Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/kernel/seccomp.c | |
3 | * | |
4 | * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com> | |
5 | * | |
e2cfabdf WD |
6 | * Copyright (C) 2012 Google, Inc. |
7 | * Will Drewry <wad@chromium.org> | |
8 | * | |
9 | * This defines a simple but solid secure-computing facility. | |
10 | * | |
11 | * Mode 1 uses a fixed list of allowed system calls. | |
12 | * Mode 2 allows user-defined system call filters in the form | |
13 | * of Berkeley Packet Filters/Linux Socket Filters. | |
1da177e4 LT |
14 | */ |
15 | ||
e2cfabdf | 16 | #include <linux/atomic.h> |
85e7bac3 | 17 | #include <linux/audit.h> |
5b101740 | 18 | #include <linux/compat.h> |
e2cfabdf WD |
19 | #include <linux/sched.h> |
20 | #include <linux/seccomp.h> | |
c8bee430 | 21 | #include <linux/slab.h> |
48dc92b9 | 22 | #include <linux/syscalls.h> |
1da177e4 | 23 | |
a4412fc9 | 24 | #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER |
e2cfabdf | 25 | #include <asm/syscall.h> |
a4412fc9 | 26 | #endif |
e2cfabdf WD |
27 | |
28 | #ifdef CONFIG_SECCOMP_FILTER | |
e2cfabdf | 29 | #include <linux/filter.h> |
c2e1f2e3 | 30 | #include <linux/pid.h> |
fb0fadf9 | 31 | #include <linux/ptrace.h> |
e2cfabdf | 32 | #include <linux/security.h> |
e2cfabdf WD |
33 | #include <linux/tracehook.h> |
34 | #include <linux/uaccess.h> | |
35 | ||
36 | /** | |
37 | * struct seccomp_filter - container for seccomp BPF programs | |
38 | * | |
39 | * @usage: reference count to manage the object lifetime. | |
40 | * get/put helpers should be used when accessing an instance | |
41 | * outside of a lifetime-guarded section. In general, this | |
42 | * is only needed for handling filters shared across tasks. | |
43 | * @prev: points to a previously installed, or inherited, filter | |
44 | * @len: the number of instructions in the program | |
119ce5c8 | 45 | * @insnsi: the BPF program instructions to evaluate |
e2cfabdf WD |
46 | * |
47 | * seccomp_filter objects are organized in a tree linked via the @prev | |
48 | * pointer. For any task, it appears to be a singly-linked list starting | |
49 | * with current->seccomp.filter, the most recently attached or inherited filter. | |
50 | * However, multiple filters may share a @prev node, by way of fork(), which | |
51 | * results in a unidirectional tree existing in memory. This is similar to | |
52 | * how namespaces work. | |
53 | * | |
54 | * seccomp_filter objects should never be modified after being attached | |
55 | * to a task_struct (other than @usage). | |
56 | */ | |
57 | struct seccomp_filter { | |
58 | atomic_t usage; | |
59 | struct seccomp_filter *prev; | |
7ae457c1 | 60 | struct bpf_prog *prog; |
e2cfabdf WD |
61 | }; |
62 | ||
63 | /* Limit any path through the tree to 256KB worth of instructions. */ | |
64 | #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter)) | |
65 | ||
bd4cf0ed | 66 | /* |
e2cfabdf WD |
67 | * Endianness is explicitly ignored and left for BPF program authors to manage |
68 | * as per the specific architecture. | |
69 | */ | |
bd4cf0ed | 70 | static void populate_seccomp_data(struct seccomp_data *sd) |
e2cfabdf | 71 | { |
bd4cf0ed AS |
72 | struct task_struct *task = current; |
73 | struct pt_regs *regs = task_pt_regs(task); | |
2eac7648 | 74 | unsigned long args[6]; |
e2cfabdf | 75 | |
bd4cf0ed | 76 | sd->nr = syscall_get_nr(task, regs); |
0b747172 | 77 | sd->arch = syscall_get_arch(); |
2eac7648 DB |
78 | syscall_get_arguments(task, regs, 0, 6, args); |
79 | sd->args[0] = args[0]; | |
80 | sd->args[1] = args[1]; | |
81 | sd->args[2] = args[2]; | |
82 | sd->args[3] = args[3]; | |
83 | sd->args[4] = args[4]; | |
84 | sd->args[5] = args[5]; | |
bd4cf0ed | 85 | sd->instruction_pointer = KSTK_EIP(task); |
e2cfabdf WD |
86 | } |
87 | ||
88 | /** | |
89 | * seccomp_check_filter - verify seccomp filter code | |
90 | * @filter: filter to verify | |
91 | * @flen: length of filter | |
92 | * | |
4df95ff4 | 93 | * Takes a previously checked filter (by bpf_check_classic) and |
e2cfabdf WD |
94 | * redirects all filter code that loads struct sk_buff data |
95 | * and related data through seccomp_bpf_load. It also | |
96 | * enforces length and alignment checking of those loads. | |
97 | * | |
98 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
99 | */ | |
100 | static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen) | |
101 | { | |
102 | int pc; | |
103 | for (pc = 0; pc < flen; pc++) { | |
104 | struct sock_filter *ftest = &filter[pc]; | |
105 | u16 code = ftest->code; | |
106 | u32 k = ftest->k; | |
107 | ||
108 | switch (code) { | |
34805931 | 109 | case BPF_LD | BPF_W | BPF_ABS: |
bd4cf0ed | 110 | ftest->code = BPF_LDX | BPF_W | BPF_ABS; |
e2cfabdf WD |
111 | /* 32-bit aligned and not out of bounds. */ |
112 | if (k >= sizeof(struct seccomp_data) || k & 3) | |
113 | return -EINVAL; | |
114 | continue; | |
34805931 | 115 | case BPF_LD | BPF_W | BPF_LEN: |
bd4cf0ed | 116 | ftest->code = BPF_LD | BPF_IMM; |
e2cfabdf WD |
117 | ftest->k = sizeof(struct seccomp_data); |
118 | continue; | |
34805931 | 119 | case BPF_LDX | BPF_W | BPF_LEN: |
bd4cf0ed | 120 | ftest->code = BPF_LDX | BPF_IMM; |
e2cfabdf WD |
121 | ftest->k = sizeof(struct seccomp_data); |
122 | continue; | |
123 | /* Explicitly include allowed calls. */ | |
34805931 DB |
124 | case BPF_RET | BPF_K: |
125 | case BPF_RET | BPF_A: | |
126 | case BPF_ALU | BPF_ADD | BPF_K: | |
127 | case BPF_ALU | BPF_ADD | BPF_X: | |
128 | case BPF_ALU | BPF_SUB | BPF_K: | |
129 | case BPF_ALU | BPF_SUB | BPF_X: | |
130 | case BPF_ALU | BPF_MUL | BPF_K: | |
131 | case BPF_ALU | BPF_MUL | BPF_X: | |
132 | case BPF_ALU | BPF_DIV | BPF_K: | |
133 | case BPF_ALU | BPF_DIV | BPF_X: | |
134 | case BPF_ALU | BPF_AND | BPF_K: | |
135 | case BPF_ALU | BPF_AND | BPF_X: | |
136 | case BPF_ALU | BPF_OR | BPF_K: | |
137 | case BPF_ALU | BPF_OR | BPF_X: | |
138 | case BPF_ALU | BPF_XOR | BPF_K: | |
139 | case BPF_ALU | BPF_XOR | BPF_X: | |
140 | case BPF_ALU | BPF_LSH | BPF_K: | |
141 | case BPF_ALU | BPF_LSH | BPF_X: | |
142 | case BPF_ALU | BPF_RSH | BPF_K: | |
143 | case BPF_ALU | BPF_RSH | BPF_X: | |
144 | case BPF_ALU | BPF_NEG: | |
145 | case BPF_LD | BPF_IMM: | |
146 | case BPF_LDX | BPF_IMM: | |
147 | case BPF_MISC | BPF_TAX: | |
148 | case BPF_MISC | BPF_TXA: | |
149 | case BPF_LD | BPF_MEM: | |
150 | case BPF_LDX | BPF_MEM: | |
151 | case BPF_ST: | |
152 | case BPF_STX: | |
153 | case BPF_JMP | BPF_JA: | |
154 | case BPF_JMP | BPF_JEQ | BPF_K: | |
155 | case BPF_JMP | BPF_JEQ | BPF_X: | |
156 | case BPF_JMP | BPF_JGE | BPF_K: | |
157 | case BPF_JMP | BPF_JGE | BPF_X: | |
158 | case BPF_JMP | BPF_JGT | BPF_K: | |
159 | case BPF_JMP | BPF_JGT | BPF_X: | |
160 | case BPF_JMP | BPF_JSET | BPF_K: | |
161 | case BPF_JMP | BPF_JSET | BPF_X: | |
e2cfabdf WD |
162 | continue; |
163 | default: | |
164 | return -EINVAL; | |
165 | } | |
166 | } | |
167 | return 0; | |
168 | } | |
169 | ||
170 | /** | |
171 | * seccomp_run_filters - evaluates all seccomp filters against @syscall | |
172 | * @syscall: number of the current system call | |
173 | * | |
174 | * Returns valid seccomp BPF response codes. | |
175 | */ | |
d39bd00d | 176 | static u32 seccomp_run_filters(struct seccomp_data *sd) |
e2cfabdf | 177 | { |
3ba2530c | 178 | struct seccomp_filter *f = ACCESS_ONCE(current->seccomp.filter); |
d39bd00d | 179 | struct seccomp_data sd_local; |
acf3b2c7 WD |
180 | u32 ret = SECCOMP_RET_ALLOW; |
181 | ||
182 | /* Ensure unexpected behavior doesn't result in failing open. */ | |
3ba2530c | 183 | if (unlikely(WARN_ON(f == NULL))) |
acf3b2c7 WD |
184 | return SECCOMP_RET_KILL; |
185 | ||
3ba2530c KC |
186 | /* Make sure cross-thread synced filter points somewhere sane. */ |
187 | smp_read_barrier_depends(); | |
188 | ||
d39bd00d AL |
189 | if (!sd) { |
190 | populate_seccomp_data(&sd_local); | |
191 | sd = &sd_local; | |
192 | } | |
bd4cf0ed | 193 | |
e2cfabdf WD |
194 | /* |
195 | * All filters in the list are evaluated and the lowest BPF return | |
acf3b2c7 | 196 | * value always takes priority (ignoring the DATA). |
e2cfabdf | 197 | */ |
3ba2530c | 198 | for (; f; f = f->prev) { |
d39bd00d | 199 | u32 cur_ret = BPF_PROG_RUN(f->prog, (void *)sd); |
8f577cad | 200 | |
acf3b2c7 WD |
201 | if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION)) |
202 | ret = cur_ret; | |
e2cfabdf WD |
203 | } |
204 | return ret; | |
205 | } | |
1f41b450 | 206 | #endif /* CONFIG_SECCOMP_FILTER */ |
e2cfabdf | 207 | |
1f41b450 KC |
208 | static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) |
209 | { | |
69f6a34b | 210 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 211 | |
1f41b450 KC |
212 | if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) |
213 | return false; | |
214 | ||
215 | return true; | |
216 | } | |
217 | ||
3ba2530c KC |
218 | static inline void seccomp_assign_mode(struct task_struct *task, |
219 | unsigned long seccomp_mode) | |
1f41b450 | 220 | { |
69f6a34b | 221 | assert_spin_locked(&task->sighand->siglock); |
dbd95212 | 222 | |
3ba2530c KC |
223 | task->seccomp.mode = seccomp_mode; |
224 | /* | |
225 | * Make sure TIF_SECCOMP cannot be set before the mode (and | |
226 | * filter) is set. | |
227 | */ | |
228 | smp_mb__before_atomic(); | |
229 | set_tsk_thread_flag(task, TIF_SECCOMP); | |
1f41b450 KC |
230 | } |
231 | ||
232 | #ifdef CONFIG_SECCOMP_FILTER | |
c2e1f2e3 KC |
233 | /* Returns 1 if the parent is an ancestor of the child. */ |
234 | static int is_ancestor(struct seccomp_filter *parent, | |
235 | struct seccomp_filter *child) | |
236 | { | |
237 | /* NULL is the root ancestor. */ | |
238 | if (parent == NULL) | |
239 | return 1; | |
240 | for (; child; child = child->prev) | |
241 | if (child == parent) | |
242 | return 1; | |
243 | return 0; | |
244 | } | |
245 | ||
246 | /** | |
247 | * seccomp_can_sync_threads: checks if all threads can be synchronized | |
248 | * | |
249 | * Expects sighand and cred_guard_mutex locks to be held. | |
250 | * | |
251 | * Returns 0 on success, -ve on error, or the pid of a thread which was | |
252 | * either not in the correct seccomp mode or it did not have an ancestral | |
253 | * seccomp filter. | |
254 | */ | |
255 | static inline pid_t seccomp_can_sync_threads(void) | |
256 | { | |
257 | struct task_struct *thread, *caller; | |
258 | ||
259 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 260 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
261 | |
262 | /* Validate all threads being eligible for synchronization. */ | |
263 | caller = current; | |
264 | for_each_thread(caller, thread) { | |
265 | pid_t failed; | |
266 | ||
267 | /* Skip current, since it is initiating the sync. */ | |
268 | if (thread == caller) | |
269 | continue; | |
270 | ||
271 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || | |
272 | (thread->seccomp.mode == SECCOMP_MODE_FILTER && | |
273 | is_ancestor(thread->seccomp.filter, | |
274 | caller->seccomp.filter))) | |
275 | continue; | |
276 | ||
277 | /* Return the first thread that cannot be synchronized. */ | |
278 | failed = task_pid_vnr(thread); | |
279 | /* If the pid cannot be resolved, then return -ESRCH */ | |
280 | if (unlikely(WARN_ON(failed == 0))) | |
281 | failed = -ESRCH; | |
282 | return failed; | |
283 | } | |
284 | ||
285 | return 0; | |
286 | } | |
287 | ||
288 | /** | |
289 | * seccomp_sync_threads: sets all threads to use current's filter | |
290 | * | |
291 | * Expects sighand and cred_guard_mutex locks to be held, and for | |
292 | * seccomp_can_sync_threads() to have returned success already | |
293 | * without dropping the locks. | |
294 | * | |
295 | */ | |
296 | static inline void seccomp_sync_threads(void) | |
297 | { | |
298 | struct task_struct *thread, *caller; | |
299 | ||
300 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 301 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
302 | |
303 | /* Synchronize all threads. */ | |
304 | caller = current; | |
305 | for_each_thread(caller, thread) { | |
306 | /* Skip current, since it needs no changes. */ | |
307 | if (thread == caller) | |
308 | continue; | |
309 | ||
310 | /* Get a task reference for the new leaf node. */ | |
311 | get_seccomp_filter(caller); | |
312 | /* | |
313 | * Drop the task reference to the shared ancestor since | |
314 | * current's path will hold a reference. (This also | |
315 | * allows a put before the assignment.) | |
316 | */ | |
317 | put_seccomp_filter(thread); | |
318 | smp_store_release(&thread->seccomp.filter, | |
319 | caller->seccomp.filter); | |
320 | /* | |
321 | * Opt the other thread into seccomp if needed. | |
322 | * As threads are considered to be trust-realm | |
323 | * equivalent (see ptrace_may_access), it is safe to | |
324 | * allow one thread to transition the other. | |
325 | */ | |
326 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) { | |
327 | /* | |
328 | * Don't let an unprivileged task work around | |
329 | * the no_new_privs restriction by creating | |
330 | * a thread that sets it up, enters seccomp, | |
331 | * then dies. | |
332 | */ | |
333 | if (task_no_new_privs(caller)) | |
334 | task_set_no_new_privs(thread); | |
335 | ||
336 | seccomp_assign_mode(thread, SECCOMP_MODE_FILTER); | |
337 | } | |
338 | } | |
339 | } | |
340 | ||
e2cfabdf | 341 | /** |
c8bee430 | 342 | * seccomp_prepare_filter: Prepares a seccomp filter for use. |
e2cfabdf WD |
343 | * @fprog: BPF program to install |
344 | * | |
c8bee430 | 345 | * Returns filter on success or an ERR_PTR on failure. |
e2cfabdf | 346 | */ |
c8bee430 | 347 | static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog) |
e2cfabdf WD |
348 | { |
349 | struct seccomp_filter *filter; | |
d9e12f42 NS |
350 | struct bpf_prog *prog; |
351 | unsigned long fsize; | |
e2cfabdf WD |
352 | |
353 | if (fprog->len == 0 || fprog->len > BPF_MAXINSNS) | |
c8bee430 | 354 | return ERR_PTR(-EINVAL); |
d9e12f42 | 355 | |
c8bee430 | 356 | BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter)); |
d9e12f42 | 357 | fsize = bpf_classic_proglen(fprog); |
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 | |
d9e12f42 NS |
370 | prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); |
371 | if (!prog) | |
c8bee430 | 372 | return ERR_PTR(-ENOMEM); |
e2cfabdf WD |
373 | |
374 | /* Copy the instructions from fprog. */ | |
d9e12f42 NS |
375 | if (copy_from_user(prog->insns, fprog->filter, fsize)) { |
376 | __bpf_prog_free(prog); | |
377 | return ERR_PTR(-EFAULT); | |
378 | } | |
e2cfabdf | 379 | |
d9e12f42 | 380 | prog->len = fprog->len; |
bd4cf0ed | 381 | |
d9e12f42 NS |
382 | /* bpf_prepare_filter() already takes care of freeing |
383 | * memory in case something goes wrong. | |
384 | */ | |
385 | prog = bpf_prepare_filter(prog, seccomp_check_filter); | |
386 | if (IS_ERR(prog)) | |
387 | return ERR_CAST(prog); | |
bd4cf0ed AS |
388 | |
389 | /* Allocate a new seccomp_filter */ | |
8f577cad | 390 | filter = kzalloc(sizeof(struct seccomp_filter), |
bd4cf0ed | 391 | GFP_KERNEL|__GFP_NOWARN); |
d9e12f42 NS |
392 | if (!filter) { |
393 | bpf_prog_destroy(prog); | |
394 | return ERR_PTR(-ENOMEM); | |
395 | } | |
bd4cf0ed | 396 | |
d9e12f42 | 397 | filter->prog = prog; |
bd4cf0ed | 398 | atomic_set(&filter->usage, 1); |
e2cfabdf | 399 | |
c8bee430 | 400 | return filter; |
e2cfabdf WD |
401 | } |
402 | ||
403 | /** | |
c8bee430 | 404 | * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog |
e2cfabdf WD |
405 | * @user_filter: pointer to the user data containing a sock_fprog. |
406 | * | |
407 | * Returns 0 on success and non-zero otherwise. | |
408 | */ | |
c8bee430 KC |
409 | static struct seccomp_filter * |
410 | seccomp_prepare_user_filter(const char __user *user_filter) | |
e2cfabdf WD |
411 | { |
412 | struct sock_fprog fprog; | |
c8bee430 | 413 | struct seccomp_filter *filter = ERR_PTR(-EFAULT); |
e2cfabdf WD |
414 | |
415 | #ifdef CONFIG_COMPAT | |
416 | if (is_compat_task()) { | |
417 | struct compat_sock_fprog fprog32; | |
418 | if (copy_from_user(&fprog32, user_filter, sizeof(fprog32))) | |
419 | goto out; | |
420 | fprog.len = fprog32.len; | |
421 | fprog.filter = compat_ptr(fprog32.filter); | |
422 | } else /* falls through to the if below. */ | |
423 | #endif | |
424 | if (copy_from_user(&fprog, user_filter, sizeof(fprog))) | |
425 | goto out; | |
c8bee430 | 426 | filter = seccomp_prepare_filter(&fprog); |
e2cfabdf | 427 | out: |
c8bee430 KC |
428 | return filter; |
429 | } | |
430 | ||
431 | /** | |
432 | * seccomp_attach_filter: validate and attach filter | |
433 | * @flags: flags to change filter behavior | |
434 | * @filter: seccomp filter to add to the current process | |
435 | * | |
dbd95212 KC |
436 | * Caller must be holding current->sighand->siglock lock. |
437 | * | |
c8bee430 KC |
438 | * Returns 0 on success, -ve on error. |
439 | */ | |
440 | static long seccomp_attach_filter(unsigned int flags, | |
441 | struct seccomp_filter *filter) | |
442 | { | |
443 | unsigned long total_insns; | |
444 | struct seccomp_filter *walker; | |
445 | ||
69f6a34b | 446 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 447 | |
c8bee430 KC |
448 | /* Validate resulting filter length. */ |
449 | total_insns = filter->prog->len; | |
450 | for (walker = current->seccomp.filter; walker; walker = walker->prev) | |
451 | total_insns += walker->prog->len + 4; /* 4 instr penalty */ | |
452 | if (total_insns > MAX_INSNS_PER_PATH) | |
453 | return -ENOMEM; | |
454 | ||
c2e1f2e3 KC |
455 | /* If thread sync has been requested, check that it is possible. */ |
456 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) { | |
457 | int ret; | |
458 | ||
459 | ret = seccomp_can_sync_threads(); | |
460 | if (ret) | |
461 | return ret; | |
462 | } | |
463 | ||
c8bee430 KC |
464 | /* |
465 | * If there is an existing filter, make it the prev and don't drop its | |
466 | * task reference. | |
467 | */ | |
468 | filter->prev = current->seccomp.filter; | |
469 | current->seccomp.filter = filter; | |
470 | ||
c2e1f2e3 KC |
471 | /* Now that the new filter is in place, synchronize to all threads. */ |
472 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) | |
473 | seccomp_sync_threads(); | |
474 | ||
c8bee430 | 475 | return 0; |
e2cfabdf WD |
476 | } |
477 | ||
478 | /* get_seccomp_filter - increments the reference count of the filter on @tsk */ | |
479 | void get_seccomp_filter(struct task_struct *tsk) | |
480 | { | |
481 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
482 | if (!orig) | |
483 | return; | |
484 | /* Reference count is bounded by the number of total processes. */ | |
485 | atomic_inc(&orig->usage); | |
486 | } | |
487 | ||
c8bee430 KC |
488 | static inline void seccomp_filter_free(struct seccomp_filter *filter) |
489 | { | |
490 | if (filter) { | |
ae045e24 | 491 | bpf_prog_free(filter->prog); |
c8bee430 KC |
492 | kfree(filter); |
493 | } | |
494 | } | |
495 | ||
e2cfabdf WD |
496 | /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */ |
497 | void put_seccomp_filter(struct task_struct *tsk) | |
498 | { | |
499 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
500 | /* Clean up single-reference branches iteratively. */ | |
501 | while (orig && atomic_dec_and_test(&orig->usage)) { | |
502 | struct seccomp_filter *freeme = orig; | |
503 | orig = orig->prev; | |
c8bee430 | 504 | seccomp_filter_free(freeme); |
e2cfabdf WD |
505 | } |
506 | } | |
bb6ea430 WD |
507 | |
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 | */ | |
534 | static int mode1_syscalls[] = { | |
535 | __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, | |
536 | 0, /* null terminated */ | |
537 | }; | |
538 | ||
5b101740 | 539 | #ifdef CONFIG_COMPAT |
1da177e4 LT |
540 | static int mode1_syscalls_32[] = { |
541 | __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32, | |
542 | 0, /* null terminated */ | |
543 | }; | |
544 | #endif | |
545 | ||
a4412fc9 | 546 | static void __secure_computing_strict(int this_syscall) |
1da177e4 | 547 | { |
a4412fc9 AL |
548 | int *syscall_whitelist = mode1_syscalls; |
549 | #ifdef CONFIG_COMPAT | |
550 | if (is_compat_task()) | |
551 | syscall_whitelist = mode1_syscalls_32; | |
552 | #endif | |
553 | do { | |
554 | if (*syscall_whitelist == this_syscall) | |
555 | return; | |
556 | } while (*++syscall_whitelist); | |
557 | ||
558 | #ifdef SECCOMP_DEBUG | |
559 | dump_stack(); | |
560 | #endif | |
561 | audit_seccomp(this_syscall, SIGKILL, SECCOMP_RET_KILL); | |
562 | do_exit(SIGKILL); | |
563 | } | |
564 | ||
565 | #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER | |
566 | void secure_computing_strict(int this_syscall) | |
567 | { | |
568 | int mode = current->seccomp.mode; | |
569 | ||
570 | if (mode == 0) | |
571 | return; | |
572 | else if (mode == SECCOMP_MODE_STRICT) | |
573 | __secure_computing_strict(this_syscall); | |
574 | else | |
575 | BUG(); | |
576 | } | |
577 | #else | |
578 | int __secure_computing(void) | |
579 | { | |
d39bd00d | 580 | u32 phase1_result = seccomp_phase1(NULL); |
13aa72f0 AL |
581 | |
582 | if (likely(phase1_result == SECCOMP_PHASE1_OK)) | |
583 | return 0; | |
584 | else if (likely(phase1_result == SECCOMP_PHASE1_SKIP)) | |
585 | return -1; | |
586 | else | |
587 | return seccomp_phase2(phase1_result); | |
588 | } | |
589 | ||
590 | #ifdef CONFIG_SECCOMP_FILTER | |
d39bd00d | 591 | static u32 __seccomp_phase1_filter(int this_syscall, struct seccomp_data *sd) |
13aa72f0 AL |
592 | { |
593 | u32 filter_ret, action; | |
594 | int data; | |
1da177e4 | 595 | |
3ba2530c KC |
596 | /* |
597 | * Make sure that any changes to mode from another thread have | |
598 | * been seen after TIF_SECCOMP was seen. | |
599 | */ | |
600 | rmb(); | |
601 | ||
d39bd00d | 602 | filter_ret = seccomp_run_filters(sd); |
13aa72f0 AL |
603 | data = filter_ret & SECCOMP_RET_DATA; |
604 | action = filter_ret & SECCOMP_RET_ACTION; | |
605 | ||
606 | switch (action) { | |
607 | case SECCOMP_RET_ERRNO: | |
580c57f1 KC |
608 | /* Set low-order bits as an errno, capped at MAX_ERRNO. */ |
609 | if (data > MAX_ERRNO) | |
610 | data = MAX_ERRNO; | |
d39bd00d | 611 | syscall_set_return_value(current, task_pt_regs(current), |
13aa72f0 AL |
612 | -data, 0); |
613 | goto skip; | |
614 | ||
615 | case SECCOMP_RET_TRAP: | |
616 | /* Show the handler the original registers. */ | |
d39bd00d | 617 | syscall_rollback(current, task_pt_regs(current)); |
13aa72f0 AL |
618 | /* Let the filter pass back 16 bits of data. */ |
619 | seccomp_send_sigsys(this_syscall, data); | |
620 | goto skip; | |
621 | ||
622 | case SECCOMP_RET_TRACE: | |
623 | return filter_ret; /* Save the rest for phase 2. */ | |
624 | ||
625 | case SECCOMP_RET_ALLOW: | |
626 | return SECCOMP_PHASE1_OK; | |
627 | ||
628 | case SECCOMP_RET_KILL: | |
629 | default: | |
630 | audit_seccomp(this_syscall, SIGSYS, action); | |
631 | do_exit(SIGSYS); | |
632 | } | |
633 | ||
634 | unreachable(); | |
635 | ||
636 | skip: | |
637 | audit_seccomp(this_syscall, 0, action); | |
638 | return SECCOMP_PHASE1_SKIP; | |
639 | } | |
1da177e4 | 640 | #endif |
13aa72f0 AL |
641 | |
642 | /** | |
643 | * seccomp_phase1() - run fast path seccomp checks on the current syscall | |
d39bd00d | 644 | * @arg sd: The seccomp_data or NULL |
13aa72f0 AL |
645 | * |
646 | * This only reads pt_regs via the syscall_xyz helpers. The only change | |
647 | * it will make to pt_regs is via syscall_set_return_value, and it will | |
648 | * only do that if it returns SECCOMP_PHASE1_SKIP. | |
649 | * | |
d39bd00d AL |
650 | * If sd is provided, it will not read pt_regs at all. |
651 | * | |
13aa72f0 AL |
652 | * It may also call do_exit or force a signal; these actions must be |
653 | * safe. | |
654 | * | |
655 | * If it returns SECCOMP_PHASE1_OK, the syscall passes checks and should | |
656 | * be processed normally. | |
657 | * | |
658 | * If it returns SECCOMP_PHASE1_SKIP, then the syscall should not be | |
659 | * invoked. In this case, seccomp_phase1 will have set the return value | |
660 | * using syscall_set_return_value. | |
661 | * | |
662 | * If it returns anything else, then the return value should be passed | |
663 | * to seccomp_phase2 from a context in which ptrace hooks are safe. | |
664 | */ | |
d39bd00d | 665 | u32 seccomp_phase1(struct seccomp_data *sd) |
13aa72f0 AL |
666 | { |
667 | int mode = current->seccomp.mode; | |
d39bd00d AL |
668 | int this_syscall = sd ? sd->nr : |
669 | syscall_get_nr(current, task_pt_regs(current)); | |
13aa72f0 AL |
670 | |
671 | switch (mode) { | |
e2cfabdf | 672 | case SECCOMP_MODE_STRICT: |
13aa72f0 AL |
673 | __secure_computing_strict(this_syscall); /* may call do_exit */ |
674 | return SECCOMP_PHASE1_OK; | |
e2cfabdf | 675 | #ifdef CONFIG_SECCOMP_FILTER |
13aa72f0 | 676 | case SECCOMP_MODE_FILTER: |
d39bd00d | 677 | return __seccomp_phase1_filter(this_syscall, sd); |
e2cfabdf | 678 | #endif |
1da177e4 LT |
679 | default: |
680 | BUG(); | |
681 | } | |
13aa72f0 | 682 | } |
1da177e4 | 683 | |
13aa72f0 AL |
684 | /** |
685 | * seccomp_phase2() - finish slow path seccomp work for the current syscall | |
686 | * @phase1_result: The return value from seccomp_phase1() | |
687 | * | |
688 | * This must be called from a context in which ptrace hooks can be used. | |
689 | * | |
690 | * Returns 0 if the syscall should be processed or -1 to skip the syscall. | |
691 | */ | |
692 | int seccomp_phase2(u32 phase1_result) | |
693 | { | |
694 | struct pt_regs *regs = task_pt_regs(current); | |
695 | u32 action = phase1_result & SECCOMP_RET_ACTION; | |
696 | int data = phase1_result & SECCOMP_RET_DATA; | |
697 | ||
698 | BUG_ON(action != SECCOMP_RET_TRACE); | |
699 | ||
700 | audit_seccomp(syscall_get_nr(current, regs), 0, action); | |
701 | ||
702 | /* Skip these calls if there is no tracer. */ | |
703 | if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { | |
704 | syscall_set_return_value(current, regs, | |
705 | -ENOSYS, 0); | |
706 | return -1; | |
707 | } | |
708 | ||
709 | /* Allow the BPF to provide the event message */ | |
710 | ptrace_event(PTRACE_EVENT_SECCOMP, data); | |
711 | /* | |
712 | * The delivery of a fatal signal during event | |
713 | * notification may silently skip tracer notification. | |
714 | * Terminating the task now avoids executing a system | |
715 | * call that may not be intended. | |
716 | */ | |
717 | if (fatal_signal_pending(current)) | |
718 | do_exit(SIGSYS); | |
719 | if (syscall_get_nr(current, regs) < 0) | |
720 | return -1; /* Explicit request to skip. */ | |
721 | ||
722 | return 0; | |
1da177e4 | 723 | } |
a4412fc9 | 724 | #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ |
1d9d02fe AA |
725 | |
726 | long prctl_get_seccomp(void) | |
727 | { | |
728 | return current->seccomp.mode; | |
729 | } | |
730 | ||
e2cfabdf | 731 | /** |
3b23dd12 | 732 | * seccomp_set_mode_strict: internal function for setting strict seccomp |
e2cfabdf WD |
733 | * |
734 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
735 | * | |
736 | * Returns 0 on success or -EINVAL on failure. | |
737 | */ | |
3b23dd12 | 738 | static long seccomp_set_mode_strict(void) |
1d9d02fe | 739 | { |
3b23dd12 | 740 | const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; |
e2cfabdf | 741 | long ret = -EINVAL; |
1d9d02fe | 742 | |
dbd95212 KC |
743 | spin_lock_irq(¤t->sighand->siglock); |
744 | ||
1f41b450 | 745 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1d9d02fe AA |
746 | goto out; |
747 | ||
cf99abac | 748 | #ifdef TIF_NOTSC |
3b23dd12 | 749 | disable_TSC(); |
cf99abac | 750 | #endif |
3ba2530c | 751 | seccomp_assign_mode(current, seccomp_mode); |
3b23dd12 KC |
752 | ret = 0; |
753 | ||
754 | out: | |
dbd95212 | 755 | spin_unlock_irq(¤t->sighand->siglock); |
3b23dd12 KC |
756 | |
757 | return ret; | |
758 | } | |
759 | ||
e2cfabdf | 760 | #ifdef CONFIG_SECCOMP_FILTER |
3b23dd12 KC |
761 | /** |
762 | * seccomp_set_mode_filter: internal function for setting seccomp filter | |
48dc92b9 | 763 | * @flags: flags to change filter behavior |
3b23dd12 KC |
764 | * @filter: struct sock_fprog containing filter |
765 | * | |
766 | * This function may be called repeatedly to install additional filters. | |
767 | * Every filter successfully installed will be evaluated (in reverse order) | |
768 | * for each system call the task makes. | |
769 | * | |
770 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
771 | * | |
772 | * Returns 0 on success or -EINVAL on failure. | |
773 | */ | |
48dc92b9 KC |
774 | static long seccomp_set_mode_filter(unsigned int flags, |
775 | const char __user *filter) | |
3b23dd12 KC |
776 | { |
777 | const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; | |
c8bee430 | 778 | struct seccomp_filter *prepared = NULL; |
3b23dd12 KC |
779 | long ret = -EINVAL; |
780 | ||
48dc92b9 | 781 | /* Validate flags. */ |
c2e1f2e3 | 782 | if (flags & ~SECCOMP_FILTER_FLAG_MASK) |
dbd95212 | 783 | return -EINVAL; |
48dc92b9 | 784 | |
c8bee430 KC |
785 | /* Prepare the new filter before holding any locks. */ |
786 | prepared = seccomp_prepare_user_filter(filter); | |
787 | if (IS_ERR(prepared)) | |
788 | return PTR_ERR(prepared); | |
789 | ||
c2e1f2e3 KC |
790 | /* |
791 | * Make sure we cannot change seccomp or nnp state via TSYNC | |
792 | * while another thread is in the middle of calling exec. | |
793 | */ | |
794 | if (flags & SECCOMP_FILTER_FLAG_TSYNC && | |
795 | mutex_lock_killable(¤t->signal->cred_guard_mutex)) | |
796 | goto out_free; | |
797 | ||
dbd95212 KC |
798 | spin_lock_irq(¤t->sighand->siglock); |
799 | ||
3b23dd12 KC |
800 | if (!seccomp_may_assign_mode(seccomp_mode)) |
801 | goto out; | |
802 | ||
c8bee430 | 803 | ret = seccomp_attach_filter(flags, prepared); |
3b23dd12 | 804 | if (ret) |
e2cfabdf | 805 | goto out; |
c8bee430 KC |
806 | /* Do not free the successfully attached filter. */ |
807 | prepared = NULL; | |
1d9d02fe | 808 | |
3ba2530c | 809 | seccomp_assign_mode(current, seccomp_mode); |
e2cfabdf | 810 | out: |
dbd95212 | 811 | spin_unlock_irq(¤t->sighand->siglock); |
c2e1f2e3 KC |
812 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
813 | mutex_unlock(¤t->signal->cred_guard_mutex); | |
814 | out_free: | |
c8bee430 | 815 | seccomp_filter_free(prepared); |
1d9d02fe AA |
816 | return ret; |
817 | } | |
3b23dd12 | 818 | #else |
48dc92b9 KC |
819 | static inline long seccomp_set_mode_filter(unsigned int flags, |
820 | const char __user *filter) | |
3b23dd12 KC |
821 | { |
822 | return -EINVAL; | |
823 | } | |
824 | #endif | |
d78ab02c | 825 | |
48dc92b9 KC |
826 | /* Common entry point for both prctl and syscall. */ |
827 | static long do_seccomp(unsigned int op, unsigned int flags, | |
828 | const char __user *uargs) | |
829 | { | |
830 | switch (op) { | |
831 | case SECCOMP_SET_MODE_STRICT: | |
832 | if (flags != 0 || uargs != NULL) | |
833 | return -EINVAL; | |
834 | return seccomp_set_mode_strict(); | |
835 | case SECCOMP_SET_MODE_FILTER: | |
836 | return seccomp_set_mode_filter(flags, uargs); | |
837 | default: | |
838 | return -EINVAL; | |
839 | } | |
840 | } | |
841 | ||
842 | SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, | |
843 | const char __user *, uargs) | |
844 | { | |
845 | return do_seccomp(op, flags, uargs); | |
846 | } | |
847 | ||
d78ab02c KC |
848 | /** |
849 | * prctl_set_seccomp: configures current->seccomp.mode | |
850 | * @seccomp_mode: requested mode to use | |
851 | * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER | |
852 | * | |
853 | * Returns 0 on success or -EINVAL on failure. | |
854 | */ | |
855 | long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter) | |
856 | { | |
48dc92b9 KC |
857 | unsigned int op; |
858 | char __user *uargs; | |
859 | ||
3b23dd12 KC |
860 | switch (seccomp_mode) { |
861 | case SECCOMP_MODE_STRICT: | |
48dc92b9 KC |
862 | op = SECCOMP_SET_MODE_STRICT; |
863 | /* | |
864 | * Setting strict mode through prctl always ignored filter, | |
865 | * so make sure it is always NULL here to pass the internal | |
866 | * check in do_seccomp(). | |
867 | */ | |
868 | uargs = NULL; | |
869 | break; | |
3b23dd12 | 870 | case SECCOMP_MODE_FILTER: |
48dc92b9 KC |
871 | op = SECCOMP_SET_MODE_FILTER; |
872 | uargs = filter; | |
873 | break; | |
3b23dd12 KC |
874 | default: |
875 | return -EINVAL; | |
876 | } | |
48dc92b9 KC |
877 | |
878 | /* prctl interface doesn't have flags, so they are always zero. */ | |
879 | return do_seccomp(op, 0, uargs); | |
d78ab02c | 880 | } |