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 LT |
23 | |
24 | /* #define SECCOMP_DEBUG 1 */ | |
e2cfabdf | 25 | |
a4412fc9 | 26 | #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER |
e2cfabdf | 27 | #include <asm/syscall.h> |
a4412fc9 AL |
28 | #endif |
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 | */ | |
a4412fc9 | 178 | static u32 seccomp_run_filters(void) |
e2cfabdf | 179 | { |
3ba2530c | 180 | struct seccomp_filter *f = ACCESS_ONCE(current->seccomp.filter); |
bd4cf0ed | 181 | struct seccomp_data sd; |
acf3b2c7 WD |
182 | u32 ret = SECCOMP_RET_ALLOW; |
183 | ||
184 | /* Ensure unexpected behavior doesn't result in failing open. */ | |
3ba2530c | 185 | if (unlikely(WARN_ON(f == NULL))) |
acf3b2c7 WD |
186 | return SECCOMP_RET_KILL; |
187 | ||
3ba2530c KC |
188 | /* Make sure cross-thread synced filter points somewhere sane. */ |
189 | smp_read_barrier_depends(); | |
190 | ||
bd4cf0ed AS |
191 | populate_seccomp_data(&sd); |
192 | ||
e2cfabdf WD |
193 | /* |
194 | * All filters in the list are evaluated and the lowest BPF return | |
acf3b2c7 | 195 | * value always takes priority (ignoring the DATA). |
e2cfabdf | 196 | */ |
3ba2530c | 197 | for (; f; f = f->prev) { |
7ae457c1 | 198 | u32 cur_ret = BPF_PROG_RUN(f->prog, (void *)&sd); |
8f577cad | 199 | |
acf3b2c7 WD |
200 | if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION)) |
201 | ret = cur_ret; | |
e2cfabdf WD |
202 | } |
203 | return ret; | |
204 | } | |
1f41b450 | 205 | #endif /* CONFIG_SECCOMP_FILTER */ |
e2cfabdf | 206 | |
1f41b450 KC |
207 | static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) |
208 | { | |
69f6a34b | 209 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 210 | |
1f41b450 KC |
211 | if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) |
212 | return false; | |
213 | ||
214 | return true; | |
215 | } | |
216 | ||
3ba2530c KC |
217 | static inline void seccomp_assign_mode(struct task_struct *task, |
218 | unsigned long seccomp_mode) | |
1f41b450 | 219 | { |
69f6a34b | 220 | assert_spin_locked(&task->sighand->siglock); |
dbd95212 | 221 | |
3ba2530c KC |
222 | task->seccomp.mode = seccomp_mode; |
223 | /* | |
224 | * Make sure TIF_SECCOMP cannot be set before the mode (and | |
225 | * filter) is set. | |
226 | */ | |
227 | smp_mb__before_atomic(); | |
228 | set_tsk_thread_flag(task, TIF_SECCOMP); | |
1f41b450 KC |
229 | } |
230 | ||
231 | #ifdef CONFIG_SECCOMP_FILTER | |
c2e1f2e3 KC |
232 | /* Returns 1 if the parent is an ancestor of the child. */ |
233 | static int is_ancestor(struct seccomp_filter *parent, | |
234 | struct seccomp_filter *child) | |
235 | { | |
236 | /* NULL is the root ancestor. */ | |
237 | if (parent == NULL) | |
238 | return 1; | |
239 | for (; child; child = child->prev) | |
240 | if (child == parent) | |
241 | return 1; | |
242 | return 0; | |
243 | } | |
244 | ||
245 | /** | |
246 | * seccomp_can_sync_threads: checks if all threads can be synchronized | |
247 | * | |
248 | * Expects sighand and cred_guard_mutex locks to be held. | |
249 | * | |
250 | * Returns 0 on success, -ve on error, or the pid of a thread which was | |
251 | * either not in the correct seccomp mode or it did not have an ancestral | |
252 | * seccomp filter. | |
253 | */ | |
254 | static inline pid_t seccomp_can_sync_threads(void) | |
255 | { | |
256 | struct task_struct *thread, *caller; | |
257 | ||
258 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 259 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
260 | |
261 | /* Validate all threads being eligible for synchronization. */ | |
262 | caller = current; | |
263 | for_each_thread(caller, thread) { | |
264 | pid_t failed; | |
265 | ||
266 | /* Skip current, since it is initiating the sync. */ | |
267 | if (thread == caller) | |
268 | continue; | |
269 | ||
270 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || | |
271 | (thread->seccomp.mode == SECCOMP_MODE_FILTER && | |
272 | is_ancestor(thread->seccomp.filter, | |
273 | caller->seccomp.filter))) | |
274 | continue; | |
275 | ||
276 | /* Return the first thread that cannot be synchronized. */ | |
277 | failed = task_pid_vnr(thread); | |
278 | /* If the pid cannot be resolved, then return -ESRCH */ | |
279 | if (unlikely(WARN_ON(failed == 0))) | |
280 | failed = -ESRCH; | |
281 | return failed; | |
282 | } | |
283 | ||
284 | return 0; | |
285 | } | |
286 | ||
287 | /** | |
288 | * seccomp_sync_threads: sets all threads to use current's filter | |
289 | * | |
290 | * Expects sighand and cred_guard_mutex locks to be held, and for | |
291 | * seccomp_can_sync_threads() to have returned success already | |
292 | * without dropping the locks. | |
293 | * | |
294 | */ | |
295 | static inline void seccomp_sync_threads(void) | |
296 | { | |
297 | struct task_struct *thread, *caller; | |
298 | ||
299 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 300 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
301 | |
302 | /* Synchronize all threads. */ | |
303 | caller = current; | |
304 | for_each_thread(caller, thread) { | |
305 | /* Skip current, since it needs no changes. */ | |
306 | if (thread == caller) | |
307 | continue; | |
308 | ||
309 | /* Get a task reference for the new leaf node. */ | |
310 | get_seccomp_filter(caller); | |
311 | /* | |
312 | * Drop the task reference to the shared ancestor since | |
313 | * current's path will hold a reference. (This also | |
314 | * allows a put before the assignment.) | |
315 | */ | |
316 | put_seccomp_filter(thread); | |
317 | smp_store_release(&thread->seccomp.filter, | |
318 | caller->seccomp.filter); | |
319 | /* | |
320 | * Opt the other thread into seccomp if needed. | |
321 | * As threads are considered to be trust-realm | |
322 | * equivalent (see ptrace_may_access), it is safe to | |
323 | * allow one thread to transition the other. | |
324 | */ | |
325 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) { | |
326 | /* | |
327 | * Don't let an unprivileged task work around | |
328 | * the no_new_privs restriction by creating | |
329 | * a thread that sets it up, enters seccomp, | |
330 | * then dies. | |
331 | */ | |
332 | if (task_no_new_privs(caller)) | |
333 | task_set_no_new_privs(thread); | |
334 | ||
335 | seccomp_assign_mode(thread, SECCOMP_MODE_FILTER); | |
336 | } | |
337 | } | |
338 | } | |
339 | ||
e2cfabdf | 340 | /** |
c8bee430 | 341 | * seccomp_prepare_filter: Prepares a seccomp filter for use. |
e2cfabdf WD |
342 | * @fprog: BPF program to install |
343 | * | |
c8bee430 | 344 | * Returns filter on success or an ERR_PTR on failure. |
e2cfabdf | 345 | */ |
c8bee430 | 346 | static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog) |
e2cfabdf WD |
347 | { |
348 | struct seccomp_filter *filter; | |
c8bee430 | 349 | unsigned long fp_size; |
bd4cf0ed AS |
350 | struct sock_filter *fp; |
351 | int new_len; | |
e2cfabdf WD |
352 | long ret; |
353 | ||
354 | if (fprog->len == 0 || fprog->len > BPF_MAXINSNS) | |
c8bee430 KC |
355 | return ERR_PTR(-EINVAL); |
356 | BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter)); | |
357 | fp_size = fprog->len * sizeof(struct sock_filter); | |
e2cfabdf WD |
358 | |
359 | /* | |
119ce5c8 | 360 | * Installing a seccomp filter requires that the task has |
e2cfabdf WD |
361 | * CAP_SYS_ADMIN in its namespace or be running with no_new_privs. |
362 | * This avoids scenarios where unprivileged tasks can affect the | |
363 | * behavior of privileged children. | |
364 | */ | |
1d4457f9 | 365 | if (!task_no_new_privs(current) && |
e2cfabdf WD |
366 | security_capable_noaudit(current_cred(), current_user_ns(), |
367 | CAP_SYS_ADMIN) != 0) | |
c8bee430 | 368 | return ERR_PTR(-EACCES); |
e2cfabdf | 369 | |
bd4cf0ed AS |
370 | fp = kzalloc(fp_size, GFP_KERNEL|__GFP_NOWARN); |
371 | if (!fp) | |
c8bee430 | 372 | return ERR_PTR(-ENOMEM); |
e2cfabdf WD |
373 | |
374 | /* Copy the instructions from fprog. */ | |
375 | ret = -EFAULT; | |
bd4cf0ed AS |
376 | if (copy_from_user(fp, fprog->filter, fp_size)) |
377 | goto free_prog; | |
e2cfabdf WD |
378 | |
379 | /* Check and rewrite the fprog via the skb checker */ | |
4df95ff4 | 380 | ret = bpf_check_classic(fp, fprog->len); |
e2cfabdf | 381 | if (ret) |
bd4cf0ed | 382 | goto free_prog; |
e2cfabdf WD |
383 | |
384 | /* Check and rewrite the fprog for seccomp use */ | |
bd4cf0ed AS |
385 | ret = seccomp_check_filter(fp, fprog->len); |
386 | if (ret) | |
387 | goto free_prog; | |
388 | ||
2695fb55 | 389 | /* Convert 'sock_filter' insns to 'bpf_insn' insns */ |
8fb575ca | 390 | ret = bpf_convert_filter(fp, fprog->len, NULL, &new_len); |
bd4cf0ed AS |
391 | if (ret) |
392 | goto free_prog; | |
393 | ||
394 | /* Allocate a new seccomp_filter */ | |
0acf07d2 | 395 | ret = -ENOMEM; |
8f577cad | 396 | filter = kzalloc(sizeof(struct seccomp_filter), |
bd4cf0ed AS |
397 | GFP_KERNEL|__GFP_NOWARN); |
398 | if (!filter) | |
399 | goto free_prog; | |
400 | ||
7ae457c1 | 401 | filter->prog = kzalloc(bpf_prog_size(new_len), |
8f577cad AS |
402 | GFP_KERNEL|__GFP_NOWARN); |
403 | if (!filter->prog) | |
bd4cf0ed | 404 | goto free_filter; |
8f577cad | 405 | |
8fb575ca | 406 | ret = bpf_convert_filter(fp, fprog->len, filter->prog->insnsi, &new_len); |
8f577cad AS |
407 | if (ret) |
408 | goto free_filter_prog; | |
0acf07d2 | 409 | kfree(fp); |
bd4cf0ed AS |
410 | |
411 | atomic_set(&filter->usage, 1); | |
8f577cad | 412 | filter->prog->len = new_len; |
8f577cad | 413 | |
7ae457c1 | 414 | bpf_prog_select_runtime(filter->prog); |
e2cfabdf | 415 | |
c8bee430 | 416 | return filter; |
bd4cf0ed | 417 | |
8f577cad AS |
418 | free_filter_prog: |
419 | kfree(filter->prog); | |
bd4cf0ed | 420 | free_filter: |
e2cfabdf | 421 | kfree(filter); |
bd4cf0ed AS |
422 | free_prog: |
423 | kfree(fp); | |
c8bee430 | 424 | return ERR_PTR(ret); |
e2cfabdf WD |
425 | } |
426 | ||
427 | /** | |
c8bee430 | 428 | * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog |
e2cfabdf WD |
429 | * @user_filter: pointer to the user data containing a sock_fprog. |
430 | * | |
431 | * Returns 0 on success and non-zero otherwise. | |
432 | */ | |
c8bee430 KC |
433 | static struct seccomp_filter * |
434 | seccomp_prepare_user_filter(const char __user *user_filter) | |
e2cfabdf WD |
435 | { |
436 | struct sock_fprog fprog; | |
c8bee430 | 437 | struct seccomp_filter *filter = ERR_PTR(-EFAULT); |
e2cfabdf WD |
438 | |
439 | #ifdef CONFIG_COMPAT | |
440 | if (is_compat_task()) { | |
441 | struct compat_sock_fprog fprog32; | |
442 | if (copy_from_user(&fprog32, user_filter, sizeof(fprog32))) | |
443 | goto out; | |
444 | fprog.len = fprog32.len; | |
445 | fprog.filter = compat_ptr(fprog32.filter); | |
446 | } else /* falls through to the if below. */ | |
447 | #endif | |
448 | if (copy_from_user(&fprog, user_filter, sizeof(fprog))) | |
449 | goto out; | |
c8bee430 | 450 | filter = seccomp_prepare_filter(&fprog); |
e2cfabdf | 451 | out: |
c8bee430 KC |
452 | return filter; |
453 | } | |
454 | ||
455 | /** | |
456 | * seccomp_attach_filter: validate and attach filter | |
457 | * @flags: flags to change filter behavior | |
458 | * @filter: seccomp filter to add to the current process | |
459 | * | |
dbd95212 KC |
460 | * Caller must be holding current->sighand->siglock lock. |
461 | * | |
c8bee430 KC |
462 | * Returns 0 on success, -ve on error. |
463 | */ | |
464 | static long seccomp_attach_filter(unsigned int flags, | |
465 | struct seccomp_filter *filter) | |
466 | { | |
467 | unsigned long total_insns; | |
468 | struct seccomp_filter *walker; | |
469 | ||
69f6a34b | 470 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 471 | |
c8bee430 KC |
472 | /* Validate resulting filter length. */ |
473 | total_insns = filter->prog->len; | |
474 | for (walker = current->seccomp.filter; walker; walker = walker->prev) | |
475 | total_insns += walker->prog->len + 4; /* 4 instr penalty */ | |
476 | if (total_insns > MAX_INSNS_PER_PATH) | |
477 | return -ENOMEM; | |
478 | ||
c2e1f2e3 KC |
479 | /* If thread sync has been requested, check that it is possible. */ |
480 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) { | |
481 | int ret; | |
482 | ||
483 | ret = seccomp_can_sync_threads(); | |
484 | if (ret) | |
485 | return ret; | |
486 | } | |
487 | ||
c8bee430 KC |
488 | /* |
489 | * If there is an existing filter, make it the prev and don't drop its | |
490 | * task reference. | |
491 | */ | |
492 | filter->prev = current->seccomp.filter; | |
493 | current->seccomp.filter = filter; | |
494 | ||
c2e1f2e3 KC |
495 | /* Now that the new filter is in place, synchronize to all threads. */ |
496 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) | |
497 | seccomp_sync_threads(); | |
498 | ||
c8bee430 | 499 | return 0; |
e2cfabdf WD |
500 | } |
501 | ||
502 | /* get_seccomp_filter - increments the reference count of the filter on @tsk */ | |
503 | void get_seccomp_filter(struct task_struct *tsk) | |
504 | { | |
505 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
506 | if (!orig) | |
507 | return; | |
508 | /* Reference count is bounded by the number of total processes. */ | |
509 | atomic_inc(&orig->usage); | |
510 | } | |
511 | ||
c8bee430 KC |
512 | static inline void seccomp_filter_free(struct seccomp_filter *filter) |
513 | { | |
514 | if (filter) { | |
ae045e24 | 515 | bpf_prog_free(filter->prog); |
c8bee430 KC |
516 | kfree(filter); |
517 | } | |
518 | } | |
519 | ||
e2cfabdf WD |
520 | /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */ |
521 | void put_seccomp_filter(struct task_struct *tsk) | |
522 | { | |
523 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
524 | /* Clean up single-reference branches iteratively. */ | |
525 | while (orig && atomic_dec_and_test(&orig->usage)) { | |
526 | struct seccomp_filter *freeme = orig; | |
527 | orig = orig->prev; | |
c8bee430 | 528 | seccomp_filter_free(freeme); |
e2cfabdf WD |
529 | } |
530 | } | |
bb6ea430 WD |
531 | |
532 | /** | |
533 | * seccomp_send_sigsys - signals the task to allow in-process syscall emulation | |
534 | * @syscall: syscall number to send to userland | |
535 | * @reason: filter-supplied reason code to send to userland (via si_errno) | |
536 | * | |
537 | * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info. | |
538 | */ | |
539 | static void seccomp_send_sigsys(int syscall, int reason) | |
540 | { | |
541 | struct siginfo info; | |
542 | memset(&info, 0, sizeof(info)); | |
543 | info.si_signo = SIGSYS; | |
544 | info.si_code = SYS_SECCOMP; | |
545 | info.si_call_addr = (void __user *)KSTK_EIP(current); | |
546 | info.si_errno = reason; | |
5e937a9a | 547 | info.si_arch = syscall_get_arch(); |
bb6ea430 WD |
548 | info.si_syscall = syscall; |
549 | force_sig_info(SIGSYS, &info, current); | |
550 | } | |
e2cfabdf | 551 | #endif /* CONFIG_SECCOMP_FILTER */ |
1da177e4 LT |
552 | |
553 | /* | |
554 | * Secure computing mode 1 allows only read/write/exit/sigreturn. | |
555 | * To be fully secure this must be combined with rlimit | |
556 | * to limit the stack allocations too. | |
557 | */ | |
558 | static int mode1_syscalls[] = { | |
559 | __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, | |
560 | 0, /* null terminated */ | |
561 | }; | |
562 | ||
5b101740 | 563 | #ifdef CONFIG_COMPAT |
1da177e4 LT |
564 | static int mode1_syscalls_32[] = { |
565 | __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32, | |
566 | 0, /* null terminated */ | |
567 | }; | |
568 | #endif | |
569 | ||
a4412fc9 | 570 | static void __secure_computing_strict(int this_syscall) |
1da177e4 | 571 | { |
a4412fc9 AL |
572 | int *syscall_whitelist = mode1_syscalls; |
573 | #ifdef CONFIG_COMPAT | |
574 | if (is_compat_task()) | |
575 | syscall_whitelist = mode1_syscalls_32; | |
576 | #endif | |
577 | do { | |
578 | if (*syscall_whitelist == this_syscall) | |
579 | return; | |
580 | } while (*++syscall_whitelist); | |
581 | ||
582 | #ifdef SECCOMP_DEBUG | |
583 | dump_stack(); | |
584 | #endif | |
585 | audit_seccomp(this_syscall, SIGKILL, SECCOMP_RET_KILL); | |
586 | do_exit(SIGKILL); | |
587 | } | |
588 | ||
589 | #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER | |
590 | void secure_computing_strict(int this_syscall) | |
591 | { | |
592 | int mode = current->seccomp.mode; | |
593 | ||
594 | if (mode == 0) | |
595 | return; | |
596 | else if (mode == SECCOMP_MODE_STRICT) | |
597 | __secure_computing_strict(this_syscall); | |
598 | else | |
599 | BUG(); | |
600 | } | |
601 | #else | |
602 | int __secure_computing(void) | |
603 | { | |
604 | struct pt_regs *regs = task_pt_regs(current); | |
605 | int this_syscall = syscall_get_nr(current, regs); | |
e2cfabdf | 606 | int exit_sig = 0; |
8156b451 | 607 | u32 ret; |
1da177e4 | 608 | |
3ba2530c KC |
609 | /* |
610 | * Make sure that any changes to mode from another thread have | |
611 | * been seen after TIF_SECCOMP was seen. | |
612 | */ | |
613 | rmb(); | |
614 | ||
615 | switch (current->seccomp.mode) { | |
e2cfabdf | 616 | case SECCOMP_MODE_STRICT: |
a4412fc9 AL |
617 | __secure_computing_strict(this_syscall); |
618 | return 0; | |
e2cfabdf | 619 | #ifdef CONFIG_SECCOMP_FILTER |
8156b451 WD |
620 | case SECCOMP_MODE_FILTER: { |
621 | int data; | |
a4412fc9 | 622 | ret = seccomp_run_filters(); |
acf3b2c7 | 623 | data = ret & SECCOMP_RET_DATA; |
8156b451 WD |
624 | ret &= SECCOMP_RET_ACTION; |
625 | switch (ret) { | |
acf3b2c7 WD |
626 | case SECCOMP_RET_ERRNO: |
627 | /* Set the low-order 16-bits as a errno. */ | |
87b526d3 | 628 | syscall_set_return_value(current, regs, |
acf3b2c7 WD |
629 | -data, 0); |
630 | goto skip; | |
bb6ea430 WD |
631 | case SECCOMP_RET_TRAP: |
632 | /* Show the handler the original registers. */ | |
87b526d3 | 633 | syscall_rollback(current, regs); |
bb6ea430 WD |
634 | /* Let the filter pass back 16 bits of data. */ |
635 | seccomp_send_sigsys(this_syscall, data); | |
636 | goto skip; | |
fb0fadf9 WD |
637 | case SECCOMP_RET_TRACE: |
638 | /* Skip these calls if there is no tracer. */ | |
87b526d3 AL |
639 | if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { |
640 | syscall_set_return_value(current, regs, | |
641 | -ENOSYS, 0); | |
fb0fadf9 | 642 | goto skip; |
87b526d3 | 643 | } |
fb0fadf9 WD |
644 | /* Allow the BPF to provide the event message */ |
645 | ptrace_event(PTRACE_EVENT_SECCOMP, data); | |
646 | /* | |
647 | * The delivery of a fatal signal during event | |
648 | * notification may silently skip tracer notification. | |
649 | * Terminating the task now avoids executing a system | |
650 | * call that may not be intended. | |
651 | */ | |
652 | if (fatal_signal_pending(current)) | |
653 | break; | |
87b526d3 AL |
654 | if (syscall_get_nr(current, regs) < 0) |
655 | goto skip; /* Explicit request to skip. */ | |
656 | ||
fb0fadf9 | 657 | return 0; |
acf3b2c7 WD |
658 | case SECCOMP_RET_ALLOW: |
659 | return 0; | |
660 | case SECCOMP_RET_KILL: | |
661 | default: | |
662 | break; | |
663 | } | |
e2cfabdf WD |
664 | exit_sig = SIGSYS; |
665 | break; | |
8156b451 | 666 | } |
e2cfabdf | 667 | #endif |
1da177e4 LT |
668 | default: |
669 | BUG(); | |
670 | } | |
671 | ||
672 | #ifdef SECCOMP_DEBUG | |
673 | dump_stack(); | |
674 | #endif | |
acf3b2c7 | 675 | audit_seccomp(this_syscall, exit_sig, ret); |
e2cfabdf | 676 | do_exit(exit_sig); |
8156b451 | 677 | #ifdef CONFIG_SECCOMP_FILTER |
acf3b2c7 WD |
678 | skip: |
679 | audit_seccomp(this_syscall, exit_sig, ret); | |
680 | return -1; | |
a4412fc9 | 681 | #endif |
1da177e4 | 682 | } |
a4412fc9 | 683 | #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ |
1d9d02fe AA |
684 | |
685 | long prctl_get_seccomp(void) | |
686 | { | |
687 | return current->seccomp.mode; | |
688 | } | |
689 | ||
e2cfabdf | 690 | /** |
3b23dd12 | 691 | * seccomp_set_mode_strict: internal function for setting strict seccomp |
e2cfabdf WD |
692 | * |
693 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
694 | * | |
695 | * Returns 0 on success or -EINVAL on failure. | |
696 | */ | |
3b23dd12 | 697 | static long seccomp_set_mode_strict(void) |
1d9d02fe | 698 | { |
3b23dd12 | 699 | const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; |
e2cfabdf | 700 | long ret = -EINVAL; |
1d9d02fe | 701 | |
dbd95212 KC |
702 | spin_lock_irq(¤t->sighand->siglock); |
703 | ||
1f41b450 | 704 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1d9d02fe AA |
705 | goto out; |
706 | ||
cf99abac | 707 | #ifdef TIF_NOTSC |
3b23dd12 | 708 | disable_TSC(); |
cf99abac | 709 | #endif |
3ba2530c | 710 | seccomp_assign_mode(current, seccomp_mode); |
3b23dd12 KC |
711 | ret = 0; |
712 | ||
713 | out: | |
dbd95212 | 714 | spin_unlock_irq(¤t->sighand->siglock); |
3b23dd12 KC |
715 | |
716 | return ret; | |
717 | } | |
718 | ||
e2cfabdf | 719 | #ifdef CONFIG_SECCOMP_FILTER |
3b23dd12 KC |
720 | /** |
721 | * seccomp_set_mode_filter: internal function for setting seccomp filter | |
48dc92b9 | 722 | * @flags: flags to change filter behavior |
3b23dd12 KC |
723 | * @filter: struct sock_fprog containing filter |
724 | * | |
725 | * This function may be called repeatedly to install additional filters. | |
726 | * Every filter successfully installed will be evaluated (in reverse order) | |
727 | * for each system call the task makes. | |
728 | * | |
729 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
730 | * | |
731 | * Returns 0 on success or -EINVAL on failure. | |
732 | */ | |
48dc92b9 KC |
733 | static long seccomp_set_mode_filter(unsigned int flags, |
734 | const char __user *filter) | |
3b23dd12 KC |
735 | { |
736 | const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; | |
c8bee430 | 737 | struct seccomp_filter *prepared = NULL; |
3b23dd12 KC |
738 | long ret = -EINVAL; |
739 | ||
48dc92b9 | 740 | /* Validate flags. */ |
c2e1f2e3 | 741 | if (flags & ~SECCOMP_FILTER_FLAG_MASK) |
dbd95212 | 742 | return -EINVAL; |
48dc92b9 | 743 | |
c8bee430 KC |
744 | /* Prepare the new filter before holding any locks. */ |
745 | prepared = seccomp_prepare_user_filter(filter); | |
746 | if (IS_ERR(prepared)) | |
747 | return PTR_ERR(prepared); | |
748 | ||
c2e1f2e3 KC |
749 | /* |
750 | * Make sure we cannot change seccomp or nnp state via TSYNC | |
751 | * while another thread is in the middle of calling exec. | |
752 | */ | |
753 | if (flags & SECCOMP_FILTER_FLAG_TSYNC && | |
754 | mutex_lock_killable(¤t->signal->cred_guard_mutex)) | |
755 | goto out_free; | |
756 | ||
dbd95212 KC |
757 | spin_lock_irq(¤t->sighand->siglock); |
758 | ||
3b23dd12 KC |
759 | if (!seccomp_may_assign_mode(seccomp_mode)) |
760 | goto out; | |
761 | ||
c8bee430 | 762 | ret = seccomp_attach_filter(flags, prepared); |
3b23dd12 | 763 | if (ret) |
e2cfabdf | 764 | goto out; |
c8bee430 KC |
765 | /* Do not free the successfully attached filter. */ |
766 | prepared = NULL; | |
1d9d02fe | 767 | |
3ba2530c | 768 | seccomp_assign_mode(current, seccomp_mode); |
e2cfabdf | 769 | out: |
dbd95212 | 770 | spin_unlock_irq(¤t->sighand->siglock); |
c2e1f2e3 KC |
771 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
772 | mutex_unlock(¤t->signal->cred_guard_mutex); | |
773 | out_free: | |
c8bee430 | 774 | seccomp_filter_free(prepared); |
1d9d02fe AA |
775 | return ret; |
776 | } | |
3b23dd12 | 777 | #else |
48dc92b9 KC |
778 | static inline long seccomp_set_mode_filter(unsigned int flags, |
779 | const char __user *filter) | |
3b23dd12 KC |
780 | { |
781 | return -EINVAL; | |
782 | } | |
783 | #endif | |
d78ab02c | 784 | |
48dc92b9 KC |
785 | /* Common entry point for both prctl and syscall. */ |
786 | static long do_seccomp(unsigned int op, unsigned int flags, | |
787 | const char __user *uargs) | |
788 | { | |
789 | switch (op) { | |
790 | case SECCOMP_SET_MODE_STRICT: | |
791 | if (flags != 0 || uargs != NULL) | |
792 | return -EINVAL; | |
793 | return seccomp_set_mode_strict(); | |
794 | case SECCOMP_SET_MODE_FILTER: | |
795 | return seccomp_set_mode_filter(flags, uargs); | |
796 | default: | |
797 | return -EINVAL; | |
798 | } | |
799 | } | |
800 | ||
801 | SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, | |
802 | const char __user *, uargs) | |
803 | { | |
804 | return do_seccomp(op, flags, uargs); | |
805 | } | |
806 | ||
d78ab02c KC |
807 | /** |
808 | * prctl_set_seccomp: configures current->seccomp.mode | |
809 | * @seccomp_mode: requested mode to use | |
810 | * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER | |
811 | * | |
812 | * Returns 0 on success or -EINVAL on failure. | |
813 | */ | |
814 | long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter) | |
815 | { | |
48dc92b9 KC |
816 | unsigned int op; |
817 | char __user *uargs; | |
818 | ||
3b23dd12 KC |
819 | switch (seccomp_mode) { |
820 | case SECCOMP_MODE_STRICT: | |
48dc92b9 KC |
821 | op = SECCOMP_SET_MODE_STRICT; |
822 | /* | |
823 | * Setting strict mode through prctl always ignored filter, | |
824 | * so make sure it is always NULL here to pass the internal | |
825 | * check in do_seccomp(). | |
826 | */ | |
827 | uargs = NULL; | |
828 | break; | |
3b23dd12 | 829 | case SECCOMP_MODE_FILTER: |
48dc92b9 KC |
830 | op = SECCOMP_SET_MODE_FILTER; |
831 | uargs = filter; | |
832 | break; | |
3b23dd12 KC |
833 | default: |
834 | return -EINVAL; | |
835 | } | |
48dc92b9 KC |
836 | |
837 | /* prctl interface doesn't have flags, so they are always zero. */ | |
838 | return do_seccomp(op, 0, uargs); | |
d78ab02c | 839 | } |