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 LT |
15 | */ |
16 | ||
0b5fa229 | 17 | #include <linux/refcount.h> |
85e7bac3 | 18 | #include <linux/audit.h> |
5b101740 | 19 | #include <linux/compat.h> |
b25e6716 | 20 | #include <linux/coredump.h> |
8e5f1ad1 | 21 | #include <linux/kmemleak.h> |
5c307089 KC |
22 | #include <linux/nospec.h> |
23 | #include <linux/prctl.h> | |
e2cfabdf | 24 | #include <linux/sched.h> |
68db0cf1 | 25 | #include <linux/sched/task_stack.h> |
e2cfabdf | 26 | #include <linux/seccomp.h> |
c8bee430 | 27 | #include <linux/slab.h> |
48dc92b9 | 28 | #include <linux/syscalls.h> |
8e5f1ad1 | 29 | #include <linux/sysctl.h> |
1da177e4 | 30 | |
a4412fc9 | 31 | #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER |
e2cfabdf | 32 | #include <asm/syscall.h> |
a4412fc9 | 33 | #endif |
e2cfabdf WD |
34 | |
35 | #ifdef CONFIG_SECCOMP_FILTER | |
6a21cc50 | 36 | #include <linux/file.h> |
e2cfabdf | 37 | #include <linux/filter.h> |
c2e1f2e3 | 38 | #include <linux/pid.h> |
fb0fadf9 | 39 | #include <linux/ptrace.h> |
e2cfabdf | 40 | #include <linux/security.h> |
e2cfabdf WD |
41 | #include <linux/tracehook.h> |
42 | #include <linux/uaccess.h> | |
6a21cc50 | 43 | #include <linux/anon_inodes.h> |
9f87dcf1 | 44 | #include <linux/lockdep.h> |
6a21cc50 TA |
45 | |
46 | enum notify_state { | |
47 | SECCOMP_NOTIFY_INIT, | |
48 | SECCOMP_NOTIFY_SENT, | |
49 | SECCOMP_NOTIFY_REPLIED, | |
50 | }; | |
51 | ||
52 | struct seccomp_knotif { | |
53 | /* The struct pid of the task whose filter triggered the notification */ | |
54 | struct task_struct *task; | |
55 | ||
56 | /* The "cookie" for this request; this is unique for this filter. */ | |
57 | u64 id; | |
58 | ||
59 | /* | |
60 | * The seccomp data. This pointer is valid the entire time this | |
61 | * notification is active, since it comes from __seccomp_filter which | |
62 | * eclipses the entire lifecycle here. | |
63 | */ | |
64 | const struct seccomp_data *data; | |
65 | ||
66 | /* | |
67 | * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a | |
68 | * struct seccomp_knotif is created and starts out in INIT. Once the | |
69 | * handler reads the notification off of an FD, it transitions to SENT. | |
70 | * If a signal is received the state transitions back to INIT and | |
71 | * another message is sent. When the userspace handler replies, state | |
72 | * transitions to REPLIED. | |
73 | */ | |
74 | enum notify_state state; | |
75 | ||
76 | /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */ | |
77 | int error; | |
78 | long val; | |
fb3c5386 | 79 | u32 flags; |
6a21cc50 TA |
80 | |
81 | /* Signals when this has entered SECCOMP_NOTIFY_REPLIED */ | |
82 | struct completion ready; | |
83 | ||
84 | struct list_head list; | |
85 | }; | |
86 | ||
87 | /** | |
88 | * struct notification - container for seccomp userspace notifications. Since | |
89 | * most seccomp filters will not have notification listeners attached and this | |
90 | * structure is fairly large, we store the notification-specific stuff in a | |
91 | * separate structure. | |
92 | * | |
93 | * @request: A semaphore that users of this notification can wait on for | |
94 | * changes. Actual reads and writes are still controlled with | |
95 | * filter->notify_lock. | |
96 | * @next_id: The id of the next request. | |
97 | * @notifications: A list of struct seccomp_knotif elements. | |
98 | * @wqh: A wait queue for poll. | |
99 | */ | |
100 | struct notification { | |
101 | struct semaphore request; | |
102 | u64 next_id; | |
103 | struct list_head notifications; | |
104 | wait_queue_head_t wqh; | |
105 | }; | |
e2cfabdf WD |
106 | |
107 | /** | |
108 | * struct seccomp_filter - container for seccomp BPF programs | |
109 | * | |
b707ddee CB |
110 | * @refs: Reference count to manage the object lifetime. |
111 | * A filter's reference count is incremented for each directly | |
112 | * attached task, once for the dependent filter, and if | |
113 | * requested for the user notifier. When @refs reaches zero, | |
114 | * the filter can be freed. | |
e66a3997 | 115 | * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged |
e2cfabdf | 116 | * @prev: points to a previously installed, or inherited, filter |
285fdfc5 | 117 | * @prog: the BPF program to evaluate |
6a21cc50 TA |
118 | * @notif: the struct that holds all notification related information |
119 | * @notify_lock: A lock for all notification-related accesses. | |
e2cfabdf WD |
120 | * |
121 | * seccomp_filter objects are organized in a tree linked via the @prev | |
122 | * pointer. For any task, it appears to be a singly-linked list starting | |
123 | * with current->seccomp.filter, the most recently attached or inherited filter. | |
124 | * However, multiple filters may share a @prev node, by way of fork(), which | |
125 | * results in a unidirectional tree existing in memory. This is similar to | |
126 | * how namespaces work. | |
127 | * | |
128 | * seccomp_filter objects should never be modified after being attached | |
b707ddee | 129 | * to a task_struct (other than @refs). |
e2cfabdf WD |
130 | */ |
131 | struct seccomp_filter { | |
b707ddee | 132 | refcount_t refs; |
e66a3997 | 133 | bool log; |
e2cfabdf | 134 | struct seccomp_filter *prev; |
7ae457c1 | 135 | struct bpf_prog *prog; |
6a21cc50 TA |
136 | struct notification *notif; |
137 | struct mutex notify_lock; | |
e2cfabdf WD |
138 | }; |
139 | ||
140 | /* Limit any path through the tree to 256KB worth of instructions. */ | |
141 | #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter)) | |
142 | ||
bd4cf0ed | 143 | /* |
e2cfabdf WD |
144 | * Endianness is explicitly ignored and left for BPF program authors to manage |
145 | * as per the specific architecture. | |
146 | */ | |
bd4cf0ed | 147 | static void populate_seccomp_data(struct seccomp_data *sd) |
e2cfabdf | 148 | { |
bd4cf0ed AS |
149 | struct task_struct *task = current; |
150 | struct pt_regs *regs = task_pt_regs(task); | |
2eac7648 | 151 | unsigned long args[6]; |
e2cfabdf | 152 | |
bd4cf0ed | 153 | sd->nr = syscall_get_nr(task, regs); |
16add411 | 154 | sd->arch = syscall_get_arch(task); |
b35f549d | 155 | syscall_get_arguments(task, regs, args); |
2eac7648 DB |
156 | sd->args[0] = args[0]; |
157 | sd->args[1] = args[1]; | |
158 | sd->args[2] = args[2]; | |
159 | sd->args[3] = args[3]; | |
160 | sd->args[4] = args[4]; | |
161 | sd->args[5] = args[5]; | |
bd4cf0ed | 162 | sd->instruction_pointer = KSTK_EIP(task); |
e2cfabdf WD |
163 | } |
164 | ||
165 | /** | |
166 | * seccomp_check_filter - verify seccomp filter code | |
167 | * @filter: filter to verify | |
168 | * @flen: length of filter | |
169 | * | |
4df95ff4 | 170 | * Takes a previously checked filter (by bpf_check_classic) and |
e2cfabdf WD |
171 | * redirects all filter code that loads struct sk_buff data |
172 | * and related data through seccomp_bpf_load. It also | |
173 | * enforces length and alignment checking of those loads. | |
174 | * | |
175 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
176 | */ | |
177 | static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen) | |
178 | { | |
179 | int pc; | |
180 | for (pc = 0; pc < flen; pc++) { | |
181 | struct sock_filter *ftest = &filter[pc]; | |
182 | u16 code = ftest->code; | |
183 | u32 k = ftest->k; | |
184 | ||
185 | switch (code) { | |
34805931 | 186 | case BPF_LD | BPF_W | BPF_ABS: |
bd4cf0ed | 187 | ftest->code = BPF_LDX | BPF_W | BPF_ABS; |
e2cfabdf WD |
188 | /* 32-bit aligned and not out of bounds. */ |
189 | if (k >= sizeof(struct seccomp_data) || k & 3) | |
190 | return -EINVAL; | |
191 | continue; | |
34805931 | 192 | case BPF_LD | BPF_W | BPF_LEN: |
bd4cf0ed | 193 | ftest->code = BPF_LD | BPF_IMM; |
e2cfabdf WD |
194 | ftest->k = sizeof(struct seccomp_data); |
195 | continue; | |
34805931 | 196 | case BPF_LDX | BPF_W | BPF_LEN: |
bd4cf0ed | 197 | ftest->code = BPF_LDX | BPF_IMM; |
e2cfabdf WD |
198 | ftest->k = sizeof(struct seccomp_data); |
199 | continue; | |
200 | /* Explicitly include allowed calls. */ | |
34805931 DB |
201 | case BPF_RET | BPF_K: |
202 | case BPF_RET | BPF_A: | |
203 | case BPF_ALU | BPF_ADD | BPF_K: | |
204 | case BPF_ALU | BPF_ADD | BPF_X: | |
205 | case BPF_ALU | BPF_SUB | BPF_K: | |
206 | case BPF_ALU | BPF_SUB | BPF_X: | |
207 | case BPF_ALU | BPF_MUL | BPF_K: | |
208 | case BPF_ALU | BPF_MUL | BPF_X: | |
209 | case BPF_ALU | BPF_DIV | BPF_K: | |
210 | case BPF_ALU | BPF_DIV | BPF_X: | |
211 | case BPF_ALU | BPF_AND | BPF_K: | |
212 | case BPF_ALU | BPF_AND | BPF_X: | |
213 | case BPF_ALU | BPF_OR | BPF_K: | |
214 | case BPF_ALU | BPF_OR | BPF_X: | |
215 | case BPF_ALU | BPF_XOR | BPF_K: | |
216 | case BPF_ALU | BPF_XOR | BPF_X: | |
217 | case BPF_ALU | BPF_LSH | BPF_K: | |
218 | case BPF_ALU | BPF_LSH | BPF_X: | |
219 | case BPF_ALU | BPF_RSH | BPF_K: | |
220 | case BPF_ALU | BPF_RSH | BPF_X: | |
221 | case BPF_ALU | BPF_NEG: | |
222 | case BPF_LD | BPF_IMM: | |
223 | case BPF_LDX | BPF_IMM: | |
224 | case BPF_MISC | BPF_TAX: | |
225 | case BPF_MISC | BPF_TXA: | |
226 | case BPF_LD | BPF_MEM: | |
227 | case BPF_LDX | BPF_MEM: | |
228 | case BPF_ST: | |
229 | case BPF_STX: | |
230 | case BPF_JMP | BPF_JA: | |
231 | case BPF_JMP | BPF_JEQ | BPF_K: | |
232 | case BPF_JMP | BPF_JEQ | BPF_X: | |
233 | case BPF_JMP | BPF_JGE | BPF_K: | |
234 | case BPF_JMP | BPF_JGE | BPF_X: | |
235 | case BPF_JMP | BPF_JGT | BPF_K: | |
236 | case BPF_JMP | BPF_JGT | BPF_X: | |
237 | case BPF_JMP | BPF_JSET | BPF_K: | |
238 | case BPF_JMP | BPF_JSET | BPF_X: | |
e2cfabdf WD |
239 | continue; |
240 | default: | |
241 | return -EINVAL; | |
242 | } | |
243 | } | |
244 | return 0; | |
245 | } | |
246 | ||
247 | /** | |
285fdfc5 MS |
248 | * seccomp_run_filters - evaluates all seccomp filters against @sd |
249 | * @sd: optional seccomp data to be passed to filters | |
deb4de8b KC |
250 | * @match: stores struct seccomp_filter that resulted in the return value, |
251 | * unless filter returned SECCOMP_RET_ALLOW, in which case it will | |
252 | * be unchanged. | |
e2cfabdf WD |
253 | * |
254 | * Returns valid seccomp BPF response codes. | |
255 | */ | |
0466bdb9 | 256 | #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL))) |
deb4de8b KC |
257 | static u32 seccomp_run_filters(const struct seccomp_data *sd, |
258 | struct seccomp_filter **match) | |
e2cfabdf | 259 | { |
acf3b2c7 | 260 | u32 ret = SECCOMP_RET_ALLOW; |
8225d385 PK |
261 | /* Make sure cross-thread synced filter points somewhere sane. */ |
262 | struct seccomp_filter *f = | |
506458ef | 263 | READ_ONCE(current->seccomp.filter); |
acf3b2c7 WD |
264 | |
265 | /* Ensure unexpected behavior doesn't result in failing open. */ | |
0d42d73a | 266 | if (WARN_ON(f == NULL)) |
4d3b0b05 | 267 | return SECCOMP_RET_KILL_PROCESS; |
acf3b2c7 | 268 | |
e2cfabdf WD |
269 | /* |
270 | * All filters in the list are evaluated and the lowest BPF return | |
acf3b2c7 | 271 | * value always takes priority (ignoring the DATA). |
e2cfabdf | 272 | */ |
3ba2530c | 273 | for (; f; f = f->prev) { |
3d9f773c | 274 | u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd); |
8f577cad | 275 | |
0466bdb9 | 276 | if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) { |
acf3b2c7 | 277 | ret = cur_ret; |
deb4de8b KC |
278 | *match = f; |
279 | } | |
e2cfabdf WD |
280 | } |
281 | return ret; | |
282 | } | |
1f41b450 | 283 | #endif /* CONFIG_SECCOMP_FILTER */ |
e2cfabdf | 284 | |
1f41b450 KC |
285 | static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) |
286 | { | |
69f6a34b | 287 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 288 | |
1f41b450 KC |
289 | if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) |
290 | return false; | |
291 | ||
292 | return true; | |
293 | } | |
294 | ||
8bf37d8c | 295 | void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { } |
5c307089 | 296 | |
3ba2530c | 297 | static inline void seccomp_assign_mode(struct task_struct *task, |
00a02d0c KC |
298 | unsigned long seccomp_mode, |
299 | unsigned long flags) | |
1f41b450 | 300 | { |
69f6a34b | 301 | assert_spin_locked(&task->sighand->siglock); |
dbd95212 | 302 | |
3ba2530c KC |
303 | task->seccomp.mode = seccomp_mode; |
304 | /* | |
305 | * Make sure TIF_SECCOMP cannot be set before the mode (and | |
306 | * filter) is set. | |
307 | */ | |
308 | smp_mb__before_atomic(); | |
00a02d0c KC |
309 | /* Assume default seccomp processes want spec flaw mitigation. */ |
310 | if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0) | |
8bf37d8c | 311 | arch_seccomp_spec_mitigate(task); |
3ba2530c | 312 | set_tsk_thread_flag(task, TIF_SECCOMP); |
1f41b450 KC |
313 | } |
314 | ||
315 | #ifdef CONFIG_SECCOMP_FILTER | |
c2e1f2e3 KC |
316 | /* Returns 1 if the parent is an ancestor of the child. */ |
317 | static int is_ancestor(struct seccomp_filter *parent, | |
318 | struct seccomp_filter *child) | |
319 | { | |
320 | /* NULL is the root ancestor. */ | |
321 | if (parent == NULL) | |
322 | return 1; | |
323 | for (; child; child = child->prev) | |
324 | if (child == parent) | |
325 | return 1; | |
326 | return 0; | |
327 | } | |
328 | ||
329 | /** | |
330 | * seccomp_can_sync_threads: checks if all threads can be synchronized | |
331 | * | |
332 | * Expects sighand and cred_guard_mutex locks to be held. | |
333 | * | |
334 | * Returns 0 on success, -ve on error, or the pid of a thread which was | |
6beff00b | 335 | * either not in the correct seccomp mode or did not have an ancestral |
c2e1f2e3 KC |
336 | * seccomp filter. |
337 | */ | |
338 | static inline pid_t seccomp_can_sync_threads(void) | |
339 | { | |
340 | struct task_struct *thread, *caller; | |
341 | ||
342 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 343 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
344 | |
345 | /* Validate all threads being eligible for synchronization. */ | |
346 | caller = current; | |
347 | for_each_thread(caller, thread) { | |
348 | pid_t failed; | |
349 | ||
350 | /* Skip current, since it is initiating the sync. */ | |
351 | if (thread == caller) | |
352 | continue; | |
353 | ||
354 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || | |
355 | (thread->seccomp.mode == SECCOMP_MODE_FILTER && | |
356 | is_ancestor(thread->seccomp.filter, | |
357 | caller->seccomp.filter))) | |
358 | continue; | |
359 | ||
360 | /* Return the first thread that cannot be synchronized. */ | |
361 | failed = task_pid_vnr(thread); | |
362 | /* If the pid cannot be resolved, then return -ESRCH */ | |
0d42d73a | 363 | if (WARN_ON(failed == 0)) |
c2e1f2e3 KC |
364 | failed = -ESRCH; |
365 | return failed; | |
366 | } | |
367 | ||
368 | return 0; | |
369 | } | |
370 | ||
3a15fb6e CB |
371 | static inline void seccomp_filter_free(struct seccomp_filter *filter) |
372 | { | |
373 | if (filter) { | |
374 | bpf_prog_destroy(filter->prog); | |
375 | kfree(filter); | |
376 | } | |
377 | } | |
378 | ||
379 | static void __put_seccomp_filter(struct seccomp_filter *orig) | |
380 | { | |
381 | /* Clean up single-reference branches iteratively. */ | |
382 | while (orig && refcount_dec_and_test(&orig->refs)) { | |
383 | struct seccomp_filter *freeme = orig; | |
384 | orig = orig->prev; | |
385 | seccomp_filter_free(freeme); | |
386 | } | |
387 | } | |
388 | ||
389 | /** | |
390 | * seccomp_filter_release - Detach the task from its filter tree | |
391 | * and drop its reference count during | |
392 | * exit. | |
393 | * | |
394 | * This function should only be called when the task is exiting as | |
395 | * it detaches it from its filter tree. As such, READ_ONCE() and | |
396 | * barriers are not needed here, as would normally be needed. | |
397 | */ | |
398 | void seccomp_filter_release(struct task_struct *tsk) | |
399 | { | |
400 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
401 | ||
402 | /* Detach task from its filter tree. */ | |
403 | tsk->seccomp.filter = NULL; | |
404 | __put_seccomp_filter(orig); | |
405 | } | |
406 | ||
c2e1f2e3 KC |
407 | /** |
408 | * seccomp_sync_threads: sets all threads to use current's filter | |
409 | * | |
410 | * Expects sighand and cred_guard_mutex locks to be held, and for | |
411 | * seccomp_can_sync_threads() to have returned success already | |
412 | * without dropping the locks. | |
413 | * | |
414 | */ | |
00a02d0c | 415 | static inline void seccomp_sync_threads(unsigned long flags) |
c2e1f2e3 KC |
416 | { |
417 | struct task_struct *thread, *caller; | |
418 | ||
419 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 420 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
421 | |
422 | /* Synchronize all threads. */ | |
423 | caller = current; | |
424 | for_each_thread(caller, thread) { | |
425 | /* Skip current, since it needs no changes. */ | |
426 | if (thread == caller) | |
427 | continue; | |
428 | ||
429 | /* Get a task reference for the new leaf node. */ | |
430 | get_seccomp_filter(caller); | |
431 | /* | |
432 | * Drop the task reference to the shared ancestor since | |
433 | * current's path will hold a reference. (This also | |
434 | * allows a put before the assignment.) | |
435 | */ | |
3a15fb6e | 436 | __put_seccomp_filter(thread->seccomp.filter); |
c2e1f2e3 KC |
437 | smp_store_release(&thread->seccomp.filter, |
438 | caller->seccomp.filter); | |
c818c03b KC |
439 | atomic_set(&thread->seccomp.filter_count, |
440 | atomic_read(&thread->seccomp.filter_count)); | |
103502a3 JH |
441 | |
442 | /* | |
443 | * Don't let an unprivileged task work around | |
444 | * the no_new_privs restriction by creating | |
445 | * a thread that sets it up, enters seccomp, | |
446 | * then dies. | |
447 | */ | |
448 | if (task_no_new_privs(caller)) | |
449 | task_set_no_new_privs(thread); | |
450 | ||
c2e1f2e3 KC |
451 | /* |
452 | * Opt the other thread into seccomp if needed. | |
453 | * As threads are considered to be trust-realm | |
454 | * equivalent (see ptrace_may_access), it is safe to | |
455 | * allow one thread to transition the other. | |
456 | */ | |
103502a3 | 457 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) |
00a02d0c KC |
458 | seccomp_assign_mode(thread, SECCOMP_MODE_FILTER, |
459 | flags); | |
c2e1f2e3 KC |
460 | } |
461 | } | |
462 | ||
e2cfabdf | 463 | /** |
c8bee430 | 464 | * seccomp_prepare_filter: Prepares a seccomp filter for use. |
e2cfabdf WD |
465 | * @fprog: BPF program to install |
466 | * | |
c8bee430 | 467 | * Returns filter on success or an ERR_PTR on failure. |
e2cfabdf | 468 | */ |
c8bee430 | 469 | static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog) |
e2cfabdf | 470 | { |
ac67eb2c DB |
471 | struct seccomp_filter *sfilter; |
472 | int ret; | |
97f2645f | 473 | const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE); |
e2cfabdf WD |
474 | |
475 | if (fprog->len == 0 || fprog->len > BPF_MAXINSNS) | |
c8bee430 | 476 | return ERR_PTR(-EINVAL); |
d9e12f42 | 477 | |
c8bee430 | 478 | BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter)); |
e2cfabdf WD |
479 | |
480 | /* | |
119ce5c8 | 481 | * Installing a seccomp filter requires that the task has |
e2cfabdf WD |
482 | * CAP_SYS_ADMIN in its namespace or be running with no_new_privs. |
483 | * This avoids scenarios where unprivileged tasks can affect the | |
484 | * behavior of privileged children. | |
485 | */ | |
1d4457f9 | 486 | if (!task_no_new_privs(current) && |
c1a85a00 MM |
487 | security_capable(current_cred(), current_user_ns(), |
488 | CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) != 0) | |
c8bee430 | 489 | return ERR_PTR(-EACCES); |
e2cfabdf | 490 | |
bd4cf0ed | 491 | /* Allocate a new seccomp_filter */ |
ac67eb2c DB |
492 | sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN); |
493 | if (!sfilter) | |
d9e12f42 | 494 | return ERR_PTR(-ENOMEM); |
ac67eb2c | 495 | |
6a21cc50 | 496 | mutex_init(&sfilter->notify_lock); |
ac67eb2c | 497 | ret = bpf_prog_create_from_user(&sfilter->prog, fprog, |
f8e529ed | 498 | seccomp_check_filter, save_orig); |
ac67eb2c DB |
499 | if (ret < 0) { |
500 | kfree(sfilter); | |
501 | return ERR_PTR(ret); | |
d9e12f42 | 502 | } |
bd4cf0ed | 503 | |
b707ddee | 504 | refcount_set(&sfilter->refs, 1); |
e2cfabdf | 505 | |
ac67eb2c | 506 | return sfilter; |
e2cfabdf WD |
507 | } |
508 | ||
509 | /** | |
c8bee430 | 510 | * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog |
e2cfabdf WD |
511 | * @user_filter: pointer to the user data containing a sock_fprog. |
512 | * | |
513 | * Returns 0 on success and non-zero otherwise. | |
514 | */ | |
c8bee430 KC |
515 | static struct seccomp_filter * |
516 | seccomp_prepare_user_filter(const char __user *user_filter) | |
e2cfabdf WD |
517 | { |
518 | struct sock_fprog fprog; | |
c8bee430 | 519 | struct seccomp_filter *filter = ERR_PTR(-EFAULT); |
e2cfabdf WD |
520 | |
521 | #ifdef CONFIG_COMPAT | |
5c38065e | 522 | if (in_compat_syscall()) { |
e2cfabdf WD |
523 | struct compat_sock_fprog fprog32; |
524 | if (copy_from_user(&fprog32, user_filter, sizeof(fprog32))) | |
525 | goto out; | |
526 | fprog.len = fprog32.len; | |
527 | fprog.filter = compat_ptr(fprog32.filter); | |
528 | } else /* falls through to the if below. */ | |
529 | #endif | |
530 | if (copy_from_user(&fprog, user_filter, sizeof(fprog))) | |
531 | goto out; | |
c8bee430 | 532 | filter = seccomp_prepare_filter(&fprog); |
e2cfabdf | 533 | out: |
c8bee430 KC |
534 | return filter; |
535 | } | |
536 | ||
537 | /** | |
538 | * seccomp_attach_filter: validate and attach filter | |
539 | * @flags: flags to change filter behavior | |
540 | * @filter: seccomp filter to add to the current process | |
541 | * | |
dbd95212 KC |
542 | * Caller must be holding current->sighand->siglock lock. |
543 | * | |
7a0df7fb TA |
544 | * Returns 0 on success, -ve on error, or |
545 | * - in TSYNC mode: the pid of a thread which was either not in the correct | |
546 | * seccomp mode or did not have an ancestral seccomp filter | |
547 | * - in NEW_LISTENER mode: the fd of the new listener | |
c8bee430 KC |
548 | */ |
549 | static long seccomp_attach_filter(unsigned int flags, | |
550 | struct seccomp_filter *filter) | |
551 | { | |
552 | unsigned long total_insns; | |
553 | struct seccomp_filter *walker; | |
554 | ||
69f6a34b | 555 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 556 | |
c8bee430 KC |
557 | /* Validate resulting filter length. */ |
558 | total_insns = filter->prog->len; | |
559 | for (walker = current->seccomp.filter; walker; walker = walker->prev) | |
560 | total_insns += walker->prog->len + 4; /* 4 instr penalty */ | |
561 | if (total_insns > MAX_INSNS_PER_PATH) | |
562 | return -ENOMEM; | |
563 | ||
c2e1f2e3 KC |
564 | /* If thread sync has been requested, check that it is possible. */ |
565 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) { | |
566 | int ret; | |
567 | ||
568 | ret = seccomp_can_sync_threads(); | |
51891498 TA |
569 | if (ret) { |
570 | if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) | |
571 | return -ESRCH; | |
572 | else | |
573 | return ret; | |
574 | } | |
c2e1f2e3 KC |
575 | } |
576 | ||
e66a3997 TH |
577 | /* Set log flag, if present. */ |
578 | if (flags & SECCOMP_FILTER_FLAG_LOG) | |
579 | filter->log = true; | |
580 | ||
c8bee430 KC |
581 | /* |
582 | * If there is an existing filter, make it the prev and don't drop its | |
583 | * task reference. | |
584 | */ | |
585 | filter->prev = current->seccomp.filter; | |
586 | current->seccomp.filter = filter; | |
c818c03b | 587 | atomic_inc(¤t->seccomp.filter_count); |
c8bee430 | 588 | |
c2e1f2e3 KC |
589 | /* Now that the new filter is in place, synchronize to all threads. */ |
590 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) | |
00a02d0c | 591 | seccomp_sync_threads(flags); |
c2e1f2e3 | 592 | |
c8bee430 | 593 | return 0; |
e2cfabdf WD |
594 | } |
595 | ||
084f5601 | 596 | static void __get_seccomp_filter(struct seccomp_filter *filter) |
66a733ea | 597 | { |
b707ddee | 598 | refcount_inc(&filter->refs); |
66a733ea ON |
599 | } |
600 | ||
e2cfabdf WD |
601 | /* get_seccomp_filter - increments the reference count of the filter on @tsk */ |
602 | void get_seccomp_filter(struct task_struct *tsk) | |
603 | { | |
604 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
605 | if (!orig) | |
606 | return; | |
66a733ea | 607 | __get_seccomp_filter(orig); |
e2cfabdf WD |
608 | } |
609 | ||
ae7795bc | 610 | static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason) |
b25e6716 | 611 | { |
3b10db2b | 612 | clear_siginfo(info); |
b25e6716 MF |
613 | info->si_signo = SIGSYS; |
614 | info->si_code = SYS_SECCOMP; | |
615 | info->si_call_addr = (void __user *)KSTK_EIP(current); | |
616 | info->si_errno = reason; | |
16add411 | 617 | info->si_arch = syscall_get_arch(current); |
b25e6716 MF |
618 | info->si_syscall = syscall; |
619 | } | |
620 | ||
bb6ea430 WD |
621 | /** |
622 | * seccomp_send_sigsys - signals the task to allow in-process syscall emulation | |
623 | * @syscall: syscall number to send to userland | |
624 | * @reason: filter-supplied reason code to send to userland (via si_errno) | |
625 | * | |
626 | * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info. | |
627 | */ | |
628 | static void seccomp_send_sigsys(int syscall, int reason) | |
629 | { | |
ae7795bc | 630 | struct kernel_siginfo info; |
b25e6716 | 631 | seccomp_init_siginfo(&info, syscall, reason); |
a89e9b8a | 632 | force_sig_info(&info); |
bb6ea430 | 633 | } |
e2cfabdf | 634 | #endif /* CONFIG_SECCOMP_FILTER */ |
1da177e4 | 635 | |
0ddec0fc | 636 | /* For use with seccomp_actions_logged */ |
4d3b0b05 KC |
637 | #define SECCOMP_LOG_KILL_PROCESS (1 << 0) |
638 | #define SECCOMP_LOG_KILL_THREAD (1 << 1) | |
0ddec0fc TH |
639 | #define SECCOMP_LOG_TRAP (1 << 2) |
640 | #define SECCOMP_LOG_ERRNO (1 << 3) | |
641 | #define SECCOMP_LOG_TRACE (1 << 4) | |
59f5cf44 TH |
642 | #define SECCOMP_LOG_LOG (1 << 5) |
643 | #define SECCOMP_LOG_ALLOW (1 << 6) | |
6a21cc50 | 644 | #define SECCOMP_LOG_USER_NOTIF (1 << 7) |
0ddec0fc | 645 | |
4d3b0b05 KC |
646 | static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS | |
647 | SECCOMP_LOG_KILL_THREAD | | |
fd76875c KC |
648 | SECCOMP_LOG_TRAP | |
649 | SECCOMP_LOG_ERRNO | | |
6a21cc50 | 650 | SECCOMP_LOG_USER_NOTIF | |
fd76875c | 651 | SECCOMP_LOG_TRACE | |
59f5cf44 | 652 | SECCOMP_LOG_LOG; |
0ddec0fc | 653 | |
e66a3997 TH |
654 | static inline void seccomp_log(unsigned long syscall, long signr, u32 action, |
655 | bool requested) | |
0ddec0fc TH |
656 | { |
657 | bool log = false; | |
658 | ||
659 | switch (action) { | |
660 | case SECCOMP_RET_ALLOW: | |
e66a3997 | 661 | break; |
0ddec0fc | 662 | case SECCOMP_RET_TRAP: |
e66a3997 TH |
663 | log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP; |
664 | break; | |
0ddec0fc | 665 | case SECCOMP_RET_ERRNO: |
e66a3997 TH |
666 | log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO; |
667 | break; | |
0ddec0fc | 668 | case SECCOMP_RET_TRACE: |
e66a3997 | 669 | log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE; |
0ddec0fc | 670 | break; |
6a21cc50 TA |
671 | case SECCOMP_RET_USER_NOTIF: |
672 | log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF; | |
673 | break; | |
59f5cf44 TH |
674 | case SECCOMP_RET_LOG: |
675 | log = seccomp_actions_logged & SECCOMP_LOG_LOG; | |
676 | break; | |
fd76875c | 677 | case SECCOMP_RET_KILL_THREAD: |
fd76875c | 678 | log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD; |
4d3b0b05 KC |
679 | break; |
680 | case SECCOMP_RET_KILL_PROCESS: | |
681 | default: | |
682 | log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS; | |
0ddec0fc TH |
683 | } |
684 | ||
685 | /* | |
326bee02 TH |
686 | * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the |
687 | * FILTER_FLAG_LOG bit was set. The admin has the ability to silence | |
688 | * any action from being logged by removing the action name from the | |
689 | * seccomp_actions_logged sysctl. | |
0ddec0fc | 690 | */ |
326bee02 TH |
691 | if (!log) |
692 | return; | |
0ddec0fc | 693 | |
326bee02 | 694 | audit_seccomp(syscall, signr, action); |
0ddec0fc TH |
695 | } |
696 | ||
1da177e4 LT |
697 | /* |
698 | * Secure computing mode 1 allows only read/write/exit/sigreturn. | |
699 | * To be fully secure this must be combined with rlimit | |
700 | * to limit the stack allocations too. | |
701 | */ | |
cb4253aa | 702 | static const int mode1_syscalls[] = { |
1da177e4 LT |
703 | __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, |
704 | 0, /* null terminated */ | |
705 | }; | |
706 | ||
a4412fc9 | 707 | static void __secure_computing_strict(int this_syscall) |
1da177e4 | 708 | { |
cb4253aa | 709 | const int *syscall_whitelist = mode1_syscalls; |
a4412fc9 | 710 | #ifdef CONFIG_COMPAT |
5c38065e | 711 | if (in_compat_syscall()) |
c983f0e8 | 712 | syscall_whitelist = get_compat_mode1_syscalls(); |
a4412fc9 AL |
713 | #endif |
714 | do { | |
715 | if (*syscall_whitelist == this_syscall) | |
716 | return; | |
717 | } while (*++syscall_whitelist); | |
718 | ||
719 | #ifdef SECCOMP_DEBUG | |
720 | dump_stack(); | |
721 | #endif | |
fd76875c | 722 | seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true); |
a4412fc9 AL |
723 | do_exit(SIGKILL); |
724 | } | |
725 | ||
726 | #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER | |
727 | void secure_computing_strict(int this_syscall) | |
728 | { | |
729 | int mode = current->seccomp.mode; | |
730 | ||
97f2645f | 731 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 TA |
732 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
733 | return; | |
734 | ||
221272f9 | 735 | if (mode == SECCOMP_MODE_DISABLED) |
a4412fc9 AL |
736 | return; |
737 | else if (mode == SECCOMP_MODE_STRICT) | |
738 | __secure_computing_strict(this_syscall); | |
739 | else | |
740 | BUG(); | |
741 | } | |
742 | #else | |
13aa72f0 AL |
743 | |
744 | #ifdef CONFIG_SECCOMP_FILTER | |
6a21cc50 TA |
745 | static u64 seccomp_next_notify_id(struct seccomp_filter *filter) |
746 | { | |
747 | /* | |
748 | * Note: overflow is ok here, the id just needs to be unique per | |
749 | * filter. | |
750 | */ | |
751 | lockdep_assert_held(&filter->notify_lock); | |
752 | return filter->notif->next_id++; | |
753 | } | |
754 | ||
fb3c5386 CB |
755 | static int seccomp_do_user_notification(int this_syscall, |
756 | struct seccomp_filter *match, | |
757 | const struct seccomp_data *sd) | |
6a21cc50 TA |
758 | { |
759 | int err; | |
fb3c5386 | 760 | u32 flags = 0; |
6a21cc50 TA |
761 | long ret = 0; |
762 | struct seccomp_knotif n = {}; | |
763 | ||
764 | mutex_lock(&match->notify_lock); | |
765 | err = -ENOSYS; | |
766 | if (!match->notif) | |
767 | goto out; | |
768 | ||
769 | n.task = current; | |
770 | n.state = SECCOMP_NOTIFY_INIT; | |
771 | n.data = sd; | |
772 | n.id = seccomp_next_notify_id(match); | |
773 | init_completion(&n.ready); | |
774 | list_add(&n.list, &match->notif->notifications); | |
775 | ||
776 | up(&match->notif->request); | |
777 | wake_up_poll(&match->notif->wqh, EPOLLIN | EPOLLRDNORM); | |
778 | mutex_unlock(&match->notify_lock); | |
779 | ||
780 | /* | |
781 | * This is where we wait for a reply from userspace. | |
782 | */ | |
783 | err = wait_for_completion_interruptible(&n.ready); | |
784 | mutex_lock(&match->notify_lock); | |
785 | if (err == 0) { | |
786 | ret = n.val; | |
787 | err = n.error; | |
fb3c5386 | 788 | flags = n.flags; |
6a21cc50 TA |
789 | } |
790 | ||
791 | /* | |
792 | * Note that it's possible the listener died in between the time when | |
793 | * we were notified of a respons (or a signal) and when we were able to | |
794 | * re-acquire the lock, so only delete from the list if the | |
795 | * notification actually exists. | |
796 | * | |
797 | * Also note that this test is only valid because there's no way to | |
798 | * *reattach* to a notifier right now. If one is added, we'll need to | |
799 | * keep track of the notif itself and make sure they match here. | |
800 | */ | |
801 | if (match->notif) | |
802 | list_del(&n.list); | |
803 | out: | |
804 | mutex_unlock(&match->notify_lock); | |
fb3c5386 CB |
805 | |
806 | /* Userspace requests to continue the syscall. */ | |
807 | if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) | |
808 | return 0; | |
809 | ||
6a21cc50 TA |
810 | syscall_set_return_value(current, task_pt_regs(current), |
811 | err, ret); | |
fb3c5386 | 812 | return -1; |
6a21cc50 TA |
813 | } |
814 | ||
ce6526e8 KC |
815 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
816 | const bool recheck_after_trace) | |
13aa72f0 AL |
817 | { |
818 | u32 filter_ret, action; | |
deb4de8b | 819 | struct seccomp_filter *match = NULL; |
13aa72f0 | 820 | int data; |
db511391 | 821 | struct seccomp_data sd_local; |
1da177e4 | 822 | |
3ba2530c KC |
823 | /* |
824 | * Make sure that any changes to mode from another thread have | |
825 | * been seen after TIF_SECCOMP was seen. | |
826 | */ | |
827 | rmb(); | |
828 | ||
db511391 TA |
829 | if (!sd) { |
830 | populate_seccomp_data(&sd_local); | |
831 | sd = &sd_local; | |
832 | } | |
833 | ||
deb4de8b | 834 | filter_ret = seccomp_run_filters(sd, &match); |
13aa72f0 | 835 | data = filter_ret & SECCOMP_RET_DATA; |
0466bdb9 | 836 | action = filter_ret & SECCOMP_RET_ACTION_FULL; |
13aa72f0 AL |
837 | |
838 | switch (action) { | |
839 | case SECCOMP_RET_ERRNO: | |
580c57f1 KC |
840 | /* Set low-order bits as an errno, capped at MAX_ERRNO. */ |
841 | if (data > MAX_ERRNO) | |
842 | data = MAX_ERRNO; | |
d39bd00d | 843 | syscall_set_return_value(current, task_pt_regs(current), |
13aa72f0 AL |
844 | -data, 0); |
845 | goto skip; | |
846 | ||
847 | case SECCOMP_RET_TRAP: | |
848 | /* Show the handler the original registers. */ | |
d39bd00d | 849 | syscall_rollback(current, task_pt_regs(current)); |
13aa72f0 AL |
850 | /* Let the filter pass back 16 bits of data. */ |
851 | seccomp_send_sigsys(this_syscall, data); | |
852 | goto skip; | |
853 | ||
854 | case SECCOMP_RET_TRACE: | |
ce6526e8 KC |
855 | /* We've been put in this state by the ptracer already. */ |
856 | if (recheck_after_trace) | |
857 | return 0; | |
858 | ||
8112c4f1 KC |
859 | /* ENOSYS these calls if there is no tracer attached. */ |
860 | if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { | |
861 | syscall_set_return_value(current, | |
862 | task_pt_regs(current), | |
863 | -ENOSYS, 0); | |
864 | goto skip; | |
865 | } | |
866 | ||
867 | /* Allow the BPF to provide the event message */ | |
868 | ptrace_event(PTRACE_EVENT_SECCOMP, data); | |
869 | /* | |
870 | * The delivery of a fatal signal during event | |
485a252a KC |
871 | * notification may silently skip tracer notification, |
872 | * which could leave us with a potentially unmodified | |
873 | * syscall that the tracer would have liked to have | |
874 | * changed. Since the process is about to die, we just | |
875 | * force the syscall to be skipped and let the signal | |
876 | * kill the process and correctly handle any tracer exit | |
877 | * notifications. | |
8112c4f1 KC |
878 | */ |
879 | if (fatal_signal_pending(current)) | |
485a252a | 880 | goto skip; |
8112c4f1 KC |
881 | /* Check if the tracer forced the syscall to be skipped. */ |
882 | this_syscall = syscall_get_nr(current, task_pt_regs(current)); | |
883 | if (this_syscall < 0) | |
884 | goto skip; | |
885 | ||
ce6526e8 KC |
886 | /* |
887 | * Recheck the syscall, since it may have changed. This | |
888 | * intentionally uses a NULL struct seccomp_data to force | |
889 | * a reload of all registers. This does not goto skip since | |
890 | * a skip would have already been reported. | |
891 | */ | |
892 | if (__seccomp_filter(this_syscall, NULL, true)) | |
893 | return -1; | |
894 | ||
8112c4f1 | 895 | return 0; |
13aa72f0 | 896 | |
6a21cc50 | 897 | case SECCOMP_RET_USER_NOTIF: |
fb3c5386 CB |
898 | if (seccomp_do_user_notification(this_syscall, match, sd)) |
899 | goto skip; | |
900 | ||
901 | return 0; | |
6a21cc50 | 902 | |
59f5cf44 TH |
903 | case SECCOMP_RET_LOG: |
904 | seccomp_log(this_syscall, 0, action, true); | |
905 | return 0; | |
906 | ||
13aa72f0 | 907 | case SECCOMP_RET_ALLOW: |
deb4de8b KC |
908 | /* |
909 | * Note that the "match" filter will always be NULL for | |
910 | * this action since SECCOMP_RET_ALLOW is the starting | |
911 | * state in seccomp_run_filters(). | |
912 | */ | |
8112c4f1 | 913 | return 0; |
13aa72f0 | 914 | |
fd76875c | 915 | case SECCOMP_RET_KILL_THREAD: |
4d3b0b05 | 916 | case SECCOMP_RET_KILL_PROCESS: |
131b6351 | 917 | default: |
e66a3997 | 918 | seccomp_log(this_syscall, SIGSYS, action, true); |
d7276e32 | 919 | /* Dump core only if this is the last remaining thread. */ |
4d3b0b05 KC |
920 | if (action == SECCOMP_RET_KILL_PROCESS || |
921 | get_nr_threads(current) == 1) { | |
ae7795bc | 922 | kernel_siginfo_t info; |
131b6351 | 923 | |
d7276e32 KC |
924 | /* Show the original registers in the dump. */ |
925 | syscall_rollback(current, task_pt_regs(current)); | |
926 | /* Trigger a manual coredump since do_exit skips it. */ | |
927 | seccomp_init_siginfo(&info, this_syscall, data); | |
928 | do_coredump(&info); | |
929 | } | |
4d3b0b05 KC |
930 | if (action == SECCOMP_RET_KILL_PROCESS) |
931 | do_group_exit(SIGSYS); | |
932 | else | |
933 | do_exit(SIGSYS); | |
13aa72f0 AL |
934 | } |
935 | ||
936 | unreachable(); | |
937 | ||
938 | skip: | |
e66a3997 | 939 | seccomp_log(this_syscall, 0, action, match ? match->log : false); |
8112c4f1 KC |
940 | return -1; |
941 | } | |
942 | #else | |
ce6526e8 KC |
943 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
944 | const bool recheck_after_trace) | |
8112c4f1 KC |
945 | { |
946 | BUG(); | |
13aa72f0 | 947 | } |
1da177e4 | 948 | #endif |
13aa72f0 | 949 | |
8112c4f1 | 950 | int __secure_computing(const struct seccomp_data *sd) |
13aa72f0 AL |
951 | { |
952 | int mode = current->seccomp.mode; | |
8112c4f1 | 953 | int this_syscall; |
13aa72f0 | 954 | |
97f2645f | 955 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 | 956 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
8112c4f1 KC |
957 | return 0; |
958 | ||
959 | this_syscall = sd ? sd->nr : | |
960 | syscall_get_nr(current, task_pt_regs(current)); | |
13c4a901 | 961 | |
13aa72f0 | 962 | switch (mode) { |
e2cfabdf | 963 | case SECCOMP_MODE_STRICT: |
13aa72f0 | 964 | __secure_computing_strict(this_syscall); /* may call do_exit */ |
8112c4f1 | 965 | return 0; |
13aa72f0 | 966 | case SECCOMP_MODE_FILTER: |
ce6526e8 | 967 | return __seccomp_filter(this_syscall, sd, false); |
1da177e4 LT |
968 | default: |
969 | BUG(); | |
970 | } | |
13aa72f0 | 971 | } |
a4412fc9 | 972 | #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ |
1d9d02fe AA |
973 | |
974 | long prctl_get_seccomp(void) | |
975 | { | |
976 | return current->seccomp.mode; | |
977 | } | |
978 | ||
e2cfabdf | 979 | /** |
3b23dd12 | 980 | * seccomp_set_mode_strict: internal function for setting strict seccomp |
e2cfabdf WD |
981 | * |
982 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
983 | * | |
984 | * Returns 0 on success or -EINVAL on failure. | |
985 | */ | |
3b23dd12 | 986 | static long seccomp_set_mode_strict(void) |
1d9d02fe | 987 | { |
3b23dd12 | 988 | const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; |
e2cfabdf | 989 | long ret = -EINVAL; |
1d9d02fe | 990 | |
dbd95212 KC |
991 | spin_lock_irq(¤t->sighand->siglock); |
992 | ||
1f41b450 | 993 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1d9d02fe AA |
994 | goto out; |
995 | ||
cf99abac | 996 | #ifdef TIF_NOTSC |
3b23dd12 | 997 | disable_TSC(); |
cf99abac | 998 | #endif |
00a02d0c | 999 | seccomp_assign_mode(current, seccomp_mode, 0); |
3b23dd12 KC |
1000 | ret = 0; |
1001 | ||
1002 | out: | |
dbd95212 | 1003 | spin_unlock_irq(¤t->sighand->siglock); |
3b23dd12 KC |
1004 | |
1005 | return ret; | |
1006 | } | |
1007 | ||
e2cfabdf | 1008 | #ifdef CONFIG_SECCOMP_FILTER |
6a21cc50 TA |
1009 | static int seccomp_notify_release(struct inode *inode, struct file *file) |
1010 | { | |
1011 | struct seccomp_filter *filter = file->private_data; | |
1012 | struct seccomp_knotif *knotif; | |
1013 | ||
a811dc61 TA |
1014 | if (!filter) |
1015 | return 0; | |
1016 | ||
6a21cc50 TA |
1017 | mutex_lock(&filter->notify_lock); |
1018 | ||
1019 | /* | |
1020 | * If this file is being closed because e.g. the task who owned it | |
1021 | * died, let's wake everyone up who was waiting on us. | |
1022 | */ | |
1023 | list_for_each_entry(knotif, &filter->notif->notifications, list) { | |
1024 | if (knotif->state == SECCOMP_NOTIFY_REPLIED) | |
1025 | continue; | |
1026 | ||
1027 | knotif->state = SECCOMP_NOTIFY_REPLIED; | |
1028 | knotif->error = -ENOSYS; | |
1029 | knotif->val = 0; | |
1030 | ||
1031 | complete(&knotif->ready); | |
1032 | } | |
1033 | ||
1034 | kfree(filter->notif); | |
1035 | filter->notif = NULL; | |
1036 | mutex_unlock(&filter->notify_lock); | |
1037 | __put_seccomp_filter(filter); | |
1038 | return 0; | |
1039 | } | |
1040 | ||
9f87dcf1 SD |
1041 | /* must be called with notif_lock held */ |
1042 | static inline struct seccomp_knotif * | |
1043 | find_notification(struct seccomp_filter *filter, u64 id) | |
1044 | { | |
1045 | struct seccomp_knotif *cur; | |
1046 | ||
1047 | lockdep_assert_held(&filter->notify_lock); | |
1048 | ||
1049 | list_for_each_entry(cur, &filter->notif->notifications, list) { | |
1050 | if (cur->id == id) | |
1051 | return cur; | |
1052 | } | |
1053 | ||
1054 | return NULL; | |
1055 | } | |
1056 | ||
1057 | ||
6a21cc50 TA |
1058 | static long seccomp_notify_recv(struct seccomp_filter *filter, |
1059 | void __user *buf) | |
1060 | { | |
1061 | struct seccomp_knotif *knotif = NULL, *cur; | |
1062 | struct seccomp_notif unotif; | |
1063 | ssize_t ret; | |
1064 | ||
2882d53c SD |
1065 | /* Verify that we're not given garbage to keep struct extensible. */ |
1066 | ret = check_zeroed_user(buf, sizeof(unotif)); | |
1067 | if (ret < 0) | |
1068 | return ret; | |
1069 | if (!ret) | |
1070 | return -EINVAL; | |
1071 | ||
6a21cc50 TA |
1072 | memset(&unotif, 0, sizeof(unotif)); |
1073 | ||
1074 | ret = down_interruptible(&filter->notif->request); | |
1075 | if (ret < 0) | |
1076 | return ret; | |
1077 | ||
1078 | mutex_lock(&filter->notify_lock); | |
1079 | list_for_each_entry(cur, &filter->notif->notifications, list) { | |
1080 | if (cur->state == SECCOMP_NOTIFY_INIT) { | |
1081 | knotif = cur; | |
1082 | break; | |
1083 | } | |
1084 | } | |
1085 | ||
1086 | /* | |
1087 | * If we didn't find a notification, it could be that the task was | |
1088 | * interrupted by a fatal signal between the time we were woken and | |
1089 | * when we were able to acquire the rw lock. | |
1090 | */ | |
1091 | if (!knotif) { | |
1092 | ret = -ENOENT; | |
1093 | goto out; | |
1094 | } | |
1095 | ||
1096 | unotif.id = knotif->id; | |
1097 | unotif.pid = task_pid_vnr(knotif->task); | |
1098 | unotif.data = *(knotif->data); | |
1099 | ||
1100 | knotif->state = SECCOMP_NOTIFY_SENT; | |
1101 | wake_up_poll(&filter->notif->wqh, EPOLLOUT | EPOLLWRNORM); | |
1102 | ret = 0; | |
1103 | out: | |
1104 | mutex_unlock(&filter->notify_lock); | |
1105 | ||
1106 | if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) { | |
1107 | ret = -EFAULT; | |
1108 | ||
1109 | /* | |
1110 | * Userspace screwed up. To make sure that we keep this | |
1111 | * notification alive, let's reset it back to INIT. It | |
1112 | * may have died when we released the lock, so we need to make | |
1113 | * sure it's still around. | |
1114 | */ | |
6a21cc50 | 1115 | mutex_lock(&filter->notify_lock); |
9f87dcf1 | 1116 | knotif = find_notification(filter, unotif.id); |
6a21cc50 TA |
1117 | if (knotif) { |
1118 | knotif->state = SECCOMP_NOTIFY_INIT; | |
1119 | up(&filter->notif->request); | |
1120 | } | |
1121 | mutex_unlock(&filter->notify_lock); | |
1122 | } | |
1123 | ||
1124 | return ret; | |
1125 | } | |
1126 | ||
1127 | static long seccomp_notify_send(struct seccomp_filter *filter, | |
1128 | void __user *buf) | |
1129 | { | |
1130 | struct seccomp_notif_resp resp = {}; | |
9f87dcf1 | 1131 | struct seccomp_knotif *knotif; |
6a21cc50 TA |
1132 | long ret; |
1133 | ||
1134 | if (copy_from_user(&resp, buf, sizeof(resp))) | |
1135 | return -EFAULT; | |
1136 | ||
fb3c5386 CB |
1137 | if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE) |
1138 | return -EINVAL; | |
1139 | ||
1140 | if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) && | |
1141 | (resp.error || resp.val)) | |
6a21cc50 TA |
1142 | return -EINVAL; |
1143 | ||
1144 | ret = mutex_lock_interruptible(&filter->notify_lock); | |
1145 | if (ret < 0) | |
1146 | return ret; | |
1147 | ||
9f87dcf1 | 1148 | knotif = find_notification(filter, resp.id); |
6a21cc50 TA |
1149 | if (!knotif) { |
1150 | ret = -ENOENT; | |
1151 | goto out; | |
1152 | } | |
1153 | ||
1154 | /* Allow exactly one reply. */ | |
1155 | if (knotif->state != SECCOMP_NOTIFY_SENT) { | |
1156 | ret = -EINPROGRESS; | |
1157 | goto out; | |
1158 | } | |
1159 | ||
1160 | ret = 0; | |
1161 | knotif->state = SECCOMP_NOTIFY_REPLIED; | |
1162 | knotif->error = resp.error; | |
1163 | knotif->val = resp.val; | |
fb3c5386 | 1164 | knotif->flags = resp.flags; |
6a21cc50 TA |
1165 | complete(&knotif->ready); |
1166 | out: | |
1167 | mutex_unlock(&filter->notify_lock); | |
1168 | return ret; | |
1169 | } | |
1170 | ||
1171 | static long seccomp_notify_id_valid(struct seccomp_filter *filter, | |
1172 | void __user *buf) | |
1173 | { | |
9f87dcf1 | 1174 | struct seccomp_knotif *knotif; |
6a21cc50 TA |
1175 | u64 id; |
1176 | long ret; | |
1177 | ||
1178 | if (copy_from_user(&id, buf, sizeof(id))) | |
1179 | return -EFAULT; | |
1180 | ||
1181 | ret = mutex_lock_interruptible(&filter->notify_lock); | |
1182 | if (ret < 0) | |
1183 | return ret; | |
1184 | ||
9f87dcf1 SD |
1185 | knotif = find_notification(filter, id); |
1186 | if (knotif && knotif->state == SECCOMP_NOTIFY_SENT) | |
1187 | ret = 0; | |
1188 | else | |
1189 | ret = -ENOENT; | |
6a21cc50 | 1190 | |
6a21cc50 TA |
1191 | mutex_unlock(&filter->notify_lock); |
1192 | return ret; | |
1193 | } | |
1194 | ||
1195 | static long seccomp_notify_ioctl(struct file *file, unsigned int cmd, | |
1196 | unsigned long arg) | |
1197 | { | |
1198 | struct seccomp_filter *filter = file->private_data; | |
1199 | void __user *buf = (void __user *)arg; | |
1200 | ||
1201 | switch (cmd) { | |
1202 | case SECCOMP_IOCTL_NOTIF_RECV: | |
1203 | return seccomp_notify_recv(filter, buf); | |
1204 | case SECCOMP_IOCTL_NOTIF_SEND: | |
1205 | return seccomp_notify_send(filter, buf); | |
1206 | case SECCOMP_IOCTL_NOTIF_ID_VALID: | |
1207 | return seccomp_notify_id_valid(filter, buf); | |
1208 | default: | |
1209 | return -EINVAL; | |
1210 | } | |
1211 | } | |
1212 | ||
1213 | static __poll_t seccomp_notify_poll(struct file *file, | |
1214 | struct poll_table_struct *poll_tab) | |
1215 | { | |
1216 | struct seccomp_filter *filter = file->private_data; | |
1217 | __poll_t ret = 0; | |
1218 | struct seccomp_knotif *cur; | |
1219 | ||
1220 | poll_wait(file, &filter->notif->wqh, poll_tab); | |
1221 | ||
319deec7 | 1222 | if (mutex_lock_interruptible(&filter->notify_lock) < 0) |
6a21cc50 TA |
1223 | return EPOLLERR; |
1224 | ||
1225 | list_for_each_entry(cur, &filter->notif->notifications, list) { | |
1226 | if (cur->state == SECCOMP_NOTIFY_INIT) | |
1227 | ret |= EPOLLIN | EPOLLRDNORM; | |
1228 | if (cur->state == SECCOMP_NOTIFY_SENT) | |
1229 | ret |= EPOLLOUT | EPOLLWRNORM; | |
1230 | if ((ret & EPOLLIN) && (ret & EPOLLOUT)) | |
1231 | break; | |
1232 | } | |
1233 | ||
1234 | mutex_unlock(&filter->notify_lock); | |
1235 | ||
1236 | return ret; | |
1237 | } | |
1238 | ||
1239 | static const struct file_operations seccomp_notify_ops = { | |
1240 | .poll = seccomp_notify_poll, | |
1241 | .release = seccomp_notify_release, | |
1242 | .unlocked_ioctl = seccomp_notify_ioctl, | |
3db81afd | 1243 | .compat_ioctl = seccomp_notify_ioctl, |
6a21cc50 TA |
1244 | }; |
1245 | ||
1246 | static struct file *init_listener(struct seccomp_filter *filter) | |
1247 | { | |
1248 | struct file *ret = ERR_PTR(-EBUSY); | |
1249 | struct seccomp_filter *cur; | |
1250 | ||
1251 | for (cur = current->seccomp.filter; cur; cur = cur->prev) { | |
1252 | if (cur->notif) | |
1253 | goto out; | |
1254 | } | |
1255 | ||
1256 | ret = ERR_PTR(-ENOMEM); | |
1257 | filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL); | |
1258 | if (!filter->notif) | |
1259 | goto out; | |
1260 | ||
1261 | sema_init(&filter->notif->request, 0); | |
1262 | filter->notif->next_id = get_random_u64(); | |
1263 | INIT_LIST_HEAD(&filter->notif->notifications); | |
1264 | init_waitqueue_head(&filter->notif->wqh); | |
1265 | ||
1266 | ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops, | |
1267 | filter, O_RDWR); | |
1268 | if (IS_ERR(ret)) | |
1269 | goto out_notif; | |
1270 | ||
1271 | /* The file has a reference to it now */ | |
1272 | __get_seccomp_filter(filter); | |
1273 | ||
1274 | out_notif: | |
1275 | if (IS_ERR(ret)) | |
1276 | kfree(filter->notif); | |
1277 | out: | |
1278 | return ret; | |
1279 | } | |
1280 | ||
3b23dd12 KC |
1281 | /** |
1282 | * seccomp_set_mode_filter: internal function for setting seccomp filter | |
48dc92b9 | 1283 | * @flags: flags to change filter behavior |
3b23dd12 KC |
1284 | * @filter: struct sock_fprog containing filter |
1285 | * | |
1286 | * This function may be called repeatedly to install additional filters. | |
1287 | * Every filter successfully installed will be evaluated (in reverse order) | |
1288 | * for each system call the task makes. | |
1289 | * | |
1290 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
1291 | * | |
1292 | * Returns 0 on success or -EINVAL on failure. | |
1293 | */ | |
48dc92b9 KC |
1294 | static long seccomp_set_mode_filter(unsigned int flags, |
1295 | const char __user *filter) | |
3b23dd12 KC |
1296 | { |
1297 | const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; | |
c8bee430 | 1298 | struct seccomp_filter *prepared = NULL; |
3b23dd12 | 1299 | long ret = -EINVAL; |
6a21cc50 TA |
1300 | int listener = -1; |
1301 | struct file *listener_f = NULL; | |
3b23dd12 | 1302 | |
48dc92b9 | 1303 | /* Validate flags. */ |
c2e1f2e3 | 1304 | if (flags & ~SECCOMP_FILTER_FLAG_MASK) |
dbd95212 | 1305 | return -EINVAL; |
48dc92b9 | 1306 | |
7a0df7fb TA |
1307 | /* |
1308 | * In the successful case, NEW_LISTENER returns the new listener fd. | |
1309 | * But in the failure case, TSYNC returns the thread that died. If you | |
1310 | * combine these two flags, there's no way to tell whether something | |
51891498 TA |
1311 | * succeeded or failed. So, let's disallow this combination if the user |
1312 | * has not explicitly requested no errors from TSYNC. | |
7a0df7fb TA |
1313 | */ |
1314 | if ((flags & SECCOMP_FILTER_FLAG_TSYNC) && | |
51891498 TA |
1315 | (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) && |
1316 | ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0)) | |
7a0df7fb TA |
1317 | return -EINVAL; |
1318 | ||
c8bee430 KC |
1319 | /* Prepare the new filter before holding any locks. */ |
1320 | prepared = seccomp_prepare_user_filter(filter); | |
1321 | if (IS_ERR(prepared)) | |
1322 | return PTR_ERR(prepared); | |
1323 | ||
6a21cc50 TA |
1324 | if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) { |
1325 | listener = get_unused_fd_flags(O_CLOEXEC); | |
1326 | if (listener < 0) { | |
1327 | ret = listener; | |
1328 | goto out_free; | |
1329 | } | |
1330 | ||
1331 | listener_f = init_listener(prepared); | |
1332 | if (IS_ERR(listener_f)) { | |
1333 | put_unused_fd(listener); | |
1334 | ret = PTR_ERR(listener_f); | |
1335 | goto out_free; | |
1336 | } | |
1337 | } | |
1338 | ||
c2e1f2e3 KC |
1339 | /* |
1340 | * Make sure we cannot change seccomp or nnp state via TSYNC | |
1341 | * while another thread is in the middle of calling exec. | |
1342 | */ | |
1343 | if (flags & SECCOMP_FILTER_FLAG_TSYNC && | |
1344 | mutex_lock_killable(¤t->signal->cred_guard_mutex)) | |
6a21cc50 | 1345 | goto out_put_fd; |
c2e1f2e3 | 1346 | |
dbd95212 KC |
1347 | spin_lock_irq(¤t->sighand->siglock); |
1348 | ||
3b23dd12 KC |
1349 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1350 | goto out; | |
1351 | ||
c8bee430 | 1352 | ret = seccomp_attach_filter(flags, prepared); |
3b23dd12 | 1353 | if (ret) |
e2cfabdf | 1354 | goto out; |
c8bee430 KC |
1355 | /* Do not free the successfully attached filter. */ |
1356 | prepared = NULL; | |
1d9d02fe | 1357 | |
00a02d0c | 1358 | seccomp_assign_mode(current, seccomp_mode, flags); |
e2cfabdf | 1359 | out: |
dbd95212 | 1360 | spin_unlock_irq(¤t->sighand->siglock); |
c2e1f2e3 KC |
1361 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
1362 | mutex_unlock(¤t->signal->cred_guard_mutex); | |
6a21cc50 TA |
1363 | out_put_fd: |
1364 | if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) { | |
7a0df7fb | 1365 | if (ret) { |
a811dc61 | 1366 | listener_f->private_data = NULL; |
6a21cc50 TA |
1367 | fput(listener_f); |
1368 | put_unused_fd(listener); | |
1369 | } else { | |
1370 | fd_install(listener, listener_f); | |
1371 | ret = listener; | |
1372 | } | |
1373 | } | |
c2e1f2e3 | 1374 | out_free: |
c8bee430 | 1375 | seccomp_filter_free(prepared); |
1d9d02fe AA |
1376 | return ret; |
1377 | } | |
3b23dd12 | 1378 | #else |
48dc92b9 KC |
1379 | static inline long seccomp_set_mode_filter(unsigned int flags, |
1380 | const char __user *filter) | |
3b23dd12 KC |
1381 | { |
1382 | return -EINVAL; | |
1383 | } | |
1384 | #endif | |
d78ab02c | 1385 | |
d612b1fd TH |
1386 | static long seccomp_get_action_avail(const char __user *uaction) |
1387 | { | |
1388 | u32 action; | |
1389 | ||
1390 | if (copy_from_user(&action, uaction, sizeof(action))) | |
1391 | return -EFAULT; | |
1392 | ||
1393 | switch (action) { | |
0466bdb9 | 1394 | case SECCOMP_RET_KILL_PROCESS: |
fd76875c | 1395 | case SECCOMP_RET_KILL_THREAD: |
d612b1fd TH |
1396 | case SECCOMP_RET_TRAP: |
1397 | case SECCOMP_RET_ERRNO: | |
6a21cc50 | 1398 | case SECCOMP_RET_USER_NOTIF: |
d612b1fd | 1399 | case SECCOMP_RET_TRACE: |
59f5cf44 | 1400 | case SECCOMP_RET_LOG: |
d612b1fd TH |
1401 | case SECCOMP_RET_ALLOW: |
1402 | break; | |
1403 | default: | |
1404 | return -EOPNOTSUPP; | |
1405 | } | |
1406 | ||
1407 | return 0; | |
1408 | } | |
1409 | ||
6a21cc50 TA |
1410 | static long seccomp_get_notif_sizes(void __user *usizes) |
1411 | { | |
1412 | struct seccomp_notif_sizes sizes = { | |
1413 | .seccomp_notif = sizeof(struct seccomp_notif), | |
1414 | .seccomp_notif_resp = sizeof(struct seccomp_notif_resp), | |
1415 | .seccomp_data = sizeof(struct seccomp_data), | |
1416 | }; | |
1417 | ||
1418 | if (copy_to_user(usizes, &sizes, sizeof(sizes))) | |
1419 | return -EFAULT; | |
1420 | ||
1421 | return 0; | |
1422 | } | |
1423 | ||
48dc92b9 KC |
1424 | /* Common entry point for both prctl and syscall. */ |
1425 | static long do_seccomp(unsigned int op, unsigned int flags, | |
a5662e4d | 1426 | void __user *uargs) |
48dc92b9 KC |
1427 | { |
1428 | switch (op) { | |
1429 | case SECCOMP_SET_MODE_STRICT: | |
1430 | if (flags != 0 || uargs != NULL) | |
1431 | return -EINVAL; | |
1432 | return seccomp_set_mode_strict(); | |
1433 | case SECCOMP_SET_MODE_FILTER: | |
1434 | return seccomp_set_mode_filter(flags, uargs); | |
d612b1fd TH |
1435 | case SECCOMP_GET_ACTION_AVAIL: |
1436 | if (flags != 0) | |
1437 | return -EINVAL; | |
1438 | ||
1439 | return seccomp_get_action_avail(uargs); | |
6a21cc50 TA |
1440 | case SECCOMP_GET_NOTIF_SIZES: |
1441 | if (flags != 0) | |
1442 | return -EINVAL; | |
1443 | ||
1444 | return seccomp_get_notif_sizes(uargs); | |
48dc92b9 KC |
1445 | default: |
1446 | return -EINVAL; | |
1447 | } | |
1448 | } | |
1449 | ||
1450 | SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, | |
a5662e4d | 1451 | void __user *, uargs) |
48dc92b9 KC |
1452 | { |
1453 | return do_seccomp(op, flags, uargs); | |
1454 | } | |
1455 | ||
d78ab02c KC |
1456 | /** |
1457 | * prctl_set_seccomp: configures current->seccomp.mode | |
1458 | * @seccomp_mode: requested mode to use | |
1459 | * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER | |
1460 | * | |
1461 | * Returns 0 on success or -EINVAL on failure. | |
1462 | */ | |
a5662e4d | 1463 | long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter) |
d78ab02c | 1464 | { |
48dc92b9 | 1465 | unsigned int op; |
a5662e4d | 1466 | void __user *uargs; |
48dc92b9 | 1467 | |
3b23dd12 KC |
1468 | switch (seccomp_mode) { |
1469 | case SECCOMP_MODE_STRICT: | |
48dc92b9 KC |
1470 | op = SECCOMP_SET_MODE_STRICT; |
1471 | /* | |
1472 | * Setting strict mode through prctl always ignored filter, | |
1473 | * so make sure it is always NULL here to pass the internal | |
1474 | * check in do_seccomp(). | |
1475 | */ | |
1476 | uargs = NULL; | |
1477 | break; | |
3b23dd12 | 1478 | case SECCOMP_MODE_FILTER: |
48dc92b9 KC |
1479 | op = SECCOMP_SET_MODE_FILTER; |
1480 | uargs = filter; | |
1481 | break; | |
3b23dd12 KC |
1482 | default: |
1483 | return -EINVAL; | |
1484 | } | |
48dc92b9 KC |
1485 | |
1486 | /* prctl interface doesn't have flags, so they are always zero. */ | |
1487 | return do_seccomp(op, 0, uargs); | |
d78ab02c | 1488 | } |
f8e529ed TA |
1489 | |
1490 | #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE) | |
f06eae83 TA |
1491 | static struct seccomp_filter *get_nth_filter(struct task_struct *task, |
1492 | unsigned long filter_off) | |
f8e529ed | 1493 | { |
f06eae83 TA |
1494 | struct seccomp_filter *orig, *filter; |
1495 | unsigned long count; | |
f8e529ed | 1496 | |
f06eae83 TA |
1497 | /* |
1498 | * Note: this is only correct because the caller should be the (ptrace) | |
1499 | * tracer of the task, otherwise lock_task_sighand is needed. | |
1500 | */ | |
f8e529ed | 1501 | spin_lock_irq(&task->sighand->siglock); |
f06eae83 | 1502 | |
f8e529ed | 1503 | if (task->seccomp.mode != SECCOMP_MODE_FILTER) { |
f06eae83 TA |
1504 | spin_unlock_irq(&task->sighand->siglock); |
1505 | return ERR_PTR(-EINVAL); | |
f8e529ed TA |
1506 | } |
1507 | ||
f06eae83 TA |
1508 | orig = task->seccomp.filter; |
1509 | __get_seccomp_filter(orig); | |
1510 | spin_unlock_irq(&task->sighand->siglock); | |
1511 | ||
1512 | count = 0; | |
1513 | for (filter = orig; filter; filter = filter->prev) | |
f8e529ed | 1514 | count++; |
f8e529ed TA |
1515 | |
1516 | if (filter_off >= count) { | |
f06eae83 | 1517 | filter = ERR_PTR(-ENOENT); |
f8e529ed TA |
1518 | goto out; |
1519 | } | |
f8e529ed | 1520 | |
f06eae83 TA |
1521 | count -= filter_off; |
1522 | for (filter = orig; filter && count > 1; filter = filter->prev) | |
f8e529ed | 1523 | count--; |
f8e529ed TA |
1524 | |
1525 | if (WARN_ON(count != 1 || !filter)) { | |
f06eae83 | 1526 | filter = ERR_PTR(-ENOENT); |
f8e529ed TA |
1527 | goto out; |
1528 | } | |
1529 | ||
f06eae83 TA |
1530 | __get_seccomp_filter(filter); |
1531 | ||
1532 | out: | |
1533 | __put_seccomp_filter(orig); | |
1534 | return filter; | |
1535 | } | |
1536 | ||
1537 | long seccomp_get_filter(struct task_struct *task, unsigned long filter_off, | |
1538 | void __user *data) | |
1539 | { | |
1540 | struct seccomp_filter *filter; | |
1541 | struct sock_fprog_kern *fprog; | |
1542 | long ret; | |
1543 | ||
1544 | if (!capable(CAP_SYS_ADMIN) || | |
1545 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { | |
1546 | return -EACCES; | |
1547 | } | |
1548 | ||
1549 | filter = get_nth_filter(task, filter_off); | |
1550 | if (IS_ERR(filter)) | |
1551 | return PTR_ERR(filter); | |
1552 | ||
f8e529ed TA |
1553 | fprog = filter->prog->orig_prog; |
1554 | if (!fprog) { | |
470bf1f2 | 1555 | /* This must be a new non-cBPF filter, since we save |
f8e529ed TA |
1556 | * every cBPF filter's orig_prog above when |
1557 | * CONFIG_CHECKPOINT_RESTORE is enabled. | |
1558 | */ | |
1559 | ret = -EMEDIUMTYPE; | |
1560 | goto out; | |
1561 | } | |
1562 | ||
1563 | ret = fprog->len; | |
1564 | if (!data) | |
1565 | goto out; | |
1566 | ||
f8e529ed TA |
1567 | if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog))) |
1568 | ret = -EFAULT; | |
1569 | ||
f8e529ed | 1570 | out: |
66a733ea | 1571 | __put_seccomp_filter(filter); |
f8e529ed | 1572 | return ret; |
f8e529ed | 1573 | } |
f8e529ed | 1574 | |
26500475 TA |
1575 | long seccomp_get_metadata(struct task_struct *task, |
1576 | unsigned long size, void __user *data) | |
1577 | { | |
1578 | long ret; | |
1579 | struct seccomp_filter *filter; | |
1580 | struct seccomp_metadata kmd = {}; | |
1581 | ||
1582 | if (!capable(CAP_SYS_ADMIN) || | |
1583 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { | |
1584 | return -EACCES; | |
1585 | } | |
1586 | ||
1587 | size = min_t(unsigned long, size, sizeof(kmd)); | |
1588 | ||
63bb0045 TA |
1589 | if (size < sizeof(kmd.filter_off)) |
1590 | return -EINVAL; | |
1591 | ||
1592 | if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off))) | |
26500475 TA |
1593 | return -EFAULT; |
1594 | ||
1595 | filter = get_nth_filter(task, kmd.filter_off); | |
1596 | if (IS_ERR(filter)) | |
1597 | return PTR_ERR(filter); | |
1598 | ||
26500475 TA |
1599 | if (filter->log) |
1600 | kmd.flags |= SECCOMP_FILTER_FLAG_LOG; | |
1601 | ||
1602 | ret = size; | |
1603 | if (copy_to_user(data, &kmd, size)) | |
1604 | ret = -EFAULT; | |
1605 | ||
1606 | __put_seccomp_filter(filter); | |
f8e529ed TA |
1607 | return ret; |
1608 | } | |
1609 | #endif | |
8e5f1ad1 TH |
1610 | |
1611 | #ifdef CONFIG_SYSCTL | |
1612 | ||
1613 | /* Human readable action names for friendly sysctl interaction */ | |
0466bdb9 | 1614 | #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process" |
fd76875c | 1615 | #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread" |
8e5f1ad1 TH |
1616 | #define SECCOMP_RET_TRAP_NAME "trap" |
1617 | #define SECCOMP_RET_ERRNO_NAME "errno" | |
6a21cc50 | 1618 | #define SECCOMP_RET_USER_NOTIF_NAME "user_notif" |
8e5f1ad1 | 1619 | #define SECCOMP_RET_TRACE_NAME "trace" |
59f5cf44 | 1620 | #define SECCOMP_RET_LOG_NAME "log" |
8e5f1ad1 TH |
1621 | #define SECCOMP_RET_ALLOW_NAME "allow" |
1622 | ||
fd76875c | 1623 | static const char seccomp_actions_avail[] = |
0466bdb9 | 1624 | SECCOMP_RET_KILL_PROCESS_NAME " " |
fd76875c KC |
1625 | SECCOMP_RET_KILL_THREAD_NAME " " |
1626 | SECCOMP_RET_TRAP_NAME " " | |
1627 | SECCOMP_RET_ERRNO_NAME " " | |
6a21cc50 | 1628 | SECCOMP_RET_USER_NOTIF_NAME " " |
fd76875c KC |
1629 | SECCOMP_RET_TRACE_NAME " " |
1630 | SECCOMP_RET_LOG_NAME " " | |
1631 | SECCOMP_RET_ALLOW_NAME; | |
8e5f1ad1 | 1632 | |
0ddec0fc TH |
1633 | struct seccomp_log_name { |
1634 | u32 log; | |
1635 | const char *name; | |
1636 | }; | |
1637 | ||
1638 | static const struct seccomp_log_name seccomp_log_names[] = { | |
0466bdb9 | 1639 | { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME }, |
fd76875c | 1640 | { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME }, |
0ddec0fc TH |
1641 | { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME }, |
1642 | { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME }, | |
6a21cc50 | 1643 | { SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME }, |
0ddec0fc | 1644 | { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME }, |
59f5cf44 | 1645 | { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME }, |
0ddec0fc TH |
1646 | { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME }, |
1647 | { } | |
1648 | }; | |
1649 | ||
1650 | static bool seccomp_names_from_actions_logged(char *names, size_t size, | |
beb44aca TH |
1651 | u32 actions_logged, |
1652 | const char *sep) | |
0ddec0fc TH |
1653 | { |
1654 | const struct seccomp_log_name *cur; | |
beb44aca | 1655 | bool append_sep = false; |
0ddec0fc TH |
1656 | |
1657 | for (cur = seccomp_log_names; cur->name && size; cur++) { | |
1658 | ssize_t ret; | |
1659 | ||
1660 | if (!(actions_logged & cur->log)) | |
1661 | continue; | |
1662 | ||
beb44aca TH |
1663 | if (append_sep) { |
1664 | ret = strscpy(names, sep, size); | |
0ddec0fc TH |
1665 | if (ret < 0) |
1666 | return false; | |
1667 | ||
1668 | names += ret; | |
1669 | size -= ret; | |
1670 | } else | |
beb44aca | 1671 | append_sep = true; |
0ddec0fc TH |
1672 | |
1673 | ret = strscpy(names, cur->name, size); | |
1674 | if (ret < 0) | |
1675 | return false; | |
1676 | ||
1677 | names += ret; | |
1678 | size -= ret; | |
1679 | } | |
1680 | ||
1681 | return true; | |
1682 | } | |
1683 | ||
1684 | static bool seccomp_action_logged_from_name(u32 *action_logged, | |
1685 | const char *name) | |
1686 | { | |
1687 | const struct seccomp_log_name *cur; | |
1688 | ||
1689 | for (cur = seccomp_log_names; cur->name; cur++) { | |
1690 | if (!strcmp(cur->name, name)) { | |
1691 | *action_logged = cur->log; | |
1692 | return true; | |
1693 | } | |
1694 | } | |
1695 | ||
1696 | return false; | |
1697 | } | |
1698 | ||
1699 | static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names) | |
1700 | { | |
1701 | char *name; | |
1702 | ||
1703 | *actions_logged = 0; | |
1704 | while ((name = strsep(&names, " ")) && *name) { | |
1705 | u32 action_logged = 0; | |
1706 | ||
1707 | if (!seccomp_action_logged_from_name(&action_logged, name)) | |
1708 | return false; | |
1709 | ||
1710 | *actions_logged |= action_logged; | |
1711 | } | |
1712 | ||
1713 | return true; | |
1714 | } | |
1715 | ||
d013db02 TH |
1716 | static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer, |
1717 | size_t *lenp, loff_t *ppos) | |
0ddec0fc TH |
1718 | { |
1719 | char names[sizeof(seccomp_actions_avail)]; | |
1720 | struct ctl_table table; | |
d013db02 TH |
1721 | |
1722 | memset(names, 0, sizeof(names)); | |
1723 | ||
1724 | if (!seccomp_names_from_actions_logged(names, sizeof(names), | |
beb44aca | 1725 | seccomp_actions_logged, " ")) |
d013db02 TH |
1726 | return -EINVAL; |
1727 | ||
1728 | table = *ro_table; | |
1729 | table.data = names; | |
1730 | table.maxlen = sizeof(names); | |
1731 | return proc_dostring(&table, 0, buffer, lenp, ppos); | |
1732 | } | |
1733 | ||
1734 | static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer, | |
ea6eca77 | 1735 | size_t *lenp, loff_t *ppos, u32 *actions_logged) |
0ddec0fc TH |
1736 | { |
1737 | char names[sizeof(seccomp_actions_avail)]; | |
1738 | struct ctl_table table; | |
1739 | int ret; | |
1740 | ||
d013db02 | 1741 | if (!capable(CAP_SYS_ADMIN)) |
0ddec0fc TH |
1742 | return -EPERM; |
1743 | ||
1744 | memset(names, 0, sizeof(names)); | |
1745 | ||
0ddec0fc TH |
1746 | table = *ro_table; |
1747 | table.data = names; | |
1748 | table.maxlen = sizeof(names); | |
d013db02 | 1749 | ret = proc_dostring(&table, 1, buffer, lenp, ppos); |
0ddec0fc TH |
1750 | if (ret) |
1751 | return ret; | |
1752 | ||
ea6eca77 | 1753 | if (!seccomp_actions_logged_from_names(actions_logged, table.data)) |
d013db02 | 1754 | return -EINVAL; |
0ddec0fc | 1755 | |
ea6eca77 | 1756 | if (*actions_logged & SECCOMP_LOG_ALLOW) |
d013db02 | 1757 | return -EINVAL; |
0ddec0fc | 1758 | |
ea6eca77 | 1759 | seccomp_actions_logged = *actions_logged; |
0ddec0fc TH |
1760 | return 0; |
1761 | } | |
0ddec0fc | 1762 | |
ea6eca77 TH |
1763 | static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged, |
1764 | int ret) | |
1765 | { | |
1766 | char names[sizeof(seccomp_actions_avail)]; | |
1767 | char old_names[sizeof(seccomp_actions_avail)]; | |
1768 | const char *new = names; | |
1769 | const char *old = old_names; | |
0ddec0fc | 1770 | |
ea6eca77 TH |
1771 | if (!audit_enabled) |
1772 | return; | |
1773 | ||
1774 | memset(names, 0, sizeof(names)); | |
1775 | memset(old_names, 0, sizeof(old_names)); | |
1776 | ||
1777 | if (ret) | |
1778 | new = "?"; | |
1779 | else if (!actions_logged) | |
1780 | new = "(none)"; | |
1781 | else if (!seccomp_names_from_actions_logged(names, sizeof(names), | |
1782 | actions_logged, ",")) | |
1783 | new = "?"; | |
1784 | ||
1785 | if (!old_actions_logged) | |
1786 | old = "(none)"; | |
1787 | else if (!seccomp_names_from_actions_logged(old_names, | |
1788 | sizeof(old_names), | |
1789 | old_actions_logged, ",")) | |
1790 | old = "?"; | |
1791 | ||
1792 | return audit_seccomp_actions_logged(new, old, !ret); | |
1793 | } | |
1794 | ||
d013db02 | 1795 | static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write, |
32927393 | 1796 | void *buffer, size_t *lenp, |
d013db02 TH |
1797 | loff_t *ppos) |
1798 | { | |
ea6eca77 TH |
1799 | int ret; |
1800 | ||
1801 | if (write) { | |
1802 | u32 actions_logged = 0; | |
1803 | u32 old_actions_logged = seccomp_actions_logged; | |
1804 | ||
1805 | ret = write_actions_logged(ro_table, buffer, lenp, ppos, | |
1806 | &actions_logged); | |
1807 | audit_actions_logged(actions_logged, old_actions_logged, ret); | |
1808 | } else | |
1809 | ret = read_actions_logged(ro_table, buffer, lenp, ppos); | |
1810 | ||
1811 | return ret; | |
0ddec0fc TH |
1812 | } |
1813 | ||
8e5f1ad1 TH |
1814 | static struct ctl_path seccomp_sysctl_path[] = { |
1815 | { .procname = "kernel", }, | |
1816 | { .procname = "seccomp", }, | |
1817 | { } | |
1818 | }; | |
1819 | ||
1820 | static struct ctl_table seccomp_sysctl_table[] = { | |
1821 | { | |
1822 | .procname = "actions_avail", | |
1823 | .data = (void *) &seccomp_actions_avail, | |
1824 | .maxlen = sizeof(seccomp_actions_avail), | |
1825 | .mode = 0444, | |
1826 | .proc_handler = proc_dostring, | |
1827 | }, | |
0ddec0fc TH |
1828 | { |
1829 | .procname = "actions_logged", | |
1830 | .mode = 0644, | |
1831 | .proc_handler = seccomp_actions_logged_handler, | |
1832 | }, | |
8e5f1ad1 TH |
1833 | { } |
1834 | }; | |
1835 | ||
1836 | static int __init seccomp_sysctl_init(void) | |
1837 | { | |
1838 | struct ctl_table_header *hdr; | |
1839 | ||
1840 | hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table); | |
1841 | if (!hdr) | |
1842 | pr_warn("seccomp: sysctl registration failed\n"); | |
1843 | else | |
1844 | kmemleak_not_leak(hdr); | |
1845 | ||
1846 | return 0; | |
1847 | } | |
1848 | ||
1849 | device_initcall(seccomp_sysctl_init) | |
1850 | ||
1851 | #endif /* CONFIG_SYSCTL */ |