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.
47 #include "opt_tmpfs.h"
49 #include <sys/cdefs.h>
50 __FBSDID("$FreeBSD$");
52 #include <sys/param.h>
53 #include <sys/systm.h>
54 #include <sys/dirent.h>
56 #include <sys/limits.h>
58 #include <sys/mount.h>
59 #include <sys/mutex.h>
62 #include <sys/kernel.h>
63 #include <sys/rwlock.h>
66 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
70 #include <vm/vm_param.h>
72 #include <vm/vm_extern.h>
73 #include <vm/vm_map.h>
74 #include <vm/vm_object.h>
75 #include <vm/vm_param.h>
77 #include <fs/tmpfs/tmpfs.h>
80 * Default permission for root node
82 #define TMPFS_DEFAULT_ROOT_MODE (S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
84 static MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
85 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
87 static int tmpfs_mount(struct mount *);
88 static int tmpfs_unmount(struct mount *, int);
89 static int tmpfs_root(struct mount *, int flags, struct vnode **);
90 static int tmpfs_fhtovp(struct mount *, struct fid *, int,
92 static int tmpfs_statfs(struct mount *, struct statfs *);
94 static const char *tmpfs_opts[] = {
95 "from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export",
96 "union", "nonc", "nomtime", NULL
99 static const char *tmpfs_updateopts[] = {
100 "from", "export", "nomtime", "size", NULL
104 tmpfs_update_mtime_lazy_filter(struct vnode *vp, void *arg)
106 struct vm_object *obj;
108 if (vp->v_type != VREG)
111 obj = atomic_load_ptr(&vp->v_object);
115 return (vm_object_mightbedirty_(obj));
119 tmpfs_update_mtime_lazy(struct mount *mp)
121 struct vnode *vp, *mvp;
123 MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, tmpfs_update_mtime_lazy_filter, NULL) {
124 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK) != 0)
126 tmpfs_check_mtime(vp);
132 tmpfs_update_mtime_all(struct mount *mp)
134 struct vnode *vp, *mvp;
136 if (VFS_TO_TMPFS(mp)->tm_nomtime)
138 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
139 if (vp->v_type != VREG) {
143 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK) != 0)
145 tmpfs_check_mtime(vp);
151 struct tmpfs_check_rw_maps_arg {
156 tmpfs_check_rw_maps_cb(struct mount *mp __unused, vm_map_t map __unused,
157 vm_map_entry_t entry __unused, void *arg)
159 struct tmpfs_check_rw_maps_arg *a;
167 * Revoke write permissions from all mappings of regular files
168 * belonging to the specified tmpfs mount.
171 tmpfs_revoke_rw_maps_cb(struct mount *mp __unused, vm_map_t map,
172 vm_map_entry_t entry, void *arg __unused)
176 * XXXKIB: might be invalidate the mapping
177 * instead ? The process is not going to be
180 entry->max_protection &= ~VM_PROT_WRITE;
181 if ((entry->protection & VM_PROT_WRITE) != 0) {
182 entry->protection &= ~VM_PROT_WRITE;
183 pmap_protect(map->pmap, entry->start, entry->end,
190 tmpfs_all_rw_maps(struct mount *mp, bool (*cb)(struct mount *mp, vm_map_t,
191 vm_map_entry_t, void *), void *cb_arg)
196 vm_map_entry_t entry;
203 sx_slock(&allproc_lock);
206 FOREACH_PROC_IN_SYSTEM(p) {
208 if (p->p_state != PRS_NORMAL || (p->p_flag & (P_INEXEC |
209 P_SYSTEM | P_WEXIT)) != 0) {
213 vm = vmspace_acquire_ref(p);
220 sx_sunlock(&allproc_lock);
225 vm_map_wait_busy(map);
226 VM_MAP_ENTRY_FOREACH(entry, map) {
227 if ((entry->eflags & (MAP_ENTRY_GUARD |
228 MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_COW)) != 0 ||
229 (entry->max_protection & VM_PROT_WRITE) == 0)
231 object = entry->object.vm_object;
232 if (object == NULL || object->type != tmpfs_pager_type)
235 * No need to dig into shadow chain, mapping
236 * of the object not at top is readonly.
239 VM_OBJECT_RLOCK(object);
240 if (object->type == OBJT_DEAD) {
241 VM_OBJECT_RUNLOCK(object);
244 MPASS(object->ref_count > 1);
245 if ((object->flags & OBJ_TMPFS) == 0) {
246 VM_OBJECT_RUNLOCK(object);
249 vp = VM_TO_TMPFS_VP(object);
250 if (vp->v_mount != mp) {
251 VM_OBJECT_RUNLOCK(object);
255 terminate = cb(mp, map, entry, cb_arg);
256 VM_OBJECT_RUNLOCK(object);
263 sx_slock(&allproc_lock);
268 if (!terminate && gen != allproc_gen)
270 sx_sunlock(&allproc_lock);
274 tmpfs_check_rw_maps(struct mount *mp)
276 struct tmpfs_check_rw_maps_arg ca;
279 tmpfs_all_rw_maps(mp, tmpfs_check_rw_maps_cb, &ca);
284 tmpfs_rw_to_ro(struct mount *mp)
289 forced = (mp->mnt_flag & MNT_FORCE) != 0;
290 flags = WRITECLOSE | (forced ? FORCECLOSE : 0);
292 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
294 error = vfs_write_suspend_umnt(mp);
297 if (!forced && tmpfs_check_rw_maps(mp)) {
301 VFS_TO_TMPFS(mp)->tm_ronly = 1;
303 mp->mnt_flag |= MNT_RDONLY;
306 tmpfs_all_rw_maps(mp, tmpfs_revoke_rw_maps_cb, NULL);
307 tmpfs_update_mtime_all(mp);
308 error = vflush(mp, 0, flags, curthread);
310 VFS_TO_TMPFS(mp)->tm_ronly = 0;
312 mp->mnt_flag &= ~MNT_RDONLY;
316 if (!tmpfs_check_rw_maps(mp))
320 vfs_write_resume(mp, 0);
325 tmpfs_mount(struct mount *mp)
327 const size_t nodes_per_page = howmany(PAGE_SIZE,
328 sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node));
329 struct tmpfs_mount *tmp;
330 struct tmpfs_node *root;
335 off_t nodes_max, size_max, maxfilesize;
337 /* Root node attributes. */
344 if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
347 if (mp->mnt_flag & MNT_UPDATE) {
348 /* Only support update mounts for certain options. */
349 if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0)
351 tmp = VFS_TO_TMPFS(mp);
352 if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) == 0) {
354 * On-the-fly resizing is not supported (yet). We still
355 * need to have "size" listed as "supported", otherwise
356 * trying to update fs that is listed in fstab with size
357 * parameter, say trying to change rw to ro or vice
358 * versa, would cause vfs_filteropt() to bail.
360 if (size_max != tmp->tm_size_max)
363 if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) &&
366 return (tmpfs_rw_to_ro(mp));
367 } else if (!vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) &&
372 mp->mnt_flag &= ~MNT_RDONLY;
375 tmp->tm_nomtime = vfs_getopt(mp->mnt_optnew, "nomtime", NULL,
378 if ((mp->mnt_flag & MNT_UNION) == 0) {
379 mp->mnt_kern_flag |= MNTK_FPLOOKUP;
381 mp->mnt_kern_flag &= ~MNTK_FPLOOKUP;
387 vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
388 error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
389 VOP_UNLOCK(mp->mnt_vnodecovered);
393 if (mp->mnt_cred->cr_ruid != 0 ||
394 vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
395 root_gid = va.va_gid;
396 if (mp->mnt_cred->cr_ruid != 0 ||
397 vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
398 root_uid = va.va_uid;
399 if (mp->mnt_cred->cr_ruid != 0 ||
400 vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
401 root_mode = va.va_mode;
402 if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0)
404 if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0)
406 if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0)
408 nonc = vfs_getopt(mp->mnt_optnew, "nonc", NULL, NULL) == 0;
409 nomtime = vfs_getopt(mp->mnt_optnew, "nomtime", NULL, NULL) == 0;
411 /* Do not allow mounts if we do not have enough memory to preserve
412 * the minimum reserved pages. */
413 if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED)
416 /* Get the maximum number of memory pages this file system is
417 * allowed to use, based on the maximum size the user passed in
418 * the mount structure. A value of zero is treated as if the
419 * maximum available space was requested. */
420 if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE ||
421 (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX))
424 size_max = roundup(size_max, PAGE_SIZE);
425 pages = howmany(size_max, PAGE_SIZE);
429 if (nodes_max <= 3) {
430 if (pages < INT_MAX / nodes_per_page)
431 nodes_max = pages * nodes_per_page;
435 if (nodes_max > INT_MAX)
437 MPASS(nodes_max >= 3);
439 /* Allocate the tmpfs mount structure and fill it. */
440 tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
441 M_TMPFSMNT, M_WAITOK | M_ZERO);
443 mtx_init(&tmp->tm_allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
444 tmp->tm_nodes_max = nodes_max;
445 tmp->tm_nodes_inuse = 0;
446 tmp->tm_refcount = 1;
447 tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX;
448 LIST_INIT(&tmp->tm_nodes_used);
450 tmp->tm_size_max = size_max;
451 tmp->tm_pages_max = pages;
452 tmp->tm_pages_used = 0;
453 new_unrhdr64(&tmp->tm_ino_unr, 2);
454 tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
456 tmp->tm_nomtime = nomtime;
458 /* Allocate the root node. */
459 error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, root_gid,
460 root_mode & ALLPERMS, NULL, NULL, VNOVAL, &root);
462 if (error != 0 || root == NULL) {
463 free(tmp, M_TMPFSMNT);
466 KASSERT(root->tn_id == 2,
467 ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
471 mp->mnt_flag |= MNT_LOCAL;
472 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
473 MNTK_TEXT_REFS | MNTK_NOMSYNC;
474 if (!nonc && (mp->mnt_flag & MNT_UNION) == 0)
475 mp->mnt_kern_flag |= MNTK_FPLOOKUP;
479 mp->mnt_stat.f_namemax = MAXNAMLEN;
481 vfs_mountedfrom(mp, "tmpfs");
488 tmpfs_unmount(struct mount *mp, int mntflags)
490 struct tmpfs_mount *tmp;
491 struct tmpfs_node *node;
494 flags = (mntflags & MNT_FORCE) != 0 ? FORCECLOSE : 0;
495 tmp = VFS_TO_TMPFS(mp);
498 error = vfs_write_suspend_umnt(mp);
502 * At this point, nodes cannot be destroyed by any other
503 * thread because write suspension is started.
507 error = vflush(mp, 0, flags, curthread);
509 vfs_write_resume(mp, VR_START_WRITE);
513 if (mp->mnt_nvnodelistsize == 0) {
518 if ((mntflags & MNT_FORCE) == 0) {
519 vfs_write_resume(mp, VR_START_WRITE);
525 while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) {
526 TMPFS_NODE_LOCK(node);
527 if (node->tn_type == VDIR)
528 tmpfs_dir_destroy(tmp, node);
529 if (tmpfs_free_node_locked(tmp, node, true))
532 TMPFS_NODE_UNLOCK(node);
537 vfs_write_resume(mp, VR_START_WRITE);
540 mp->mnt_flag &= ~MNT_LOCAL;
547 tmpfs_free_tmp(struct tmpfs_mount *tmp)
549 TMPFS_MP_ASSERT_LOCKED(tmp);
550 MPASS(tmp->tm_refcount > 0);
553 if (tmp->tm_refcount > 0) {
559 mtx_destroy(&tmp->tm_allnode_lock);
560 MPASS(tmp->tm_pages_used == 0);
561 MPASS(tmp->tm_nodes_inuse == 0);
563 free(tmp, M_TMPFSMNT);
567 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp)
571 error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp);
573 (*vpp)->v_vflag |= VV_ROOT;
578 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags,
581 struct tmpfs_fid_data tfd;
582 struct tmpfs_mount *tmp;
583 struct tmpfs_node *node;
586 if (fhp->fid_len != sizeof(tfd))
590 * Copy from fid_data onto the stack to avoid unaligned pointer use.
591 * See the comment in sys/mount.h on struct fid for details.
593 memcpy(&tfd, fhp->fid_data, fhp->fid_len);
595 tmp = VFS_TO_TMPFS(mp);
597 if (tfd.tfd_id >= tmp->tm_nodes_max)
601 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
602 if (node->tn_id == tfd.tfd_id &&
603 node->tn_gen == tfd.tfd_gen) {
604 tmpfs_ref_node(node);
611 error = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp);
612 tmpfs_free_node(tmp, node);
620 tmpfs_statfs(struct mount *mp, struct statfs *sbp)
622 struct tmpfs_mount *tmp;
625 tmp = VFS_TO_TMPFS(mp);
627 sbp->f_iosize = PAGE_SIZE;
628 sbp->f_bsize = PAGE_SIZE;
630 used = tmpfs_pages_used(tmp);
631 if (tmp->tm_pages_max != ULONG_MAX)
632 sbp->f_blocks = tmp->tm_pages_max;
634 sbp->f_blocks = used + tmpfs_mem_avail();
635 if (sbp->f_blocks <= used)
638 sbp->f_bavail = sbp->f_blocks - used;
639 sbp->f_bfree = sbp->f_bavail;
640 used = tmp->tm_nodes_inuse;
641 sbp->f_files = tmp->tm_nodes_max;
642 if (sbp->f_files <= used)
645 sbp->f_ffree = sbp->f_files - used;
646 /* sbp->f_owner = tmp->tn_uid; */
652 tmpfs_sync(struct mount *mp, int waitfor)
655 if (waitfor == MNT_SUSPEND) {
657 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
659 } else if (waitfor == MNT_LAZY) {
660 tmpfs_update_mtime_lazy(mp);
666 tmpfs_init(struct vfsconf *conf)
670 res = tmpfs_subr_init();
673 memcpy(&tmpfs_fnops, &vnops, sizeof(struct fileops));
674 tmpfs_fnops.fo_close = tmpfs_fo_close;
679 tmpfs_uninit(struct vfsconf *conf)
686 * tmpfs vfs operations.
688 struct vfsops tmpfs_vfsops = {
689 .vfs_mount = tmpfs_mount,
690 .vfs_unmount = tmpfs_unmount,
691 .vfs_root = vfs_cache_root,
692 .vfs_cachedroot = tmpfs_root,
693 .vfs_statfs = tmpfs_statfs,
694 .vfs_fhtovp = tmpfs_fhtovp,
695 .vfs_sync = tmpfs_sync,
696 .vfs_init = tmpfs_init,
697 .vfs_uninit = tmpfs_uninit,
699 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL);
705 db_print_tmpfs(struct mount *mp, struct tmpfs_mount *tmp)
707 db_printf("mp %p (%s) tmp %p\n", mp,
708 mp->mnt_stat.f_mntonname, tmp);
710 "\tsize max %ju pages max %lu pages used %lu\n"
711 "\tinodes max %ju inodes inuse %ju refcount %ju\n"
712 "\tmaxfilesize %ju r%c %snamecache %smtime\n",
713 (uintmax_t)tmp->tm_size_max, tmp->tm_pages_max, tmp->tm_pages_used,
714 (uintmax_t)tmp->tm_nodes_max, (uintmax_t)tmp->tm_nodes_inuse,
715 (uintmax_t)tmp->tm_refcount, (uintmax_t)tmp->tm_maxfilesize,
716 tmp->tm_ronly ? 'o' : 'w', tmp->tm_nonc ? "no" : "",
717 tmp->tm_nomtime ? "no" : "");
720 DB_SHOW_COMMAND(tmpfs, db_show_tmpfs)
723 struct tmpfs_mount *tmp;
726 mp = (struct mount *)addr;
727 tmp = VFS_TO_TMPFS(mp);
728 db_print_tmpfs(mp, tmp);
732 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
733 if (strcmp(mp->mnt_stat.f_fstypename, tmpfs_vfsconf.vfc_name) ==
735 tmp = VFS_TO_TMPFS(mp);
736 db_print_tmpfs(mp, tmp);