xref: /freebsd/sys/fs/tmpfs/tmpfs_vfsops.c (revision 4928135658a9d0eaee37003df6137ab363fcb0b4)
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/dirent.h>
50 #include <sys/limits.h>
51 #include <sys/lock.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
54 #include <sys/proc.h>
55 #include <sys/jail.h>
56 #include <sys/kernel.h>
57 #include <sys/rwlock.h>
58 #include <sys/stat.h>
59 #include <sys/systm.h>
60 #include <sys/sysctl.h>
61 #include <sys/vnode.h>
62 
63 #include <vm/vm.h>
64 #include <vm/vm_object.h>
65 #include <vm/vm_param.h>
66 
67 #include <fs/tmpfs/tmpfs.h>
68 
69 /*
70  * Default permission for root node
71  */
72 #define TMPFS_DEFAULT_ROOT_MODE	(S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
73 
74 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
75 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
76 
77 static int	tmpfs_mount(struct mount *);
78 static int	tmpfs_unmount(struct mount *, int);
79 static int	tmpfs_root(struct mount *, int flags, struct vnode **);
80 static int	tmpfs_fhtovp(struct mount *, struct fid *, int,
81 		    struct vnode **);
82 static int	tmpfs_statfs(struct mount *, struct statfs *);
83 static void	tmpfs_susp_clean(struct mount *);
84 
85 static const char *tmpfs_opts[] = {
86 	"from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export",
87 	"union", "nonc", NULL
88 };
89 
90 static const char *tmpfs_updateopts[] = {
91 	"from", "export", NULL
92 };
93 
94 static int
95 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
96 {
97 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
98 
99 	node->tn_gen++;
100 	node->tn_size = 0;
101 	node->tn_status = 0;
102 	node->tn_flags = 0;
103 	node->tn_links = 0;
104 	node->tn_vnode = NULL;
105 	node->tn_vpstate = 0;
106 
107 	return (0);
108 }
109 
110 static void
111 tmpfs_node_dtor(void *mem, int size, void *arg)
112 {
113 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
114 	node->tn_type = VNON;
115 }
116 
117 static int
118 tmpfs_node_init(void *mem, int size, int flags)
119 {
120 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
121 	node->tn_id = 0;
122 
123 	mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF);
124 	node->tn_gen = arc4random();
125 
126 	return (0);
127 }
128 
129 static void
130 tmpfs_node_fini(void *mem, int size)
131 {
132 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
133 
134 	mtx_destroy(&node->tn_interlock);
135 }
136 
137 static int
138 tmpfs_mount(struct mount *mp)
139 {
140 	const size_t nodes_per_page = howmany(PAGE_SIZE,
141 	    sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node));
142 	struct tmpfs_mount *tmp;
143 	struct tmpfs_node *root;
144 	int error;
145 	bool nonc;
146 	/* Size counters. */
147 	u_quad_t pages;
148 	off_t nodes_max, size_max, maxfilesize;
149 
150 	/* Root node attributes. */
151 	uid_t root_uid;
152 	gid_t root_gid;
153 	mode_t root_mode;
154 
155 	struct vattr va;
156 
157 	if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
158 		return (EINVAL);
159 
160 	if (mp->mnt_flag & MNT_UPDATE) {
161 		/* Only support update mounts for certain options. */
162 		if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0)
163 			return (EOPNOTSUPP);
164 		if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) !=
165 		    ((struct tmpfs_mount *)mp->mnt_data)->tm_ronly)
166 			return (EOPNOTSUPP);
167 		return (0);
168 	}
169 
170 	vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
171 	error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
172 	VOP_UNLOCK(mp->mnt_vnodecovered, 0);
173 	if (error)
174 		return (error);
175 
176 	if (mp->mnt_cred->cr_ruid != 0 ||
177 	    vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
178 		root_gid = va.va_gid;
179 	if (mp->mnt_cred->cr_ruid != 0 ||
180 	    vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
181 		root_uid = va.va_uid;
182 	if (mp->mnt_cred->cr_ruid != 0 ||
183 	    vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
184 		root_mode = va.va_mode;
185 	if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0)
186 		nodes_max = 0;
187 	if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0)
188 		size_max = 0;
189 	if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0)
190 		maxfilesize = 0;
191 	nonc = vfs_getopt(mp->mnt_optnew, "nonc", NULL, NULL) == 0;
192 
193 	/* Do not allow mounts if we do not have enough memory to preserve
194 	 * the minimum reserved pages. */
195 	if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED)
196 		return (ENOSPC);
197 
198 	/* Get the maximum number of memory pages this file system is
199 	 * allowed to use, based on the maximum size the user passed in
200 	 * the mount structure.  A value of zero is treated as if the
201 	 * maximum available space was requested. */
202 	if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE ||
203 	    (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX))
204 		pages = SIZE_MAX;
205 	else {
206 		size_max = roundup(size_max, PAGE_SIZE);
207 		pages = howmany(size_max, PAGE_SIZE);
208 	}
209 	MPASS(pages > 0);
210 
211 	if (nodes_max <= 3) {
212 		if (pages < INT_MAX / nodes_per_page)
213 			nodes_max = pages * nodes_per_page;
214 		else
215 			nodes_max = INT_MAX;
216 	}
217 	if (nodes_max > INT_MAX)
218 		nodes_max = INT_MAX;
219 	MPASS(nodes_max >= 3);
220 
221 	/* Allocate the tmpfs mount structure and fill it. */
222 	tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
223 	    M_TMPFSMNT, M_WAITOK | M_ZERO);
224 
225 	mtx_init(&tmp->tm_allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
226 	tmp->tm_nodes_max = nodes_max;
227 	tmp->tm_nodes_inuse = 0;
228 	tmp->tm_refcount = 1;
229 	tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX;
230 	LIST_INIT(&tmp->tm_nodes_used);
231 
232 	tmp->tm_pages_max = pages;
233 	tmp->tm_pages_used = 0;
234 	tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->tm_allnode_lock);
235 	tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent",
236 	    sizeof(struct tmpfs_dirent), NULL, NULL, NULL, NULL,
237 	    UMA_ALIGN_PTR, 0);
238 	tmp->tm_node_pool = uma_zcreate("TMPFS node",
239 	    sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor,
240 	    tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0);
241 	tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
242 	tmp->tm_nonc = nonc;
243 
244 	/* Allocate the root node. */
245 	error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, root_gid,
246 	    root_mode & ALLPERMS, NULL, NULL, VNOVAL, &root);
247 
248 	if (error != 0 || root == NULL) {
249 		uma_zdestroy(tmp->tm_node_pool);
250 		uma_zdestroy(tmp->tm_dirent_pool);
251 		delete_unrhdr(tmp->tm_ino_unr);
252 		free(tmp, M_TMPFSMNT);
253 		return (error);
254 	}
255 	KASSERT(root->tn_id == 2,
256 	    ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
257 	tmp->tm_root = root;
258 
259 	MNT_ILOCK(mp);
260 	mp->mnt_flag |= MNT_LOCAL;
261 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED;
262 	MNT_IUNLOCK(mp);
263 
264 	mp->mnt_data = tmp;
265 	mp->mnt_stat.f_namemax = MAXNAMLEN;
266 	vfs_getnewfsid(mp);
267 	vfs_mountedfrom(mp, "tmpfs");
268 
269 	return 0;
270 }
271 
272 /* ARGSUSED2 */
273 static int
274 tmpfs_unmount(struct mount *mp, int mntflags)
275 {
276 	struct tmpfs_mount *tmp;
277 	struct tmpfs_node *node;
278 	int error, flags;
279 
280 	flags = (mntflags & MNT_FORCE) != 0 ? FORCECLOSE : 0;
281 	tmp = VFS_TO_TMPFS(mp);
282 
283 	/* Stop writers */
284 	error = vfs_write_suspend_umnt(mp);
285 	if (error != 0)
286 		return (error);
287 	/*
288 	 * At this point, nodes cannot be destroyed by any other
289 	 * thread because write suspension is started.
290 	 */
291 
292 	for (;;) {
293 		error = vflush(mp, 0, flags, curthread);
294 		if (error != 0) {
295 			vfs_write_resume(mp, VR_START_WRITE);
296 			return (error);
297 		}
298 		MNT_ILOCK(mp);
299 		if (mp->mnt_nvnodelistsize == 0) {
300 			MNT_IUNLOCK(mp);
301 			break;
302 		}
303 		MNT_IUNLOCK(mp);
304 		if ((mntflags & MNT_FORCE) == 0) {
305 			vfs_write_resume(mp, VR_START_WRITE);
306 			return (EBUSY);
307 		}
308 	}
309 
310 	TMPFS_LOCK(tmp);
311 	while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) {
312 		TMPFS_NODE_LOCK(node);
313 		if (node->tn_type == VDIR)
314 			tmpfs_dir_destroy(tmp, node);
315 		if (tmpfs_free_node_locked(tmp, node, true))
316 			TMPFS_LOCK(tmp);
317 		else
318 			TMPFS_NODE_UNLOCK(node);
319 	}
320 
321 	mp->mnt_data = NULL;
322 	tmpfs_free_tmp(tmp);
323 	vfs_write_resume(mp, VR_START_WRITE);
324 
325 	MNT_ILOCK(mp);
326 	mp->mnt_flag &= ~MNT_LOCAL;
327 	MNT_IUNLOCK(mp);
328 
329 	return (0);
330 }
331 
332 void
333 tmpfs_free_tmp(struct tmpfs_mount *tmp)
334 {
335 
336 	MPASS(tmp->tm_refcount > 0);
337 	tmp->tm_refcount--;
338 	if (tmp->tm_refcount > 0) {
339 		TMPFS_UNLOCK(tmp);
340 		return;
341 	}
342 	TMPFS_UNLOCK(tmp);
343 
344 	uma_zdestroy(tmp->tm_dirent_pool);
345 	uma_zdestroy(tmp->tm_node_pool);
346 	clear_unrhdr(tmp->tm_ino_unr);
347 	delete_unrhdr(tmp->tm_ino_unr);
348 
349 	mtx_destroy(&tmp->tm_allnode_lock);
350 	MPASS(tmp->tm_pages_used == 0);
351 	MPASS(tmp->tm_nodes_inuse == 0);
352 
353 	free(tmp, M_TMPFSMNT);
354 }
355 
356 static int
357 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp)
358 {
359 	int error;
360 
361 	error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp);
362 	if (error == 0)
363 		(*vpp)->v_vflag |= VV_ROOT;
364 	return (error);
365 }
366 
367 static int
368 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags,
369     struct vnode **vpp)
370 {
371 	struct tmpfs_fid *tfhp;
372 	struct tmpfs_mount *tmp;
373 	struct tmpfs_node *node;
374 	int error;
375 
376 	tmp = VFS_TO_TMPFS(mp);
377 
378 	tfhp = (struct tmpfs_fid *)fhp;
379 	if (tfhp->tf_len != sizeof(struct tmpfs_fid))
380 		return (EINVAL);
381 
382 	if (tfhp->tf_id >= tmp->tm_nodes_max)
383 		return (EINVAL);
384 
385 	TMPFS_LOCK(tmp);
386 	LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
387 		if (node->tn_id == tfhp->tf_id &&
388 		    node->tn_gen == tfhp->tf_gen) {
389 			tmpfs_ref_node(node);
390 			break;
391 		}
392 	}
393 	TMPFS_UNLOCK(tmp);
394 
395 	if (node != NULL) {
396 		error = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp);
397 		tmpfs_free_node(tmp, node);
398 	} else
399 		error = EINVAL;
400 	return (error);
401 }
402 
403 /* ARGSUSED2 */
404 static int
405 tmpfs_statfs(struct mount *mp, struct statfs *sbp)
406 {
407 	struct tmpfs_mount *tmp;
408 	size_t used;
409 
410 	tmp = VFS_TO_TMPFS(mp);
411 
412 	sbp->f_iosize = PAGE_SIZE;
413 	sbp->f_bsize = PAGE_SIZE;
414 
415 	used = tmpfs_pages_used(tmp);
416 	if (tmp->tm_pages_max != ULONG_MAX)
417 		 sbp->f_blocks = tmp->tm_pages_max;
418 	else
419 		 sbp->f_blocks = used + tmpfs_mem_avail();
420 	if (sbp->f_blocks <= used)
421 		sbp->f_bavail = 0;
422 	else
423 		sbp->f_bavail = sbp->f_blocks - used;
424 	sbp->f_bfree = sbp->f_bavail;
425 	used = tmp->tm_nodes_inuse;
426 	sbp->f_files = tmp->tm_nodes_max;
427 	if (sbp->f_files <= used)
428 		sbp->f_ffree = 0;
429 	else
430 		sbp->f_ffree = sbp->f_files - used;
431 	/* sbp->f_owner = tmp->tn_uid; */
432 
433 	return 0;
434 }
435 
436 static int
437 tmpfs_sync(struct mount *mp, int waitfor)
438 {
439 	struct vnode *vp, *mvp;
440 	struct vm_object *obj;
441 
442 	if (waitfor == MNT_SUSPEND) {
443 		MNT_ILOCK(mp);
444 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
445 		MNT_IUNLOCK(mp);
446 	} else if (waitfor == MNT_LAZY) {
447 		/*
448 		 * Handle lazy updates of mtime from writes to mmaped
449 		 * regions.  Use MNT_VNODE_FOREACH_ALL instead of
450 		 * MNT_VNODE_FOREACH_ACTIVE, since unmap of the
451 		 * tmpfs-backed vnode does not call vinactive(), due
452 		 * to vm object type is OBJT_SWAP.
453 		 */
454 		MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
455 			if (vp->v_type != VREG) {
456 				VI_UNLOCK(vp);
457 				continue;
458 			}
459 			obj = vp->v_object;
460 			KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
461 			    (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
462 
463 			/*
464 			 * Unlocked read, avoid taking vnode lock if
465 			 * not needed.  Lost update will be handled on
466 			 * the next call.
467 			 */
468 			if ((obj->flags & OBJ_TMPFS_DIRTY) == 0) {
469 				VI_UNLOCK(vp);
470 				continue;
471 			}
472 			if (vget(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
473 			    curthread) != 0)
474 				continue;
475 			tmpfs_check_mtime(vp);
476 			vput(vp);
477 		}
478 	}
479 	return (0);
480 }
481 
482 /*
483  * The presence of a susp_clean method tells the VFS to track writes.
484  */
485 static void
486 tmpfs_susp_clean(struct mount *mp __unused)
487 {
488 }
489 
490 /*
491  * tmpfs vfs operations.
492  */
493 
494 struct vfsops tmpfs_vfsops = {
495 	.vfs_mount =			tmpfs_mount,
496 	.vfs_unmount =			tmpfs_unmount,
497 	.vfs_root =			tmpfs_root,
498 	.vfs_statfs =			tmpfs_statfs,
499 	.vfs_fhtovp =			tmpfs_fhtovp,
500 	.vfs_sync =			tmpfs_sync,
501 	.vfs_susp_clean =		tmpfs_susp_clean,
502 };
503 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL);
504