1 /* $NetBSD: tmpfs_vfsops.c,v 1.10 2005/12/11 12:24:29 christos Exp $ */
4 * SPDX-License-Identifier: BSD-2-Clause-NetBSD
6 * Copyright (c) 2005 The NetBSD Foundation, Inc.
9 * This code is derived from software contributed to The NetBSD Foundation
10 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
22 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
24 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
25 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
26 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32 * POSSIBILITY OF SUCH DAMAGE.
36 * Efficient memory file system.
38 * tmpfs is a file system that uses FreeBSD's virtual memory
39 * sub-system to store file data and metadata in an efficient way.
40 * This means that it does not follow the structure of an on-disk file
41 * system because it simply does not need to. Instead, it uses
42 * memory-specific data structures and algorithms to automatically
43 * allocate and release resources.
46 #include "opt_tmpfs.h"
48 #include <sys/cdefs.h>
49 __FBSDID("$FreeBSD$");
51 #include <sys/param.h>
52 #include <sys/systm.h>
53 #include <sys/dirent.h>
54 #include <sys/limits.h>
56 #include <sys/mount.h>
57 #include <sys/mutex.h>
60 #include <sys/kernel.h>
61 #include <sys/rwlock.h>
64 #include <sys/sysctl.h>
65 #include <sys/vnode.h>
68 #include <vm/vm_param.h>
70 #include <vm/vm_extern.h>
71 #include <vm/vm_map.h>
72 #include <vm/vm_object.h>
73 #include <vm/vm_param.h>
75 #include <fs/tmpfs/tmpfs.h>
78 * Default permission for root node
80 #define TMPFS_DEFAULT_ROOT_MODE (S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
82 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
83 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
85 static int tmpfs_mount(struct mount *);
86 static int tmpfs_unmount(struct mount *, int);
87 static int tmpfs_root(struct mount *, int flags, struct vnode **);
88 static int tmpfs_fhtovp(struct mount *, struct fid *, int,
90 static int tmpfs_statfs(struct mount *, struct statfs *);
92 static const char *tmpfs_opts[] = {
93 "from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export",
94 "union", "nonc", "nomtime", NULL
97 static const char *tmpfs_updateopts[] = {
98 "from", "export", "nomtime", "size", NULL
102 * Handle updates of time from writes to mmaped regions, if allowed.
103 * Use MNT_VNODE_FOREACH_ALL instead of MNT_VNODE_FOREACH_LAZY, since
104 * unmap of the tmpfs-backed vnode does not call vinactive(), due to
105 * vm object type is OBJT_SWAP. If lazy, only handle delayed update
106 * of mtime due to the writes to mapped files.
109 tmpfs_update_mtime(struct mount *mp, bool lazy)
111 struct vnode *vp, *mvp;
112 struct vm_object *obj;
114 if (VFS_TO_TMPFS(mp)->tm_nomtime)
116 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
117 if (vp->v_type != VREG) {
122 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
123 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
126 * In lazy case, do unlocked read, avoid taking vnode
127 * lock if not needed. Lost update will be handled on
129 * For non-lazy case, we must flush all pending
130 * metadata changes now.
132 if (!lazy || obj->generation != obj->cleangeneration) {
133 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK,
136 tmpfs_check_mtime(vp);
147 struct tmpfs_check_rw_maps_arg {
152 tmpfs_check_rw_maps_cb(struct mount *mp __unused, vm_map_t map __unused,
153 vm_map_entry_t entry __unused, void *arg)
155 struct tmpfs_check_rw_maps_arg *a;
163 * Revoke write permissions from all mappings of regular files
164 * belonging to the specified tmpfs mount.
167 tmpfs_revoke_rw_maps_cb(struct mount *mp __unused, vm_map_t map,
168 vm_map_entry_t entry, void *arg __unused)
172 * XXXKIB: might be invalidate the mapping
173 * instead ? The process is not going to be
176 entry->max_protection &= ~VM_PROT_WRITE;
177 if ((entry->protection & VM_PROT_WRITE) != 0) {
178 entry->protection &= ~VM_PROT_WRITE;
179 pmap_protect(map->pmap, entry->start, entry->end,
186 tmpfs_all_rw_maps(struct mount *mp, bool (*cb)(struct mount *mp, vm_map_t,
187 vm_map_entry_t, void *), void *cb_arg)
192 vm_map_entry_t entry;
199 sx_slock(&allproc_lock);
202 FOREACH_PROC_IN_SYSTEM(p) {
204 if (p->p_state != PRS_NORMAL || (p->p_flag & (P_INEXEC |
205 P_SYSTEM | P_WEXIT)) != 0) {
209 vm = vmspace_acquire_ref(p);
216 sx_sunlock(&allproc_lock);
221 vm_map_wait_busy(map);
222 VM_MAP_ENTRY_FOREACH(entry, map) {
223 if ((entry->eflags & (MAP_ENTRY_GUARD |
224 MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_COW)) != 0 ||
225 (entry->max_protection & VM_PROT_WRITE) == 0)
227 object = entry->object.vm_object;
228 if (object == NULL || object->type != OBJT_SWAP ||
229 (object->flags & OBJ_TMPFS_NODE) == 0)
232 * No need to dig into shadow chain, mapping
233 * of the object not at top is readonly.
236 VM_OBJECT_RLOCK(object);
237 if (object->type == OBJT_DEAD) {
238 VM_OBJECT_RUNLOCK(object);
241 MPASS(object->ref_count > 1);
242 if ((object->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) !=
243 (OBJ_TMPFS_NODE | OBJ_TMPFS)) {
244 VM_OBJECT_RUNLOCK(object);
247 vp = object->un_pager.swp.swp_tmpfs;
248 if (vp->v_mount != mp) {
249 VM_OBJECT_RUNLOCK(object);
253 terminate = cb(mp, map, entry, cb_arg);
254 VM_OBJECT_RUNLOCK(object);
261 sx_slock(&allproc_lock);
266 if (!terminate && gen != allproc_gen)
268 sx_sunlock(&allproc_lock);
272 tmpfs_check_rw_maps(struct mount *mp)
274 struct tmpfs_check_rw_maps_arg ca;
277 tmpfs_all_rw_maps(mp, tmpfs_check_rw_maps_cb, &ca);
282 tmpfs_rw_to_ro(struct mount *mp)
287 forced = (mp->mnt_flag & MNT_FORCE) != 0;
288 flags = WRITECLOSE | (forced ? FORCECLOSE : 0);
290 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
292 error = vfs_write_suspend_umnt(mp);
295 if (!forced && tmpfs_check_rw_maps(mp)) {
299 VFS_TO_TMPFS(mp)->tm_ronly = 1;
301 mp->mnt_flag |= MNT_RDONLY;
304 tmpfs_all_rw_maps(mp, tmpfs_revoke_rw_maps_cb, NULL);
305 tmpfs_update_mtime(mp, false);
306 error = vflush(mp, 0, flags, curthread);
308 VFS_TO_TMPFS(mp)->tm_ronly = 0;
310 mp->mnt_flag &= ~MNT_RDONLY;
314 if (!tmpfs_check_rw_maps(mp))
318 vfs_write_resume(mp, 0);
323 tmpfs_mount(struct mount *mp)
325 const size_t nodes_per_page = howmany(PAGE_SIZE,
326 sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node));
327 struct tmpfs_mount *tmp;
328 struct tmpfs_node *root;
333 off_t nodes_max, size_max, maxfilesize;
335 /* Root node attributes. */
342 if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
345 if (mp->mnt_flag & MNT_UPDATE) {
346 /* Only support update mounts for certain options. */
347 if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0)
349 tmp = VFS_TO_TMPFS(mp);
350 if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) == 0) {
352 * On-the-fly resizing is not supported (yet). We still
353 * need to have "size" listed as "supported", otherwise
354 * trying to update fs that is listed in fstab with size
355 * parameter, say trying to change rw to ro or vice
356 * versa, would cause vfs_filteropt() to bail.
358 if (size_max != tmp->tm_size_max)
361 if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) &&
364 return (tmpfs_rw_to_ro(mp));
365 } else if (!vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) &&
370 mp->mnt_flag &= ~MNT_RDONLY;
373 tmp->tm_nomtime = vfs_getopt(mp->mnt_optnew, "nomtime", NULL,
376 if ((mp->mnt_flag & MNT_UNION) == 0) {
377 mp->mnt_kern_flag |= MNTK_FPLOOKUP;
379 mp->mnt_kern_flag &= ~MNTK_FPLOOKUP;
385 vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
386 error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
387 VOP_UNLOCK(mp->mnt_vnodecovered);
391 if (mp->mnt_cred->cr_ruid != 0 ||
392 vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
393 root_gid = va.va_gid;
394 if (mp->mnt_cred->cr_ruid != 0 ||
395 vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
396 root_uid = va.va_uid;
397 if (mp->mnt_cred->cr_ruid != 0 ||
398 vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
399 root_mode = va.va_mode;
400 if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0)
402 if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0)
404 if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0)
406 nonc = vfs_getopt(mp->mnt_optnew, "nonc", NULL, NULL) == 0;
407 nomtime = vfs_getopt(mp->mnt_optnew, "nomtime", NULL, NULL) == 0;
409 /* Do not allow mounts if we do not have enough memory to preserve
410 * the minimum reserved pages. */
411 if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED)
414 /* Get the maximum number of memory pages this file system is
415 * allowed to use, based on the maximum size the user passed in
416 * the mount structure. A value of zero is treated as if the
417 * maximum available space was requested. */
418 if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE ||
419 (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX))
422 size_max = roundup(size_max, PAGE_SIZE);
423 pages = howmany(size_max, PAGE_SIZE);
427 if (nodes_max <= 3) {
428 if (pages < INT_MAX / nodes_per_page)
429 nodes_max = pages * nodes_per_page;
433 if (nodes_max > INT_MAX)
435 MPASS(nodes_max >= 3);
437 /* Allocate the tmpfs mount structure and fill it. */
438 tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
439 M_TMPFSMNT, M_WAITOK | M_ZERO);
441 mtx_init(&tmp->tm_allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
442 tmp->tm_nodes_max = nodes_max;
443 tmp->tm_nodes_inuse = 0;
444 tmp->tm_refcount = 1;
445 tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX;
446 LIST_INIT(&tmp->tm_nodes_used);
448 tmp->tm_size_max = size_max;
449 tmp->tm_pages_max = pages;
450 tmp->tm_pages_used = 0;
451 new_unrhdr64(&tmp->tm_ino_unr, 2);
452 tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
454 tmp->tm_nomtime = nomtime;
456 /* Allocate the root node. */
457 error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, root_gid,
458 root_mode & ALLPERMS, NULL, NULL, VNOVAL, &root);
460 if (error != 0 || root == NULL) {
461 free(tmp, M_TMPFSMNT);
464 KASSERT(root->tn_id == 2,
465 ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
469 mp->mnt_flag |= MNT_LOCAL;
470 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
471 MNTK_TEXT_REFS | MNTK_NOMSYNC;
472 if (!nonc && (mp->mnt_flag & MNT_UNION) == 0)
473 mp->mnt_kern_flag |= MNTK_FPLOOKUP;
477 mp->mnt_stat.f_namemax = MAXNAMLEN;
479 vfs_mountedfrom(mp, "tmpfs");
486 tmpfs_unmount(struct mount *mp, int mntflags)
488 struct tmpfs_mount *tmp;
489 struct tmpfs_node *node;
492 flags = (mntflags & MNT_FORCE) != 0 ? FORCECLOSE : 0;
493 tmp = VFS_TO_TMPFS(mp);
496 error = vfs_write_suspend_umnt(mp);
500 * At this point, nodes cannot be destroyed by any other
501 * thread because write suspension is started.
505 error = vflush(mp, 0, flags, curthread);
507 vfs_write_resume(mp, VR_START_WRITE);
511 if (mp->mnt_nvnodelistsize == 0) {
516 if ((mntflags & MNT_FORCE) == 0) {
517 vfs_write_resume(mp, VR_START_WRITE);
523 while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) {
524 TMPFS_NODE_LOCK(node);
525 if (node->tn_type == VDIR)
526 tmpfs_dir_destroy(tmp, node);
527 if (tmpfs_free_node_locked(tmp, node, true))
530 TMPFS_NODE_UNLOCK(node);
535 vfs_write_resume(mp, VR_START_WRITE);
538 mp->mnt_flag &= ~MNT_LOCAL;
545 tmpfs_free_tmp(struct tmpfs_mount *tmp)
548 MPASS(tmp->tm_refcount > 0);
550 if (tmp->tm_refcount > 0) {
556 mtx_destroy(&tmp->tm_allnode_lock);
557 MPASS(tmp->tm_pages_used == 0);
558 MPASS(tmp->tm_nodes_inuse == 0);
560 free(tmp, M_TMPFSMNT);
564 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp)
568 error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp);
570 (*vpp)->v_vflag |= VV_ROOT;
575 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags,
578 struct tmpfs_fid_data tfd;
579 struct tmpfs_mount *tmp;
580 struct tmpfs_node *node;
583 if (fhp->fid_len != sizeof(tfd))
587 * Copy from fid_data onto the stack to avoid unaligned pointer use.
588 * See the comment in sys/mount.h on struct fid for details.
590 memcpy(&tfd, fhp->fid_data, fhp->fid_len);
592 tmp = VFS_TO_TMPFS(mp);
594 if (tfd.tfd_id >= tmp->tm_nodes_max)
598 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
599 if (node->tn_id == tfd.tfd_id &&
600 node->tn_gen == tfd.tfd_gen) {
601 tmpfs_ref_node(node);
608 error = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp);
609 tmpfs_free_node(tmp, node);
617 tmpfs_statfs(struct mount *mp, struct statfs *sbp)
619 struct tmpfs_mount *tmp;
622 tmp = VFS_TO_TMPFS(mp);
624 sbp->f_iosize = PAGE_SIZE;
625 sbp->f_bsize = PAGE_SIZE;
627 used = tmpfs_pages_used(tmp);
628 if (tmp->tm_pages_max != ULONG_MAX)
629 sbp->f_blocks = tmp->tm_pages_max;
631 sbp->f_blocks = used + tmpfs_mem_avail();
632 if (sbp->f_blocks <= used)
635 sbp->f_bavail = sbp->f_blocks - used;
636 sbp->f_bfree = sbp->f_bavail;
637 used = tmp->tm_nodes_inuse;
638 sbp->f_files = tmp->tm_nodes_max;
639 if (sbp->f_files <= used)
642 sbp->f_ffree = sbp->f_files - used;
643 /* sbp->f_owner = tmp->tn_uid; */
649 tmpfs_sync(struct mount *mp, int waitfor)
652 if (waitfor == MNT_SUSPEND) {
654 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
656 } else if (waitfor == MNT_LAZY) {
657 tmpfs_update_mtime(mp, true);
663 tmpfs_init(struct vfsconf *conf)
670 tmpfs_uninit(struct vfsconf *conf)
677 * tmpfs vfs operations.
679 struct vfsops tmpfs_vfsops = {
680 .vfs_mount = tmpfs_mount,
681 .vfs_unmount = tmpfs_unmount,
682 .vfs_root = vfs_cache_root,
683 .vfs_cachedroot = tmpfs_root,
684 .vfs_statfs = tmpfs_statfs,
685 .vfs_fhtovp = tmpfs_fhtovp,
686 .vfs_sync = tmpfs_sync,
687 .vfs_init = tmpfs_init,
688 .vfs_uninit = tmpfs_uninit,
690 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL);