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 *);
91 static void tmpfs_susp_clean(struct mount *);
93 static const char *tmpfs_opts[] = {
94 "from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export",
98 static const char *tmpfs_updateopts[] = {
99 "from", "export", "size", NULL
103 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
105 struct tmpfs_node *node = (struct tmpfs_node *)mem;
112 node->tn_vnode = NULL;
113 node->tn_vpstate = 0;
119 tmpfs_node_dtor(void *mem, int size, void *arg)
121 struct tmpfs_node *node = (struct tmpfs_node *)mem;
122 node->tn_type = VNON;
126 tmpfs_node_init(void *mem, int size, int flags)
128 struct tmpfs_node *node = (struct tmpfs_node *)mem;
131 mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF);
132 node->tn_gen = arc4random();
138 tmpfs_node_fini(void *mem, int size)
140 struct tmpfs_node *node = (struct tmpfs_node *)mem;
142 mtx_destroy(&node->tn_interlock);
146 * Handle updates of time from writes to mmaped regions. Use
147 * MNT_VNODE_FOREACH_ALL instead of MNT_VNODE_FOREACH_ACTIVE, since
148 * unmap of the tmpfs-backed vnode does not call vinactive(), due to
149 * vm object type is OBJT_SWAP.
150 * If lazy, only handle delayed update of mtime due to the writes to
154 tmpfs_update_mtime(struct mount *mp, bool lazy)
156 struct vnode *vp, *mvp;
157 struct vm_object *obj;
159 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
160 if (vp->v_type != VREG) {
165 KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
166 (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
169 * In lazy case, do unlocked read, avoid taking vnode
170 * lock if not needed. Lost update will be handled on
172 * For non-lazy case, we must flush all pending
173 * metadata changes now.
175 if (!lazy || (obj->flags & OBJ_TMPFS_DIRTY) != 0) {
176 if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK,
179 tmpfs_check_mtime(vp);
190 struct tmpfs_check_rw_maps_arg {
195 tmpfs_check_rw_maps_cb(struct mount *mp __unused, vm_map_t map __unused,
196 vm_map_entry_t entry __unused, void *arg)
198 struct tmpfs_check_rw_maps_arg *a;
206 * Revoke write permissions from all mappings of regular files
207 * belonging to the specified tmpfs mount.
210 tmpfs_revoke_rw_maps_cb(struct mount *mp __unused, vm_map_t map,
211 vm_map_entry_t entry, void *arg __unused)
215 * XXXKIB: might be invalidate the mapping
216 * instead ? The process is not going to be
219 entry->max_protection &= ~VM_PROT_WRITE;
220 if ((entry->protection & VM_PROT_WRITE) != 0) {
221 entry->protection &= ~VM_PROT_WRITE;
222 pmap_protect(map->pmap, entry->start, entry->end,
229 tmpfs_all_rw_maps(struct mount *mp, bool (*cb)(struct mount *mp, vm_map_t,
230 vm_map_entry_t, void *), void *cb_arg)
235 vm_map_entry_t entry;
242 sx_slock(&allproc_lock);
245 FOREACH_PROC_IN_SYSTEM(p) {
247 if (p->p_state != PRS_NORMAL || (p->p_flag & (P_INEXEC |
248 P_SYSTEM | P_WEXIT)) != 0) {
252 vm = vmspace_acquire_ref(p);
259 sx_sunlock(&allproc_lock);
264 vm_map_wait_busy(map);
265 VM_MAP_ENTRY_FOREACH(entry, map) {
266 if ((entry->eflags & (MAP_ENTRY_GUARD |
267 MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_COW)) != 0 ||
268 (entry->max_protection & VM_PROT_WRITE) == 0)
270 object = entry->object.vm_object;
271 if (object == NULL || object->type != OBJT_SWAP ||
272 (object->flags & OBJ_TMPFS_NODE) == 0)
275 * No need to dig into shadow chain, mapping
276 * of the object not at top is readonly.
279 VM_OBJECT_RLOCK(object);
280 if (object->type == OBJT_DEAD) {
281 VM_OBJECT_RUNLOCK(object);
284 MPASS(object->ref_count > 1);
285 if ((object->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) !=
286 (OBJ_TMPFS_NODE | OBJ_TMPFS)) {
287 VM_OBJECT_RUNLOCK(object);
290 vp = object->un_pager.swp.swp_tmpfs;
291 if (vp->v_mount != mp) {
292 VM_OBJECT_RUNLOCK(object);
296 terminate = cb(mp, map, entry, cb_arg);
297 VM_OBJECT_RUNLOCK(object);
304 sx_slock(&allproc_lock);
309 if (!terminate && gen != allproc_gen)
311 sx_sunlock(&allproc_lock);
315 tmpfs_check_rw_maps(struct mount *mp)
317 struct tmpfs_check_rw_maps_arg ca;
320 tmpfs_all_rw_maps(mp, tmpfs_check_rw_maps_cb, &ca);
325 tmpfs_rw_to_ro(struct mount *mp)
330 forced = (mp->mnt_flag & MNT_FORCE) != 0;
331 flags = WRITECLOSE | (forced ? FORCECLOSE : 0);
333 if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
335 error = vfs_write_suspend_umnt(mp);
338 if (!forced && tmpfs_check_rw_maps(mp)) {
342 VFS_TO_TMPFS(mp)->tm_ronly = 1;
344 mp->mnt_flag |= MNT_RDONLY;
347 tmpfs_all_rw_maps(mp, tmpfs_revoke_rw_maps_cb, NULL);
348 tmpfs_update_mtime(mp, false);
349 error = vflush(mp, 0, flags, curthread);
351 VFS_TO_TMPFS(mp)->tm_ronly = 0;
353 mp->mnt_flag &= ~MNT_RDONLY;
357 if (!tmpfs_check_rw_maps(mp))
361 vfs_write_resume(mp, 0);
366 tmpfs_mount(struct mount *mp)
368 const size_t nodes_per_page = howmany(PAGE_SIZE,
369 sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node));
370 struct tmpfs_mount *tmp;
371 struct tmpfs_node *root;
376 off_t nodes_max, size_max, maxfilesize;
378 /* Root node attributes. */
385 if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
388 if (mp->mnt_flag & MNT_UPDATE) {
389 /* Only support update mounts for certain options. */
390 if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0)
392 if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) == 0) {
394 * On-the-fly resizing is not supported (yet). We still
395 * need to have "size" listed as "supported", otherwise
396 * trying to update fs that is listed in fstab with size
397 * parameter, say trying to change rw to ro or vice
398 * versa, would cause vfs_filteropt() to bail.
400 if (size_max != VFS_TO_TMPFS(mp)->tm_size_max)
403 if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) &&
404 !(VFS_TO_TMPFS(mp)->tm_ronly)) {
406 return (tmpfs_rw_to_ro(mp));
407 } else if (!vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) &&
408 VFS_TO_TMPFS(mp)->tm_ronly) {
410 VFS_TO_TMPFS(mp)->tm_ronly = 0;
412 mp->mnt_flag &= ~MNT_RDONLY;
418 vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
419 error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
420 VOP_UNLOCK(mp->mnt_vnodecovered, 0);
424 if (mp->mnt_cred->cr_ruid != 0 ||
425 vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
426 root_gid = va.va_gid;
427 if (mp->mnt_cred->cr_ruid != 0 ||
428 vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
429 root_uid = va.va_uid;
430 if (mp->mnt_cred->cr_ruid != 0 ||
431 vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
432 root_mode = va.va_mode;
433 if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0)
435 if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0)
437 if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0)
439 nonc = vfs_getopt(mp->mnt_optnew, "nonc", NULL, NULL) == 0;
441 /* Do not allow mounts if we do not have enough memory to preserve
442 * the minimum reserved pages. */
443 if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED)
446 /* Get the maximum number of memory pages this file system is
447 * allowed to use, based on the maximum size the user passed in
448 * the mount structure. A value of zero is treated as if the
449 * maximum available space was requested. */
450 if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE ||
451 (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX))
454 size_max = roundup(size_max, PAGE_SIZE);
455 pages = howmany(size_max, PAGE_SIZE);
459 if (nodes_max <= 3) {
460 if (pages < INT_MAX / nodes_per_page)
461 nodes_max = pages * nodes_per_page;
465 if (nodes_max > INT_MAX)
467 MPASS(nodes_max >= 3);
469 /* Allocate the tmpfs mount structure and fill it. */
470 tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
471 M_TMPFSMNT, M_WAITOK | M_ZERO);
473 mtx_init(&tmp->tm_allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
474 tmp->tm_nodes_max = nodes_max;
475 tmp->tm_nodes_inuse = 0;
476 tmp->tm_refcount = 1;
477 tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX;
478 LIST_INIT(&tmp->tm_nodes_used);
480 tmp->tm_size_max = size_max;
481 tmp->tm_pages_max = pages;
482 tmp->tm_pages_used = 0;
483 new_unrhdr64(&tmp->tm_ino_unr, 2);
484 tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent",
485 sizeof(struct tmpfs_dirent), NULL, NULL, NULL, NULL,
487 tmp->tm_node_pool = uma_zcreate("TMPFS node",
488 sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor,
489 tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0);
490 tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
493 /* Allocate the root node. */
494 error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, root_gid,
495 root_mode & ALLPERMS, NULL, NULL, VNOVAL, &root);
497 if (error != 0 || root == NULL) {
498 uma_zdestroy(tmp->tm_node_pool);
499 uma_zdestroy(tmp->tm_dirent_pool);
500 free(tmp, M_TMPFSMNT);
503 KASSERT(root->tn_id == 2,
504 ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
508 mp->mnt_flag |= MNT_LOCAL;
509 mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED |
510 MNTK_TEXT_REFS | MNTK_NOMSYNC;
514 mp->mnt_stat.f_namemax = MAXNAMLEN;
516 vfs_mountedfrom(mp, "tmpfs");
523 tmpfs_unmount(struct mount *mp, int mntflags)
525 struct tmpfs_mount *tmp;
526 struct tmpfs_node *node;
529 flags = (mntflags & MNT_FORCE) != 0 ? FORCECLOSE : 0;
530 tmp = VFS_TO_TMPFS(mp);
533 error = vfs_write_suspend_umnt(mp);
537 * At this point, nodes cannot be destroyed by any other
538 * thread because write suspension is started.
542 error = vflush(mp, 0, flags, curthread);
544 vfs_write_resume(mp, VR_START_WRITE);
548 if (mp->mnt_nvnodelistsize == 0) {
553 if ((mntflags & MNT_FORCE) == 0) {
554 vfs_write_resume(mp, VR_START_WRITE);
560 while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) {
561 TMPFS_NODE_LOCK(node);
562 if (node->tn_type == VDIR)
563 tmpfs_dir_destroy(tmp, node);
564 if (tmpfs_free_node_locked(tmp, node, true))
567 TMPFS_NODE_UNLOCK(node);
572 vfs_write_resume(mp, VR_START_WRITE);
575 mp->mnt_flag &= ~MNT_LOCAL;
582 tmpfs_free_tmp(struct tmpfs_mount *tmp)
585 MPASS(tmp->tm_refcount > 0);
587 if (tmp->tm_refcount > 0) {
593 uma_zdestroy(tmp->tm_dirent_pool);
594 uma_zdestroy(tmp->tm_node_pool);
596 mtx_destroy(&tmp->tm_allnode_lock);
597 MPASS(tmp->tm_pages_used == 0);
598 MPASS(tmp->tm_nodes_inuse == 0);
600 free(tmp, M_TMPFSMNT);
604 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp)
608 error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp);
610 (*vpp)->v_vflag |= VV_ROOT;
615 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags,
618 struct tmpfs_fid *tfhp;
619 struct tmpfs_mount *tmp;
620 struct tmpfs_node *node;
623 tmp = VFS_TO_TMPFS(mp);
625 tfhp = (struct tmpfs_fid *)fhp;
626 if (tfhp->tf_len != sizeof(struct tmpfs_fid))
629 if (tfhp->tf_id >= tmp->tm_nodes_max)
633 LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
634 if (node->tn_id == tfhp->tf_id &&
635 node->tn_gen == tfhp->tf_gen) {
636 tmpfs_ref_node(node);
643 error = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp);
644 tmpfs_free_node(tmp, node);
652 tmpfs_statfs(struct mount *mp, struct statfs *sbp)
654 struct tmpfs_mount *tmp;
657 tmp = VFS_TO_TMPFS(mp);
659 sbp->f_iosize = PAGE_SIZE;
660 sbp->f_bsize = PAGE_SIZE;
662 used = tmpfs_pages_used(tmp);
663 if (tmp->tm_pages_max != ULONG_MAX)
664 sbp->f_blocks = tmp->tm_pages_max;
666 sbp->f_blocks = used + tmpfs_mem_avail();
667 if (sbp->f_blocks <= used)
670 sbp->f_bavail = sbp->f_blocks - used;
671 sbp->f_bfree = sbp->f_bavail;
672 used = tmp->tm_nodes_inuse;
673 sbp->f_files = tmp->tm_nodes_max;
674 if (sbp->f_files <= used)
677 sbp->f_ffree = sbp->f_files - used;
678 /* sbp->f_owner = tmp->tn_uid; */
684 tmpfs_sync(struct mount *mp, int waitfor)
687 if (waitfor == MNT_SUSPEND) {
689 mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
691 } else if (waitfor == MNT_LAZY) {
692 tmpfs_update_mtime(mp, true);
698 * The presence of a susp_clean method tells the VFS to track writes.
701 tmpfs_susp_clean(struct mount *mp __unused)
706 * tmpfs vfs operations.
709 struct vfsops tmpfs_vfsops = {
710 .vfs_mount = tmpfs_mount,
711 .vfs_unmount = tmpfs_unmount,
712 .vfs_root = vfs_cache_root,
713 .vfs_cachedroot = tmpfs_root,
714 .vfs_statfs = tmpfs_statfs,
715 .vfs_fhtovp = tmpfs_fhtovp,
716 .vfs_sync = tmpfs_sync,
717 .vfs_susp_clean = tmpfs_susp_clean,
719 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL);