]>
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> |
9ca58825 | 20 | #include <linux/kmemleak.h> |
7d85e158 KC |
21 | #include <linux/nospec.h> |
22 | #include <linux/prctl.h> | |
e2cfabdf | 23 | #include <linux/sched.h> |
68db0cf1 | 24 | #include <linux/sched/task_stack.h> |
e2cfabdf | 25 | #include <linux/seccomp.h> |
c8bee430 | 26 | #include <linux/slab.h> |
48dc92b9 | 27 | #include <linux/syscalls.h> |
9ca58825 | 28 | #include <linux/sysctl.h> |
1da177e4 | 29 | |
a4412fc9 | 30 | #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER |
e2cfabdf | 31 | #include <asm/syscall.h> |
a4412fc9 | 32 | #endif |
e2cfabdf WD |
33 | |
34 | #ifdef CONFIG_SECCOMP_FILTER | |
e2cfabdf | 35 | #include <linux/filter.h> |
c2e1f2e3 | 36 | #include <linux/pid.h> |
fb0fadf9 | 37 | #include <linux/ptrace.h> |
e2cfabdf | 38 | #include <linux/security.h> |
e2cfabdf WD |
39 | #include <linux/tracehook.h> |
40 | #include <linux/uaccess.h> | |
41 | ||
42 | /** | |
43 | * struct seccomp_filter - container for seccomp BPF programs | |
44 | * | |
45 | * @usage: reference count to manage the object lifetime. | |
46 | * get/put helpers should be used when accessing an instance | |
47 | * outside of a lifetime-guarded section. In general, this | |
48 | * is only needed for handling filters shared across tasks. | |
8ac8bbe1 | 49 | * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged |
e2cfabdf | 50 | * @prev: points to a previously installed, or inherited, filter |
285fdfc5 | 51 | * @prog: the BPF program to evaluate |
e2cfabdf WD |
52 | * |
53 | * seccomp_filter objects are organized in a tree linked via the @prev | |
54 | * pointer. For any task, it appears to be a singly-linked list starting | |
55 | * with current->seccomp.filter, the most recently attached or inherited filter. | |
56 | * However, multiple filters may share a @prev node, by way of fork(), which | |
57 | * results in a unidirectional tree existing in memory. This is similar to | |
58 | * how namespaces work. | |
59 | * | |
60 | * seccomp_filter objects should never be modified after being attached | |
61 | * to a task_struct (other than @usage). | |
62 | */ | |
63 | struct seccomp_filter { | |
0b5fa229 | 64 | refcount_t usage; |
8ac8bbe1 | 65 | bool log; |
e2cfabdf | 66 | struct seccomp_filter *prev; |
7ae457c1 | 67 | struct bpf_prog *prog; |
e2cfabdf WD |
68 | }; |
69 | ||
70 | /* Limit any path through the tree to 256KB worth of instructions. */ | |
71 | #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter)) | |
72 | ||
bd4cf0ed | 73 | /* |
e2cfabdf WD |
74 | * Endianness is explicitly ignored and left for BPF program authors to manage |
75 | * as per the specific architecture. | |
76 | */ | |
bd4cf0ed | 77 | static void populate_seccomp_data(struct seccomp_data *sd) |
e2cfabdf | 78 | { |
bd4cf0ed AS |
79 | struct task_struct *task = current; |
80 | struct pt_regs *regs = task_pt_regs(task); | |
2eac7648 | 81 | unsigned long args[6]; |
e2cfabdf | 82 | |
bd4cf0ed | 83 | sd->nr = syscall_get_nr(task, regs); |
0b747172 | 84 | sd->arch = syscall_get_arch(); |
2eac7648 DB |
85 | syscall_get_arguments(task, regs, 0, 6, args); |
86 | sd->args[0] = args[0]; | |
87 | sd->args[1] = args[1]; | |
88 | sd->args[2] = args[2]; | |
89 | sd->args[3] = args[3]; | |
90 | sd->args[4] = args[4]; | |
91 | sd->args[5] = args[5]; | |
bd4cf0ed | 92 | sd->instruction_pointer = KSTK_EIP(task); |
e2cfabdf WD |
93 | } |
94 | ||
95 | /** | |
96 | * seccomp_check_filter - verify seccomp filter code | |
97 | * @filter: filter to verify | |
98 | * @flen: length of filter | |
99 | * | |
4df95ff4 | 100 | * Takes a previously checked filter (by bpf_check_classic) and |
e2cfabdf WD |
101 | * redirects all filter code that loads struct sk_buff data |
102 | * and related data through seccomp_bpf_load. It also | |
103 | * enforces length and alignment checking of those loads. | |
104 | * | |
105 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
106 | */ | |
107 | static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen) | |
108 | { | |
109 | int pc; | |
110 | for (pc = 0; pc < flen; pc++) { | |
111 | struct sock_filter *ftest = &filter[pc]; | |
112 | u16 code = ftest->code; | |
113 | u32 k = ftest->k; | |
114 | ||
115 | switch (code) { | |
34805931 | 116 | case BPF_LD | BPF_W | BPF_ABS: |
bd4cf0ed | 117 | ftest->code = BPF_LDX | BPF_W | BPF_ABS; |
e2cfabdf WD |
118 | /* 32-bit aligned and not out of bounds. */ |
119 | if (k >= sizeof(struct seccomp_data) || k & 3) | |
120 | return -EINVAL; | |
121 | continue; | |
34805931 | 122 | case BPF_LD | BPF_W | BPF_LEN: |
bd4cf0ed | 123 | ftest->code = BPF_LD | BPF_IMM; |
e2cfabdf WD |
124 | ftest->k = sizeof(struct seccomp_data); |
125 | continue; | |
34805931 | 126 | case BPF_LDX | BPF_W | BPF_LEN: |
bd4cf0ed | 127 | ftest->code = BPF_LDX | BPF_IMM; |
e2cfabdf WD |
128 | ftest->k = sizeof(struct seccomp_data); |
129 | continue; | |
130 | /* Explicitly include allowed calls. */ | |
34805931 DB |
131 | case BPF_RET | BPF_K: |
132 | case BPF_RET | BPF_A: | |
133 | case BPF_ALU | BPF_ADD | BPF_K: | |
134 | case BPF_ALU | BPF_ADD | BPF_X: | |
135 | case BPF_ALU | BPF_SUB | BPF_K: | |
136 | case BPF_ALU | BPF_SUB | BPF_X: | |
137 | case BPF_ALU | BPF_MUL | BPF_K: | |
138 | case BPF_ALU | BPF_MUL | BPF_X: | |
139 | case BPF_ALU | BPF_DIV | BPF_K: | |
140 | case BPF_ALU | BPF_DIV | BPF_X: | |
141 | case BPF_ALU | BPF_AND | BPF_K: | |
142 | case BPF_ALU | BPF_AND | BPF_X: | |
143 | case BPF_ALU | BPF_OR | BPF_K: | |
144 | case BPF_ALU | BPF_OR | BPF_X: | |
145 | case BPF_ALU | BPF_XOR | BPF_K: | |
146 | case BPF_ALU | BPF_XOR | BPF_X: | |
147 | case BPF_ALU | BPF_LSH | BPF_K: | |
148 | case BPF_ALU | BPF_LSH | BPF_X: | |
149 | case BPF_ALU | BPF_RSH | BPF_K: | |
150 | case BPF_ALU | BPF_RSH | BPF_X: | |
151 | case BPF_ALU | BPF_NEG: | |
152 | case BPF_LD | BPF_IMM: | |
153 | case BPF_LDX | BPF_IMM: | |
154 | case BPF_MISC | BPF_TAX: | |
155 | case BPF_MISC | BPF_TXA: | |
156 | case BPF_LD | BPF_MEM: | |
157 | case BPF_LDX | BPF_MEM: | |
158 | case BPF_ST: | |
159 | case BPF_STX: | |
160 | case BPF_JMP | BPF_JA: | |
161 | case BPF_JMP | BPF_JEQ | BPF_K: | |
162 | case BPF_JMP | BPF_JEQ | BPF_X: | |
163 | case BPF_JMP | BPF_JGE | BPF_K: | |
164 | case BPF_JMP | BPF_JGE | BPF_X: | |
165 | case BPF_JMP | BPF_JGT | BPF_K: | |
166 | case BPF_JMP | BPF_JGT | BPF_X: | |
167 | case BPF_JMP | BPF_JSET | BPF_K: | |
168 | case BPF_JMP | BPF_JSET | BPF_X: | |
e2cfabdf WD |
169 | continue; |
170 | default: | |
171 | return -EINVAL; | |
172 | } | |
173 | } | |
174 | return 0; | |
175 | } | |
176 | ||
177 | /** | |
285fdfc5 MS |
178 | * seccomp_run_filters - evaluates all seccomp filters against @sd |
179 | * @sd: optional seccomp data to be passed to filters | |
e95596f2 KC |
180 | * @match: stores struct seccomp_filter that resulted in the return value, |
181 | * unless filter returned SECCOMP_RET_ALLOW, in which case it will | |
182 | * be unchanged. | |
e2cfabdf WD |
183 | * |
184 | * Returns valid seccomp BPF response codes. | |
185 | */ | |
e95596f2 KC |
186 | static u32 seccomp_run_filters(const struct seccomp_data *sd, |
187 | struct seccomp_filter **match) | |
e2cfabdf | 188 | { |
d39bd00d | 189 | struct seccomp_data sd_local; |
acf3b2c7 | 190 | u32 ret = SECCOMP_RET_ALLOW; |
8225d385 PK |
191 | /* Make sure cross-thread synced filter points somewhere sane. */ |
192 | struct seccomp_filter *f = | |
7252704b | 193 | READ_ONCE(current->seccomp.filter); |
acf3b2c7 WD |
194 | |
195 | /* Ensure unexpected behavior doesn't result in failing open. */ | |
3ba2530c | 196 | if (unlikely(WARN_ON(f == NULL))) |
acf3b2c7 WD |
197 | return SECCOMP_RET_KILL; |
198 | ||
d39bd00d AL |
199 | if (!sd) { |
200 | populate_seccomp_data(&sd_local); | |
201 | sd = &sd_local; | |
202 | } | |
bd4cf0ed | 203 | |
e2cfabdf WD |
204 | /* |
205 | * All filters in the list are evaluated and the lowest BPF return | |
acf3b2c7 | 206 | * value always takes priority (ignoring the DATA). |
e2cfabdf | 207 | */ |
3ba2530c | 208 | for (; f; f = f->prev) { |
88575199 | 209 | u32 cur_ret = BPF_PROG_RUN(f->prog, sd); |
8f577cad | 210 | |
e95596f2 | 211 | if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION)) { |
acf3b2c7 | 212 | ret = cur_ret; |
e95596f2 KC |
213 | *match = f; |
214 | } | |
e2cfabdf WD |
215 | } |
216 | return ret; | |
217 | } | |
1f41b450 | 218 | #endif /* CONFIG_SECCOMP_FILTER */ |
e2cfabdf | 219 | |
1f41b450 KC |
220 | static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) |
221 | { | |
69f6a34b | 222 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 223 | |
1f41b450 KC |
224 | if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) |
225 | return false; | |
226 | ||
227 | return true; | |
228 | } | |
229 | ||
7d85e158 KC |
230 | /* |
231 | * If a given speculation mitigation is opt-in (prctl()-controlled), | |
232 | * select it, by disabling speculation (enabling mitigation). | |
233 | */ | |
234 | static inline void spec_mitigate(struct task_struct *task, | |
235 | unsigned long which) | |
236 | { | |
237 | int state = arch_prctl_spec_ctrl_get(task, which); | |
238 | ||
239 | if (state > 0 && (state & PR_SPEC_PRCTL)) | |
240 | arch_prctl_spec_ctrl_set(task, which, PR_SPEC_DISABLE); | |
241 | } | |
242 | ||
3ba2530c KC |
243 | static inline void seccomp_assign_mode(struct task_struct *task, |
244 | unsigned long seccomp_mode) | |
1f41b450 | 245 | { |
69f6a34b | 246 | assert_spin_locked(&task->sighand->siglock); |
dbd95212 | 247 | |
3ba2530c KC |
248 | task->seccomp.mode = seccomp_mode; |
249 | /* | |
250 | * Make sure TIF_SECCOMP cannot be set before the mode (and | |
251 | * filter) is set. | |
252 | */ | |
253 | smp_mb__before_atomic(); | |
7d85e158 KC |
254 | /* Assume seccomp processes want speculation flaw mitigation. */ |
255 | spec_mitigate(task, PR_SPEC_STORE_BYPASS); | |
3ba2530c | 256 | set_tsk_thread_flag(task, TIF_SECCOMP); |
1f41b450 KC |
257 | } |
258 | ||
259 | #ifdef CONFIG_SECCOMP_FILTER | |
c2e1f2e3 KC |
260 | /* Returns 1 if the parent is an ancestor of the child. */ |
261 | static int is_ancestor(struct seccomp_filter *parent, | |
262 | struct seccomp_filter *child) | |
263 | { | |
264 | /* NULL is the root ancestor. */ | |
265 | if (parent == NULL) | |
266 | return 1; | |
267 | for (; child; child = child->prev) | |
268 | if (child == parent) | |
269 | return 1; | |
270 | return 0; | |
271 | } | |
272 | ||
273 | /** | |
274 | * seccomp_can_sync_threads: checks if all threads can be synchronized | |
275 | * | |
276 | * Expects sighand and cred_guard_mutex locks to be held. | |
277 | * | |
278 | * Returns 0 on success, -ve on error, or the pid of a thread which was | |
279 | * either not in the correct seccomp mode or it did not have an ancestral | |
280 | * seccomp filter. | |
281 | */ | |
282 | static inline pid_t seccomp_can_sync_threads(void) | |
283 | { | |
284 | struct task_struct *thread, *caller; | |
285 | ||
286 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 287 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
288 | |
289 | /* Validate all threads being eligible for synchronization. */ | |
290 | caller = current; | |
291 | for_each_thread(caller, thread) { | |
292 | pid_t failed; | |
293 | ||
294 | /* Skip current, since it is initiating the sync. */ | |
295 | if (thread == caller) | |
296 | continue; | |
297 | ||
298 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || | |
299 | (thread->seccomp.mode == SECCOMP_MODE_FILTER && | |
300 | is_ancestor(thread->seccomp.filter, | |
301 | caller->seccomp.filter))) | |
302 | continue; | |
303 | ||
304 | /* Return the first thread that cannot be synchronized. */ | |
305 | failed = task_pid_vnr(thread); | |
306 | /* If the pid cannot be resolved, then return -ESRCH */ | |
307 | if (unlikely(WARN_ON(failed == 0))) | |
308 | failed = -ESRCH; | |
309 | return failed; | |
310 | } | |
311 | ||
312 | return 0; | |
313 | } | |
314 | ||
315 | /** | |
316 | * seccomp_sync_threads: sets all threads to use current's filter | |
317 | * | |
318 | * Expects sighand and cred_guard_mutex locks to be held, and for | |
319 | * seccomp_can_sync_threads() to have returned success already | |
320 | * without dropping the locks. | |
321 | * | |
322 | */ | |
323 | static inline void seccomp_sync_threads(void) | |
324 | { | |
325 | struct task_struct *thread, *caller; | |
326 | ||
327 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 328 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
329 | |
330 | /* Synchronize all threads. */ | |
331 | caller = current; | |
332 | for_each_thread(caller, thread) { | |
333 | /* Skip current, since it needs no changes. */ | |
334 | if (thread == caller) | |
335 | continue; | |
336 | ||
337 | /* Get a task reference for the new leaf node. */ | |
338 | get_seccomp_filter(caller); | |
339 | /* | |
340 | * Drop the task reference to the shared ancestor since | |
341 | * current's path will hold a reference. (This also | |
342 | * allows a put before the assignment.) | |
343 | */ | |
344 | put_seccomp_filter(thread); | |
345 | smp_store_release(&thread->seccomp.filter, | |
346 | caller->seccomp.filter); | |
103502a3 JH |
347 | |
348 | /* | |
349 | * Don't let an unprivileged task work around | |
350 | * the no_new_privs restriction by creating | |
351 | * a thread that sets it up, enters seccomp, | |
352 | * then dies. | |
353 | */ | |
354 | if (task_no_new_privs(caller)) | |
355 | task_set_no_new_privs(thread); | |
356 | ||
c2e1f2e3 KC |
357 | /* |
358 | * Opt the other thread into seccomp if needed. | |
359 | * As threads are considered to be trust-realm | |
360 | * equivalent (see ptrace_may_access), it is safe to | |
361 | * allow one thread to transition the other. | |
362 | */ | |
103502a3 | 363 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) |
c2e1f2e3 | 364 | seccomp_assign_mode(thread, SECCOMP_MODE_FILTER); |
c2e1f2e3 KC |
365 | } |
366 | } | |
367 | ||
e2cfabdf | 368 | /** |
c8bee430 | 369 | * seccomp_prepare_filter: Prepares a seccomp filter for use. |
e2cfabdf WD |
370 | * @fprog: BPF program to install |
371 | * | |
c8bee430 | 372 | * Returns filter on success or an ERR_PTR on failure. |
e2cfabdf | 373 | */ |
c8bee430 | 374 | static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog) |
e2cfabdf | 375 | { |
ac67eb2c DB |
376 | struct seccomp_filter *sfilter; |
377 | int ret; | |
97f2645f | 378 | const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE); |
e2cfabdf WD |
379 | |
380 | if (fprog->len == 0 || fprog->len > BPF_MAXINSNS) | |
c8bee430 | 381 | return ERR_PTR(-EINVAL); |
d9e12f42 | 382 | |
c8bee430 | 383 | BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter)); |
e2cfabdf WD |
384 | |
385 | /* | |
119ce5c8 | 386 | * Installing a seccomp filter requires that the task has |
e2cfabdf WD |
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 | */ | |
1d4457f9 | 391 | if (!task_no_new_privs(current) && |
e2cfabdf WD |
392 | security_capable_noaudit(current_cred(), current_user_ns(), |
393 | CAP_SYS_ADMIN) != 0) | |
c8bee430 | 394 | return ERR_PTR(-EACCES); |
e2cfabdf | 395 | |
bd4cf0ed | 396 | /* Allocate a new seccomp_filter */ |
ac67eb2c DB |
397 | sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN); |
398 | if (!sfilter) | |
d9e12f42 | 399 | return ERR_PTR(-ENOMEM); |
ac67eb2c DB |
400 | |
401 | ret = bpf_prog_create_from_user(&sfilter->prog, fprog, | |
f8e529ed | 402 | seccomp_check_filter, save_orig); |
ac67eb2c DB |
403 | if (ret < 0) { |
404 | kfree(sfilter); | |
405 | return ERR_PTR(ret); | |
d9e12f42 | 406 | } |
bd4cf0ed | 407 | |
0b5fa229 | 408 | refcount_set(&sfilter->usage, 1); |
e2cfabdf | 409 | |
ac67eb2c | 410 | return sfilter; |
e2cfabdf WD |
411 | } |
412 | ||
413 | /** | |
c8bee430 | 414 | * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog |
e2cfabdf WD |
415 | * @user_filter: pointer to the user data containing a sock_fprog. |
416 | * | |
417 | * Returns 0 on success and non-zero otherwise. | |
418 | */ | |
c8bee430 KC |
419 | static struct seccomp_filter * |
420 | seccomp_prepare_user_filter(const char __user *user_filter) | |
e2cfabdf WD |
421 | { |
422 | struct sock_fprog fprog; | |
c8bee430 | 423 | struct seccomp_filter *filter = ERR_PTR(-EFAULT); |
e2cfabdf WD |
424 | |
425 | #ifdef CONFIG_COMPAT | |
5c38065e | 426 | if (in_compat_syscall()) { |
e2cfabdf WD |
427 | struct compat_sock_fprog fprog32; |
428 | if (copy_from_user(&fprog32, user_filter, sizeof(fprog32))) | |
429 | goto out; | |
430 | fprog.len = fprog32.len; | |
431 | fprog.filter = compat_ptr(fprog32.filter); | |
432 | } else /* falls through to the if below. */ | |
433 | #endif | |
434 | if (copy_from_user(&fprog, user_filter, sizeof(fprog))) | |
435 | goto out; | |
c8bee430 | 436 | filter = seccomp_prepare_filter(&fprog); |
e2cfabdf | 437 | out: |
c8bee430 KC |
438 | return filter; |
439 | } | |
440 | ||
441 | /** | |
442 | * seccomp_attach_filter: validate and attach filter | |
443 | * @flags: flags to change filter behavior | |
444 | * @filter: seccomp filter to add to the current process | |
445 | * | |
dbd95212 KC |
446 | * Caller must be holding current->sighand->siglock lock. |
447 | * | |
c8bee430 KC |
448 | * Returns 0 on success, -ve on error. |
449 | */ | |
450 | static long seccomp_attach_filter(unsigned int flags, | |
451 | struct seccomp_filter *filter) | |
452 | { | |
453 | unsigned long total_insns; | |
454 | struct seccomp_filter *walker; | |
455 | ||
69f6a34b | 456 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 457 | |
c8bee430 KC |
458 | /* Validate resulting filter length. */ |
459 | total_insns = filter->prog->len; | |
460 | for (walker = current->seccomp.filter; walker; walker = walker->prev) | |
461 | total_insns += walker->prog->len + 4; /* 4 instr penalty */ | |
462 | if (total_insns > MAX_INSNS_PER_PATH) | |
463 | return -ENOMEM; | |
464 | ||
c2e1f2e3 KC |
465 | /* If thread sync has been requested, check that it is possible. */ |
466 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) { | |
467 | int ret; | |
468 | ||
469 | ret = seccomp_can_sync_threads(); | |
470 | if (ret) | |
471 | return ret; | |
472 | } | |
473 | ||
8ac8bbe1 TH |
474 | /* Set log flag, if present. */ |
475 | if (flags & SECCOMP_FILTER_FLAG_LOG) | |
476 | filter->log = true; | |
477 | ||
c8bee430 KC |
478 | /* |
479 | * If there is an existing filter, make it the prev and don't drop its | |
480 | * task reference. | |
481 | */ | |
482 | filter->prev = current->seccomp.filter; | |
483 | current->seccomp.filter = filter; | |
484 | ||
c2e1f2e3 KC |
485 | /* Now that the new filter is in place, synchronize to all threads. */ |
486 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) | |
487 | seccomp_sync_threads(); | |
488 | ||
c8bee430 | 489 | return 0; |
e2cfabdf WD |
490 | } |
491 | ||
1eaabc57 ON |
492 | void __get_seccomp_filter(struct seccomp_filter *filter) |
493 | { | |
494 | /* Reference count is bounded by the number of total processes. */ | |
495 | refcount_inc(&filter->usage); | |
496 | } | |
497 | ||
e2cfabdf WD |
498 | /* get_seccomp_filter - increments the reference count of the filter on @tsk */ |
499 | void get_seccomp_filter(struct task_struct *tsk) | |
500 | { | |
501 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
502 | if (!orig) | |
503 | return; | |
1eaabc57 | 504 | __get_seccomp_filter(orig); |
e2cfabdf WD |
505 | } |
506 | ||
c8bee430 KC |
507 | static inline void seccomp_filter_free(struct seccomp_filter *filter) |
508 | { | |
509 | if (filter) { | |
bab18991 | 510 | bpf_prog_destroy(filter->prog); |
c8bee430 KC |
511 | kfree(filter); |
512 | } | |
513 | } | |
514 | ||
1eaabc57 | 515 | static void __put_seccomp_filter(struct seccomp_filter *orig) |
e2cfabdf | 516 | { |
e2cfabdf | 517 | /* Clean up single-reference branches iteratively. */ |
0b5fa229 | 518 | while (orig && refcount_dec_and_test(&orig->usage)) { |
e2cfabdf WD |
519 | struct seccomp_filter *freeme = orig; |
520 | orig = orig->prev; | |
c8bee430 | 521 | seccomp_filter_free(freeme); |
e2cfabdf WD |
522 | } |
523 | } | |
bb6ea430 | 524 | |
1eaabc57 ON |
525 | /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */ |
526 | void put_seccomp_filter(struct task_struct *tsk) | |
527 | { | |
528 | __put_seccomp_filter(tsk->seccomp.filter); | |
529 | } | |
530 | ||
b25e6716 MF |
531 | static void seccomp_init_siginfo(siginfo_t *info, int syscall, int reason) |
532 | { | |
533 | memset(info, 0, sizeof(*info)); | |
534 | info->si_signo = SIGSYS; | |
535 | info->si_code = SYS_SECCOMP; | |
536 | info->si_call_addr = (void __user *)KSTK_EIP(current); | |
537 | info->si_errno = reason; | |
538 | info->si_arch = syscall_get_arch(); | |
539 | info->si_syscall = syscall; | |
540 | } | |
541 | ||
bb6ea430 WD |
542 | /** |
543 | * seccomp_send_sigsys - signals the task to allow in-process syscall emulation | |
544 | * @syscall: syscall number to send to userland | |
545 | * @reason: filter-supplied reason code to send to userland (via si_errno) | |
546 | * | |
547 | * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info. | |
548 | */ | |
549 | static void seccomp_send_sigsys(int syscall, int reason) | |
550 | { | |
551 | struct siginfo info; | |
b25e6716 | 552 | seccomp_init_siginfo(&info, syscall, reason); |
bb6ea430 WD |
553 | force_sig_info(SIGSYS, &info, current); |
554 | } | |
e2cfabdf | 555 | #endif /* CONFIG_SECCOMP_FILTER */ |
1da177e4 | 556 | |
56eb631b TH |
557 | /* For use with seccomp_actions_logged */ |
558 | #define SECCOMP_LOG_KILL (1 << 0) | |
559 | #define SECCOMP_LOG_TRAP (1 << 2) | |
560 | #define SECCOMP_LOG_ERRNO (1 << 3) | |
561 | #define SECCOMP_LOG_TRACE (1 << 4) | |
4267083e TH |
562 | #define SECCOMP_LOG_LOG (1 << 5) |
563 | #define SECCOMP_LOG_ALLOW (1 << 6) | |
56eb631b TH |
564 | |
565 | static u32 seccomp_actions_logged = SECCOMP_LOG_KILL | SECCOMP_LOG_TRAP | | |
4267083e TH |
566 | SECCOMP_LOG_ERRNO | SECCOMP_LOG_TRACE | |
567 | SECCOMP_LOG_LOG; | |
56eb631b | 568 | |
8ac8bbe1 TH |
569 | static inline void seccomp_log(unsigned long syscall, long signr, u32 action, |
570 | bool requested) | |
56eb631b TH |
571 | { |
572 | bool log = false; | |
573 | ||
574 | switch (action) { | |
575 | case SECCOMP_RET_ALLOW: | |
8ac8bbe1 | 576 | break; |
56eb631b | 577 | case SECCOMP_RET_TRAP: |
8ac8bbe1 TH |
578 | log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP; |
579 | break; | |
56eb631b | 580 | case SECCOMP_RET_ERRNO: |
8ac8bbe1 TH |
581 | log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO; |
582 | break; | |
56eb631b | 583 | case SECCOMP_RET_TRACE: |
8ac8bbe1 | 584 | log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE; |
56eb631b | 585 | break; |
4267083e TH |
586 | case SECCOMP_RET_LOG: |
587 | log = seccomp_actions_logged & SECCOMP_LOG_LOG; | |
588 | break; | |
56eb631b TH |
589 | case SECCOMP_RET_KILL: |
590 | default: | |
591 | log = seccomp_actions_logged & SECCOMP_LOG_KILL; | |
592 | } | |
593 | ||
594 | /* | |
4267083e TH |
595 | * Force an audit message to be emitted when the action is RET_KILL, |
596 | * RET_LOG, or the FILTER_FLAG_LOG bit was set and the action is | |
597 | * allowed to be logged by the admin. | |
56eb631b TH |
598 | */ |
599 | if (log) | |
600 | return __audit_seccomp(syscall, signr, action); | |
601 | ||
602 | /* | |
603 | * Let the audit subsystem decide if the action should be audited based | |
604 | * on whether the current task itself is being audited. | |
605 | */ | |
606 | return audit_seccomp(syscall, signr, action); | |
607 | } | |
608 | ||
1da177e4 LT |
609 | /* |
610 | * Secure computing mode 1 allows only read/write/exit/sigreturn. | |
611 | * To be fully secure this must be combined with rlimit | |
612 | * to limit the stack allocations too. | |
613 | */ | |
cb4253aa | 614 | static const int mode1_syscalls[] = { |
1da177e4 LT |
615 | __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, |
616 | 0, /* null terminated */ | |
617 | }; | |
618 | ||
a4412fc9 | 619 | static void __secure_computing_strict(int this_syscall) |
1da177e4 | 620 | { |
cb4253aa | 621 | const int *syscall_whitelist = mode1_syscalls; |
a4412fc9 | 622 | #ifdef CONFIG_COMPAT |
5c38065e | 623 | if (in_compat_syscall()) |
c983f0e8 | 624 | syscall_whitelist = get_compat_mode1_syscalls(); |
a4412fc9 AL |
625 | #endif |
626 | do { | |
627 | if (*syscall_whitelist == this_syscall) | |
628 | return; | |
629 | } while (*++syscall_whitelist); | |
630 | ||
631 | #ifdef SECCOMP_DEBUG | |
632 | dump_stack(); | |
633 | #endif | |
8ac8bbe1 | 634 | seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL, true); |
a4412fc9 AL |
635 | do_exit(SIGKILL); |
636 | } | |
637 | ||
638 | #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER | |
639 | void secure_computing_strict(int this_syscall) | |
640 | { | |
641 | int mode = current->seccomp.mode; | |
642 | ||
97f2645f | 643 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 TA |
644 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
645 | return; | |
646 | ||
221272f9 | 647 | if (mode == SECCOMP_MODE_DISABLED) |
a4412fc9 AL |
648 | return; |
649 | else if (mode == SECCOMP_MODE_STRICT) | |
650 | __secure_computing_strict(this_syscall); | |
651 | else | |
652 | BUG(); | |
653 | } | |
654 | #else | |
13aa72f0 AL |
655 | |
656 | #ifdef CONFIG_SECCOMP_FILTER | |
ce6526e8 KC |
657 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
658 | const bool recheck_after_trace) | |
13aa72f0 AL |
659 | { |
660 | u32 filter_ret, action; | |
e95596f2 | 661 | struct seccomp_filter *match = NULL; |
13aa72f0 | 662 | int data; |
1da177e4 | 663 | |
3ba2530c KC |
664 | /* |
665 | * Make sure that any changes to mode from another thread have | |
666 | * been seen after TIF_SECCOMP was seen. | |
667 | */ | |
668 | rmb(); | |
669 | ||
e95596f2 | 670 | filter_ret = seccomp_run_filters(sd, &match); |
13aa72f0 AL |
671 | data = filter_ret & SECCOMP_RET_DATA; |
672 | action = filter_ret & SECCOMP_RET_ACTION; | |
673 | ||
674 | switch (action) { | |
675 | case SECCOMP_RET_ERRNO: | |
580c57f1 KC |
676 | /* Set low-order bits as an errno, capped at MAX_ERRNO. */ |
677 | if (data > MAX_ERRNO) | |
678 | data = MAX_ERRNO; | |
d39bd00d | 679 | syscall_set_return_value(current, task_pt_regs(current), |
13aa72f0 AL |
680 | -data, 0); |
681 | goto skip; | |
682 | ||
683 | case SECCOMP_RET_TRAP: | |
684 | /* Show the handler the original registers. */ | |
d39bd00d | 685 | syscall_rollback(current, task_pt_regs(current)); |
13aa72f0 AL |
686 | /* Let the filter pass back 16 bits of data. */ |
687 | seccomp_send_sigsys(this_syscall, data); | |
688 | goto skip; | |
689 | ||
690 | case SECCOMP_RET_TRACE: | |
ce6526e8 KC |
691 | /* We've been put in this state by the ptracer already. */ |
692 | if (recheck_after_trace) | |
693 | return 0; | |
694 | ||
8112c4f1 KC |
695 | /* ENOSYS these calls if there is no tracer attached. */ |
696 | if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { | |
697 | syscall_set_return_value(current, | |
698 | task_pt_regs(current), | |
699 | -ENOSYS, 0); | |
700 | goto skip; | |
701 | } | |
702 | ||
703 | /* Allow the BPF to provide the event message */ | |
704 | ptrace_event(PTRACE_EVENT_SECCOMP, data); | |
705 | /* | |
706 | * The delivery of a fatal signal during event | |
485a252a KC |
707 | * notification may silently skip tracer notification, |
708 | * which could leave us with a potentially unmodified | |
709 | * syscall that the tracer would have liked to have | |
710 | * changed. Since the process is about to die, we just | |
711 | * force the syscall to be skipped and let the signal | |
712 | * kill the process and correctly handle any tracer exit | |
713 | * notifications. | |
8112c4f1 KC |
714 | */ |
715 | if (fatal_signal_pending(current)) | |
485a252a | 716 | goto skip; |
8112c4f1 KC |
717 | /* Check if the tracer forced the syscall to be skipped. */ |
718 | this_syscall = syscall_get_nr(current, task_pt_regs(current)); | |
719 | if (this_syscall < 0) | |
720 | goto skip; | |
721 | ||
ce6526e8 KC |
722 | /* |
723 | * Recheck the syscall, since it may have changed. This | |
724 | * intentionally uses a NULL struct seccomp_data to force | |
725 | * a reload of all registers. This does not goto skip since | |
726 | * a skip would have already been reported. | |
727 | */ | |
728 | if (__seccomp_filter(this_syscall, NULL, true)) | |
729 | return -1; | |
730 | ||
8112c4f1 | 731 | return 0; |
13aa72f0 | 732 | |
4267083e TH |
733 | case SECCOMP_RET_LOG: |
734 | seccomp_log(this_syscall, 0, action, true); | |
735 | return 0; | |
736 | ||
13aa72f0 | 737 | case SECCOMP_RET_ALLOW: |
e95596f2 KC |
738 | /* |
739 | * Note that the "match" filter will always be NULL for | |
740 | * this action since SECCOMP_RET_ALLOW is the starting | |
741 | * state in seccomp_run_filters(). | |
742 | */ | |
8112c4f1 | 743 | return 0; |
13aa72f0 AL |
744 | |
745 | case SECCOMP_RET_KILL: | |
131b6351 | 746 | default: |
8ac8bbe1 | 747 | seccomp_log(this_syscall, SIGSYS, action, true); |
d7276e32 KC |
748 | /* Dump core only if this is the last remaining thread. */ |
749 | if (get_nr_threads(current) == 1) { | |
131b6351 KC |
750 | siginfo_t info; |
751 | ||
d7276e32 KC |
752 | /* Show the original registers in the dump. */ |
753 | syscall_rollback(current, task_pt_regs(current)); | |
754 | /* Trigger a manual coredump since do_exit skips it. */ | |
755 | seccomp_init_siginfo(&info, this_syscall, data); | |
756 | do_coredump(&info); | |
757 | } | |
13aa72f0 AL |
758 | do_exit(SIGSYS); |
759 | } | |
760 | ||
761 | unreachable(); | |
762 | ||
763 | skip: | |
8ac8bbe1 | 764 | seccomp_log(this_syscall, 0, action, match ? match->log : false); |
8112c4f1 KC |
765 | return -1; |
766 | } | |
767 | #else | |
ce6526e8 KC |
768 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
769 | const bool recheck_after_trace) | |
8112c4f1 KC |
770 | { |
771 | BUG(); | |
13aa72f0 | 772 | } |
1da177e4 | 773 | #endif |
13aa72f0 | 774 | |
8112c4f1 | 775 | int __secure_computing(const struct seccomp_data *sd) |
13aa72f0 AL |
776 | { |
777 | int mode = current->seccomp.mode; | |
8112c4f1 | 778 | int this_syscall; |
13aa72f0 | 779 | |
97f2645f | 780 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 | 781 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
8112c4f1 KC |
782 | return 0; |
783 | ||
784 | this_syscall = sd ? sd->nr : | |
785 | syscall_get_nr(current, task_pt_regs(current)); | |
13c4a901 | 786 | |
13aa72f0 | 787 | switch (mode) { |
e2cfabdf | 788 | case SECCOMP_MODE_STRICT: |
13aa72f0 | 789 | __secure_computing_strict(this_syscall); /* may call do_exit */ |
8112c4f1 | 790 | return 0; |
13aa72f0 | 791 | case SECCOMP_MODE_FILTER: |
ce6526e8 | 792 | return __seccomp_filter(this_syscall, sd, false); |
1da177e4 LT |
793 | default: |
794 | BUG(); | |
795 | } | |
13aa72f0 | 796 | } |
a4412fc9 | 797 | #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ |
1d9d02fe AA |
798 | |
799 | long prctl_get_seccomp(void) | |
800 | { | |
801 | return current->seccomp.mode; | |
802 | } | |
803 | ||
e2cfabdf | 804 | /** |
3b23dd12 | 805 | * seccomp_set_mode_strict: internal function for setting strict seccomp |
e2cfabdf WD |
806 | * |
807 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
808 | * | |
809 | * Returns 0 on success or -EINVAL on failure. | |
810 | */ | |
3b23dd12 | 811 | static long seccomp_set_mode_strict(void) |
1d9d02fe | 812 | { |
3b23dd12 | 813 | const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; |
e2cfabdf | 814 | long ret = -EINVAL; |
1d9d02fe | 815 | |
dbd95212 KC |
816 | spin_lock_irq(¤t->sighand->siglock); |
817 | ||
1f41b450 | 818 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1d9d02fe AA |
819 | goto out; |
820 | ||
cf99abac | 821 | #ifdef TIF_NOTSC |
3b23dd12 | 822 | disable_TSC(); |
cf99abac | 823 | #endif |
3ba2530c | 824 | seccomp_assign_mode(current, seccomp_mode); |
3b23dd12 KC |
825 | ret = 0; |
826 | ||
827 | out: | |
dbd95212 | 828 | spin_unlock_irq(¤t->sighand->siglock); |
3b23dd12 KC |
829 | |
830 | return ret; | |
831 | } | |
832 | ||
e2cfabdf | 833 | #ifdef CONFIG_SECCOMP_FILTER |
3b23dd12 KC |
834 | /** |
835 | * seccomp_set_mode_filter: internal function for setting seccomp filter | |
48dc92b9 | 836 | * @flags: flags to change filter behavior |
3b23dd12 KC |
837 | * @filter: struct sock_fprog containing filter |
838 | * | |
839 | * This function may be called repeatedly to install additional filters. | |
840 | * Every filter successfully installed will be evaluated (in reverse order) | |
841 | * for each system call the task makes. | |
842 | * | |
843 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
844 | * | |
845 | * Returns 0 on success or -EINVAL on failure. | |
846 | */ | |
48dc92b9 KC |
847 | static long seccomp_set_mode_filter(unsigned int flags, |
848 | const char __user *filter) | |
3b23dd12 KC |
849 | { |
850 | const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; | |
c8bee430 | 851 | struct seccomp_filter *prepared = NULL; |
3b23dd12 KC |
852 | long ret = -EINVAL; |
853 | ||
48dc92b9 | 854 | /* Validate flags. */ |
c2e1f2e3 | 855 | if (flags & ~SECCOMP_FILTER_FLAG_MASK) |
dbd95212 | 856 | return -EINVAL; |
48dc92b9 | 857 | |
c8bee430 KC |
858 | /* Prepare the new filter before holding any locks. */ |
859 | prepared = seccomp_prepare_user_filter(filter); | |
860 | if (IS_ERR(prepared)) | |
861 | return PTR_ERR(prepared); | |
862 | ||
c2e1f2e3 KC |
863 | /* |
864 | * Make sure we cannot change seccomp or nnp state via TSYNC | |
865 | * while another thread is in the middle of calling exec. | |
866 | */ | |
867 | if (flags & SECCOMP_FILTER_FLAG_TSYNC && | |
868 | mutex_lock_killable(¤t->signal->cred_guard_mutex)) | |
869 | goto out_free; | |
870 | ||
dbd95212 KC |
871 | spin_lock_irq(¤t->sighand->siglock); |
872 | ||
3b23dd12 KC |
873 | if (!seccomp_may_assign_mode(seccomp_mode)) |
874 | goto out; | |
875 | ||
c8bee430 | 876 | ret = seccomp_attach_filter(flags, prepared); |
3b23dd12 | 877 | if (ret) |
e2cfabdf | 878 | goto out; |
c8bee430 KC |
879 | /* Do not free the successfully attached filter. */ |
880 | prepared = NULL; | |
1d9d02fe | 881 | |
3ba2530c | 882 | seccomp_assign_mode(current, seccomp_mode); |
e2cfabdf | 883 | out: |
dbd95212 | 884 | spin_unlock_irq(¤t->sighand->siglock); |
c2e1f2e3 KC |
885 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
886 | mutex_unlock(¤t->signal->cred_guard_mutex); | |
887 | out_free: | |
c8bee430 | 888 | seccomp_filter_free(prepared); |
1d9d02fe AA |
889 | return ret; |
890 | } | |
3b23dd12 | 891 | #else |
48dc92b9 KC |
892 | static inline long seccomp_set_mode_filter(unsigned int flags, |
893 | const char __user *filter) | |
3b23dd12 KC |
894 | { |
895 | return -EINVAL; | |
896 | } | |
897 | #endif | |
d78ab02c | 898 | |
4f65cecf TH |
899 | static long seccomp_get_action_avail(const char __user *uaction) |
900 | { | |
901 | u32 action; | |
902 | ||
903 | if (copy_from_user(&action, uaction, sizeof(action))) | |
904 | return -EFAULT; | |
905 | ||
906 | switch (action) { | |
907 | case SECCOMP_RET_KILL: | |
908 | case SECCOMP_RET_TRAP: | |
909 | case SECCOMP_RET_ERRNO: | |
910 | case SECCOMP_RET_TRACE: | |
4267083e | 911 | case SECCOMP_RET_LOG: |
4f65cecf TH |
912 | case SECCOMP_RET_ALLOW: |
913 | break; | |
914 | default: | |
915 | return -EOPNOTSUPP; | |
916 | } | |
917 | ||
918 | return 0; | |
919 | } | |
920 | ||
48dc92b9 KC |
921 | /* Common entry point for both prctl and syscall. */ |
922 | static long do_seccomp(unsigned int op, unsigned int flags, | |
923 | const char __user *uargs) | |
924 | { | |
925 | switch (op) { | |
926 | case SECCOMP_SET_MODE_STRICT: | |
927 | if (flags != 0 || uargs != NULL) | |
928 | return -EINVAL; | |
929 | return seccomp_set_mode_strict(); | |
930 | case SECCOMP_SET_MODE_FILTER: | |
931 | return seccomp_set_mode_filter(flags, uargs); | |
4f65cecf TH |
932 | case SECCOMP_GET_ACTION_AVAIL: |
933 | if (flags != 0) | |
934 | return -EINVAL; | |
935 | ||
936 | return seccomp_get_action_avail(uargs); | |
48dc92b9 KC |
937 | default: |
938 | return -EINVAL; | |
939 | } | |
940 | } | |
941 | ||
942 | SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, | |
943 | const char __user *, uargs) | |
944 | { | |
945 | return do_seccomp(op, flags, uargs); | |
946 | } | |
947 | ||
d78ab02c KC |
948 | /** |
949 | * prctl_set_seccomp: configures current->seccomp.mode | |
950 | * @seccomp_mode: requested mode to use | |
951 | * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER | |
952 | * | |
953 | * Returns 0 on success or -EINVAL on failure. | |
954 | */ | |
955 | long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter) | |
956 | { | |
48dc92b9 KC |
957 | unsigned int op; |
958 | char __user *uargs; | |
959 | ||
3b23dd12 KC |
960 | switch (seccomp_mode) { |
961 | case SECCOMP_MODE_STRICT: | |
48dc92b9 KC |
962 | op = SECCOMP_SET_MODE_STRICT; |
963 | /* | |
964 | * Setting strict mode through prctl always ignored filter, | |
965 | * so make sure it is always NULL here to pass the internal | |
966 | * check in do_seccomp(). | |
967 | */ | |
968 | uargs = NULL; | |
969 | break; | |
3b23dd12 | 970 | case SECCOMP_MODE_FILTER: |
48dc92b9 KC |
971 | op = SECCOMP_SET_MODE_FILTER; |
972 | uargs = filter; | |
973 | break; | |
3b23dd12 KC |
974 | default: |
975 | return -EINVAL; | |
976 | } | |
48dc92b9 KC |
977 | |
978 | /* prctl interface doesn't have flags, so they are always zero. */ | |
979 | return do_seccomp(op, 0, uargs); | |
d78ab02c | 980 | } |
f8e529ed TA |
981 | |
982 | #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE) | |
983 | long seccomp_get_filter(struct task_struct *task, unsigned long filter_off, | |
984 | void __user *data) | |
985 | { | |
986 | struct seccomp_filter *filter; | |
987 | struct sock_fprog_kern *fprog; | |
988 | long ret; | |
989 | unsigned long count = 0; | |
990 | ||
991 | if (!capable(CAP_SYS_ADMIN) || | |
992 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { | |
993 | return -EACCES; | |
994 | } | |
995 | ||
996 | spin_lock_irq(&task->sighand->siglock); | |
997 | if (task->seccomp.mode != SECCOMP_MODE_FILTER) { | |
998 | ret = -EINVAL; | |
999 | goto out; | |
1000 | } | |
1001 | ||
1002 | filter = task->seccomp.filter; | |
1003 | while (filter) { | |
1004 | filter = filter->prev; | |
1005 | count++; | |
1006 | } | |
1007 | ||
1008 | if (filter_off >= count) { | |
1009 | ret = -ENOENT; | |
1010 | goto out; | |
1011 | } | |
1012 | count -= filter_off; | |
1013 | ||
1014 | filter = task->seccomp.filter; | |
1015 | while (filter && count > 1) { | |
1016 | filter = filter->prev; | |
1017 | count--; | |
1018 | } | |
1019 | ||
1020 | if (WARN_ON(count != 1 || !filter)) { | |
1021 | /* The filter tree shouldn't shrink while we're using it. */ | |
1022 | ret = -ENOENT; | |
1023 | goto out; | |
1024 | } | |
1025 | ||
1026 | fprog = filter->prog->orig_prog; | |
1027 | if (!fprog) { | |
470bf1f2 | 1028 | /* This must be a new non-cBPF filter, since we save |
f8e529ed TA |
1029 | * every cBPF filter's orig_prog above when |
1030 | * CONFIG_CHECKPOINT_RESTORE is enabled. | |
1031 | */ | |
1032 | ret = -EMEDIUMTYPE; | |
1033 | goto out; | |
1034 | } | |
1035 | ||
1036 | ret = fprog->len; | |
1037 | if (!data) | |
1038 | goto out; | |
1039 | ||
1eaabc57 | 1040 | __get_seccomp_filter(filter); |
f8e529ed TA |
1041 | spin_unlock_irq(&task->sighand->siglock); |
1042 | ||
1043 | if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog))) | |
1044 | ret = -EFAULT; | |
1045 | ||
1eaabc57 | 1046 | __put_seccomp_filter(filter); |
f8e529ed TA |
1047 | return ret; |
1048 | ||
1049 | out: | |
1050 | spin_unlock_irq(&task->sighand->siglock); | |
1051 | return ret; | |
1052 | } | |
1053 | #endif | |
9ca58825 TH |
1054 | |
1055 | #ifdef CONFIG_SYSCTL | |
1056 | ||
1057 | /* Human readable action names for friendly sysctl interaction */ | |
1058 | #define SECCOMP_RET_KILL_NAME "kill" | |
1059 | #define SECCOMP_RET_TRAP_NAME "trap" | |
1060 | #define SECCOMP_RET_ERRNO_NAME "errno" | |
1061 | #define SECCOMP_RET_TRACE_NAME "trace" | |
4267083e | 1062 | #define SECCOMP_RET_LOG_NAME "log" |
9ca58825 TH |
1063 | #define SECCOMP_RET_ALLOW_NAME "allow" |
1064 | ||
1065 | static const char seccomp_actions_avail[] = SECCOMP_RET_KILL_NAME " " | |
1066 | SECCOMP_RET_TRAP_NAME " " | |
1067 | SECCOMP_RET_ERRNO_NAME " " | |
1068 | SECCOMP_RET_TRACE_NAME " " | |
4267083e | 1069 | SECCOMP_RET_LOG_NAME " " |
9ca58825 TH |
1070 | SECCOMP_RET_ALLOW_NAME; |
1071 | ||
56eb631b TH |
1072 | struct seccomp_log_name { |
1073 | u32 log; | |
1074 | const char *name; | |
1075 | }; | |
1076 | ||
1077 | static const struct seccomp_log_name seccomp_log_names[] = { | |
1078 | { SECCOMP_LOG_KILL, SECCOMP_RET_KILL_NAME }, | |
1079 | { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME }, | |
1080 | { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME }, | |
1081 | { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME }, | |
4267083e | 1082 | { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME }, |
56eb631b TH |
1083 | { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME }, |
1084 | { } | |
1085 | }; | |
1086 | ||
1087 | static bool seccomp_names_from_actions_logged(char *names, size_t size, | |
1088 | u32 actions_logged) | |
1089 | { | |
1090 | const struct seccomp_log_name *cur; | |
1091 | bool append_space = false; | |
1092 | ||
1093 | for (cur = seccomp_log_names; cur->name && size; cur++) { | |
1094 | ssize_t ret; | |
1095 | ||
1096 | if (!(actions_logged & cur->log)) | |
1097 | continue; | |
1098 | ||
1099 | if (append_space) { | |
1100 | ret = strscpy(names, " ", size); | |
1101 | if (ret < 0) | |
1102 | return false; | |
1103 | ||
1104 | names += ret; | |
1105 | size -= ret; | |
1106 | } else | |
1107 | append_space = true; | |
1108 | ||
1109 | ret = strscpy(names, cur->name, size); | |
1110 | if (ret < 0) | |
1111 | return false; | |
1112 | ||
1113 | names += ret; | |
1114 | size -= ret; | |
1115 | } | |
1116 | ||
1117 | return true; | |
1118 | } | |
1119 | ||
1120 | static bool seccomp_action_logged_from_name(u32 *action_logged, | |
1121 | const char *name) | |
1122 | { | |
1123 | const struct seccomp_log_name *cur; | |
1124 | ||
1125 | for (cur = seccomp_log_names; cur->name; cur++) { | |
1126 | if (!strcmp(cur->name, name)) { | |
1127 | *action_logged = cur->log; | |
1128 | return true; | |
1129 | } | |
1130 | } | |
1131 | ||
1132 | return false; | |
1133 | } | |
1134 | ||
1135 | static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names) | |
1136 | { | |
1137 | char *name; | |
1138 | ||
1139 | *actions_logged = 0; | |
1140 | while ((name = strsep(&names, " ")) && *name) { | |
1141 | u32 action_logged = 0; | |
1142 | ||
1143 | if (!seccomp_action_logged_from_name(&action_logged, name)) | |
1144 | return false; | |
1145 | ||
1146 | *actions_logged |= action_logged; | |
1147 | } | |
1148 | ||
1149 | return true; | |
1150 | } | |
1151 | ||
1152 | static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write, | |
1153 | void __user *buffer, size_t *lenp, | |
1154 | loff_t *ppos) | |
1155 | { | |
1156 | char names[sizeof(seccomp_actions_avail)]; | |
1157 | struct ctl_table table; | |
1158 | int ret; | |
1159 | ||
1160 | if (write && !capable(CAP_SYS_ADMIN)) | |
1161 | return -EPERM; | |
1162 | ||
1163 | memset(names, 0, sizeof(names)); | |
1164 | ||
1165 | if (!write) { | |
1166 | if (!seccomp_names_from_actions_logged(names, sizeof(names), | |
1167 | seccomp_actions_logged)) | |
1168 | return -EINVAL; | |
1169 | } | |
1170 | ||
1171 | table = *ro_table; | |
1172 | table.data = names; | |
1173 | table.maxlen = sizeof(names); | |
1174 | ret = proc_dostring(&table, write, buffer, lenp, ppos); | |
1175 | if (ret) | |
1176 | return ret; | |
1177 | ||
1178 | if (write) { | |
1179 | u32 actions_logged; | |
1180 | ||
1181 | if (!seccomp_actions_logged_from_names(&actions_logged, | |
1182 | table.data)) | |
1183 | return -EINVAL; | |
1184 | ||
1185 | if (actions_logged & SECCOMP_LOG_ALLOW) | |
1186 | return -EINVAL; | |
1187 | ||
1188 | seccomp_actions_logged = actions_logged; | |
1189 | } | |
1190 | ||
1191 | return 0; | |
1192 | } | |
1193 | ||
9ca58825 TH |
1194 | static struct ctl_path seccomp_sysctl_path[] = { |
1195 | { .procname = "kernel", }, | |
1196 | { .procname = "seccomp", }, | |
1197 | { } | |
1198 | }; | |
1199 | ||
1200 | static struct ctl_table seccomp_sysctl_table[] = { | |
1201 | { | |
1202 | .procname = "actions_avail", | |
1203 | .data = (void *) &seccomp_actions_avail, | |
1204 | .maxlen = sizeof(seccomp_actions_avail), | |
1205 | .mode = 0444, | |
1206 | .proc_handler = proc_dostring, | |
1207 | }, | |
56eb631b TH |
1208 | { |
1209 | .procname = "actions_logged", | |
1210 | .mode = 0644, | |
1211 | .proc_handler = seccomp_actions_logged_handler, | |
1212 | }, | |
9ca58825 TH |
1213 | { } |
1214 | }; | |
1215 | ||
1216 | static int __init seccomp_sysctl_init(void) | |
1217 | { | |
1218 | struct ctl_table_header *hdr; | |
1219 | ||
1220 | hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table); | |
1221 | if (!hdr) | |
1222 | pr_warn("seccomp: sysctl registration failed\n"); | |
1223 | else | |
1224 | kmemleak_not_leak(hdr); | |
1225 | ||
1226 | return 0; | |
1227 | } | |
1228 | ||
1229 | device_initcall(seccomp_sysctl_init) | |
1230 | ||
1231 | #endif /* CONFIG_SYSCTL */ |