]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/kernel/seccomp.c | |
3 | * | |
4 | * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com> | |
5 | * | |
e2cfabdf WD |
6 | * Copyright (C) 2012 Google, Inc. |
7 | * Will Drewry <wad@chromium.org> | |
8 | * | |
9 | * This defines a simple but solid secure-computing facility. | |
10 | * | |
11 | * Mode 1 uses a fixed list of allowed system calls. | |
12 | * Mode 2 allows user-defined system call filters in the form | |
13 | * of Berkeley Packet Filters/Linux Socket Filters. | |
1da177e4 LT |
14 | */ |
15 | ||
e2cfabdf | 16 | #include <linux/atomic.h> |
85e7bac3 | 17 | #include <linux/audit.h> |
5b101740 | 18 | #include <linux/compat.h> |
e2cfabdf WD |
19 | #include <linux/sched.h> |
20 | #include <linux/seccomp.h> | |
c8bee430 | 21 | #include <linux/slab.h> |
48dc92b9 | 22 | #include <linux/syscalls.h> |
1da177e4 | 23 | |
a4412fc9 | 24 | #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER |
e2cfabdf | 25 | #include <asm/syscall.h> |
a4412fc9 | 26 | #endif |
e2cfabdf WD |
27 | |
28 | #ifdef CONFIG_SECCOMP_FILTER | |
e2cfabdf | 29 | #include <linux/filter.h> |
c2e1f2e3 | 30 | #include <linux/pid.h> |
fb0fadf9 | 31 | #include <linux/ptrace.h> |
e2cfabdf | 32 | #include <linux/security.h> |
e2cfabdf WD |
33 | #include <linux/tracehook.h> |
34 | #include <linux/uaccess.h> | |
35 | ||
36 | /** | |
37 | * struct seccomp_filter - container for seccomp BPF programs | |
38 | * | |
39 | * @usage: reference count to manage the object lifetime. | |
40 | * get/put helpers should be used when accessing an instance | |
41 | * outside of a lifetime-guarded section. In general, this | |
42 | * is only needed for handling filters shared across tasks. | |
43 | * @prev: points to a previously installed, or inherited, filter | |
285fdfc5 | 44 | * @prog: the BPF program to evaluate |
e2cfabdf WD |
45 | * |
46 | * seccomp_filter objects are organized in a tree linked via the @prev | |
47 | * pointer. For any task, it appears to be a singly-linked list starting | |
48 | * with current->seccomp.filter, the most recently attached or inherited filter. | |
49 | * However, multiple filters may share a @prev node, by way of fork(), which | |
50 | * results in a unidirectional tree existing in memory. This is similar to | |
51 | * how namespaces work. | |
52 | * | |
53 | * seccomp_filter objects should never be modified after being attached | |
54 | * to a task_struct (other than @usage). | |
55 | */ | |
56 | struct seccomp_filter { | |
57 | atomic_t usage; | |
58 | struct seccomp_filter *prev; | |
7ae457c1 | 59 | struct bpf_prog *prog; |
e2cfabdf WD |
60 | }; |
61 | ||
62 | /* Limit any path through the tree to 256KB worth of instructions. */ | |
63 | #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter)) | |
64 | ||
bd4cf0ed | 65 | /* |
e2cfabdf WD |
66 | * Endianness is explicitly ignored and left for BPF program authors to manage |
67 | * as per the specific architecture. | |
68 | */ | |
bd4cf0ed | 69 | static void populate_seccomp_data(struct seccomp_data *sd) |
e2cfabdf | 70 | { |
bd4cf0ed AS |
71 | struct task_struct *task = current; |
72 | struct pt_regs *regs = task_pt_regs(task); | |
2eac7648 | 73 | unsigned long args[6]; |
e2cfabdf | 74 | |
bd4cf0ed | 75 | sd->nr = syscall_get_nr(task, regs); |
0b747172 | 76 | sd->arch = syscall_get_arch(); |
2eac7648 DB |
77 | syscall_get_arguments(task, regs, 0, 6, args); |
78 | sd->args[0] = args[0]; | |
79 | sd->args[1] = args[1]; | |
80 | sd->args[2] = args[2]; | |
81 | sd->args[3] = args[3]; | |
82 | sd->args[4] = args[4]; | |
83 | sd->args[5] = args[5]; | |
bd4cf0ed | 84 | sd->instruction_pointer = KSTK_EIP(task); |
e2cfabdf WD |
85 | } |
86 | ||
87 | /** | |
88 | * seccomp_check_filter - verify seccomp filter code | |
89 | * @filter: filter to verify | |
90 | * @flen: length of filter | |
91 | * | |
4df95ff4 | 92 | * Takes a previously checked filter (by bpf_check_classic) and |
e2cfabdf WD |
93 | * redirects all filter code that loads struct sk_buff data |
94 | * and related data through seccomp_bpf_load. It also | |
95 | * enforces length and alignment checking of those loads. | |
96 | * | |
97 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
98 | */ | |
99 | static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen) | |
100 | { | |
101 | int pc; | |
102 | for (pc = 0; pc < flen; pc++) { | |
103 | struct sock_filter *ftest = &filter[pc]; | |
104 | u16 code = ftest->code; | |
105 | u32 k = ftest->k; | |
106 | ||
107 | switch (code) { | |
34805931 | 108 | case BPF_LD | BPF_W | BPF_ABS: |
bd4cf0ed | 109 | ftest->code = BPF_LDX | BPF_W | BPF_ABS; |
e2cfabdf WD |
110 | /* 32-bit aligned and not out of bounds. */ |
111 | if (k >= sizeof(struct seccomp_data) || k & 3) | |
112 | return -EINVAL; | |
113 | continue; | |
34805931 | 114 | case BPF_LD | BPF_W | BPF_LEN: |
bd4cf0ed | 115 | ftest->code = BPF_LD | BPF_IMM; |
e2cfabdf WD |
116 | ftest->k = sizeof(struct seccomp_data); |
117 | continue; | |
34805931 | 118 | case BPF_LDX | BPF_W | BPF_LEN: |
bd4cf0ed | 119 | ftest->code = BPF_LDX | BPF_IMM; |
e2cfabdf WD |
120 | ftest->k = sizeof(struct seccomp_data); |
121 | continue; | |
122 | /* Explicitly include allowed calls. */ | |
34805931 DB |
123 | case BPF_RET | BPF_K: |
124 | case BPF_RET | BPF_A: | |
125 | case BPF_ALU | BPF_ADD | BPF_K: | |
126 | case BPF_ALU | BPF_ADD | BPF_X: | |
127 | case BPF_ALU | BPF_SUB | BPF_K: | |
128 | case BPF_ALU | BPF_SUB | BPF_X: | |
129 | case BPF_ALU | BPF_MUL | BPF_K: | |
130 | case BPF_ALU | BPF_MUL | BPF_X: | |
131 | case BPF_ALU | BPF_DIV | BPF_K: | |
132 | case BPF_ALU | BPF_DIV | BPF_X: | |
133 | case BPF_ALU | BPF_AND | BPF_K: | |
134 | case BPF_ALU | BPF_AND | BPF_X: | |
135 | case BPF_ALU | BPF_OR | BPF_K: | |
136 | case BPF_ALU | BPF_OR | BPF_X: | |
137 | case BPF_ALU | BPF_XOR | BPF_K: | |
138 | case BPF_ALU | BPF_XOR | BPF_X: | |
139 | case BPF_ALU | BPF_LSH | BPF_K: | |
140 | case BPF_ALU | BPF_LSH | BPF_X: | |
141 | case BPF_ALU | BPF_RSH | BPF_K: | |
142 | case BPF_ALU | BPF_RSH | BPF_X: | |
143 | case BPF_ALU | BPF_NEG: | |
144 | case BPF_LD | BPF_IMM: | |
145 | case BPF_LDX | BPF_IMM: | |
146 | case BPF_MISC | BPF_TAX: | |
147 | case BPF_MISC | BPF_TXA: | |
148 | case BPF_LD | BPF_MEM: | |
149 | case BPF_LDX | BPF_MEM: | |
150 | case BPF_ST: | |
151 | case BPF_STX: | |
152 | case BPF_JMP | BPF_JA: | |
153 | case BPF_JMP | BPF_JEQ | BPF_K: | |
154 | case BPF_JMP | BPF_JEQ | BPF_X: | |
155 | case BPF_JMP | BPF_JGE | BPF_K: | |
156 | case BPF_JMP | BPF_JGE | BPF_X: | |
157 | case BPF_JMP | BPF_JGT | BPF_K: | |
158 | case BPF_JMP | BPF_JGT | BPF_X: | |
159 | case BPF_JMP | BPF_JSET | BPF_K: | |
160 | case BPF_JMP | BPF_JSET | BPF_X: | |
e2cfabdf WD |
161 | continue; |
162 | default: | |
163 | return -EINVAL; | |
164 | } | |
165 | } | |
166 | return 0; | |
167 | } | |
168 | ||
169 | /** | |
285fdfc5 MS |
170 | * seccomp_run_filters - evaluates all seccomp filters against @sd |
171 | * @sd: optional seccomp data to be passed to filters | |
e2cfabdf WD |
172 | * |
173 | * Returns valid seccomp BPF response codes. | |
174 | */ | |
8112c4f1 | 175 | static u32 seccomp_run_filters(const struct seccomp_data *sd) |
e2cfabdf | 176 | { |
d39bd00d | 177 | struct seccomp_data sd_local; |
acf3b2c7 | 178 | u32 ret = SECCOMP_RET_ALLOW; |
8225d385 PK |
179 | /* Make sure cross-thread synced filter points somewhere sane. */ |
180 | struct seccomp_filter *f = | |
181 | lockless_dereference(current->seccomp.filter); | |
acf3b2c7 WD |
182 | |
183 | /* Ensure unexpected behavior doesn't result in failing open. */ | |
3ba2530c | 184 | if (unlikely(WARN_ON(f == NULL))) |
acf3b2c7 WD |
185 | return SECCOMP_RET_KILL; |
186 | ||
d39bd00d AL |
187 | if (!sd) { |
188 | populate_seccomp_data(&sd_local); | |
189 | sd = &sd_local; | |
190 | } | |
bd4cf0ed | 191 | |
e2cfabdf WD |
192 | /* |
193 | * All filters in the list are evaluated and the lowest BPF return | |
acf3b2c7 | 194 | * value always takes priority (ignoring the DATA). |
e2cfabdf | 195 | */ |
3ba2530c | 196 | for (; f; f = f->prev) { |
88575199 | 197 | u32 cur_ret = BPF_PROG_RUN(f->prog, sd); |
8f577cad | 198 | |
acf3b2c7 WD |
199 | if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION)) |
200 | ret = cur_ret; | |
e2cfabdf WD |
201 | } |
202 | return ret; | |
203 | } | |
1f41b450 | 204 | #endif /* CONFIG_SECCOMP_FILTER */ |
e2cfabdf | 205 | |
1f41b450 KC |
206 | static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) |
207 | { | |
69f6a34b | 208 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 209 | |
1f41b450 KC |
210 | if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) |
211 | return false; | |
212 | ||
213 | return true; | |
214 | } | |
215 | ||
3ba2530c KC |
216 | static inline void seccomp_assign_mode(struct task_struct *task, |
217 | unsigned long seccomp_mode) | |
1f41b450 | 218 | { |
69f6a34b | 219 | assert_spin_locked(&task->sighand->siglock); |
dbd95212 | 220 | |
3ba2530c KC |
221 | task->seccomp.mode = seccomp_mode; |
222 | /* | |
223 | * Make sure TIF_SECCOMP cannot be set before the mode (and | |
224 | * filter) is set. | |
225 | */ | |
226 | smp_mb__before_atomic(); | |
227 | set_tsk_thread_flag(task, TIF_SECCOMP); | |
1f41b450 KC |
228 | } |
229 | ||
230 | #ifdef CONFIG_SECCOMP_FILTER | |
c2e1f2e3 KC |
231 | /* Returns 1 if the parent is an ancestor of the child. */ |
232 | static int is_ancestor(struct seccomp_filter *parent, | |
233 | struct seccomp_filter *child) | |
234 | { | |
235 | /* NULL is the root ancestor. */ | |
236 | if (parent == NULL) | |
237 | return 1; | |
238 | for (; child; child = child->prev) | |
239 | if (child == parent) | |
240 | return 1; | |
241 | return 0; | |
242 | } | |
243 | ||
244 | /** | |
245 | * seccomp_can_sync_threads: checks if all threads can be synchronized | |
246 | * | |
247 | * Expects sighand and cred_guard_mutex locks to be held. | |
248 | * | |
249 | * Returns 0 on success, -ve on error, or the pid of a thread which was | |
250 | * either not in the correct seccomp mode or it did not have an ancestral | |
251 | * seccomp filter. | |
252 | */ | |
253 | static inline pid_t seccomp_can_sync_threads(void) | |
254 | { | |
255 | struct task_struct *thread, *caller; | |
256 | ||
257 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 258 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
259 | |
260 | /* Validate all threads being eligible for synchronization. */ | |
261 | caller = current; | |
262 | for_each_thread(caller, thread) { | |
263 | pid_t failed; | |
264 | ||
265 | /* Skip current, since it is initiating the sync. */ | |
266 | if (thread == caller) | |
267 | continue; | |
268 | ||
269 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || | |
270 | (thread->seccomp.mode == SECCOMP_MODE_FILTER && | |
271 | is_ancestor(thread->seccomp.filter, | |
272 | caller->seccomp.filter))) | |
273 | continue; | |
274 | ||
275 | /* Return the first thread that cannot be synchronized. */ | |
276 | failed = task_pid_vnr(thread); | |
277 | /* If the pid cannot be resolved, then return -ESRCH */ | |
278 | if (unlikely(WARN_ON(failed == 0))) | |
279 | failed = -ESRCH; | |
280 | return failed; | |
281 | } | |
282 | ||
283 | return 0; | |
284 | } | |
285 | ||
286 | /** | |
287 | * seccomp_sync_threads: sets all threads to use current's filter | |
288 | * | |
289 | * Expects sighand and cred_guard_mutex locks to be held, and for | |
290 | * seccomp_can_sync_threads() to have returned success already | |
291 | * without dropping the locks. | |
292 | * | |
293 | */ | |
294 | static inline void seccomp_sync_threads(void) | |
295 | { | |
296 | struct task_struct *thread, *caller; | |
297 | ||
298 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 299 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
300 | |
301 | /* Synchronize all threads. */ | |
302 | caller = current; | |
303 | for_each_thread(caller, thread) { | |
304 | /* Skip current, since it needs no changes. */ | |
305 | if (thread == caller) | |
306 | continue; | |
307 | ||
308 | /* Get a task reference for the new leaf node. */ | |
309 | get_seccomp_filter(caller); | |
310 | /* | |
311 | * Drop the task reference to the shared ancestor since | |
312 | * current's path will hold a reference. (This also | |
313 | * allows a put before the assignment.) | |
314 | */ | |
315 | put_seccomp_filter(thread); | |
316 | smp_store_release(&thread->seccomp.filter, | |
317 | caller->seccomp.filter); | |
103502a3 JH |
318 | |
319 | /* | |
320 | * Don't let an unprivileged task work around | |
321 | * the no_new_privs restriction by creating | |
322 | * a thread that sets it up, enters seccomp, | |
323 | * then dies. | |
324 | */ | |
325 | if (task_no_new_privs(caller)) | |
326 | task_set_no_new_privs(thread); | |
327 | ||
c2e1f2e3 KC |
328 | /* |
329 | * Opt the other thread into seccomp if needed. | |
330 | * As threads are considered to be trust-realm | |
331 | * equivalent (see ptrace_may_access), it is safe to | |
332 | * allow one thread to transition the other. | |
333 | */ | |
103502a3 | 334 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) |
c2e1f2e3 | 335 | seccomp_assign_mode(thread, SECCOMP_MODE_FILTER); |
c2e1f2e3 KC |
336 | } |
337 | } | |
338 | ||
e2cfabdf | 339 | /** |
c8bee430 | 340 | * seccomp_prepare_filter: Prepares a seccomp filter for use. |
e2cfabdf WD |
341 | * @fprog: BPF program to install |
342 | * | |
c8bee430 | 343 | * Returns filter on success or an ERR_PTR on failure. |
e2cfabdf | 344 | */ |
c8bee430 | 345 | static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog) |
e2cfabdf | 346 | { |
ac67eb2c DB |
347 | struct seccomp_filter *sfilter; |
348 | int ret; | |
97f2645f | 349 | const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE); |
e2cfabdf WD |
350 | |
351 | if (fprog->len == 0 || fprog->len > BPF_MAXINSNS) | |
c8bee430 | 352 | return ERR_PTR(-EINVAL); |
d9e12f42 | 353 | |
c8bee430 | 354 | BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter)); |
e2cfabdf WD |
355 | |
356 | /* | |
119ce5c8 | 357 | * Installing a seccomp filter requires that the task has |
e2cfabdf WD |
358 | * CAP_SYS_ADMIN in its namespace or be running with no_new_privs. |
359 | * This avoids scenarios where unprivileged tasks can affect the | |
360 | * behavior of privileged children. | |
361 | */ | |
1d4457f9 | 362 | if (!task_no_new_privs(current) && |
e2cfabdf WD |
363 | security_capable_noaudit(current_cred(), current_user_ns(), |
364 | CAP_SYS_ADMIN) != 0) | |
c8bee430 | 365 | return ERR_PTR(-EACCES); |
e2cfabdf | 366 | |
bd4cf0ed | 367 | /* Allocate a new seccomp_filter */ |
ac67eb2c DB |
368 | sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN); |
369 | if (!sfilter) | |
d9e12f42 | 370 | return ERR_PTR(-ENOMEM); |
ac67eb2c DB |
371 | |
372 | ret = bpf_prog_create_from_user(&sfilter->prog, fprog, | |
f8e529ed | 373 | seccomp_check_filter, save_orig); |
ac67eb2c DB |
374 | if (ret < 0) { |
375 | kfree(sfilter); | |
376 | return ERR_PTR(ret); | |
d9e12f42 | 377 | } |
bd4cf0ed | 378 | |
ac67eb2c | 379 | atomic_set(&sfilter->usage, 1); |
e2cfabdf | 380 | |
ac67eb2c | 381 | return sfilter; |
e2cfabdf WD |
382 | } |
383 | ||
384 | /** | |
c8bee430 | 385 | * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog |
e2cfabdf WD |
386 | * @user_filter: pointer to the user data containing a sock_fprog. |
387 | * | |
388 | * Returns 0 on success and non-zero otherwise. | |
389 | */ | |
c8bee430 KC |
390 | static struct seccomp_filter * |
391 | seccomp_prepare_user_filter(const char __user *user_filter) | |
e2cfabdf WD |
392 | { |
393 | struct sock_fprog fprog; | |
c8bee430 | 394 | struct seccomp_filter *filter = ERR_PTR(-EFAULT); |
e2cfabdf WD |
395 | |
396 | #ifdef CONFIG_COMPAT | |
5c38065e | 397 | if (in_compat_syscall()) { |
e2cfabdf WD |
398 | struct compat_sock_fprog fprog32; |
399 | if (copy_from_user(&fprog32, user_filter, sizeof(fprog32))) | |
400 | goto out; | |
401 | fprog.len = fprog32.len; | |
402 | fprog.filter = compat_ptr(fprog32.filter); | |
403 | } else /* falls through to the if below. */ | |
404 | #endif | |
405 | if (copy_from_user(&fprog, user_filter, sizeof(fprog))) | |
406 | goto out; | |
c8bee430 | 407 | filter = seccomp_prepare_filter(&fprog); |
e2cfabdf | 408 | out: |
c8bee430 KC |
409 | return filter; |
410 | } | |
411 | ||
412 | /** | |
413 | * seccomp_attach_filter: validate and attach filter | |
414 | * @flags: flags to change filter behavior | |
415 | * @filter: seccomp filter to add to the current process | |
416 | * | |
dbd95212 KC |
417 | * Caller must be holding current->sighand->siglock lock. |
418 | * | |
c8bee430 KC |
419 | * Returns 0 on success, -ve on error. |
420 | */ | |
421 | static long seccomp_attach_filter(unsigned int flags, | |
422 | struct seccomp_filter *filter) | |
423 | { | |
424 | unsigned long total_insns; | |
425 | struct seccomp_filter *walker; | |
426 | ||
69f6a34b | 427 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 428 | |
c8bee430 KC |
429 | /* Validate resulting filter length. */ |
430 | total_insns = filter->prog->len; | |
431 | for (walker = current->seccomp.filter; walker; walker = walker->prev) | |
432 | total_insns += walker->prog->len + 4; /* 4 instr penalty */ | |
433 | if (total_insns > MAX_INSNS_PER_PATH) | |
434 | return -ENOMEM; | |
435 | ||
c2e1f2e3 KC |
436 | /* If thread sync has been requested, check that it is possible. */ |
437 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) { | |
438 | int ret; | |
439 | ||
440 | ret = seccomp_can_sync_threads(); | |
441 | if (ret) | |
442 | return ret; | |
443 | } | |
444 | ||
c8bee430 KC |
445 | /* |
446 | * If there is an existing filter, make it the prev and don't drop its | |
447 | * task reference. | |
448 | */ | |
449 | filter->prev = current->seccomp.filter; | |
450 | current->seccomp.filter = filter; | |
451 | ||
c2e1f2e3 KC |
452 | /* Now that the new filter is in place, synchronize to all threads. */ |
453 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) | |
454 | seccomp_sync_threads(); | |
455 | ||
c8bee430 | 456 | return 0; |
e2cfabdf WD |
457 | } |
458 | ||
459 | /* get_seccomp_filter - increments the reference count of the filter on @tsk */ | |
460 | void get_seccomp_filter(struct task_struct *tsk) | |
461 | { | |
462 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
463 | if (!orig) | |
464 | return; | |
465 | /* Reference count is bounded by the number of total processes. */ | |
466 | atomic_inc(&orig->usage); | |
467 | } | |
468 | ||
c8bee430 KC |
469 | static inline void seccomp_filter_free(struct seccomp_filter *filter) |
470 | { | |
471 | if (filter) { | |
bab18991 | 472 | bpf_prog_destroy(filter->prog); |
c8bee430 KC |
473 | kfree(filter); |
474 | } | |
475 | } | |
476 | ||
e2cfabdf WD |
477 | /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */ |
478 | void put_seccomp_filter(struct task_struct *tsk) | |
479 | { | |
480 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
481 | /* Clean up single-reference branches iteratively. */ | |
482 | while (orig && atomic_dec_and_test(&orig->usage)) { | |
483 | struct seccomp_filter *freeme = orig; | |
484 | orig = orig->prev; | |
c8bee430 | 485 | seccomp_filter_free(freeme); |
e2cfabdf WD |
486 | } |
487 | } | |
bb6ea430 WD |
488 | |
489 | /** | |
490 | * seccomp_send_sigsys - signals the task to allow in-process syscall emulation | |
491 | * @syscall: syscall number to send to userland | |
492 | * @reason: filter-supplied reason code to send to userland (via si_errno) | |
493 | * | |
494 | * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info. | |
495 | */ | |
496 | static void seccomp_send_sigsys(int syscall, int reason) | |
497 | { | |
498 | struct siginfo info; | |
499 | memset(&info, 0, sizeof(info)); | |
500 | info.si_signo = SIGSYS; | |
501 | info.si_code = SYS_SECCOMP; | |
502 | info.si_call_addr = (void __user *)KSTK_EIP(current); | |
503 | info.si_errno = reason; | |
5e937a9a | 504 | info.si_arch = syscall_get_arch(); |
bb6ea430 WD |
505 | info.si_syscall = syscall; |
506 | force_sig_info(SIGSYS, &info, current); | |
507 | } | |
e2cfabdf | 508 | #endif /* CONFIG_SECCOMP_FILTER */ |
1da177e4 LT |
509 | |
510 | /* | |
511 | * Secure computing mode 1 allows only read/write/exit/sigreturn. | |
512 | * To be fully secure this must be combined with rlimit | |
513 | * to limit the stack allocations too. | |
514 | */ | |
cb4253aa | 515 | static const int mode1_syscalls[] = { |
1da177e4 LT |
516 | __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, |
517 | 0, /* null terminated */ | |
518 | }; | |
519 | ||
a4412fc9 | 520 | static void __secure_computing_strict(int this_syscall) |
1da177e4 | 521 | { |
cb4253aa | 522 | const int *syscall_whitelist = mode1_syscalls; |
a4412fc9 | 523 | #ifdef CONFIG_COMPAT |
5c38065e | 524 | if (in_compat_syscall()) |
c983f0e8 | 525 | syscall_whitelist = get_compat_mode1_syscalls(); |
a4412fc9 AL |
526 | #endif |
527 | do { | |
528 | if (*syscall_whitelist == this_syscall) | |
529 | return; | |
530 | } while (*++syscall_whitelist); | |
531 | ||
532 | #ifdef SECCOMP_DEBUG | |
533 | dump_stack(); | |
534 | #endif | |
535 | audit_seccomp(this_syscall, SIGKILL, SECCOMP_RET_KILL); | |
536 | do_exit(SIGKILL); | |
537 | } | |
538 | ||
539 | #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER | |
540 | void secure_computing_strict(int this_syscall) | |
541 | { | |
542 | int mode = current->seccomp.mode; | |
543 | ||
97f2645f | 544 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 TA |
545 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
546 | return; | |
547 | ||
221272f9 | 548 | if (mode == SECCOMP_MODE_DISABLED) |
a4412fc9 AL |
549 | return; |
550 | else if (mode == SECCOMP_MODE_STRICT) | |
551 | __secure_computing_strict(this_syscall); | |
552 | else | |
553 | BUG(); | |
554 | } | |
555 | #else | |
13aa72f0 AL |
556 | |
557 | #ifdef CONFIG_SECCOMP_FILTER | |
ce6526e8 KC |
558 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
559 | const bool recheck_after_trace) | |
13aa72f0 AL |
560 | { |
561 | u32 filter_ret, action; | |
562 | int data; | |
1da177e4 | 563 | |
3ba2530c KC |
564 | /* |
565 | * Make sure that any changes to mode from another thread have | |
566 | * been seen after TIF_SECCOMP was seen. | |
567 | */ | |
568 | rmb(); | |
569 | ||
d39bd00d | 570 | filter_ret = seccomp_run_filters(sd); |
13aa72f0 AL |
571 | data = filter_ret & SECCOMP_RET_DATA; |
572 | action = filter_ret & SECCOMP_RET_ACTION; | |
573 | ||
574 | switch (action) { | |
575 | case SECCOMP_RET_ERRNO: | |
580c57f1 KC |
576 | /* Set low-order bits as an errno, capped at MAX_ERRNO. */ |
577 | if (data > MAX_ERRNO) | |
578 | data = MAX_ERRNO; | |
d39bd00d | 579 | syscall_set_return_value(current, task_pt_regs(current), |
13aa72f0 AL |
580 | -data, 0); |
581 | goto skip; | |
582 | ||
583 | case SECCOMP_RET_TRAP: | |
584 | /* Show the handler the original registers. */ | |
d39bd00d | 585 | syscall_rollback(current, task_pt_regs(current)); |
13aa72f0 AL |
586 | /* Let the filter pass back 16 bits of data. */ |
587 | seccomp_send_sigsys(this_syscall, data); | |
588 | goto skip; | |
589 | ||
590 | case SECCOMP_RET_TRACE: | |
ce6526e8 KC |
591 | /* We've been put in this state by the ptracer already. */ |
592 | if (recheck_after_trace) | |
593 | return 0; | |
594 | ||
8112c4f1 KC |
595 | /* ENOSYS these calls if there is no tracer attached. */ |
596 | if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { | |
597 | syscall_set_return_value(current, | |
598 | task_pt_regs(current), | |
599 | -ENOSYS, 0); | |
600 | goto skip; | |
601 | } | |
602 | ||
603 | /* Allow the BPF to provide the event message */ | |
604 | ptrace_event(PTRACE_EVENT_SECCOMP, data); | |
605 | /* | |
606 | * The delivery of a fatal signal during event | |
485a252a KC |
607 | * notification may silently skip tracer notification, |
608 | * which could leave us with a potentially unmodified | |
609 | * syscall that the tracer would have liked to have | |
610 | * changed. Since the process is about to die, we just | |
611 | * force the syscall to be skipped and let the signal | |
612 | * kill the process and correctly handle any tracer exit | |
613 | * notifications. | |
8112c4f1 KC |
614 | */ |
615 | if (fatal_signal_pending(current)) | |
485a252a | 616 | goto skip; |
8112c4f1 KC |
617 | /* Check if the tracer forced the syscall to be skipped. */ |
618 | this_syscall = syscall_get_nr(current, task_pt_regs(current)); | |
619 | if (this_syscall < 0) | |
620 | goto skip; | |
621 | ||
ce6526e8 KC |
622 | /* |
623 | * Recheck the syscall, since it may have changed. This | |
624 | * intentionally uses a NULL struct seccomp_data to force | |
625 | * a reload of all registers. This does not goto skip since | |
626 | * a skip would have already been reported. | |
627 | */ | |
628 | if (__seccomp_filter(this_syscall, NULL, true)) | |
629 | return -1; | |
630 | ||
8112c4f1 | 631 | return 0; |
13aa72f0 AL |
632 | |
633 | case SECCOMP_RET_ALLOW: | |
8112c4f1 | 634 | return 0; |
13aa72f0 AL |
635 | |
636 | case SECCOMP_RET_KILL: | |
637 | default: | |
638 | audit_seccomp(this_syscall, SIGSYS, action); | |
639 | do_exit(SIGSYS); | |
640 | } | |
641 | ||
642 | unreachable(); | |
643 | ||
644 | skip: | |
645 | audit_seccomp(this_syscall, 0, action); | |
8112c4f1 KC |
646 | return -1; |
647 | } | |
648 | #else | |
ce6526e8 KC |
649 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
650 | const bool recheck_after_trace) | |
8112c4f1 KC |
651 | { |
652 | BUG(); | |
13aa72f0 | 653 | } |
1da177e4 | 654 | #endif |
13aa72f0 | 655 | |
8112c4f1 | 656 | int __secure_computing(const struct seccomp_data *sd) |
13aa72f0 AL |
657 | { |
658 | int mode = current->seccomp.mode; | |
8112c4f1 | 659 | int this_syscall; |
13aa72f0 | 660 | |
97f2645f | 661 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 | 662 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
8112c4f1 KC |
663 | return 0; |
664 | ||
665 | this_syscall = sd ? sd->nr : | |
666 | syscall_get_nr(current, task_pt_regs(current)); | |
13c4a901 | 667 | |
13aa72f0 | 668 | switch (mode) { |
e2cfabdf | 669 | case SECCOMP_MODE_STRICT: |
13aa72f0 | 670 | __secure_computing_strict(this_syscall); /* may call do_exit */ |
8112c4f1 | 671 | return 0; |
13aa72f0 | 672 | case SECCOMP_MODE_FILTER: |
ce6526e8 | 673 | return __seccomp_filter(this_syscall, sd, false); |
1da177e4 LT |
674 | default: |
675 | BUG(); | |
676 | } | |
13aa72f0 | 677 | } |
a4412fc9 | 678 | #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ |
1d9d02fe AA |
679 | |
680 | long prctl_get_seccomp(void) | |
681 | { | |
682 | return current->seccomp.mode; | |
683 | } | |
684 | ||
e2cfabdf | 685 | /** |
3b23dd12 | 686 | * seccomp_set_mode_strict: internal function for setting strict seccomp |
e2cfabdf WD |
687 | * |
688 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
689 | * | |
690 | * Returns 0 on success or -EINVAL on failure. | |
691 | */ | |
3b23dd12 | 692 | static long seccomp_set_mode_strict(void) |
1d9d02fe | 693 | { |
3b23dd12 | 694 | const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; |
e2cfabdf | 695 | long ret = -EINVAL; |
1d9d02fe | 696 | |
dbd95212 KC |
697 | spin_lock_irq(¤t->sighand->siglock); |
698 | ||
1f41b450 | 699 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1d9d02fe AA |
700 | goto out; |
701 | ||
cf99abac | 702 | #ifdef TIF_NOTSC |
3b23dd12 | 703 | disable_TSC(); |
cf99abac | 704 | #endif |
3ba2530c | 705 | seccomp_assign_mode(current, seccomp_mode); |
3b23dd12 KC |
706 | ret = 0; |
707 | ||
708 | out: | |
dbd95212 | 709 | spin_unlock_irq(¤t->sighand->siglock); |
3b23dd12 KC |
710 | |
711 | return ret; | |
712 | } | |
713 | ||
e2cfabdf | 714 | #ifdef CONFIG_SECCOMP_FILTER |
3b23dd12 KC |
715 | /** |
716 | * seccomp_set_mode_filter: internal function for setting seccomp filter | |
48dc92b9 | 717 | * @flags: flags to change filter behavior |
3b23dd12 KC |
718 | * @filter: struct sock_fprog containing filter |
719 | * | |
720 | * This function may be called repeatedly to install additional filters. | |
721 | * Every filter successfully installed will be evaluated (in reverse order) | |
722 | * for each system call the task makes. | |
723 | * | |
724 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
725 | * | |
726 | * Returns 0 on success or -EINVAL on failure. | |
727 | */ | |
48dc92b9 KC |
728 | static long seccomp_set_mode_filter(unsigned int flags, |
729 | const char __user *filter) | |
3b23dd12 KC |
730 | { |
731 | const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; | |
c8bee430 | 732 | struct seccomp_filter *prepared = NULL; |
3b23dd12 KC |
733 | long ret = -EINVAL; |
734 | ||
48dc92b9 | 735 | /* Validate flags. */ |
c2e1f2e3 | 736 | if (flags & ~SECCOMP_FILTER_FLAG_MASK) |
dbd95212 | 737 | return -EINVAL; |
48dc92b9 | 738 | |
c8bee430 KC |
739 | /* Prepare the new filter before holding any locks. */ |
740 | prepared = seccomp_prepare_user_filter(filter); | |
741 | if (IS_ERR(prepared)) | |
742 | return PTR_ERR(prepared); | |
743 | ||
c2e1f2e3 KC |
744 | /* |
745 | * Make sure we cannot change seccomp or nnp state via TSYNC | |
746 | * while another thread is in the middle of calling exec. | |
747 | */ | |
748 | if (flags & SECCOMP_FILTER_FLAG_TSYNC && | |
749 | mutex_lock_killable(¤t->signal->cred_guard_mutex)) | |
750 | goto out_free; | |
751 | ||
dbd95212 KC |
752 | spin_lock_irq(¤t->sighand->siglock); |
753 | ||
3b23dd12 KC |
754 | if (!seccomp_may_assign_mode(seccomp_mode)) |
755 | goto out; | |
756 | ||
c8bee430 | 757 | ret = seccomp_attach_filter(flags, prepared); |
3b23dd12 | 758 | if (ret) |
e2cfabdf | 759 | goto out; |
c8bee430 KC |
760 | /* Do not free the successfully attached filter. */ |
761 | prepared = NULL; | |
1d9d02fe | 762 | |
3ba2530c | 763 | seccomp_assign_mode(current, seccomp_mode); |
e2cfabdf | 764 | out: |
dbd95212 | 765 | spin_unlock_irq(¤t->sighand->siglock); |
c2e1f2e3 KC |
766 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
767 | mutex_unlock(¤t->signal->cred_guard_mutex); | |
768 | out_free: | |
c8bee430 | 769 | seccomp_filter_free(prepared); |
1d9d02fe AA |
770 | return ret; |
771 | } | |
3b23dd12 | 772 | #else |
48dc92b9 KC |
773 | static inline long seccomp_set_mode_filter(unsigned int flags, |
774 | const char __user *filter) | |
3b23dd12 KC |
775 | { |
776 | return -EINVAL; | |
777 | } | |
778 | #endif | |
d78ab02c | 779 | |
48dc92b9 KC |
780 | /* Common entry point for both prctl and syscall. */ |
781 | static long do_seccomp(unsigned int op, unsigned int flags, | |
782 | const char __user *uargs) | |
783 | { | |
784 | switch (op) { | |
785 | case SECCOMP_SET_MODE_STRICT: | |
786 | if (flags != 0 || uargs != NULL) | |
787 | return -EINVAL; | |
788 | return seccomp_set_mode_strict(); | |
789 | case SECCOMP_SET_MODE_FILTER: | |
790 | return seccomp_set_mode_filter(flags, uargs); | |
791 | default: | |
792 | return -EINVAL; | |
793 | } | |
794 | } | |
795 | ||
796 | SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, | |
797 | const char __user *, uargs) | |
798 | { | |
799 | return do_seccomp(op, flags, uargs); | |
800 | } | |
801 | ||
d78ab02c KC |
802 | /** |
803 | * prctl_set_seccomp: configures current->seccomp.mode | |
804 | * @seccomp_mode: requested mode to use | |
805 | * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER | |
806 | * | |
807 | * Returns 0 on success or -EINVAL on failure. | |
808 | */ | |
809 | long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter) | |
810 | { | |
48dc92b9 KC |
811 | unsigned int op; |
812 | char __user *uargs; | |
813 | ||
3b23dd12 KC |
814 | switch (seccomp_mode) { |
815 | case SECCOMP_MODE_STRICT: | |
48dc92b9 KC |
816 | op = SECCOMP_SET_MODE_STRICT; |
817 | /* | |
818 | * Setting strict mode through prctl always ignored filter, | |
819 | * so make sure it is always NULL here to pass the internal | |
820 | * check in do_seccomp(). | |
821 | */ | |
822 | uargs = NULL; | |
823 | break; | |
3b23dd12 | 824 | case SECCOMP_MODE_FILTER: |
48dc92b9 KC |
825 | op = SECCOMP_SET_MODE_FILTER; |
826 | uargs = filter; | |
827 | break; | |
3b23dd12 KC |
828 | default: |
829 | return -EINVAL; | |
830 | } | |
48dc92b9 KC |
831 | |
832 | /* prctl interface doesn't have flags, so they are always zero. */ | |
833 | return do_seccomp(op, 0, uargs); | |
d78ab02c | 834 | } |
f8e529ed TA |
835 | |
836 | #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE) | |
837 | long seccomp_get_filter(struct task_struct *task, unsigned long filter_off, | |
838 | void __user *data) | |
839 | { | |
840 | struct seccomp_filter *filter; | |
841 | struct sock_fprog_kern *fprog; | |
842 | long ret; | |
843 | unsigned long count = 0; | |
844 | ||
845 | if (!capable(CAP_SYS_ADMIN) || | |
846 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { | |
847 | return -EACCES; | |
848 | } | |
849 | ||
850 | spin_lock_irq(&task->sighand->siglock); | |
851 | if (task->seccomp.mode != SECCOMP_MODE_FILTER) { | |
852 | ret = -EINVAL; | |
853 | goto out; | |
854 | } | |
855 | ||
856 | filter = task->seccomp.filter; | |
857 | while (filter) { | |
858 | filter = filter->prev; | |
859 | count++; | |
860 | } | |
861 | ||
862 | if (filter_off >= count) { | |
863 | ret = -ENOENT; | |
864 | goto out; | |
865 | } | |
866 | count -= filter_off; | |
867 | ||
868 | filter = task->seccomp.filter; | |
869 | while (filter && count > 1) { | |
870 | filter = filter->prev; | |
871 | count--; | |
872 | } | |
873 | ||
874 | if (WARN_ON(count != 1 || !filter)) { | |
875 | /* The filter tree shouldn't shrink while we're using it. */ | |
876 | ret = -ENOENT; | |
877 | goto out; | |
878 | } | |
879 | ||
880 | fprog = filter->prog->orig_prog; | |
881 | if (!fprog) { | |
470bf1f2 | 882 | /* This must be a new non-cBPF filter, since we save |
f8e529ed TA |
883 | * every cBPF filter's orig_prog above when |
884 | * CONFIG_CHECKPOINT_RESTORE is enabled. | |
885 | */ | |
886 | ret = -EMEDIUMTYPE; | |
887 | goto out; | |
888 | } | |
889 | ||
890 | ret = fprog->len; | |
891 | if (!data) | |
892 | goto out; | |
893 | ||
894 | get_seccomp_filter(task); | |
895 | spin_unlock_irq(&task->sighand->siglock); | |
896 | ||
897 | if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog))) | |
898 | ret = -EFAULT; | |
899 | ||
900 | put_seccomp_filter(task); | |
901 | return ret; | |
902 | ||
903 | out: | |
904 | spin_unlock_irq(&task->sighand->siglock); | |
905 | return ret; | |
906 | } | |
907 | #endif |