overlayfs: Implement splice-read
[linux-block.git] / fs / fsopen.c
CommitLineData
b4d0d230 1// SPDX-License-Identifier: GPL-2.0-or-later
24dcb3d9
DH
2/* Filesystem access-by-fd.
3 *
4 * Copyright (C) 2017 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
24dcb3d9
DH
6 */
7
8#include <linux/fs_context.h>
ecdab150 9#include <linux/fs_parser.h>
24dcb3d9
DH
10#include <linux/slab.h>
11#include <linux/uaccess.h>
12#include <linux/syscalls.h>
13#include <linux/security.h>
14#include <linux/anon_inodes.h>
15#include <linux/namei.h>
16#include <linux/file.h>
17#include <uapi/linux/mount.h>
ecdab150 18#include "internal.h"
24dcb3d9
DH
19#include "mount.h"
20
007ec26c
DH
21/*
22 * Allow the user to read back any error, warning or informational messages.
23 */
24static ssize_t fscontext_read(struct file *file,
25 char __user *_buf, size_t len, loff_t *pos)
26{
27 struct fs_context *fc = file->private_data;
cc3c0b53 28 struct fc_log *log = fc->log.log;
007ec26c
DH
29 unsigned int logsize = ARRAY_SIZE(log->buffer);
30 ssize_t ret;
31 char *p;
32 bool need_free;
33 int index, n;
34
35 ret = mutex_lock_interruptible(&fc->uapi_mutex);
36 if (ret < 0)
37 return ret;
38
39 if (log->head == log->tail) {
40 mutex_unlock(&fc->uapi_mutex);
41 return -ENODATA;
42 }
43
44 index = log->tail & (logsize - 1);
45 p = log->buffer[index];
46 need_free = log->need_free & (1 << index);
47 log->buffer[index] = NULL;
48 log->need_free &= ~(1 << index);
49 log->tail++;
50 mutex_unlock(&fc->uapi_mutex);
51
52 ret = -EMSGSIZE;
53 n = strlen(p);
54 if (n > len)
55 goto err_free;
56 ret = -EFAULT;
57 if (copy_to_user(_buf, p, n) != 0)
58 goto err_free;
59 ret = n;
60
61err_free:
62 if (need_free)
63 kfree(p);
64 return ret;
65}
66
24dcb3d9
DH
67static int fscontext_release(struct inode *inode, struct file *file)
68{
69 struct fs_context *fc = file->private_data;
70
71 if (fc) {
72 file->private_data = NULL;
73 put_fs_context(fc);
74 }
75 return 0;
76}
77
78const struct file_operations fscontext_fops = {
007ec26c 79 .read = fscontext_read,
24dcb3d9
DH
80 .release = fscontext_release,
81 .llseek = no_llseek,
82};
83
84/*
85 * Attach a filesystem context to a file and an fd.
86 */
87static int fscontext_create_fd(struct fs_context *fc, unsigned int o_flags)
88{
89 int fd;
90
1cdc415f 91 fd = anon_inode_getfd("[fscontext]", &fscontext_fops, fc,
24dcb3d9
DH
92 O_RDWR | o_flags);
93 if (fd < 0)
94 put_fs_context(fc);
95 return fd;
96}
97
007ec26c
DH
98static int fscontext_alloc_log(struct fs_context *fc)
99{
cc3c0b53
AV
100 fc->log.log = kzalloc(sizeof(*fc->log.log), GFP_KERNEL);
101 if (!fc->log.log)
007ec26c 102 return -ENOMEM;
cc3c0b53
AV
103 refcount_set(&fc->log.log->usage, 1);
104 fc->log.log->owner = fc->fs_type->owner;
007ec26c
DH
105 return 0;
106}
107
24dcb3d9
DH
108/*
109 * Open a filesystem by name so that it can be configured for mounting.
110 *
111 * We are allowed to specify a container in which the filesystem will be
112 * opened, thereby indicating which namespaces will be used (notably, which
113 * network namespace will be used for network filesystems).
114 */
115SYSCALL_DEFINE2(fsopen, const char __user *, _fs_name, unsigned int, flags)
116{
117 struct file_system_type *fs_type;
118 struct fs_context *fc;
119 const char *fs_name;
007ec26c 120 int ret;
24dcb3d9 121
a5f85d78 122 if (!may_mount())
24dcb3d9
DH
123 return -EPERM;
124
125 if (flags & ~FSOPEN_CLOEXEC)
126 return -EINVAL;
127
128 fs_name = strndup_user(_fs_name, PAGE_SIZE);
129 if (IS_ERR(fs_name))
130 return PTR_ERR(fs_name);
131
132 fs_type = get_fs_type(fs_name);
133 kfree(fs_name);
134 if (!fs_type)
135 return -ENODEV;
136
137 fc = fs_context_for_mount(fs_type, 0);
138 put_filesystem(fs_type);
139 if (IS_ERR(fc))
140 return PTR_ERR(fc);
141
142 fc->phase = FS_CONTEXT_CREATE_PARAMS;
007ec26c
DH
143
144 ret = fscontext_alloc_log(fc);
145 if (ret < 0)
146 goto err_fc;
147
24dcb3d9 148 return fscontext_create_fd(fc, flags & FSOPEN_CLOEXEC ? O_CLOEXEC : 0);
007ec26c
DH
149
150err_fc:
151 put_fs_context(fc);
152 return ret;
24dcb3d9 153}
ecdab150 154
cf3cba4a
DH
155/*
156 * Pick a superblock into a context for reconfiguration.
157 */
158SYSCALL_DEFINE3(fspick, int, dfd, const char __user *, path, unsigned int, flags)
159{
160 struct fs_context *fc;
161 struct path target;
162 unsigned int lookup_flags;
163 int ret;
164
a5f85d78 165 if (!may_mount())
cf3cba4a
DH
166 return -EPERM;
167
168 if ((flags & ~(FSPICK_CLOEXEC |
169 FSPICK_SYMLINK_NOFOLLOW |
170 FSPICK_NO_AUTOMOUNT |
171 FSPICK_EMPTY_PATH)) != 0)
172 return -EINVAL;
173
174 lookup_flags = LOOKUP_FOLLOW | LOOKUP_AUTOMOUNT;
175 if (flags & FSPICK_SYMLINK_NOFOLLOW)
176 lookup_flags &= ~LOOKUP_FOLLOW;
177 if (flags & FSPICK_NO_AUTOMOUNT)
178 lookup_flags &= ~LOOKUP_AUTOMOUNT;
179 if (flags & FSPICK_EMPTY_PATH)
180 lookup_flags |= LOOKUP_EMPTY;
181 ret = user_path_at(dfd, path, lookup_flags, &target);
182 if (ret < 0)
183 goto err;
184
185 ret = -EINVAL;
186 if (target.mnt->mnt_root != target.dentry)
187 goto err_path;
188
189 fc = fs_context_for_reconfigure(target.dentry, 0, 0);
190 if (IS_ERR(fc)) {
191 ret = PTR_ERR(fc);
192 goto err_path;
193 }
194
195 fc->phase = FS_CONTEXT_RECONF_PARAMS;
196
197 ret = fscontext_alloc_log(fc);
198 if (ret < 0)
199 goto err_fc;
200
201 path_put(&target);
202 return fscontext_create_fd(fc, flags & FSPICK_CLOEXEC ? O_CLOEXEC : 0);
203
204err_fc:
205 put_fs_context(fc);
206err_path:
207 path_put(&target);
208err:
209 return ret;
210}
211
ecdab150
DH
212/*
213 * Check the state and apply the configuration. Note that this function is
214 * allowed to 'steal' the value by setting param->xxx to NULL before returning.
215 */
216static int vfs_fsconfig_locked(struct fs_context *fc, int cmd,
217 struct fs_parameter *param)
218{
219 struct super_block *sb;
220 int ret;
221
222 ret = finish_clean_context(fc);
223 if (ret)
224 return ret;
225 switch (cmd) {
226 case FSCONFIG_CMD_CREATE:
227 if (fc->phase != FS_CONTEXT_CREATE_PARAMS)
228 return -EBUSY;
c3aabf07
AV
229 if (!mount_capable(fc))
230 return -EPERM;
ecdab150
DH
231 fc->phase = FS_CONTEXT_CREATING;
232 ret = vfs_get_tree(fc);
233 if (ret)
234 break;
235 sb = fc->root->d_sb;
236 ret = security_sb_kern_mount(sb);
237 if (unlikely(ret)) {
238 fc_drop_locked(fc);
239 break;
240 }
241 up_write(&sb->s_umount);
242 fc->phase = FS_CONTEXT_AWAITING_MOUNT;
243 return 0;
244 case FSCONFIG_CMD_RECONFIGURE:
245 if (fc->phase != FS_CONTEXT_RECONF_PARAMS)
246 return -EBUSY;
247 fc->phase = FS_CONTEXT_RECONFIGURING;
248 sb = fc->root->d_sb;
249 if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
250 ret = -EPERM;
251 break;
252 }
253 down_write(&sb->s_umount);
254 ret = reconfigure_super(fc);
255 up_write(&sb->s_umount);
256 if (ret)
257 break;
258 vfs_clean_context(fc);
259 return 0;
260 default:
261 if (fc->phase != FS_CONTEXT_CREATE_PARAMS &&
262 fc->phase != FS_CONTEXT_RECONF_PARAMS)
263 return -EBUSY;
264
265 return vfs_parse_fs_param(fc, param);
266 }
267 fc->phase = FS_CONTEXT_FAILED;
268 return ret;
269}
270
271/**
272 * sys_fsconfig - Set parameters and trigger actions on a context
273 * @fd: The filesystem context to act upon
274 * @cmd: The action to take
275 * @_key: Where appropriate, the parameter key to set
276 * @_value: Where appropriate, the parameter value to set
277 * @aux: Additional information for the value
278 *
279 * This system call is used to set parameters on a context, including
280 * superblock settings, data source and security labelling.
281 *
282 * Actions include triggering the creation of a superblock and the
283 * reconfiguration of the superblock attached to the specified context.
284 *
285 * When setting a parameter, @cmd indicates the type of value being proposed
286 * and @_key indicates the parameter to be altered.
287 *
288 * @_value and @aux are used to specify the value, should a value be required:
289 *
290 * (*) fsconfig_set_flag: No value is specified. The parameter must be boolean
291 * in nature. The key may be prefixed with "no" to invert the
292 * setting. @_value must be NULL and @aux must be 0.
293 *
294 * (*) fsconfig_set_string: A string value is specified. The parameter can be
295 * expecting boolean, integer, string or take a path. A conversion to an
296 * appropriate type will be attempted (which may include looking up as a
297 * path). @_value points to a NUL-terminated string and @aux must be 0.
298 *
299 * (*) fsconfig_set_binary: A binary blob is specified. @_value points to the
300 * blob and @aux indicates its size. The parameter must be expecting a
301 * blob.
302 *
303 * (*) fsconfig_set_path: A non-empty path is specified. The parameter must be
304 * expecting a path object. @_value points to a NUL-terminated string that
305 * is the path and @aux is a file descriptor at which to start a relative
306 * lookup or AT_FDCWD.
307 *
308 * (*) fsconfig_set_path_empty: As fsconfig_set_path, but with AT_EMPTY_PATH
309 * implied.
310 *
311 * (*) fsconfig_set_fd: An open file descriptor is specified. @_value must be
312 * NULL and @aux indicates the file descriptor.
313 */
314SYSCALL_DEFINE5(fsconfig,
315 int, fd,
316 unsigned int, cmd,
317 const char __user *, _key,
318 const void __user *, _value,
319 int, aux)
320{
321 struct fs_context *fc;
322 struct fd f;
323 int ret;
aa1918f9 324 int lookup_flags = 0;
ecdab150
DH
325
326 struct fs_parameter param = {
327 .type = fs_value_is_undefined,
328 };
329
330 if (fd < 0)
331 return -EINVAL;
332
333 switch (cmd) {
334 case FSCONFIG_SET_FLAG:
335 if (!_key || _value || aux)
336 return -EINVAL;
337 break;
338 case FSCONFIG_SET_STRING:
339 if (!_key || !_value || aux)
340 return -EINVAL;
341 break;
342 case FSCONFIG_SET_BINARY:
343 if (!_key || !_value || aux <= 0 || aux > 1024 * 1024)
344 return -EINVAL;
345 break;
346 case FSCONFIG_SET_PATH:
347 case FSCONFIG_SET_PATH_EMPTY:
348 if (!_key || !_value || (aux != AT_FDCWD && aux < 0))
349 return -EINVAL;
350 break;
351 case FSCONFIG_SET_FD:
352 if (!_key || _value || aux < 0)
353 return -EINVAL;
354 break;
355 case FSCONFIG_CMD_CREATE:
356 case FSCONFIG_CMD_RECONFIGURE:
357 if (_key || _value || aux)
358 return -EINVAL;
359 break;
360 default:
361 return -EOPNOTSUPP;
362 }
363
364 f = fdget(fd);
365 if (!f.file)
366 return -EBADF;
367 ret = -EINVAL;
368 if (f.file->f_op != &fscontext_fops)
369 goto out_f;
370
371 fc = f.file->private_data;
372 if (fc->ops == &legacy_fs_context_ops) {
373 switch (cmd) {
374 case FSCONFIG_SET_BINARY:
375 case FSCONFIG_SET_PATH:
376 case FSCONFIG_SET_PATH_EMPTY:
377 case FSCONFIG_SET_FD:
378 ret = -EOPNOTSUPP;
379 goto out_f;
380 }
381 }
382
383 if (_key) {
384 param.key = strndup_user(_key, 256);
385 if (IS_ERR(param.key)) {
386 ret = PTR_ERR(param.key);
387 goto out_f;
388 }
389 }
390
391 switch (cmd) {
392 case FSCONFIG_SET_FLAG:
393 param.type = fs_value_is_flag;
394 break;
395 case FSCONFIG_SET_STRING:
396 param.type = fs_value_is_string;
397 param.string = strndup_user(_value, 256);
398 if (IS_ERR(param.string)) {
399 ret = PTR_ERR(param.string);
400 goto out_key;
401 }
402 param.size = strlen(param.string);
403 break;
404 case FSCONFIG_SET_BINARY:
405 param.type = fs_value_is_blob;
406 param.size = aux;
407 param.blob = memdup_user_nul(_value, aux);
408 if (IS_ERR(param.blob)) {
409 ret = PTR_ERR(param.blob);
410 goto out_key;
411 }
412 break;
aa1918f9
AV
413 case FSCONFIG_SET_PATH_EMPTY:
414 lookup_flags = LOOKUP_EMPTY;
df561f66 415 fallthrough;
ecdab150
DH
416 case FSCONFIG_SET_PATH:
417 param.type = fs_value_is_filename;
aa1918f9 418 param.name = getname_flags(_value, lookup_flags, NULL);
ecdab150
DH
419 if (IS_ERR(param.name)) {
420 ret = PTR_ERR(param.name);
421 goto out_key;
422 }
423 param.dirfd = aux;
424 param.size = strlen(param.name->name);
425 break;
426 case FSCONFIG_SET_FD:
427 param.type = fs_value_is_file;
428 ret = -EBADF;
429 param.file = fget(aux);
430 if (!param.file)
431 goto out_key;
432 break;
433 default:
434 break;
435 }
436
437 ret = mutex_lock_interruptible(&fc->uapi_mutex);
438 if (ret == 0) {
439 ret = vfs_fsconfig_locked(fc, cmd, &param);
440 mutex_unlock(&fc->uapi_mutex);
441 }
442
443 /* Clean up the our record of any value that we obtained from
444 * userspace. Note that the value may have been stolen by the LSM or
445 * filesystem, in which case the value pointer will have been cleared.
446 */
447 switch (cmd) {
448 case FSCONFIG_SET_STRING:
449 case FSCONFIG_SET_BINARY:
450 kfree(param.string);
451 break;
452 case FSCONFIG_SET_PATH:
453 case FSCONFIG_SET_PATH_EMPTY:
454 if (param.name)
455 putname(param.name);
456 break;
457 case FSCONFIG_SET_FD:
458 if (param.file)
459 fput(param.file);
460 break;
461 default:
462 break;
463 }
464out_key:
465 kfree(param.key);
466out_f:
467 fdput(f);
468 return ret;
469}