xref: /freebsd/sys/fs/tmpfs/tmpfs_vfsops.c (revision 49ba44e765f46fa747b69640e09ceb9303f53b09)
1 /*	$NetBSD: tmpfs_vfsops.c,v 1.10 2005/12/11 12:24:29 christos Exp $	*/
2 
3 /*-
4  * SPDX-License-Identifier: BSD-2-Clause-NetBSD
5  *
6  * Copyright (c) 2005 The NetBSD Foundation, Inc.
7  * All rights reserved.
8  *
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
11  * 2005 program.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
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.
21  *
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.
33  */
34 
35 /*
36  * Efficient memory file system.
37  *
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.
44  */
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47 
48 #include <sys/param.h>
49 #include <sys/limits.h>
50 #include <sys/lock.h>
51 #include <sys/mutex.h>
52 #include <sys/proc.h>
53 #include <sys/jail.h>
54 #include <sys/kernel.h>
55 #include <sys/rwlock.h>
56 #include <sys/stat.h>
57 #include <sys/systm.h>
58 #include <sys/sysctl.h>
59 
60 #include <vm/vm.h>
61 #include <vm/vm_object.h>
62 #include <vm/vm_param.h>
63 
64 #include <fs/tmpfs/tmpfs.h>
65 
66 /*
67  * Default permission for root node
68  */
69 #define TMPFS_DEFAULT_ROOT_MODE	(S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
70 
71 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
72 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
73 
74 static int	tmpfs_mount(struct mount *);
75 static int	tmpfs_unmount(struct mount *, int);
76 static int	tmpfs_root(struct mount *, int flags, struct vnode **);
77 static int	tmpfs_fhtovp(struct mount *, struct fid *, int,
78 		    struct vnode **);
79 static int	tmpfs_statfs(struct mount *, struct statfs *);
80 static void	tmpfs_susp_clean(struct mount *);
81 
82 static const char *tmpfs_opts[] = {
83 	"from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export",
84 	"union", "nonc", NULL
85 };
86 
87 static const char *tmpfs_updateopts[] = {
88 	"from", "export", NULL
89 };
90 
91 static int
92 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
93 {
94 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
95 
96 	node->tn_gen++;
97 	node->tn_size = 0;
98 	node->tn_status = 0;
99 	node->tn_flags = 0;
100 	node->tn_links = 0;
101 	node->tn_vnode = NULL;
102 	node->tn_vpstate = 0;
103 
104 	return (0);
105 }
106 
107 static void
108 tmpfs_node_dtor(void *mem, int size, void *arg)
109 {
110 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
111 	node->tn_type = VNON;
112 }
113 
114 static int
115 tmpfs_node_init(void *mem, int size, int flags)
116 {
117 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
118 	node->tn_id = 0;
119 
120 	mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF);
121 	node->tn_gen = arc4random();
122 
123 	return (0);
124 }
125 
126 static void
127 tmpfs_node_fini(void *mem, int size)
128 {
129 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
130 
131 	mtx_destroy(&node->tn_interlock);
132 }
133 
134 static int
135 tmpfs_mount(struct mount *mp)
136 {
137 	const size_t nodes_per_page = howmany(PAGE_SIZE,
138 	    sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node));
139 	struct tmpfs_mount *tmp;
140 	struct tmpfs_node *root;
141 	struct thread *td = curthread;
142 	int error;
143 	bool nonc;
144 	/* Size counters. */
145 	u_quad_t pages;
146 	off_t nodes_max, size_max, maxfilesize;
147 
148 	/* Root node attributes. */
149 	uid_t root_uid;
150 	gid_t root_gid;
151 	mode_t root_mode;
152 
153 	struct vattr va;
154 
155 	if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_TMPFS))
156 		return (EPERM);
157 
158 	if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
159 		return (EINVAL);
160 
161 	if (mp->mnt_flag & MNT_UPDATE) {
162 		/* Only support update mounts for certain options. */
163 		if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0)
164 			return (EOPNOTSUPP);
165 		if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) !=
166 		    ((struct tmpfs_mount *)mp->mnt_data)->tm_ronly)
167 			return (EOPNOTSUPP);
168 		return (0);
169 	}
170 
171 	vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
172 	error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
173 	VOP_UNLOCK(mp->mnt_vnodecovered, 0);
174 	if (error)
175 		return (error);
176 
177 	if (mp->mnt_cred->cr_ruid != 0 ||
178 	    vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
179 		root_gid = va.va_gid;
180 	if (mp->mnt_cred->cr_ruid != 0 ||
181 	    vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
182 		root_uid = va.va_uid;
183 	if (mp->mnt_cred->cr_ruid != 0 ||
184 	    vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
185 		root_mode = va.va_mode;
186 	if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0)
187 		nodes_max = 0;
188 	if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0)
189 		size_max = 0;
190 	if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0)
191 		maxfilesize = 0;
192 	nonc = vfs_getopt(mp->mnt_optnew, "nonc", NULL, NULL) == 0;
193 
194 	/* Do not allow mounts if we do not have enough memory to preserve
195 	 * the minimum reserved pages. */
196 	if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED)
197 		return (ENOSPC);
198 
199 	/* Get the maximum number of memory pages this file system is
200 	 * allowed to use, based on the maximum size the user passed in
201 	 * the mount structure.  A value of zero is treated as if the
202 	 * maximum available space was requested. */
203 	if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE ||
204 	    (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX))
205 		pages = SIZE_MAX;
206 	else {
207 		size_max = roundup(size_max, PAGE_SIZE);
208 		pages = howmany(size_max, PAGE_SIZE);
209 	}
210 	MPASS(pages > 0);
211 
212 	if (nodes_max <= 3) {
213 		if (pages < INT_MAX / nodes_per_page)
214 			nodes_max = pages * nodes_per_page;
215 		else
216 			nodes_max = INT_MAX;
217 	}
218 	if (nodes_max > INT_MAX)
219 		nodes_max = INT_MAX;
220 	MPASS(nodes_max >= 3);
221 
222 	/* Allocate the tmpfs mount structure and fill it. */
223 	tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
224 	    M_TMPFSMNT, M_WAITOK | M_ZERO);
225 
226 	mtx_init(&tmp->tm_allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
227 	tmp->tm_nodes_max = nodes_max;
228 	tmp->tm_nodes_inuse = 0;
229 	tmp->tm_refcount = 1;
230 	tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX;
231 	LIST_INIT(&tmp->tm_nodes_used);
232 
233 	tmp->tm_pages_max = pages;
234 	tmp->tm_pages_used = 0;
235 	tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->tm_allnode_lock);
236 	tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent",
237 	    sizeof(struct tmpfs_dirent), NULL, NULL, NULL, NULL,
238 	    UMA_ALIGN_PTR, 0);
239 	tmp->tm_node_pool = uma_zcreate("TMPFS node",
240 	    sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor,
241 	    tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0);
242 	tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
243 	tmp->tm_nonc = nonc;
244 
245 	/* Allocate the root node. */
246 	error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, root_gid,
247 	    root_mode & ALLPERMS, NULL, NULL, VNOVAL, &root);
248 
249 	if (error != 0 || root == NULL) {
250 		uma_zdestroy(tmp->tm_node_pool);
251 		uma_zdestroy(tmp->tm_dirent_pool);
252 		delete_unrhdr(tmp->tm_ino_unr);
253 		free(tmp, M_TMPFSMNT);
254 		return (error);
255 	}
256 	KASSERT(root->tn_id == 2,
257 	    ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
258 	tmp->tm_root = root;
259 
260 	MNT_ILOCK(mp);
261 	mp->mnt_flag |= MNT_LOCAL;
262 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED;
263 	MNT_IUNLOCK(mp);
264 
265 	mp->mnt_data = tmp;
266 	mp->mnt_stat.f_namemax = MAXNAMLEN;
267 	vfs_getnewfsid(mp);
268 	vfs_mountedfrom(mp, "tmpfs");
269 
270 	return 0;
271 }
272 
273 /* ARGSUSED2 */
274 static int
275 tmpfs_unmount(struct mount *mp, int mntflags)
276 {
277 	struct tmpfs_mount *tmp;
278 	struct tmpfs_node *node;
279 	int error, flags;
280 
281 	flags = (mntflags & MNT_FORCE) != 0 ? FORCECLOSE : 0;
282 	tmp = VFS_TO_TMPFS(mp);
283 
284 	/* Stop writers */
285 	error = vfs_write_suspend_umnt(mp);
286 	if (error != 0)
287 		return (error);
288 	/*
289 	 * At this point, nodes cannot be destroyed by any other
290 	 * thread because write suspension is started.
291 	 */
292 
293 	for (;;) {
294 		error = vflush(mp, 0, flags, curthread);
295 		if (error != 0) {
296 			vfs_write_resume(mp, VR_START_WRITE);
297 			return (error);
298 		}
299 		MNT_ILOCK(mp);
300 		if (mp->mnt_nvnodelistsize == 0) {
301 			MNT_IUNLOCK(mp);
302 			break;
303 		}
304 		MNT_IUNLOCK(mp);
305 		if ((mntflags & MNT_FORCE) == 0) {
306 			vfs_write_resume(mp, VR_START_WRITE);
307 			return (EBUSY);
308 		}
309 	}
310 
311 	TMPFS_LOCK(tmp);
312 	while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) {
313 		TMPFS_NODE_LOCK(node);
314 		if (node->tn_type == VDIR)
315 			tmpfs_dir_destroy(tmp, node);
316 		if (tmpfs_free_node_locked(tmp, node, true))
317 			TMPFS_LOCK(tmp);
318 		else
319 			TMPFS_NODE_UNLOCK(node);
320 	}
321 
322 	mp->mnt_data = NULL;
323 	tmpfs_free_tmp(tmp);
324 	vfs_write_resume(mp, VR_START_WRITE);
325 
326 	MNT_ILOCK(mp);
327 	mp->mnt_flag &= ~MNT_LOCAL;
328 	MNT_IUNLOCK(mp);
329 
330 	return (0);
331 }
332 
333 void
334 tmpfs_free_tmp(struct tmpfs_mount *tmp)
335 {
336 
337 	MPASS(tmp->tm_refcount > 0);
338 	tmp->tm_refcount--;
339 	if (tmp->tm_refcount > 0) {
340 		TMPFS_UNLOCK(tmp);
341 		return;
342 	}
343 	TMPFS_UNLOCK(tmp);
344 
345 	uma_zdestroy(tmp->tm_dirent_pool);
346 	uma_zdestroy(tmp->tm_node_pool);
347 	clear_unrhdr(tmp->tm_ino_unr);
348 	delete_unrhdr(tmp->tm_ino_unr);
349 
350 	mtx_destroy(&tmp->tm_allnode_lock);
351 	MPASS(tmp->tm_pages_used == 0);
352 	MPASS(tmp->tm_nodes_inuse == 0);
353 
354 	free(tmp, M_TMPFSMNT);
355 }
356 
357 static int
358 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp)
359 {
360 	int error;
361 
362 	error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp);
363 	if (error == 0)
364 		(*vpp)->v_vflag |= VV_ROOT;
365 	return (error);
366 }
367 
368 static int
369 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags,
370     struct vnode **vpp)
371 {
372 	struct tmpfs_fid *tfhp;
373 	struct tmpfs_mount *tmp;
374 	struct tmpfs_node *node;
375 	int error;
376 
377 	tmp = VFS_TO_TMPFS(mp);
378 
379 	tfhp = (struct tmpfs_fid *)fhp;
380 	if (tfhp->tf_len != sizeof(struct tmpfs_fid))
381 		return (EINVAL);
382 
383 	if (tfhp->tf_id >= tmp->tm_nodes_max)
384 		return (EINVAL);
385 
386 	TMPFS_LOCK(tmp);
387 	LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
388 		if (node->tn_id == tfhp->tf_id &&
389 		    node->tn_gen == tfhp->tf_gen) {
390 			tmpfs_ref_node(node);
391 			break;
392 		}
393 	}
394 	TMPFS_UNLOCK(tmp);
395 
396 	if (node != NULL) {
397 		error = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp);
398 		tmpfs_free_node(tmp, node);
399 	} else
400 		error = EINVAL;
401 	return (error);
402 }
403 
404 /* ARGSUSED2 */
405 static int
406 tmpfs_statfs(struct mount *mp, struct statfs *sbp)
407 {
408 	struct tmpfs_mount *tmp;
409 	size_t used;
410 
411 	tmp = VFS_TO_TMPFS(mp);
412 
413 	sbp->f_iosize = PAGE_SIZE;
414 	sbp->f_bsize = PAGE_SIZE;
415 
416 	used = tmpfs_pages_used(tmp);
417 	if (tmp->tm_pages_max != ULONG_MAX)
418 		 sbp->f_blocks = tmp->tm_pages_max;
419 	else
420 		 sbp->f_blocks = used + tmpfs_mem_avail();
421 	if (sbp->f_blocks <= used)
422 		sbp->f_bavail = 0;
423 	else
424 		sbp->f_bavail = sbp->f_blocks - used;
425 	sbp->f_bfree = sbp->f_bavail;
426 	used = tmp->tm_nodes_inuse;
427 	sbp->f_files = tmp->tm_nodes_max;
428 	if (sbp->f_files <= used)
429 		sbp->f_ffree = 0;
430 	else
431 		sbp->f_ffree = sbp->f_files - used;
432 	/* sbp->f_owner = tmp->tn_uid; */
433 
434 	return 0;
435 }
436 
437 static int
438 tmpfs_sync(struct mount *mp, int waitfor)
439 {
440 	struct vnode *vp, *mvp;
441 	struct vm_object *obj;
442 
443 	if (waitfor == MNT_SUSPEND) {
444 		MNT_ILOCK(mp);
445 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
446 		MNT_IUNLOCK(mp);
447 	} else if (waitfor == MNT_LAZY) {
448 		/*
449 		 * Handle lazy updates of mtime from writes to mmaped
450 		 * regions.  Use MNT_VNODE_FOREACH_ALL instead of
451 		 * MNT_VNODE_FOREACH_ACTIVE, since unmap of the
452 		 * tmpfs-backed vnode does not call vinactive(), due
453 		 * to vm object type is OBJT_SWAP.
454 		 */
455 		MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
456 			if (vp->v_type != VREG) {
457 				VI_UNLOCK(vp);
458 				continue;
459 			}
460 			obj = vp->v_object;
461 			KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
462 			    (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
463 
464 			/*
465 			 * Unlocked read, avoid taking vnode lock if
466 			 * not needed.  Lost update will be handled on
467 			 * the next call.
468 			 */
469 			if ((obj->flags & OBJ_TMPFS_DIRTY) == 0) {
470 				VI_UNLOCK(vp);
471 				continue;
472 			}
473 			if (vget(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
474 			    curthread) != 0)
475 				continue;
476 			tmpfs_check_mtime(vp);
477 			vput(vp);
478 		}
479 	}
480 	return (0);
481 }
482 
483 /*
484  * The presence of a susp_clean method tells the VFS to track writes.
485  */
486 static void
487 tmpfs_susp_clean(struct mount *mp __unused)
488 {
489 }
490 
491 /*
492  * tmpfs vfs operations.
493  */
494 
495 struct vfsops tmpfs_vfsops = {
496 	.vfs_mount =			tmpfs_mount,
497 	.vfs_unmount =			tmpfs_unmount,
498 	.vfs_root =			tmpfs_root,
499 	.vfs_statfs =			tmpfs_statfs,
500 	.vfs_fhtovp =			tmpfs_fhtovp,
501 	.vfs_sync =			tmpfs_sync,
502 	.vfs_susp_clean =		tmpfs_susp_clean,
503 };
504 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL);
505