]>
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> |
0a112f10 KC |
19 | #include <linux/nospec.h> |
20 | #include <linux/prctl.h> | |
e2cfabdf WD |
21 | #include <linux/sched.h> |
22 | #include <linux/seccomp.h> | |
c8bee430 | 23 | #include <linux/slab.h> |
48dc92b9 | 24 | #include <linux/syscalls.h> |
1da177e4 | 25 | |
a4412fc9 | 26 | #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER |
e2cfabdf | 27 | #include <asm/syscall.h> |
a4412fc9 | 28 | #endif |
e2cfabdf WD |
29 | |
30 | #ifdef CONFIG_SECCOMP_FILTER | |
e2cfabdf | 31 | #include <linux/filter.h> |
c2e1f2e3 | 32 | #include <linux/pid.h> |
fb0fadf9 | 33 | #include <linux/ptrace.h> |
e2cfabdf | 34 | #include <linux/security.h> |
e2cfabdf WD |
35 | #include <linux/tracehook.h> |
36 | #include <linux/uaccess.h> | |
37 | ||
38 | /** | |
39 | * struct seccomp_filter - container for seccomp BPF programs | |
40 | * | |
41 | * @usage: reference count to manage the object lifetime. | |
42 | * get/put helpers should be used when accessing an instance | |
43 | * outside of a lifetime-guarded section. In general, this | |
44 | * is only needed for handling filters shared across tasks. | |
45 | * @prev: points to a previously installed, or inherited, filter | |
46 | * @len: the number of instructions in the program | |
119ce5c8 | 47 | * @insnsi: the BPF program instructions to evaluate |
e2cfabdf WD |
48 | * |
49 | * seccomp_filter objects are organized in a tree linked via the @prev | |
50 | * pointer. For any task, it appears to be a singly-linked list starting | |
51 | * with current->seccomp.filter, the most recently attached or inherited filter. | |
52 | * However, multiple filters may share a @prev node, by way of fork(), which | |
53 | * results in a unidirectional tree existing in memory. This is similar to | |
54 | * how namespaces work. | |
55 | * | |
56 | * seccomp_filter objects should never be modified after being attached | |
57 | * to a task_struct (other than @usage). | |
58 | */ | |
59 | struct seccomp_filter { | |
60 | atomic_t usage; | |
61 | struct seccomp_filter *prev; | |
7ae457c1 | 62 | struct bpf_prog *prog; |
e2cfabdf WD |
63 | }; |
64 | ||
65 | /* Limit any path through the tree to 256KB worth of instructions. */ | |
66 | #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter)) | |
67 | ||
bd4cf0ed | 68 | /* |
e2cfabdf WD |
69 | * Endianness is explicitly ignored and left for BPF program authors to manage |
70 | * as per the specific architecture. | |
71 | */ | |
bd4cf0ed | 72 | static void populate_seccomp_data(struct seccomp_data *sd) |
e2cfabdf | 73 | { |
bd4cf0ed AS |
74 | struct task_struct *task = current; |
75 | struct pt_regs *regs = task_pt_regs(task); | |
2eac7648 | 76 | unsigned long args[6]; |
e2cfabdf | 77 | |
bd4cf0ed | 78 | sd->nr = syscall_get_nr(task, regs); |
0b747172 | 79 | sd->arch = syscall_get_arch(); |
2eac7648 DB |
80 | syscall_get_arguments(task, regs, 0, 6, args); |
81 | sd->args[0] = args[0]; | |
82 | sd->args[1] = args[1]; | |
83 | sd->args[2] = args[2]; | |
84 | sd->args[3] = args[3]; | |
85 | sd->args[4] = args[4]; | |
86 | sd->args[5] = args[5]; | |
bd4cf0ed | 87 | sd->instruction_pointer = KSTK_EIP(task); |
e2cfabdf WD |
88 | } |
89 | ||
90 | /** | |
91 | * seccomp_check_filter - verify seccomp filter code | |
92 | * @filter: filter to verify | |
93 | * @flen: length of filter | |
94 | * | |
4df95ff4 | 95 | * Takes a previously checked filter (by bpf_check_classic) and |
e2cfabdf WD |
96 | * redirects all filter code that loads struct sk_buff data |
97 | * and related data through seccomp_bpf_load. It also | |
98 | * enforces length and alignment checking of those loads. | |
99 | * | |
100 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
101 | */ | |
102 | static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen) | |
103 | { | |
104 | int pc; | |
105 | for (pc = 0; pc < flen; pc++) { | |
106 | struct sock_filter *ftest = &filter[pc]; | |
107 | u16 code = ftest->code; | |
108 | u32 k = ftest->k; | |
109 | ||
110 | switch (code) { | |
34805931 | 111 | case BPF_LD | BPF_W | BPF_ABS: |
bd4cf0ed | 112 | ftest->code = BPF_LDX | BPF_W | BPF_ABS; |
e2cfabdf WD |
113 | /* 32-bit aligned and not out of bounds. */ |
114 | if (k >= sizeof(struct seccomp_data) || k & 3) | |
115 | return -EINVAL; | |
116 | continue; | |
34805931 | 117 | case BPF_LD | BPF_W | BPF_LEN: |
bd4cf0ed | 118 | ftest->code = BPF_LD | BPF_IMM; |
e2cfabdf WD |
119 | ftest->k = sizeof(struct seccomp_data); |
120 | continue; | |
34805931 | 121 | case BPF_LDX | BPF_W | BPF_LEN: |
bd4cf0ed | 122 | ftest->code = BPF_LDX | BPF_IMM; |
e2cfabdf WD |
123 | ftest->k = sizeof(struct seccomp_data); |
124 | continue; | |
125 | /* Explicitly include allowed calls. */ | |
34805931 DB |
126 | case BPF_RET | BPF_K: |
127 | case BPF_RET | BPF_A: | |
128 | case BPF_ALU | BPF_ADD | BPF_K: | |
129 | case BPF_ALU | BPF_ADD | BPF_X: | |
130 | case BPF_ALU | BPF_SUB | BPF_K: | |
131 | case BPF_ALU | BPF_SUB | BPF_X: | |
132 | case BPF_ALU | BPF_MUL | BPF_K: | |
133 | case BPF_ALU | BPF_MUL | BPF_X: | |
134 | case BPF_ALU | BPF_DIV | BPF_K: | |
135 | case BPF_ALU | BPF_DIV | BPF_X: | |
136 | case BPF_ALU | BPF_AND | BPF_K: | |
137 | case BPF_ALU | BPF_AND | BPF_X: | |
138 | case BPF_ALU | BPF_OR | BPF_K: | |
139 | case BPF_ALU | BPF_OR | BPF_X: | |
140 | case BPF_ALU | BPF_XOR | BPF_K: | |
141 | case BPF_ALU | BPF_XOR | BPF_X: | |
142 | case BPF_ALU | BPF_LSH | BPF_K: | |
143 | case BPF_ALU | BPF_LSH | BPF_X: | |
144 | case BPF_ALU | BPF_RSH | BPF_K: | |
145 | case BPF_ALU | BPF_RSH | BPF_X: | |
146 | case BPF_ALU | BPF_NEG: | |
147 | case BPF_LD | BPF_IMM: | |
148 | case BPF_LDX | BPF_IMM: | |
149 | case BPF_MISC | BPF_TAX: | |
150 | case BPF_MISC | BPF_TXA: | |
151 | case BPF_LD | BPF_MEM: | |
152 | case BPF_LDX | BPF_MEM: | |
153 | case BPF_ST: | |
154 | case BPF_STX: | |
155 | case BPF_JMP | BPF_JA: | |
156 | case BPF_JMP | BPF_JEQ | BPF_K: | |
157 | case BPF_JMP | BPF_JEQ | BPF_X: | |
158 | case BPF_JMP | BPF_JGE | BPF_K: | |
159 | case BPF_JMP | BPF_JGE | BPF_X: | |
160 | case BPF_JMP | BPF_JGT | BPF_K: | |
161 | case BPF_JMP | BPF_JGT | BPF_X: | |
162 | case BPF_JMP | BPF_JSET | BPF_K: | |
163 | case BPF_JMP | BPF_JSET | BPF_X: | |
e2cfabdf WD |
164 | continue; |
165 | default: | |
166 | return -EINVAL; | |
167 | } | |
168 | } | |
169 | return 0; | |
170 | } | |
171 | ||
172 | /** | |
173 | * seccomp_run_filters - evaluates all seccomp filters against @syscall | |
174 | * @syscall: number of the current system call | |
175 | * | |
176 | * Returns valid seccomp BPF response codes. | |
177 | */ | |
8112c4f1 | 178 | static u32 seccomp_run_filters(const struct seccomp_data *sd) |
e2cfabdf | 179 | { |
d39bd00d | 180 | struct seccomp_data sd_local; |
acf3b2c7 | 181 | u32 ret = SECCOMP_RET_ALLOW; |
8225d385 PK |
182 | /* Make sure cross-thread synced filter points somewhere sane. */ |
183 | struct seccomp_filter *f = | |
184 | lockless_dereference(current->seccomp.filter); | |
acf3b2c7 WD |
185 | |
186 | /* Ensure unexpected behavior doesn't result in failing open. */ | |
3ba2530c | 187 | if (unlikely(WARN_ON(f == NULL))) |
acf3b2c7 WD |
188 | return SECCOMP_RET_KILL; |
189 | ||
d39bd00d AL |
190 | if (!sd) { |
191 | populate_seccomp_data(&sd_local); | |
192 | sd = &sd_local; | |
193 | } | |
bd4cf0ed | 194 | |
e2cfabdf WD |
195 | /* |
196 | * All filters in the list are evaluated and the lowest BPF return | |
acf3b2c7 | 197 | * value always takes priority (ignoring the DATA). |
e2cfabdf | 198 | */ |
3ba2530c | 199 | for (; f; f = f->prev) { |
d39bd00d | 200 | u32 cur_ret = BPF_PROG_RUN(f->prog, (void *)sd); |
8f577cad | 201 | |
acf3b2c7 WD |
202 | if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION)) |
203 | ret = cur_ret; | |
e2cfabdf WD |
204 | } |
205 | return ret; | |
206 | } | |
1f41b450 | 207 | #endif /* CONFIG_SECCOMP_FILTER */ |
e2cfabdf | 208 | |
1f41b450 KC |
209 | static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) |
210 | { | |
69f6a34b | 211 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 212 | |
1f41b450 KC |
213 | if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) |
214 | return false; | |
215 | ||
216 | return true; | |
217 | } | |
218 | ||
094c2767 | 219 | void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { } |
0a112f10 | 220 | |
3ba2530c | 221 | static inline void seccomp_assign_mode(struct task_struct *task, |
ab677c2a KC |
222 | unsigned long seccomp_mode, |
223 | unsigned long flags) | |
1f41b450 | 224 | { |
69f6a34b | 225 | assert_spin_locked(&task->sighand->siglock); |
dbd95212 | 226 | |
3ba2530c KC |
227 | task->seccomp.mode = seccomp_mode; |
228 | /* | |
229 | * Make sure TIF_SECCOMP cannot be set before the mode (and | |
230 | * filter) is set. | |
231 | */ | |
232 | smp_mb__before_atomic(); | |
ab677c2a KC |
233 | /* Assume default seccomp processes want spec flaw mitigation. */ |
234 | if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0) | |
094c2767 | 235 | arch_seccomp_spec_mitigate(task); |
3ba2530c | 236 | set_tsk_thread_flag(task, TIF_SECCOMP); |
1f41b450 KC |
237 | } |
238 | ||
239 | #ifdef CONFIG_SECCOMP_FILTER | |
c2e1f2e3 KC |
240 | /* Returns 1 if the parent is an ancestor of the child. */ |
241 | static int is_ancestor(struct seccomp_filter *parent, | |
242 | struct seccomp_filter *child) | |
243 | { | |
244 | /* NULL is the root ancestor. */ | |
245 | if (parent == NULL) | |
246 | return 1; | |
247 | for (; child; child = child->prev) | |
248 | if (child == parent) | |
249 | return 1; | |
250 | return 0; | |
251 | } | |
252 | ||
253 | /** | |
254 | * seccomp_can_sync_threads: checks if all threads can be synchronized | |
255 | * | |
256 | * Expects sighand and cred_guard_mutex locks to be held. | |
257 | * | |
258 | * Returns 0 on success, -ve on error, or the pid of a thread which was | |
259 | * either not in the correct seccomp mode or it did not have an ancestral | |
260 | * seccomp filter. | |
261 | */ | |
262 | static inline pid_t seccomp_can_sync_threads(void) | |
263 | { | |
264 | struct task_struct *thread, *caller; | |
265 | ||
266 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 267 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
268 | |
269 | /* Validate all threads being eligible for synchronization. */ | |
270 | caller = current; | |
271 | for_each_thread(caller, thread) { | |
272 | pid_t failed; | |
273 | ||
274 | /* Skip current, since it is initiating the sync. */ | |
275 | if (thread == caller) | |
276 | continue; | |
277 | ||
278 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || | |
279 | (thread->seccomp.mode == SECCOMP_MODE_FILTER && | |
280 | is_ancestor(thread->seccomp.filter, | |
281 | caller->seccomp.filter))) | |
282 | continue; | |
283 | ||
284 | /* Return the first thread that cannot be synchronized. */ | |
285 | failed = task_pid_vnr(thread); | |
286 | /* If the pid cannot be resolved, then return -ESRCH */ | |
287 | if (unlikely(WARN_ON(failed == 0))) | |
288 | failed = -ESRCH; | |
289 | return failed; | |
290 | } | |
291 | ||
292 | return 0; | |
293 | } | |
294 | ||
295 | /** | |
296 | * seccomp_sync_threads: sets all threads to use current's filter | |
297 | * | |
298 | * Expects sighand and cred_guard_mutex locks to be held, and for | |
299 | * seccomp_can_sync_threads() to have returned success already | |
300 | * without dropping the locks. | |
301 | * | |
302 | */ | |
ab677c2a | 303 | static inline void seccomp_sync_threads(unsigned long flags) |
c2e1f2e3 KC |
304 | { |
305 | struct task_struct *thread, *caller; | |
306 | ||
307 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 308 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
309 | |
310 | /* Synchronize all threads. */ | |
311 | caller = current; | |
312 | for_each_thread(caller, thread) { | |
313 | /* Skip current, since it needs no changes. */ | |
314 | if (thread == caller) | |
315 | continue; | |
316 | ||
317 | /* Get a task reference for the new leaf node. */ | |
318 | get_seccomp_filter(caller); | |
319 | /* | |
320 | * Drop the task reference to the shared ancestor since | |
321 | * current's path will hold a reference. (This also | |
322 | * allows a put before the assignment.) | |
323 | */ | |
324 | put_seccomp_filter(thread); | |
325 | smp_store_release(&thread->seccomp.filter, | |
326 | caller->seccomp.filter); | |
103502a3 JH |
327 | |
328 | /* | |
329 | * Don't let an unprivileged task work around | |
330 | * the no_new_privs restriction by creating | |
331 | * a thread that sets it up, enters seccomp, | |
332 | * then dies. | |
333 | */ | |
334 | if (task_no_new_privs(caller)) | |
335 | task_set_no_new_privs(thread); | |
336 | ||
c2e1f2e3 KC |
337 | /* |
338 | * Opt the other thread into seccomp if needed. | |
339 | * As threads are considered to be trust-realm | |
340 | * equivalent (see ptrace_may_access), it is safe to | |
341 | * allow one thread to transition the other. | |
342 | */ | |
103502a3 | 343 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) |
ab677c2a KC |
344 | seccomp_assign_mode(thread, SECCOMP_MODE_FILTER, |
345 | flags); | |
c2e1f2e3 KC |
346 | } |
347 | } | |
348 | ||
e2cfabdf | 349 | /** |
c8bee430 | 350 | * seccomp_prepare_filter: Prepares a seccomp filter for use. |
e2cfabdf WD |
351 | * @fprog: BPF program to install |
352 | * | |
c8bee430 | 353 | * Returns filter on success or an ERR_PTR on failure. |
e2cfabdf | 354 | */ |
c8bee430 | 355 | static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog) |
e2cfabdf | 356 | { |
ac67eb2c DB |
357 | struct seccomp_filter *sfilter; |
358 | int ret; | |
97f2645f | 359 | const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE); |
e2cfabdf WD |
360 | |
361 | if (fprog->len == 0 || fprog->len > BPF_MAXINSNS) | |
c8bee430 | 362 | return ERR_PTR(-EINVAL); |
d9e12f42 | 363 | |
c8bee430 | 364 | BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter)); |
e2cfabdf WD |
365 | |
366 | /* | |
119ce5c8 | 367 | * Installing a seccomp filter requires that the task has |
e2cfabdf WD |
368 | * CAP_SYS_ADMIN in its namespace or be running with no_new_privs. |
369 | * This avoids scenarios where unprivileged tasks can affect the | |
370 | * behavior of privileged children. | |
371 | */ | |
1d4457f9 | 372 | if (!task_no_new_privs(current) && |
e2cfabdf WD |
373 | security_capable_noaudit(current_cred(), current_user_ns(), |
374 | CAP_SYS_ADMIN) != 0) | |
c8bee430 | 375 | return ERR_PTR(-EACCES); |
e2cfabdf | 376 | |
bd4cf0ed | 377 | /* Allocate a new seccomp_filter */ |
ac67eb2c DB |
378 | sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN); |
379 | if (!sfilter) | |
d9e12f42 | 380 | return ERR_PTR(-ENOMEM); |
ac67eb2c DB |
381 | |
382 | ret = bpf_prog_create_from_user(&sfilter->prog, fprog, | |
f8e529ed | 383 | seccomp_check_filter, save_orig); |
ac67eb2c DB |
384 | if (ret < 0) { |
385 | kfree(sfilter); | |
386 | return ERR_PTR(ret); | |
d9e12f42 | 387 | } |
bd4cf0ed | 388 | |
ac67eb2c | 389 | atomic_set(&sfilter->usage, 1); |
e2cfabdf | 390 | |
ac67eb2c | 391 | return sfilter; |
e2cfabdf WD |
392 | } |
393 | ||
394 | /** | |
c8bee430 | 395 | * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog |
e2cfabdf WD |
396 | * @user_filter: pointer to the user data containing a sock_fprog. |
397 | * | |
398 | * Returns 0 on success and non-zero otherwise. | |
399 | */ | |
c8bee430 KC |
400 | static struct seccomp_filter * |
401 | seccomp_prepare_user_filter(const char __user *user_filter) | |
e2cfabdf WD |
402 | { |
403 | struct sock_fprog fprog; | |
c8bee430 | 404 | struct seccomp_filter *filter = ERR_PTR(-EFAULT); |
e2cfabdf WD |
405 | |
406 | #ifdef CONFIG_COMPAT | |
5c38065e | 407 | if (in_compat_syscall()) { |
e2cfabdf WD |
408 | struct compat_sock_fprog fprog32; |
409 | if (copy_from_user(&fprog32, user_filter, sizeof(fprog32))) | |
410 | goto out; | |
411 | fprog.len = fprog32.len; | |
412 | fprog.filter = compat_ptr(fprog32.filter); | |
413 | } else /* falls through to the if below. */ | |
414 | #endif | |
415 | if (copy_from_user(&fprog, user_filter, sizeof(fprog))) | |
416 | goto out; | |
c8bee430 | 417 | filter = seccomp_prepare_filter(&fprog); |
e2cfabdf | 418 | out: |
c8bee430 KC |
419 | return filter; |
420 | } | |
421 | ||
422 | /** | |
423 | * seccomp_attach_filter: validate and attach filter | |
424 | * @flags: flags to change filter behavior | |
425 | * @filter: seccomp filter to add to the current process | |
426 | * | |
dbd95212 KC |
427 | * Caller must be holding current->sighand->siglock lock. |
428 | * | |
c8bee430 KC |
429 | * Returns 0 on success, -ve on error. |
430 | */ | |
431 | static long seccomp_attach_filter(unsigned int flags, | |
432 | struct seccomp_filter *filter) | |
433 | { | |
434 | unsigned long total_insns; | |
435 | struct seccomp_filter *walker; | |
436 | ||
69f6a34b | 437 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 438 | |
c8bee430 KC |
439 | /* Validate resulting filter length. */ |
440 | total_insns = filter->prog->len; | |
441 | for (walker = current->seccomp.filter; walker; walker = walker->prev) | |
442 | total_insns += walker->prog->len + 4; /* 4 instr penalty */ | |
443 | if (total_insns > MAX_INSNS_PER_PATH) | |
444 | return -ENOMEM; | |
445 | ||
c2e1f2e3 KC |
446 | /* If thread sync has been requested, check that it is possible. */ |
447 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) { | |
448 | int ret; | |
449 | ||
450 | ret = seccomp_can_sync_threads(); | |
451 | if (ret) | |
452 | return ret; | |
453 | } | |
454 | ||
c8bee430 KC |
455 | /* |
456 | * If there is an existing filter, make it the prev and don't drop its | |
457 | * task reference. | |
458 | */ | |
459 | filter->prev = current->seccomp.filter; | |
460 | current->seccomp.filter = filter; | |
461 | ||
c2e1f2e3 KC |
462 | /* Now that the new filter is in place, synchronize to all threads. */ |
463 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) | |
ab677c2a | 464 | seccomp_sync_threads(flags); |
c2e1f2e3 | 465 | |
c8bee430 | 466 | return 0; |
e2cfabdf WD |
467 | } |
468 | ||
be69c4c0 ON |
469 | void __get_seccomp_filter(struct seccomp_filter *filter) |
470 | { | |
471 | /* Reference count is bounded by the number of total processes. */ | |
472 | atomic_inc(&filter->usage); | |
473 | } | |
474 | ||
e2cfabdf WD |
475 | /* get_seccomp_filter - increments the reference count of the filter on @tsk */ |
476 | void get_seccomp_filter(struct task_struct *tsk) | |
477 | { | |
478 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
479 | if (!orig) | |
480 | return; | |
be69c4c0 | 481 | __get_seccomp_filter(orig); |
e2cfabdf WD |
482 | } |
483 | ||
c8bee430 KC |
484 | static inline void seccomp_filter_free(struct seccomp_filter *filter) |
485 | { | |
486 | if (filter) { | |
bab18991 | 487 | bpf_prog_destroy(filter->prog); |
c8bee430 KC |
488 | kfree(filter); |
489 | } | |
490 | } | |
491 | ||
be69c4c0 | 492 | static void __put_seccomp_filter(struct seccomp_filter *orig) |
e2cfabdf | 493 | { |
e2cfabdf WD |
494 | /* Clean up single-reference branches iteratively. */ |
495 | while (orig && atomic_dec_and_test(&orig->usage)) { | |
496 | struct seccomp_filter *freeme = orig; | |
497 | orig = orig->prev; | |
c8bee430 | 498 | seccomp_filter_free(freeme); |
e2cfabdf WD |
499 | } |
500 | } | |
bb6ea430 | 501 | |
be69c4c0 ON |
502 | /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */ |
503 | void put_seccomp_filter(struct task_struct *tsk) | |
504 | { | |
505 | __put_seccomp_filter(tsk->seccomp.filter); | |
506 | } | |
507 | ||
bb6ea430 WD |
508 | /** |
509 | * seccomp_send_sigsys - signals the task to allow in-process syscall emulation | |
510 | * @syscall: syscall number to send to userland | |
511 | * @reason: filter-supplied reason code to send to userland (via si_errno) | |
512 | * | |
513 | * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info. | |
514 | */ | |
515 | static void seccomp_send_sigsys(int syscall, int reason) | |
516 | { | |
517 | struct siginfo info; | |
518 | memset(&info, 0, sizeof(info)); | |
519 | info.si_signo = SIGSYS; | |
520 | info.si_code = SYS_SECCOMP; | |
521 | info.si_call_addr = (void __user *)KSTK_EIP(current); | |
522 | info.si_errno = reason; | |
5e937a9a | 523 | info.si_arch = syscall_get_arch(); |
bb6ea430 WD |
524 | info.si_syscall = syscall; |
525 | force_sig_info(SIGSYS, &info, current); | |
526 | } | |
e2cfabdf | 527 | #endif /* CONFIG_SECCOMP_FILTER */ |
1da177e4 LT |
528 | |
529 | /* | |
530 | * Secure computing mode 1 allows only read/write/exit/sigreturn. | |
531 | * To be fully secure this must be combined with rlimit | |
532 | * to limit the stack allocations too. | |
533 | */ | |
cb4253aa | 534 | static const int mode1_syscalls[] = { |
1da177e4 LT |
535 | __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, |
536 | 0, /* null terminated */ | |
537 | }; | |
538 | ||
a4412fc9 | 539 | static void __secure_computing_strict(int this_syscall) |
1da177e4 | 540 | { |
cb4253aa | 541 | const int *syscall_whitelist = mode1_syscalls; |
a4412fc9 | 542 | #ifdef CONFIG_COMPAT |
5c38065e | 543 | if (in_compat_syscall()) |
c983f0e8 | 544 | syscall_whitelist = get_compat_mode1_syscalls(); |
a4412fc9 AL |
545 | #endif |
546 | do { | |
547 | if (*syscall_whitelist == this_syscall) | |
548 | return; | |
549 | } while (*++syscall_whitelist); | |
550 | ||
551 | #ifdef SECCOMP_DEBUG | |
552 | dump_stack(); | |
553 | #endif | |
554 | audit_seccomp(this_syscall, SIGKILL, SECCOMP_RET_KILL); | |
555 | do_exit(SIGKILL); | |
556 | } | |
557 | ||
558 | #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER | |
559 | void secure_computing_strict(int this_syscall) | |
560 | { | |
561 | int mode = current->seccomp.mode; | |
562 | ||
97f2645f | 563 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 TA |
564 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
565 | return; | |
566 | ||
221272f9 | 567 | if (mode == SECCOMP_MODE_DISABLED) |
a4412fc9 AL |
568 | return; |
569 | else if (mode == SECCOMP_MODE_STRICT) | |
570 | __secure_computing_strict(this_syscall); | |
571 | else | |
572 | BUG(); | |
573 | } | |
574 | #else | |
13aa72f0 AL |
575 | |
576 | #ifdef CONFIG_SECCOMP_FILTER | |
ce6526e8 KC |
577 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
578 | const bool recheck_after_trace) | |
13aa72f0 AL |
579 | { |
580 | u32 filter_ret, action; | |
581 | int data; | |
1da177e4 | 582 | |
3ba2530c KC |
583 | /* |
584 | * Make sure that any changes to mode from another thread have | |
585 | * been seen after TIF_SECCOMP was seen. | |
586 | */ | |
587 | rmb(); | |
588 | ||
d39bd00d | 589 | filter_ret = seccomp_run_filters(sd); |
13aa72f0 AL |
590 | data = filter_ret & SECCOMP_RET_DATA; |
591 | action = filter_ret & SECCOMP_RET_ACTION; | |
592 | ||
593 | switch (action) { | |
594 | case SECCOMP_RET_ERRNO: | |
580c57f1 KC |
595 | /* Set low-order bits as an errno, capped at MAX_ERRNO. */ |
596 | if (data > MAX_ERRNO) | |
597 | data = MAX_ERRNO; | |
d39bd00d | 598 | syscall_set_return_value(current, task_pt_regs(current), |
13aa72f0 AL |
599 | -data, 0); |
600 | goto skip; | |
601 | ||
602 | case SECCOMP_RET_TRAP: | |
603 | /* Show the handler the original registers. */ | |
d39bd00d | 604 | syscall_rollback(current, task_pt_regs(current)); |
13aa72f0 AL |
605 | /* Let the filter pass back 16 bits of data. */ |
606 | seccomp_send_sigsys(this_syscall, data); | |
607 | goto skip; | |
608 | ||
609 | case SECCOMP_RET_TRACE: | |
ce6526e8 KC |
610 | /* We've been put in this state by the ptracer already. */ |
611 | if (recheck_after_trace) | |
612 | return 0; | |
613 | ||
8112c4f1 KC |
614 | /* ENOSYS these calls if there is no tracer attached. */ |
615 | if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { | |
616 | syscall_set_return_value(current, | |
617 | task_pt_regs(current), | |
618 | -ENOSYS, 0); | |
619 | goto skip; | |
620 | } | |
621 | ||
622 | /* Allow the BPF to provide the event message */ | |
623 | ptrace_event(PTRACE_EVENT_SECCOMP, data); | |
624 | /* | |
625 | * The delivery of a fatal signal during event | |
485a252a KC |
626 | * notification may silently skip tracer notification, |
627 | * which could leave us with a potentially unmodified | |
628 | * syscall that the tracer would have liked to have | |
629 | * changed. Since the process is about to die, we just | |
630 | * force the syscall to be skipped and let the signal | |
631 | * kill the process and correctly handle any tracer exit | |
632 | * notifications. | |
8112c4f1 KC |
633 | */ |
634 | if (fatal_signal_pending(current)) | |
485a252a | 635 | goto skip; |
8112c4f1 KC |
636 | /* Check if the tracer forced the syscall to be skipped. */ |
637 | this_syscall = syscall_get_nr(current, task_pt_regs(current)); | |
638 | if (this_syscall < 0) | |
639 | goto skip; | |
640 | ||
ce6526e8 KC |
641 | /* |
642 | * Recheck the syscall, since it may have changed. This | |
643 | * intentionally uses a NULL struct seccomp_data to force | |
644 | * a reload of all registers. This does not goto skip since | |
645 | * a skip would have already been reported. | |
646 | */ | |
647 | if (__seccomp_filter(this_syscall, NULL, true)) | |
648 | return -1; | |
649 | ||
8112c4f1 | 650 | return 0; |
13aa72f0 AL |
651 | |
652 | case SECCOMP_RET_ALLOW: | |
8112c4f1 | 653 | return 0; |
13aa72f0 AL |
654 | |
655 | case SECCOMP_RET_KILL: | |
656 | default: | |
657 | audit_seccomp(this_syscall, SIGSYS, action); | |
658 | do_exit(SIGSYS); | |
659 | } | |
660 | ||
661 | unreachable(); | |
662 | ||
663 | skip: | |
664 | audit_seccomp(this_syscall, 0, action); | |
8112c4f1 KC |
665 | return -1; |
666 | } | |
667 | #else | |
ce6526e8 KC |
668 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
669 | const bool recheck_after_trace) | |
8112c4f1 KC |
670 | { |
671 | BUG(); | |
e2e1857a PC |
672 | |
673 | return -1; | |
13aa72f0 | 674 | } |
1da177e4 | 675 | #endif |
13aa72f0 | 676 | |
8112c4f1 | 677 | int __secure_computing(const struct seccomp_data *sd) |
13aa72f0 AL |
678 | { |
679 | int mode = current->seccomp.mode; | |
8112c4f1 | 680 | int this_syscall; |
13aa72f0 | 681 | |
97f2645f | 682 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 | 683 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
8112c4f1 KC |
684 | return 0; |
685 | ||
686 | this_syscall = sd ? sd->nr : | |
687 | syscall_get_nr(current, task_pt_regs(current)); | |
13c4a901 | 688 | |
13aa72f0 | 689 | switch (mode) { |
e2cfabdf | 690 | case SECCOMP_MODE_STRICT: |
13aa72f0 | 691 | __secure_computing_strict(this_syscall); /* may call do_exit */ |
8112c4f1 | 692 | return 0; |
13aa72f0 | 693 | case SECCOMP_MODE_FILTER: |
ce6526e8 | 694 | return __seccomp_filter(this_syscall, sd, false); |
1da177e4 LT |
695 | default: |
696 | BUG(); | |
697 | } | |
13aa72f0 | 698 | } |
a4412fc9 | 699 | #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ |
1d9d02fe AA |
700 | |
701 | long prctl_get_seccomp(void) | |
702 | { | |
703 | return current->seccomp.mode; | |
704 | } | |
705 | ||
e2cfabdf | 706 | /** |
3b23dd12 | 707 | * seccomp_set_mode_strict: internal function for setting strict seccomp |
e2cfabdf WD |
708 | * |
709 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
710 | * | |
711 | * Returns 0 on success or -EINVAL on failure. | |
712 | */ | |
3b23dd12 | 713 | static long seccomp_set_mode_strict(void) |
1d9d02fe | 714 | { |
3b23dd12 | 715 | const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; |
e2cfabdf | 716 | long ret = -EINVAL; |
1d9d02fe | 717 | |
dbd95212 KC |
718 | spin_lock_irq(¤t->sighand->siglock); |
719 | ||
1f41b450 | 720 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1d9d02fe AA |
721 | goto out; |
722 | ||
cf99abac | 723 | #ifdef TIF_NOTSC |
3b23dd12 | 724 | disable_TSC(); |
cf99abac | 725 | #endif |
ab677c2a | 726 | seccomp_assign_mode(current, seccomp_mode, 0); |
3b23dd12 KC |
727 | ret = 0; |
728 | ||
729 | out: | |
dbd95212 | 730 | spin_unlock_irq(¤t->sighand->siglock); |
3b23dd12 KC |
731 | |
732 | return ret; | |
733 | } | |
734 | ||
e2cfabdf | 735 | #ifdef CONFIG_SECCOMP_FILTER |
3b23dd12 KC |
736 | /** |
737 | * seccomp_set_mode_filter: internal function for setting seccomp filter | |
48dc92b9 | 738 | * @flags: flags to change filter behavior |
3b23dd12 KC |
739 | * @filter: struct sock_fprog containing filter |
740 | * | |
741 | * This function may be called repeatedly to install additional filters. | |
742 | * Every filter successfully installed will be evaluated (in reverse order) | |
743 | * for each system call the task makes. | |
744 | * | |
745 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
746 | * | |
747 | * Returns 0 on success or -EINVAL on failure. | |
748 | */ | |
48dc92b9 KC |
749 | static long seccomp_set_mode_filter(unsigned int flags, |
750 | const char __user *filter) | |
3b23dd12 KC |
751 | { |
752 | const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; | |
c8bee430 | 753 | struct seccomp_filter *prepared = NULL; |
3b23dd12 KC |
754 | long ret = -EINVAL; |
755 | ||
48dc92b9 | 756 | /* Validate flags. */ |
c2e1f2e3 | 757 | if (flags & ~SECCOMP_FILTER_FLAG_MASK) |
dbd95212 | 758 | return -EINVAL; |
48dc92b9 | 759 | |
c8bee430 KC |
760 | /* Prepare the new filter before holding any locks. */ |
761 | prepared = seccomp_prepare_user_filter(filter); | |
762 | if (IS_ERR(prepared)) | |
763 | return PTR_ERR(prepared); | |
764 | ||
c2e1f2e3 KC |
765 | /* |
766 | * Make sure we cannot change seccomp or nnp state via TSYNC | |
767 | * while another thread is in the middle of calling exec. | |
768 | */ | |
769 | if (flags & SECCOMP_FILTER_FLAG_TSYNC && | |
770 | mutex_lock_killable(¤t->signal->cred_guard_mutex)) | |
771 | goto out_free; | |
772 | ||
dbd95212 KC |
773 | spin_lock_irq(¤t->sighand->siglock); |
774 | ||
3b23dd12 KC |
775 | if (!seccomp_may_assign_mode(seccomp_mode)) |
776 | goto out; | |
777 | ||
c8bee430 | 778 | ret = seccomp_attach_filter(flags, prepared); |
3b23dd12 | 779 | if (ret) |
e2cfabdf | 780 | goto out; |
c8bee430 KC |
781 | /* Do not free the successfully attached filter. */ |
782 | prepared = NULL; | |
1d9d02fe | 783 | |
ab677c2a | 784 | seccomp_assign_mode(current, seccomp_mode, flags); |
e2cfabdf | 785 | out: |
dbd95212 | 786 | spin_unlock_irq(¤t->sighand->siglock); |
c2e1f2e3 KC |
787 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
788 | mutex_unlock(¤t->signal->cred_guard_mutex); | |
789 | out_free: | |
c8bee430 | 790 | seccomp_filter_free(prepared); |
1d9d02fe AA |
791 | return ret; |
792 | } | |
3b23dd12 | 793 | #else |
48dc92b9 KC |
794 | static inline long seccomp_set_mode_filter(unsigned int flags, |
795 | const char __user *filter) | |
3b23dd12 KC |
796 | { |
797 | return -EINVAL; | |
798 | } | |
799 | #endif | |
d78ab02c | 800 | |
48dc92b9 KC |
801 | /* Common entry point for both prctl and syscall. */ |
802 | static long do_seccomp(unsigned int op, unsigned int flags, | |
803 | const char __user *uargs) | |
804 | { | |
805 | switch (op) { | |
806 | case SECCOMP_SET_MODE_STRICT: | |
807 | if (flags != 0 || uargs != NULL) | |
808 | return -EINVAL; | |
809 | return seccomp_set_mode_strict(); | |
810 | case SECCOMP_SET_MODE_FILTER: | |
811 | return seccomp_set_mode_filter(flags, uargs); | |
812 | default: | |
813 | return -EINVAL; | |
814 | } | |
815 | } | |
816 | ||
817 | SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, | |
818 | const char __user *, uargs) | |
819 | { | |
820 | return do_seccomp(op, flags, uargs); | |
821 | } | |
822 | ||
d78ab02c KC |
823 | /** |
824 | * prctl_set_seccomp: configures current->seccomp.mode | |
825 | * @seccomp_mode: requested mode to use | |
826 | * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER | |
827 | * | |
828 | * Returns 0 on success or -EINVAL on failure. | |
829 | */ | |
830 | long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter) | |
831 | { | |
48dc92b9 KC |
832 | unsigned int op; |
833 | char __user *uargs; | |
834 | ||
3b23dd12 KC |
835 | switch (seccomp_mode) { |
836 | case SECCOMP_MODE_STRICT: | |
48dc92b9 KC |
837 | op = SECCOMP_SET_MODE_STRICT; |
838 | /* | |
839 | * Setting strict mode through prctl always ignored filter, | |
840 | * so make sure it is always NULL here to pass the internal | |
841 | * check in do_seccomp(). | |
842 | */ | |
843 | uargs = NULL; | |
844 | break; | |
3b23dd12 | 845 | case SECCOMP_MODE_FILTER: |
48dc92b9 KC |
846 | op = SECCOMP_SET_MODE_FILTER; |
847 | uargs = filter; | |
848 | break; | |
3b23dd12 KC |
849 | default: |
850 | return -EINVAL; | |
851 | } | |
48dc92b9 KC |
852 | |
853 | /* prctl interface doesn't have flags, so they are always zero. */ | |
854 | return do_seccomp(op, 0, uargs); | |
d78ab02c | 855 | } |
f8e529ed TA |
856 | |
857 | #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE) | |
858 | long seccomp_get_filter(struct task_struct *task, unsigned long filter_off, | |
859 | void __user *data) | |
860 | { | |
861 | struct seccomp_filter *filter; | |
862 | struct sock_fprog_kern *fprog; | |
863 | long ret; | |
864 | unsigned long count = 0; | |
865 | ||
866 | if (!capable(CAP_SYS_ADMIN) || | |
867 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { | |
868 | return -EACCES; | |
869 | } | |
870 | ||
871 | spin_lock_irq(&task->sighand->siglock); | |
872 | if (task->seccomp.mode != SECCOMP_MODE_FILTER) { | |
873 | ret = -EINVAL; | |
874 | goto out; | |
875 | } | |
876 | ||
877 | filter = task->seccomp.filter; | |
878 | while (filter) { | |
879 | filter = filter->prev; | |
880 | count++; | |
881 | } | |
882 | ||
883 | if (filter_off >= count) { | |
884 | ret = -ENOENT; | |
885 | goto out; | |
886 | } | |
887 | count -= filter_off; | |
888 | ||
889 | filter = task->seccomp.filter; | |
890 | while (filter && count > 1) { | |
891 | filter = filter->prev; | |
892 | count--; | |
893 | } | |
894 | ||
895 | if (WARN_ON(count != 1 || !filter)) { | |
896 | /* The filter tree shouldn't shrink while we're using it. */ | |
897 | ret = -ENOENT; | |
898 | goto out; | |
899 | } | |
900 | ||
901 | fprog = filter->prog->orig_prog; | |
902 | if (!fprog) { | |
470bf1f2 | 903 | /* This must be a new non-cBPF filter, since we save |
f8e529ed TA |
904 | * every cBPF filter's orig_prog above when |
905 | * CONFIG_CHECKPOINT_RESTORE is enabled. | |
906 | */ | |
907 | ret = -EMEDIUMTYPE; | |
908 | goto out; | |
909 | } | |
910 | ||
911 | ret = fprog->len; | |
912 | if (!data) | |
913 | goto out; | |
914 | ||
be69c4c0 | 915 | __get_seccomp_filter(filter); |
f8e529ed TA |
916 | spin_unlock_irq(&task->sighand->siglock); |
917 | ||
918 | if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog))) | |
919 | ret = -EFAULT; | |
920 | ||
be69c4c0 | 921 | __put_seccomp_filter(filter); |
f8e529ed TA |
922 | return ret; |
923 | ||
924 | out: | |
925 | spin_unlock_irq(&task->sighand->siglock); | |
926 | return ret; | |
927 | } | |
928 | #endif |