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 | |
e2cfabdf | 36 | #include <linux/filter.h> |
c2e1f2e3 | 37 | #include <linux/pid.h> |
fb0fadf9 | 38 | #include <linux/ptrace.h> |
e2cfabdf | 39 | #include <linux/security.h> |
e2cfabdf WD |
40 | #include <linux/tracehook.h> |
41 | #include <linux/uaccess.h> | |
42 | ||
43 | /** | |
44 | * struct seccomp_filter - container for seccomp BPF programs | |
45 | * | |
46 | * @usage: reference count to manage the object lifetime. | |
47 | * get/put helpers should be used when accessing an instance | |
48 | * outside of a lifetime-guarded section. In general, this | |
49 | * is only needed for handling filters shared across tasks. | |
e66a3997 | 50 | * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged |
e2cfabdf | 51 | * @prev: points to a previously installed, or inherited, filter |
285fdfc5 | 52 | * @prog: the BPF program to evaluate |
e2cfabdf WD |
53 | * |
54 | * seccomp_filter objects are organized in a tree linked via the @prev | |
55 | * pointer. For any task, it appears to be a singly-linked list starting | |
56 | * with current->seccomp.filter, the most recently attached or inherited filter. | |
57 | * However, multiple filters may share a @prev node, by way of fork(), which | |
58 | * results in a unidirectional tree existing in memory. This is similar to | |
59 | * how namespaces work. | |
60 | * | |
61 | * seccomp_filter objects should never be modified after being attached | |
62 | * to a task_struct (other than @usage). | |
63 | */ | |
64 | struct seccomp_filter { | |
0b5fa229 | 65 | refcount_t usage; |
e66a3997 | 66 | bool log; |
e2cfabdf | 67 | struct seccomp_filter *prev; |
7ae457c1 | 68 | struct bpf_prog *prog; |
e2cfabdf WD |
69 | }; |
70 | ||
71 | /* Limit any path through the tree to 256KB worth of instructions. */ | |
72 | #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter)) | |
73 | ||
bd4cf0ed | 74 | /* |
e2cfabdf WD |
75 | * Endianness is explicitly ignored and left for BPF program authors to manage |
76 | * as per the specific architecture. | |
77 | */ | |
bd4cf0ed | 78 | static void populate_seccomp_data(struct seccomp_data *sd) |
e2cfabdf | 79 | { |
bd4cf0ed AS |
80 | struct task_struct *task = current; |
81 | struct pt_regs *regs = task_pt_regs(task); | |
2eac7648 | 82 | unsigned long args[6]; |
e2cfabdf | 83 | |
bd4cf0ed | 84 | sd->nr = syscall_get_nr(task, regs); |
0b747172 | 85 | sd->arch = syscall_get_arch(); |
2eac7648 DB |
86 | syscall_get_arguments(task, regs, 0, 6, args); |
87 | sd->args[0] = args[0]; | |
88 | sd->args[1] = args[1]; | |
89 | sd->args[2] = args[2]; | |
90 | sd->args[3] = args[3]; | |
91 | sd->args[4] = args[4]; | |
92 | sd->args[5] = args[5]; | |
bd4cf0ed | 93 | sd->instruction_pointer = KSTK_EIP(task); |
e2cfabdf WD |
94 | } |
95 | ||
96 | /** | |
97 | * seccomp_check_filter - verify seccomp filter code | |
98 | * @filter: filter to verify | |
99 | * @flen: length of filter | |
100 | * | |
4df95ff4 | 101 | * Takes a previously checked filter (by bpf_check_classic) and |
e2cfabdf WD |
102 | * redirects all filter code that loads struct sk_buff data |
103 | * and related data through seccomp_bpf_load. It also | |
104 | * enforces length and alignment checking of those loads. | |
105 | * | |
106 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
107 | */ | |
108 | static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen) | |
109 | { | |
110 | int pc; | |
111 | for (pc = 0; pc < flen; pc++) { | |
112 | struct sock_filter *ftest = &filter[pc]; | |
113 | u16 code = ftest->code; | |
114 | u32 k = ftest->k; | |
115 | ||
116 | switch (code) { | |
34805931 | 117 | case BPF_LD | BPF_W | BPF_ABS: |
bd4cf0ed | 118 | ftest->code = BPF_LDX | BPF_W | BPF_ABS; |
e2cfabdf WD |
119 | /* 32-bit aligned and not out of bounds. */ |
120 | if (k >= sizeof(struct seccomp_data) || k & 3) | |
121 | return -EINVAL; | |
122 | continue; | |
34805931 | 123 | case BPF_LD | BPF_W | BPF_LEN: |
bd4cf0ed | 124 | ftest->code = BPF_LD | BPF_IMM; |
e2cfabdf WD |
125 | ftest->k = sizeof(struct seccomp_data); |
126 | continue; | |
34805931 | 127 | case BPF_LDX | BPF_W | BPF_LEN: |
bd4cf0ed | 128 | ftest->code = BPF_LDX | BPF_IMM; |
e2cfabdf WD |
129 | ftest->k = sizeof(struct seccomp_data); |
130 | continue; | |
131 | /* Explicitly include allowed calls. */ | |
34805931 DB |
132 | case BPF_RET | BPF_K: |
133 | case BPF_RET | BPF_A: | |
134 | case BPF_ALU | BPF_ADD | BPF_K: | |
135 | case BPF_ALU | BPF_ADD | BPF_X: | |
136 | case BPF_ALU | BPF_SUB | BPF_K: | |
137 | case BPF_ALU | BPF_SUB | BPF_X: | |
138 | case BPF_ALU | BPF_MUL | BPF_K: | |
139 | case BPF_ALU | BPF_MUL | BPF_X: | |
140 | case BPF_ALU | BPF_DIV | BPF_K: | |
141 | case BPF_ALU | BPF_DIV | BPF_X: | |
142 | case BPF_ALU | BPF_AND | BPF_K: | |
143 | case BPF_ALU | BPF_AND | BPF_X: | |
144 | case BPF_ALU | BPF_OR | BPF_K: | |
145 | case BPF_ALU | BPF_OR | BPF_X: | |
146 | case BPF_ALU | BPF_XOR | BPF_K: | |
147 | case BPF_ALU | BPF_XOR | BPF_X: | |
148 | case BPF_ALU | BPF_LSH | BPF_K: | |
149 | case BPF_ALU | BPF_LSH | BPF_X: | |
150 | case BPF_ALU | BPF_RSH | BPF_K: | |
151 | case BPF_ALU | BPF_RSH | BPF_X: | |
152 | case BPF_ALU | BPF_NEG: | |
153 | case BPF_LD | BPF_IMM: | |
154 | case BPF_LDX | BPF_IMM: | |
155 | case BPF_MISC | BPF_TAX: | |
156 | case BPF_MISC | BPF_TXA: | |
157 | case BPF_LD | BPF_MEM: | |
158 | case BPF_LDX | BPF_MEM: | |
159 | case BPF_ST: | |
160 | case BPF_STX: | |
161 | case BPF_JMP | BPF_JA: | |
162 | case BPF_JMP | BPF_JEQ | BPF_K: | |
163 | case BPF_JMP | BPF_JEQ | BPF_X: | |
164 | case BPF_JMP | BPF_JGE | BPF_K: | |
165 | case BPF_JMP | BPF_JGE | BPF_X: | |
166 | case BPF_JMP | BPF_JGT | BPF_K: | |
167 | case BPF_JMP | BPF_JGT | BPF_X: | |
168 | case BPF_JMP | BPF_JSET | BPF_K: | |
169 | case BPF_JMP | BPF_JSET | BPF_X: | |
e2cfabdf WD |
170 | continue; |
171 | default: | |
172 | return -EINVAL; | |
173 | } | |
174 | } | |
175 | return 0; | |
176 | } | |
177 | ||
178 | /** | |
285fdfc5 MS |
179 | * seccomp_run_filters - evaluates all seccomp filters against @sd |
180 | * @sd: optional seccomp data to be passed to filters | |
deb4de8b KC |
181 | * @match: stores struct seccomp_filter that resulted in the return value, |
182 | * unless filter returned SECCOMP_RET_ALLOW, in which case it will | |
183 | * be unchanged. | |
e2cfabdf WD |
184 | * |
185 | * Returns valid seccomp BPF response codes. | |
186 | */ | |
0466bdb9 | 187 | #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL))) |
deb4de8b KC |
188 | static u32 seccomp_run_filters(const struct seccomp_data *sd, |
189 | struct seccomp_filter **match) | |
e2cfabdf | 190 | { |
d39bd00d | 191 | struct seccomp_data sd_local; |
acf3b2c7 | 192 | u32 ret = SECCOMP_RET_ALLOW; |
8225d385 PK |
193 | /* Make sure cross-thread synced filter points somewhere sane. */ |
194 | struct seccomp_filter *f = | |
506458ef | 195 | READ_ONCE(current->seccomp.filter); |
acf3b2c7 WD |
196 | |
197 | /* Ensure unexpected behavior doesn't result in failing open. */ | |
3ba2530c | 198 | if (unlikely(WARN_ON(f == NULL))) |
4d3b0b05 | 199 | return SECCOMP_RET_KILL_PROCESS; |
acf3b2c7 | 200 | |
d39bd00d AL |
201 | if (!sd) { |
202 | populate_seccomp_data(&sd_local); | |
203 | sd = &sd_local; | |
204 | } | |
bd4cf0ed | 205 | |
e2cfabdf WD |
206 | /* |
207 | * All filters in the list are evaluated and the lowest BPF return | |
acf3b2c7 | 208 | * value always takes priority (ignoring the DATA). |
e2cfabdf | 209 | */ |
3ba2530c | 210 | for (; f; f = f->prev) { |
88575199 | 211 | u32 cur_ret = BPF_PROG_RUN(f->prog, sd); |
8f577cad | 212 | |
0466bdb9 | 213 | if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) { |
acf3b2c7 | 214 | ret = cur_ret; |
deb4de8b KC |
215 | *match = f; |
216 | } | |
e2cfabdf WD |
217 | } |
218 | return ret; | |
219 | } | |
1f41b450 | 220 | #endif /* CONFIG_SECCOMP_FILTER */ |
e2cfabdf | 221 | |
1f41b450 KC |
222 | static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) |
223 | { | |
69f6a34b | 224 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 225 | |
1f41b450 KC |
226 | if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) |
227 | return false; | |
228 | ||
229 | return true; | |
230 | } | |
231 | ||
5c307089 KC |
232 | /* |
233 | * If a given speculation mitigation is opt-in (prctl()-controlled), | |
234 | * select it, by disabling speculation (enabling mitigation). | |
235 | */ | |
236 | static inline void spec_mitigate(struct task_struct *task, | |
237 | unsigned long which) | |
238 | { | |
239 | int state = arch_prctl_spec_ctrl_get(task, which); | |
240 | ||
241 | if (state > 0 && (state & PR_SPEC_PRCTL)) | |
b849a812 | 242 | arch_prctl_spec_ctrl_set(task, which, PR_SPEC_FORCE_DISABLE); |
5c307089 KC |
243 | } |
244 | ||
3ba2530c | 245 | static inline void seccomp_assign_mode(struct task_struct *task, |
00a02d0c KC |
246 | unsigned long seccomp_mode, |
247 | unsigned long flags) | |
1f41b450 | 248 | { |
69f6a34b | 249 | assert_spin_locked(&task->sighand->siglock); |
dbd95212 | 250 | |
3ba2530c KC |
251 | task->seccomp.mode = seccomp_mode; |
252 | /* | |
253 | * Make sure TIF_SECCOMP cannot be set before the mode (and | |
254 | * filter) is set. | |
255 | */ | |
256 | smp_mb__before_atomic(); | |
00a02d0c KC |
257 | /* Assume default seccomp processes want spec flaw mitigation. */ |
258 | if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0) | |
259 | spec_mitigate(task, PR_SPEC_STORE_BYPASS); | |
3ba2530c | 260 | set_tsk_thread_flag(task, TIF_SECCOMP); |
1f41b450 KC |
261 | } |
262 | ||
263 | #ifdef CONFIG_SECCOMP_FILTER | |
c2e1f2e3 KC |
264 | /* Returns 1 if the parent is an ancestor of the child. */ |
265 | static int is_ancestor(struct seccomp_filter *parent, | |
266 | struct seccomp_filter *child) | |
267 | { | |
268 | /* NULL is the root ancestor. */ | |
269 | if (parent == NULL) | |
270 | return 1; | |
271 | for (; child; child = child->prev) | |
272 | if (child == parent) | |
273 | return 1; | |
274 | return 0; | |
275 | } | |
276 | ||
277 | /** | |
278 | * seccomp_can_sync_threads: checks if all threads can be synchronized | |
279 | * | |
280 | * Expects sighand and cred_guard_mutex locks to be held. | |
281 | * | |
282 | * Returns 0 on success, -ve on error, or the pid of a thread which was | |
283 | * either not in the correct seccomp mode or it did not have an ancestral | |
284 | * seccomp filter. | |
285 | */ | |
286 | static inline pid_t seccomp_can_sync_threads(void) | |
287 | { | |
288 | struct task_struct *thread, *caller; | |
289 | ||
290 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 291 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
292 | |
293 | /* Validate all threads being eligible for synchronization. */ | |
294 | caller = current; | |
295 | for_each_thread(caller, thread) { | |
296 | pid_t failed; | |
297 | ||
298 | /* Skip current, since it is initiating the sync. */ | |
299 | if (thread == caller) | |
300 | continue; | |
301 | ||
302 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || | |
303 | (thread->seccomp.mode == SECCOMP_MODE_FILTER && | |
304 | is_ancestor(thread->seccomp.filter, | |
305 | caller->seccomp.filter))) | |
306 | continue; | |
307 | ||
308 | /* Return the first thread that cannot be synchronized. */ | |
309 | failed = task_pid_vnr(thread); | |
310 | /* If the pid cannot be resolved, then return -ESRCH */ | |
311 | if (unlikely(WARN_ON(failed == 0))) | |
312 | failed = -ESRCH; | |
313 | return failed; | |
314 | } | |
315 | ||
316 | return 0; | |
317 | } | |
318 | ||
319 | /** | |
320 | * seccomp_sync_threads: sets all threads to use current's filter | |
321 | * | |
322 | * Expects sighand and cred_guard_mutex locks to be held, and for | |
323 | * seccomp_can_sync_threads() to have returned success already | |
324 | * without dropping the locks. | |
325 | * | |
326 | */ | |
00a02d0c | 327 | static inline void seccomp_sync_threads(unsigned long flags) |
c2e1f2e3 KC |
328 | { |
329 | struct task_struct *thread, *caller; | |
330 | ||
331 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 332 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
333 | |
334 | /* Synchronize all threads. */ | |
335 | caller = current; | |
336 | for_each_thread(caller, thread) { | |
337 | /* Skip current, since it needs no changes. */ | |
338 | if (thread == caller) | |
339 | continue; | |
340 | ||
341 | /* Get a task reference for the new leaf node. */ | |
342 | get_seccomp_filter(caller); | |
343 | /* | |
344 | * Drop the task reference to the shared ancestor since | |
345 | * current's path will hold a reference. (This also | |
346 | * allows a put before the assignment.) | |
347 | */ | |
348 | put_seccomp_filter(thread); | |
349 | smp_store_release(&thread->seccomp.filter, | |
350 | caller->seccomp.filter); | |
103502a3 JH |
351 | |
352 | /* | |
353 | * Don't let an unprivileged task work around | |
354 | * the no_new_privs restriction by creating | |
355 | * a thread that sets it up, enters seccomp, | |
356 | * then dies. | |
357 | */ | |
358 | if (task_no_new_privs(caller)) | |
359 | task_set_no_new_privs(thread); | |
360 | ||
c2e1f2e3 KC |
361 | /* |
362 | * Opt the other thread into seccomp if needed. | |
363 | * As threads are considered to be trust-realm | |
364 | * equivalent (see ptrace_may_access), it is safe to | |
365 | * allow one thread to transition the other. | |
366 | */ | |
103502a3 | 367 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) |
00a02d0c KC |
368 | seccomp_assign_mode(thread, SECCOMP_MODE_FILTER, |
369 | flags); | |
c2e1f2e3 KC |
370 | } |
371 | } | |
372 | ||
e2cfabdf | 373 | /** |
c8bee430 | 374 | * seccomp_prepare_filter: Prepares a seccomp filter for use. |
e2cfabdf WD |
375 | * @fprog: BPF program to install |
376 | * | |
c8bee430 | 377 | * Returns filter on success or an ERR_PTR on failure. |
e2cfabdf | 378 | */ |
c8bee430 | 379 | static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog) |
e2cfabdf | 380 | { |
ac67eb2c DB |
381 | struct seccomp_filter *sfilter; |
382 | int ret; | |
97f2645f | 383 | const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE); |
e2cfabdf WD |
384 | |
385 | if (fprog->len == 0 || fprog->len > BPF_MAXINSNS) | |
c8bee430 | 386 | return ERR_PTR(-EINVAL); |
d9e12f42 | 387 | |
c8bee430 | 388 | BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter)); |
e2cfabdf WD |
389 | |
390 | /* | |
119ce5c8 | 391 | * Installing a seccomp filter requires that the task has |
e2cfabdf WD |
392 | * CAP_SYS_ADMIN in its namespace or be running with no_new_privs. |
393 | * This avoids scenarios where unprivileged tasks can affect the | |
394 | * behavior of privileged children. | |
395 | */ | |
1d4457f9 | 396 | if (!task_no_new_privs(current) && |
e2cfabdf WD |
397 | security_capable_noaudit(current_cred(), current_user_ns(), |
398 | CAP_SYS_ADMIN) != 0) | |
c8bee430 | 399 | return ERR_PTR(-EACCES); |
e2cfabdf | 400 | |
bd4cf0ed | 401 | /* Allocate a new seccomp_filter */ |
ac67eb2c DB |
402 | sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN); |
403 | if (!sfilter) | |
d9e12f42 | 404 | return ERR_PTR(-ENOMEM); |
ac67eb2c DB |
405 | |
406 | ret = bpf_prog_create_from_user(&sfilter->prog, fprog, | |
f8e529ed | 407 | seccomp_check_filter, save_orig); |
ac67eb2c DB |
408 | if (ret < 0) { |
409 | kfree(sfilter); | |
410 | return ERR_PTR(ret); | |
d9e12f42 | 411 | } |
bd4cf0ed | 412 | |
0b5fa229 | 413 | refcount_set(&sfilter->usage, 1); |
e2cfabdf | 414 | |
ac67eb2c | 415 | return sfilter; |
e2cfabdf WD |
416 | } |
417 | ||
418 | /** | |
c8bee430 | 419 | * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog |
e2cfabdf WD |
420 | * @user_filter: pointer to the user data containing a sock_fprog. |
421 | * | |
422 | * Returns 0 on success and non-zero otherwise. | |
423 | */ | |
c8bee430 KC |
424 | static struct seccomp_filter * |
425 | seccomp_prepare_user_filter(const char __user *user_filter) | |
e2cfabdf WD |
426 | { |
427 | struct sock_fprog fprog; | |
c8bee430 | 428 | struct seccomp_filter *filter = ERR_PTR(-EFAULT); |
e2cfabdf WD |
429 | |
430 | #ifdef CONFIG_COMPAT | |
5c38065e | 431 | if (in_compat_syscall()) { |
e2cfabdf WD |
432 | struct compat_sock_fprog fprog32; |
433 | if (copy_from_user(&fprog32, user_filter, sizeof(fprog32))) | |
434 | goto out; | |
435 | fprog.len = fprog32.len; | |
436 | fprog.filter = compat_ptr(fprog32.filter); | |
437 | } else /* falls through to the if below. */ | |
438 | #endif | |
439 | if (copy_from_user(&fprog, user_filter, sizeof(fprog))) | |
440 | goto out; | |
c8bee430 | 441 | filter = seccomp_prepare_filter(&fprog); |
e2cfabdf | 442 | out: |
c8bee430 KC |
443 | return filter; |
444 | } | |
445 | ||
446 | /** | |
447 | * seccomp_attach_filter: validate and attach filter | |
448 | * @flags: flags to change filter behavior | |
449 | * @filter: seccomp filter to add to the current process | |
450 | * | |
dbd95212 KC |
451 | * Caller must be holding current->sighand->siglock lock. |
452 | * | |
c8bee430 KC |
453 | * Returns 0 on success, -ve on error. |
454 | */ | |
455 | static long seccomp_attach_filter(unsigned int flags, | |
456 | struct seccomp_filter *filter) | |
457 | { | |
458 | unsigned long total_insns; | |
459 | struct seccomp_filter *walker; | |
460 | ||
69f6a34b | 461 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 462 | |
c8bee430 KC |
463 | /* Validate resulting filter length. */ |
464 | total_insns = filter->prog->len; | |
465 | for (walker = current->seccomp.filter; walker; walker = walker->prev) | |
466 | total_insns += walker->prog->len + 4; /* 4 instr penalty */ | |
467 | if (total_insns > MAX_INSNS_PER_PATH) | |
468 | return -ENOMEM; | |
469 | ||
c2e1f2e3 KC |
470 | /* If thread sync has been requested, check that it is possible. */ |
471 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) { | |
472 | int ret; | |
473 | ||
474 | ret = seccomp_can_sync_threads(); | |
475 | if (ret) | |
476 | return ret; | |
477 | } | |
478 | ||
e66a3997 TH |
479 | /* Set log flag, if present. */ |
480 | if (flags & SECCOMP_FILTER_FLAG_LOG) | |
481 | filter->log = true; | |
482 | ||
c8bee430 KC |
483 | /* |
484 | * If there is an existing filter, make it the prev and don't drop its | |
485 | * task reference. | |
486 | */ | |
487 | filter->prev = current->seccomp.filter; | |
488 | current->seccomp.filter = filter; | |
489 | ||
c2e1f2e3 KC |
490 | /* Now that the new filter is in place, synchronize to all threads. */ |
491 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) | |
00a02d0c | 492 | seccomp_sync_threads(flags); |
c2e1f2e3 | 493 | |
c8bee430 | 494 | return 0; |
e2cfabdf WD |
495 | } |
496 | ||
084f5601 | 497 | static void __get_seccomp_filter(struct seccomp_filter *filter) |
66a733ea ON |
498 | { |
499 | /* Reference count is bounded by the number of total processes. */ | |
500 | refcount_inc(&filter->usage); | |
501 | } | |
502 | ||
e2cfabdf WD |
503 | /* get_seccomp_filter - increments the reference count of the filter on @tsk */ |
504 | void get_seccomp_filter(struct task_struct *tsk) | |
505 | { | |
506 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
507 | if (!orig) | |
508 | return; | |
66a733ea | 509 | __get_seccomp_filter(orig); |
e2cfabdf WD |
510 | } |
511 | ||
c8bee430 KC |
512 | static inline void seccomp_filter_free(struct seccomp_filter *filter) |
513 | { | |
514 | if (filter) { | |
bab18991 | 515 | bpf_prog_destroy(filter->prog); |
c8bee430 KC |
516 | kfree(filter); |
517 | } | |
518 | } | |
519 | ||
66a733ea | 520 | static void __put_seccomp_filter(struct seccomp_filter *orig) |
e2cfabdf | 521 | { |
e2cfabdf | 522 | /* Clean up single-reference branches iteratively. */ |
0b5fa229 | 523 | while (orig && refcount_dec_and_test(&orig->usage)) { |
e2cfabdf WD |
524 | struct seccomp_filter *freeme = orig; |
525 | orig = orig->prev; | |
c8bee430 | 526 | seccomp_filter_free(freeme); |
e2cfabdf WD |
527 | } |
528 | } | |
bb6ea430 | 529 | |
66a733ea ON |
530 | /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */ |
531 | void put_seccomp_filter(struct task_struct *tsk) | |
532 | { | |
533 | __put_seccomp_filter(tsk->seccomp.filter); | |
534 | } | |
535 | ||
b25e6716 MF |
536 | static void seccomp_init_siginfo(siginfo_t *info, int syscall, int reason) |
537 | { | |
3b10db2b | 538 | clear_siginfo(info); |
b25e6716 MF |
539 | info->si_signo = SIGSYS; |
540 | info->si_code = SYS_SECCOMP; | |
541 | info->si_call_addr = (void __user *)KSTK_EIP(current); | |
542 | info->si_errno = reason; | |
543 | info->si_arch = syscall_get_arch(); | |
544 | info->si_syscall = syscall; | |
545 | } | |
546 | ||
bb6ea430 WD |
547 | /** |
548 | * seccomp_send_sigsys - signals the task to allow in-process syscall emulation | |
549 | * @syscall: syscall number to send to userland | |
550 | * @reason: filter-supplied reason code to send to userland (via si_errno) | |
551 | * | |
552 | * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info. | |
553 | */ | |
554 | static void seccomp_send_sigsys(int syscall, int reason) | |
555 | { | |
556 | struct siginfo info; | |
b25e6716 | 557 | seccomp_init_siginfo(&info, syscall, reason); |
bb6ea430 WD |
558 | force_sig_info(SIGSYS, &info, current); |
559 | } | |
e2cfabdf | 560 | #endif /* CONFIG_SECCOMP_FILTER */ |
1da177e4 | 561 | |
0ddec0fc | 562 | /* For use with seccomp_actions_logged */ |
4d3b0b05 KC |
563 | #define SECCOMP_LOG_KILL_PROCESS (1 << 0) |
564 | #define SECCOMP_LOG_KILL_THREAD (1 << 1) | |
0ddec0fc TH |
565 | #define SECCOMP_LOG_TRAP (1 << 2) |
566 | #define SECCOMP_LOG_ERRNO (1 << 3) | |
567 | #define SECCOMP_LOG_TRACE (1 << 4) | |
59f5cf44 TH |
568 | #define SECCOMP_LOG_LOG (1 << 5) |
569 | #define SECCOMP_LOG_ALLOW (1 << 6) | |
0ddec0fc | 570 | |
4d3b0b05 KC |
571 | static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS | |
572 | SECCOMP_LOG_KILL_THREAD | | |
fd76875c KC |
573 | SECCOMP_LOG_TRAP | |
574 | SECCOMP_LOG_ERRNO | | |
575 | SECCOMP_LOG_TRACE | | |
59f5cf44 | 576 | SECCOMP_LOG_LOG; |
0ddec0fc | 577 | |
e66a3997 TH |
578 | static inline void seccomp_log(unsigned long syscall, long signr, u32 action, |
579 | bool requested) | |
0ddec0fc TH |
580 | { |
581 | bool log = false; | |
582 | ||
583 | switch (action) { | |
584 | case SECCOMP_RET_ALLOW: | |
e66a3997 | 585 | break; |
0ddec0fc | 586 | case SECCOMP_RET_TRAP: |
e66a3997 TH |
587 | log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP; |
588 | break; | |
0ddec0fc | 589 | case SECCOMP_RET_ERRNO: |
e66a3997 TH |
590 | log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO; |
591 | break; | |
0ddec0fc | 592 | case SECCOMP_RET_TRACE: |
e66a3997 | 593 | log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE; |
0ddec0fc | 594 | break; |
59f5cf44 TH |
595 | case SECCOMP_RET_LOG: |
596 | log = seccomp_actions_logged & SECCOMP_LOG_LOG; | |
597 | break; | |
fd76875c | 598 | case SECCOMP_RET_KILL_THREAD: |
fd76875c | 599 | log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD; |
4d3b0b05 KC |
600 | break; |
601 | case SECCOMP_RET_KILL_PROCESS: | |
602 | default: | |
603 | log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS; | |
0ddec0fc TH |
604 | } |
605 | ||
606 | /* | |
fd76875c | 607 | * Force an audit message to be emitted when the action is RET_KILL_*, |
59f5cf44 TH |
608 | * RET_LOG, or the FILTER_FLAG_LOG bit was set and the action is |
609 | * allowed to be logged by the admin. | |
0ddec0fc TH |
610 | */ |
611 | if (log) | |
612 | return __audit_seccomp(syscall, signr, action); | |
613 | ||
614 | /* | |
615 | * Let the audit subsystem decide if the action should be audited based | |
616 | * on whether the current task itself is being audited. | |
617 | */ | |
618 | return audit_seccomp(syscall, signr, action); | |
619 | } | |
620 | ||
1da177e4 LT |
621 | /* |
622 | * Secure computing mode 1 allows only read/write/exit/sigreturn. | |
623 | * To be fully secure this must be combined with rlimit | |
624 | * to limit the stack allocations too. | |
625 | */ | |
cb4253aa | 626 | static const int mode1_syscalls[] = { |
1da177e4 LT |
627 | __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, |
628 | 0, /* null terminated */ | |
629 | }; | |
630 | ||
a4412fc9 | 631 | static void __secure_computing_strict(int this_syscall) |
1da177e4 | 632 | { |
cb4253aa | 633 | const int *syscall_whitelist = mode1_syscalls; |
a4412fc9 | 634 | #ifdef CONFIG_COMPAT |
5c38065e | 635 | if (in_compat_syscall()) |
c983f0e8 | 636 | syscall_whitelist = get_compat_mode1_syscalls(); |
a4412fc9 AL |
637 | #endif |
638 | do { | |
639 | if (*syscall_whitelist == this_syscall) | |
640 | return; | |
641 | } while (*++syscall_whitelist); | |
642 | ||
643 | #ifdef SECCOMP_DEBUG | |
644 | dump_stack(); | |
645 | #endif | |
fd76875c | 646 | seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true); |
a4412fc9 AL |
647 | do_exit(SIGKILL); |
648 | } | |
649 | ||
650 | #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER | |
651 | void secure_computing_strict(int this_syscall) | |
652 | { | |
653 | int mode = current->seccomp.mode; | |
654 | ||
97f2645f | 655 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 TA |
656 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
657 | return; | |
658 | ||
221272f9 | 659 | if (mode == SECCOMP_MODE_DISABLED) |
a4412fc9 AL |
660 | return; |
661 | else if (mode == SECCOMP_MODE_STRICT) | |
662 | __secure_computing_strict(this_syscall); | |
663 | else | |
664 | BUG(); | |
665 | } | |
666 | #else | |
13aa72f0 AL |
667 | |
668 | #ifdef CONFIG_SECCOMP_FILTER | |
ce6526e8 KC |
669 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
670 | const bool recheck_after_trace) | |
13aa72f0 AL |
671 | { |
672 | u32 filter_ret, action; | |
deb4de8b | 673 | struct seccomp_filter *match = NULL; |
13aa72f0 | 674 | int data; |
1da177e4 | 675 | |
3ba2530c KC |
676 | /* |
677 | * Make sure that any changes to mode from another thread have | |
678 | * been seen after TIF_SECCOMP was seen. | |
679 | */ | |
680 | rmb(); | |
681 | ||
deb4de8b | 682 | filter_ret = seccomp_run_filters(sd, &match); |
13aa72f0 | 683 | data = filter_ret & SECCOMP_RET_DATA; |
0466bdb9 | 684 | action = filter_ret & SECCOMP_RET_ACTION_FULL; |
13aa72f0 AL |
685 | |
686 | switch (action) { | |
687 | case SECCOMP_RET_ERRNO: | |
580c57f1 KC |
688 | /* Set low-order bits as an errno, capped at MAX_ERRNO. */ |
689 | if (data > MAX_ERRNO) | |
690 | data = MAX_ERRNO; | |
d39bd00d | 691 | syscall_set_return_value(current, task_pt_regs(current), |
13aa72f0 AL |
692 | -data, 0); |
693 | goto skip; | |
694 | ||
695 | case SECCOMP_RET_TRAP: | |
696 | /* Show the handler the original registers. */ | |
d39bd00d | 697 | syscall_rollback(current, task_pt_regs(current)); |
13aa72f0 AL |
698 | /* Let the filter pass back 16 bits of data. */ |
699 | seccomp_send_sigsys(this_syscall, data); | |
700 | goto skip; | |
701 | ||
702 | case SECCOMP_RET_TRACE: | |
ce6526e8 KC |
703 | /* We've been put in this state by the ptracer already. */ |
704 | if (recheck_after_trace) | |
705 | return 0; | |
706 | ||
8112c4f1 KC |
707 | /* ENOSYS these calls if there is no tracer attached. */ |
708 | if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { | |
709 | syscall_set_return_value(current, | |
710 | task_pt_regs(current), | |
711 | -ENOSYS, 0); | |
712 | goto skip; | |
713 | } | |
714 | ||
715 | /* Allow the BPF to provide the event message */ | |
716 | ptrace_event(PTRACE_EVENT_SECCOMP, data); | |
717 | /* | |
718 | * The delivery of a fatal signal during event | |
485a252a KC |
719 | * notification may silently skip tracer notification, |
720 | * which could leave us with a potentially unmodified | |
721 | * syscall that the tracer would have liked to have | |
722 | * changed. Since the process is about to die, we just | |
723 | * force the syscall to be skipped and let the signal | |
724 | * kill the process and correctly handle any tracer exit | |
725 | * notifications. | |
8112c4f1 KC |
726 | */ |
727 | if (fatal_signal_pending(current)) | |
485a252a | 728 | goto skip; |
8112c4f1 KC |
729 | /* Check if the tracer forced the syscall to be skipped. */ |
730 | this_syscall = syscall_get_nr(current, task_pt_regs(current)); | |
731 | if (this_syscall < 0) | |
732 | goto skip; | |
733 | ||
ce6526e8 KC |
734 | /* |
735 | * Recheck the syscall, since it may have changed. This | |
736 | * intentionally uses a NULL struct seccomp_data to force | |
737 | * a reload of all registers. This does not goto skip since | |
738 | * a skip would have already been reported. | |
739 | */ | |
740 | if (__seccomp_filter(this_syscall, NULL, true)) | |
741 | return -1; | |
742 | ||
8112c4f1 | 743 | return 0; |
13aa72f0 | 744 | |
59f5cf44 TH |
745 | case SECCOMP_RET_LOG: |
746 | seccomp_log(this_syscall, 0, action, true); | |
747 | return 0; | |
748 | ||
13aa72f0 | 749 | case SECCOMP_RET_ALLOW: |
deb4de8b KC |
750 | /* |
751 | * Note that the "match" filter will always be NULL for | |
752 | * this action since SECCOMP_RET_ALLOW is the starting | |
753 | * state in seccomp_run_filters(). | |
754 | */ | |
8112c4f1 | 755 | return 0; |
13aa72f0 | 756 | |
fd76875c | 757 | case SECCOMP_RET_KILL_THREAD: |
4d3b0b05 | 758 | case SECCOMP_RET_KILL_PROCESS: |
131b6351 | 759 | default: |
e66a3997 | 760 | seccomp_log(this_syscall, SIGSYS, action, true); |
d7276e32 | 761 | /* Dump core only if this is the last remaining thread. */ |
4d3b0b05 KC |
762 | if (action == SECCOMP_RET_KILL_PROCESS || |
763 | get_nr_threads(current) == 1) { | |
131b6351 KC |
764 | siginfo_t info; |
765 | ||
d7276e32 KC |
766 | /* Show the original registers in the dump. */ |
767 | syscall_rollback(current, task_pt_regs(current)); | |
768 | /* Trigger a manual coredump since do_exit skips it. */ | |
769 | seccomp_init_siginfo(&info, this_syscall, data); | |
770 | do_coredump(&info); | |
771 | } | |
4d3b0b05 KC |
772 | if (action == SECCOMP_RET_KILL_PROCESS) |
773 | do_group_exit(SIGSYS); | |
774 | else | |
775 | do_exit(SIGSYS); | |
13aa72f0 AL |
776 | } |
777 | ||
778 | unreachable(); | |
779 | ||
780 | skip: | |
e66a3997 | 781 | seccomp_log(this_syscall, 0, action, match ? match->log : false); |
8112c4f1 KC |
782 | return -1; |
783 | } | |
784 | #else | |
ce6526e8 KC |
785 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
786 | const bool recheck_after_trace) | |
8112c4f1 KC |
787 | { |
788 | BUG(); | |
13aa72f0 | 789 | } |
1da177e4 | 790 | #endif |
13aa72f0 | 791 | |
8112c4f1 | 792 | int __secure_computing(const struct seccomp_data *sd) |
13aa72f0 AL |
793 | { |
794 | int mode = current->seccomp.mode; | |
8112c4f1 | 795 | int this_syscall; |
13aa72f0 | 796 | |
97f2645f | 797 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 | 798 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
8112c4f1 KC |
799 | return 0; |
800 | ||
801 | this_syscall = sd ? sd->nr : | |
802 | syscall_get_nr(current, task_pt_regs(current)); | |
13c4a901 | 803 | |
13aa72f0 | 804 | switch (mode) { |
e2cfabdf | 805 | case SECCOMP_MODE_STRICT: |
13aa72f0 | 806 | __secure_computing_strict(this_syscall); /* may call do_exit */ |
8112c4f1 | 807 | return 0; |
13aa72f0 | 808 | case SECCOMP_MODE_FILTER: |
ce6526e8 | 809 | return __seccomp_filter(this_syscall, sd, false); |
1da177e4 LT |
810 | default: |
811 | BUG(); | |
812 | } | |
13aa72f0 | 813 | } |
a4412fc9 | 814 | #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ |
1d9d02fe AA |
815 | |
816 | long prctl_get_seccomp(void) | |
817 | { | |
818 | return current->seccomp.mode; | |
819 | } | |
820 | ||
e2cfabdf | 821 | /** |
3b23dd12 | 822 | * seccomp_set_mode_strict: internal function for setting strict seccomp |
e2cfabdf WD |
823 | * |
824 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
825 | * | |
826 | * Returns 0 on success or -EINVAL on failure. | |
827 | */ | |
3b23dd12 | 828 | static long seccomp_set_mode_strict(void) |
1d9d02fe | 829 | { |
3b23dd12 | 830 | const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; |
e2cfabdf | 831 | long ret = -EINVAL; |
1d9d02fe | 832 | |
dbd95212 KC |
833 | spin_lock_irq(¤t->sighand->siglock); |
834 | ||
1f41b450 | 835 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1d9d02fe AA |
836 | goto out; |
837 | ||
cf99abac | 838 | #ifdef TIF_NOTSC |
3b23dd12 | 839 | disable_TSC(); |
cf99abac | 840 | #endif |
00a02d0c | 841 | seccomp_assign_mode(current, seccomp_mode, 0); |
3b23dd12 KC |
842 | ret = 0; |
843 | ||
844 | out: | |
dbd95212 | 845 | spin_unlock_irq(¤t->sighand->siglock); |
3b23dd12 KC |
846 | |
847 | return ret; | |
848 | } | |
849 | ||
e2cfabdf | 850 | #ifdef CONFIG_SECCOMP_FILTER |
3b23dd12 KC |
851 | /** |
852 | * seccomp_set_mode_filter: internal function for setting seccomp filter | |
48dc92b9 | 853 | * @flags: flags to change filter behavior |
3b23dd12 KC |
854 | * @filter: struct sock_fprog containing filter |
855 | * | |
856 | * This function may be called repeatedly to install additional filters. | |
857 | * Every filter successfully installed will be evaluated (in reverse order) | |
858 | * for each system call the task makes. | |
859 | * | |
860 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
861 | * | |
862 | * Returns 0 on success or -EINVAL on failure. | |
863 | */ | |
48dc92b9 KC |
864 | static long seccomp_set_mode_filter(unsigned int flags, |
865 | const char __user *filter) | |
3b23dd12 KC |
866 | { |
867 | const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; | |
c8bee430 | 868 | struct seccomp_filter *prepared = NULL; |
3b23dd12 KC |
869 | long ret = -EINVAL; |
870 | ||
48dc92b9 | 871 | /* Validate flags. */ |
c2e1f2e3 | 872 | if (flags & ~SECCOMP_FILTER_FLAG_MASK) |
dbd95212 | 873 | return -EINVAL; |
48dc92b9 | 874 | |
c8bee430 KC |
875 | /* Prepare the new filter before holding any locks. */ |
876 | prepared = seccomp_prepare_user_filter(filter); | |
877 | if (IS_ERR(prepared)) | |
878 | return PTR_ERR(prepared); | |
879 | ||
c2e1f2e3 KC |
880 | /* |
881 | * Make sure we cannot change seccomp or nnp state via TSYNC | |
882 | * while another thread is in the middle of calling exec. | |
883 | */ | |
884 | if (flags & SECCOMP_FILTER_FLAG_TSYNC && | |
885 | mutex_lock_killable(¤t->signal->cred_guard_mutex)) | |
886 | goto out_free; | |
887 | ||
dbd95212 KC |
888 | spin_lock_irq(¤t->sighand->siglock); |
889 | ||
3b23dd12 KC |
890 | if (!seccomp_may_assign_mode(seccomp_mode)) |
891 | goto out; | |
892 | ||
c8bee430 | 893 | ret = seccomp_attach_filter(flags, prepared); |
3b23dd12 | 894 | if (ret) |
e2cfabdf | 895 | goto out; |
c8bee430 KC |
896 | /* Do not free the successfully attached filter. */ |
897 | prepared = NULL; | |
1d9d02fe | 898 | |
00a02d0c | 899 | seccomp_assign_mode(current, seccomp_mode, flags); |
e2cfabdf | 900 | out: |
dbd95212 | 901 | spin_unlock_irq(¤t->sighand->siglock); |
c2e1f2e3 KC |
902 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
903 | mutex_unlock(¤t->signal->cred_guard_mutex); | |
904 | out_free: | |
c8bee430 | 905 | seccomp_filter_free(prepared); |
1d9d02fe AA |
906 | return ret; |
907 | } | |
3b23dd12 | 908 | #else |
48dc92b9 KC |
909 | static inline long seccomp_set_mode_filter(unsigned int flags, |
910 | const char __user *filter) | |
3b23dd12 KC |
911 | { |
912 | return -EINVAL; | |
913 | } | |
914 | #endif | |
d78ab02c | 915 | |
d612b1fd TH |
916 | static long seccomp_get_action_avail(const char __user *uaction) |
917 | { | |
918 | u32 action; | |
919 | ||
920 | if (copy_from_user(&action, uaction, sizeof(action))) | |
921 | return -EFAULT; | |
922 | ||
923 | switch (action) { | |
0466bdb9 | 924 | case SECCOMP_RET_KILL_PROCESS: |
fd76875c | 925 | case SECCOMP_RET_KILL_THREAD: |
d612b1fd TH |
926 | case SECCOMP_RET_TRAP: |
927 | case SECCOMP_RET_ERRNO: | |
928 | case SECCOMP_RET_TRACE: | |
59f5cf44 | 929 | case SECCOMP_RET_LOG: |
d612b1fd TH |
930 | case SECCOMP_RET_ALLOW: |
931 | break; | |
932 | default: | |
933 | return -EOPNOTSUPP; | |
934 | } | |
935 | ||
936 | return 0; | |
937 | } | |
938 | ||
48dc92b9 KC |
939 | /* Common entry point for both prctl and syscall. */ |
940 | static long do_seccomp(unsigned int op, unsigned int flags, | |
941 | const char __user *uargs) | |
942 | { | |
943 | switch (op) { | |
944 | case SECCOMP_SET_MODE_STRICT: | |
945 | if (flags != 0 || uargs != NULL) | |
946 | return -EINVAL; | |
947 | return seccomp_set_mode_strict(); | |
948 | case SECCOMP_SET_MODE_FILTER: | |
949 | return seccomp_set_mode_filter(flags, uargs); | |
d612b1fd TH |
950 | case SECCOMP_GET_ACTION_AVAIL: |
951 | if (flags != 0) | |
952 | return -EINVAL; | |
953 | ||
954 | return seccomp_get_action_avail(uargs); | |
48dc92b9 KC |
955 | default: |
956 | return -EINVAL; | |
957 | } | |
958 | } | |
959 | ||
960 | SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, | |
961 | const char __user *, uargs) | |
962 | { | |
963 | return do_seccomp(op, flags, uargs); | |
964 | } | |
965 | ||
d78ab02c KC |
966 | /** |
967 | * prctl_set_seccomp: configures current->seccomp.mode | |
968 | * @seccomp_mode: requested mode to use | |
969 | * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER | |
970 | * | |
971 | * Returns 0 on success or -EINVAL on failure. | |
972 | */ | |
973 | long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter) | |
974 | { | |
48dc92b9 KC |
975 | unsigned int op; |
976 | char __user *uargs; | |
977 | ||
3b23dd12 KC |
978 | switch (seccomp_mode) { |
979 | case SECCOMP_MODE_STRICT: | |
48dc92b9 KC |
980 | op = SECCOMP_SET_MODE_STRICT; |
981 | /* | |
982 | * Setting strict mode through prctl always ignored filter, | |
983 | * so make sure it is always NULL here to pass the internal | |
984 | * check in do_seccomp(). | |
985 | */ | |
986 | uargs = NULL; | |
987 | break; | |
3b23dd12 | 988 | case SECCOMP_MODE_FILTER: |
48dc92b9 KC |
989 | op = SECCOMP_SET_MODE_FILTER; |
990 | uargs = filter; | |
991 | break; | |
3b23dd12 KC |
992 | default: |
993 | return -EINVAL; | |
994 | } | |
48dc92b9 KC |
995 | |
996 | /* prctl interface doesn't have flags, so they are always zero. */ | |
997 | return do_seccomp(op, 0, uargs); | |
d78ab02c | 998 | } |
f8e529ed TA |
999 | |
1000 | #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE) | |
f06eae83 TA |
1001 | static struct seccomp_filter *get_nth_filter(struct task_struct *task, |
1002 | unsigned long filter_off) | |
f8e529ed | 1003 | { |
f06eae83 TA |
1004 | struct seccomp_filter *orig, *filter; |
1005 | unsigned long count; | |
f8e529ed | 1006 | |
f06eae83 TA |
1007 | /* |
1008 | * Note: this is only correct because the caller should be the (ptrace) | |
1009 | * tracer of the task, otherwise lock_task_sighand is needed. | |
1010 | */ | |
f8e529ed | 1011 | spin_lock_irq(&task->sighand->siglock); |
f06eae83 | 1012 | |
f8e529ed | 1013 | if (task->seccomp.mode != SECCOMP_MODE_FILTER) { |
f06eae83 TA |
1014 | spin_unlock_irq(&task->sighand->siglock); |
1015 | return ERR_PTR(-EINVAL); | |
f8e529ed TA |
1016 | } |
1017 | ||
f06eae83 TA |
1018 | orig = task->seccomp.filter; |
1019 | __get_seccomp_filter(orig); | |
1020 | spin_unlock_irq(&task->sighand->siglock); | |
1021 | ||
1022 | count = 0; | |
1023 | for (filter = orig; filter; filter = filter->prev) | |
f8e529ed | 1024 | count++; |
f8e529ed TA |
1025 | |
1026 | if (filter_off >= count) { | |
f06eae83 | 1027 | filter = ERR_PTR(-ENOENT); |
f8e529ed TA |
1028 | goto out; |
1029 | } | |
f8e529ed | 1030 | |
f06eae83 TA |
1031 | count -= filter_off; |
1032 | for (filter = orig; filter && count > 1; filter = filter->prev) | |
f8e529ed | 1033 | count--; |
f8e529ed TA |
1034 | |
1035 | if (WARN_ON(count != 1 || !filter)) { | |
f06eae83 | 1036 | filter = ERR_PTR(-ENOENT); |
f8e529ed TA |
1037 | goto out; |
1038 | } | |
1039 | ||
f06eae83 TA |
1040 | __get_seccomp_filter(filter); |
1041 | ||
1042 | out: | |
1043 | __put_seccomp_filter(orig); | |
1044 | return filter; | |
1045 | } | |
1046 | ||
1047 | long seccomp_get_filter(struct task_struct *task, unsigned long filter_off, | |
1048 | void __user *data) | |
1049 | { | |
1050 | struct seccomp_filter *filter; | |
1051 | struct sock_fprog_kern *fprog; | |
1052 | long ret; | |
1053 | ||
1054 | if (!capable(CAP_SYS_ADMIN) || | |
1055 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { | |
1056 | return -EACCES; | |
1057 | } | |
1058 | ||
1059 | filter = get_nth_filter(task, filter_off); | |
1060 | if (IS_ERR(filter)) | |
1061 | return PTR_ERR(filter); | |
1062 | ||
f8e529ed TA |
1063 | fprog = filter->prog->orig_prog; |
1064 | if (!fprog) { | |
470bf1f2 | 1065 | /* This must be a new non-cBPF filter, since we save |
f8e529ed TA |
1066 | * every cBPF filter's orig_prog above when |
1067 | * CONFIG_CHECKPOINT_RESTORE is enabled. | |
1068 | */ | |
1069 | ret = -EMEDIUMTYPE; | |
1070 | goto out; | |
1071 | } | |
1072 | ||
1073 | ret = fprog->len; | |
1074 | if (!data) | |
1075 | goto out; | |
1076 | ||
f8e529ed TA |
1077 | if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog))) |
1078 | ret = -EFAULT; | |
1079 | ||
f8e529ed | 1080 | out: |
66a733ea | 1081 | __put_seccomp_filter(filter); |
f8e529ed | 1082 | return ret; |
f8e529ed | 1083 | } |
f8e529ed | 1084 | |
26500475 TA |
1085 | long seccomp_get_metadata(struct task_struct *task, |
1086 | unsigned long size, void __user *data) | |
1087 | { | |
1088 | long ret; | |
1089 | struct seccomp_filter *filter; | |
1090 | struct seccomp_metadata kmd = {}; | |
1091 | ||
1092 | if (!capable(CAP_SYS_ADMIN) || | |
1093 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { | |
1094 | return -EACCES; | |
1095 | } | |
1096 | ||
1097 | size = min_t(unsigned long, size, sizeof(kmd)); | |
1098 | ||
63bb0045 TA |
1099 | if (size < sizeof(kmd.filter_off)) |
1100 | return -EINVAL; | |
1101 | ||
1102 | if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off))) | |
26500475 TA |
1103 | return -EFAULT; |
1104 | ||
1105 | filter = get_nth_filter(task, kmd.filter_off); | |
1106 | if (IS_ERR(filter)) | |
1107 | return PTR_ERR(filter); | |
1108 | ||
26500475 TA |
1109 | if (filter->log) |
1110 | kmd.flags |= SECCOMP_FILTER_FLAG_LOG; | |
1111 | ||
1112 | ret = size; | |
1113 | if (copy_to_user(data, &kmd, size)) | |
1114 | ret = -EFAULT; | |
1115 | ||
1116 | __put_seccomp_filter(filter); | |
f8e529ed TA |
1117 | return ret; |
1118 | } | |
1119 | #endif | |
8e5f1ad1 TH |
1120 | |
1121 | #ifdef CONFIG_SYSCTL | |
1122 | ||
1123 | /* Human readable action names for friendly sysctl interaction */ | |
0466bdb9 | 1124 | #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process" |
fd76875c | 1125 | #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread" |
8e5f1ad1 TH |
1126 | #define SECCOMP_RET_TRAP_NAME "trap" |
1127 | #define SECCOMP_RET_ERRNO_NAME "errno" | |
1128 | #define SECCOMP_RET_TRACE_NAME "trace" | |
59f5cf44 | 1129 | #define SECCOMP_RET_LOG_NAME "log" |
8e5f1ad1 TH |
1130 | #define SECCOMP_RET_ALLOW_NAME "allow" |
1131 | ||
fd76875c | 1132 | static const char seccomp_actions_avail[] = |
0466bdb9 | 1133 | SECCOMP_RET_KILL_PROCESS_NAME " " |
fd76875c KC |
1134 | SECCOMP_RET_KILL_THREAD_NAME " " |
1135 | SECCOMP_RET_TRAP_NAME " " | |
1136 | SECCOMP_RET_ERRNO_NAME " " | |
1137 | SECCOMP_RET_TRACE_NAME " " | |
1138 | SECCOMP_RET_LOG_NAME " " | |
1139 | SECCOMP_RET_ALLOW_NAME; | |
8e5f1ad1 | 1140 | |
0ddec0fc TH |
1141 | struct seccomp_log_name { |
1142 | u32 log; | |
1143 | const char *name; | |
1144 | }; | |
1145 | ||
1146 | static const struct seccomp_log_name seccomp_log_names[] = { | |
0466bdb9 | 1147 | { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME }, |
fd76875c | 1148 | { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME }, |
0ddec0fc TH |
1149 | { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME }, |
1150 | { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME }, | |
1151 | { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME }, | |
59f5cf44 | 1152 | { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME }, |
0ddec0fc TH |
1153 | { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME }, |
1154 | { } | |
1155 | }; | |
1156 | ||
1157 | static bool seccomp_names_from_actions_logged(char *names, size_t size, | |
1158 | u32 actions_logged) | |
1159 | { | |
1160 | const struct seccomp_log_name *cur; | |
1161 | bool append_space = false; | |
1162 | ||
1163 | for (cur = seccomp_log_names; cur->name && size; cur++) { | |
1164 | ssize_t ret; | |
1165 | ||
1166 | if (!(actions_logged & cur->log)) | |
1167 | continue; | |
1168 | ||
1169 | if (append_space) { | |
1170 | ret = strscpy(names, " ", size); | |
1171 | if (ret < 0) | |
1172 | return false; | |
1173 | ||
1174 | names += ret; | |
1175 | size -= ret; | |
1176 | } else | |
1177 | append_space = true; | |
1178 | ||
1179 | ret = strscpy(names, cur->name, size); | |
1180 | if (ret < 0) | |
1181 | return false; | |
1182 | ||
1183 | names += ret; | |
1184 | size -= ret; | |
1185 | } | |
1186 | ||
1187 | return true; | |
1188 | } | |
1189 | ||
1190 | static bool seccomp_action_logged_from_name(u32 *action_logged, | |
1191 | const char *name) | |
1192 | { | |
1193 | const struct seccomp_log_name *cur; | |
1194 | ||
1195 | for (cur = seccomp_log_names; cur->name; cur++) { | |
1196 | if (!strcmp(cur->name, name)) { | |
1197 | *action_logged = cur->log; | |
1198 | return true; | |
1199 | } | |
1200 | } | |
1201 | ||
1202 | return false; | |
1203 | } | |
1204 | ||
1205 | static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names) | |
1206 | { | |
1207 | char *name; | |
1208 | ||
1209 | *actions_logged = 0; | |
1210 | while ((name = strsep(&names, " ")) && *name) { | |
1211 | u32 action_logged = 0; | |
1212 | ||
1213 | if (!seccomp_action_logged_from_name(&action_logged, name)) | |
1214 | return false; | |
1215 | ||
1216 | *actions_logged |= action_logged; | |
1217 | } | |
1218 | ||
1219 | return true; | |
1220 | } | |
1221 | ||
1222 | static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write, | |
1223 | void __user *buffer, size_t *lenp, | |
1224 | loff_t *ppos) | |
1225 | { | |
1226 | char names[sizeof(seccomp_actions_avail)]; | |
1227 | struct ctl_table table; | |
1228 | int ret; | |
1229 | ||
1230 | if (write && !capable(CAP_SYS_ADMIN)) | |
1231 | return -EPERM; | |
1232 | ||
1233 | memset(names, 0, sizeof(names)); | |
1234 | ||
1235 | if (!write) { | |
1236 | if (!seccomp_names_from_actions_logged(names, sizeof(names), | |
1237 | seccomp_actions_logged)) | |
1238 | return -EINVAL; | |
1239 | } | |
1240 | ||
1241 | table = *ro_table; | |
1242 | table.data = names; | |
1243 | table.maxlen = sizeof(names); | |
1244 | ret = proc_dostring(&table, write, buffer, lenp, ppos); | |
1245 | if (ret) | |
1246 | return ret; | |
1247 | ||
1248 | if (write) { | |
1249 | u32 actions_logged; | |
1250 | ||
1251 | if (!seccomp_actions_logged_from_names(&actions_logged, | |
1252 | table.data)) | |
1253 | return -EINVAL; | |
1254 | ||
1255 | if (actions_logged & SECCOMP_LOG_ALLOW) | |
1256 | return -EINVAL; | |
1257 | ||
1258 | seccomp_actions_logged = actions_logged; | |
1259 | } | |
1260 | ||
1261 | return 0; | |
1262 | } | |
1263 | ||
8e5f1ad1 TH |
1264 | static struct ctl_path seccomp_sysctl_path[] = { |
1265 | { .procname = "kernel", }, | |
1266 | { .procname = "seccomp", }, | |
1267 | { } | |
1268 | }; | |
1269 | ||
1270 | static struct ctl_table seccomp_sysctl_table[] = { | |
1271 | { | |
1272 | .procname = "actions_avail", | |
1273 | .data = (void *) &seccomp_actions_avail, | |
1274 | .maxlen = sizeof(seccomp_actions_avail), | |
1275 | .mode = 0444, | |
1276 | .proc_handler = proc_dostring, | |
1277 | }, | |
0ddec0fc TH |
1278 | { |
1279 | .procname = "actions_logged", | |
1280 | .mode = 0644, | |
1281 | .proc_handler = seccomp_actions_logged_handler, | |
1282 | }, | |
8e5f1ad1 TH |
1283 | { } |
1284 | }; | |
1285 | ||
1286 | static int __init seccomp_sysctl_init(void) | |
1287 | { | |
1288 | struct ctl_table_header *hdr; | |
1289 | ||
1290 | hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table); | |
1291 | if (!hdr) | |
1292 | pr_warn("seccomp: sysctl registration failed\n"); | |
1293 | else | |
1294 | kmemleak_not_leak(hdr); | |
1295 | ||
1296 | return 0; | |
1297 | } | |
1298 | ||
1299 | device_initcall(seccomp_sysctl_init) | |
1300 | ||
1301 | #endif /* CONFIG_SYSCTL */ |