check_avail(tmp, 1) == 0)
169 nnode = (struct tmpfs_node *)uma_zalloc_arg(
170 tmp->tm_node_pool, tmp, M_WAITOK);
172 /* Generic initialization. */
173 nnode->tn_type = type;
174 vfs_timestamp(&nnode->tn_atime);
175 nnode->tn_birthtime = nnode->tn_ctime = nnode->tn_mtime =
179 nnode->tn_mode = mode;
180 nnode->tn_id = alloc_unr(tmp->tm_ino_unr);
182 /* Type-specific initialization. */
183 switch (nnode->tn_type) {
186 nnode->tn_rdev = rdev;
190 TAILQ_INIT(&nnode->tn_dir.tn_dirhead);
191 MPASS(parent != nnode);
192 MPASS(IMPLIES(parent == NULL, tmp->tm_root == NULL));
193 nnode->tn_dir.tn_parent = (parent == NULL) ? nnode : parent;
194 nnode->tn_dir.tn_readdir_lastn = 0;
195 nnode->tn_dir.tn_readdir_lastp = NULL;
197 TMPFS_NODE_LOCK(nnode->tn_dir.tn_parent);
198 nnode->tn_dir.tn_parent->tn_links++;
199 TMPFS_NODE_UNLOCK(nnode->tn_dir.tn_parent);
208 MPASS(strlen(target) < MAXPATHLEN);
209 nnode->tn_size = strlen(target);
210 nnode->tn_link = malloc(nnode->tn_size, M_TMPFSNAME,
212 memcpy(nnode->tn_link, target, nnode->tn_size);
216 nnode->tn_reg.tn_aobj =
217 vm_pager_allocate(OBJT_SWAP, NULL, 0, VM_PROT_DEFAULT, 0,
218 NULL /* XXXKIB - tmpfs needs swap reservation */);
222 panic("tmpfs_alloc_node: type %p %d", nnode, (int)nnode->tn_type);
226 LIST_INSERT_HEAD(&tmp->tm_nodes_used, nnode, tn_entries);
227 tmp->tm_nodes_inuse++;
234 /* --------------------------------------------------------------------- */
237 * Destroys the node pointed to by node from the file system 'tmp'.
238 * If the node does not belong to the given mount point, the results are
241 * If the node references a directory; no entries are allowed because
242 * their removal could need a recursive algorithm, something forbidden in
243 * kernel space. Furthermore, there is not need to provide such
244 * functionality (recursive removal) because the only primitives offered
245 * to the user are the removal of empty directories and the deletion of
248 * Note that nodes are not really deleted; in fact, when a node has been
249 * allocated, it cannot be deleted during the whole life of the file
250 * system. Instead, they are moved to the available list and remain there
254 tmpfs_free_node(struct tmpfs_mount *tmp, struct tmpfs_node *node)
259 TMPFS_NODE_LOCK(node);
260 MPASS(node->tn_vnode == NULL);
261 MPASS((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0);
262 TMPFS_NODE_UNLOCK(node);
266 LIST_REMOVE(node, tn_entries);
267 tmp->tm_nodes_inuse--;
270 switch (node->tn_type) {
272 /* Do not do anything. VNON is provided to let the
273 * allocation routine clean itself easily by avoiding
274 * duplicating code in it. */
288 free(node->tn_link, M_TMPFSNAME);
292 uobj = node->tn_reg.tn_aobj;
295 tmp->tm_pages_used -= uobj->size;
297 vm_object_deallocate(uobj);
302 panic("tmpfs_free_node: type %p %d", node, (int)node->tn_type);
305 free_unr(tmp->tm_ino_unr, node->tn_id);
306 uma_zfree(tmp->tm_node_pool, node);
309 /* --------------------------------------------------------------------- */
312 * Allocates a new directory entry for the node node with a name of name.
313 * The new directory entry is returned in *de.
315 * The link count of node is increased by one to reflect the new object
318 * Returns zero on success or an appropriate error code on failure.
321 tmpfs_alloc_dirent(struct tmpfs_mount *tmp, struct tmpfs_node *node,
322 const char *name, uint16_t len, struct tmpfs_dirent **de)
324 struct tmpfs_dirent *nde;
326 nde = (struct tmpfs_dirent *)uma_zalloc(
327 tmp->tm_dirent_pool, M_WAITOK);
328 nde->td_name = malloc(len, M_TMPFSNAME, M_WAITOK);
329 nde->td_namelen = len;
330 memcpy(nde->td_name, name, len);
341 /* --------------------------------------------------------------------- */
344 * Frees a directory entry. It is the caller's responsibility to destroy
345 * the node referenced by it if needed.
347 * The link count of node is decreased by one to reflect the removal of an
348 * object that referenced it. This only happens if 'node_exists' is true;
349 * otherwise the function will not access the node referred to by the
350 * directory entry, as it may already have been released from the outside.
353 tmpfs_free_dirent(struct tmpfs_mount *tmp, struct tmpfs_dirent *de,
354 boolean_t node_exists)
357 struct tmpfs_node *node;
361 MPASS(node->tn_links > 0);
366 free(de->td_name, M_TMPFSNAME);
367 uma_zfree(tmp->tm_dirent_pool, de);
370 /* --------------------------------------------------------------------- */
373 * Allocates a new vnode for the node node or returns a new reference to
374 * an existing one if the node had already a vnode referencing it. The
375 * resulting locked vnode is returned in *vpp.
377 * Returns zero on success or an appropriate error code on failure.
380 tmpfs_alloc_vp(struct mount *mp, struct tmpfs_node *node, int lkflag,
387 TMPFS_NODE_LOCK(node);
388 if ((vp = node->tn_vnode) != NULL) {
389 MPASS((node->tn_vpstate & TMPFS_VNODE_DOOMED) == 0);
391 TMPFS_NODE_UNLOCK(node);
392 error = vget(vp, lkflag | LK_INTERLOCK, curthread);
399 * Make sure the vnode is still there after
400 * getting the interlock to avoid racing a free.
402 if (node->tn_vnode == NULL || node->tn_vnode != vp) {
410 if ((node->tn_vpstate & TMPFS_VNODE_DOOMED) ||
411 (node->tn_type == VDIR && node->tn_dir.tn_parent == NULL)) {
412 TMPFS_NODE_UNLOCK(node);
419 * otherwise lock the vp list while we call getnewvnode
420 * since that can block.
422 if (node->tn_vpstate & TMPFS_VNODE_ALLOCATING) {
423 node->tn_vpstate |= TMPFS_VNODE_WANT;
424 error = msleep((caddr_t) &node->tn_vpstate,
425 TMPFS_NODE_MTX(node), PDROP | PCATCH,
426 "tmpfs_alloc_vp", 0);
432 node->tn_vpstate |= TMPFS_VNODE_ALLOCATING;
434 TMPFS_NODE_UNLOCK(node);
436 /* Get a new vnode and associate it with our node. */
437 error = getnewvnode("tmpfs", mp, &tmpfs_vnodeop_entries, &vp);
442 (void) vn_lock(vp, lkflag | LK_RETRY);
445 vp->v_type = node->tn_type;
447 /* Type-specific initialization. */
448 switch (node->tn_type) {
460 vp->v_op = &tmpfs_fifoop_entries;
463 MPASS(node->tn_dir.tn_parent != NULL);
464 if (node->tn_dir.tn_parent == node)
465 vp->v_vflag |= VV_ROOT;
469 panic("tmpfs_alloc_vp: type %p %d", node, (int)node->tn_type);
472 vnode_pager_setsize(vp, node->tn_size);
473 error = insmntque(vp, mp);
478 TMPFS_NODE_LOCK(node);
480 MPASS(node->tn_vpstate & TMPFS_VNODE_ALLOCATING);
481 node->tn_vpstate &= ~TMPFS_VNODE_ALLOCATING;
484 if (node->tn_vpstate & TMPFS_VNODE_WANT) {
485 node->tn_vpstate &= ~TMPFS_VNODE_WANT;
486 TMPFS_NODE_UNLOCK(node);
487 wakeup((caddr_t) &node->tn_vpstate);
489 TMPFS_NODE_UNLOCK(node);
496 MPASS(*vpp != NULL && VOP_ISLOCKED(*vpp));
497 TMPFS_NODE_LOCK(node);
498 MPASS(*vpp == node->tn_vnode);
499 TMPFS_NODE_UNLOCK(node);
506 /* --------------------------------------------------------------------- */
509 * Destroys the association between the vnode vp and the node it
513 tmpfs_free_vp(struct vnode *vp)
515 struct tmpfs_node *node;
517 node = VP_TO_TMPFS_NODE(vp);
519 mtx_assert(TMPFS_NODE_MTX(node), MA_OWNED);
520 node->tn_vnode = NULL;
524 /* --------------------------------------------------------------------- */
527 * Allocates a new file of type 'type' and adds it to the parent directory
528 * 'dvp'; this addition is done using the component name given in 'cnp'.
529 * The ownership of the new file is automatically assigned based on the
530 * credentials of the caller (through 'cnp'), the group is set based on
531 * the parent directory and the mode is determined from the 'vap' argument.
532 * If successful, *vpp holds a vnode to the newly created file and zero
533 * is returned. Otherwise *vpp is NULL and the function returns an
534 * appropriate error code.
537 tmpfs_alloc_file(struct vnode *dvp, struct vnode **vpp, struct vattr *vap,
538 struct componentname *cnp, char *target)
541 struct tmpfs_dirent *de;
542 struct tmpfs_mount *tmp;
543 struct tmpfs_node *dnode;
544 struct tmpfs_node *node;
545 struct tmpfs_node *parent;
547 MPASS(VOP_ISLOCKED(dvp));
548 MPASS(cnp->cn_flags & HASBUF);
550 tmp = VFS_TO_TMPFS(dvp->v_mount);
551 dnode = VP_TO_TMPFS_DIR(dvp);
554 /* If the entry we are creating is a directory, we cannot overflow
555 * the number of links of its parent, because it will get a new
557 if (vap->va_type == VDIR) {
558 /* Ensure that we do not overflow the maximum number of links
559 * imposed by the system. */
560 MPASS(dnode->tn_links <= LINK_MAX);
561 if (dnode->tn_links == LINK_MAX) {
567 MPASS(parent != NULL);
571 /* Allocate a node that represents the new file. */
572 error = tmpfs_alloc_node(tmp, vap->va_type, cnp->cn_cred->cr_uid,
573 dnode->tn_gid, vap->va_mode, parent, target, vap->va_rdev, &node);
577 /* Allocate a directory entry that points to the new file. */
578 error = tmpfs_alloc_dirent(tmp, node, cnp->cn_nameptr, cnp->cn_namelen,
581 tmpfs_free_node(tmp, node);
585 /* Allocate a vnode for the new file. */
586 error = tmpfs_alloc_vp(dvp->v_mount, node, LK_EXCLUSIVE, vpp);
588 tmpfs_free_dirent(tmp, de, TRUE);
589 tmpfs_free_node(tmp, node);
593 /* Now that all required items are allocated, we can proceed to
594 * insert the new node into the directory, an operation that
596 if (cnp->cn_flags & ISWHITEOUT)
597 tmpfs_dir_whiteout_remove(dvp, cnp);
598 tmpfs_dir_attach(dvp, de);
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 vnode *vp, struct tmpfs_dirent *de)
615 struct tmpfs_node *dnode;
617 ASSERT_VOP_ELOCKED(vp, __func__);
618 dnode = VP_TO_TMPFS_DIR(vp);
619 TAILQ_INSERT_TAIL(&dnode->tn_dir.tn_dirhead, de, td_entries);
620 dnode->tn_size += sizeof(struct tmpfs_dirent);
621 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
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 vnode *vp, struct tmpfs_dirent *de)
635 struct tmpfs_node *dnode;
637 ASSERT_VOP_ELOCKED(vp, __func__);
638 dnode = VP_TO_TMPFS_DIR(vp);
640 if (dnode->tn_dir.tn_readdir_lastp == de) {
641 dnode->tn_dir.tn_readdir_lastn = 0;
642 dnode->tn_dir.tn_readdir_lastp = NULL;
645 TAILQ_REMOVE(&dnode->tn_dir.tn_dirhead, de, td_entries);
646 dnode->tn_size -= sizeof(struct tmpfs_dirent);
647 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \
651 /* --------------------------------------------------------------------- */
654 * Looks for a directory entry in the directory represented by node.
655 * 'cnp' describes the name of the entry to look for. Note that the .
656 * and .. components are not allowed as they do not physically exist
657 * within directories.
659 * Returns a pointer to the entry when found, otherwise NULL.
661 struct tmpfs_dirent *
662 tmpfs_dir_lookup(struct tmpfs_node *node, struct tmpfs_node *f,
663 struct componentname *cnp)
666 struct tmpfs_dirent *de;
668 MPASS(IMPLIES(cnp->cn_namelen == 1, cnp->cn_nameptr[0] != '.'));
669 MPASS(IMPLIES(cnp->cn_namelen == 2, !(cnp->cn_nameptr[0] == '.' &&
670 cnp->cn_nameptr[1] == '.')));
671 TMPFS_VALIDATE_DIR(node);
674 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
675 if (f != NULL && de->td_node != f)
677 MPASS(cnp->cn_namelen < 0xffff);
678 if (de->td_namelen == (uint16_t)cnp->cn_namelen &&
679 bcmp(de->td_name, cnp->cn_nameptr, de->td_namelen) == 0) {
684 node->tn_status |= TMPFS_NODE_ACCESSED;
686 return found ? de : NULL;
689 /* --------------------------------------------------------------------- */
692 * Helper function for tmpfs_readdir. Creates a '.' entry for the given
693 * directory and returns it in the uio space. The function returns 0
694 * on success, -1 if there was not enough space in the uio structure to
695 * hold the directory entry or an appropriate error code if another
699 tmpfs_dir_getdotdent(struct tmpfs_node *node, struct uio *uio)
704 TMPFS_VALIDATE_DIR(node);
705 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOT);
707 dent.d_fileno = node->tn_id;
708 dent.d_type = DT_DIR;
710 dent.d_name[0] = '.';
711 dent.d_name[1] = '\0';
712 dent.d_reclen = GENERIC_DIRSIZ(&dent);
714 if (dent.d_reclen > uio->uio_resid)
717 error = uiomove(&dent, dent.d_reclen, uio);
719 uio->uio_offset = TMPFS_DIRCOOKIE_DOTDOT;
722 node->tn_status |= TMPFS_NODE_ACCESSED;
727 /* --------------------------------------------------------------------- */
730 * Helper function for tmpfs_readdir. Creates a '..' entry for the given
731 * directory and returns it in the uio space. The function returns 0
732 * on success, -1 if there was not enough space in the uio structure to
733 * hold the directory entry or an appropriate error code if another
737 tmpfs_dir_getdotdotdent(struct tmpfs_node *node, struct uio *uio)
742 TMPFS_VALIDATE_DIR(node);
743 MPASS(uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT);
746 * Return ENOENT if the current node is already removed.
748 TMPFS_ASSERT_LOCKED(node);
749 if (node->tn_dir.tn_parent == NULL) {
753 TMPFS_NODE_LOCK(node->tn_dir.tn_parent);
754 dent.d_fileno = node->tn_dir.tn_parent->tn_id;
755 TMPFS_NODE_UNLOCK(node->tn_dir.tn_parent);
757 dent.d_type = DT_DIR;
759 dent.d_name[0] = '.';
760 dent.d_name[1] = '.';
761 dent.d_name[2] = '\0';
762 dent.d_reclen = GENERIC_DIRSIZ(&dent);
764 if (dent.d_reclen > uio->uio_resid)
767 error = uiomove(&dent, dent.d_reclen, uio);
769 struct tmpfs_dirent *de;
771 de = TAILQ_FIRST(&node->tn_dir.tn_dirhead);
773 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
775 uio->uio_offset = tmpfs_dircookie(de);
779 node->tn_status |= TMPFS_NODE_ACCESSED;
784 /* --------------------------------------------------------------------- */
787 * Lookup a directory entry by its associated cookie.
789 struct tmpfs_dirent *
790 tmpfs_dir_lookupbycookie(struct tmpfs_node *node, off_t cookie)
792 struct tmpfs_dirent *de;
794 if (cookie == node->tn_dir.tn_readdir_lastn &&
795 node->tn_dir.tn_readdir_lastp != NULL) {
796 return node->tn_dir.tn_readdir_lastp;
799 TAILQ_FOREACH(de, &node->tn_dir.tn_dirhead, td_entries) {
800 if (tmpfs_dircookie(de) == cookie) {
808 /* --------------------------------------------------------------------- */
811 * Helper function for tmpfs_readdir. Returns as much directory entries
812 * as can fit in the uio space. The read starts at uio->uio_offset.
813 * The function returns 0 on success, -1 if there was not enough space
814 * in the uio structure to hold the directory entry or an appropriate
815 * error code if another error happens.
818 tmpfs_dir_getdents(struct tmpfs_node *node, struct uio *uio, off_t *cntp)
822 struct tmpfs_dirent *de;
824 TMPFS_VALIDATE_DIR(node);
826 /* Locate the first directory entry we have to return. We have cached
827 * the last readdir in the node, so use those values if appropriate.
828 * Otherwise do a linear scan to find the requested entry. */
829 startcookie = uio->uio_offset;
830 MPASS(startcookie != TMPFS_DIRCOOKIE_DOT);
831 MPASS(startcookie != TMPFS_DIRCOOKIE_DOTDOT);
832 if (startcookie == TMPFS_DIRCOOKIE_EOF) {
835 de = tmpfs_dir_lookupbycookie(node, startcookie);
841 /* Read as much entries as possible; i.e., until we reach the end of
842 * the directory or we exhaust uio space. */
846 /* Create a dirent structure representing the current
847 * tmpfs_node and fill it. */
848 if (de->td_node == NULL) {
852 d.d_fileno = de->td_node->tn_id;
853 switch (de->td_node->tn_type) {
883 panic("tmpfs_dir_getdents: type %p %d",
884 de->td_node, (int)de->td_node->tn_type);
887 d.d_namlen = de->td_namelen;
888 MPASS(de->td_namelen < sizeof(d.d_name));
889 (void)memcpy(d.d_name, de->td_name, de->td_namelen);
890 d.d_name[de->td_namelen] = '\0';
891 d.d_reclen = GENERIC_DIRSIZ(&d);
893 /* Stop reading if the directory entry we are treating is
894 * bigger than the amount of data that can be returned. */
895 if (d.d_reclen > uio->uio_resid) {
900 /* Copy the new dirent structure into the output buffer and
901 * advance pointers. */
902 error = uiomove(&d, d.d_reclen, uio);
905 de = TAILQ_NEXT(de, td_entries);
907 } while (error == 0 && uio->uio_resid > 0 && de != NULL);
909 /* Update the offset and cache. */
911 uio->uio_offset = TMPFS_DIRCOOKIE_EOF;
912 node->tn_dir.tn_readdir_lastn = 0;
913 node->tn_dir.tn_readdir_lastp = NULL;
915 node->tn_dir.tn_readdir_lastn = uio->uio_offset = tmpfs_dircookie(de);
916 node->tn_dir.tn_readdir_lastp = de;
919 node->tn_status |= TMPFS_NODE_ACCESSED;
924 tmpfs_dir_whiteout_add(struct vnode *dvp, struct componentname *cnp)
926 struct tmpfs_dirent *de;
929 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(dvp->v_mount), NULL,
930 cnp->cn_nameptr, cnp->cn_namelen, &de);
933 tmpfs_dir_attach(dvp, de);
938 tmpfs_dir_whiteout_remove(struct vnode *dvp, struct componentname *cnp)
940 struct tmpfs_dirent *de;
942 de = tmpfs_dir_lookup(VP_TO_TMPFS_DIR(dvp), NULL, cnp);
943 MPASS(de != NULL && de->td_node == NULL);
944 tmpfs_dir_detach(dvp, de);
945 tmpfs_free_dirent(VFS_TO_TMPFS(dvp->v_mount), de, TRUE);
948 /* --------------------------------------------------------------------- */
951 * Resizes the aobj associated with the regular file pointed to by 'vp' to the
952 * size 'newsize'. 'vp' must point to a vnode that represents a regular file.
953 * 'newsize' must be positive.
955 * Returns zero on success or an appropriate error code on failure.
958 tmpfs_reg_resize(struct vnode *vp, off_t newsize, boolean_t ignerr)
960 struct tmpfs_mount *tmp;
961 struct tmpfs_node *node;
964 vm_pindex_t idx, newpages, oldpages;
968 MPASS(vp->v_type == VREG);
971 node = VP_TO_TMPFS_NODE(vp);
972 uobj = node->tn_reg.tn_aobj;
973 tmp = VFS_TO_TMPFS(vp->v_mount);
976 * Convert the old and new sizes to the number of pages needed to
977 * store them. It may happen that we do not need to do anything
978 * because the last allocated page can accommodate the change on
981 oldsize = node->tn_size;
982 oldpages = OFF_TO_IDX(oldsize + PAGE_MASK);
983 MPASS(oldpages == uobj->size);
984 newpages = OFF_TO_IDX(newsize + PAGE_MASK);
985 if (newpages > oldpages &&
986 tmpfs_pages_check_avail(tmp, newpages - oldpages) == 0)
989 VM_OBJECT_LOCK(uobj);
990 if (newsize < oldsize) {
992 * Zero the truncated part of the last page.
994 base = newsize & PAGE_MASK;
996 idx = OFF_TO_IDX(newsize);
998 m = vm_page_lookup(uobj, idx);
1000 if ((m->oflags & VPO_BUSY) != 0 ||
1002 vm_page_sleep(m, "tmfssz");
1005 MPASS(m->valid == VM_PAGE_BITS_ALL);
1006 } else if (vm_pager_has_page(uobj, idx, NULL, NULL)) {
1007 m = vm_page_alloc(uobj, idx, VM_ALLOC_NORMAL);
1009 VM_OBJECT_UNLOCK(uobj);
1011 VM_OBJECT_LOCK(uobj);
1013 } else if (m->valid != VM_PAGE_BITS_ALL) {
1015 rv = vm_pager_get_pages(uobj, ma, 1, 0);
1016 m = vm_page_lookup(uobj, idx);
1018 /* A cached page was reactivated. */
1021 if (rv == VM_PAGER_OK) {
1022 vm_page_deactivate(m);
1031 VM_OBJECT_UNLOCK(uobj);
1037 pmap_zero_page_area(m, base, PAGE_SIZE - base);
1039 vm_pager_page_unswapped(m);
1044 * Release any swap space and free any whole pages.
1046 if (newpages < oldpages) {
1047 swap_pager_freespace(uobj, newpages, oldpages -
1049 vm_object_page_remove(uobj, newpages, 0, 0);
1052 uobj->size = newpages;
1053 VM_OBJECT_UNLOCK(uobj);
1056 tmp->tm_pages_used += (newpages - oldpages);
1059 node->tn_size = newsize;
1060 vnode_pager_setsize(vp, newsize);
1064 /* --------------------------------------------------------------------- */
1067 * Change flags of the given vnode.
1068 * Caller should execute tmpfs_update on vp after a successful execution.
1069 * The vnode must be locked on entry and remain locked on exit.
1072 tmpfs_chflags(struct vnode *vp, int flags, struct ucred *cred, struct thread *p)
1075 struct tmpfs_node *node;
1077 MPASS(VOP_ISLOCKED(vp));
1079 node = VP_TO_TMPFS_NODE(vp);
1081 /* Disallow this operation if the file system is mounted read-only. */
1082 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1086 * Callers may only modify the file flags on objects they
1087 * have VADMIN rights for.
1089 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1092 * Unprivileged processes are not permitted to unset system
1093 * flags, or modify flags if any system flags are set.
1095 if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) {
1097 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND)) {
1098 error = securelevel_gt(cred, 0);
1102 /* Snapshot flag cannot be set or cleared */
1103 if (((flags & SF_SNAPSHOT) != 0 &&
1104 (node->tn_flags & SF_SNAPSHOT) == 0) ||
1105 ((flags & SF_SNAPSHOT) == 0 &&
1106 (node->tn_flags & SF_SNAPSHOT) != 0))
1108 node->tn_flags = flags;
1111 & (SF_NOUNLINK | SF_IMMUTABLE | SF_APPEND) ||
1112 (flags & UF_SETTABLE) != flags)
1114 node->tn_flags &= SF_SETTABLE;
1115 node->tn_flags |= (flags & UF_SETTABLE);
1117 node->tn_status |= TMPFS_NODE_CHANGED;
1119 MPASS(VOP_ISLOCKED(vp));
1124 /* --------------------------------------------------------------------- */
1127 * Change access mode on the given vnode.
1128 * Caller should execute tmpfs_update on vp after a successful execution.
1129 * The vnode must be locked on entry and remain locked on exit.
1132 tmpfs_chmod(struct vnode *vp, mode_t mode, struct ucred *cred, struct thread *p)
1135 struct tmpfs_node *node;
1137 MPASS(VOP_ISLOCKED(vp));
1139 node = VP_TO_TMPFS_NODE(vp);
1141 /* Disallow this operation if the file system is mounted read-only. */
1142 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1145 /* Immutable or append-only files cannot be modified, either. */
1146 if (node->tn_flags & (IMMUTABLE | APPEND))
1150 * To modify the permissions on a file, must possess VADMIN
1153 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1157 * Privileged processes may set the sticky bit on non-directories,
1158 * as well as set the setgid bit on a file with a group that the
1159 * process is not a member of.
1161 if (vp->v_type != VDIR && (mode & S_ISTXT)) {
1162 if (priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0))
1165 if (!groupmember(node->tn_gid, cred) && (mode & S_ISGID)) {
1166 error = priv_check_cred(cred, PRIV_VFS_SETGID, 0);
1172 node->tn_mode &= ~ALLPERMS;
1173 node->tn_mode |= mode & ALLPERMS;
1175 node->tn_status |= TMPFS_NODE_CHANGED;
1177 MPASS(VOP_ISLOCKED(vp));
1182 /* --------------------------------------------------------------------- */
1185 * Change ownership of the given vnode. At least one of uid or gid must
1186 * be different than VNOVAL. If one is set to that value, the attribute
1188 * Caller should execute tmpfs_update on vp after a successful execution.
1189 * The vnode must be locked on entry and remain locked on exit.
1192 tmpfs_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred,
1196 struct tmpfs_node *node;
1200 MPASS(VOP_ISLOCKED(vp));
1202 node = VP_TO_TMPFS_NODE(vp);
1204 /* Assign default values if they are unknown. */
1205 MPASS(uid != VNOVAL || gid != VNOVAL);
1210 MPASS(uid != VNOVAL && gid != VNOVAL);
1212 /* Disallow this operation if the file system is mounted read-only. */
1213 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1216 /* Immutable or append-only files cannot be modified, either. */
1217 if (node->tn_flags & (IMMUTABLE | APPEND))
1221 * To modify the ownership of a file, must possess VADMIN for that
1224 if ((error = VOP_ACCESS(vp, VADMIN, cred, p)))
1228 * To change the owner of a file, or change the group of a file to a
1229 * group of which we are not a member, the caller must have
1232 if ((uid != node->tn_uid ||
1233 (gid != node->tn_gid && !groupmember(gid, cred))) &&
1234 (error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0)))
1237 ogid = node->tn_gid;
1238 ouid = node->tn_uid;
1243 node->tn_status |= TMPFS_NODE_CHANGED;
1245 if ((node->tn_mode & (S_ISUID | S_ISGID)) && (ouid != uid || ogid != gid)) {
1246 if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0))
1247 node->tn_mode &= ~(S_ISUID | S_ISGID);
1250 MPASS(VOP_ISLOCKED(vp));
1255 /* --------------------------------------------------------------------- */
1258 * Change size of the given vnode.
1259 * Caller should execute tmpfs_update on vp after a successful execution.
1260 * The vnode must be locked on entry and remain locked on exit.
1263 tmpfs_chsize(struct vnode *vp, u_quad_t size, struct ucred *cred,
1267 struct tmpfs_node *node;
1269 MPASS(VOP_ISLOCKED(vp));
1271 node = VP_TO_TMPFS_NODE(vp);
1273 /* Decide whether this is a valid operation based on the file type. */
1275 switch (vp->v_type) {
1280 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1289 /* Allow modifications of special files even if in the file
1290 * system is mounted read-only (we are not modifying the
1291 * files themselves, but the objects they represent). */
1295 /* Anything else is unsupported. */
1299 /* Immutable or append-only files cannot be modified, either. */
1300 if (node->tn_flags & (IMMUTABLE | APPEND))
1303 error = tmpfs_truncate(vp, size);
1304 /* tmpfs_truncate will raise the NOTE_EXTEND and NOTE_ATTRIB kevents
1305 * for us, as will update tn_status; no need to do that here. */
1307 MPASS(VOP_ISLOCKED(vp));
1312 /* --------------------------------------------------------------------- */
1315 * Change access and modification times of the given vnode.
1316 * Caller should execute tmpfs_update on vp after a successful execution.
1317 * The vnode must be locked on entry and remain locked on exit.
1320 tmpfs_chtimes(struct vnode *vp, struct timespec *atime, struct timespec *mtime,
1321 struct timespec *birthtime, int vaflags, struct ucred *cred, struct thread *l)
1324 struct tmpfs_node *node;
1326 MPASS(VOP_ISLOCKED(vp));
1328 node = VP_TO_TMPFS_NODE(vp);
1330 /* Disallow this operation if the file system is mounted read-only. */
1331 if (vp->v_mount->mnt_flag & MNT_RDONLY)
1334 /* Immutable or append-only files cannot be modified, either. */
1335 if (node->tn_flags & (IMMUTABLE | APPEND))
1338 /* Determine if the user have proper privilege to update time. */
1339 if (vaflags & VA_UTIMES_NULL) {
1340 error = VOP_ACCESS(vp, VADMIN, cred, l);
1342 error = VOP_ACCESS(vp, VWRITE, cred, l);
1344 error = VOP_ACCESS(vp, VADMIN, cred, l);
1348 if (atime->tv_sec != VNOVAL && atime->tv_nsec != VNOVAL)
1349 node->tn_status |= TMPFS_NODE_ACCESSED;
1351 if (mtime->tv_sec != VNOVAL && mtime->tv_nsec != VNOVAL)
1352 node->tn_status |= TMPFS_NODE_MODIFIED;
1354 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1355 node->tn_status |= TMPFS_NODE_MODIFIED;
1357 tmpfs_itimes(vp, atime, mtime);
1359 if (birthtime->tv_nsec != VNOVAL && birthtime->tv_nsec != VNOVAL)
1360 node->tn_birthtime = *birthtime;
1361 MPASS(VOP_ISLOCKED(vp));
1366 /* --------------------------------------------------------------------- */
1367 /* Sync timestamps */
1369 tmpfs_itimes(struct vnode *vp, const struct timespec *acc,
1370 const struct timespec *mod)
1372 struct tmpfs_node *node;
1373 struct timespec now;
1375 node = VP_TO_TMPFS_NODE(vp);
1377 if ((node->tn_status & (TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED |
1378 TMPFS_NODE_CHANGED)) == 0)
1381 vfs_timestamp(&now);
1382 if (node->tn_status & TMPFS_NODE_ACCESSED) {
1385 node->tn_atime = *acc;
1387 if (node->tn_status & TMPFS_NODE_MODIFIED) {
1390 node->tn_mtime = *mod;
1392 if (node->tn_status & TMPFS_NODE_CHANGED) {
1393 node->tn_ctime = now;
1396 ~(TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED | TMPFS_NODE_CHANGED);
1399 /* --------------------------------------------------------------------- */
1402 tmpfs_update(struct vnode *vp)
1405 tmpfs_itimes(vp, NULL, NULL);
1408 /* --------------------------------------------------------------------- */
1411 tmpfs_truncate(struct vnode *vp, off_t length)
1414 struct tmpfs_node *node;
1416 node = VP_TO_TMPFS_NODE(vp);
1423 if (node->tn_size == length) {
1428 if (length > VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize)
1431 error = tmpfs_reg_resize(vp, length, FALSE);
1433 node->tn_status |= TMPFS_NODE_CHANGED | TMPFS_NODE_MODIFIED;