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 | ||
0b5fa229 | 16 | #include <linux/refcount.h> |
85e7bac3 | 17 | #include <linux/audit.h> |
5b101740 | 18 | #include <linux/compat.h> |
b25e6716 | 19 | #include <linux/coredump.h> |
8e5f1ad1 | 20 | #include <linux/kmemleak.h> |
e2cfabdf | 21 | #include <linux/sched.h> |
68db0cf1 | 22 | #include <linux/sched/task_stack.h> |
e2cfabdf | 23 | #include <linux/seccomp.h> |
c8bee430 | 24 | #include <linux/slab.h> |
48dc92b9 | 25 | #include <linux/syscalls.h> |
8e5f1ad1 | 26 | #include <linux/sysctl.h> |
1da177e4 | 27 | |
a4412fc9 | 28 | #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER |
e2cfabdf | 29 | #include <asm/syscall.h> |
a4412fc9 | 30 | #endif |
e2cfabdf WD |
31 | |
32 | #ifdef CONFIG_SECCOMP_FILTER | |
e2cfabdf | 33 | #include <linux/filter.h> |
c2e1f2e3 | 34 | #include <linux/pid.h> |
fb0fadf9 | 35 | #include <linux/ptrace.h> |
e2cfabdf | 36 | #include <linux/security.h> |
e2cfabdf WD |
37 | #include <linux/tracehook.h> |
38 | #include <linux/uaccess.h> | |
39 | ||
40 | /** | |
41 | * struct seccomp_filter - container for seccomp BPF programs | |
42 | * | |
43 | * @usage: reference count to manage the object lifetime. | |
44 | * get/put helpers should be used when accessing an instance | |
45 | * outside of a lifetime-guarded section. In general, this | |
46 | * is only needed for handling filters shared across tasks. | |
47 | * @prev: points to a previously installed, or inherited, filter | |
285fdfc5 | 48 | * @prog: the BPF program to evaluate |
e2cfabdf WD |
49 | * |
50 | * seccomp_filter objects are organized in a tree linked via the @prev | |
51 | * pointer. For any task, it appears to be a singly-linked list starting | |
52 | * with current->seccomp.filter, the most recently attached or inherited filter. | |
53 | * However, multiple filters may share a @prev node, by way of fork(), which | |
54 | * results in a unidirectional tree existing in memory. This is similar to | |
55 | * how namespaces work. | |
56 | * | |
57 | * seccomp_filter objects should never be modified after being attached | |
58 | * to a task_struct (other than @usage). | |
59 | */ | |
60 | struct seccomp_filter { | |
0b5fa229 | 61 | refcount_t usage; |
e2cfabdf | 62 | struct seccomp_filter *prev; |
7ae457c1 | 63 | struct bpf_prog *prog; |
e2cfabdf WD |
64 | }; |
65 | ||
66 | /* Limit any path through the tree to 256KB worth of instructions. */ | |
67 | #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter)) | |
68 | ||
bd4cf0ed | 69 | /* |
e2cfabdf WD |
70 | * Endianness is explicitly ignored and left for BPF program authors to manage |
71 | * as per the specific architecture. | |
72 | */ | |
bd4cf0ed | 73 | static void populate_seccomp_data(struct seccomp_data *sd) |
e2cfabdf | 74 | { |
bd4cf0ed AS |
75 | struct task_struct *task = current; |
76 | struct pt_regs *regs = task_pt_regs(task); | |
2eac7648 | 77 | unsigned long args[6]; |
e2cfabdf | 78 | |
bd4cf0ed | 79 | sd->nr = syscall_get_nr(task, regs); |
0b747172 | 80 | sd->arch = syscall_get_arch(); |
2eac7648 DB |
81 | syscall_get_arguments(task, regs, 0, 6, args); |
82 | sd->args[0] = args[0]; | |
83 | sd->args[1] = args[1]; | |
84 | sd->args[2] = args[2]; | |
85 | sd->args[3] = args[3]; | |
86 | sd->args[4] = args[4]; | |
87 | sd->args[5] = args[5]; | |
bd4cf0ed | 88 | sd->instruction_pointer = KSTK_EIP(task); |
e2cfabdf WD |
89 | } |
90 | ||
91 | /** | |
92 | * seccomp_check_filter - verify seccomp filter code | |
93 | * @filter: filter to verify | |
94 | * @flen: length of filter | |
95 | * | |
4df95ff4 | 96 | * Takes a previously checked filter (by bpf_check_classic) and |
e2cfabdf WD |
97 | * redirects all filter code that loads struct sk_buff data |
98 | * and related data through seccomp_bpf_load. It also | |
99 | * enforces length and alignment checking of those loads. | |
100 | * | |
101 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
102 | */ | |
103 | static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen) | |
104 | { | |
105 | int pc; | |
106 | for (pc = 0; pc < flen; pc++) { | |
107 | struct sock_filter *ftest = &filter[pc]; | |
108 | u16 code = ftest->code; | |
109 | u32 k = ftest->k; | |
110 | ||
111 | switch (code) { | |
34805931 | 112 | case BPF_LD | BPF_W | BPF_ABS: |
bd4cf0ed | 113 | ftest->code = BPF_LDX | BPF_W | BPF_ABS; |
e2cfabdf WD |
114 | /* 32-bit aligned and not out of bounds. */ |
115 | if (k >= sizeof(struct seccomp_data) || k & 3) | |
116 | return -EINVAL; | |
117 | continue; | |
34805931 | 118 | case BPF_LD | BPF_W | BPF_LEN: |
bd4cf0ed | 119 | ftest->code = BPF_LD | BPF_IMM; |
e2cfabdf WD |
120 | ftest->k = sizeof(struct seccomp_data); |
121 | continue; | |
34805931 | 122 | case BPF_LDX | BPF_W | BPF_LEN: |
bd4cf0ed | 123 | ftest->code = BPF_LDX | BPF_IMM; |
e2cfabdf WD |
124 | ftest->k = sizeof(struct seccomp_data); |
125 | continue; | |
126 | /* Explicitly include allowed calls. */ | |
34805931 DB |
127 | case BPF_RET | BPF_K: |
128 | case BPF_RET | BPF_A: | |
129 | case BPF_ALU | BPF_ADD | BPF_K: | |
130 | case BPF_ALU | BPF_ADD | BPF_X: | |
131 | case BPF_ALU | BPF_SUB | BPF_K: | |
132 | case BPF_ALU | BPF_SUB | BPF_X: | |
133 | case BPF_ALU | BPF_MUL | BPF_K: | |
134 | case BPF_ALU | BPF_MUL | BPF_X: | |
135 | case BPF_ALU | BPF_DIV | BPF_K: | |
136 | case BPF_ALU | BPF_DIV | BPF_X: | |
137 | case BPF_ALU | BPF_AND | BPF_K: | |
138 | case BPF_ALU | BPF_AND | BPF_X: | |
139 | case BPF_ALU | BPF_OR | BPF_K: | |
140 | case BPF_ALU | BPF_OR | BPF_X: | |
141 | case BPF_ALU | BPF_XOR | BPF_K: | |
142 | case BPF_ALU | BPF_XOR | BPF_X: | |
143 | case BPF_ALU | BPF_LSH | BPF_K: | |
144 | case BPF_ALU | BPF_LSH | BPF_X: | |
145 | case BPF_ALU | BPF_RSH | BPF_K: | |
146 | case BPF_ALU | BPF_RSH | BPF_X: | |
147 | case BPF_ALU | BPF_NEG: | |
148 | case BPF_LD | BPF_IMM: | |
149 | case BPF_LDX | BPF_IMM: | |
150 | case BPF_MISC | BPF_TAX: | |
151 | case BPF_MISC | BPF_TXA: | |
152 | case BPF_LD | BPF_MEM: | |
153 | case BPF_LDX | BPF_MEM: | |
154 | case BPF_ST: | |
155 | case BPF_STX: | |
156 | case BPF_JMP | BPF_JA: | |
157 | case BPF_JMP | BPF_JEQ | BPF_K: | |
158 | case BPF_JMP | BPF_JEQ | BPF_X: | |
159 | case BPF_JMP | BPF_JGE | BPF_K: | |
160 | case BPF_JMP | BPF_JGE | BPF_X: | |
161 | case BPF_JMP | BPF_JGT | BPF_K: | |
162 | case BPF_JMP | BPF_JGT | BPF_X: | |
163 | case BPF_JMP | BPF_JSET | BPF_K: | |
164 | case BPF_JMP | BPF_JSET | BPF_X: | |
e2cfabdf WD |
165 | continue; |
166 | default: | |
167 | return -EINVAL; | |
168 | } | |
169 | } | |
170 | return 0; | |
171 | } | |
172 | ||
173 | /** | |
285fdfc5 MS |
174 | * seccomp_run_filters - evaluates all seccomp filters against @sd |
175 | * @sd: optional seccomp data to be passed to filters | |
deb4de8b KC |
176 | * @match: stores struct seccomp_filter that resulted in the return value, |
177 | * unless filter returned SECCOMP_RET_ALLOW, in which case it will | |
178 | * be unchanged. | |
e2cfabdf WD |
179 | * |
180 | * Returns valid seccomp BPF response codes. | |
181 | */ | |
deb4de8b KC |
182 | static u32 seccomp_run_filters(const struct seccomp_data *sd, |
183 | struct seccomp_filter **match) | |
e2cfabdf | 184 | { |
d39bd00d | 185 | struct seccomp_data sd_local; |
acf3b2c7 | 186 | u32 ret = SECCOMP_RET_ALLOW; |
8225d385 PK |
187 | /* Make sure cross-thread synced filter points somewhere sane. */ |
188 | struct seccomp_filter *f = | |
189 | lockless_dereference(current->seccomp.filter); | |
acf3b2c7 WD |
190 | |
191 | /* Ensure unexpected behavior doesn't result in failing open. */ | |
3ba2530c | 192 | if (unlikely(WARN_ON(f == NULL))) |
acf3b2c7 WD |
193 | return SECCOMP_RET_KILL; |
194 | ||
d39bd00d AL |
195 | if (!sd) { |
196 | populate_seccomp_data(&sd_local); | |
197 | sd = &sd_local; | |
198 | } | |
bd4cf0ed | 199 | |
e2cfabdf WD |
200 | /* |
201 | * All filters in the list are evaluated and the lowest BPF return | |
acf3b2c7 | 202 | * value always takes priority (ignoring the DATA). |
e2cfabdf | 203 | */ |
3ba2530c | 204 | for (; f; f = f->prev) { |
88575199 | 205 | u32 cur_ret = BPF_PROG_RUN(f->prog, sd); |
8f577cad | 206 | |
deb4de8b | 207 | if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION)) { |
acf3b2c7 | 208 | ret = cur_ret; |
deb4de8b KC |
209 | *match = f; |
210 | } | |
e2cfabdf WD |
211 | } |
212 | return ret; | |
213 | } | |
1f41b450 | 214 | #endif /* CONFIG_SECCOMP_FILTER */ |
e2cfabdf | 215 | |
1f41b450 KC |
216 | static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) |
217 | { | |
69f6a34b | 218 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 219 | |
1f41b450 KC |
220 | if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) |
221 | return false; | |
222 | ||
223 | return true; | |
224 | } | |
225 | ||
3ba2530c KC |
226 | static inline void seccomp_assign_mode(struct task_struct *task, |
227 | unsigned long seccomp_mode) | |
1f41b450 | 228 | { |
69f6a34b | 229 | assert_spin_locked(&task->sighand->siglock); |
dbd95212 | 230 | |
3ba2530c KC |
231 | task->seccomp.mode = seccomp_mode; |
232 | /* | |
233 | * Make sure TIF_SECCOMP cannot be set before the mode (and | |
234 | * filter) is set. | |
235 | */ | |
236 | smp_mb__before_atomic(); | |
237 | set_tsk_thread_flag(task, TIF_SECCOMP); | |
1f41b450 KC |
238 | } |
239 | ||
240 | #ifdef CONFIG_SECCOMP_FILTER | |
c2e1f2e3 KC |
241 | /* Returns 1 if the parent is an ancestor of the child. */ |
242 | static int is_ancestor(struct seccomp_filter *parent, | |
243 | struct seccomp_filter *child) | |
244 | { | |
245 | /* NULL is the root ancestor. */ | |
246 | if (parent == NULL) | |
247 | return 1; | |
248 | for (; child; child = child->prev) | |
249 | if (child == parent) | |
250 | return 1; | |
251 | return 0; | |
252 | } | |
253 | ||
254 | /** | |
255 | * seccomp_can_sync_threads: checks if all threads can be synchronized | |
256 | * | |
257 | * Expects sighand and cred_guard_mutex locks to be held. | |
258 | * | |
259 | * Returns 0 on success, -ve on error, or the pid of a thread which was | |
260 | * either not in the correct seccomp mode or it did not have an ancestral | |
261 | * seccomp filter. | |
262 | */ | |
263 | static inline pid_t seccomp_can_sync_threads(void) | |
264 | { | |
265 | struct task_struct *thread, *caller; | |
266 | ||
267 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 268 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
269 | |
270 | /* Validate all threads being eligible for synchronization. */ | |
271 | caller = current; | |
272 | for_each_thread(caller, thread) { | |
273 | pid_t failed; | |
274 | ||
275 | /* Skip current, since it is initiating the sync. */ | |
276 | if (thread == caller) | |
277 | continue; | |
278 | ||
279 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || | |
280 | (thread->seccomp.mode == SECCOMP_MODE_FILTER && | |
281 | is_ancestor(thread->seccomp.filter, | |
282 | caller->seccomp.filter))) | |
283 | continue; | |
284 | ||
285 | /* Return the first thread that cannot be synchronized. */ | |
286 | failed = task_pid_vnr(thread); | |
287 | /* If the pid cannot be resolved, then return -ESRCH */ | |
288 | if (unlikely(WARN_ON(failed == 0))) | |
289 | failed = -ESRCH; | |
290 | return failed; | |
291 | } | |
292 | ||
293 | return 0; | |
294 | } | |
295 | ||
296 | /** | |
297 | * seccomp_sync_threads: sets all threads to use current's filter | |
298 | * | |
299 | * Expects sighand and cred_guard_mutex locks to be held, and for | |
300 | * seccomp_can_sync_threads() to have returned success already | |
301 | * without dropping the locks. | |
302 | * | |
303 | */ | |
304 | static inline void seccomp_sync_threads(void) | |
305 | { | |
306 | struct task_struct *thread, *caller; | |
307 | ||
308 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 309 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
310 | |
311 | /* Synchronize all threads. */ | |
312 | caller = current; | |
313 | for_each_thread(caller, thread) { | |
314 | /* Skip current, since it needs no changes. */ | |
315 | if (thread == caller) | |
316 | continue; | |
317 | ||
318 | /* Get a task reference for the new leaf node. */ | |
319 | get_seccomp_filter(caller); | |
320 | /* | |
321 | * Drop the task reference to the shared ancestor since | |
322 | * current's path will hold a reference. (This also | |
323 | * allows a put before the assignment.) | |
324 | */ | |
325 | put_seccomp_filter(thread); | |
326 | smp_store_release(&thread->seccomp.filter, | |
327 | caller->seccomp.filter); | |
103502a3 JH |
328 | |
329 | /* | |
330 | * Don't let an unprivileged task work around | |
331 | * the no_new_privs restriction by creating | |
332 | * a thread that sets it up, enters seccomp, | |
333 | * then dies. | |
334 | */ | |
335 | if (task_no_new_privs(caller)) | |
336 | task_set_no_new_privs(thread); | |
337 | ||
c2e1f2e3 KC |
338 | /* |
339 | * Opt the other thread into seccomp if needed. | |
340 | * As threads are considered to be trust-realm | |
341 | * equivalent (see ptrace_may_access), it is safe to | |
342 | * allow one thread to transition the other. | |
343 | */ | |
103502a3 | 344 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) |
c2e1f2e3 | 345 | seccomp_assign_mode(thread, SECCOMP_MODE_FILTER); |
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 | |
0b5fa229 | 389 | refcount_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) | |
464 | seccomp_sync_threads(); | |
465 | ||
c8bee430 | 466 | return 0; |
e2cfabdf WD |
467 | } |
468 | ||
469 | /* get_seccomp_filter - increments the reference count of the filter on @tsk */ | |
470 | void get_seccomp_filter(struct task_struct *tsk) | |
471 | { | |
472 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
473 | if (!orig) | |
474 | return; | |
475 | /* Reference count is bounded by the number of total processes. */ | |
0b5fa229 | 476 | refcount_inc(&orig->usage); |
e2cfabdf WD |
477 | } |
478 | ||
c8bee430 KC |
479 | static inline void seccomp_filter_free(struct seccomp_filter *filter) |
480 | { | |
481 | if (filter) { | |
bab18991 | 482 | bpf_prog_destroy(filter->prog); |
c8bee430 KC |
483 | kfree(filter); |
484 | } | |
485 | } | |
486 | ||
e2cfabdf WD |
487 | /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */ |
488 | void put_seccomp_filter(struct task_struct *tsk) | |
489 | { | |
490 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
491 | /* Clean up single-reference branches iteratively. */ | |
0b5fa229 | 492 | while (orig && refcount_dec_and_test(&orig->usage)) { |
e2cfabdf WD |
493 | struct seccomp_filter *freeme = orig; |
494 | orig = orig->prev; | |
c8bee430 | 495 | seccomp_filter_free(freeme); |
e2cfabdf WD |
496 | } |
497 | } | |
bb6ea430 | 498 | |
b25e6716 MF |
499 | static void seccomp_init_siginfo(siginfo_t *info, int syscall, int reason) |
500 | { | |
501 | memset(info, 0, sizeof(*info)); | |
502 | info->si_signo = SIGSYS; | |
503 | info->si_code = SYS_SECCOMP; | |
504 | info->si_call_addr = (void __user *)KSTK_EIP(current); | |
505 | info->si_errno = reason; | |
506 | info->si_arch = syscall_get_arch(); | |
507 | info->si_syscall = syscall; | |
508 | } | |
509 | ||
bb6ea430 WD |
510 | /** |
511 | * seccomp_send_sigsys - signals the task to allow in-process syscall emulation | |
512 | * @syscall: syscall number to send to userland | |
513 | * @reason: filter-supplied reason code to send to userland (via si_errno) | |
514 | * | |
515 | * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info. | |
516 | */ | |
517 | static void seccomp_send_sigsys(int syscall, int reason) | |
518 | { | |
519 | struct siginfo info; | |
b25e6716 | 520 | seccomp_init_siginfo(&info, syscall, reason); |
bb6ea430 WD |
521 | force_sig_info(SIGSYS, &info, current); |
522 | } | |
e2cfabdf | 523 | #endif /* CONFIG_SECCOMP_FILTER */ |
1da177e4 LT |
524 | |
525 | /* | |
526 | * Secure computing mode 1 allows only read/write/exit/sigreturn. | |
527 | * To be fully secure this must be combined with rlimit | |
528 | * to limit the stack allocations too. | |
529 | */ | |
cb4253aa | 530 | static const int mode1_syscalls[] = { |
1da177e4 LT |
531 | __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, |
532 | 0, /* null terminated */ | |
533 | }; | |
534 | ||
a4412fc9 | 535 | static void __secure_computing_strict(int this_syscall) |
1da177e4 | 536 | { |
cb4253aa | 537 | const int *syscall_whitelist = mode1_syscalls; |
a4412fc9 | 538 | #ifdef CONFIG_COMPAT |
5c38065e | 539 | if (in_compat_syscall()) |
c983f0e8 | 540 | syscall_whitelist = get_compat_mode1_syscalls(); |
a4412fc9 AL |
541 | #endif |
542 | do { | |
543 | if (*syscall_whitelist == this_syscall) | |
544 | return; | |
545 | } while (*++syscall_whitelist); | |
546 | ||
547 | #ifdef SECCOMP_DEBUG | |
548 | dump_stack(); | |
549 | #endif | |
550 | audit_seccomp(this_syscall, SIGKILL, SECCOMP_RET_KILL); | |
551 | do_exit(SIGKILL); | |
552 | } | |
553 | ||
554 | #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER | |
555 | void secure_computing_strict(int this_syscall) | |
556 | { | |
557 | int mode = current->seccomp.mode; | |
558 | ||
97f2645f | 559 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 TA |
560 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
561 | return; | |
562 | ||
221272f9 | 563 | if (mode == SECCOMP_MODE_DISABLED) |
a4412fc9 AL |
564 | return; |
565 | else if (mode == SECCOMP_MODE_STRICT) | |
566 | __secure_computing_strict(this_syscall); | |
567 | else | |
568 | BUG(); | |
569 | } | |
570 | #else | |
13aa72f0 AL |
571 | |
572 | #ifdef CONFIG_SECCOMP_FILTER | |
ce6526e8 KC |
573 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
574 | const bool recheck_after_trace) | |
13aa72f0 AL |
575 | { |
576 | u32 filter_ret, action; | |
deb4de8b | 577 | struct seccomp_filter *match = NULL; |
13aa72f0 | 578 | int data; |
1da177e4 | 579 | |
3ba2530c KC |
580 | /* |
581 | * Make sure that any changes to mode from another thread have | |
582 | * been seen after TIF_SECCOMP was seen. | |
583 | */ | |
584 | rmb(); | |
585 | ||
deb4de8b | 586 | filter_ret = seccomp_run_filters(sd, &match); |
13aa72f0 AL |
587 | data = filter_ret & SECCOMP_RET_DATA; |
588 | action = filter_ret & SECCOMP_RET_ACTION; | |
589 | ||
590 | switch (action) { | |
591 | case SECCOMP_RET_ERRNO: | |
580c57f1 KC |
592 | /* Set low-order bits as an errno, capped at MAX_ERRNO. */ |
593 | if (data > MAX_ERRNO) | |
594 | data = MAX_ERRNO; | |
d39bd00d | 595 | syscall_set_return_value(current, task_pt_regs(current), |
13aa72f0 AL |
596 | -data, 0); |
597 | goto skip; | |
598 | ||
599 | case SECCOMP_RET_TRAP: | |
600 | /* Show the handler the original registers. */ | |
d39bd00d | 601 | syscall_rollback(current, task_pt_regs(current)); |
13aa72f0 AL |
602 | /* Let the filter pass back 16 bits of data. */ |
603 | seccomp_send_sigsys(this_syscall, data); | |
604 | goto skip; | |
605 | ||
606 | case SECCOMP_RET_TRACE: | |
ce6526e8 KC |
607 | /* We've been put in this state by the ptracer already. */ |
608 | if (recheck_after_trace) | |
609 | return 0; | |
610 | ||
8112c4f1 KC |
611 | /* ENOSYS these calls if there is no tracer attached. */ |
612 | if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { | |
613 | syscall_set_return_value(current, | |
614 | task_pt_regs(current), | |
615 | -ENOSYS, 0); | |
616 | goto skip; | |
617 | } | |
618 | ||
619 | /* Allow the BPF to provide the event message */ | |
620 | ptrace_event(PTRACE_EVENT_SECCOMP, data); | |
621 | /* | |
622 | * The delivery of a fatal signal during event | |
485a252a KC |
623 | * notification may silently skip tracer notification, |
624 | * which could leave us with a potentially unmodified | |
625 | * syscall that the tracer would have liked to have | |
626 | * changed. Since the process is about to die, we just | |
627 | * force the syscall to be skipped and let the signal | |
628 | * kill the process and correctly handle any tracer exit | |
629 | * notifications. | |
8112c4f1 KC |
630 | */ |
631 | if (fatal_signal_pending(current)) | |
485a252a | 632 | goto skip; |
8112c4f1 KC |
633 | /* Check if the tracer forced the syscall to be skipped. */ |
634 | this_syscall = syscall_get_nr(current, task_pt_regs(current)); | |
635 | if (this_syscall < 0) | |
636 | goto skip; | |
637 | ||
ce6526e8 KC |
638 | /* |
639 | * Recheck the syscall, since it may have changed. This | |
640 | * intentionally uses a NULL struct seccomp_data to force | |
641 | * a reload of all registers. This does not goto skip since | |
642 | * a skip would have already been reported. | |
643 | */ | |
644 | if (__seccomp_filter(this_syscall, NULL, true)) | |
645 | return -1; | |
646 | ||
8112c4f1 | 647 | return 0; |
13aa72f0 AL |
648 | |
649 | case SECCOMP_RET_ALLOW: | |
deb4de8b KC |
650 | /* |
651 | * Note that the "match" filter will always be NULL for | |
652 | * this action since SECCOMP_RET_ALLOW is the starting | |
653 | * state in seccomp_run_filters(). | |
654 | */ | |
8112c4f1 | 655 | return 0; |
13aa72f0 AL |
656 | |
657 | case SECCOMP_RET_KILL: | |
131b6351 | 658 | default: |
13aa72f0 | 659 | audit_seccomp(this_syscall, SIGSYS, action); |
d7276e32 KC |
660 | /* Dump core only if this is the last remaining thread. */ |
661 | if (get_nr_threads(current) == 1) { | |
131b6351 KC |
662 | siginfo_t info; |
663 | ||
d7276e32 KC |
664 | /* Show the original registers in the dump. */ |
665 | syscall_rollback(current, task_pt_regs(current)); | |
666 | /* Trigger a manual coredump since do_exit skips it. */ | |
667 | seccomp_init_siginfo(&info, this_syscall, data); | |
668 | do_coredump(&info); | |
669 | } | |
13aa72f0 AL |
670 | do_exit(SIGSYS); |
671 | } | |
672 | ||
673 | unreachable(); | |
674 | ||
675 | skip: | |
676 | audit_seccomp(this_syscall, 0, action); | |
8112c4f1 KC |
677 | return -1; |
678 | } | |
679 | #else | |
ce6526e8 KC |
680 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
681 | const bool recheck_after_trace) | |
8112c4f1 KC |
682 | { |
683 | BUG(); | |
13aa72f0 | 684 | } |
1da177e4 | 685 | #endif |
13aa72f0 | 686 | |
8112c4f1 | 687 | int __secure_computing(const struct seccomp_data *sd) |
13aa72f0 AL |
688 | { |
689 | int mode = current->seccomp.mode; | |
8112c4f1 | 690 | int this_syscall; |
13aa72f0 | 691 | |
97f2645f | 692 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 | 693 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
8112c4f1 KC |
694 | return 0; |
695 | ||
696 | this_syscall = sd ? sd->nr : | |
697 | syscall_get_nr(current, task_pt_regs(current)); | |
13c4a901 | 698 | |
13aa72f0 | 699 | switch (mode) { |
e2cfabdf | 700 | case SECCOMP_MODE_STRICT: |
13aa72f0 | 701 | __secure_computing_strict(this_syscall); /* may call do_exit */ |
8112c4f1 | 702 | return 0; |
13aa72f0 | 703 | case SECCOMP_MODE_FILTER: |
ce6526e8 | 704 | return __seccomp_filter(this_syscall, sd, false); |
1da177e4 LT |
705 | default: |
706 | BUG(); | |
707 | } | |
13aa72f0 | 708 | } |
a4412fc9 | 709 | #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ |
1d9d02fe AA |
710 | |
711 | long prctl_get_seccomp(void) | |
712 | { | |
713 | return current->seccomp.mode; | |
714 | } | |
715 | ||
e2cfabdf | 716 | /** |
3b23dd12 | 717 | * seccomp_set_mode_strict: internal function for setting strict seccomp |
e2cfabdf WD |
718 | * |
719 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
720 | * | |
721 | * Returns 0 on success or -EINVAL on failure. | |
722 | */ | |
3b23dd12 | 723 | static long seccomp_set_mode_strict(void) |
1d9d02fe | 724 | { |
3b23dd12 | 725 | const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; |
e2cfabdf | 726 | long ret = -EINVAL; |
1d9d02fe | 727 | |
dbd95212 KC |
728 | spin_lock_irq(¤t->sighand->siglock); |
729 | ||
1f41b450 | 730 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1d9d02fe AA |
731 | goto out; |
732 | ||
cf99abac | 733 | #ifdef TIF_NOTSC |
3b23dd12 | 734 | disable_TSC(); |
cf99abac | 735 | #endif |
3ba2530c | 736 | seccomp_assign_mode(current, seccomp_mode); |
3b23dd12 KC |
737 | ret = 0; |
738 | ||
739 | out: | |
dbd95212 | 740 | spin_unlock_irq(¤t->sighand->siglock); |
3b23dd12 KC |
741 | |
742 | return ret; | |
743 | } | |
744 | ||
e2cfabdf | 745 | #ifdef CONFIG_SECCOMP_FILTER |
3b23dd12 KC |
746 | /** |
747 | * seccomp_set_mode_filter: internal function for setting seccomp filter | |
48dc92b9 | 748 | * @flags: flags to change filter behavior |
3b23dd12 KC |
749 | * @filter: struct sock_fprog containing filter |
750 | * | |
751 | * This function may be called repeatedly to install additional filters. | |
752 | * Every filter successfully installed will be evaluated (in reverse order) | |
753 | * for each system call the task makes. | |
754 | * | |
755 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
756 | * | |
757 | * Returns 0 on success or -EINVAL on failure. | |
758 | */ | |
48dc92b9 KC |
759 | static long seccomp_set_mode_filter(unsigned int flags, |
760 | const char __user *filter) | |
3b23dd12 KC |
761 | { |
762 | const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; | |
c8bee430 | 763 | struct seccomp_filter *prepared = NULL; |
3b23dd12 KC |
764 | long ret = -EINVAL; |
765 | ||
48dc92b9 | 766 | /* Validate flags. */ |
c2e1f2e3 | 767 | if (flags & ~SECCOMP_FILTER_FLAG_MASK) |
dbd95212 | 768 | return -EINVAL; |
48dc92b9 | 769 | |
c8bee430 KC |
770 | /* Prepare the new filter before holding any locks. */ |
771 | prepared = seccomp_prepare_user_filter(filter); | |
772 | if (IS_ERR(prepared)) | |
773 | return PTR_ERR(prepared); | |
774 | ||
c2e1f2e3 KC |
775 | /* |
776 | * Make sure we cannot change seccomp or nnp state via TSYNC | |
777 | * while another thread is in the middle of calling exec. | |
778 | */ | |
779 | if (flags & SECCOMP_FILTER_FLAG_TSYNC && | |
780 | mutex_lock_killable(¤t->signal->cred_guard_mutex)) | |
781 | goto out_free; | |
782 | ||
dbd95212 KC |
783 | spin_lock_irq(¤t->sighand->siglock); |
784 | ||
3b23dd12 KC |
785 | if (!seccomp_may_assign_mode(seccomp_mode)) |
786 | goto out; | |
787 | ||
c8bee430 | 788 | ret = seccomp_attach_filter(flags, prepared); |
3b23dd12 | 789 | if (ret) |
e2cfabdf | 790 | goto out; |
c8bee430 KC |
791 | /* Do not free the successfully attached filter. */ |
792 | prepared = NULL; | |
1d9d02fe | 793 | |
3ba2530c | 794 | seccomp_assign_mode(current, seccomp_mode); |
e2cfabdf | 795 | out: |
dbd95212 | 796 | spin_unlock_irq(¤t->sighand->siglock); |
c2e1f2e3 KC |
797 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
798 | mutex_unlock(¤t->signal->cred_guard_mutex); | |
799 | out_free: | |
c8bee430 | 800 | seccomp_filter_free(prepared); |
1d9d02fe AA |
801 | return ret; |
802 | } | |
3b23dd12 | 803 | #else |
48dc92b9 KC |
804 | static inline long seccomp_set_mode_filter(unsigned int flags, |
805 | const char __user *filter) | |
3b23dd12 KC |
806 | { |
807 | return -EINVAL; | |
808 | } | |
809 | #endif | |
d78ab02c | 810 | |
48dc92b9 KC |
811 | /* Common entry point for both prctl and syscall. */ |
812 | static long do_seccomp(unsigned int op, unsigned int flags, | |
813 | const char __user *uargs) | |
814 | { | |
815 | switch (op) { | |
816 | case SECCOMP_SET_MODE_STRICT: | |
817 | if (flags != 0 || uargs != NULL) | |
818 | return -EINVAL; | |
819 | return seccomp_set_mode_strict(); | |
820 | case SECCOMP_SET_MODE_FILTER: | |
821 | return seccomp_set_mode_filter(flags, uargs); | |
822 | default: | |
823 | return -EINVAL; | |
824 | } | |
825 | } | |
826 | ||
827 | SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, | |
828 | const char __user *, uargs) | |
829 | { | |
830 | return do_seccomp(op, flags, uargs); | |
831 | } | |
832 | ||
d78ab02c KC |
833 | /** |
834 | * prctl_set_seccomp: configures current->seccomp.mode | |
835 | * @seccomp_mode: requested mode to use | |
836 | * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER | |
837 | * | |
838 | * Returns 0 on success or -EINVAL on failure. | |
839 | */ | |
840 | long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter) | |
841 | { | |
48dc92b9 KC |
842 | unsigned int op; |
843 | char __user *uargs; | |
844 | ||
3b23dd12 KC |
845 | switch (seccomp_mode) { |
846 | case SECCOMP_MODE_STRICT: | |
48dc92b9 KC |
847 | op = SECCOMP_SET_MODE_STRICT; |
848 | /* | |
849 | * Setting strict mode through prctl always ignored filter, | |
850 | * so make sure it is always NULL here to pass the internal | |
851 | * check in do_seccomp(). | |
852 | */ | |
853 | uargs = NULL; | |
854 | break; | |
3b23dd12 | 855 | case SECCOMP_MODE_FILTER: |
48dc92b9 KC |
856 | op = SECCOMP_SET_MODE_FILTER; |
857 | uargs = filter; | |
858 | break; | |
3b23dd12 KC |
859 | default: |
860 | return -EINVAL; | |
861 | } | |
48dc92b9 KC |
862 | |
863 | /* prctl interface doesn't have flags, so they are always zero. */ | |
864 | return do_seccomp(op, 0, uargs); | |
d78ab02c | 865 | } |
f8e529ed TA |
866 | |
867 | #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE) | |
868 | long seccomp_get_filter(struct task_struct *task, unsigned long filter_off, | |
869 | void __user *data) | |
870 | { | |
871 | struct seccomp_filter *filter; | |
872 | struct sock_fprog_kern *fprog; | |
873 | long ret; | |
874 | unsigned long count = 0; | |
875 | ||
876 | if (!capable(CAP_SYS_ADMIN) || | |
877 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { | |
878 | return -EACCES; | |
879 | } | |
880 | ||
881 | spin_lock_irq(&task->sighand->siglock); | |
882 | if (task->seccomp.mode != SECCOMP_MODE_FILTER) { | |
883 | ret = -EINVAL; | |
884 | goto out; | |
885 | } | |
886 | ||
887 | filter = task->seccomp.filter; | |
888 | while (filter) { | |
889 | filter = filter->prev; | |
890 | count++; | |
891 | } | |
892 | ||
893 | if (filter_off >= count) { | |
894 | ret = -ENOENT; | |
895 | goto out; | |
896 | } | |
897 | count -= filter_off; | |
898 | ||
899 | filter = task->seccomp.filter; | |
900 | while (filter && count > 1) { | |
901 | filter = filter->prev; | |
902 | count--; | |
903 | } | |
904 | ||
905 | if (WARN_ON(count != 1 || !filter)) { | |
906 | /* The filter tree shouldn't shrink while we're using it. */ | |
907 | ret = -ENOENT; | |
908 | goto out; | |
909 | } | |
910 | ||
911 | fprog = filter->prog->orig_prog; | |
912 | if (!fprog) { | |
470bf1f2 | 913 | /* This must be a new non-cBPF filter, since we save |
f8e529ed TA |
914 | * every cBPF filter's orig_prog above when |
915 | * CONFIG_CHECKPOINT_RESTORE is enabled. | |
916 | */ | |
917 | ret = -EMEDIUMTYPE; | |
918 | goto out; | |
919 | } | |
920 | ||
921 | ret = fprog->len; | |
922 | if (!data) | |
923 | goto out; | |
924 | ||
925 | get_seccomp_filter(task); | |
926 | spin_unlock_irq(&task->sighand->siglock); | |
927 | ||
928 | if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog))) | |
929 | ret = -EFAULT; | |
930 | ||
931 | put_seccomp_filter(task); | |
932 | return ret; | |
933 | ||
934 | out: | |
935 | spin_unlock_irq(&task->sighand->siglock); | |
936 | return ret; | |
937 | } | |
938 | #endif | |
8e5f1ad1 TH |
939 | |
940 | #ifdef CONFIG_SYSCTL | |
941 | ||
942 | /* Human readable action names for friendly sysctl interaction */ | |
943 | #define SECCOMP_RET_KILL_NAME "kill" | |
944 | #define SECCOMP_RET_TRAP_NAME "trap" | |
945 | #define SECCOMP_RET_ERRNO_NAME "errno" | |
946 | #define SECCOMP_RET_TRACE_NAME "trace" | |
947 | #define SECCOMP_RET_ALLOW_NAME "allow" | |
948 | ||
949 | static const char seccomp_actions_avail[] = SECCOMP_RET_KILL_NAME " " | |
950 | SECCOMP_RET_TRAP_NAME " " | |
951 | SECCOMP_RET_ERRNO_NAME " " | |
952 | SECCOMP_RET_TRACE_NAME " " | |
953 | SECCOMP_RET_ALLOW_NAME; | |
954 | ||
955 | static struct ctl_path seccomp_sysctl_path[] = { | |
956 | { .procname = "kernel", }, | |
957 | { .procname = "seccomp", }, | |
958 | { } | |
959 | }; | |
960 | ||
961 | static struct ctl_table seccomp_sysctl_table[] = { | |
962 | { | |
963 | .procname = "actions_avail", | |
964 | .data = (void *) &seccomp_actions_avail, | |
965 | .maxlen = sizeof(seccomp_actions_avail), | |
966 | .mode = 0444, | |
967 | .proc_handler = proc_dostring, | |
968 | }, | |
969 | { } | |
970 | }; | |
971 | ||
972 | static int __init seccomp_sysctl_init(void) | |
973 | { | |
974 | struct ctl_table_header *hdr; | |
975 | ||
976 | hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table); | |
977 | if (!hdr) | |
978 | pr_warn("seccomp: sysctl registration failed\n"); | |
979 | else | |
980 | kmemleak_not_leak(hdr); | |
981 | ||
982 | return 0; | |
983 | } | |
984 | ||
985 | device_initcall(seccomp_sysctl_init) | |
986 | ||
987 | #endif /* CONFIG_SYSCTL */ |