]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
33d3dfff | 2 | #include <linux/fanotify.h> |
11637e4b | 3 | #include <linux/fcntl.h> |
2a3edf86 | 4 | #include <linux/file.h> |
11637e4b | 5 | #include <linux/fs.h> |
52c923dd | 6 | #include <linux/anon_inodes.h> |
11637e4b | 7 | #include <linux/fsnotify_backend.h> |
2a3edf86 | 8 | #include <linux/init.h> |
a1014f10 | 9 | #include <linux/mount.h> |
2a3edf86 | 10 | #include <linux/namei.h> |
a1014f10 | 11 | #include <linux/poll.h> |
11637e4b EP |
12 | #include <linux/security.h> |
13 | #include <linux/syscalls.h> | |
e4e047a2 | 14 | #include <linux/slab.h> |
2a3edf86 | 15 | #include <linux/types.h> |
a1014f10 | 16 | #include <linux/uaccess.h> |
91c2e0bc | 17 | #include <linux/compat.h> |
174cd4b1 | 18 | #include <linux/sched/signal.h> |
d46eb14b | 19 | #include <linux/memcontrol.h> |
a8b13aa2 AG |
20 | #include <linux/statfs.h> |
21 | #include <linux/exportfs.h> | |
a1014f10 EP |
22 | |
23 | #include <asm/ioctls.h> | |
11637e4b | 24 | |
c63181e6 | 25 | #include "../../mount.h" |
be77196b | 26 | #include "../fdinfo.h" |
7053aee2 | 27 | #include "fanotify.h" |
c63181e6 | 28 | |
2529a0df | 29 | #define FANOTIFY_DEFAULT_MAX_EVENTS 16384 |
e7099d8a | 30 | #define FANOTIFY_DEFAULT_MAX_MARKS 8192 |
4afeff85 | 31 | #define FANOTIFY_DEFAULT_MAX_LISTENERS 128 |
2529a0df | 32 | |
48149e9d HS |
33 | /* |
34 | * All flags that may be specified in parameter event_f_flags of fanotify_init. | |
35 | * | |
36 | * Internal and external open flags are stored together in field f_flags of | |
37 | * struct file. Only external open flags shall be allowed in event_f_flags. | |
38 | * Internal flags like FMODE_NONOTIFY, FMODE_EXEC, FMODE_NOCMTIME shall be | |
39 | * excluded. | |
40 | */ | |
41 | #define FANOTIFY_INIT_ALL_EVENT_F_BITS ( \ | |
42 | O_ACCMODE | O_APPEND | O_NONBLOCK | \ | |
43 | __O_SYNC | O_DSYNC | O_CLOEXEC | \ | |
44 | O_LARGEFILE | O_NOATIME ) | |
45 | ||
33d3dfff | 46 | extern const struct fsnotify_ops fanotify_fsnotify_ops; |
11637e4b | 47 | |
054c636e | 48 | struct kmem_cache *fanotify_mark_cache __read_mostly; |
7088f357 JK |
49 | struct kmem_cache *fanotify_fid_event_cachep __read_mostly; |
50 | struct kmem_cache *fanotify_path_event_cachep __read_mostly; | |
f083441b | 51 | struct kmem_cache *fanotify_perm_event_cachep __read_mostly; |
2a3edf86 | 52 | |
5e469c83 | 53 | #define FANOTIFY_EVENT_ALIGN 4 |
44d705b0 AG |
54 | #define FANOTIFY_INFO_HDR_LEN \ |
55 | (sizeof(struct fanotify_event_info_fid) + sizeof(struct file_handle)) | |
5e469c83 | 56 | |
44d705b0 | 57 | static int fanotify_fid_info_len(int fh_len, int name_len) |
d766b553 | 58 | { |
44d705b0 AG |
59 | int info_len = fh_len; |
60 | ||
61 | if (name_len) | |
62 | info_len += name_len + 1; | |
63 | ||
64 | return roundup(FANOTIFY_INFO_HDR_LEN + info_len, FANOTIFY_EVENT_ALIGN); | |
d766b553 AG |
65 | } |
66 | ||
929943b3 AG |
67 | static int fanotify_event_info_len(unsigned int fid_mode, |
68 | struct fanotify_event *event) | |
5e469c83 | 69 | { |
f454fa61 AG |
70 | struct fanotify_info *info = fanotify_event_info(event); |
71 | int dir_fh_len = fanotify_event_dir_fh_len(event); | |
afc894c7 | 72 | int fh_len = fanotify_event_object_fh_len(event); |
f454fa61 | 73 | int info_len = 0; |
929943b3 | 74 | int dot_len = 0; |
f454fa61 | 75 | |
929943b3 | 76 | if (dir_fh_len) { |
f454fa61 | 77 | info_len += fanotify_fid_info_len(dir_fh_len, info->name_len); |
929943b3 AG |
78 | } else if ((fid_mode & FAN_REPORT_NAME) && (event->mask & FAN_ONDIR)) { |
79 | /* | |
80 | * With group flag FAN_REPORT_NAME, if name was not recorded in | |
81 | * event on a directory, we will report the name ".". | |
82 | */ | |
83 | dot_len = 1; | |
84 | } | |
afc894c7 | 85 | |
44d705b0 | 86 | if (fh_len) |
929943b3 | 87 | info_len += fanotify_fid_info_len(fh_len, dot_len); |
44d705b0 | 88 | |
44d705b0 | 89 | return info_len; |
5e469c83 AG |
90 | } |
91 | ||
a1014f10 | 92 | /* |
7088f357 | 93 | * Get an fanotify notification event if one exists and is small |
a1014f10 | 94 | * enough to fit in "count". Return an error pointer if the count |
40873284 JK |
95 | * is not large enough. When permission event is dequeued, its state is |
96 | * updated accordingly. | |
a1014f10 | 97 | */ |
7088f357 | 98 | static struct fanotify_event *get_one_event(struct fsnotify_group *group, |
a1014f10 EP |
99 | size_t count) |
100 | { | |
5e469c83 | 101 | size_t event_size = FAN_EVENT_METADATA_LEN; |
7088f357 | 102 | struct fanotify_event *event = NULL; |
929943b3 | 103 | unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS); |
a1014f10 EP |
104 | |
105 | pr_debug("%s: group=%p count=%zd\n", __func__, group, count); | |
106 | ||
8c554466 | 107 | spin_lock(&group->notification_lock); |
a1014f10 | 108 | if (fsnotify_notify_queue_is_empty(group)) |
8c554466 | 109 | goto out; |
a1014f10 | 110 | |
929943b3 AG |
111 | if (fid_mode) { |
112 | event_size += fanotify_event_info_len(fid_mode, | |
8c554466 | 113 | FANOTIFY_E(fsnotify_peek_first_event(group))); |
5e469c83 AG |
114 | } |
115 | ||
8c554466 | 116 | if (event_size > count) { |
7088f357 | 117 | event = ERR_PTR(-EINVAL); |
8c554466 JK |
118 | goto out; |
119 | } | |
7088f357 JK |
120 | event = FANOTIFY_E(fsnotify_remove_first_event(group)); |
121 | if (fanotify_is_perm_event(event->mask)) | |
122 | FANOTIFY_PERM(event)->state = FAN_EVENT_REPORTED; | |
8c554466 JK |
123 | out: |
124 | spin_unlock(&group->notification_lock); | |
7088f357 | 125 | return event; |
a1014f10 EP |
126 | } |
127 | ||
a741c2fe | 128 | static int create_fd(struct fsnotify_group *group, struct path *path, |
7053aee2 | 129 | struct file **file) |
a1014f10 EP |
130 | { |
131 | int client_fd; | |
a1014f10 EP |
132 | struct file *new_file; |
133 | ||
0b37e097 | 134 | client_fd = get_unused_fd_flags(group->fanotify_data.f_flags); |
a1014f10 EP |
135 | if (client_fd < 0) |
136 | return client_fd; | |
137 | ||
a1014f10 EP |
138 | /* |
139 | * we need a new file handle for the userspace program so it can read even if it was | |
140 | * originally opened O_WRONLY. | |
141 | */ | |
a741c2fe JK |
142 | new_file = dentry_open(path, |
143 | group->fanotify_data.f_flags | FMODE_NONOTIFY, | |
144 | current_cred()); | |
a1014f10 EP |
145 | if (IS_ERR(new_file)) { |
146 | /* | |
147 | * we still send an event even if we can't open the file. this | |
148 | * can happen when say tasks are gone and we try to open their | |
149 | * /proc files or we try to open a WRONLY file like in sysfs | |
150 | * we just send the errno to userspace since there isn't much | |
151 | * else we can do. | |
152 | */ | |
153 | put_unused_fd(client_fd); | |
154 | client_fd = PTR_ERR(new_file); | |
155 | } else { | |
352e3b24 | 156 | *file = new_file; |
a1014f10 EP |
157 | } |
158 | ||
22aa425d | 159 | return client_fd; |
a1014f10 EP |
160 | } |
161 | ||
40873284 JK |
162 | /* |
163 | * Finish processing of permission event by setting it to ANSWERED state and | |
164 | * drop group->notification_lock. | |
165 | */ | |
166 | static void finish_permission_event(struct fsnotify_group *group, | |
167 | struct fanotify_perm_event *event, | |
168 | unsigned int response) | |
169 | __releases(&group->notification_lock) | |
170 | { | |
fabf7f29 JK |
171 | bool destroy = false; |
172 | ||
40873284 JK |
173 | assert_spin_locked(&group->notification_lock); |
174 | event->response = response; | |
fabf7f29 JK |
175 | if (event->state == FAN_EVENT_CANCELED) |
176 | destroy = true; | |
177 | else | |
178 | event->state = FAN_EVENT_ANSWERED; | |
40873284 | 179 | spin_unlock(&group->notification_lock); |
fabf7f29 JK |
180 | if (destroy) |
181 | fsnotify_destroy_event(group, &event->fae.fse); | |
40873284 JK |
182 | } |
183 | ||
b2d87909 EP |
184 | static int process_access_response(struct fsnotify_group *group, |
185 | struct fanotify_response *response_struct) | |
186 | { | |
33913997 | 187 | struct fanotify_perm_event *event; |
f083441b JK |
188 | int fd = response_struct->fd; |
189 | int response = response_struct->response; | |
b2d87909 EP |
190 | |
191 | pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group, | |
192 | fd, response); | |
193 | /* | |
194 | * make sure the response is valid, if invalid we do nothing and either | |
25985edc | 195 | * userspace can send a valid response or we will clean it up after the |
b2d87909 EP |
196 | * timeout |
197 | */ | |
de8cd83e | 198 | switch (response & ~FAN_AUDIT) { |
b2d87909 EP |
199 | case FAN_ALLOW: |
200 | case FAN_DENY: | |
201 | break; | |
202 | default: | |
203 | return -EINVAL; | |
204 | } | |
205 | ||
206 | if (fd < 0) | |
207 | return -EINVAL; | |
208 | ||
96a71f21 | 209 | if ((response & FAN_AUDIT) && !FAN_GROUP_FLAG(group, FAN_ENABLE_AUDIT)) |
de8cd83e SG |
210 | return -EINVAL; |
211 | ||
af6a5113 JK |
212 | spin_lock(&group->notification_lock); |
213 | list_for_each_entry(event, &group->fanotify_data.access_list, | |
214 | fae.fse.list) { | |
215 | if (event->fd != fd) | |
216 | continue; | |
b2d87909 | 217 | |
af6a5113 | 218 | list_del_init(&event->fae.fse.list); |
40873284 | 219 | finish_permission_event(group, event, response); |
af6a5113 JK |
220 | wake_up(&group->fanotify_data.access_waitq); |
221 | return 0; | |
222 | } | |
223 | spin_unlock(&group->notification_lock); | |
b2d87909 | 224 | |
af6a5113 | 225 | return -ENOENT; |
b2d87909 | 226 | } |
b2d87909 | 227 | |
44d705b0 | 228 | static int copy_info_to_user(__kernel_fsid_t *fsid, struct fanotify_fh *fh, |
83b7a598 | 229 | int info_type, const char *name, size_t name_len, |
44d705b0 | 230 | char __user *buf, size_t count) |
5e469c83 AG |
231 | { |
232 | struct fanotify_event_info_fid info = { }; | |
233 | struct file_handle handle = { }; | |
afc894c7 | 234 | unsigned char bounce[FANOTIFY_INLINE_FH_LEN], *fh_buf; |
cacfb956 | 235 | size_t fh_len = fh ? fh->len : 0; |
44d705b0 AG |
236 | size_t info_len = fanotify_fid_info_len(fh_len, name_len); |
237 | size_t len = info_len; | |
5e469c83 | 238 | |
44d705b0 AG |
239 | pr_debug("%s: fh_len=%zu name_len=%zu, info_len=%zu, count=%zu\n", |
240 | __func__, fh_len, name_len, info_len, count); | |
241 | ||
83b7a598 | 242 | if (!fh_len) |
5e469c83 AG |
243 | return 0; |
244 | ||
44d705b0 | 245 | if (WARN_ON_ONCE(len < sizeof(info) || len > count)) |
5e469c83 AG |
246 | return -EFAULT; |
247 | ||
44d705b0 AG |
248 | /* |
249 | * Copy event info fid header followed by variable sized file handle | |
250 | * and optionally followed by variable sized filename. | |
251 | */ | |
83b7a598 AG |
252 | switch (info_type) { |
253 | case FAN_EVENT_INFO_TYPE_FID: | |
254 | case FAN_EVENT_INFO_TYPE_DFID: | |
255 | if (WARN_ON_ONCE(name_len)) | |
256 | return -EFAULT; | |
257 | break; | |
258 | case FAN_EVENT_INFO_TYPE_DFID_NAME: | |
259 | if (WARN_ON_ONCE(!name || !name_len)) | |
260 | return -EFAULT; | |
261 | break; | |
262 | default: | |
263 | return -EFAULT; | |
264 | } | |
265 | ||
266 | info.hdr.info_type = info_type; | |
5e469c83 | 267 | info.hdr.len = len; |
d766b553 | 268 | info.fsid = *fsid; |
5e469c83 AG |
269 | if (copy_to_user(buf, &info, sizeof(info))) |
270 | return -EFAULT; | |
271 | ||
272 | buf += sizeof(info); | |
273 | len -= sizeof(info); | |
44d705b0 AG |
274 | if (WARN_ON_ONCE(len < sizeof(handle))) |
275 | return -EFAULT; | |
276 | ||
afc894c7 | 277 | handle.handle_type = fh->type; |
5e469c83 AG |
278 | handle.handle_bytes = fh_len; |
279 | if (copy_to_user(buf, &handle, sizeof(handle))) | |
280 | return -EFAULT; | |
281 | ||
282 | buf += sizeof(handle); | |
283 | len -= sizeof(handle); | |
44d705b0 AG |
284 | if (WARN_ON_ONCE(len < fh_len)) |
285 | return -EFAULT; | |
286 | ||
b2d22b6b | 287 | /* |
44d705b0 AG |
288 | * For an inline fh and inline file name, copy through stack to exclude |
289 | * the copy from usercopy hardening protections. | |
b2d22b6b | 290 | */ |
afc894c7 | 291 | fh_buf = fanotify_fh_buf(fh); |
b2d22b6b | 292 | if (fh_len <= FANOTIFY_INLINE_FH_LEN) { |
afc894c7 JK |
293 | memcpy(bounce, fh_buf, fh_len); |
294 | fh_buf = bounce; | |
b2d22b6b | 295 | } |
afc894c7 | 296 | if (copy_to_user(buf, fh_buf, fh_len)) |
5e469c83 AG |
297 | return -EFAULT; |
298 | ||
5e469c83 AG |
299 | buf += fh_len; |
300 | len -= fh_len; | |
44d705b0 AG |
301 | |
302 | if (name_len) { | |
303 | /* Copy the filename with terminating null */ | |
304 | name_len++; | |
305 | if (WARN_ON_ONCE(len < name_len)) | |
306 | return -EFAULT; | |
307 | ||
308 | if (copy_to_user(buf, name, name_len)) | |
309 | return -EFAULT; | |
310 | ||
311 | buf += name_len; | |
312 | len -= name_len; | |
313 | } | |
314 | ||
315 | /* Pad with 0's */ | |
5e469c83 AG |
316 | WARN_ON_ONCE(len < 0 || len >= FANOTIFY_EVENT_ALIGN); |
317 | if (len > 0 && clear_user(buf, len)) | |
318 | return -EFAULT; | |
319 | ||
44d705b0 | 320 | return info_len; |
5e469c83 AG |
321 | } |
322 | ||
a1014f10 | 323 | static ssize_t copy_event_to_user(struct fsnotify_group *group, |
7088f357 | 324 | struct fanotify_event *event, |
5b03a472 | 325 | char __user *buf, size_t count) |
a1014f10 | 326 | { |
bb2f7b45 | 327 | struct fanotify_event_metadata metadata; |
7088f357 | 328 | struct path *path = fanotify_event_path(event); |
f454fa61 | 329 | struct fanotify_info *info = fanotify_event_info(event); |
83b7a598 | 330 | unsigned int fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS); |
bb2f7b45 | 331 | struct file *f = NULL; |
e9e0c890 | 332 | int ret, fd = FAN_NOFD; |
83b7a598 | 333 | int info_type = 0; |
a1014f10 | 334 | |
7088f357 | 335 | pr_debug("%s: group=%p event=%p\n", __func__, group, event); |
a1014f10 | 336 | |
44d705b0 | 337 | metadata.event_len = FAN_EVENT_METADATA_LEN + |
929943b3 | 338 | fanotify_event_info_len(fid_mode, event); |
bb2f7b45 AG |
339 | metadata.metadata_len = FAN_EVENT_METADATA_LEN; |
340 | metadata.vers = FANOTIFY_METADATA_VERSION; | |
341 | metadata.reserved = 0; | |
342 | metadata.mask = event->mask & FANOTIFY_OUTGOING_EVENTS; | |
343 | metadata.pid = pid_vnr(event->pid); | |
344 | ||
44d705b0 | 345 | if (path && path->mnt && path->dentry) { |
afc894c7 JK |
346 | fd = create_fd(group, path, &f); |
347 | if (fd < 0) | |
348 | return fd; | |
bb2f7b45 AG |
349 | } |
350 | metadata.fd = fd; | |
b2d87909 | 351 | |
b2d87909 | 352 | ret = -EFAULT; |
5b03a472 KC |
353 | /* |
354 | * Sanity check copy size in case get_one_event() and | |
c5e443cb | 355 | * event_len sizes ever get out of sync. |
5b03a472 | 356 | */ |
bb2f7b45 | 357 | if (WARN_ON_ONCE(metadata.event_len > count)) |
5b03a472 | 358 | goto out_close_fd; |
bb2f7b45 | 359 | |
5e469c83 | 360 | if (copy_to_user(buf, &metadata, FAN_EVENT_METADATA_LEN)) |
352e3b24 AV |
361 | goto out_close_fd; |
362 | ||
44d705b0 AG |
363 | buf += FAN_EVENT_METADATA_LEN; |
364 | count -= FAN_EVENT_METADATA_LEN; | |
365 | ||
bb2f7b45 | 366 | if (fanotify_is_perm_event(event->mask)) |
7088f357 | 367 | FANOTIFY_PERM(event)->fd = fd; |
a1014f10 | 368 | |
44d705b0 | 369 | if (f) |
3587b1b0 | 370 | fd_install(fd, f); |
44d705b0 AG |
371 | |
372 | /* Event info records order is: dir fid + name, child fid */ | |
f454fa61 | 373 | if (fanotify_event_dir_fh_len(event)) { |
691d9763 AG |
374 | info_type = info->name_len ? FAN_EVENT_INFO_TYPE_DFID_NAME : |
375 | FAN_EVENT_INFO_TYPE_DFID; | |
44d705b0 | 376 | ret = copy_info_to_user(fanotify_event_fsid(event), |
f454fa61 | 377 | fanotify_info_dir_fh(info), |
83b7a598 | 378 | info_type, fanotify_info_name(info), |
f454fa61 | 379 | info->name_len, buf, count); |
5e469c83 AG |
380 | if (ret < 0) |
381 | return ret; | |
44d705b0 AG |
382 | |
383 | buf += ret; | |
384 | count -= ret; | |
385 | } | |
386 | ||
387 | if (fanotify_event_object_fh_len(event)) { | |
929943b3 AG |
388 | const char *dot = NULL; |
389 | int dot_len = 0; | |
390 | ||
83b7a598 AG |
391 | if (fid_mode == FAN_REPORT_FID || info_type) { |
392 | /* | |
393 | * With only group flag FAN_REPORT_FID only type FID is | |
394 | * reported. Second info record type is always FID. | |
395 | */ | |
396 | info_type = FAN_EVENT_INFO_TYPE_FID; | |
929943b3 AG |
397 | } else if ((fid_mode & FAN_REPORT_NAME) && |
398 | (event->mask & FAN_ONDIR)) { | |
399 | /* | |
400 | * With group flag FAN_REPORT_NAME, if name was not | |
401 | * recorded in an event on a directory, report the | |
402 | * name "." with info type DFID_NAME. | |
403 | */ | |
404 | info_type = FAN_EVENT_INFO_TYPE_DFID_NAME; | |
405 | dot = "."; | |
406 | dot_len = 1; | |
83b7a598 AG |
407 | } else if ((event->mask & ALL_FSNOTIFY_DIRENT_EVENTS) || |
408 | (event->mask & FAN_ONDIR)) { | |
409 | /* | |
410 | * With group flag FAN_REPORT_DIR_FID, a single info | |
411 | * record has type DFID for directory entry modification | |
412 | * event and for event on a directory. | |
413 | */ | |
414 | info_type = FAN_EVENT_INFO_TYPE_DFID; | |
415 | } else { | |
416 | /* | |
417 | * With group flags FAN_REPORT_DIR_FID|FAN_REPORT_FID, | |
418 | * a single info record has type FID for event on a | |
419 | * non-directory, when there is no directory to report. | |
420 | * For example, on FAN_DELETE_SELF event. | |
421 | */ | |
422 | info_type = FAN_EVENT_INFO_TYPE_FID; | |
423 | } | |
424 | ||
44d705b0 AG |
425 | ret = copy_info_to_user(fanotify_event_fsid(event), |
426 | fanotify_event_object_fh(event), | |
929943b3 | 427 | info_type, dot, dot_len, buf, count); |
44d705b0 AG |
428 | if (ret < 0) |
429 | return ret; | |
430 | ||
431 | buf += ret; | |
432 | count -= ret; | |
5e469c83 AG |
433 | } |
434 | ||
bb2f7b45 | 435 | return metadata.event_len; |
b2d87909 | 436 | |
b2d87909 | 437 | out_close_fd: |
352e3b24 AV |
438 | if (fd != FAN_NOFD) { |
439 | put_unused_fd(fd); | |
440 | fput(f); | |
441 | } | |
b2d87909 | 442 | return ret; |
a1014f10 EP |
443 | } |
444 | ||
445 | /* intofiy userspace file descriptor functions */ | |
076ccb76 | 446 | static __poll_t fanotify_poll(struct file *file, poll_table *wait) |
a1014f10 EP |
447 | { |
448 | struct fsnotify_group *group = file->private_data; | |
076ccb76 | 449 | __poll_t ret = 0; |
a1014f10 EP |
450 | |
451 | poll_wait(file, &group->notification_waitq, wait); | |
c21dbe20 | 452 | spin_lock(&group->notification_lock); |
a1014f10 | 453 | if (!fsnotify_notify_queue_is_empty(group)) |
a9a08845 | 454 | ret = EPOLLIN | EPOLLRDNORM; |
c21dbe20 | 455 | spin_unlock(&group->notification_lock); |
a1014f10 EP |
456 | |
457 | return ret; | |
458 | } | |
459 | ||
460 | static ssize_t fanotify_read(struct file *file, char __user *buf, | |
461 | size_t count, loff_t *pos) | |
462 | { | |
463 | struct fsnotify_group *group; | |
7088f357 | 464 | struct fanotify_event *event; |
a1014f10 EP |
465 | char __user *start; |
466 | int ret; | |
536ebe9c | 467 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
a1014f10 EP |
468 | |
469 | start = buf; | |
470 | group = file->private_data; | |
471 | ||
472 | pr_debug("%s: group=%p\n", __func__, group); | |
473 | ||
536ebe9c | 474 | add_wait_queue(&group->notification_waitq, &wait); |
a1014f10 | 475 | while (1) { |
47aaabde JK |
476 | /* |
477 | * User can supply arbitrarily large buffer. Avoid softlockups | |
478 | * in case there are lots of available events. | |
479 | */ | |
480 | cond_resched(); | |
7088f357 JK |
481 | event = get_one_event(group, count); |
482 | if (IS_ERR(event)) { | |
483 | ret = PTR_ERR(event); | |
d8aaab4f JK |
484 | break; |
485 | } | |
486 | ||
7088f357 | 487 | if (!event) { |
d8aaab4f JK |
488 | ret = -EAGAIN; |
489 | if (file->f_flags & O_NONBLOCK) | |
a1014f10 | 490 | break; |
d8aaab4f JK |
491 | |
492 | ret = -ERESTARTSYS; | |
493 | if (signal_pending(current)) | |
494 | break; | |
495 | ||
496 | if (start != buf) | |
a1014f10 | 497 | break; |
536ebe9c PZ |
498 | |
499 | wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
a1014f10 EP |
500 | continue; |
501 | } | |
502 | ||
7088f357 | 503 | ret = copy_event_to_user(group, event, buf, count); |
4ff33aaf AG |
504 | if (unlikely(ret == -EOPENSTALE)) { |
505 | /* | |
506 | * We cannot report events with stale fd so drop it. | |
507 | * Setting ret to 0 will continue the event loop and | |
508 | * do the right thing if there are no more events to | |
509 | * read (i.e. return bytes read, -EAGAIN or wait). | |
510 | */ | |
511 | ret = 0; | |
512 | } | |
513 | ||
d8aaab4f JK |
514 | /* |
515 | * Permission events get queued to wait for response. Other | |
516 | * events can be destroyed now. | |
517 | */ | |
7088f357 JK |
518 | if (!fanotify_is_perm_event(event->mask)) { |
519 | fsnotify_destroy_event(group, &event->fse); | |
d507816b | 520 | } else { |
4ff33aaf | 521 | if (ret <= 0) { |
40873284 JK |
522 | spin_lock(&group->notification_lock); |
523 | finish_permission_event(group, | |
7088f357 | 524 | FANOTIFY_PERM(event), FAN_DENY); |
d507816b | 525 | wake_up(&group->fanotify_data.access_waitq); |
4ff33aaf AG |
526 | } else { |
527 | spin_lock(&group->notification_lock); | |
7088f357 | 528 | list_add_tail(&event->fse.list, |
4ff33aaf AG |
529 | &group->fanotify_data.access_list); |
530 | spin_unlock(&group->notification_lock); | |
d507816b | 531 | } |
d507816b | 532 | } |
4ff33aaf AG |
533 | if (ret < 0) |
534 | break; | |
d8aaab4f JK |
535 | buf += ret; |
536 | count -= ret; | |
a1014f10 | 537 | } |
536ebe9c | 538 | remove_wait_queue(&group->notification_waitq, &wait); |
a1014f10 | 539 | |
a1014f10 EP |
540 | if (start != buf && ret != -EFAULT) |
541 | ret = buf - start; | |
542 | return ret; | |
543 | } | |
544 | ||
b2d87909 EP |
545 | static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos) |
546 | { | |
b2d87909 EP |
547 | struct fanotify_response response = { .fd = -1, .response = -1 }; |
548 | struct fsnotify_group *group; | |
549 | int ret; | |
550 | ||
6685df31 MS |
551 | if (!IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) |
552 | return -EINVAL; | |
553 | ||
b2d87909 EP |
554 | group = file->private_data; |
555 | ||
5e23663b FF |
556 | if (count < sizeof(response)) |
557 | return -EINVAL; | |
558 | ||
559 | count = sizeof(response); | |
b2d87909 EP |
560 | |
561 | pr_debug("%s: group=%p count=%zu\n", __func__, group, count); | |
562 | ||
563 | if (copy_from_user(&response, buf, count)) | |
564 | return -EFAULT; | |
565 | ||
566 | ret = process_access_response(group, &response); | |
567 | if (ret < 0) | |
568 | count = ret; | |
569 | ||
570 | return count; | |
b2d87909 EP |
571 | } |
572 | ||
52c923dd EP |
573 | static int fanotify_release(struct inode *ignored, struct file *file) |
574 | { | |
575 | struct fsnotify_group *group = file->private_data; | |
19ba54f4 | 576 | |
5838d444 | 577 | /* |
96d41019 JK |
578 | * Stop new events from arriving in the notification queue. since |
579 | * userspace cannot use fanotify fd anymore, no event can enter or | |
580 | * leave access_list by now either. | |
5838d444 | 581 | */ |
96d41019 | 582 | fsnotify_group_stop_queueing(group); |
2eebf582 | 583 | |
96d41019 JK |
584 | /* |
585 | * Process all permission events on access_list and notification queue | |
586 | * and simulate reply from userspace. | |
587 | */ | |
073f6552 | 588 | spin_lock(&group->notification_lock); |
ca6f8699 | 589 | while (!list_empty(&group->fanotify_data.access_list)) { |
7088f357 JK |
590 | struct fanotify_perm_event *event; |
591 | ||
ca6f8699 JK |
592 | event = list_first_entry(&group->fanotify_data.access_list, |
593 | struct fanotify_perm_event, fae.fse.list); | |
f083441b | 594 | list_del_init(&event->fae.fse.list); |
40873284 JK |
595 | finish_permission_event(group, event, FAN_ALLOW); |
596 | spin_lock(&group->notification_lock); | |
2eebf582 | 597 | } |
2eebf582 | 598 | |
5838d444 | 599 | /* |
96d41019 JK |
600 | * Destroy all non-permission events. For permission events just |
601 | * dequeue them and set the response. They will be freed once the | |
602 | * response is consumed and fanotify_get_response() returns. | |
5838d444 | 603 | */ |
96d41019 | 604 | while (!fsnotify_notify_queue_is_empty(group)) { |
7088f357 JK |
605 | struct fanotify_event *event; |
606 | ||
607 | event = FANOTIFY_E(fsnotify_remove_first_event(group)); | |
608 | if (!(event->mask & FANOTIFY_PERM_EVENTS)) { | |
c21dbe20 | 609 | spin_unlock(&group->notification_lock); |
7088f357 | 610 | fsnotify_destroy_event(group, &event->fse); |
6685df31 | 611 | } else { |
7088f357 | 612 | finish_permission_event(group, FANOTIFY_PERM(event), |
40873284 | 613 | FAN_ALLOW); |
6685df31 | 614 | } |
40873284 | 615 | spin_lock(&group->notification_lock); |
96d41019 | 616 | } |
c21dbe20 | 617 | spin_unlock(&group->notification_lock); |
96d41019 JK |
618 | |
619 | /* Response for all permission events it set, wakeup waiters */ | |
2eebf582 | 620 | wake_up(&group->fanotify_data.access_waitq); |
0a6b6bd5 | 621 | |
52c923dd | 622 | /* matches the fanotify_init->fsnotify_alloc_group */ |
d8153d4d | 623 | fsnotify_destroy_group(group); |
52c923dd EP |
624 | |
625 | return 0; | |
626 | } | |
627 | ||
a1014f10 EP |
628 | static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg) |
629 | { | |
630 | struct fsnotify_group *group; | |
7053aee2 | 631 | struct fsnotify_event *fsn_event; |
a1014f10 EP |
632 | void __user *p; |
633 | int ret = -ENOTTY; | |
634 | size_t send_len = 0; | |
635 | ||
636 | group = file->private_data; | |
637 | ||
638 | p = (void __user *) arg; | |
639 | ||
640 | switch (cmd) { | |
641 | case FIONREAD: | |
c21dbe20 | 642 | spin_lock(&group->notification_lock); |
7053aee2 | 643 | list_for_each_entry(fsn_event, &group->notification_list, list) |
a1014f10 | 644 | send_len += FAN_EVENT_METADATA_LEN; |
c21dbe20 | 645 | spin_unlock(&group->notification_lock); |
a1014f10 EP |
646 | ret = put_user(send_len, (int __user *) p); |
647 | break; | |
648 | } | |
649 | ||
650 | return ret; | |
651 | } | |
652 | ||
52c923dd | 653 | static const struct file_operations fanotify_fops = { |
be77196b | 654 | .show_fdinfo = fanotify_show_fdinfo, |
a1014f10 EP |
655 | .poll = fanotify_poll, |
656 | .read = fanotify_read, | |
b2d87909 | 657 | .write = fanotify_write, |
52c923dd EP |
658 | .fasync = NULL, |
659 | .release = fanotify_release, | |
a1014f10 | 660 | .unlocked_ioctl = fanotify_ioctl, |
1832f2d8 | 661 | .compat_ioctl = compat_ptr_ioctl, |
6038f373 | 662 | .llseek = noop_llseek, |
52c923dd EP |
663 | }; |
664 | ||
2a3edf86 | 665 | static int fanotify_find_path(int dfd, const char __user *filename, |
ac5656d8 AG |
666 | struct path *path, unsigned int flags, __u64 mask, |
667 | unsigned int obj_type) | |
2a3edf86 EP |
668 | { |
669 | int ret; | |
670 | ||
671 | pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__, | |
672 | dfd, filename, flags); | |
673 | ||
674 | if (filename == NULL) { | |
2903ff01 | 675 | struct fd f = fdget(dfd); |
2a3edf86 EP |
676 | |
677 | ret = -EBADF; | |
2903ff01 | 678 | if (!f.file) |
2a3edf86 EP |
679 | goto out; |
680 | ||
681 | ret = -ENOTDIR; | |
682 | if ((flags & FAN_MARK_ONLYDIR) && | |
496ad9aa | 683 | !(S_ISDIR(file_inode(f.file)->i_mode))) { |
2903ff01 | 684 | fdput(f); |
2a3edf86 EP |
685 | goto out; |
686 | } | |
687 | ||
2903ff01 | 688 | *path = f.file->f_path; |
2a3edf86 | 689 | path_get(path); |
2903ff01 | 690 | fdput(f); |
2a3edf86 EP |
691 | } else { |
692 | unsigned int lookup_flags = 0; | |
693 | ||
694 | if (!(flags & FAN_MARK_DONT_FOLLOW)) | |
695 | lookup_flags |= LOOKUP_FOLLOW; | |
696 | if (flags & FAN_MARK_ONLYDIR) | |
697 | lookup_flags |= LOOKUP_DIRECTORY; | |
698 | ||
699 | ret = user_path_at(dfd, filename, lookup_flags, path); | |
700 | if (ret) | |
701 | goto out; | |
702 | } | |
703 | ||
704 | /* you can only watch an inode if you have read permissions on it */ | |
705 | ret = inode_permission(path->dentry->d_inode, MAY_READ); | |
ac5656d8 AG |
706 | if (ret) { |
707 | path_put(path); | |
708 | goto out; | |
709 | } | |
710 | ||
711 | ret = security_path_notify(path, mask, obj_type); | |
2a3edf86 EP |
712 | if (ret) |
713 | path_put(path); | |
ac5656d8 | 714 | |
2a3edf86 EP |
715 | out: |
716 | return ret; | |
717 | } | |
718 | ||
b9e4e3bd | 719 | static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark, |
4ed6814a AG |
720 | __u32 mask, unsigned int flags, |
721 | __u32 umask, int *destroy) | |
088b09b0 | 722 | { |
d2c1874c | 723 | __u32 oldmask = 0; |
088b09b0 | 724 | |
4ed6814a AG |
725 | /* umask bits cannot be removed by user */ |
726 | mask &= ~umask; | |
088b09b0 | 727 | spin_lock(&fsn_mark->lock); |
b9e4e3bd EP |
728 | if (!(flags & FAN_MARK_IGNORED_MASK)) { |
729 | oldmask = fsn_mark->mask; | |
a72fd224 | 730 | fsn_mark->mask &= ~mask; |
b9e4e3bd | 731 | } else { |
a72fd224 | 732 | fsn_mark->ignored_mask &= ~mask; |
b9e4e3bd | 733 | } |
4ed6814a AG |
734 | /* |
735 | * We need to keep the mark around even if remaining mask cannot | |
736 | * result in any events (e.g. mask == FAN_ONDIR) to support incremenal | |
737 | * changes to the mask. | |
738 | * Destroy mark when only umask bits remain. | |
739 | */ | |
740 | *destroy = !((fsn_mark->mask | fsn_mark->ignored_mask) & ~umask); | |
088b09b0 AG |
741 | spin_unlock(&fsn_mark->lock); |
742 | ||
088b09b0 AG |
743 | return mask & oldmask; |
744 | } | |
745 | ||
eaa2c6b0 AG |
746 | static int fanotify_remove_mark(struct fsnotify_group *group, |
747 | fsnotify_connp_t *connp, __u32 mask, | |
4ed6814a | 748 | unsigned int flags, __u32 umask) |
88826276 EP |
749 | { |
750 | struct fsnotify_mark *fsn_mark = NULL; | |
088b09b0 | 751 | __u32 removed; |
6dfbd149 | 752 | int destroy_mark; |
88826276 | 753 | |
7b18527c | 754 | mutex_lock(&group->mark_mutex); |
eaa2c6b0 | 755 | fsn_mark = fsnotify_find_mark(connp, group); |
7b18527c LS |
756 | if (!fsn_mark) { |
757 | mutex_unlock(&group->mark_mutex); | |
f3640192 | 758 | return -ENOENT; |
7b18527c | 759 | } |
88826276 | 760 | |
6dfbd149 | 761 | removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags, |
4ed6814a | 762 | umask, &destroy_mark); |
3ac70bfc AG |
763 | if (removed & fsnotify_conn_mask(fsn_mark->connector)) |
764 | fsnotify_recalc_mask(fsn_mark->connector); | |
6dfbd149 | 765 | if (destroy_mark) |
4712e722 | 766 | fsnotify_detach_mark(fsn_mark); |
7b18527c | 767 | mutex_unlock(&group->mark_mutex); |
4712e722 JK |
768 | if (destroy_mark) |
769 | fsnotify_free_mark(fsn_mark); | |
6dfbd149 | 770 | |
eaa2c6b0 | 771 | /* matches the fsnotify_find_mark() */ |
f3640192 | 772 | fsnotify_put_mark(fsn_mark); |
f3640192 AG |
773 | return 0; |
774 | } | |
2a3edf86 | 775 | |
eaa2c6b0 AG |
776 | static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group, |
777 | struct vfsmount *mnt, __u32 mask, | |
4ed6814a | 778 | unsigned int flags, __u32 umask) |
eaa2c6b0 AG |
779 | { |
780 | return fanotify_remove_mark(group, &real_mount(mnt)->mnt_fsnotify_marks, | |
4ed6814a | 781 | mask, flags, umask); |
eaa2c6b0 AG |
782 | } |
783 | ||
d54f4fba | 784 | static int fanotify_remove_sb_mark(struct fsnotify_group *group, |
4ed6814a AG |
785 | struct super_block *sb, __u32 mask, |
786 | unsigned int flags, __u32 umask) | |
d54f4fba | 787 | { |
4ed6814a AG |
788 | return fanotify_remove_mark(group, &sb->s_fsnotify_marks, mask, |
789 | flags, umask); | |
d54f4fba AG |
790 | } |
791 | ||
f3640192 | 792 | static int fanotify_remove_inode_mark(struct fsnotify_group *group, |
b9e4e3bd | 793 | struct inode *inode, __u32 mask, |
4ed6814a | 794 | unsigned int flags, __u32 umask) |
f3640192 | 795 | { |
eaa2c6b0 | 796 | return fanotify_remove_mark(group, &inode->i_fsnotify_marks, mask, |
4ed6814a | 797 | flags, umask); |
2a3edf86 EP |
798 | } |
799 | ||
b9e4e3bd EP |
800 | static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark, |
801 | __u32 mask, | |
802 | unsigned int flags) | |
912ee394 | 803 | { |
192ca4d1 | 804 | __u32 oldmask = -1; |
912ee394 AG |
805 | |
806 | spin_lock(&fsn_mark->lock); | |
b9e4e3bd EP |
807 | if (!(flags & FAN_MARK_IGNORED_MASK)) { |
808 | oldmask = fsn_mark->mask; | |
a72fd224 | 809 | fsn_mark->mask |= mask; |
b9e4e3bd | 810 | } else { |
a72fd224 | 811 | fsn_mark->ignored_mask |= mask; |
c9778a98 EP |
812 | if (flags & FAN_MARK_IGNORED_SURV_MODIFY) |
813 | fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY; | |
b9e4e3bd | 814 | } |
912ee394 AG |
815 | spin_unlock(&fsn_mark->lock); |
816 | ||
817 | return mask & ~oldmask; | |
818 | } | |
819 | ||
5e9c070c | 820 | static struct fsnotify_mark *fanotify_add_new_mark(struct fsnotify_group *group, |
b812a9f5 | 821 | fsnotify_connp_t *connp, |
77115225 AG |
822 | unsigned int type, |
823 | __kernel_fsid_t *fsid) | |
5e9c070c LS |
824 | { |
825 | struct fsnotify_mark *mark; | |
826 | int ret; | |
827 | ||
828 | if (atomic_read(&group->num_marks) > group->fanotify_data.max_marks) | |
829 | return ERR_PTR(-ENOSPC); | |
830 | ||
831 | mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL); | |
832 | if (!mark) | |
833 | return ERR_PTR(-ENOMEM); | |
834 | ||
054c636e | 835 | fsnotify_init_mark(mark, group); |
77115225 | 836 | ret = fsnotify_add_mark_locked(mark, connp, type, 0, fsid); |
5e9c070c LS |
837 | if (ret) { |
838 | fsnotify_put_mark(mark); | |
839 | return ERR_PTR(ret); | |
840 | } | |
841 | ||
842 | return mark; | |
843 | } | |
844 | ||
845 | ||
eaa2c6b0 AG |
846 | static int fanotify_add_mark(struct fsnotify_group *group, |
847 | fsnotify_connp_t *connp, unsigned int type, | |
77115225 AG |
848 | __u32 mask, unsigned int flags, |
849 | __kernel_fsid_t *fsid) | |
2a3edf86 EP |
850 | { |
851 | struct fsnotify_mark *fsn_mark; | |
912ee394 | 852 | __u32 added; |
2a3edf86 | 853 | |
7b18527c | 854 | mutex_lock(&group->mark_mutex); |
b812a9f5 | 855 | fsn_mark = fsnotify_find_mark(connp, group); |
88826276 | 856 | if (!fsn_mark) { |
77115225 | 857 | fsn_mark = fanotify_add_new_mark(group, connp, type, fsid); |
5e9c070c | 858 | if (IS_ERR(fsn_mark)) { |
7b18527c | 859 | mutex_unlock(&group->mark_mutex); |
5e9c070c | 860 | return PTR_ERR(fsn_mark); |
7b18527c | 861 | } |
88826276 | 862 | } |
b9e4e3bd | 863 | added = fanotify_mark_add_to_mask(fsn_mark, mask, flags); |
3ac70bfc AG |
864 | if (added & ~fsnotify_conn_mask(fsn_mark->connector)) |
865 | fsnotify_recalc_mask(fsn_mark->connector); | |
c9747640 | 866 | mutex_unlock(&group->mark_mutex); |
5e9c070c | 867 | |
fa218ab9 | 868 | fsnotify_put_mark(fsn_mark); |
5e9c070c | 869 | return 0; |
88826276 EP |
870 | } |
871 | ||
eaa2c6b0 AG |
872 | static int fanotify_add_vfsmount_mark(struct fsnotify_group *group, |
873 | struct vfsmount *mnt, __u32 mask, | |
77115225 | 874 | unsigned int flags, __kernel_fsid_t *fsid) |
eaa2c6b0 AG |
875 | { |
876 | return fanotify_add_mark(group, &real_mount(mnt)->mnt_fsnotify_marks, | |
77115225 | 877 | FSNOTIFY_OBJ_TYPE_VFSMOUNT, mask, flags, fsid); |
eaa2c6b0 AG |
878 | } |
879 | ||
d54f4fba | 880 | static int fanotify_add_sb_mark(struct fsnotify_group *group, |
77115225 AG |
881 | struct super_block *sb, __u32 mask, |
882 | unsigned int flags, __kernel_fsid_t *fsid) | |
d54f4fba AG |
883 | { |
884 | return fanotify_add_mark(group, &sb->s_fsnotify_marks, | |
77115225 | 885 | FSNOTIFY_OBJ_TYPE_SB, mask, flags, fsid); |
d54f4fba AG |
886 | } |
887 | ||
52202dfb | 888 | static int fanotify_add_inode_mark(struct fsnotify_group *group, |
b9e4e3bd | 889 | struct inode *inode, __u32 mask, |
77115225 | 890 | unsigned int flags, __kernel_fsid_t *fsid) |
88826276 | 891 | { |
88826276 | 892 | pr_debug("%s: group=%p inode=%p\n", __func__, group, inode); |
2a3edf86 | 893 | |
5322a59f EP |
894 | /* |
895 | * If some other task has this inode open for write we should not add | |
896 | * an ignored mark, unless that ignored mark is supposed to survive | |
897 | * modification changes anyway. | |
898 | */ | |
899 | if ((flags & FAN_MARK_IGNORED_MASK) && | |
900 | !(flags & FAN_MARK_IGNORED_SURV_MODIFY) && | |
ac9498d6 | 901 | inode_is_open_for_write(inode)) |
5322a59f EP |
902 | return 0; |
903 | ||
eaa2c6b0 | 904 | return fanotify_add_mark(group, &inode->i_fsnotify_marks, |
77115225 | 905 | FSNOTIFY_OBJ_TYPE_INODE, mask, flags, fsid); |
88826276 | 906 | } |
2a3edf86 | 907 | |
b8a6c3a2 AG |
908 | static struct fsnotify_event *fanotify_alloc_overflow_event(void) |
909 | { | |
910 | struct fanotify_event *oevent; | |
911 | ||
912 | oevent = kmalloc(sizeof(*oevent), GFP_KERNEL_ACCOUNT); | |
913 | if (!oevent) | |
914 | return NULL; | |
915 | ||
916 | fanotify_init_event(oevent, 0, FS_Q_OVERFLOW); | |
917 | oevent->type = FANOTIFY_EVENT_TYPE_OVERFLOW; | |
918 | ||
919 | return &oevent->fse; | |
920 | } | |
921 | ||
52c923dd | 922 | /* fanotify syscalls */ |
08ae8938 | 923 | SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags) |
11637e4b | 924 | { |
52c923dd EP |
925 | struct fsnotify_group *group; |
926 | int f_flags, fd; | |
4afeff85 | 927 | struct user_struct *user; |
83b7a598 AG |
928 | unsigned int fid_mode = flags & FANOTIFY_FID_BITS; |
929 | unsigned int class = flags & FANOTIFY_CLASS_BITS; | |
52c923dd | 930 | |
96a71f21 AG |
931 | pr_debug("%s: flags=%x event_f_flags=%x\n", |
932 | __func__, flags, event_f_flags); | |
52c923dd | 933 | |
52c923dd | 934 | if (!capable(CAP_SYS_ADMIN)) |
a2f13ad0 | 935 | return -EPERM; |
52c923dd | 936 | |
de8cd83e | 937 | #ifdef CONFIG_AUDITSYSCALL |
23c9deeb | 938 | if (flags & ~(FANOTIFY_INIT_FLAGS | FAN_ENABLE_AUDIT)) |
de8cd83e | 939 | #else |
23c9deeb | 940 | if (flags & ~FANOTIFY_INIT_FLAGS) |
de8cd83e | 941 | #endif |
52c923dd EP |
942 | return -EINVAL; |
943 | ||
48149e9d HS |
944 | if (event_f_flags & ~FANOTIFY_INIT_ALL_EVENT_F_BITS) |
945 | return -EINVAL; | |
946 | ||
947 | switch (event_f_flags & O_ACCMODE) { | |
948 | case O_RDONLY: | |
949 | case O_RDWR: | |
950 | case O_WRONLY: | |
951 | break; | |
952 | default: | |
953 | return -EINVAL; | |
954 | } | |
955 | ||
83b7a598 | 956 | if (fid_mode && class != FAN_CLASS_NOTIF) |
a8b13aa2 AG |
957 | return -EINVAL; |
958 | ||
929943b3 | 959 | /* |
929943b3 | 960 | * Child name is reported with parent fid so requires dir fid. |
691d9763 | 961 | * We can report both child fid and dir fid with or without name. |
929943b3 | 962 | */ |
691d9763 | 963 | if ((fid_mode & FAN_REPORT_NAME) && !(fid_mode & FAN_REPORT_DIR_FID)) |
83b7a598 | 964 | return -EINVAL; |
83b7a598 | 965 | |
4afeff85 EP |
966 | user = get_current_user(); |
967 | if (atomic_read(&user->fanotify_listeners) > FANOTIFY_DEFAULT_MAX_LISTENERS) { | |
968 | free_uid(user); | |
969 | return -EMFILE; | |
970 | } | |
971 | ||
b2d87909 | 972 | f_flags = O_RDWR | FMODE_NONOTIFY; |
52c923dd EP |
973 | if (flags & FAN_CLOEXEC) |
974 | f_flags |= O_CLOEXEC; | |
975 | if (flags & FAN_NONBLOCK) | |
976 | f_flags |= O_NONBLOCK; | |
977 | ||
978 | /* fsnotify_alloc_group takes a ref. Dropped in fanotify_release */ | |
979 | group = fsnotify_alloc_group(&fanotify_fsnotify_ops); | |
26379198 EP |
980 | if (IS_ERR(group)) { |
981 | free_uid(user); | |
52c923dd | 982 | return PTR_ERR(group); |
26379198 | 983 | } |
52c923dd | 984 | |
4afeff85 | 985 | group->fanotify_data.user = user; |
96a71f21 | 986 | group->fanotify_data.flags = flags; |
4afeff85 | 987 | atomic_inc(&user->fanotify_listeners); |
d46eb14b | 988 | group->memcg = get_mem_cgroup_from_mm(current->mm); |
4afeff85 | 989 | |
b8a6c3a2 AG |
990 | group->overflow_event = fanotify_alloc_overflow_event(); |
991 | if (unlikely(!group->overflow_event)) { | |
ff57cd58 JK |
992 | fd = -ENOMEM; |
993 | goto out_destroy_group; | |
994 | } | |
ff57cd58 | 995 | |
1e2ee49f WW |
996 | if (force_o_largefile()) |
997 | event_f_flags |= O_LARGEFILE; | |
80af2588 | 998 | group->fanotify_data.f_flags = event_f_flags; |
9e66e423 EP |
999 | init_waitqueue_head(&group->fanotify_data.access_waitq); |
1000 | INIT_LIST_HEAD(&group->fanotify_data.access_list); | |
83b7a598 | 1001 | switch (class) { |
4231a235 EP |
1002 | case FAN_CLASS_NOTIF: |
1003 | group->priority = FS_PRIO_0; | |
1004 | break; | |
1005 | case FAN_CLASS_CONTENT: | |
1006 | group->priority = FS_PRIO_1; | |
1007 | break; | |
1008 | case FAN_CLASS_PRE_CONTENT: | |
1009 | group->priority = FS_PRIO_2; | |
1010 | break; | |
1011 | default: | |
1012 | fd = -EINVAL; | |
d8153d4d | 1013 | goto out_destroy_group; |
4231a235 | 1014 | } |
cb2d429f | 1015 | |
5dd03f55 EP |
1016 | if (flags & FAN_UNLIMITED_QUEUE) { |
1017 | fd = -EPERM; | |
1018 | if (!capable(CAP_SYS_ADMIN)) | |
d8153d4d | 1019 | goto out_destroy_group; |
5dd03f55 EP |
1020 | group->max_events = UINT_MAX; |
1021 | } else { | |
1022 | group->max_events = FANOTIFY_DEFAULT_MAX_EVENTS; | |
1023 | } | |
2529a0df | 1024 | |
ac7e22dc EP |
1025 | if (flags & FAN_UNLIMITED_MARKS) { |
1026 | fd = -EPERM; | |
1027 | if (!capable(CAP_SYS_ADMIN)) | |
d8153d4d | 1028 | goto out_destroy_group; |
ac7e22dc EP |
1029 | group->fanotify_data.max_marks = UINT_MAX; |
1030 | } else { | |
1031 | group->fanotify_data.max_marks = FANOTIFY_DEFAULT_MAX_MARKS; | |
1032 | } | |
e7099d8a | 1033 | |
de8cd83e SG |
1034 | if (flags & FAN_ENABLE_AUDIT) { |
1035 | fd = -EPERM; | |
1036 | if (!capable(CAP_AUDIT_WRITE)) | |
1037 | goto out_destroy_group; | |
de8cd83e SG |
1038 | } |
1039 | ||
52c923dd EP |
1040 | fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags); |
1041 | if (fd < 0) | |
d8153d4d | 1042 | goto out_destroy_group; |
52c923dd EP |
1043 | |
1044 | return fd; | |
1045 | ||
d8153d4d LS |
1046 | out_destroy_group: |
1047 | fsnotify_destroy_group(group); | |
52c923dd | 1048 | return fd; |
11637e4b | 1049 | } |
bbaa4168 | 1050 | |
a8b13aa2 | 1051 | /* Check if filesystem can encode a unique fid */ |
73072283 | 1052 | static int fanotify_test_fid(struct path *path, __kernel_fsid_t *fsid) |
a8b13aa2 | 1053 | { |
73072283 | 1054 | __kernel_fsid_t root_fsid; |
a8b13aa2 AG |
1055 | int err; |
1056 | ||
1057 | /* | |
1058 | * Make sure path is not in filesystem with zero fsid (e.g. tmpfs). | |
1059 | */ | |
73072283 | 1060 | err = vfs_get_fsid(path->dentry, fsid); |
a8b13aa2 AG |
1061 | if (err) |
1062 | return err; | |
1063 | ||
73072283 | 1064 | if (!fsid->val[0] && !fsid->val[1]) |
a8b13aa2 AG |
1065 | return -ENODEV; |
1066 | ||
1067 | /* | |
1068 | * Make sure path is not inside a filesystem subvolume (e.g. btrfs) | |
1069 | * which uses a different fsid than sb root. | |
1070 | */ | |
73072283 | 1071 | err = vfs_get_fsid(path->dentry->d_sb->s_root, &root_fsid); |
a8b13aa2 AG |
1072 | if (err) |
1073 | return err; | |
1074 | ||
73072283 AG |
1075 | if (root_fsid.val[0] != fsid->val[0] || |
1076 | root_fsid.val[1] != fsid->val[1]) | |
a8b13aa2 AG |
1077 | return -EXDEV; |
1078 | ||
1079 | /* | |
1080 | * We need to make sure that the file system supports at least | |
1081 | * encoding a file handle so user can use name_to_handle_at() to | |
1082 | * compare fid returned with event to the file handle of watched | |
1083 | * objects. However, name_to_handle_at() requires that the | |
1084 | * filesystem also supports decoding file handles. | |
1085 | */ | |
1086 | if (!path->dentry->d_sb->s_export_op || | |
1087 | !path->dentry->d_sb->s_export_op->fh_to_dentry) | |
1088 | return -EOPNOTSUPP; | |
1089 | ||
1090 | return 0; | |
1091 | } | |
1092 | ||
0b3b094a JK |
1093 | static int fanotify_events_supported(struct path *path, __u64 mask) |
1094 | { | |
1095 | /* | |
1096 | * Some filesystems such as 'proc' acquire unusual locks when opening | |
1097 | * files. For them fanotify permission events have high chances of | |
1098 | * deadlocking the system - open done when reporting fanotify event | |
1099 | * blocks on this "unusual" lock while another process holding the lock | |
1100 | * waits for fanotify permission event to be answered. Just disallow | |
1101 | * permission events for such filesystems. | |
1102 | */ | |
1103 | if (mask & FANOTIFY_PERM_EVENTS && | |
1104 | path->mnt->mnt_sb->s_type->fs_flags & FS_DISALLOW_NOTIFY_PERM) | |
1105 | return -EINVAL; | |
1106 | return 0; | |
1107 | } | |
1108 | ||
183caa3c DB |
1109 | static int do_fanotify_mark(int fanotify_fd, unsigned int flags, __u64 mask, |
1110 | int dfd, const char __user *pathname) | |
bbaa4168 | 1111 | { |
0ff21db9 EP |
1112 | struct inode *inode = NULL; |
1113 | struct vfsmount *mnt = NULL; | |
2a3edf86 | 1114 | struct fsnotify_group *group; |
2903ff01 | 1115 | struct fd f; |
2a3edf86 | 1116 | struct path path; |
73072283 | 1117 | __kernel_fsid_t __fsid, *fsid = NULL; |
bdd5a46f | 1118 | u32 valid_mask = FANOTIFY_EVENTS | FANOTIFY_EVENT_FLAGS; |
23c9deeb | 1119 | unsigned int mark_type = flags & FANOTIFY_MARK_TYPE_BITS; |
3ef86653 | 1120 | bool ignored = flags & FAN_MARK_IGNORED_MASK; |
d809daf1 | 1121 | unsigned int obj_type, fid_mode; |
85af5d92 | 1122 | u32 umask = 0; |
2903ff01 | 1123 | int ret; |
2a3edf86 EP |
1124 | |
1125 | pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n", | |
1126 | __func__, fanotify_fd, flags, dfd, pathname, mask); | |
1127 | ||
1128 | /* we only use the lower 32 bits as of right now. */ | |
1129 | if (mask & ((__u64)0xffffffff << 32)) | |
1130 | return -EINVAL; | |
1131 | ||
23c9deeb | 1132 | if (flags & ~FANOTIFY_MARK_FLAGS) |
88380fe6 | 1133 | return -EINVAL; |
d54f4fba AG |
1134 | |
1135 | switch (mark_type) { | |
1136 | case FAN_MARK_INODE: | |
ac5656d8 AG |
1137 | obj_type = FSNOTIFY_OBJ_TYPE_INODE; |
1138 | break; | |
d54f4fba | 1139 | case FAN_MARK_MOUNT: |
ac5656d8 AG |
1140 | obj_type = FSNOTIFY_OBJ_TYPE_VFSMOUNT; |
1141 | break; | |
d54f4fba | 1142 | case FAN_MARK_FILESYSTEM: |
ac5656d8 | 1143 | obj_type = FSNOTIFY_OBJ_TYPE_SB; |
d54f4fba AG |
1144 | break; |
1145 | default: | |
1146 | return -EINVAL; | |
1147 | } | |
1148 | ||
4d92604c | 1149 | switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) { |
df561f66 | 1150 | case FAN_MARK_ADD: |
88380fe6 | 1151 | case FAN_MARK_REMOVE: |
1734dee4 LS |
1152 | if (!mask) |
1153 | return -EINVAL; | |
cc299a98 | 1154 | break; |
4d92604c | 1155 | case FAN_MARK_FLUSH: |
23c9deeb | 1156 | if (flags & ~(FANOTIFY_MARK_TYPE_BITS | FAN_MARK_FLUSH)) |
cc299a98 | 1157 | return -EINVAL; |
88380fe6 AG |
1158 | break; |
1159 | default: | |
1160 | return -EINVAL; | |
1161 | } | |
8fcd6528 | 1162 | |
6685df31 | 1163 | if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) |
23c9deeb | 1164 | valid_mask |= FANOTIFY_PERM_EVENTS; |
6685df31 MS |
1165 | |
1166 | if (mask & ~valid_mask) | |
2a3edf86 EP |
1167 | return -EINVAL; |
1168 | ||
3ef86653 AG |
1169 | /* Event flags (ONDIR, ON_CHILD) are meaningless in ignored mask */ |
1170 | if (ignored) | |
1171 | mask &= ~FANOTIFY_EVENT_FLAGS; | |
1172 | ||
2903ff01 AV |
1173 | f = fdget(fanotify_fd); |
1174 | if (unlikely(!f.file)) | |
2a3edf86 EP |
1175 | return -EBADF; |
1176 | ||
1177 | /* verify that this is indeed an fanotify instance */ | |
1178 | ret = -EINVAL; | |
2903ff01 | 1179 | if (unlikely(f.file->f_op != &fanotify_fops)) |
2a3edf86 | 1180 | goto fput_and_out; |
2903ff01 | 1181 | group = f.file->private_data; |
4231a235 EP |
1182 | |
1183 | /* | |
1184 | * group->priority == FS_PRIO_0 == FAN_CLASS_NOTIF. These are not | |
1185 | * allowed to set permissions events. | |
1186 | */ | |
1187 | ret = -EINVAL; | |
23c9deeb | 1188 | if (mask & FANOTIFY_PERM_EVENTS && |
4231a235 EP |
1189 | group->priority == FS_PRIO_0) |
1190 | goto fput_and_out; | |
2a3edf86 | 1191 | |
235328d1 AG |
1192 | /* |
1193 | * Events with data type inode do not carry enough information to report | |
1194 | * event->fd, so we do not allow setting a mask for inode events unless | |
1195 | * group supports reporting fid. | |
1196 | * inode events are not supported on a mount mark, because they do not | |
1197 | * carry enough information (i.e. path) to be filtered by mount point. | |
1198 | */ | |
d809daf1 | 1199 | fid_mode = FAN_GROUP_FLAG(group, FANOTIFY_FID_BITS); |
235328d1 | 1200 | if (mask & FANOTIFY_INODE_EVENTS && |
d809daf1 | 1201 | (!fid_mode || mark_type == FAN_MARK_MOUNT)) |
235328d1 AG |
1202 | goto fput_and_out; |
1203 | ||
0a8dd2db HS |
1204 | if (flags & FAN_MARK_FLUSH) { |
1205 | ret = 0; | |
d54f4fba | 1206 | if (mark_type == FAN_MARK_MOUNT) |
0a8dd2db | 1207 | fsnotify_clear_vfsmount_marks_by_group(group); |
d54f4fba AG |
1208 | else if (mark_type == FAN_MARK_FILESYSTEM) |
1209 | fsnotify_clear_sb_marks_by_group(group); | |
0a8dd2db HS |
1210 | else |
1211 | fsnotify_clear_inode_marks_by_group(group); | |
1212 | goto fput_and_out; | |
1213 | } | |
1214 | ||
ac5656d8 AG |
1215 | ret = fanotify_find_path(dfd, pathname, &path, flags, |
1216 | (mask & ALL_FSNOTIFY_EVENTS), obj_type); | |
2a3edf86 EP |
1217 | if (ret) |
1218 | goto fput_and_out; | |
1219 | ||
0b3b094a JK |
1220 | if (flags & FAN_MARK_ADD) { |
1221 | ret = fanotify_events_supported(&path, mask); | |
1222 | if (ret) | |
1223 | goto path_put_and_out; | |
1224 | } | |
1225 | ||
d809daf1 | 1226 | if (fid_mode) { |
73072283 | 1227 | ret = fanotify_test_fid(&path, &__fsid); |
a8b13aa2 AG |
1228 | if (ret) |
1229 | goto path_put_and_out; | |
77115225 | 1230 | |
73072283 | 1231 | fsid = &__fsid; |
a8b13aa2 AG |
1232 | } |
1233 | ||
2a3edf86 | 1234 | /* inode held in place by reference to path; group by fget on fd */ |
d54f4fba | 1235 | if (mark_type == FAN_MARK_INODE) |
0ff21db9 EP |
1236 | inode = path.dentry->d_inode; |
1237 | else | |
1238 | mnt = path.mnt; | |
2a3edf86 | 1239 | |
85af5d92 AG |
1240 | /* Mask out FAN_EVENT_ON_CHILD flag for sb/mount/non-dir marks */ |
1241 | if (mnt || !S_ISDIR(inode->i_mode)) { | |
1242 | mask &= ~FAN_EVENT_ON_CHILD; | |
1243 | umask = FAN_EVENT_ON_CHILD; | |
51280637 AG |
1244 | /* |
1245 | * If group needs to report parent fid, register for getting | |
1246 | * events with parent/name info for non-directory. | |
1247 | */ | |
1248 | if ((fid_mode & FAN_REPORT_DIR_FID) && | |
1249 | (flags & FAN_MARK_ADD) && !ignored) | |
1250 | mask |= FAN_EVENT_ON_CHILD; | |
85af5d92 AG |
1251 | } |
1252 | ||
2a3edf86 | 1253 | /* create/update an inode mark */ |
0a8dd2db | 1254 | switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE)) { |
c6223f46 | 1255 | case FAN_MARK_ADD: |
d54f4fba | 1256 | if (mark_type == FAN_MARK_MOUNT) |
77115225 AG |
1257 | ret = fanotify_add_vfsmount_mark(group, mnt, mask, |
1258 | flags, fsid); | |
d54f4fba | 1259 | else if (mark_type == FAN_MARK_FILESYSTEM) |
77115225 AG |
1260 | ret = fanotify_add_sb_mark(group, mnt->mnt_sb, mask, |
1261 | flags, fsid); | |
0ff21db9 | 1262 | else |
77115225 AG |
1263 | ret = fanotify_add_inode_mark(group, inode, mask, |
1264 | flags, fsid); | |
c6223f46 AG |
1265 | break; |
1266 | case FAN_MARK_REMOVE: | |
d54f4fba | 1267 | if (mark_type == FAN_MARK_MOUNT) |
77115225 | 1268 | ret = fanotify_remove_vfsmount_mark(group, mnt, mask, |
85af5d92 | 1269 | flags, umask); |
d54f4fba | 1270 | else if (mark_type == FAN_MARK_FILESYSTEM) |
77115225 | 1271 | ret = fanotify_remove_sb_mark(group, mnt->mnt_sb, mask, |
85af5d92 | 1272 | flags, umask); |
f3640192 | 1273 | else |
77115225 | 1274 | ret = fanotify_remove_inode_mark(group, inode, mask, |
85af5d92 | 1275 | flags, umask); |
c6223f46 AG |
1276 | break; |
1277 | default: | |
1278 | ret = -EINVAL; | |
1279 | } | |
2a3edf86 | 1280 | |
a8b13aa2 | 1281 | path_put_and_out: |
2a3edf86 EP |
1282 | path_put(&path); |
1283 | fput_and_out: | |
2903ff01 | 1284 | fdput(f); |
2a3edf86 EP |
1285 | return ret; |
1286 | } | |
1287 | ||
2ca408d9 | 1288 | #ifndef CONFIG_ARCH_SPLIT_ARG64 |
183caa3c DB |
1289 | SYSCALL_DEFINE5(fanotify_mark, int, fanotify_fd, unsigned int, flags, |
1290 | __u64, mask, int, dfd, | |
1291 | const char __user *, pathname) | |
1292 | { | |
1293 | return do_fanotify_mark(fanotify_fd, flags, mask, dfd, pathname); | |
1294 | } | |
2ca408d9 | 1295 | #endif |
183caa3c | 1296 | |
2ca408d9 BG |
1297 | #if defined(CONFIG_ARCH_SPLIT_ARG64) || defined(CONFIG_COMPAT) |
1298 | SYSCALL32_DEFINE6(fanotify_mark, | |
91c2e0bc | 1299 | int, fanotify_fd, unsigned int, flags, |
2ca408d9 | 1300 | SC_ARG64(mask), int, dfd, |
91c2e0bc AV |
1301 | const char __user *, pathname) |
1302 | { | |
2ca408d9 BG |
1303 | return do_fanotify_mark(fanotify_fd, flags, SC_VAL64(__u64, mask), |
1304 | dfd, pathname); | |
91c2e0bc AV |
1305 | } |
1306 | #endif | |
1307 | ||
2a3edf86 | 1308 | /* |
ae0e47f0 | 1309 | * fanotify_user_setup - Our initialization function. Note that we cannot return |
2a3edf86 EP |
1310 | * error because we have compiled-in VFS hooks. So an (unlikely) failure here |
1311 | * must result in panic(). | |
1312 | */ | |
1313 | static int __init fanotify_user_setup(void) | |
1314 | { | |
929943b3 | 1315 | BUILD_BUG_ON(HWEIGHT32(FANOTIFY_INIT_FLAGS) != 10); |
bdd5a46f AG |
1316 | BUILD_BUG_ON(HWEIGHT32(FANOTIFY_MARK_FLAGS) != 9); |
1317 | ||
d46eb14b SB |
1318 | fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, |
1319 | SLAB_PANIC|SLAB_ACCOUNT); | |
7088f357 JK |
1320 | fanotify_fid_event_cachep = KMEM_CACHE(fanotify_fid_event, |
1321 | SLAB_PANIC); | |
1322 | fanotify_path_event_cachep = KMEM_CACHE(fanotify_path_event, | |
1323 | SLAB_PANIC); | |
6685df31 MS |
1324 | if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) { |
1325 | fanotify_perm_event_cachep = | |
33913997 | 1326 | KMEM_CACHE(fanotify_perm_event, SLAB_PANIC); |
6685df31 | 1327 | } |
2a3edf86 EP |
1328 | |
1329 | return 0; | |
bbaa4168 | 1330 | } |
2a3edf86 | 1331 | device_initcall(fanotify_user_setup); |