Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/kernel/seccomp.c | |
4 | * | |
5 | * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com> | |
6 | * | |
e2cfabdf WD |
7 | * Copyright (C) 2012 Google, Inc. |
8 | * Will Drewry <wad@chromium.org> | |
9 | * | |
10 | * This defines a simple but solid secure-computing facility. | |
11 | * | |
12 | * Mode 1 uses a fixed list of allowed system calls. | |
13 | * Mode 2 allows user-defined system call filters in the form | |
14 | * of Berkeley Packet Filters/Linux Socket Filters. | |
1da177e4 | 15 | */ |
e68f9d49 | 16 | #define pr_fmt(fmt) "seccomp: " fmt |
1da177e4 | 17 | |
0b5fa229 | 18 | #include <linux/refcount.h> |
85e7bac3 | 19 | #include <linux/audit.h> |
5b101740 | 20 | #include <linux/compat.h> |
b25e6716 | 21 | #include <linux/coredump.h> |
8e5f1ad1 | 22 | #include <linux/kmemleak.h> |
5c307089 KC |
23 | #include <linux/nospec.h> |
24 | #include <linux/prctl.h> | |
e2cfabdf | 25 | #include <linux/sched.h> |
68db0cf1 | 26 | #include <linux/sched/task_stack.h> |
e2cfabdf | 27 | #include <linux/seccomp.h> |
c8bee430 | 28 | #include <linux/slab.h> |
48dc92b9 | 29 | #include <linux/syscalls.h> |
8e5f1ad1 | 30 | #include <linux/sysctl.h> |
1da177e4 | 31 | |
a4412fc9 | 32 | #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER |
e2cfabdf | 33 | #include <asm/syscall.h> |
a4412fc9 | 34 | #endif |
e2cfabdf WD |
35 | |
36 | #ifdef CONFIG_SECCOMP_FILTER | |
6a21cc50 | 37 | #include <linux/file.h> |
e2cfabdf | 38 | #include <linux/filter.h> |
c2e1f2e3 | 39 | #include <linux/pid.h> |
fb0fadf9 | 40 | #include <linux/ptrace.h> |
e2cfabdf | 41 | #include <linux/security.h> |
e2cfabdf WD |
42 | #include <linux/tracehook.h> |
43 | #include <linux/uaccess.h> | |
6a21cc50 | 44 | #include <linux/anon_inodes.h> |
9f87dcf1 | 45 | #include <linux/lockdep.h> |
6a21cc50 | 46 | |
47e33c05 KC |
47 | /* |
48 | * When SECCOMP_IOCTL_NOTIF_ID_VALID was first introduced, it had the | |
49 | * wrong direction flag in the ioctl number. This is the broken one, | |
50 | * which the kernel needs to keep supporting until all userspaces stop | |
51 | * using the wrong command number. | |
52 | */ | |
53 | #define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR SECCOMP_IOR(2, __u64) | |
54 | ||
6a21cc50 TA |
55 | enum notify_state { |
56 | SECCOMP_NOTIFY_INIT, | |
57 | SECCOMP_NOTIFY_SENT, | |
58 | SECCOMP_NOTIFY_REPLIED, | |
59 | }; | |
60 | ||
61 | struct seccomp_knotif { | |
62 | /* The struct pid of the task whose filter triggered the notification */ | |
63 | struct task_struct *task; | |
64 | ||
65 | /* The "cookie" for this request; this is unique for this filter. */ | |
66 | u64 id; | |
67 | ||
68 | /* | |
69 | * The seccomp data. This pointer is valid the entire time this | |
70 | * notification is active, since it comes from __seccomp_filter which | |
71 | * eclipses the entire lifecycle here. | |
72 | */ | |
73 | const struct seccomp_data *data; | |
74 | ||
75 | /* | |
76 | * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a | |
77 | * struct seccomp_knotif is created and starts out in INIT. Once the | |
78 | * handler reads the notification off of an FD, it transitions to SENT. | |
79 | * If a signal is received the state transitions back to INIT and | |
80 | * another message is sent. When the userspace handler replies, state | |
81 | * transitions to REPLIED. | |
82 | */ | |
83 | enum notify_state state; | |
84 | ||
85 | /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */ | |
86 | int error; | |
87 | long val; | |
fb3c5386 | 88 | u32 flags; |
6a21cc50 TA |
89 | |
90 | /* Signals when this has entered SECCOMP_NOTIFY_REPLIED */ | |
91 | struct completion ready; | |
92 | ||
93 | struct list_head list; | |
94 | }; | |
95 | ||
96 | /** | |
97 | * struct notification - container for seccomp userspace notifications. Since | |
98 | * most seccomp filters will not have notification listeners attached and this | |
99 | * structure is fairly large, we store the notification-specific stuff in a | |
100 | * separate structure. | |
101 | * | |
102 | * @request: A semaphore that users of this notification can wait on for | |
103 | * changes. Actual reads and writes are still controlled with | |
104 | * filter->notify_lock. | |
105 | * @next_id: The id of the next request. | |
106 | * @notifications: A list of struct seccomp_knotif elements. | |
6a21cc50 TA |
107 | */ |
108 | struct notification { | |
109 | struct semaphore request; | |
110 | u64 next_id; | |
111 | struct list_head notifications; | |
6a21cc50 | 112 | }; |
e2cfabdf WD |
113 | |
114 | /** | |
115 | * struct seccomp_filter - container for seccomp BPF programs | |
116 | * | |
b707ddee CB |
117 | * @refs: Reference count to manage the object lifetime. |
118 | * A filter's reference count is incremented for each directly | |
119 | * attached task, once for the dependent filter, and if | |
120 | * requested for the user notifier. When @refs reaches zero, | |
121 | * the filter can be freed. | |
99cdb8b9 CB |
122 | * @users: A filter's @users count is incremented for each directly |
123 | * attached task (filter installation, fork(), thread_sync), | |
124 | * and once for the dependent filter (tracked in filter->prev). | |
125 | * When it reaches zero it indicates that no direct or indirect | |
126 | * users of that filter exist. No new tasks can get associated with | |
127 | * this filter after reaching 0. The @users count is always smaller | |
128 | * or equal to @refs. Hence, reaching 0 for @users does not mean | |
129 | * the filter can be freed. | |
e66a3997 | 130 | * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged |
e2cfabdf | 131 | * @prev: points to a previously installed, or inherited, filter |
285fdfc5 | 132 | * @prog: the BPF program to evaluate |
6a21cc50 TA |
133 | * @notif: the struct that holds all notification related information |
134 | * @notify_lock: A lock for all notification-related accesses. | |
76194c4e | 135 | * @wqh: A wait queue for poll if a notifier is in use. |
e2cfabdf WD |
136 | * |
137 | * seccomp_filter objects are organized in a tree linked via the @prev | |
138 | * pointer. For any task, it appears to be a singly-linked list starting | |
139 | * with current->seccomp.filter, the most recently attached or inherited filter. | |
140 | * However, multiple filters may share a @prev node, by way of fork(), which | |
141 | * results in a unidirectional tree existing in memory. This is similar to | |
142 | * how namespaces work. | |
143 | * | |
144 | * seccomp_filter objects should never be modified after being attached | |
b707ddee | 145 | * to a task_struct (other than @refs). |
e2cfabdf WD |
146 | */ |
147 | struct seccomp_filter { | |
b707ddee | 148 | refcount_t refs; |
99cdb8b9 | 149 | refcount_t users; |
e66a3997 | 150 | bool log; |
e2cfabdf | 151 | struct seccomp_filter *prev; |
7ae457c1 | 152 | struct bpf_prog *prog; |
6a21cc50 TA |
153 | struct notification *notif; |
154 | struct mutex notify_lock; | |
76194c4e | 155 | wait_queue_head_t wqh; |
e2cfabdf WD |
156 | }; |
157 | ||
158 | /* Limit any path through the tree to 256KB worth of instructions. */ | |
159 | #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter)) | |
160 | ||
bd4cf0ed | 161 | /* |
e2cfabdf WD |
162 | * Endianness is explicitly ignored and left for BPF program authors to manage |
163 | * as per the specific architecture. | |
164 | */ | |
bd4cf0ed | 165 | static void populate_seccomp_data(struct seccomp_data *sd) |
e2cfabdf | 166 | { |
bd4cf0ed AS |
167 | struct task_struct *task = current; |
168 | struct pt_regs *regs = task_pt_regs(task); | |
2eac7648 | 169 | unsigned long args[6]; |
e2cfabdf | 170 | |
bd4cf0ed | 171 | sd->nr = syscall_get_nr(task, regs); |
16add411 | 172 | sd->arch = syscall_get_arch(task); |
b35f549d | 173 | syscall_get_arguments(task, regs, args); |
2eac7648 DB |
174 | sd->args[0] = args[0]; |
175 | sd->args[1] = args[1]; | |
176 | sd->args[2] = args[2]; | |
177 | sd->args[3] = args[3]; | |
178 | sd->args[4] = args[4]; | |
179 | sd->args[5] = args[5]; | |
bd4cf0ed | 180 | sd->instruction_pointer = KSTK_EIP(task); |
e2cfabdf WD |
181 | } |
182 | ||
183 | /** | |
184 | * seccomp_check_filter - verify seccomp filter code | |
185 | * @filter: filter to verify | |
186 | * @flen: length of filter | |
187 | * | |
4df95ff4 | 188 | * Takes a previously checked filter (by bpf_check_classic) and |
e2cfabdf WD |
189 | * redirects all filter code that loads struct sk_buff data |
190 | * and related data through seccomp_bpf_load. It also | |
191 | * enforces length and alignment checking of those loads. | |
192 | * | |
193 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
194 | */ | |
195 | static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen) | |
196 | { | |
197 | int pc; | |
198 | for (pc = 0; pc < flen; pc++) { | |
199 | struct sock_filter *ftest = &filter[pc]; | |
200 | u16 code = ftest->code; | |
201 | u32 k = ftest->k; | |
202 | ||
203 | switch (code) { | |
34805931 | 204 | case BPF_LD | BPF_W | BPF_ABS: |
bd4cf0ed | 205 | ftest->code = BPF_LDX | BPF_W | BPF_ABS; |
e2cfabdf WD |
206 | /* 32-bit aligned and not out of bounds. */ |
207 | if (k >= sizeof(struct seccomp_data) || k & 3) | |
208 | return -EINVAL; | |
209 | continue; | |
34805931 | 210 | case BPF_LD | BPF_W | BPF_LEN: |
bd4cf0ed | 211 | ftest->code = BPF_LD | BPF_IMM; |
e2cfabdf WD |
212 | ftest->k = sizeof(struct seccomp_data); |
213 | continue; | |
34805931 | 214 | case BPF_LDX | BPF_W | BPF_LEN: |
bd4cf0ed | 215 | ftest->code = BPF_LDX | BPF_IMM; |
e2cfabdf WD |
216 | ftest->k = sizeof(struct seccomp_data); |
217 | continue; | |
218 | /* Explicitly include allowed calls. */ | |
34805931 DB |
219 | case BPF_RET | BPF_K: |
220 | case BPF_RET | BPF_A: | |
221 | case BPF_ALU | BPF_ADD | BPF_K: | |
222 | case BPF_ALU | BPF_ADD | BPF_X: | |
223 | case BPF_ALU | BPF_SUB | BPF_K: | |
224 | case BPF_ALU | BPF_SUB | BPF_X: | |
225 | case BPF_ALU | BPF_MUL | BPF_K: | |
226 | case BPF_ALU | BPF_MUL | BPF_X: | |
227 | case BPF_ALU | BPF_DIV | BPF_K: | |
228 | case BPF_ALU | BPF_DIV | BPF_X: | |
229 | case BPF_ALU | BPF_AND | BPF_K: | |
230 | case BPF_ALU | BPF_AND | BPF_X: | |
231 | case BPF_ALU | BPF_OR | BPF_K: | |
232 | case BPF_ALU | BPF_OR | BPF_X: | |
233 | case BPF_ALU | BPF_XOR | BPF_K: | |
234 | case BPF_ALU | BPF_XOR | BPF_X: | |
235 | case BPF_ALU | BPF_LSH | BPF_K: | |
236 | case BPF_ALU | BPF_LSH | BPF_X: | |
237 | case BPF_ALU | BPF_RSH | BPF_K: | |
238 | case BPF_ALU | BPF_RSH | BPF_X: | |
239 | case BPF_ALU | BPF_NEG: | |
240 | case BPF_LD | BPF_IMM: | |
241 | case BPF_LDX | BPF_IMM: | |
242 | case BPF_MISC | BPF_TAX: | |
243 | case BPF_MISC | BPF_TXA: | |
244 | case BPF_LD | BPF_MEM: | |
245 | case BPF_LDX | BPF_MEM: | |
246 | case BPF_ST: | |
247 | case BPF_STX: | |
248 | case BPF_JMP | BPF_JA: | |
249 | case BPF_JMP | BPF_JEQ | BPF_K: | |
250 | case BPF_JMP | BPF_JEQ | BPF_X: | |
251 | case BPF_JMP | BPF_JGE | BPF_K: | |
252 | case BPF_JMP | BPF_JGE | BPF_X: | |
253 | case BPF_JMP | BPF_JGT | BPF_K: | |
254 | case BPF_JMP | BPF_JGT | BPF_X: | |
255 | case BPF_JMP | BPF_JSET | BPF_K: | |
256 | case BPF_JMP | BPF_JSET | BPF_X: | |
e2cfabdf WD |
257 | continue; |
258 | default: | |
259 | return -EINVAL; | |
260 | } | |
261 | } | |
262 | return 0; | |
263 | } | |
264 | ||
265 | /** | |
285fdfc5 MS |
266 | * seccomp_run_filters - evaluates all seccomp filters against @sd |
267 | * @sd: optional seccomp data to be passed to filters | |
deb4de8b KC |
268 | * @match: stores struct seccomp_filter that resulted in the return value, |
269 | * unless filter returned SECCOMP_RET_ALLOW, in which case it will | |
270 | * be unchanged. | |
e2cfabdf WD |
271 | * |
272 | * Returns valid seccomp BPF response codes. | |
273 | */ | |
0466bdb9 | 274 | #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL))) |
deb4de8b KC |
275 | static u32 seccomp_run_filters(const struct seccomp_data *sd, |
276 | struct seccomp_filter **match) | |
e2cfabdf | 277 | { |
acf3b2c7 | 278 | u32 ret = SECCOMP_RET_ALLOW; |
8225d385 PK |
279 | /* Make sure cross-thread synced filter points somewhere sane. */ |
280 | struct seccomp_filter *f = | |
506458ef | 281 | READ_ONCE(current->seccomp.filter); |
acf3b2c7 WD |
282 | |
283 | /* Ensure unexpected behavior doesn't result in failing open. */ | |
0d42d73a | 284 | if (WARN_ON(f == NULL)) |
4d3b0b05 | 285 | return SECCOMP_RET_KILL_PROCESS; |
acf3b2c7 | 286 | |
e2cfabdf WD |
287 | /* |
288 | * All filters in the list are evaluated and the lowest BPF return | |
acf3b2c7 | 289 | * value always takes priority (ignoring the DATA). |
e2cfabdf | 290 | */ |
3ba2530c | 291 | for (; f; f = f->prev) { |
3d9f773c | 292 | u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd); |
8f577cad | 293 | |
0466bdb9 | 294 | if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) { |
acf3b2c7 | 295 | ret = cur_ret; |
deb4de8b KC |
296 | *match = f; |
297 | } | |
e2cfabdf WD |
298 | } |
299 | return ret; | |
300 | } | |
1f41b450 | 301 | #endif /* CONFIG_SECCOMP_FILTER */ |
e2cfabdf | 302 | |
1f41b450 KC |
303 | static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) |
304 | { | |
69f6a34b | 305 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 306 | |
1f41b450 KC |
307 | if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) |
308 | return false; | |
309 | ||
310 | return true; | |
311 | } | |
312 | ||
8bf37d8c | 313 | void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { } |
5c307089 | 314 | |
3ba2530c | 315 | static inline void seccomp_assign_mode(struct task_struct *task, |
00a02d0c KC |
316 | unsigned long seccomp_mode, |
317 | unsigned long flags) | |
1f41b450 | 318 | { |
69f6a34b | 319 | assert_spin_locked(&task->sighand->siglock); |
dbd95212 | 320 | |
3ba2530c KC |
321 | task->seccomp.mode = seccomp_mode; |
322 | /* | |
323 | * Make sure TIF_SECCOMP cannot be set before the mode (and | |
324 | * filter) is set. | |
325 | */ | |
326 | smp_mb__before_atomic(); | |
00a02d0c KC |
327 | /* Assume default seccomp processes want spec flaw mitigation. */ |
328 | if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0) | |
8bf37d8c | 329 | arch_seccomp_spec_mitigate(task); |
3ba2530c | 330 | set_tsk_thread_flag(task, TIF_SECCOMP); |
1f41b450 KC |
331 | } |
332 | ||
333 | #ifdef CONFIG_SECCOMP_FILTER | |
c2e1f2e3 KC |
334 | /* Returns 1 if the parent is an ancestor of the child. */ |
335 | static int is_ancestor(struct seccomp_filter *parent, | |
336 | struct seccomp_filter *child) | |
337 | { | |
338 | /* NULL is the root ancestor. */ | |
339 | if (parent == NULL) | |
340 | return 1; | |
341 | for (; child; child = child->prev) | |
342 | if (child == parent) | |
343 | return 1; | |
344 | return 0; | |
345 | } | |
346 | ||
347 | /** | |
348 | * seccomp_can_sync_threads: checks if all threads can be synchronized | |
349 | * | |
350 | * Expects sighand and cred_guard_mutex locks to be held. | |
351 | * | |
352 | * Returns 0 on success, -ve on error, or the pid of a thread which was | |
6beff00b | 353 | * either not in the correct seccomp mode or did not have an ancestral |
c2e1f2e3 KC |
354 | * seccomp filter. |
355 | */ | |
356 | static inline pid_t seccomp_can_sync_threads(void) | |
357 | { | |
358 | struct task_struct *thread, *caller; | |
359 | ||
360 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 361 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
362 | |
363 | /* Validate all threads being eligible for synchronization. */ | |
364 | caller = current; | |
365 | for_each_thread(caller, thread) { | |
366 | pid_t failed; | |
367 | ||
368 | /* Skip current, since it is initiating the sync. */ | |
369 | if (thread == caller) | |
370 | continue; | |
371 | ||
372 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || | |
373 | (thread->seccomp.mode == SECCOMP_MODE_FILTER && | |
374 | is_ancestor(thread->seccomp.filter, | |
375 | caller->seccomp.filter))) | |
376 | continue; | |
377 | ||
378 | /* Return the first thread that cannot be synchronized. */ | |
379 | failed = task_pid_vnr(thread); | |
380 | /* If the pid cannot be resolved, then return -ESRCH */ | |
0d42d73a | 381 | if (WARN_ON(failed == 0)) |
c2e1f2e3 KC |
382 | failed = -ESRCH; |
383 | return failed; | |
384 | } | |
385 | ||
386 | return 0; | |
387 | } | |
388 | ||
3a15fb6e CB |
389 | static inline void seccomp_filter_free(struct seccomp_filter *filter) |
390 | { | |
391 | if (filter) { | |
392 | bpf_prog_destroy(filter->prog); | |
393 | kfree(filter); | |
394 | } | |
395 | } | |
396 | ||
99cdb8b9 CB |
397 | static void __seccomp_filter_orphan(struct seccomp_filter *orig) |
398 | { | |
399 | while (orig && refcount_dec_and_test(&orig->users)) { | |
400 | if (waitqueue_active(&orig->wqh)) | |
401 | wake_up_poll(&orig->wqh, EPOLLHUP); | |
402 | orig = orig->prev; | |
403 | } | |
404 | } | |
405 | ||
3a15fb6e CB |
406 | static void __put_seccomp_filter(struct seccomp_filter *orig) |
407 | { | |
408 | /* Clean up single-reference branches iteratively. */ | |
409 | while (orig && refcount_dec_and_test(&orig->refs)) { | |
410 | struct seccomp_filter *freeme = orig; | |
411 | orig = orig->prev; | |
412 | seccomp_filter_free(freeme); | |
413 | } | |
414 | } | |
415 | ||
99cdb8b9 CB |
416 | static void __seccomp_filter_release(struct seccomp_filter *orig) |
417 | { | |
418 | /* Notify about any unused filters in the task's former filter tree. */ | |
419 | __seccomp_filter_orphan(orig); | |
420 | /* Finally drop all references to the task's former tree. */ | |
421 | __put_seccomp_filter(orig); | |
422 | } | |
423 | ||
3a15fb6e | 424 | /** |
99cdb8b9 CB |
425 | * seccomp_filter_release - Detach the task from its filter tree, |
426 | * drop its reference count, and notify | |
427 | * about unused filters | |
3a15fb6e CB |
428 | * |
429 | * This function should only be called when the task is exiting as | |
430 | * it detaches it from its filter tree. As such, READ_ONCE() and | |
431 | * barriers are not needed here, as would normally be needed. | |
432 | */ | |
433 | void seccomp_filter_release(struct task_struct *tsk) | |
434 | { | |
435 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
436 | ||
437 | /* Detach task from its filter tree. */ | |
438 | tsk->seccomp.filter = NULL; | |
99cdb8b9 | 439 | __seccomp_filter_release(orig); |
3a15fb6e CB |
440 | } |
441 | ||
c2e1f2e3 KC |
442 | /** |
443 | * seccomp_sync_threads: sets all threads to use current's filter | |
444 | * | |
445 | * Expects sighand and cred_guard_mutex locks to be held, and for | |
446 | * seccomp_can_sync_threads() to have returned success already | |
447 | * without dropping the locks. | |
448 | * | |
449 | */ | |
00a02d0c | 450 | static inline void seccomp_sync_threads(unsigned long flags) |
c2e1f2e3 KC |
451 | { |
452 | struct task_struct *thread, *caller; | |
453 | ||
454 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 455 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
456 | |
457 | /* Synchronize all threads. */ | |
458 | caller = current; | |
459 | for_each_thread(caller, thread) { | |
460 | /* Skip current, since it needs no changes. */ | |
461 | if (thread == caller) | |
462 | continue; | |
463 | ||
464 | /* Get a task reference for the new leaf node. */ | |
465 | get_seccomp_filter(caller); | |
99cdb8b9 | 466 | |
c2e1f2e3 KC |
467 | /* |
468 | * Drop the task reference to the shared ancestor since | |
469 | * current's path will hold a reference. (This also | |
470 | * allows a put before the assignment.) | |
471 | */ | |
99cdb8b9 CB |
472 | __seccomp_filter_release(thread->seccomp.filter); |
473 | ||
474 | /* Make our new filter tree visible. */ | |
c2e1f2e3 KC |
475 | smp_store_release(&thread->seccomp.filter, |
476 | caller->seccomp.filter); | |
c818c03b KC |
477 | atomic_set(&thread->seccomp.filter_count, |
478 | atomic_read(&thread->seccomp.filter_count)); | |
103502a3 JH |
479 | |
480 | /* | |
481 | * Don't let an unprivileged task work around | |
482 | * the no_new_privs restriction by creating | |
483 | * a thread that sets it up, enters seccomp, | |
484 | * then dies. | |
485 | */ | |
486 | if (task_no_new_privs(caller)) | |
487 | task_set_no_new_privs(thread); | |
488 | ||
c2e1f2e3 KC |
489 | /* |
490 | * Opt the other thread into seccomp if needed. | |
491 | * As threads are considered to be trust-realm | |
492 | * equivalent (see ptrace_may_access), it is safe to | |
493 | * allow one thread to transition the other. | |
494 | */ | |
103502a3 | 495 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) |
00a02d0c KC |
496 | seccomp_assign_mode(thread, SECCOMP_MODE_FILTER, |
497 | flags); | |
c2e1f2e3 KC |
498 | } |
499 | } | |
500 | ||
e2cfabdf | 501 | /** |
c8bee430 | 502 | * seccomp_prepare_filter: Prepares a seccomp filter for use. |
e2cfabdf WD |
503 | * @fprog: BPF program to install |
504 | * | |
c8bee430 | 505 | * Returns filter on success or an ERR_PTR on failure. |
e2cfabdf | 506 | */ |
c8bee430 | 507 | static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog) |
e2cfabdf | 508 | { |
ac67eb2c DB |
509 | struct seccomp_filter *sfilter; |
510 | int ret; | |
97f2645f | 511 | const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE); |
e2cfabdf WD |
512 | |
513 | if (fprog->len == 0 || fprog->len > BPF_MAXINSNS) | |
c8bee430 | 514 | return ERR_PTR(-EINVAL); |
d9e12f42 | 515 | |
c8bee430 | 516 | BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter)); |
e2cfabdf WD |
517 | |
518 | /* | |
119ce5c8 | 519 | * Installing a seccomp filter requires that the task has |
e2cfabdf WD |
520 | * CAP_SYS_ADMIN in its namespace or be running with no_new_privs. |
521 | * This avoids scenarios where unprivileged tasks can affect the | |
522 | * behavior of privileged children. | |
523 | */ | |
1d4457f9 | 524 | if (!task_no_new_privs(current) && |
c1a85a00 MM |
525 | security_capable(current_cred(), current_user_ns(), |
526 | CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) != 0) | |
c8bee430 | 527 | return ERR_PTR(-EACCES); |
e2cfabdf | 528 | |
bd4cf0ed | 529 | /* Allocate a new seccomp_filter */ |
ac67eb2c DB |
530 | sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN); |
531 | if (!sfilter) | |
d9e12f42 | 532 | return ERR_PTR(-ENOMEM); |
ac67eb2c | 533 | |
6a21cc50 | 534 | mutex_init(&sfilter->notify_lock); |
ac67eb2c | 535 | ret = bpf_prog_create_from_user(&sfilter->prog, fprog, |
f8e529ed | 536 | seccomp_check_filter, save_orig); |
ac67eb2c DB |
537 | if (ret < 0) { |
538 | kfree(sfilter); | |
539 | return ERR_PTR(ret); | |
d9e12f42 | 540 | } |
bd4cf0ed | 541 | |
b707ddee | 542 | refcount_set(&sfilter->refs, 1); |
99cdb8b9 | 543 | refcount_set(&sfilter->users, 1); |
76194c4e | 544 | init_waitqueue_head(&sfilter->wqh); |
e2cfabdf | 545 | |
ac67eb2c | 546 | return sfilter; |
e2cfabdf WD |
547 | } |
548 | ||
549 | /** | |
c8bee430 | 550 | * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog |
e2cfabdf WD |
551 | * @user_filter: pointer to the user data containing a sock_fprog. |
552 | * | |
553 | * Returns 0 on success and non-zero otherwise. | |
554 | */ | |
c8bee430 KC |
555 | static struct seccomp_filter * |
556 | seccomp_prepare_user_filter(const char __user *user_filter) | |
e2cfabdf WD |
557 | { |
558 | struct sock_fprog fprog; | |
c8bee430 | 559 | struct seccomp_filter *filter = ERR_PTR(-EFAULT); |
e2cfabdf WD |
560 | |
561 | #ifdef CONFIG_COMPAT | |
5c38065e | 562 | if (in_compat_syscall()) { |
e2cfabdf WD |
563 | struct compat_sock_fprog fprog32; |
564 | if (copy_from_user(&fprog32, user_filter, sizeof(fprog32))) | |
565 | goto out; | |
566 | fprog.len = fprog32.len; | |
567 | fprog.filter = compat_ptr(fprog32.filter); | |
568 | } else /* falls through to the if below. */ | |
569 | #endif | |
570 | if (copy_from_user(&fprog, user_filter, sizeof(fprog))) | |
571 | goto out; | |
c8bee430 | 572 | filter = seccomp_prepare_filter(&fprog); |
e2cfabdf | 573 | out: |
c8bee430 KC |
574 | return filter; |
575 | } | |
576 | ||
577 | /** | |
578 | * seccomp_attach_filter: validate and attach filter | |
579 | * @flags: flags to change filter behavior | |
580 | * @filter: seccomp filter to add to the current process | |
581 | * | |
dbd95212 KC |
582 | * Caller must be holding current->sighand->siglock lock. |
583 | * | |
7a0df7fb TA |
584 | * Returns 0 on success, -ve on error, or |
585 | * - in TSYNC mode: the pid of a thread which was either not in the correct | |
586 | * seccomp mode or did not have an ancestral seccomp filter | |
587 | * - in NEW_LISTENER mode: the fd of the new listener | |
c8bee430 KC |
588 | */ |
589 | static long seccomp_attach_filter(unsigned int flags, | |
590 | struct seccomp_filter *filter) | |
591 | { | |
592 | unsigned long total_insns; | |
593 | struct seccomp_filter *walker; | |
594 | ||
69f6a34b | 595 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 596 | |
c8bee430 KC |
597 | /* Validate resulting filter length. */ |
598 | total_insns = filter->prog->len; | |
599 | for (walker = current->seccomp.filter; walker; walker = walker->prev) | |
600 | total_insns += walker->prog->len + 4; /* 4 instr penalty */ | |
601 | if (total_insns > MAX_INSNS_PER_PATH) | |
602 | return -ENOMEM; | |
603 | ||
c2e1f2e3 KC |
604 | /* If thread sync has been requested, check that it is possible. */ |
605 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) { | |
606 | int ret; | |
607 | ||
608 | ret = seccomp_can_sync_threads(); | |
51891498 TA |
609 | if (ret) { |
610 | if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) | |
611 | return -ESRCH; | |
612 | else | |
613 | return ret; | |
614 | } | |
c2e1f2e3 KC |
615 | } |
616 | ||
e66a3997 TH |
617 | /* Set log flag, if present. */ |
618 | if (flags & SECCOMP_FILTER_FLAG_LOG) | |
619 | filter->log = true; | |
620 | ||
c8bee430 KC |
621 | /* |
622 | * If there is an existing filter, make it the prev and don't drop its | |
623 | * task reference. | |
624 | */ | |
625 | filter->prev = current->seccomp.filter; | |
626 | current->seccomp.filter = filter; | |
c818c03b | 627 | atomic_inc(¤t->seccomp.filter_count); |
c8bee430 | 628 | |
c2e1f2e3 KC |
629 | /* Now that the new filter is in place, synchronize to all threads. */ |
630 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) | |
00a02d0c | 631 | seccomp_sync_threads(flags); |
c2e1f2e3 | 632 | |
c8bee430 | 633 | return 0; |
e2cfabdf WD |
634 | } |
635 | ||
084f5601 | 636 | static void __get_seccomp_filter(struct seccomp_filter *filter) |
66a733ea | 637 | { |
b707ddee | 638 | refcount_inc(&filter->refs); |
66a733ea ON |
639 | } |
640 | ||
e2cfabdf WD |
641 | /* get_seccomp_filter - increments the reference count of the filter on @tsk */ |
642 | void get_seccomp_filter(struct task_struct *tsk) | |
643 | { | |
644 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
645 | if (!orig) | |
646 | return; | |
66a733ea | 647 | __get_seccomp_filter(orig); |
99cdb8b9 | 648 | refcount_inc(&orig->users); |
e2cfabdf WD |
649 | } |
650 | ||
ae7795bc | 651 | static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason) |
b25e6716 | 652 | { |
3b10db2b | 653 | clear_siginfo(info); |
b25e6716 MF |
654 | info->si_signo = SIGSYS; |
655 | info->si_code = SYS_SECCOMP; | |
656 | info->si_call_addr = (void __user *)KSTK_EIP(current); | |
657 | info->si_errno = reason; | |
16add411 | 658 | info->si_arch = syscall_get_arch(current); |
b25e6716 MF |
659 | info->si_syscall = syscall; |
660 | } | |
661 | ||
bb6ea430 WD |
662 | /** |
663 | * seccomp_send_sigsys - signals the task to allow in-process syscall emulation | |
664 | * @syscall: syscall number to send to userland | |
665 | * @reason: filter-supplied reason code to send to userland (via si_errno) | |
666 | * | |
667 | * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info. | |
668 | */ | |
669 | static void seccomp_send_sigsys(int syscall, int reason) | |
670 | { | |
ae7795bc | 671 | struct kernel_siginfo info; |
b25e6716 | 672 | seccomp_init_siginfo(&info, syscall, reason); |
a89e9b8a | 673 | force_sig_info(&info); |
bb6ea430 | 674 | } |
e2cfabdf | 675 | #endif /* CONFIG_SECCOMP_FILTER */ |
1da177e4 | 676 | |
0ddec0fc | 677 | /* For use with seccomp_actions_logged */ |
4d3b0b05 KC |
678 | #define SECCOMP_LOG_KILL_PROCESS (1 << 0) |
679 | #define SECCOMP_LOG_KILL_THREAD (1 << 1) | |
0ddec0fc TH |
680 | #define SECCOMP_LOG_TRAP (1 << 2) |
681 | #define SECCOMP_LOG_ERRNO (1 << 3) | |
682 | #define SECCOMP_LOG_TRACE (1 << 4) | |
59f5cf44 TH |
683 | #define SECCOMP_LOG_LOG (1 << 5) |
684 | #define SECCOMP_LOG_ALLOW (1 << 6) | |
6a21cc50 | 685 | #define SECCOMP_LOG_USER_NOTIF (1 << 7) |
0ddec0fc | 686 | |
4d3b0b05 KC |
687 | static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS | |
688 | SECCOMP_LOG_KILL_THREAD | | |
fd76875c KC |
689 | SECCOMP_LOG_TRAP | |
690 | SECCOMP_LOG_ERRNO | | |
6a21cc50 | 691 | SECCOMP_LOG_USER_NOTIF | |
fd76875c | 692 | SECCOMP_LOG_TRACE | |
59f5cf44 | 693 | SECCOMP_LOG_LOG; |
0ddec0fc | 694 | |
e66a3997 TH |
695 | static inline void seccomp_log(unsigned long syscall, long signr, u32 action, |
696 | bool requested) | |
0ddec0fc TH |
697 | { |
698 | bool log = false; | |
699 | ||
700 | switch (action) { | |
701 | case SECCOMP_RET_ALLOW: | |
e66a3997 | 702 | break; |
0ddec0fc | 703 | case SECCOMP_RET_TRAP: |
e66a3997 TH |
704 | log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP; |
705 | break; | |
0ddec0fc | 706 | case SECCOMP_RET_ERRNO: |
e66a3997 TH |
707 | log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO; |
708 | break; | |
0ddec0fc | 709 | case SECCOMP_RET_TRACE: |
e66a3997 | 710 | log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE; |
0ddec0fc | 711 | break; |
6a21cc50 TA |
712 | case SECCOMP_RET_USER_NOTIF: |
713 | log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF; | |
714 | break; | |
59f5cf44 TH |
715 | case SECCOMP_RET_LOG: |
716 | log = seccomp_actions_logged & SECCOMP_LOG_LOG; | |
717 | break; | |
fd76875c | 718 | case SECCOMP_RET_KILL_THREAD: |
fd76875c | 719 | log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD; |
4d3b0b05 KC |
720 | break; |
721 | case SECCOMP_RET_KILL_PROCESS: | |
722 | default: | |
723 | log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS; | |
0ddec0fc TH |
724 | } |
725 | ||
726 | /* | |
326bee02 TH |
727 | * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the |
728 | * FILTER_FLAG_LOG bit was set. The admin has the ability to silence | |
729 | * any action from being logged by removing the action name from the | |
730 | * seccomp_actions_logged sysctl. | |
0ddec0fc | 731 | */ |
326bee02 TH |
732 | if (!log) |
733 | return; | |
0ddec0fc | 734 | |
326bee02 | 735 | audit_seccomp(syscall, signr, action); |
0ddec0fc TH |
736 | } |
737 | ||
1da177e4 LT |
738 | /* |
739 | * Secure computing mode 1 allows only read/write/exit/sigreturn. | |
740 | * To be fully secure this must be combined with rlimit | |
741 | * to limit the stack allocations too. | |
742 | */ | |
cb4253aa | 743 | static const int mode1_syscalls[] = { |
1da177e4 | 744 | __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, |
fe4bfff8 | 745 | -1, /* negative terminated */ |
1da177e4 LT |
746 | }; |
747 | ||
a4412fc9 | 748 | static void __secure_computing_strict(int this_syscall) |
1da177e4 | 749 | { |
fe4bfff8 | 750 | const int *allowed_syscalls = mode1_syscalls; |
a4412fc9 | 751 | #ifdef CONFIG_COMPAT |
5c38065e | 752 | if (in_compat_syscall()) |
fe4bfff8 | 753 | allowed_syscalls = get_compat_mode1_syscalls(); |
a4412fc9 AL |
754 | #endif |
755 | do { | |
fe4bfff8 | 756 | if (*allowed_syscalls == this_syscall) |
a4412fc9 | 757 | return; |
fe4bfff8 | 758 | } while (*++allowed_syscalls != -1); |
a4412fc9 AL |
759 | |
760 | #ifdef SECCOMP_DEBUG | |
761 | dump_stack(); | |
762 | #endif | |
fd76875c | 763 | seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true); |
a4412fc9 AL |
764 | do_exit(SIGKILL); |
765 | } | |
766 | ||
767 | #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER | |
768 | void secure_computing_strict(int this_syscall) | |
769 | { | |
770 | int mode = current->seccomp.mode; | |
771 | ||
97f2645f | 772 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 TA |
773 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
774 | return; | |
775 | ||
221272f9 | 776 | if (mode == SECCOMP_MODE_DISABLED) |
a4412fc9 AL |
777 | return; |
778 | else if (mode == SECCOMP_MODE_STRICT) | |
779 | __secure_computing_strict(this_syscall); | |
780 | else | |
781 | BUG(); | |
782 | } | |
783 | #else | |
13aa72f0 AL |
784 | |
785 | #ifdef CONFIG_SECCOMP_FILTER | |
6a21cc50 TA |
786 | static u64 seccomp_next_notify_id(struct seccomp_filter *filter) |
787 | { | |
788 | /* | |
789 | * Note: overflow is ok here, the id just needs to be unique per | |
790 | * filter. | |
791 | */ | |
792 | lockdep_assert_held(&filter->notify_lock); | |
793 | return filter->notif->next_id++; | |
794 | } | |
795 | ||
fb3c5386 CB |
796 | static int seccomp_do_user_notification(int this_syscall, |
797 | struct seccomp_filter *match, | |
798 | const struct seccomp_data *sd) | |
6a21cc50 TA |
799 | { |
800 | int err; | |
fb3c5386 | 801 | u32 flags = 0; |
6a21cc50 TA |
802 | long ret = 0; |
803 | struct seccomp_knotif n = {}; | |
804 | ||
805 | mutex_lock(&match->notify_lock); | |
806 | err = -ENOSYS; | |
807 | if (!match->notif) | |
808 | goto out; | |
809 | ||
810 | n.task = current; | |
811 | n.state = SECCOMP_NOTIFY_INIT; | |
812 | n.data = sd; | |
813 | n.id = seccomp_next_notify_id(match); | |
814 | init_completion(&n.ready); | |
815 | list_add(&n.list, &match->notif->notifications); | |
816 | ||
817 | up(&match->notif->request); | |
76194c4e | 818 | wake_up_poll(&match->wqh, EPOLLIN | EPOLLRDNORM); |
6a21cc50 TA |
819 | mutex_unlock(&match->notify_lock); |
820 | ||
821 | /* | |
822 | * This is where we wait for a reply from userspace. | |
823 | */ | |
824 | err = wait_for_completion_interruptible(&n.ready); | |
825 | mutex_lock(&match->notify_lock); | |
826 | if (err == 0) { | |
827 | ret = n.val; | |
828 | err = n.error; | |
fb3c5386 | 829 | flags = n.flags; |
6a21cc50 TA |
830 | } |
831 | ||
832 | /* | |
833 | * Note that it's possible the listener died in between the time when | |
834 | * we were notified of a respons (or a signal) and when we were able to | |
835 | * re-acquire the lock, so only delete from the list if the | |
836 | * notification actually exists. | |
837 | * | |
838 | * Also note that this test is only valid because there's no way to | |
839 | * *reattach* to a notifier right now. If one is added, we'll need to | |
840 | * keep track of the notif itself and make sure they match here. | |
841 | */ | |
842 | if (match->notif) | |
843 | list_del(&n.list); | |
844 | out: | |
845 | mutex_unlock(&match->notify_lock); | |
fb3c5386 CB |
846 | |
847 | /* Userspace requests to continue the syscall. */ | |
848 | if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) | |
849 | return 0; | |
850 | ||
6a21cc50 TA |
851 | syscall_set_return_value(current, task_pt_regs(current), |
852 | err, ret); | |
fb3c5386 | 853 | return -1; |
6a21cc50 TA |
854 | } |
855 | ||
ce6526e8 KC |
856 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
857 | const bool recheck_after_trace) | |
13aa72f0 AL |
858 | { |
859 | u32 filter_ret, action; | |
deb4de8b | 860 | struct seccomp_filter *match = NULL; |
13aa72f0 | 861 | int data; |
db511391 | 862 | struct seccomp_data sd_local; |
1da177e4 | 863 | |
3ba2530c KC |
864 | /* |
865 | * Make sure that any changes to mode from another thread have | |
866 | * been seen after TIF_SECCOMP was seen. | |
867 | */ | |
868 | rmb(); | |
869 | ||
db511391 TA |
870 | if (!sd) { |
871 | populate_seccomp_data(&sd_local); | |
872 | sd = &sd_local; | |
873 | } | |
874 | ||
deb4de8b | 875 | filter_ret = seccomp_run_filters(sd, &match); |
13aa72f0 | 876 | data = filter_ret & SECCOMP_RET_DATA; |
0466bdb9 | 877 | action = filter_ret & SECCOMP_RET_ACTION_FULL; |
13aa72f0 AL |
878 | |
879 | switch (action) { | |
880 | case SECCOMP_RET_ERRNO: | |
580c57f1 KC |
881 | /* Set low-order bits as an errno, capped at MAX_ERRNO. */ |
882 | if (data > MAX_ERRNO) | |
883 | data = MAX_ERRNO; | |
d39bd00d | 884 | syscall_set_return_value(current, task_pt_regs(current), |
13aa72f0 AL |
885 | -data, 0); |
886 | goto skip; | |
887 | ||
888 | case SECCOMP_RET_TRAP: | |
889 | /* Show the handler the original registers. */ | |
d39bd00d | 890 | syscall_rollback(current, task_pt_regs(current)); |
13aa72f0 AL |
891 | /* Let the filter pass back 16 bits of data. */ |
892 | seccomp_send_sigsys(this_syscall, data); | |
893 | goto skip; | |
894 | ||
895 | case SECCOMP_RET_TRACE: | |
ce6526e8 KC |
896 | /* We've been put in this state by the ptracer already. */ |
897 | if (recheck_after_trace) | |
898 | return 0; | |
899 | ||
8112c4f1 KC |
900 | /* ENOSYS these calls if there is no tracer attached. */ |
901 | if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { | |
902 | syscall_set_return_value(current, | |
903 | task_pt_regs(current), | |
904 | -ENOSYS, 0); | |
905 | goto skip; | |
906 | } | |
907 | ||
908 | /* Allow the BPF to provide the event message */ | |
909 | ptrace_event(PTRACE_EVENT_SECCOMP, data); | |
910 | /* | |
911 | * The delivery of a fatal signal during event | |
485a252a KC |
912 | * notification may silently skip tracer notification, |
913 | * which could leave us with a potentially unmodified | |
914 | * syscall that the tracer would have liked to have | |
915 | * changed. Since the process is about to die, we just | |
916 | * force the syscall to be skipped and let the signal | |
917 | * kill the process and correctly handle any tracer exit | |
918 | * notifications. | |
8112c4f1 KC |
919 | */ |
920 | if (fatal_signal_pending(current)) | |
485a252a | 921 | goto skip; |
8112c4f1 KC |
922 | /* Check if the tracer forced the syscall to be skipped. */ |
923 | this_syscall = syscall_get_nr(current, task_pt_regs(current)); | |
924 | if (this_syscall < 0) | |
925 | goto skip; | |
926 | ||
ce6526e8 KC |
927 | /* |
928 | * Recheck the syscall, since it may have changed. This | |
929 | * intentionally uses a NULL struct seccomp_data to force | |
930 | * a reload of all registers. This does not goto skip since | |
931 | * a skip would have already been reported. | |
932 | */ | |
933 | if (__seccomp_filter(this_syscall, NULL, true)) | |
934 | return -1; | |
935 | ||
8112c4f1 | 936 | return 0; |
13aa72f0 | 937 | |
6a21cc50 | 938 | case SECCOMP_RET_USER_NOTIF: |
fb3c5386 CB |
939 | if (seccomp_do_user_notification(this_syscall, match, sd)) |
940 | goto skip; | |
941 | ||
942 | return 0; | |
6a21cc50 | 943 | |
59f5cf44 TH |
944 | case SECCOMP_RET_LOG: |
945 | seccomp_log(this_syscall, 0, action, true); | |
946 | return 0; | |
947 | ||
13aa72f0 | 948 | case SECCOMP_RET_ALLOW: |
deb4de8b KC |
949 | /* |
950 | * Note that the "match" filter will always be NULL for | |
951 | * this action since SECCOMP_RET_ALLOW is the starting | |
952 | * state in seccomp_run_filters(). | |
953 | */ | |
8112c4f1 | 954 | return 0; |
13aa72f0 | 955 | |
fd76875c | 956 | case SECCOMP_RET_KILL_THREAD: |
4d3b0b05 | 957 | case SECCOMP_RET_KILL_PROCESS: |
131b6351 | 958 | default: |
e66a3997 | 959 | seccomp_log(this_syscall, SIGSYS, action, true); |
d7276e32 | 960 | /* Dump core only if this is the last remaining thread. */ |
4d3b0b05 KC |
961 | if (action == SECCOMP_RET_KILL_PROCESS || |
962 | get_nr_threads(current) == 1) { | |
ae7795bc | 963 | kernel_siginfo_t info; |
131b6351 | 964 | |
d7276e32 KC |
965 | /* Show the original registers in the dump. */ |
966 | syscall_rollback(current, task_pt_regs(current)); | |
967 | /* Trigger a manual coredump since do_exit skips it. */ | |
968 | seccomp_init_siginfo(&info, this_syscall, data); | |
969 | do_coredump(&info); | |
970 | } | |
4d3b0b05 KC |
971 | if (action == SECCOMP_RET_KILL_PROCESS) |
972 | do_group_exit(SIGSYS); | |
973 | else | |
974 | do_exit(SIGSYS); | |
13aa72f0 AL |
975 | } |
976 | ||
977 | unreachable(); | |
978 | ||
979 | skip: | |
e66a3997 | 980 | seccomp_log(this_syscall, 0, action, match ? match->log : false); |
8112c4f1 KC |
981 | return -1; |
982 | } | |
983 | #else | |
ce6526e8 KC |
984 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
985 | const bool recheck_after_trace) | |
8112c4f1 KC |
986 | { |
987 | BUG(); | |
13aa72f0 | 988 | } |
1da177e4 | 989 | #endif |
13aa72f0 | 990 | |
8112c4f1 | 991 | int __secure_computing(const struct seccomp_data *sd) |
13aa72f0 AL |
992 | { |
993 | int mode = current->seccomp.mode; | |
8112c4f1 | 994 | int this_syscall; |
13aa72f0 | 995 | |
97f2645f | 996 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 | 997 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
8112c4f1 KC |
998 | return 0; |
999 | ||
1000 | this_syscall = sd ? sd->nr : | |
1001 | syscall_get_nr(current, task_pt_regs(current)); | |
13c4a901 | 1002 | |
13aa72f0 | 1003 | switch (mode) { |
e2cfabdf | 1004 | case SECCOMP_MODE_STRICT: |
13aa72f0 | 1005 | __secure_computing_strict(this_syscall); /* may call do_exit */ |
8112c4f1 | 1006 | return 0; |
13aa72f0 | 1007 | case SECCOMP_MODE_FILTER: |
ce6526e8 | 1008 | return __seccomp_filter(this_syscall, sd, false); |
1da177e4 LT |
1009 | default: |
1010 | BUG(); | |
1011 | } | |
13aa72f0 | 1012 | } |
a4412fc9 | 1013 | #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ |
1d9d02fe AA |
1014 | |
1015 | long prctl_get_seccomp(void) | |
1016 | { | |
1017 | return current->seccomp.mode; | |
1018 | } | |
1019 | ||
e2cfabdf | 1020 | /** |
3b23dd12 | 1021 | * seccomp_set_mode_strict: internal function for setting strict seccomp |
e2cfabdf WD |
1022 | * |
1023 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
1024 | * | |
1025 | * Returns 0 on success or -EINVAL on failure. | |
1026 | */ | |
3b23dd12 | 1027 | static long seccomp_set_mode_strict(void) |
1d9d02fe | 1028 | { |
3b23dd12 | 1029 | const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; |
e2cfabdf | 1030 | long ret = -EINVAL; |
1d9d02fe | 1031 | |
dbd95212 KC |
1032 | spin_lock_irq(¤t->sighand->siglock); |
1033 | ||
1f41b450 | 1034 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1d9d02fe AA |
1035 | goto out; |
1036 | ||
cf99abac | 1037 | #ifdef TIF_NOTSC |
3b23dd12 | 1038 | disable_TSC(); |
cf99abac | 1039 | #endif |
00a02d0c | 1040 | seccomp_assign_mode(current, seccomp_mode, 0); |
3b23dd12 KC |
1041 | ret = 0; |
1042 | ||
1043 | out: | |
dbd95212 | 1044 | spin_unlock_irq(¤t->sighand->siglock); |
3b23dd12 KC |
1045 | |
1046 | return ret; | |
1047 | } | |
1048 | ||
e2cfabdf | 1049 | #ifdef CONFIG_SECCOMP_FILTER |
6a21cc50 TA |
1050 | static int seccomp_notify_release(struct inode *inode, struct file *file) |
1051 | { | |
1052 | struct seccomp_filter *filter = file->private_data; | |
1053 | struct seccomp_knotif *knotif; | |
1054 | ||
a811dc61 TA |
1055 | if (!filter) |
1056 | return 0; | |
1057 | ||
6a21cc50 TA |
1058 | mutex_lock(&filter->notify_lock); |
1059 | ||
1060 | /* | |
1061 | * If this file is being closed because e.g. the task who owned it | |
1062 | * died, let's wake everyone up who was waiting on us. | |
1063 | */ | |
1064 | list_for_each_entry(knotif, &filter->notif->notifications, list) { | |
1065 | if (knotif->state == SECCOMP_NOTIFY_REPLIED) | |
1066 | continue; | |
1067 | ||
1068 | knotif->state = SECCOMP_NOTIFY_REPLIED; | |
1069 | knotif->error = -ENOSYS; | |
1070 | knotif->val = 0; | |
1071 | ||
1072 | complete(&knotif->ready); | |
1073 | } | |
1074 | ||
1075 | kfree(filter->notif); | |
1076 | filter->notif = NULL; | |
1077 | mutex_unlock(&filter->notify_lock); | |
1078 | __put_seccomp_filter(filter); | |
1079 | return 0; | |
1080 | } | |
1081 | ||
9f87dcf1 SD |
1082 | /* must be called with notif_lock held */ |
1083 | static inline struct seccomp_knotif * | |
1084 | find_notification(struct seccomp_filter *filter, u64 id) | |
1085 | { | |
1086 | struct seccomp_knotif *cur; | |
1087 | ||
1088 | lockdep_assert_held(&filter->notify_lock); | |
1089 | ||
1090 | list_for_each_entry(cur, &filter->notif->notifications, list) { | |
1091 | if (cur->id == id) | |
1092 | return cur; | |
1093 | } | |
1094 | ||
1095 | return NULL; | |
1096 | } | |
1097 | ||
1098 | ||
6a21cc50 TA |
1099 | static long seccomp_notify_recv(struct seccomp_filter *filter, |
1100 | void __user *buf) | |
1101 | { | |
1102 | struct seccomp_knotif *knotif = NULL, *cur; | |
1103 | struct seccomp_notif unotif; | |
1104 | ssize_t ret; | |
1105 | ||
2882d53c SD |
1106 | /* Verify that we're not given garbage to keep struct extensible. */ |
1107 | ret = check_zeroed_user(buf, sizeof(unotif)); | |
1108 | if (ret < 0) | |
1109 | return ret; | |
1110 | if (!ret) | |
1111 | return -EINVAL; | |
1112 | ||
6a21cc50 TA |
1113 | memset(&unotif, 0, sizeof(unotif)); |
1114 | ||
1115 | ret = down_interruptible(&filter->notif->request); | |
1116 | if (ret < 0) | |
1117 | return ret; | |
1118 | ||
1119 | mutex_lock(&filter->notify_lock); | |
1120 | list_for_each_entry(cur, &filter->notif->notifications, list) { | |
1121 | if (cur->state == SECCOMP_NOTIFY_INIT) { | |
1122 | knotif = cur; | |
1123 | break; | |
1124 | } | |
1125 | } | |
1126 | ||
1127 | /* | |
1128 | * If we didn't find a notification, it could be that the task was | |
1129 | * interrupted by a fatal signal between the time we were woken and | |
1130 | * when we were able to acquire the rw lock. | |
1131 | */ | |
1132 | if (!knotif) { | |
1133 | ret = -ENOENT; | |
1134 | goto out; | |
1135 | } | |
1136 | ||
1137 | unotif.id = knotif->id; | |
1138 | unotif.pid = task_pid_vnr(knotif->task); | |
1139 | unotif.data = *(knotif->data); | |
1140 | ||
1141 | knotif->state = SECCOMP_NOTIFY_SENT; | |
76194c4e | 1142 | wake_up_poll(&filter->wqh, EPOLLOUT | EPOLLWRNORM); |
6a21cc50 TA |
1143 | ret = 0; |
1144 | out: | |
1145 | mutex_unlock(&filter->notify_lock); | |
1146 | ||
1147 | if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) { | |
1148 | ret = -EFAULT; | |
1149 | ||
1150 | /* | |
1151 | * Userspace screwed up. To make sure that we keep this | |
1152 | * notification alive, let's reset it back to INIT. It | |
1153 | * may have died when we released the lock, so we need to make | |
1154 | * sure it's still around. | |
1155 | */ | |
6a21cc50 | 1156 | mutex_lock(&filter->notify_lock); |
9f87dcf1 | 1157 | knotif = find_notification(filter, unotif.id); |
6a21cc50 TA |
1158 | if (knotif) { |
1159 | knotif->state = SECCOMP_NOTIFY_INIT; | |
1160 | up(&filter->notif->request); | |
1161 | } | |
1162 | mutex_unlock(&filter->notify_lock); | |
1163 | } | |
1164 | ||
1165 | return ret; | |
1166 | } | |
1167 | ||
1168 | static long seccomp_notify_send(struct seccomp_filter *filter, | |
1169 | void __user *buf) | |
1170 | { | |
1171 | struct seccomp_notif_resp resp = {}; | |
9f87dcf1 | 1172 | struct seccomp_knotif *knotif; |
6a21cc50 TA |
1173 | long ret; |
1174 | ||
1175 | if (copy_from_user(&resp, buf, sizeof(resp))) | |
1176 | return -EFAULT; | |
1177 | ||
fb3c5386 CB |
1178 | if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE) |
1179 | return -EINVAL; | |
1180 | ||
1181 | if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) && | |
1182 | (resp.error || resp.val)) | |
6a21cc50 TA |
1183 | return -EINVAL; |
1184 | ||
1185 | ret = mutex_lock_interruptible(&filter->notify_lock); | |
1186 | if (ret < 0) | |
1187 | return ret; | |
1188 | ||
9f87dcf1 | 1189 | knotif = find_notification(filter, resp.id); |
6a21cc50 TA |
1190 | if (!knotif) { |
1191 | ret = -ENOENT; | |
1192 | goto out; | |
1193 | } | |
1194 | ||
1195 | /* Allow exactly one reply. */ | |
1196 | if (knotif->state != SECCOMP_NOTIFY_SENT) { | |
1197 | ret = -EINPROGRESS; | |
1198 | goto out; | |
1199 | } | |
1200 | ||
1201 | ret = 0; | |
1202 | knotif->state = SECCOMP_NOTIFY_REPLIED; | |
1203 | knotif->error = resp.error; | |
1204 | knotif->val = resp.val; | |
fb3c5386 | 1205 | knotif->flags = resp.flags; |
6a21cc50 TA |
1206 | complete(&knotif->ready); |
1207 | out: | |
1208 | mutex_unlock(&filter->notify_lock); | |
1209 | return ret; | |
1210 | } | |
1211 | ||
1212 | static long seccomp_notify_id_valid(struct seccomp_filter *filter, | |
1213 | void __user *buf) | |
1214 | { | |
9f87dcf1 | 1215 | struct seccomp_knotif *knotif; |
6a21cc50 TA |
1216 | u64 id; |
1217 | long ret; | |
1218 | ||
1219 | if (copy_from_user(&id, buf, sizeof(id))) | |
1220 | return -EFAULT; | |
1221 | ||
1222 | ret = mutex_lock_interruptible(&filter->notify_lock); | |
1223 | if (ret < 0) | |
1224 | return ret; | |
1225 | ||
9f87dcf1 SD |
1226 | knotif = find_notification(filter, id); |
1227 | if (knotif && knotif->state == SECCOMP_NOTIFY_SENT) | |
1228 | ret = 0; | |
1229 | else | |
1230 | ret = -ENOENT; | |
6a21cc50 | 1231 | |
6a21cc50 TA |
1232 | mutex_unlock(&filter->notify_lock); |
1233 | return ret; | |
1234 | } | |
1235 | ||
1236 | static long seccomp_notify_ioctl(struct file *file, unsigned int cmd, | |
1237 | unsigned long arg) | |
1238 | { | |
1239 | struct seccomp_filter *filter = file->private_data; | |
1240 | void __user *buf = (void __user *)arg; | |
1241 | ||
1242 | switch (cmd) { | |
1243 | case SECCOMP_IOCTL_NOTIF_RECV: | |
1244 | return seccomp_notify_recv(filter, buf); | |
1245 | case SECCOMP_IOCTL_NOTIF_SEND: | |
1246 | return seccomp_notify_send(filter, buf); | |
47e33c05 | 1247 | case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR: |
6a21cc50 TA |
1248 | case SECCOMP_IOCTL_NOTIF_ID_VALID: |
1249 | return seccomp_notify_id_valid(filter, buf); | |
1250 | default: | |
1251 | return -EINVAL; | |
1252 | } | |
1253 | } | |
1254 | ||
1255 | static __poll_t seccomp_notify_poll(struct file *file, | |
1256 | struct poll_table_struct *poll_tab) | |
1257 | { | |
1258 | struct seccomp_filter *filter = file->private_data; | |
1259 | __poll_t ret = 0; | |
1260 | struct seccomp_knotif *cur; | |
1261 | ||
76194c4e | 1262 | poll_wait(file, &filter->wqh, poll_tab); |
6a21cc50 | 1263 | |
319deec7 | 1264 | if (mutex_lock_interruptible(&filter->notify_lock) < 0) |
6a21cc50 TA |
1265 | return EPOLLERR; |
1266 | ||
1267 | list_for_each_entry(cur, &filter->notif->notifications, list) { | |
1268 | if (cur->state == SECCOMP_NOTIFY_INIT) | |
1269 | ret |= EPOLLIN | EPOLLRDNORM; | |
1270 | if (cur->state == SECCOMP_NOTIFY_SENT) | |
1271 | ret |= EPOLLOUT | EPOLLWRNORM; | |
1272 | if ((ret & EPOLLIN) && (ret & EPOLLOUT)) | |
1273 | break; | |
1274 | } | |
1275 | ||
1276 | mutex_unlock(&filter->notify_lock); | |
1277 | ||
99cdb8b9 CB |
1278 | if (refcount_read(&filter->users) == 0) |
1279 | ret |= EPOLLHUP; | |
1280 | ||
6a21cc50 TA |
1281 | return ret; |
1282 | } | |
1283 | ||
1284 | static const struct file_operations seccomp_notify_ops = { | |
1285 | .poll = seccomp_notify_poll, | |
1286 | .release = seccomp_notify_release, | |
1287 | .unlocked_ioctl = seccomp_notify_ioctl, | |
3db81afd | 1288 | .compat_ioctl = seccomp_notify_ioctl, |
6a21cc50 TA |
1289 | }; |
1290 | ||
1291 | static struct file *init_listener(struct seccomp_filter *filter) | |
1292 | { | |
1293 | struct file *ret = ERR_PTR(-EBUSY); | |
1294 | struct seccomp_filter *cur; | |
1295 | ||
1296 | for (cur = current->seccomp.filter; cur; cur = cur->prev) { | |
1297 | if (cur->notif) | |
1298 | goto out; | |
1299 | } | |
1300 | ||
1301 | ret = ERR_PTR(-ENOMEM); | |
1302 | filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL); | |
1303 | if (!filter->notif) | |
1304 | goto out; | |
1305 | ||
1306 | sema_init(&filter->notif->request, 0); | |
1307 | filter->notif->next_id = get_random_u64(); | |
1308 | INIT_LIST_HEAD(&filter->notif->notifications); | |
6a21cc50 TA |
1309 | |
1310 | ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops, | |
1311 | filter, O_RDWR); | |
1312 | if (IS_ERR(ret)) | |
1313 | goto out_notif; | |
1314 | ||
1315 | /* The file has a reference to it now */ | |
1316 | __get_seccomp_filter(filter); | |
1317 | ||
1318 | out_notif: | |
1319 | if (IS_ERR(ret)) | |
1320 | kfree(filter->notif); | |
1321 | out: | |
1322 | return ret; | |
1323 | } | |
1324 | ||
3b23dd12 KC |
1325 | /** |
1326 | * seccomp_set_mode_filter: internal function for setting seccomp filter | |
48dc92b9 | 1327 | * @flags: flags to change filter behavior |
3b23dd12 KC |
1328 | * @filter: struct sock_fprog containing filter |
1329 | * | |
1330 | * This function may be called repeatedly to install additional filters. | |
1331 | * Every filter successfully installed will be evaluated (in reverse order) | |
1332 | * for each system call the task makes. | |
1333 | * | |
1334 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
1335 | * | |
1336 | * Returns 0 on success or -EINVAL on failure. | |
1337 | */ | |
48dc92b9 KC |
1338 | static long seccomp_set_mode_filter(unsigned int flags, |
1339 | const char __user *filter) | |
3b23dd12 KC |
1340 | { |
1341 | const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; | |
c8bee430 | 1342 | struct seccomp_filter *prepared = NULL; |
3b23dd12 | 1343 | long ret = -EINVAL; |
6a21cc50 TA |
1344 | int listener = -1; |
1345 | struct file *listener_f = NULL; | |
3b23dd12 | 1346 | |
48dc92b9 | 1347 | /* Validate flags. */ |
c2e1f2e3 | 1348 | if (flags & ~SECCOMP_FILTER_FLAG_MASK) |
dbd95212 | 1349 | return -EINVAL; |
48dc92b9 | 1350 | |
7a0df7fb TA |
1351 | /* |
1352 | * In the successful case, NEW_LISTENER returns the new listener fd. | |
1353 | * But in the failure case, TSYNC returns the thread that died. If you | |
1354 | * combine these two flags, there's no way to tell whether something | |
51891498 TA |
1355 | * succeeded or failed. So, let's disallow this combination if the user |
1356 | * has not explicitly requested no errors from TSYNC. | |
7a0df7fb TA |
1357 | */ |
1358 | if ((flags & SECCOMP_FILTER_FLAG_TSYNC) && | |
51891498 TA |
1359 | (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) && |
1360 | ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0)) | |
7a0df7fb TA |
1361 | return -EINVAL; |
1362 | ||
c8bee430 KC |
1363 | /* Prepare the new filter before holding any locks. */ |
1364 | prepared = seccomp_prepare_user_filter(filter); | |
1365 | if (IS_ERR(prepared)) | |
1366 | return PTR_ERR(prepared); | |
1367 | ||
6a21cc50 TA |
1368 | if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) { |
1369 | listener = get_unused_fd_flags(O_CLOEXEC); | |
1370 | if (listener < 0) { | |
1371 | ret = listener; | |
1372 | goto out_free; | |
1373 | } | |
1374 | ||
1375 | listener_f = init_listener(prepared); | |
1376 | if (IS_ERR(listener_f)) { | |
1377 | put_unused_fd(listener); | |
1378 | ret = PTR_ERR(listener_f); | |
1379 | goto out_free; | |
1380 | } | |
1381 | } | |
1382 | ||
c2e1f2e3 KC |
1383 | /* |
1384 | * Make sure we cannot change seccomp or nnp state via TSYNC | |
1385 | * while another thread is in the middle of calling exec. | |
1386 | */ | |
1387 | if (flags & SECCOMP_FILTER_FLAG_TSYNC && | |
1388 | mutex_lock_killable(¤t->signal->cred_guard_mutex)) | |
6a21cc50 | 1389 | goto out_put_fd; |
c2e1f2e3 | 1390 | |
dbd95212 KC |
1391 | spin_lock_irq(¤t->sighand->siglock); |
1392 | ||
3b23dd12 KC |
1393 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1394 | goto out; | |
1395 | ||
c8bee430 | 1396 | ret = seccomp_attach_filter(flags, prepared); |
3b23dd12 | 1397 | if (ret) |
e2cfabdf | 1398 | goto out; |
c8bee430 KC |
1399 | /* Do not free the successfully attached filter. */ |
1400 | prepared = NULL; | |
1d9d02fe | 1401 | |
00a02d0c | 1402 | seccomp_assign_mode(current, seccomp_mode, flags); |
e2cfabdf | 1403 | out: |
dbd95212 | 1404 | spin_unlock_irq(¤t->sighand->siglock); |
c2e1f2e3 KC |
1405 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
1406 | mutex_unlock(¤t->signal->cred_guard_mutex); | |
6a21cc50 TA |
1407 | out_put_fd: |
1408 | if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) { | |
7a0df7fb | 1409 | if (ret) { |
a811dc61 | 1410 | listener_f->private_data = NULL; |
6a21cc50 TA |
1411 | fput(listener_f); |
1412 | put_unused_fd(listener); | |
1413 | } else { | |
1414 | fd_install(listener, listener_f); | |
1415 | ret = listener; | |
1416 | } | |
1417 | } | |
c2e1f2e3 | 1418 | out_free: |
c8bee430 | 1419 | seccomp_filter_free(prepared); |
1d9d02fe AA |
1420 | return ret; |
1421 | } | |
3b23dd12 | 1422 | #else |
48dc92b9 KC |
1423 | static inline long seccomp_set_mode_filter(unsigned int flags, |
1424 | const char __user *filter) | |
3b23dd12 KC |
1425 | { |
1426 | return -EINVAL; | |
1427 | } | |
1428 | #endif | |
d78ab02c | 1429 | |
d612b1fd TH |
1430 | static long seccomp_get_action_avail(const char __user *uaction) |
1431 | { | |
1432 | u32 action; | |
1433 | ||
1434 | if (copy_from_user(&action, uaction, sizeof(action))) | |
1435 | return -EFAULT; | |
1436 | ||
1437 | switch (action) { | |
0466bdb9 | 1438 | case SECCOMP_RET_KILL_PROCESS: |
fd76875c | 1439 | case SECCOMP_RET_KILL_THREAD: |
d612b1fd TH |
1440 | case SECCOMP_RET_TRAP: |
1441 | case SECCOMP_RET_ERRNO: | |
6a21cc50 | 1442 | case SECCOMP_RET_USER_NOTIF: |
d612b1fd | 1443 | case SECCOMP_RET_TRACE: |
59f5cf44 | 1444 | case SECCOMP_RET_LOG: |
d612b1fd TH |
1445 | case SECCOMP_RET_ALLOW: |
1446 | break; | |
1447 | default: | |
1448 | return -EOPNOTSUPP; | |
1449 | } | |
1450 | ||
1451 | return 0; | |
1452 | } | |
1453 | ||
6a21cc50 TA |
1454 | static long seccomp_get_notif_sizes(void __user *usizes) |
1455 | { | |
1456 | struct seccomp_notif_sizes sizes = { | |
1457 | .seccomp_notif = sizeof(struct seccomp_notif), | |
1458 | .seccomp_notif_resp = sizeof(struct seccomp_notif_resp), | |
1459 | .seccomp_data = sizeof(struct seccomp_data), | |
1460 | }; | |
1461 | ||
1462 | if (copy_to_user(usizes, &sizes, sizeof(sizes))) | |
1463 | return -EFAULT; | |
1464 | ||
1465 | return 0; | |
1466 | } | |
1467 | ||
48dc92b9 KC |
1468 | /* Common entry point for both prctl and syscall. */ |
1469 | static long do_seccomp(unsigned int op, unsigned int flags, | |
a5662e4d | 1470 | void __user *uargs) |
48dc92b9 KC |
1471 | { |
1472 | switch (op) { | |
1473 | case SECCOMP_SET_MODE_STRICT: | |
1474 | if (flags != 0 || uargs != NULL) | |
1475 | return -EINVAL; | |
1476 | return seccomp_set_mode_strict(); | |
1477 | case SECCOMP_SET_MODE_FILTER: | |
1478 | return seccomp_set_mode_filter(flags, uargs); | |
d612b1fd TH |
1479 | case SECCOMP_GET_ACTION_AVAIL: |
1480 | if (flags != 0) | |
1481 | return -EINVAL; | |
1482 | ||
1483 | return seccomp_get_action_avail(uargs); | |
6a21cc50 TA |
1484 | case SECCOMP_GET_NOTIF_SIZES: |
1485 | if (flags != 0) | |
1486 | return -EINVAL; | |
1487 | ||
1488 | return seccomp_get_notif_sizes(uargs); | |
48dc92b9 KC |
1489 | default: |
1490 | return -EINVAL; | |
1491 | } | |
1492 | } | |
1493 | ||
1494 | SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, | |
a5662e4d | 1495 | void __user *, uargs) |
48dc92b9 KC |
1496 | { |
1497 | return do_seccomp(op, flags, uargs); | |
1498 | } | |
1499 | ||
d78ab02c KC |
1500 | /** |
1501 | * prctl_set_seccomp: configures current->seccomp.mode | |
1502 | * @seccomp_mode: requested mode to use | |
1503 | * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER | |
1504 | * | |
1505 | * Returns 0 on success or -EINVAL on failure. | |
1506 | */ | |
a5662e4d | 1507 | long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter) |
d78ab02c | 1508 | { |
48dc92b9 | 1509 | unsigned int op; |
a5662e4d | 1510 | void __user *uargs; |
48dc92b9 | 1511 | |
3b23dd12 KC |
1512 | switch (seccomp_mode) { |
1513 | case SECCOMP_MODE_STRICT: | |
48dc92b9 KC |
1514 | op = SECCOMP_SET_MODE_STRICT; |
1515 | /* | |
1516 | * Setting strict mode through prctl always ignored filter, | |
1517 | * so make sure it is always NULL here to pass the internal | |
1518 | * check in do_seccomp(). | |
1519 | */ | |
1520 | uargs = NULL; | |
1521 | break; | |
3b23dd12 | 1522 | case SECCOMP_MODE_FILTER: |
48dc92b9 KC |
1523 | op = SECCOMP_SET_MODE_FILTER; |
1524 | uargs = filter; | |
1525 | break; | |
3b23dd12 KC |
1526 | default: |
1527 | return -EINVAL; | |
1528 | } | |
48dc92b9 KC |
1529 | |
1530 | /* prctl interface doesn't have flags, so they are always zero. */ | |
1531 | return do_seccomp(op, 0, uargs); | |
d78ab02c | 1532 | } |
f8e529ed TA |
1533 | |
1534 | #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE) | |
f06eae83 TA |
1535 | static struct seccomp_filter *get_nth_filter(struct task_struct *task, |
1536 | unsigned long filter_off) | |
f8e529ed | 1537 | { |
f06eae83 TA |
1538 | struct seccomp_filter *orig, *filter; |
1539 | unsigned long count; | |
f8e529ed | 1540 | |
f06eae83 TA |
1541 | /* |
1542 | * Note: this is only correct because the caller should be the (ptrace) | |
1543 | * tracer of the task, otherwise lock_task_sighand is needed. | |
1544 | */ | |
f8e529ed | 1545 | spin_lock_irq(&task->sighand->siglock); |
f06eae83 | 1546 | |
f8e529ed | 1547 | if (task->seccomp.mode != SECCOMP_MODE_FILTER) { |
f06eae83 TA |
1548 | spin_unlock_irq(&task->sighand->siglock); |
1549 | return ERR_PTR(-EINVAL); | |
f8e529ed TA |
1550 | } |
1551 | ||
f06eae83 TA |
1552 | orig = task->seccomp.filter; |
1553 | __get_seccomp_filter(orig); | |
1554 | spin_unlock_irq(&task->sighand->siglock); | |
1555 | ||
1556 | count = 0; | |
1557 | for (filter = orig; filter; filter = filter->prev) | |
f8e529ed | 1558 | count++; |
f8e529ed TA |
1559 | |
1560 | if (filter_off >= count) { | |
f06eae83 | 1561 | filter = ERR_PTR(-ENOENT); |
f8e529ed TA |
1562 | goto out; |
1563 | } | |
f8e529ed | 1564 | |
f06eae83 TA |
1565 | count -= filter_off; |
1566 | for (filter = orig; filter && count > 1; filter = filter->prev) | |
f8e529ed | 1567 | count--; |
f8e529ed TA |
1568 | |
1569 | if (WARN_ON(count != 1 || !filter)) { | |
f06eae83 | 1570 | filter = ERR_PTR(-ENOENT); |
f8e529ed TA |
1571 | goto out; |
1572 | } | |
1573 | ||
f06eae83 TA |
1574 | __get_seccomp_filter(filter); |
1575 | ||
1576 | out: | |
1577 | __put_seccomp_filter(orig); | |
1578 | return filter; | |
1579 | } | |
1580 | ||
1581 | long seccomp_get_filter(struct task_struct *task, unsigned long filter_off, | |
1582 | void __user *data) | |
1583 | { | |
1584 | struct seccomp_filter *filter; | |
1585 | struct sock_fprog_kern *fprog; | |
1586 | long ret; | |
1587 | ||
1588 | if (!capable(CAP_SYS_ADMIN) || | |
1589 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { | |
1590 | return -EACCES; | |
1591 | } | |
1592 | ||
1593 | filter = get_nth_filter(task, filter_off); | |
1594 | if (IS_ERR(filter)) | |
1595 | return PTR_ERR(filter); | |
1596 | ||
f8e529ed TA |
1597 | fprog = filter->prog->orig_prog; |
1598 | if (!fprog) { | |
470bf1f2 | 1599 | /* This must be a new non-cBPF filter, since we save |
f8e529ed TA |
1600 | * every cBPF filter's orig_prog above when |
1601 | * CONFIG_CHECKPOINT_RESTORE is enabled. | |
1602 | */ | |
1603 | ret = -EMEDIUMTYPE; | |
1604 | goto out; | |
1605 | } | |
1606 | ||
1607 | ret = fprog->len; | |
1608 | if (!data) | |
1609 | goto out; | |
1610 | ||
f8e529ed TA |
1611 | if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog))) |
1612 | ret = -EFAULT; | |
1613 | ||
f8e529ed | 1614 | out: |
66a733ea | 1615 | __put_seccomp_filter(filter); |
f8e529ed | 1616 | return ret; |
f8e529ed | 1617 | } |
f8e529ed | 1618 | |
26500475 TA |
1619 | long seccomp_get_metadata(struct task_struct *task, |
1620 | unsigned long size, void __user *data) | |
1621 | { | |
1622 | long ret; | |
1623 | struct seccomp_filter *filter; | |
1624 | struct seccomp_metadata kmd = {}; | |
1625 | ||
1626 | if (!capable(CAP_SYS_ADMIN) || | |
1627 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { | |
1628 | return -EACCES; | |
1629 | } | |
1630 | ||
1631 | size = min_t(unsigned long, size, sizeof(kmd)); | |
1632 | ||
63bb0045 TA |
1633 | if (size < sizeof(kmd.filter_off)) |
1634 | return -EINVAL; | |
1635 | ||
1636 | if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off))) | |
26500475 TA |
1637 | return -EFAULT; |
1638 | ||
1639 | filter = get_nth_filter(task, kmd.filter_off); | |
1640 | if (IS_ERR(filter)) | |
1641 | return PTR_ERR(filter); | |
1642 | ||
26500475 TA |
1643 | if (filter->log) |
1644 | kmd.flags |= SECCOMP_FILTER_FLAG_LOG; | |
1645 | ||
1646 | ret = size; | |
1647 | if (copy_to_user(data, &kmd, size)) | |
1648 | ret = -EFAULT; | |
1649 | ||
1650 | __put_seccomp_filter(filter); | |
f8e529ed TA |
1651 | return ret; |
1652 | } | |
1653 | #endif | |
8e5f1ad1 TH |
1654 | |
1655 | #ifdef CONFIG_SYSCTL | |
1656 | ||
1657 | /* Human readable action names for friendly sysctl interaction */ | |
0466bdb9 | 1658 | #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process" |
fd76875c | 1659 | #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread" |
8e5f1ad1 TH |
1660 | #define SECCOMP_RET_TRAP_NAME "trap" |
1661 | #define SECCOMP_RET_ERRNO_NAME "errno" | |
6a21cc50 | 1662 | #define SECCOMP_RET_USER_NOTIF_NAME "user_notif" |
8e5f1ad1 | 1663 | #define SECCOMP_RET_TRACE_NAME "trace" |
59f5cf44 | 1664 | #define SECCOMP_RET_LOG_NAME "log" |
8e5f1ad1 TH |
1665 | #define SECCOMP_RET_ALLOW_NAME "allow" |
1666 | ||
fd76875c | 1667 | static const char seccomp_actions_avail[] = |
0466bdb9 | 1668 | SECCOMP_RET_KILL_PROCESS_NAME " " |
fd76875c KC |
1669 | SECCOMP_RET_KILL_THREAD_NAME " " |
1670 | SECCOMP_RET_TRAP_NAME " " | |
1671 | SECCOMP_RET_ERRNO_NAME " " | |
6a21cc50 | 1672 | SECCOMP_RET_USER_NOTIF_NAME " " |
fd76875c KC |
1673 | SECCOMP_RET_TRACE_NAME " " |
1674 | SECCOMP_RET_LOG_NAME " " | |
1675 | SECCOMP_RET_ALLOW_NAME; | |
8e5f1ad1 | 1676 | |
0ddec0fc TH |
1677 | struct seccomp_log_name { |
1678 | u32 log; | |
1679 | const char *name; | |
1680 | }; | |
1681 | ||
1682 | static const struct seccomp_log_name seccomp_log_names[] = { | |
0466bdb9 | 1683 | { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME }, |
fd76875c | 1684 | { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME }, |
0ddec0fc TH |
1685 | { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME }, |
1686 | { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME }, | |
6a21cc50 | 1687 | { SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME }, |
0ddec0fc | 1688 | { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME }, |
59f5cf44 | 1689 | { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME }, |
0ddec0fc TH |
1690 | { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME }, |
1691 | { } | |
1692 | }; | |
1693 | ||
1694 | static bool seccomp_names_from_actions_logged(char *names, size_t size, | |
beb44aca TH |
1695 | u32 actions_logged, |
1696 | const char *sep) | |
0ddec0fc TH |
1697 | { |
1698 | const struct seccomp_log_name *cur; | |
beb44aca | 1699 | bool append_sep = false; |
0ddec0fc TH |
1700 | |
1701 | for (cur = seccomp_log_names; cur->name && size; cur++) { | |
1702 | ssize_t ret; | |
1703 | ||
1704 | if (!(actions_logged & cur->log)) | |
1705 | continue; | |
1706 | ||
beb44aca TH |
1707 | if (append_sep) { |
1708 | ret = strscpy(names, sep, size); | |
0ddec0fc TH |
1709 | if (ret < 0) |
1710 | return false; | |
1711 | ||
1712 | names += ret; | |
1713 | size -= ret; | |
1714 | } else | |
beb44aca | 1715 | append_sep = true; |
0ddec0fc TH |
1716 | |
1717 | ret = strscpy(names, cur->name, size); | |
1718 | if (ret < 0) | |
1719 | return false; | |
1720 | ||
1721 | names += ret; | |
1722 | size -= ret; | |
1723 | } | |
1724 | ||
1725 | return true; | |
1726 | } | |
1727 | ||
1728 | static bool seccomp_action_logged_from_name(u32 *action_logged, | |
1729 | const char *name) | |
1730 | { | |
1731 | const struct seccomp_log_name *cur; | |
1732 | ||
1733 | for (cur = seccomp_log_names; cur->name; cur++) { | |
1734 | if (!strcmp(cur->name, name)) { | |
1735 | *action_logged = cur->log; | |
1736 | return true; | |
1737 | } | |
1738 | } | |
1739 | ||
1740 | return false; | |
1741 | } | |
1742 | ||
1743 | static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names) | |
1744 | { | |
1745 | char *name; | |
1746 | ||
1747 | *actions_logged = 0; | |
1748 | while ((name = strsep(&names, " ")) && *name) { | |
1749 | u32 action_logged = 0; | |
1750 | ||
1751 | if (!seccomp_action_logged_from_name(&action_logged, name)) | |
1752 | return false; | |
1753 | ||
1754 | *actions_logged |= action_logged; | |
1755 | } | |
1756 | ||
1757 | return true; | |
1758 | } | |
1759 | ||
d013db02 TH |
1760 | static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer, |
1761 | size_t *lenp, loff_t *ppos) | |
0ddec0fc TH |
1762 | { |
1763 | char names[sizeof(seccomp_actions_avail)]; | |
1764 | struct ctl_table table; | |
d013db02 TH |
1765 | |
1766 | memset(names, 0, sizeof(names)); | |
1767 | ||
1768 | if (!seccomp_names_from_actions_logged(names, sizeof(names), | |
beb44aca | 1769 | seccomp_actions_logged, " ")) |
d013db02 TH |
1770 | return -EINVAL; |
1771 | ||
1772 | table = *ro_table; | |
1773 | table.data = names; | |
1774 | table.maxlen = sizeof(names); | |
1775 | return proc_dostring(&table, 0, buffer, lenp, ppos); | |
1776 | } | |
1777 | ||
1778 | static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer, | |
ea6eca77 | 1779 | size_t *lenp, loff_t *ppos, u32 *actions_logged) |
0ddec0fc TH |
1780 | { |
1781 | char names[sizeof(seccomp_actions_avail)]; | |
1782 | struct ctl_table table; | |
1783 | int ret; | |
1784 | ||
d013db02 | 1785 | if (!capable(CAP_SYS_ADMIN)) |
0ddec0fc TH |
1786 | return -EPERM; |
1787 | ||
1788 | memset(names, 0, sizeof(names)); | |
1789 | ||
0ddec0fc TH |
1790 | table = *ro_table; |
1791 | table.data = names; | |
1792 | table.maxlen = sizeof(names); | |
d013db02 | 1793 | ret = proc_dostring(&table, 1, buffer, lenp, ppos); |
0ddec0fc TH |
1794 | if (ret) |
1795 | return ret; | |
1796 | ||
ea6eca77 | 1797 | if (!seccomp_actions_logged_from_names(actions_logged, table.data)) |
d013db02 | 1798 | return -EINVAL; |
0ddec0fc | 1799 | |
ea6eca77 | 1800 | if (*actions_logged & SECCOMP_LOG_ALLOW) |
d013db02 | 1801 | return -EINVAL; |
0ddec0fc | 1802 | |
ea6eca77 | 1803 | seccomp_actions_logged = *actions_logged; |
0ddec0fc TH |
1804 | return 0; |
1805 | } | |
0ddec0fc | 1806 | |
ea6eca77 TH |
1807 | static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged, |
1808 | int ret) | |
1809 | { | |
1810 | char names[sizeof(seccomp_actions_avail)]; | |
1811 | char old_names[sizeof(seccomp_actions_avail)]; | |
1812 | const char *new = names; | |
1813 | const char *old = old_names; | |
0ddec0fc | 1814 | |
ea6eca77 TH |
1815 | if (!audit_enabled) |
1816 | return; | |
1817 | ||
1818 | memset(names, 0, sizeof(names)); | |
1819 | memset(old_names, 0, sizeof(old_names)); | |
1820 | ||
1821 | if (ret) | |
1822 | new = "?"; | |
1823 | else if (!actions_logged) | |
1824 | new = "(none)"; | |
1825 | else if (!seccomp_names_from_actions_logged(names, sizeof(names), | |
1826 | actions_logged, ",")) | |
1827 | new = "?"; | |
1828 | ||
1829 | if (!old_actions_logged) | |
1830 | old = "(none)"; | |
1831 | else if (!seccomp_names_from_actions_logged(old_names, | |
1832 | sizeof(old_names), | |
1833 | old_actions_logged, ",")) | |
1834 | old = "?"; | |
1835 | ||
1836 | return audit_seccomp_actions_logged(new, old, !ret); | |
1837 | } | |
1838 | ||
d013db02 | 1839 | static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write, |
32927393 | 1840 | void *buffer, size_t *lenp, |
d013db02 TH |
1841 | loff_t *ppos) |
1842 | { | |
ea6eca77 TH |
1843 | int ret; |
1844 | ||
1845 | if (write) { | |
1846 | u32 actions_logged = 0; | |
1847 | u32 old_actions_logged = seccomp_actions_logged; | |
1848 | ||
1849 | ret = write_actions_logged(ro_table, buffer, lenp, ppos, | |
1850 | &actions_logged); | |
1851 | audit_actions_logged(actions_logged, old_actions_logged, ret); | |
1852 | } else | |
1853 | ret = read_actions_logged(ro_table, buffer, lenp, ppos); | |
1854 | ||
1855 | return ret; | |
0ddec0fc TH |
1856 | } |
1857 | ||
8e5f1ad1 TH |
1858 | static struct ctl_path seccomp_sysctl_path[] = { |
1859 | { .procname = "kernel", }, | |
1860 | { .procname = "seccomp", }, | |
1861 | { } | |
1862 | }; | |
1863 | ||
1864 | static struct ctl_table seccomp_sysctl_table[] = { | |
1865 | { | |
1866 | .procname = "actions_avail", | |
1867 | .data = (void *) &seccomp_actions_avail, | |
1868 | .maxlen = sizeof(seccomp_actions_avail), | |
1869 | .mode = 0444, | |
1870 | .proc_handler = proc_dostring, | |
1871 | }, | |
0ddec0fc TH |
1872 | { |
1873 | .procname = "actions_logged", | |
1874 | .mode = 0644, | |
1875 | .proc_handler = seccomp_actions_logged_handler, | |
1876 | }, | |
8e5f1ad1 TH |
1877 | { } |
1878 | }; | |
1879 | ||
1880 | static int __init seccomp_sysctl_init(void) | |
1881 | { | |
1882 | struct ctl_table_header *hdr; | |
1883 | ||
1884 | hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table); | |
1885 | if (!hdr) | |
e68f9d49 | 1886 | pr_warn("sysctl registration failed\n"); |
8e5f1ad1 TH |
1887 | else |
1888 | kmemleak_not_leak(hdr); | |
1889 | ||
1890 | return 0; | |
1891 | } | |
1892 | ||
1893 | device_initcall(seccomp_sysctl_init) | |
1894 | ||
1895 | #endif /* CONFIG_SYSCTL */ |