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> | |
6fa3eb70 S |
21 | #include <linux/slab.h> |
22 | #include <linux/syscalls.h> | |
1da177e4 LT |
23 | |
24 | /* #define SECCOMP_DEBUG 1 */ | |
e2cfabdf WD |
25 | |
26 | #ifdef CONFIG_SECCOMP_FILTER | |
27 | #include <asm/syscall.h> | |
28 | #include <linux/filter.h> | |
6fa3eb70 | 29 | #include <linux/pid.h> |
fb0fadf9 | 30 | #include <linux/ptrace.h> |
e2cfabdf | 31 | #include <linux/security.h> |
e2cfabdf WD |
32 | #include <linux/tracehook.h> |
33 | #include <linux/uaccess.h> | |
34 | ||
35 | /** | |
36 | * struct seccomp_filter - container for seccomp BPF programs | |
37 | * | |
38 | * @usage: reference count to manage the object lifetime. | |
39 | * get/put helpers should be used when accessing an instance | |
40 | * outside of a lifetime-guarded section. In general, this | |
41 | * is only needed for handling filters shared across tasks. | |
42 | * @prev: points to a previously installed, or inherited, filter | |
43 | * @len: the number of instructions in the program | |
44 | * @insns: the BPF program instructions to evaluate | |
45 | * | |
46 | * seccomp_filter objects are organized in a tree linked via the @prev | |
47 | * pointer. For any task, it appears to be a singly-linked list starting | |
48 | * with current->seccomp.filter, the most recently attached or inherited filter. | |
49 | * However, multiple filters may share a @prev node, by way of fork(), which | |
50 | * results in a unidirectional tree existing in memory. This is similar to | |
51 | * how namespaces work. | |
52 | * | |
53 | * seccomp_filter objects should never be modified after being attached | |
54 | * to a task_struct (other than @usage). | |
55 | */ | |
56 | struct seccomp_filter { | |
57 | atomic_t usage; | |
58 | struct seccomp_filter *prev; | |
59 | unsigned short len; /* Instruction count */ | |
60 | struct sock_filter insns[]; | |
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 | ||
e2cfabdf WD |
66 | /** |
67 | * get_u32 - returns a u32 offset into data | |
68 | * @data: a unsigned 64 bit value | |
69 | * @index: 0 or 1 to return the first or second 32-bits | |
70 | * | |
71 | * This inline exists to hide the length of unsigned long. If a 32-bit | |
72 | * unsigned long is passed in, it will be extended and the top 32-bits will be | |
73 | * 0. If it is a 64-bit unsigned long, then whatever data is resident will be | |
74 | * properly returned. | |
75 | * | |
76 | * Endianness is explicitly ignored and left for BPF program authors to manage | |
77 | * as per the specific architecture. | |
78 | */ | |
79 | static inline u32 get_u32(u64 data, int index) | |
80 | { | |
81 | return ((u32 *)&data)[index]; | |
82 | } | |
83 | ||
84 | /* Helper for bpf_load below. */ | |
85 | #define BPF_DATA(_name) offsetof(struct seccomp_data, _name) | |
86 | /** | |
87 | * bpf_load: checks and returns a pointer to the requested offset | |
88 | * @off: offset into struct seccomp_data to load from | |
89 | * | |
90 | * Returns the requested 32-bits of data. | |
91 | * seccomp_check_filter() should assure that @off is 32-bit aligned | |
92 | * and not out of bounds. Failure to do so is a BUG. | |
93 | */ | |
94 | u32 seccomp_bpf_load(int off) | |
95 | { | |
96 | struct pt_regs *regs = task_pt_regs(current); | |
97 | if (off == BPF_DATA(nr)) | |
98 | return syscall_get_nr(current, regs); | |
99 | if (off == BPF_DATA(arch)) | |
6fa3eb70 | 100 | return syscall_get_arch(); |
e2cfabdf WD |
101 | if (off >= BPF_DATA(args[0]) && off < BPF_DATA(args[6])) { |
102 | unsigned long value; | |
103 | int arg = (off - BPF_DATA(args[0])) / sizeof(u64); | |
104 | int index = !!(off % sizeof(u64)); | |
105 | syscall_get_arguments(current, regs, arg, 1, &value); | |
106 | return get_u32(value, index); | |
107 | } | |
108 | if (off == BPF_DATA(instruction_pointer)) | |
109 | return get_u32(KSTK_EIP(current), 0); | |
110 | if (off == BPF_DATA(instruction_pointer) + sizeof(u32)) | |
111 | return get_u32(KSTK_EIP(current), 1); | |
112 | /* seccomp_check_filter should make this impossible. */ | |
113 | BUG(); | |
114 | } | |
115 | ||
116 | /** | |
117 | * seccomp_check_filter - verify seccomp filter code | |
118 | * @filter: filter to verify | |
119 | * @flen: length of filter | |
120 | * | |
121 | * Takes a previously checked filter (by sk_chk_filter) and | |
122 | * redirects all filter code that loads struct sk_buff data | |
123 | * and related data through seccomp_bpf_load. It also | |
124 | * enforces length and alignment checking of those loads. | |
125 | * | |
126 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
127 | */ | |
128 | static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen) | |
129 | { | |
130 | int pc; | |
131 | for (pc = 0; pc < flen; pc++) { | |
132 | struct sock_filter *ftest = &filter[pc]; | |
133 | u16 code = ftest->code; | |
134 | u32 k = ftest->k; | |
135 | ||
136 | switch (code) { | |
137 | case BPF_S_LD_W_ABS: | |
138 | ftest->code = BPF_S_ANC_SECCOMP_LD_W; | |
139 | /* 32-bit aligned and not out of bounds. */ | |
140 | if (k >= sizeof(struct seccomp_data) || k & 3) | |
141 | return -EINVAL; | |
142 | continue; | |
143 | case BPF_S_LD_W_LEN: | |
144 | ftest->code = BPF_S_LD_IMM; | |
145 | ftest->k = sizeof(struct seccomp_data); | |
146 | continue; | |
147 | case BPF_S_LDX_W_LEN: | |
148 | ftest->code = BPF_S_LDX_IMM; | |
149 | ftest->k = sizeof(struct seccomp_data); | |
150 | continue; | |
151 | /* Explicitly include allowed calls. */ | |
152 | case BPF_S_RET_K: | |
153 | case BPF_S_RET_A: | |
154 | case BPF_S_ALU_ADD_K: | |
155 | case BPF_S_ALU_ADD_X: | |
156 | case BPF_S_ALU_SUB_K: | |
157 | case BPF_S_ALU_SUB_X: | |
158 | case BPF_S_ALU_MUL_K: | |
159 | case BPF_S_ALU_MUL_X: | |
160 | case BPF_S_ALU_DIV_X: | |
161 | case BPF_S_ALU_AND_K: | |
162 | case BPF_S_ALU_AND_X: | |
163 | case BPF_S_ALU_OR_K: | |
164 | case BPF_S_ALU_OR_X: | |
d1327479 NS |
165 | case BPF_S_ALU_XOR_K: |
166 | case BPF_S_ALU_XOR_X: | |
e2cfabdf WD |
167 | case BPF_S_ALU_LSH_K: |
168 | case BPF_S_ALU_LSH_X: | |
169 | case BPF_S_ALU_RSH_K: | |
170 | case BPF_S_ALU_RSH_X: | |
171 | case BPF_S_ALU_NEG: | |
172 | case BPF_S_LD_IMM: | |
173 | case BPF_S_LDX_IMM: | |
174 | case BPF_S_MISC_TAX: | |
175 | case BPF_S_MISC_TXA: | |
176 | case BPF_S_ALU_DIV_K: | |
177 | case BPF_S_LD_MEM: | |
178 | case BPF_S_LDX_MEM: | |
179 | case BPF_S_ST: | |
180 | case BPF_S_STX: | |
181 | case BPF_S_JMP_JA: | |
182 | case BPF_S_JMP_JEQ_K: | |
183 | case BPF_S_JMP_JEQ_X: | |
184 | case BPF_S_JMP_JGE_K: | |
185 | case BPF_S_JMP_JGE_X: | |
186 | case BPF_S_JMP_JGT_K: | |
187 | case BPF_S_JMP_JGT_X: | |
188 | case BPF_S_JMP_JSET_K: | |
189 | case BPF_S_JMP_JSET_X: | |
190 | continue; | |
191 | default: | |
192 | return -EINVAL; | |
193 | } | |
194 | } | |
195 | return 0; | |
196 | } | |
197 | ||
198 | /** | |
199 | * seccomp_run_filters - evaluates all seccomp filters against @syscall | |
200 | * @syscall: number of the current system call | |
201 | * | |
202 | * Returns valid seccomp BPF response codes. | |
203 | */ | |
204 | static u32 seccomp_run_filters(int syscall) | |
205 | { | |
6fa3eb70 | 206 | struct seccomp_filter *f = ACCESS_ONCE(current->seccomp.filter); |
acf3b2c7 WD |
207 | u32 ret = SECCOMP_RET_ALLOW; |
208 | ||
209 | /* Ensure unexpected behavior doesn't result in failing open. */ | |
6fa3eb70 | 210 | if (unlikely(WARN_ON(f == NULL))) |
acf3b2c7 WD |
211 | return SECCOMP_RET_KILL; |
212 | ||
6fa3eb70 S |
213 | /* Make sure cross-thread synced filter points somewhere sane. */ |
214 | smp_read_barrier_depends(); | |
215 | ||
e2cfabdf WD |
216 | /* |
217 | * All filters in the list are evaluated and the lowest BPF return | |
acf3b2c7 | 218 | * value always takes priority (ignoring the DATA). |
e2cfabdf | 219 | */ |
6fa3eb70 | 220 | for (; f; f = f->prev) { |
acf3b2c7 | 221 | u32 cur_ret = sk_run_filter(NULL, f->insns); |
6fa3eb70 | 222 | |
acf3b2c7 WD |
223 | if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION)) |
224 | ret = cur_ret; | |
e2cfabdf WD |
225 | } |
226 | return ret; | |
227 | } | |
6fa3eb70 S |
228 | #endif /* CONFIG_SECCOMP_FILTER */ |
229 | ||
230 | static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) | |
231 | { | |
232 | assert_spin_locked(¤t->sighand->siglock); | |
233 | ||
234 | if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) | |
235 | return false; | |
236 | ||
237 | return true; | |
238 | } | |
239 | ||
240 | static inline void seccomp_assign_mode(struct task_struct *task, | |
241 | unsigned long seccomp_mode) | |
242 | { | |
243 | assert_spin_locked(&task->sighand->siglock); | |
244 | ||
245 | task->seccomp.mode = seccomp_mode; | |
246 | /* | |
247 | * Make sure TIF_SECCOMP cannot be set before the mode (and | |
248 | * filter) is set. | |
249 | */ | |
250 | smp_mb(); | |
251 | set_tsk_thread_flag(task, TIF_SECCOMP); | |
252 | } | |
253 | ||
254 | #ifdef CONFIG_SECCOMP_FILTER | |
255 | /* Returns 1 if the parent is an ancestor of the child. */ | |
256 | static int is_ancestor(struct seccomp_filter *parent, | |
257 | struct seccomp_filter *child) | |
258 | { | |
259 | /* NULL is the root ancestor. */ | |
260 | if (parent == NULL) | |
261 | return 1; | |
262 | for (; child; child = child->prev) | |
263 | if (child == parent) | |
264 | return 1; | |
265 | return 0; | |
266 | } | |
e2cfabdf WD |
267 | |
268 | /** | |
6fa3eb70 S |
269 | * seccomp_can_sync_threads: checks if all threads can be synchronized |
270 | * | |
271 | * Expects sighand and cred_guard_mutex locks to be held. | |
272 | * | |
273 | * Returns 0 on success, -ve on error, or the pid of a thread which was | |
274 | * either not in the correct seccomp mode or it did not have an ancestral | |
275 | * seccomp filter. | |
276 | */ | |
277 | static inline pid_t seccomp_can_sync_threads(void) | |
278 | { | |
279 | struct task_struct *thread, *caller; | |
280 | ||
281 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
282 | assert_spin_locked(¤t->sighand->siglock); | |
283 | ||
284 | /* Validate all threads being eligible for synchronization. */ | |
285 | caller = current; | |
286 | for_each_thread(caller, thread) { | |
287 | pid_t failed; | |
288 | ||
289 | /* Skip current, since it is initiating the sync. */ | |
290 | if (thread == caller) | |
291 | continue; | |
292 | ||
293 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || | |
294 | (thread->seccomp.mode == SECCOMP_MODE_FILTER && | |
295 | is_ancestor(thread->seccomp.filter, | |
296 | caller->seccomp.filter))) | |
297 | continue; | |
298 | ||
299 | /* Return the first thread that cannot be synchronized. */ | |
300 | failed = task_pid_vnr(thread); | |
301 | /* If the pid cannot be resolved, then return -ESRCH */ | |
302 | if (unlikely(WARN_ON(failed == 0))) | |
303 | failed = -ESRCH; | |
304 | return failed; | |
305 | } | |
306 | ||
307 | return 0; | |
308 | } | |
309 | ||
310 | /** | |
311 | * seccomp_sync_threads: sets all threads to use current's filter | |
312 | * | |
313 | * Expects sighand and cred_guard_mutex locks to be held, and for | |
314 | * seccomp_can_sync_threads() to have returned success already | |
315 | * without dropping the locks. | |
316 | * | |
317 | */ | |
318 | static inline void seccomp_sync_threads(void) | |
319 | { | |
320 | struct task_struct *thread, *caller; | |
321 | ||
322 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
323 | assert_spin_locked(¤t->sighand->siglock); | |
324 | ||
325 | /* Synchronize all threads. */ | |
326 | caller = current; | |
327 | for_each_thread(caller, thread) { | |
328 | /* Skip current, since it needs no changes. */ | |
329 | if (thread == caller) | |
330 | continue; | |
331 | ||
332 | /* Get a task reference for the new leaf node. */ | |
333 | get_seccomp_filter(caller); | |
334 | /* | |
335 | * Drop the task reference to the shared ancestor since | |
336 | * current's path will hold a reference. (This also | |
337 | * allows a put before the assignment.) | |
338 | */ | |
339 | put_seccomp_filter(thread); | |
340 | smp_store_release(&thread->seccomp.filter, | |
341 | caller->seccomp.filter); | |
342 | /* | |
343 | * Opt the other thread into seccomp if needed. | |
344 | * As threads are considered to be trust-realm | |
345 | * equivalent (see ptrace_may_access), it is safe to | |
346 | * allow one thread to transition the other. | |
347 | */ | |
348 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) { | |
349 | /* | |
350 | * Don't let an unprivileged task work around | |
351 | * the no_new_privs restriction by creating | |
352 | * a thread that sets it up, enters seccomp, | |
353 | * then dies. | |
354 | */ | |
355 | if (task_no_new_privs(caller)) | |
356 | task_set_no_new_privs(thread); | |
357 | ||
358 | seccomp_assign_mode(thread, SECCOMP_MODE_FILTER); | |
359 | } | |
360 | } | |
361 | } | |
362 | ||
363 | /** | |
364 | * seccomp_prepare_filter: Prepares a seccomp filter for use. | |
e2cfabdf WD |
365 | * @fprog: BPF program to install |
366 | * | |
6fa3eb70 | 367 | * Returns filter on success or an ERR_PTR on failure. |
e2cfabdf | 368 | */ |
6fa3eb70 | 369 | static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog) |
e2cfabdf WD |
370 | { |
371 | struct seccomp_filter *filter; | |
372 | unsigned long fp_size = fprog->len * sizeof(struct sock_filter); | |
373 | unsigned long total_insns = fprog->len; | |
374 | long ret; | |
375 | ||
376 | if (fprog->len == 0 || fprog->len > BPF_MAXINSNS) | |
6fa3eb70 S |
377 | return ERR_PTR(-EINVAL); |
378 | BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter)); | |
e2cfabdf WD |
379 | |
380 | for (filter = current->seccomp.filter; filter; filter = filter->prev) | |
381 | total_insns += filter->len + 4; /* include a 4 instr penalty */ | |
382 | if (total_insns > MAX_INSNS_PER_PATH) | |
6fa3eb70 | 383 | return ERR_PTR(-ENOMEM); |
e2cfabdf WD |
384 | |
385 | /* | |
386 | * Installing a seccomp filter requires that the task have | |
387 | * CAP_SYS_ADMIN in its namespace or be running with no_new_privs. | |
388 | * This avoids scenarios where unprivileged tasks can affect the | |
389 | * behavior of privileged children. | |
390 | */ | |
6fa3eb70 | 391 | if (!task_no_new_privs(current) && |
e2cfabdf WD |
392 | security_capable_noaudit(current_cred(), current_user_ns(), |
393 | CAP_SYS_ADMIN) != 0) | |
6fa3eb70 | 394 | return ERR_PTR(-EACCES); |
e2cfabdf WD |
395 | |
396 | /* Allocate a new seccomp_filter */ | |
397 | filter = kzalloc(sizeof(struct seccomp_filter) + fp_size, | |
398 | GFP_KERNEL|__GFP_NOWARN); | |
399 | if (!filter) | |
6fa3eb70 | 400 | return ERR_PTR(-ENOMEM);; |
e2cfabdf WD |
401 | atomic_set(&filter->usage, 1); |
402 | filter->len = fprog->len; | |
403 | ||
404 | /* Copy the instructions from fprog. */ | |
405 | ret = -EFAULT; | |
406 | if (copy_from_user(filter->insns, fprog->filter, fp_size)) | |
407 | goto fail; | |
408 | ||
409 | /* Check and rewrite the fprog via the skb checker */ | |
410 | ret = sk_chk_filter(filter->insns, filter->len); | |
411 | if (ret) | |
412 | goto fail; | |
413 | ||
414 | /* Check and rewrite the fprog for seccomp use */ | |
415 | ret = seccomp_check_filter(filter->insns, filter->len); | |
416 | if (ret) | |
417 | goto fail; | |
418 | ||
6fa3eb70 S |
419 | return filter; |
420 | ||
e2cfabdf WD |
421 | fail: |
422 | kfree(filter); | |
6fa3eb70 | 423 | return ERR_PTR(ret); |
e2cfabdf WD |
424 | } |
425 | ||
426 | /** | |
6fa3eb70 | 427 | * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog |
e2cfabdf WD |
428 | * @user_filter: pointer to the user data containing a sock_fprog. |
429 | * | |
430 | * Returns 0 on success and non-zero otherwise. | |
431 | */ | |
6fa3eb70 S |
432 | static struct seccomp_filter * |
433 | seccomp_prepare_user_filter(const char __user *user_filter) | |
e2cfabdf WD |
434 | { |
435 | struct sock_fprog fprog; | |
6fa3eb70 | 436 | struct seccomp_filter *filter = ERR_PTR(-EFAULT); |
e2cfabdf WD |
437 | |
438 | #ifdef CONFIG_COMPAT | |
439 | if (is_compat_task()) { | |
440 | struct compat_sock_fprog fprog32; | |
441 | if (copy_from_user(&fprog32, user_filter, sizeof(fprog32))) | |
442 | goto out; | |
443 | fprog.len = fprog32.len; | |
444 | fprog.filter = compat_ptr(fprog32.filter); | |
445 | } else /* falls through to the if below. */ | |
446 | #endif | |
447 | if (copy_from_user(&fprog, user_filter, sizeof(fprog))) | |
448 | goto out; | |
6fa3eb70 | 449 | filter = seccomp_prepare_filter(&fprog); |
e2cfabdf | 450 | out: |
6fa3eb70 S |
451 | return filter; |
452 | } | |
453 | ||
454 | /** | |
455 | * seccomp_attach_filter: validate and attach filter | |
456 | * @flags: flags to change filter behavior | |
457 | * @filter: seccomp filter to add to the current process | |
458 | * | |
459 | * Caller must be holding current->sighand->siglock lock. | |
460 | * | |
461 | * Returns 0 on success, -ve on error. | |
462 | */ | |
463 | static long seccomp_attach_filter(unsigned int flags, | |
464 | struct seccomp_filter *filter) | |
465 | { | |
466 | unsigned long total_insns; | |
467 | struct seccomp_filter *walker; | |
468 | ||
469 | assert_spin_locked(¤t->sighand->siglock); | |
470 | ||
471 | /* Validate resulting filter length. */ | |
472 | total_insns = filter->len; | |
473 | for (walker = current->seccomp.filter; walker; walker = walker->prev) | |
474 | total_insns += walker->len + 4; /* 4 instr penalty */ | |
475 | if (total_insns > MAX_INSNS_PER_PATH) | |
476 | return -ENOMEM; | |
477 | ||
478 | /* If thread sync has been requested, check that it is possible. */ | |
479 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) { | |
480 | int ret; | |
481 | ||
482 | ret = seccomp_can_sync_threads(); | |
483 | if (ret) | |
484 | return ret; | |
485 | } | |
486 | ||
487 | /* | |
488 | * If there is an existing filter, make it the prev and don't drop its | |
489 | * task reference. | |
490 | */ | |
491 | filter->prev = current->seccomp.filter; | |
492 | current->seccomp.filter = filter; | |
493 | ||
494 | /* Now that the new filter is in place, synchronize to all threads. */ | |
495 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) | |
496 | seccomp_sync_threads(); | |
497 | ||
498 | return 0; | |
e2cfabdf WD |
499 | } |
500 | ||
501 | /* get_seccomp_filter - increments the reference count of the filter on @tsk */ | |
502 | void get_seccomp_filter(struct task_struct *tsk) | |
503 | { | |
504 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
505 | if (!orig) | |
506 | return; | |
507 | /* Reference count is bounded by the number of total processes. */ | |
508 | atomic_inc(&orig->usage); | |
509 | } | |
510 | ||
6fa3eb70 S |
511 | static inline void seccomp_filter_free(struct seccomp_filter *filter) |
512 | { | |
513 | if (filter) { | |
514 | kfree(filter); | |
515 | } | |
516 | } | |
517 | ||
e2cfabdf WD |
518 | /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */ |
519 | void put_seccomp_filter(struct task_struct *tsk) | |
520 | { | |
521 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
522 | /* Clean up single-reference branches iteratively. */ | |
523 | while (orig && atomic_dec_and_test(&orig->usage)) { | |
524 | struct seccomp_filter *freeme = orig; | |
525 | orig = orig->prev; | |
6fa3eb70 | 526 | seccomp_filter_free(freeme); |
e2cfabdf WD |
527 | } |
528 | } | |
bb6ea430 WD |
529 | |
530 | /** | |
531 | * seccomp_send_sigsys - signals the task to allow in-process syscall emulation | |
532 | * @syscall: syscall number to send to userland | |
533 | * @reason: filter-supplied reason code to send to userland (via si_errno) | |
534 | * | |
535 | * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info. | |
536 | */ | |
537 | static void seccomp_send_sigsys(int syscall, int reason) | |
538 | { | |
539 | struct siginfo info; | |
540 | memset(&info, 0, sizeof(info)); | |
541 | info.si_signo = SIGSYS; | |
542 | info.si_code = SYS_SECCOMP; | |
543 | info.si_call_addr = (void __user *)KSTK_EIP(current); | |
544 | info.si_errno = reason; | |
6fa3eb70 | 545 | info.si_arch = syscall_get_arch(); |
bb6ea430 WD |
546 | info.si_syscall = syscall; |
547 | force_sig_info(SIGSYS, &info, current); | |
548 | } | |
e2cfabdf | 549 | #endif /* CONFIG_SECCOMP_FILTER */ |
1da177e4 LT |
550 | |
551 | /* | |
552 | * Secure computing mode 1 allows only read/write/exit/sigreturn. | |
553 | * To be fully secure this must be combined with rlimit | |
554 | * to limit the stack allocations too. | |
555 | */ | |
556 | static int mode1_syscalls[] = { | |
557 | __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, | |
558 | 0, /* null terminated */ | |
559 | }; | |
560 | ||
5b101740 | 561 | #ifdef CONFIG_COMPAT |
1da177e4 LT |
562 | static int mode1_syscalls_32[] = { |
563 | __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32, | |
564 | 0, /* null terminated */ | |
565 | }; | |
566 | #endif | |
567 | ||
acf3b2c7 | 568 | int __secure_computing(int this_syscall) |
1da177e4 | 569 | { |
e2cfabdf WD |
570 | int exit_sig = 0; |
571 | int *syscall; | |
8156b451 | 572 | u32 ret; |
1da177e4 | 573 | |
6fa3eb70 S |
574 | /* |
575 | * Make sure that any changes to mode from another thread have | |
576 | * been seen after TIF_SECCOMP was seen. | |
577 | */ | |
578 | rmb(); | |
579 | ||
580 | switch (current->seccomp.mode) { | |
e2cfabdf | 581 | case SECCOMP_MODE_STRICT: |
1da177e4 | 582 | syscall = mode1_syscalls; |
5b101740 RM |
583 | #ifdef CONFIG_COMPAT |
584 | if (is_compat_task()) | |
1da177e4 LT |
585 | syscall = mode1_syscalls_32; |
586 | #endif | |
587 | do { | |
588 | if (*syscall == this_syscall) | |
acf3b2c7 | 589 | return 0; |
1da177e4 | 590 | } while (*++syscall); |
e2cfabdf | 591 | exit_sig = SIGKILL; |
8156b451 | 592 | ret = SECCOMP_RET_KILL; |
1da177e4 | 593 | break; |
e2cfabdf | 594 | #ifdef CONFIG_SECCOMP_FILTER |
8156b451 WD |
595 | case SECCOMP_MODE_FILTER: { |
596 | int data; | |
87b526d3 | 597 | struct pt_regs *regs = task_pt_regs(current); |
acf3b2c7 WD |
598 | ret = seccomp_run_filters(this_syscall); |
599 | data = ret & SECCOMP_RET_DATA; | |
8156b451 WD |
600 | ret &= SECCOMP_RET_ACTION; |
601 | switch (ret) { | |
acf3b2c7 WD |
602 | case SECCOMP_RET_ERRNO: |
603 | /* Set the low-order 16-bits as a errno. */ | |
87b526d3 | 604 | syscall_set_return_value(current, regs, |
acf3b2c7 WD |
605 | -data, 0); |
606 | goto skip; | |
bb6ea430 WD |
607 | case SECCOMP_RET_TRAP: |
608 | /* Show the handler the original registers. */ | |
87b526d3 | 609 | syscall_rollback(current, regs); |
bb6ea430 WD |
610 | /* Let the filter pass back 16 bits of data. */ |
611 | seccomp_send_sigsys(this_syscall, data); | |
612 | goto skip; | |
fb0fadf9 WD |
613 | case SECCOMP_RET_TRACE: |
614 | /* Skip these calls if there is no tracer. */ | |
87b526d3 AL |
615 | if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { |
616 | syscall_set_return_value(current, regs, | |
617 | -ENOSYS, 0); | |
fb0fadf9 | 618 | goto skip; |
87b526d3 | 619 | } |
fb0fadf9 WD |
620 | /* Allow the BPF to provide the event message */ |
621 | ptrace_event(PTRACE_EVENT_SECCOMP, data); | |
622 | /* | |
623 | * The delivery of a fatal signal during event | |
624 | * notification may silently skip tracer notification. | |
625 | * Terminating the task now avoids executing a system | |
626 | * call that may not be intended. | |
627 | */ | |
628 | if (fatal_signal_pending(current)) | |
629 | break; | |
87b526d3 AL |
630 | if (syscall_get_nr(current, regs) < 0) |
631 | goto skip; /* Explicit request to skip. */ | |
632 | ||
fb0fadf9 | 633 | return 0; |
acf3b2c7 WD |
634 | case SECCOMP_RET_ALLOW: |
635 | return 0; | |
636 | case SECCOMP_RET_KILL: | |
637 | default: | |
638 | break; | |
639 | } | |
e2cfabdf WD |
640 | exit_sig = SIGSYS; |
641 | break; | |
8156b451 | 642 | } |
e2cfabdf | 643 | #endif |
1da177e4 LT |
644 | default: |
645 | BUG(); | |
646 | } | |
647 | ||
648 | #ifdef SECCOMP_DEBUG | |
649 | dump_stack(); | |
650 | #endif | |
acf3b2c7 | 651 | audit_seccomp(this_syscall, exit_sig, ret); |
e2cfabdf | 652 | do_exit(exit_sig); |
8156b451 | 653 | #ifdef CONFIG_SECCOMP_FILTER |
acf3b2c7 WD |
654 | skip: |
655 | audit_seccomp(this_syscall, exit_sig, ret); | |
8156b451 | 656 | #endif |
acf3b2c7 | 657 | return -1; |
1da177e4 | 658 | } |
1d9d02fe AA |
659 | |
660 | long prctl_get_seccomp(void) | |
661 | { | |
662 | return current->seccomp.mode; | |
663 | } | |
664 | ||
e2cfabdf | 665 | /** |
6fa3eb70 S |
666 | * seccomp_set_mode_strict: internal function for setting strict seccomp |
667 | * | |
668 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
669 | * | |
670 | * Returns 0 on success or -EINVAL on failure. | |
671 | */ | |
672 | static long seccomp_set_mode_strict(void) | |
673 | { | |
674 | const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; | |
675 | long ret = -EINVAL; | |
676 | ||
677 | spin_lock_irq(¤t->sighand->siglock); | |
678 | ||
679 | if (!seccomp_may_assign_mode(seccomp_mode)) | |
680 | goto out; | |
681 | ||
682 | #ifdef TIF_NOTSC | |
683 | disable_TSC(); | |
684 | #endif | |
685 | seccomp_assign_mode(current, seccomp_mode); | |
686 | ret = 0; | |
687 | ||
688 | out: | |
689 | spin_unlock_irq(¤t->sighand->siglock); | |
690 | ||
691 | return ret; | |
692 | } | |
693 | ||
694 | #ifdef CONFIG_SECCOMP_FILTER | |
695 | /** | |
696 | * seccomp_set_mode_filter: internal function for setting seccomp filter | |
697 | * @flags: flags to change filter behavior | |
698 | * @filter: struct sock_fprog containing filter | |
e2cfabdf | 699 | * |
6fa3eb70 S |
700 | * This function may be called repeatedly to install additional filters. |
701 | * Every filter successfully installed will be evaluated (in reverse order) | |
702 | * for each system call the task makes. | |
e2cfabdf WD |
703 | * |
704 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
705 | * | |
706 | * Returns 0 on success or -EINVAL on failure. | |
707 | */ | |
6fa3eb70 S |
708 | static long seccomp_set_mode_filter(unsigned int flags, |
709 | const char __user *filter) | |
1d9d02fe | 710 | { |
6fa3eb70 S |
711 | const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; |
712 | struct seccomp_filter *prepared = NULL; | |
e2cfabdf | 713 | long ret = -EINVAL; |
1d9d02fe | 714 | |
6fa3eb70 S |
715 | /* Validate flags. */ |
716 | if (flags & ~SECCOMP_FILTER_FLAG_MASK) | |
717 | return -EINVAL; | |
718 | ||
719 | /* Prepare the new filter before holding any locks. */ | |
720 | prepared = seccomp_prepare_user_filter(filter); | |
721 | if (IS_ERR(prepared)) | |
722 | return PTR_ERR(prepared); | |
723 | ||
724 | /* | |
725 | * Make sure we cannot change seccomp or nnp state via TSYNC | |
726 | * while another thread is in the middle of calling exec. | |
727 | */ | |
728 | if (flags & SECCOMP_FILTER_FLAG_TSYNC && | |
729 | mutex_lock_killable(¤t->signal->cred_guard_mutex)) | |
730 | goto out_free; | |
731 | ||
732 | spin_lock_irq(¤t->sighand->siglock); | |
733 | ||
734 | if (!seccomp_may_assign_mode(seccomp_mode)) | |
735 | goto out; | |
736 | ||
737 | ret = seccomp_attach_filter(flags, prepared); | |
738 | if (ret) | |
1d9d02fe | 739 | goto out; |
6fa3eb70 S |
740 | /* Do not free the successfully attached filter. */ |
741 | prepared = NULL; | |
742 | ||
743 | seccomp_assign_mode(current, seccomp_mode); | |
744 | out: | |
745 | spin_unlock_irq(¤t->sighand->siglock); | |
746 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) | |
747 | mutex_unlock(¤t->signal->cred_guard_mutex); | |
748 | out_free: | |
749 | seccomp_filter_free(prepared); | |
750 | return ret; | |
751 | } | |
752 | #else | |
753 | static inline long seccomp_set_mode_filter(unsigned int flags, | |
754 | const char __user *filter) | |
755 | { | |
756 | return -EINVAL; | |
757 | } | |
758 | #endif | |
759 | ||
760 | /* Common entry point for both prctl and syscall. */ | |
761 | static long do_seccomp(unsigned int op, unsigned int flags, | |
762 | const char __user *uargs) | |
763 | { | |
764 | switch (op) { | |
765 | case SECCOMP_SET_MODE_STRICT: | |
766 | if (flags != 0 || uargs != NULL) | |
767 | return -EINVAL; | |
768 | return seccomp_set_mode_strict(); | |
769 | case SECCOMP_SET_MODE_FILTER: | |
770 | return seccomp_set_mode_filter(flags, uargs); | |
771 | default: | |
772 | return -EINVAL; | |
773 | } | |
774 | } | |
775 | ||
776 | SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, | |
777 | const char __user *, uargs) | |
778 | { | |
779 | return do_seccomp(op, flags, uargs); | |
780 | } | |
781 | ||
782 | /** | |
783 | * prctl_set_seccomp: configures current->seccomp.mode | |
784 | * @seccomp_mode: requested mode to use | |
785 | * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER | |
786 | * | |
787 | * Returns 0 on success or -EINVAL on failure. | |
788 | */ | |
789 | long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter) | |
790 | { | |
791 | unsigned int op; | |
792 | char __user *uargs; | |
1d9d02fe | 793 | |
e2cfabdf WD |
794 | switch (seccomp_mode) { |
795 | case SECCOMP_MODE_STRICT: | |
6fa3eb70 S |
796 | op = SECCOMP_SET_MODE_STRICT; |
797 | /* | |
798 | * Setting strict mode through prctl always ignored filter, | |
799 | * so make sure it is always NULL here to pass the internal | |
800 | * check in do_seccomp(). | |
801 | */ | |
802 | uargs = NULL; | |
e2cfabdf | 803 | break; |
e2cfabdf | 804 | case SECCOMP_MODE_FILTER: |
6fa3eb70 S |
805 | op = SECCOMP_SET_MODE_FILTER; |
806 | uargs = filter; | |
e2cfabdf | 807 | break; |
e2cfabdf | 808 | default: |
6fa3eb70 | 809 | return -EINVAL; |
1d9d02fe AA |
810 | } |
811 | ||
6fa3eb70 S |
812 | /* prctl interface doesn't have flags, so they are always zero. */ |
813 | return do_seccomp(op, 0, uargs); | |
1d9d02fe | 814 | } |