xref: /netbsd-src/sys/fs/tmpfs/tmpfs_vfsops.c (revision ed75d7a867996c84cfa88e3b8906816277e957f7)
1 /*	$NetBSD: tmpfs_vfsops.c,v 1.76 2020/01/17 20:08:08 ad Exp $	*/
2 
3 /*
4  * Copyright (c) 2005, 2006, 2007 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Julio M. Merino Vidal, developed as part of Google's Summer of Code
9  * 2005 program.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Efficient memory file system.
35  *
36  * tmpfs is a file system that uses NetBSD's virtual memory sub-system
37  * (the well-known UVM) to store file data and metadata in an efficient
38  * way.  This means that it does not follow the structure of an on-disk
39  * file system because it simply does not need to.  Instead, it uses
40  * memory-specific data structures and algorithms to automatically
41  * allocate and release resources.
42  */
43 
44 #include <sys/cdefs.h>
45 __KERNEL_RCSID(0, "$NetBSD: tmpfs_vfsops.c,v 1.76 2020/01/17 20:08:08 ad Exp $");
46 
47 #include <sys/param.h>
48 #include <sys/atomic.h>
49 #include <sys/types.h>
50 #include <sys/kmem.h>
51 #include <sys/mount.h>
52 #include <sys/stat.h>
53 #include <sys/systm.h>
54 #include <sys/vnode.h>
55 #include <sys/kauth.h>
56 #include <sys/module.h>
57 
58 #include <miscfs/genfs/genfs.h>
59 #include <fs/tmpfs/tmpfs.h>
60 #include <fs/tmpfs/tmpfs_args.h>
61 
62 MODULE(MODULE_CLASS_VFS, tmpfs, NULL);
63 
64 struct pool	tmpfs_dirent_pool;
65 struct pool	tmpfs_node_pool;
66 
67 void
68 tmpfs_init(void)
69 {
70 
71 	pool_init(&tmpfs_dirent_pool, sizeof(tmpfs_dirent_t), 0, 0, 0,
72 	    "tmpfs_dirent", &pool_allocator_nointr, IPL_NONE);
73 	pool_init(&tmpfs_node_pool, sizeof(tmpfs_node_t), 0, 0, 0,
74 	    "tmpfs_node", &pool_allocator_nointr, IPL_NONE);
75 }
76 
77 void
78 tmpfs_done(void)
79 {
80 
81 	pool_destroy(&tmpfs_dirent_pool);
82 	pool_destroy(&tmpfs_node_pool);
83 }
84 
85 int
86 tmpfs_mount(struct mount *mp, const char *path, void *data, size_t *data_len)
87 {
88 	struct tmpfs_args *args = data;
89 	tmpfs_mount_t *tmp;
90 	tmpfs_node_t *root;
91 	struct vattr va;
92 	struct vnode *vp;
93 	uint64_t memlimit;
94 	ino_t nodes;
95 	int error, flags;
96 	bool set_memlimit;
97 	bool set_nodes;
98 
99 	if (args == NULL)
100 		return EINVAL;
101 
102 	/* Validate the version. */
103 	if (*data_len < sizeof(*args) ||
104 	    args->ta_version != TMPFS_ARGS_VERSION)
105 		return EINVAL;
106 
107 	/* Handle retrieval of mount point arguments. */
108 	if (mp->mnt_flag & MNT_GETARGS) {
109 		if (mp->mnt_data == NULL)
110 			return EIO;
111 		tmp = VFS_TO_TMPFS(mp);
112 
113 		args->ta_version = TMPFS_ARGS_VERSION;
114 		args->ta_nodes_max = tmp->tm_nodes_max;
115 		args->ta_size_max = tmp->tm_mem_limit;
116 
117 		root = tmp->tm_root;
118 		args->ta_root_uid = root->tn_uid;
119 		args->ta_root_gid = root->tn_gid;
120 		args->ta_root_mode = root->tn_mode;
121 
122 		*data_len = sizeof(*args);
123 		return 0;
124 	}
125 
126 
127 	/* Prohibit mounts if there is not enough memory. */
128 	if (tmpfs_mem_info(true) < uvmexp.freetarg)
129 		return EINVAL;
130 
131 	/* Check for invalid uid and gid arguments */
132 	if (args->ta_root_uid == VNOVAL || args->ta_root_gid == VNOVAL)
133 		return EINVAL;
134 
135 	/* Get the memory usage limit for this file-system. */
136 	if (args->ta_size_max < PAGE_SIZE) {
137 		memlimit = UINT64_MAX;
138 		set_memlimit = false;
139 	} else {
140 		memlimit = args->ta_size_max;
141 		set_memlimit = true;
142 	}
143 	KASSERT(memlimit > 0);
144 
145 	if (args->ta_nodes_max <= 3) {
146 		nodes = 3 + (memlimit / 1024);
147 		set_nodes = false;
148 	} else {
149 		nodes = args->ta_nodes_max;
150 		set_nodes = true;
151 	}
152 	nodes = MIN(nodes, INT_MAX);
153 	KASSERT(nodes >= 3);
154 
155 	if (mp->mnt_flag & MNT_UPDATE) {
156 		tmp = VFS_TO_TMPFS(mp);
157 		if (set_nodes && nodes < tmp->tm_nodes_cnt)
158 			return EBUSY;
159 		if ((mp->mnt_iflag & IMNT_WANTRDONLY)) {
160 			/* Changing from read/write to read-only. */
161 			flags = WRITECLOSE;
162 			if ((mp->mnt_flag & MNT_FORCE))
163 				flags |= FORCECLOSE;
164 			error = vflush(mp, NULL, flags);
165 			if (error)
166 				return error;
167 		}
168 		if (set_memlimit) {
169 			if ((error = tmpfs_mntmem_set(tmp, memlimit)) != 0)
170 				return error;
171 		}
172 		if (set_nodes)
173 			tmp->tm_nodes_max = nodes;
174 		root = tmp->tm_root;
175 		root->tn_uid = args->ta_root_uid;
176 		root->tn_gid = args->ta_root_gid;
177 		root->tn_mode = args->ta_root_mode;
178 		return 0;
179 	}
180 
181 	mp->mnt_flag |= MNT_LOCAL;
182 	mp->mnt_stat.f_namemax = TMPFS_MAXNAMLEN;
183 	mp->mnt_fs_bshift = PAGE_SHIFT;
184 	mp->mnt_dev_bshift = DEV_BSHIFT;
185 	mp->mnt_iflag |= IMNT_MPSAFE | IMNT_CAN_RWTORO;
186 	vfs_getnewfsid(mp);
187 
188 	/* Allocate the tmpfs mount structure and fill it. */
189 	tmp = kmem_zalloc(sizeof(tmpfs_mount_t), KM_SLEEP);
190 	tmp->tm_nodes_max = nodes;
191 	tmp->tm_nodes_cnt = 0;
192 	LIST_INIT(&tmp->tm_nodes);
193 
194 	mutex_init(&tmp->tm_lock, MUTEX_DEFAULT, IPL_NONE);
195 	tmpfs_mntmem_init(tmp, memlimit);
196 	mp->mnt_data = tmp;
197 
198 	/* Allocate the root node. */
199 	vattr_null(&va);
200 	va.va_type = VDIR;
201 	va.va_mode = args->ta_root_mode & ALLPERMS;
202 	va.va_uid = args->ta_root_uid;
203 	va.va_gid = args->ta_root_gid;
204 	error = vcache_new(mp, NULL, &va, NOCRED, NULL, &vp);
205 	if (error) {
206 		mp->mnt_data = NULL;
207 		tmpfs_mntmem_destroy(tmp);
208 		mutex_destroy(&tmp->tm_lock);
209 		kmem_free(tmp, sizeof(*tmp));
210 		return error;
211 	}
212 	KASSERT(vp != NULL);
213 	root = VP_TO_TMPFS_NODE(vp);
214 	KASSERT(root != NULL);
215 
216 	/*
217 	 * Parent of the root inode is itself.  Also, root inode has no
218 	 * directory entry (i.e. is never attached), thus hold an extra
219 	 * reference (link) for it.
220 	 */
221 	root->tn_links++;
222 	root->tn_spec.tn_dir.tn_parent = root;
223 	tmp->tm_root = root;
224 	vrele(vp);
225 
226 	error = set_statvfs_info(path, UIO_USERSPACE, "tmpfs", UIO_SYSSPACE,
227 	    mp->mnt_op->vfs_name, mp, curlwp);
228 	if (error) {
229 		(void)tmpfs_unmount(mp, MNT_FORCE);
230 	}
231 	return error;
232 }
233 
234 int
235 tmpfs_start(struct mount *mp, int flags)
236 {
237 
238 	return 0;
239 }
240 
241 int
242 tmpfs_unmount(struct mount *mp, int mntflags)
243 {
244 	tmpfs_mount_t *tmp = VFS_TO_TMPFS(mp);
245 	tmpfs_node_t *node, *cnode;
246 	int error, flags = 0;
247 
248 	/* Handle forced unmounts. */
249 	if (mntflags & MNT_FORCE)
250 		flags |= FORCECLOSE;
251 
252 	/* Finalize all pending I/O. */
253 	error = vflush(mp, NULL, flags);
254 	if (error != 0)
255 		return error;
256 
257 	/*
258 	 * First round, detach and destroy all directory entries.
259 	 * Also, clear the pointers to the vnodes - they are gone.
260 	 */
261 	LIST_FOREACH(node, &tmp->tm_nodes, tn_entries) {
262 		tmpfs_dirent_t *de;
263 
264 		node->tn_vnode = NULL;
265 		if (node->tn_type != VDIR) {
266 			continue;
267 		}
268 		while ((de = TAILQ_FIRST(&node->tn_spec.tn_dir.tn_dir)) != NULL) {
269 			cnode = de->td_node;
270 			if (cnode && cnode != TMPFS_NODE_WHITEOUT) {
271 				cnode->tn_vnode = NULL;
272 			}
273 			tmpfs_dir_detach(node, de);
274 			tmpfs_free_dirent(tmp, de);
275 		}
276 		/* Extra virtual entry (itself for the root). */
277 		node->tn_links--;
278 	}
279 
280 	/* Release the reference on root (diagnostic). */
281 	node = tmp->tm_root;
282 	node->tn_links--;
283 
284 	/* Second round, destroy all inodes. */
285 	while ((node = LIST_FIRST(&tmp->tm_nodes)) != NULL) {
286 		tmpfs_free_node(tmp, node);
287 	}
288 
289 	/* Throw away the tmpfs_mount structure. */
290 	tmpfs_mntmem_destroy(tmp);
291 	mutex_destroy(&tmp->tm_lock);
292 	kmem_free(tmp, sizeof(*tmp));
293 	mp->mnt_data = NULL;
294 
295 	return 0;
296 }
297 
298 int
299 tmpfs_root(struct mount *mp, int lktype, vnode_t **vpp)
300 {
301 	tmpfs_node_t *node = VFS_TO_TMPFS(mp)->tm_root;
302 	int error;
303 
304 	error = vcache_get(mp, &node, sizeof(node), vpp);
305 	if (error)
306 		return error;
307 	error = vn_lock(*vpp, lktype);
308 	if (error) {
309 		vrele(*vpp);
310 		*vpp = NULL;
311 		return error;
312 	}
313 
314 	return 0;
315 }
316 
317 int
318 tmpfs_vget(struct mount *mp, ino_t ino, int lktype, vnode_t **vpp)
319 {
320 
321 	return EOPNOTSUPP;
322 }
323 
324 int
325 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int lktype, vnode_t **vpp)
326 {
327 	tmpfs_mount_t *tmp = VFS_TO_TMPFS(mp);
328 	tmpfs_node_t *node;
329 	tmpfs_fid_t tfh;
330 	int error;
331 
332 	if (fhp->fid_len != sizeof(tmpfs_fid_t)) {
333 		return EINVAL;
334 	}
335 	memcpy(&tfh, fhp, sizeof(tmpfs_fid_t));
336 
337 	mutex_enter(&tmp->tm_lock);
338 	/* XXX big oof .. use a better data structure */
339 	LIST_FOREACH(node, &tmp->tm_nodes, tn_entries) {
340 		if (node->tn_id == tfh.tf_id) {
341 			/* Prevent this node from disappearing. */
342 			atomic_inc_32(&node->tn_holdcount);
343 			break;
344 		}
345 	}
346 	mutex_exit(&tmp->tm_lock);
347 	if (node == NULL)
348 		return ESTALE;
349 
350 	error = vcache_get(mp, &node, sizeof(node), vpp);
351 	/* If this node has been reclaimed free it now. */
352 	if (atomic_dec_32_nv(&node->tn_holdcount) == TMPFS_NODE_RECLAIMED) {
353 		KASSERT(error != 0);
354 		tmpfs_free_node(tmp, node);
355 	}
356 	if (error)
357 		return (error == ENOENT ? ESTALE : error);
358 	error = vn_lock(*vpp, lktype);
359 	if (error) {
360 		vrele(*vpp);
361 		*vpp = NULL;
362 		return error;
363 	}
364 	if (TMPFS_NODE_GEN(node) != tfh.tf_gen) {
365 		vput(*vpp);
366 		*vpp = NULL;
367 		return ESTALE;
368 	}
369 
370 	return 0;
371 }
372 
373 int
374 tmpfs_vptofh(vnode_t *vp, struct fid *fhp, size_t *fh_size)
375 {
376 	tmpfs_fid_t tfh;
377 	tmpfs_node_t *node;
378 
379 	if (*fh_size < sizeof(tmpfs_fid_t)) {
380 		*fh_size = sizeof(tmpfs_fid_t);
381 		return E2BIG;
382 	}
383 	*fh_size = sizeof(tmpfs_fid_t);
384 	node = VP_TO_TMPFS_NODE(vp);
385 
386 	memset(&tfh, 0, sizeof(tfh));
387 	tfh.tf_len = sizeof(tmpfs_fid_t);
388 	tfh.tf_gen = TMPFS_NODE_GEN(node);
389 	tfh.tf_id = node->tn_id;
390 	memcpy(fhp, &tfh, sizeof(tfh));
391 
392 	return 0;
393 }
394 
395 int
396 tmpfs_statvfs(struct mount *mp, struct statvfs *sbp)
397 {
398 	tmpfs_mount_t *tmp;
399 	fsfilcnt_t freenodes;
400 	size_t avail;
401 
402 	tmp = VFS_TO_TMPFS(mp);
403 
404 	sbp->f_iosize = sbp->f_frsize = sbp->f_bsize = PAGE_SIZE;
405 
406 	mutex_enter(&tmp->tm_acc_lock);
407 	avail =  tmpfs_pages_avail(tmp);
408 	sbp->f_blocks = (tmpfs_bytes_max(tmp) >> PAGE_SHIFT);
409 	sbp->f_bavail = sbp->f_bfree = avail;
410 	sbp->f_bresvd = 0;
411 
412 	freenodes = MIN(tmp->tm_nodes_max - tmp->tm_nodes_cnt,
413 	    avail * PAGE_SIZE / sizeof(tmpfs_node_t));
414 
415 	sbp->f_files = tmp->tm_nodes_cnt + freenodes;
416 	sbp->f_favail = sbp->f_ffree = freenodes;
417 	sbp->f_fresvd = 0;
418 	mutex_exit(&tmp->tm_acc_lock);
419 
420 	copy_statvfs_info(sbp, mp);
421 
422 	return 0;
423 }
424 
425 int
426 tmpfs_sync(struct mount *mp, int waitfor, kauth_cred_t uc)
427 {
428 
429 	return 0;
430 }
431 
432 int
433 tmpfs_snapshot(struct mount *mp, vnode_t *vp, struct timespec *ctime)
434 {
435 
436 	return EOPNOTSUPP;
437 }
438 
439 /*
440  * tmpfs vfs operations.
441  */
442 
443 extern const struct vnodeopv_desc tmpfs_fifoop_opv_desc;
444 extern const struct vnodeopv_desc tmpfs_specop_opv_desc;
445 extern const struct vnodeopv_desc tmpfs_vnodeop_opv_desc;
446 
447 const struct vnodeopv_desc * const tmpfs_vnodeopv_descs[] = {
448 	&tmpfs_fifoop_opv_desc,
449 	&tmpfs_specop_opv_desc,
450 	&tmpfs_vnodeop_opv_desc,
451 	NULL,
452 };
453 
454 struct vfsops tmpfs_vfsops = {
455 	.vfs_name = MOUNT_TMPFS,
456 	.vfs_min_mount_data = sizeof (struct tmpfs_args),
457 	.vfs_mount = tmpfs_mount,
458 	.vfs_start = tmpfs_start,
459 	.vfs_unmount = tmpfs_unmount,
460 	.vfs_root = tmpfs_root,
461 	.vfs_quotactl = (void *)eopnotsupp,
462 	.vfs_statvfs = tmpfs_statvfs,
463 	.vfs_sync = tmpfs_sync,
464 	.vfs_vget = tmpfs_vget,
465 	.vfs_loadvnode = tmpfs_loadvnode,
466 	.vfs_newvnode = tmpfs_newvnode,
467 	.vfs_fhtovp = tmpfs_fhtovp,
468 	.vfs_vptofh = tmpfs_vptofh,
469 	.vfs_init = tmpfs_init,
470 	.vfs_done = tmpfs_done,
471 	.vfs_snapshot = tmpfs_snapshot,
472 	.vfs_extattrctl = vfs_stdextattrctl,
473 	.vfs_suspendctl = genfs_suspendctl,
474 	.vfs_renamelock_enter = genfs_renamelock_enter,
475 	.vfs_renamelock_exit = genfs_renamelock_exit,
476 	.vfs_fsync = (void *)eopnotsupp,
477 	.vfs_opv_descs = tmpfs_vnodeopv_descs
478 };
479 
480 static int
481 tmpfs_modcmd(modcmd_t cmd, void *arg)
482 {
483 
484 	switch (cmd) {
485 	case MODULE_CMD_INIT:
486 		return vfs_attach(&tmpfs_vfsops);
487 	case MODULE_CMD_FINI:
488 		return vfs_detach(&tmpfs_vfsops);
489 	default:
490 		return ENOTTY;
491 	}
492 }
493