xref: /freebsd/sys/fs/tmpfs/tmpfs_vfsops.c (revision 3416500aef140042c64bc149cb1ec6620483bc44)
1 /*	$NetBSD: tmpfs_vfsops.c,v 1.10 2005/12/11 12:24:29 christos Exp $	*/
2 
3 /*-
4  * Copyright (c) 2005 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 FreeBSD's virtual memory
37  * sub-system to store file data and metadata in an efficient way.
38  * This means that it does not follow the structure of an on-disk file
39  * 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 #include <sys/cdefs.h>
44 __FBSDID("$FreeBSD$");
45 
46 #include <sys/param.h>
47 #include <sys/limits.h>
48 #include <sys/lock.h>
49 #include <sys/mutex.h>
50 #include <sys/proc.h>
51 #include <sys/jail.h>
52 #include <sys/kernel.h>
53 #include <sys/rwlock.h>
54 #include <sys/stat.h>
55 #include <sys/systm.h>
56 #include <sys/sysctl.h>
57 
58 #include <vm/vm.h>
59 #include <vm/vm_object.h>
60 #include <vm/vm_param.h>
61 
62 #include <fs/tmpfs/tmpfs.h>
63 
64 /*
65  * Default permission for root node
66  */
67 #define TMPFS_DEFAULT_ROOT_MODE	(S_IRWXU|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH)
68 
69 MALLOC_DEFINE(M_TMPFSMNT, "tmpfs mount", "tmpfs mount structures");
70 MALLOC_DEFINE(M_TMPFSNAME, "tmpfs name", "tmpfs file names");
71 
72 static int	tmpfs_mount(struct mount *);
73 static int	tmpfs_unmount(struct mount *, int);
74 static int	tmpfs_root(struct mount *, int flags, struct vnode **);
75 static int	tmpfs_fhtovp(struct mount *, struct fid *, int,
76 		    struct vnode **);
77 static int	tmpfs_statfs(struct mount *, struct statfs *);
78 static void	tmpfs_susp_clean(struct mount *);
79 
80 static const char *tmpfs_opts[] = {
81 	"from", "size", "maxfilesize", "inodes", "uid", "gid", "mode", "export",
82 	"union", "nonc", NULL
83 };
84 
85 static const char *tmpfs_updateopts[] = {
86 	"from", "export", NULL
87 };
88 
89 static int
90 tmpfs_node_ctor(void *mem, int size, void *arg, int flags)
91 {
92 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
93 
94 	node->tn_gen++;
95 	node->tn_size = 0;
96 	node->tn_status = 0;
97 	node->tn_flags = 0;
98 	node->tn_links = 0;
99 	node->tn_vnode = NULL;
100 	node->tn_vpstate = 0;
101 
102 	return (0);
103 }
104 
105 static void
106 tmpfs_node_dtor(void *mem, int size, void *arg)
107 {
108 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
109 	node->tn_type = VNON;
110 }
111 
112 static int
113 tmpfs_node_init(void *mem, int size, int flags)
114 {
115 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
116 	node->tn_id = 0;
117 
118 	mtx_init(&node->tn_interlock, "tmpfs node interlock", NULL, MTX_DEF);
119 	node->tn_gen = arc4random();
120 
121 	return (0);
122 }
123 
124 static void
125 tmpfs_node_fini(void *mem, int size)
126 {
127 	struct tmpfs_node *node = (struct tmpfs_node *)mem;
128 
129 	mtx_destroy(&node->tn_interlock);
130 }
131 
132 static int
133 tmpfs_mount(struct mount *mp)
134 {
135 	const size_t nodes_per_page = howmany(PAGE_SIZE,
136 	    sizeof(struct tmpfs_dirent) + sizeof(struct tmpfs_node));
137 	struct tmpfs_mount *tmp;
138 	struct tmpfs_node *root;
139 	struct thread *td = curthread;
140 	int error;
141 	bool nonc;
142 	/* Size counters. */
143 	u_quad_t pages;
144 	off_t nodes_max, size_max, maxfilesize;
145 
146 	/* Root node attributes. */
147 	uid_t root_uid;
148 	gid_t root_gid;
149 	mode_t root_mode;
150 
151 	struct vattr va;
152 
153 	if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_TMPFS))
154 		return (EPERM);
155 
156 	if (vfs_filteropt(mp->mnt_optnew, tmpfs_opts))
157 		return (EINVAL);
158 
159 	if (mp->mnt_flag & MNT_UPDATE) {
160 		/* Only support update mounts for certain options. */
161 		if (vfs_filteropt(mp->mnt_optnew, tmpfs_updateopts) != 0)
162 			return (EOPNOTSUPP);
163 		if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) !=
164 		    ((struct tmpfs_mount *)mp->mnt_data)->tm_ronly)
165 			return (EOPNOTSUPP);
166 		return (0);
167 	}
168 
169 	vn_lock(mp->mnt_vnodecovered, LK_SHARED | LK_RETRY);
170 	error = VOP_GETATTR(mp->mnt_vnodecovered, &va, mp->mnt_cred);
171 	VOP_UNLOCK(mp->mnt_vnodecovered, 0);
172 	if (error)
173 		return (error);
174 
175 	if (mp->mnt_cred->cr_ruid != 0 ||
176 	    vfs_scanopt(mp->mnt_optnew, "gid", "%d", &root_gid) != 1)
177 		root_gid = va.va_gid;
178 	if (mp->mnt_cred->cr_ruid != 0 ||
179 	    vfs_scanopt(mp->mnt_optnew, "uid", "%d", &root_uid) != 1)
180 		root_uid = va.va_uid;
181 	if (mp->mnt_cred->cr_ruid != 0 ||
182 	    vfs_scanopt(mp->mnt_optnew, "mode", "%ho", &root_mode) != 1)
183 		root_mode = va.va_mode;
184 	if (vfs_getopt_size(mp->mnt_optnew, "inodes", &nodes_max) != 0)
185 		nodes_max = 0;
186 	if (vfs_getopt_size(mp->mnt_optnew, "size", &size_max) != 0)
187 		size_max = 0;
188 	if (vfs_getopt_size(mp->mnt_optnew, "maxfilesize", &maxfilesize) != 0)
189 		maxfilesize = 0;
190 	nonc = vfs_getopt(mp->mnt_optnew, "nonc", NULL, NULL) == 0;
191 
192 	/* Do not allow mounts if we do not have enough memory to preserve
193 	 * the minimum reserved pages. */
194 	if (tmpfs_mem_avail() < TMPFS_PAGES_MINRESERVED)
195 		return (ENOSPC);
196 
197 	/* Get the maximum number of memory pages this file system is
198 	 * allowed to use, based on the maximum size the user passed in
199 	 * the mount structure.  A value of zero is treated as if the
200 	 * maximum available space was requested. */
201 	if (size_max == 0 || size_max > OFF_MAX - PAGE_SIZE ||
202 	    (SIZE_MAX < OFF_MAX && size_max / PAGE_SIZE >= SIZE_MAX))
203 		pages = SIZE_MAX;
204 	else {
205 		size_max = roundup(size_max, PAGE_SIZE);
206 		pages = howmany(size_max, PAGE_SIZE);
207 	}
208 	MPASS(pages > 0);
209 
210 	if (nodes_max <= 3) {
211 		if (pages < INT_MAX / nodes_per_page)
212 			nodes_max = pages * nodes_per_page;
213 		else
214 			nodes_max = INT_MAX;
215 	}
216 	if (nodes_max > INT_MAX)
217 		nodes_max = INT_MAX;
218 	MPASS(nodes_max >= 3);
219 
220 	/* Allocate the tmpfs mount structure and fill it. */
221 	tmp = (struct tmpfs_mount *)malloc(sizeof(struct tmpfs_mount),
222 	    M_TMPFSMNT, M_WAITOK | M_ZERO);
223 
224 	mtx_init(&tmp->tm_allnode_lock, "tmpfs allnode lock", NULL, MTX_DEF);
225 	tmp->tm_nodes_max = nodes_max;
226 	tmp->tm_nodes_inuse = 0;
227 	tmp->tm_refcount = 1;
228 	tmp->tm_maxfilesize = maxfilesize > 0 ? maxfilesize : OFF_MAX;
229 	LIST_INIT(&tmp->tm_nodes_used);
230 
231 	tmp->tm_pages_max = pages;
232 	tmp->tm_pages_used = 0;
233 	tmp->tm_ino_unr = new_unrhdr(2, INT_MAX, &tmp->tm_allnode_lock);
234 	tmp->tm_dirent_pool = uma_zcreate("TMPFS dirent",
235 	    sizeof(struct tmpfs_dirent), NULL, NULL, NULL, NULL,
236 	    UMA_ALIGN_PTR, 0);
237 	tmp->tm_node_pool = uma_zcreate("TMPFS node",
238 	    sizeof(struct tmpfs_node), tmpfs_node_ctor, tmpfs_node_dtor,
239 	    tmpfs_node_init, tmpfs_node_fini, UMA_ALIGN_PTR, 0);
240 	tmp->tm_ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
241 	tmp->tm_nonc = nonc;
242 
243 	/* Allocate the root node. */
244 	error = tmpfs_alloc_node(mp, tmp, VDIR, root_uid, root_gid,
245 	    root_mode & ALLPERMS, NULL, NULL, VNOVAL, &root);
246 
247 	if (error != 0 || root == NULL) {
248 		uma_zdestroy(tmp->tm_node_pool);
249 		uma_zdestroy(tmp->tm_dirent_pool);
250 		delete_unrhdr(tmp->tm_ino_unr);
251 		free(tmp, M_TMPFSMNT);
252 		return (error);
253 	}
254 	KASSERT(root->tn_id == 2,
255 	    ("tmpfs root with invalid ino: %ju", (uintmax_t)root->tn_id));
256 	tmp->tm_root = root;
257 
258 	MNT_ILOCK(mp);
259 	mp->mnt_flag |= MNT_LOCAL;
260 	mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED;
261 	MNT_IUNLOCK(mp);
262 
263 	mp->mnt_data = tmp;
264 	mp->mnt_stat.f_namemax = MAXNAMLEN;
265 	vfs_getnewfsid(mp);
266 	vfs_mountedfrom(mp, "tmpfs");
267 
268 	return 0;
269 }
270 
271 /* ARGSUSED2 */
272 static int
273 tmpfs_unmount(struct mount *mp, int mntflags)
274 {
275 	struct tmpfs_mount *tmp;
276 	struct tmpfs_node *node;
277 	int error, flags;
278 
279 	flags = (mntflags & MNT_FORCE) != 0 ? FORCECLOSE : 0;
280 	tmp = VFS_TO_TMPFS(mp);
281 
282 	/* Stop writers */
283 	error = vfs_write_suspend_umnt(mp);
284 	if (error != 0)
285 		return (error);
286 	/*
287 	 * At this point, nodes cannot be destroyed by any other
288 	 * thread because write suspension is started.
289 	 */
290 
291 	for (;;) {
292 		error = vflush(mp, 0, flags, curthread);
293 		if (error != 0) {
294 			vfs_write_resume(mp, VR_START_WRITE);
295 			return (error);
296 		}
297 		MNT_ILOCK(mp);
298 		if (mp->mnt_nvnodelistsize == 0) {
299 			MNT_IUNLOCK(mp);
300 			break;
301 		}
302 		MNT_IUNLOCK(mp);
303 		if ((mntflags & MNT_FORCE) == 0) {
304 			vfs_write_resume(mp, VR_START_WRITE);
305 			return (EBUSY);
306 		}
307 	}
308 
309 	TMPFS_LOCK(tmp);
310 	while ((node = LIST_FIRST(&tmp->tm_nodes_used)) != NULL) {
311 		TMPFS_NODE_LOCK(node);
312 		if (node->tn_type == VDIR)
313 			tmpfs_dir_destroy(tmp, node);
314 		if (tmpfs_free_node_locked(tmp, node, true))
315 			TMPFS_LOCK(tmp);
316 		else
317 			TMPFS_NODE_UNLOCK(node);
318 	}
319 
320 	mp->mnt_data = NULL;
321 	tmpfs_free_tmp(tmp);
322 	vfs_write_resume(mp, VR_START_WRITE);
323 
324 	MNT_ILOCK(mp);
325 	mp->mnt_flag &= ~MNT_LOCAL;
326 	MNT_IUNLOCK(mp);
327 
328 	return (0);
329 }
330 
331 void
332 tmpfs_free_tmp(struct tmpfs_mount *tmp)
333 {
334 
335 	MPASS(tmp->tm_refcount > 0);
336 	tmp->tm_refcount--;
337 	if (tmp->tm_refcount > 0) {
338 		TMPFS_UNLOCK(tmp);
339 		return;
340 	}
341 	TMPFS_UNLOCK(tmp);
342 
343 	uma_zdestroy(tmp->tm_dirent_pool);
344 	uma_zdestroy(tmp->tm_node_pool);
345 	delete_unrhdr(tmp->tm_ino_unr);
346 
347 	mtx_destroy(&tmp->tm_allnode_lock);
348 	MPASS(tmp->tm_pages_used == 0);
349 	MPASS(tmp->tm_nodes_inuse == 0);
350 
351 	free(tmp, M_TMPFSMNT);
352 }
353 
354 static int
355 tmpfs_root(struct mount *mp, int flags, struct vnode **vpp)
356 {
357 	int error;
358 
359 	error = tmpfs_alloc_vp(mp, VFS_TO_TMPFS(mp)->tm_root, flags, vpp);
360 	if (error == 0)
361 		(*vpp)->v_vflag |= VV_ROOT;
362 	return (error);
363 }
364 
365 static int
366 tmpfs_fhtovp(struct mount *mp, struct fid *fhp, int flags,
367     struct vnode **vpp)
368 {
369 	struct tmpfs_fid *tfhp;
370 	struct tmpfs_mount *tmp;
371 	struct tmpfs_node *node;
372 	int error;
373 
374 	tmp = VFS_TO_TMPFS(mp);
375 
376 	tfhp = (struct tmpfs_fid *)fhp;
377 	if (tfhp->tf_len != sizeof(struct tmpfs_fid))
378 		return (EINVAL);
379 
380 	if (tfhp->tf_id >= tmp->tm_nodes_max)
381 		return (EINVAL);
382 
383 	TMPFS_LOCK(tmp);
384 	LIST_FOREACH(node, &tmp->tm_nodes_used, tn_entries) {
385 		if (node->tn_id == tfhp->tf_id &&
386 		    node->tn_gen == tfhp->tf_gen) {
387 			tmpfs_ref_node(node);
388 			break;
389 		}
390 	}
391 	TMPFS_UNLOCK(tmp);
392 
393 	if (node != NULL) {
394 		error = tmpfs_alloc_vp(mp, node, LK_EXCLUSIVE, vpp);
395 		tmpfs_free_node(tmp, node);
396 	} else
397 		error = EINVAL;
398 	return (error);
399 }
400 
401 /* ARGSUSED2 */
402 static int
403 tmpfs_statfs(struct mount *mp, struct statfs *sbp)
404 {
405 	struct tmpfs_mount *tmp;
406 	size_t used;
407 
408 	tmp = VFS_TO_TMPFS(mp);
409 
410 	sbp->f_iosize = PAGE_SIZE;
411 	sbp->f_bsize = PAGE_SIZE;
412 
413 	used = tmpfs_pages_used(tmp);
414 	if (tmp->tm_pages_max != ULONG_MAX)
415 		 sbp->f_blocks = tmp->tm_pages_max;
416 	else
417 		 sbp->f_blocks = used + tmpfs_mem_avail();
418 	if (sbp->f_blocks <= used)
419 		sbp->f_bavail = 0;
420 	else
421 		sbp->f_bavail = sbp->f_blocks - used;
422 	sbp->f_bfree = sbp->f_bavail;
423 	used = tmp->tm_nodes_inuse;
424 	sbp->f_files = tmp->tm_nodes_max;
425 	if (sbp->f_files <= used)
426 		sbp->f_ffree = 0;
427 	else
428 		sbp->f_ffree = sbp->f_files - used;
429 	/* sbp->f_owner = tmp->tn_uid; */
430 
431 	return 0;
432 }
433 
434 static int
435 tmpfs_sync(struct mount *mp, int waitfor)
436 {
437 	struct vnode *vp, *mvp;
438 	struct vm_object *obj;
439 
440 	if (waitfor == MNT_SUSPEND) {
441 		MNT_ILOCK(mp);
442 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
443 		MNT_IUNLOCK(mp);
444 	} else if (waitfor == MNT_LAZY) {
445 		/*
446 		 * Handle lazy updates of mtime from writes to mmaped
447 		 * regions.  Use MNT_VNODE_FOREACH_ALL instead of
448 		 * MNT_VNODE_FOREACH_ACTIVE, since unmap of the
449 		 * tmpfs-backed vnode does not call vinactive(), due
450 		 * to vm object type is OBJT_SWAP.
451 		 */
452 		MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
453 			if (vp->v_type != VREG) {
454 				VI_UNLOCK(vp);
455 				continue;
456 			}
457 			obj = vp->v_object;
458 			KASSERT((obj->flags & (OBJ_TMPFS_NODE | OBJ_TMPFS)) ==
459 			    (OBJ_TMPFS_NODE | OBJ_TMPFS), ("non-tmpfs obj"));
460 
461 			/*
462 			 * Unlocked read, avoid taking vnode lock if
463 			 * not needed.  Lost update will be handled on
464 			 * the next call.
465 			 */
466 			if ((obj->flags & OBJ_TMPFS_DIRTY) == 0) {
467 				VI_UNLOCK(vp);
468 				continue;
469 			}
470 			if (vget(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
471 			    curthread) != 0)
472 				continue;
473 			tmpfs_check_mtime(vp);
474 			vput(vp);
475 		}
476 	}
477 	return (0);
478 }
479 
480 /*
481  * The presence of a susp_clean method tells the VFS to track writes.
482  */
483 static void
484 tmpfs_susp_clean(struct mount *mp __unused)
485 {
486 }
487 
488 /*
489  * tmpfs vfs operations.
490  */
491 
492 struct vfsops tmpfs_vfsops = {
493 	.vfs_mount =			tmpfs_mount,
494 	.vfs_unmount =			tmpfs_unmount,
495 	.vfs_root =			tmpfs_root,
496 	.vfs_statfs =			tmpfs_statfs,
497 	.vfs_fhtovp =			tmpfs_fhtovp,
498 	.vfs_sync =			tmpfs_sync,
499 	.vfs_susp_clean =		tmpfs_susp_clean,
500 };
501 VFS_SET(tmpfs_vfsops, tmpfs, VFCF_JAIL);
502