]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/kernel/seccomp.c | |
4 | * | |
5 | * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com> | |
6 | * | |
e2cfabdf WD |
7 | * Copyright (C) 2012 Google, Inc. |
8 | * Will Drewry <wad@chromium.org> | |
9 | * | |
10 | * This defines a simple but solid secure-computing facility. | |
11 | * | |
12 | * Mode 1 uses a fixed list of allowed system calls. | |
13 | * Mode 2 allows user-defined system call filters in the form | |
14 | * of Berkeley Packet Filters/Linux Socket Filters. | |
1da177e4 LT |
15 | */ |
16 | ||
0b5fa229 | 17 | #include <linux/refcount.h> |
85e7bac3 | 18 | #include <linux/audit.h> |
5b101740 | 19 | #include <linux/compat.h> |
b25e6716 | 20 | #include <linux/coredump.h> |
8e5f1ad1 | 21 | #include <linux/kmemleak.h> |
5c307089 KC |
22 | #include <linux/nospec.h> |
23 | #include <linux/prctl.h> | |
e2cfabdf | 24 | #include <linux/sched.h> |
68db0cf1 | 25 | #include <linux/sched/task_stack.h> |
e2cfabdf | 26 | #include <linux/seccomp.h> |
c8bee430 | 27 | #include <linux/slab.h> |
48dc92b9 | 28 | #include <linux/syscalls.h> |
8e5f1ad1 | 29 | #include <linux/sysctl.h> |
1da177e4 | 30 | |
a4412fc9 | 31 | #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER |
e2cfabdf | 32 | #include <asm/syscall.h> |
a4412fc9 | 33 | #endif |
e2cfabdf WD |
34 | |
35 | #ifdef CONFIG_SECCOMP_FILTER | |
6a21cc50 | 36 | #include <linux/file.h> |
e2cfabdf | 37 | #include <linux/filter.h> |
c2e1f2e3 | 38 | #include <linux/pid.h> |
fb0fadf9 | 39 | #include <linux/ptrace.h> |
e2cfabdf | 40 | #include <linux/security.h> |
e2cfabdf WD |
41 | #include <linux/tracehook.h> |
42 | #include <linux/uaccess.h> | |
6a21cc50 TA |
43 | #include <linux/anon_inodes.h> |
44 | ||
45 | enum notify_state { | |
46 | SECCOMP_NOTIFY_INIT, | |
47 | SECCOMP_NOTIFY_SENT, | |
48 | SECCOMP_NOTIFY_REPLIED, | |
49 | }; | |
50 | ||
51 | struct seccomp_knotif { | |
52 | /* The struct pid of the task whose filter triggered the notification */ | |
53 | struct task_struct *task; | |
54 | ||
55 | /* The "cookie" for this request; this is unique for this filter. */ | |
56 | u64 id; | |
57 | ||
58 | /* | |
59 | * The seccomp data. This pointer is valid the entire time this | |
60 | * notification is active, since it comes from __seccomp_filter which | |
61 | * eclipses the entire lifecycle here. | |
62 | */ | |
63 | const struct seccomp_data *data; | |
64 | ||
65 | /* | |
66 | * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a | |
67 | * struct seccomp_knotif is created and starts out in INIT. Once the | |
68 | * handler reads the notification off of an FD, it transitions to SENT. | |
69 | * If a signal is received the state transitions back to INIT and | |
70 | * another message is sent. When the userspace handler replies, state | |
71 | * transitions to REPLIED. | |
72 | */ | |
73 | enum notify_state state; | |
74 | ||
75 | /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */ | |
76 | int error; | |
77 | long val; | |
fb3c5386 | 78 | u32 flags; |
6a21cc50 TA |
79 | |
80 | /* Signals when this has entered SECCOMP_NOTIFY_REPLIED */ | |
81 | struct completion ready; | |
82 | ||
83 | struct list_head list; | |
84 | }; | |
85 | ||
86 | /** | |
87 | * struct notification - container for seccomp userspace notifications. Since | |
88 | * most seccomp filters will not have notification listeners attached and this | |
89 | * structure is fairly large, we store the notification-specific stuff in a | |
90 | * separate structure. | |
91 | * | |
92 | * @request: A semaphore that users of this notification can wait on for | |
93 | * changes. Actual reads and writes are still controlled with | |
94 | * filter->notify_lock. | |
95 | * @next_id: The id of the next request. | |
96 | * @notifications: A list of struct seccomp_knotif elements. | |
97 | * @wqh: A wait queue for poll. | |
98 | */ | |
99 | struct notification { | |
100 | struct semaphore request; | |
101 | u64 next_id; | |
102 | struct list_head notifications; | |
103 | wait_queue_head_t wqh; | |
104 | }; | |
e2cfabdf WD |
105 | |
106 | /** | |
107 | * struct seccomp_filter - container for seccomp BPF programs | |
108 | * | |
109 | * @usage: reference count to manage the object lifetime. | |
110 | * get/put helpers should be used when accessing an instance | |
111 | * outside of a lifetime-guarded section. In general, this | |
112 | * is only needed for handling filters shared across tasks. | |
e66a3997 | 113 | * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged |
e2cfabdf | 114 | * @prev: points to a previously installed, or inherited, filter |
285fdfc5 | 115 | * @prog: the BPF program to evaluate |
6a21cc50 TA |
116 | * @notif: the struct that holds all notification related information |
117 | * @notify_lock: A lock for all notification-related accesses. | |
e2cfabdf WD |
118 | * |
119 | * seccomp_filter objects are organized in a tree linked via the @prev | |
120 | * pointer. For any task, it appears to be a singly-linked list starting | |
121 | * with current->seccomp.filter, the most recently attached or inherited filter. | |
122 | * However, multiple filters may share a @prev node, by way of fork(), which | |
123 | * results in a unidirectional tree existing in memory. This is similar to | |
124 | * how namespaces work. | |
125 | * | |
126 | * seccomp_filter objects should never be modified after being attached | |
127 | * to a task_struct (other than @usage). | |
128 | */ | |
129 | struct seccomp_filter { | |
0b5fa229 | 130 | refcount_t usage; |
e66a3997 | 131 | bool log; |
e2cfabdf | 132 | struct seccomp_filter *prev; |
7ae457c1 | 133 | struct bpf_prog *prog; |
6a21cc50 TA |
134 | struct notification *notif; |
135 | struct mutex notify_lock; | |
e2cfabdf WD |
136 | }; |
137 | ||
138 | /* Limit any path through the tree to 256KB worth of instructions. */ | |
139 | #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter)) | |
140 | ||
bd4cf0ed | 141 | /* |
e2cfabdf WD |
142 | * Endianness is explicitly ignored and left for BPF program authors to manage |
143 | * as per the specific architecture. | |
144 | */ | |
bd4cf0ed | 145 | static void populate_seccomp_data(struct seccomp_data *sd) |
e2cfabdf | 146 | { |
bd4cf0ed AS |
147 | struct task_struct *task = current; |
148 | struct pt_regs *regs = task_pt_regs(task); | |
2eac7648 | 149 | unsigned long args[6]; |
e2cfabdf | 150 | |
bd4cf0ed | 151 | sd->nr = syscall_get_nr(task, regs); |
16add411 | 152 | sd->arch = syscall_get_arch(task); |
b35f549d | 153 | syscall_get_arguments(task, regs, args); |
2eac7648 DB |
154 | sd->args[0] = args[0]; |
155 | sd->args[1] = args[1]; | |
156 | sd->args[2] = args[2]; | |
157 | sd->args[3] = args[3]; | |
158 | sd->args[4] = args[4]; | |
159 | sd->args[5] = args[5]; | |
bd4cf0ed | 160 | sd->instruction_pointer = KSTK_EIP(task); |
e2cfabdf WD |
161 | } |
162 | ||
163 | /** | |
164 | * seccomp_check_filter - verify seccomp filter code | |
165 | * @filter: filter to verify | |
166 | * @flen: length of filter | |
167 | * | |
4df95ff4 | 168 | * Takes a previously checked filter (by bpf_check_classic) and |
e2cfabdf WD |
169 | * redirects all filter code that loads struct sk_buff data |
170 | * and related data through seccomp_bpf_load. It also | |
171 | * enforces length and alignment checking of those loads. | |
172 | * | |
173 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
174 | */ | |
175 | static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen) | |
176 | { | |
177 | int pc; | |
178 | for (pc = 0; pc < flen; pc++) { | |
179 | struct sock_filter *ftest = &filter[pc]; | |
180 | u16 code = ftest->code; | |
181 | u32 k = ftest->k; | |
182 | ||
183 | switch (code) { | |
34805931 | 184 | case BPF_LD | BPF_W | BPF_ABS: |
bd4cf0ed | 185 | ftest->code = BPF_LDX | BPF_W | BPF_ABS; |
e2cfabdf WD |
186 | /* 32-bit aligned and not out of bounds. */ |
187 | if (k >= sizeof(struct seccomp_data) || k & 3) | |
188 | return -EINVAL; | |
189 | continue; | |
34805931 | 190 | case BPF_LD | BPF_W | BPF_LEN: |
bd4cf0ed | 191 | ftest->code = BPF_LD | BPF_IMM; |
e2cfabdf WD |
192 | ftest->k = sizeof(struct seccomp_data); |
193 | continue; | |
34805931 | 194 | case BPF_LDX | BPF_W | BPF_LEN: |
bd4cf0ed | 195 | ftest->code = BPF_LDX | BPF_IMM; |
e2cfabdf WD |
196 | ftest->k = sizeof(struct seccomp_data); |
197 | continue; | |
198 | /* Explicitly include allowed calls. */ | |
34805931 DB |
199 | case BPF_RET | BPF_K: |
200 | case BPF_RET | BPF_A: | |
201 | case BPF_ALU | BPF_ADD | BPF_K: | |
202 | case BPF_ALU | BPF_ADD | BPF_X: | |
203 | case BPF_ALU | BPF_SUB | BPF_K: | |
204 | case BPF_ALU | BPF_SUB | BPF_X: | |
205 | case BPF_ALU | BPF_MUL | BPF_K: | |
206 | case BPF_ALU | BPF_MUL | BPF_X: | |
207 | case BPF_ALU | BPF_DIV | BPF_K: | |
208 | case BPF_ALU | BPF_DIV | BPF_X: | |
209 | case BPF_ALU | BPF_AND | BPF_K: | |
210 | case BPF_ALU | BPF_AND | BPF_X: | |
211 | case BPF_ALU | BPF_OR | BPF_K: | |
212 | case BPF_ALU | BPF_OR | BPF_X: | |
213 | case BPF_ALU | BPF_XOR | BPF_K: | |
214 | case BPF_ALU | BPF_XOR | BPF_X: | |
215 | case BPF_ALU | BPF_LSH | BPF_K: | |
216 | case BPF_ALU | BPF_LSH | BPF_X: | |
217 | case BPF_ALU | BPF_RSH | BPF_K: | |
218 | case BPF_ALU | BPF_RSH | BPF_X: | |
219 | case BPF_ALU | BPF_NEG: | |
220 | case BPF_LD | BPF_IMM: | |
221 | case BPF_LDX | BPF_IMM: | |
222 | case BPF_MISC | BPF_TAX: | |
223 | case BPF_MISC | BPF_TXA: | |
224 | case BPF_LD | BPF_MEM: | |
225 | case BPF_LDX | BPF_MEM: | |
226 | case BPF_ST: | |
227 | case BPF_STX: | |
228 | case BPF_JMP | BPF_JA: | |
229 | case BPF_JMP | BPF_JEQ | BPF_K: | |
230 | case BPF_JMP | BPF_JEQ | BPF_X: | |
231 | case BPF_JMP | BPF_JGE | BPF_K: | |
232 | case BPF_JMP | BPF_JGE | BPF_X: | |
233 | case BPF_JMP | BPF_JGT | BPF_K: | |
234 | case BPF_JMP | BPF_JGT | BPF_X: | |
235 | case BPF_JMP | BPF_JSET | BPF_K: | |
236 | case BPF_JMP | BPF_JSET | BPF_X: | |
e2cfabdf WD |
237 | continue; |
238 | default: | |
239 | return -EINVAL; | |
240 | } | |
241 | } | |
242 | return 0; | |
243 | } | |
244 | ||
245 | /** | |
285fdfc5 MS |
246 | * seccomp_run_filters - evaluates all seccomp filters against @sd |
247 | * @sd: optional seccomp data to be passed to filters | |
deb4de8b KC |
248 | * @match: stores struct seccomp_filter that resulted in the return value, |
249 | * unless filter returned SECCOMP_RET_ALLOW, in which case it will | |
250 | * be unchanged. | |
e2cfabdf WD |
251 | * |
252 | * Returns valid seccomp BPF response codes. | |
253 | */ | |
0466bdb9 | 254 | #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL))) |
deb4de8b KC |
255 | static u32 seccomp_run_filters(const struct seccomp_data *sd, |
256 | struct seccomp_filter **match) | |
e2cfabdf | 257 | { |
acf3b2c7 | 258 | u32 ret = SECCOMP_RET_ALLOW; |
8225d385 PK |
259 | /* Make sure cross-thread synced filter points somewhere sane. */ |
260 | struct seccomp_filter *f = | |
506458ef | 261 | READ_ONCE(current->seccomp.filter); |
acf3b2c7 WD |
262 | |
263 | /* Ensure unexpected behavior doesn't result in failing open. */ | |
0d42d73a | 264 | if (WARN_ON(f == NULL)) |
4d3b0b05 | 265 | return SECCOMP_RET_KILL_PROCESS; |
acf3b2c7 | 266 | |
e2cfabdf WD |
267 | /* |
268 | * All filters in the list are evaluated and the lowest BPF return | |
acf3b2c7 | 269 | * value always takes priority (ignoring the DATA). |
e2cfabdf | 270 | */ |
3ba2530c | 271 | for (; f; f = f->prev) { |
3d9f773c | 272 | u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd); |
8f577cad | 273 | |
0466bdb9 | 274 | if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) { |
acf3b2c7 | 275 | ret = cur_ret; |
deb4de8b KC |
276 | *match = f; |
277 | } | |
e2cfabdf WD |
278 | } |
279 | return ret; | |
280 | } | |
1f41b450 | 281 | #endif /* CONFIG_SECCOMP_FILTER */ |
e2cfabdf | 282 | |
1f41b450 KC |
283 | static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) |
284 | { | |
69f6a34b | 285 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 286 | |
1f41b450 KC |
287 | if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) |
288 | return false; | |
289 | ||
290 | return true; | |
291 | } | |
292 | ||
8bf37d8c | 293 | void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { } |
5c307089 | 294 | |
3ba2530c | 295 | static inline void seccomp_assign_mode(struct task_struct *task, |
00a02d0c KC |
296 | unsigned long seccomp_mode, |
297 | unsigned long flags) | |
1f41b450 | 298 | { |
69f6a34b | 299 | assert_spin_locked(&task->sighand->siglock); |
dbd95212 | 300 | |
3ba2530c KC |
301 | task->seccomp.mode = seccomp_mode; |
302 | /* | |
303 | * Make sure TIF_SECCOMP cannot be set before the mode (and | |
304 | * filter) is set. | |
305 | */ | |
306 | smp_mb__before_atomic(); | |
00a02d0c KC |
307 | /* Assume default seccomp processes want spec flaw mitigation. */ |
308 | if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0) | |
8bf37d8c | 309 | arch_seccomp_spec_mitigate(task); |
3ba2530c | 310 | set_tsk_thread_flag(task, TIF_SECCOMP); |
1f41b450 KC |
311 | } |
312 | ||
313 | #ifdef CONFIG_SECCOMP_FILTER | |
c2e1f2e3 KC |
314 | /* Returns 1 if the parent is an ancestor of the child. */ |
315 | static int is_ancestor(struct seccomp_filter *parent, | |
316 | struct seccomp_filter *child) | |
317 | { | |
318 | /* NULL is the root ancestor. */ | |
319 | if (parent == NULL) | |
320 | return 1; | |
321 | for (; child; child = child->prev) | |
322 | if (child == parent) | |
323 | return 1; | |
324 | return 0; | |
325 | } | |
326 | ||
327 | /** | |
328 | * seccomp_can_sync_threads: checks if all threads can be synchronized | |
329 | * | |
330 | * Expects sighand and cred_guard_mutex locks to be held. | |
331 | * | |
332 | * Returns 0 on success, -ve on error, or the pid of a thread which was | |
6beff00b | 333 | * either not in the correct seccomp mode or did not have an ancestral |
c2e1f2e3 KC |
334 | * seccomp filter. |
335 | */ | |
336 | static inline pid_t seccomp_can_sync_threads(void) | |
337 | { | |
338 | struct task_struct *thread, *caller; | |
339 | ||
340 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 341 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
342 | |
343 | /* Validate all threads being eligible for synchronization. */ | |
344 | caller = current; | |
345 | for_each_thread(caller, thread) { | |
346 | pid_t failed; | |
347 | ||
348 | /* Skip current, since it is initiating the sync. */ | |
349 | if (thread == caller) | |
350 | continue; | |
351 | ||
352 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || | |
353 | (thread->seccomp.mode == SECCOMP_MODE_FILTER && | |
354 | is_ancestor(thread->seccomp.filter, | |
355 | caller->seccomp.filter))) | |
356 | continue; | |
357 | ||
358 | /* Return the first thread that cannot be synchronized. */ | |
359 | failed = task_pid_vnr(thread); | |
360 | /* If the pid cannot be resolved, then return -ESRCH */ | |
0d42d73a | 361 | if (WARN_ON(failed == 0)) |
c2e1f2e3 KC |
362 | failed = -ESRCH; |
363 | return failed; | |
364 | } | |
365 | ||
366 | return 0; | |
367 | } | |
368 | ||
369 | /** | |
370 | * seccomp_sync_threads: sets all threads to use current's filter | |
371 | * | |
372 | * Expects sighand and cred_guard_mutex locks to be held, and for | |
373 | * seccomp_can_sync_threads() to have returned success already | |
374 | * without dropping the locks. | |
375 | * | |
376 | */ | |
00a02d0c | 377 | static inline void seccomp_sync_threads(unsigned long flags) |
c2e1f2e3 KC |
378 | { |
379 | struct task_struct *thread, *caller; | |
380 | ||
381 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 382 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
383 | |
384 | /* Synchronize all threads. */ | |
385 | caller = current; | |
386 | for_each_thread(caller, thread) { | |
387 | /* Skip current, since it needs no changes. */ | |
388 | if (thread == caller) | |
389 | continue; | |
390 | ||
391 | /* Get a task reference for the new leaf node. */ | |
392 | get_seccomp_filter(caller); | |
393 | /* | |
394 | * Drop the task reference to the shared ancestor since | |
395 | * current's path will hold a reference. (This also | |
396 | * allows a put before the assignment.) | |
397 | */ | |
398 | put_seccomp_filter(thread); | |
399 | smp_store_release(&thread->seccomp.filter, | |
400 | caller->seccomp.filter); | |
103502a3 JH |
401 | |
402 | /* | |
403 | * Don't let an unprivileged task work around | |
404 | * the no_new_privs restriction by creating | |
405 | * a thread that sets it up, enters seccomp, | |
406 | * then dies. | |
407 | */ | |
408 | if (task_no_new_privs(caller)) | |
409 | task_set_no_new_privs(thread); | |
410 | ||
c2e1f2e3 KC |
411 | /* |
412 | * Opt the other thread into seccomp if needed. | |
413 | * As threads are considered to be trust-realm | |
414 | * equivalent (see ptrace_may_access), it is safe to | |
415 | * allow one thread to transition the other. | |
416 | */ | |
103502a3 | 417 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) |
00a02d0c KC |
418 | seccomp_assign_mode(thread, SECCOMP_MODE_FILTER, |
419 | flags); | |
c2e1f2e3 KC |
420 | } |
421 | } | |
422 | ||
e2cfabdf | 423 | /** |
c8bee430 | 424 | * seccomp_prepare_filter: Prepares a seccomp filter for use. |
e2cfabdf WD |
425 | * @fprog: BPF program to install |
426 | * | |
c8bee430 | 427 | * Returns filter on success or an ERR_PTR on failure. |
e2cfabdf | 428 | */ |
c8bee430 | 429 | static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog) |
e2cfabdf | 430 | { |
ac67eb2c DB |
431 | struct seccomp_filter *sfilter; |
432 | int ret; | |
97f2645f | 433 | const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE); |
e2cfabdf WD |
434 | |
435 | if (fprog->len == 0 || fprog->len > BPF_MAXINSNS) | |
c8bee430 | 436 | return ERR_PTR(-EINVAL); |
d9e12f42 | 437 | |
c8bee430 | 438 | BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter)); |
e2cfabdf WD |
439 | |
440 | /* | |
119ce5c8 | 441 | * Installing a seccomp filter requires that the task has |
e2cfabdf WD |
442 | * CAP_SYS_ADMIN in its namespace or be running with no_new_privs. |
443 | * This avoids scenarios where unprivileged tasks can affect the | |
444 | * behavior of privileged children. | |
445 | */ | |
1d4457f9 | 446 | if (!task_no_new_privs(current) && |
c1a85a00 MM |
447 | security_capable(current_cred(), current_user_ns(), |
448 | CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) != 0) | |
c8bee430 | 449 | return ERR_PTR(-EACCES); |
e2cfabdf | 450 | |
bd4cf0ed | 451 | /* Allocate a new seccomp_filter */ |
ac67eb2c DB |
452 | sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN); |
453 | if (!sfilter) | |
d9e12f42 | 454 | return ERR_PTR(-ENOMEM); |
ac67eb2c | 455 | |
6a21cc50 | 456 | mutex_init(&sfilter->notify_lock); |
ac67eb2c | 457 | ret = bpf_prog_create_from_user(&sfilter->prog, fprog, |
f8e529ed | 458 | seccomp_check_filter, save_orig); |
ac67eb2c DB |
459 | if (ret < 0) { |
460 | kfree(sfilter); | |
461 | return ERR_PTR(ret); | |
d9e12f42 | 462 | } |
bd4cf0ed | 463 | |
0b5fa229 | 464 | refcount_set(&sfilter->usage, 1); |
e2cfabdf | 465 | |
ac67eb2c | 466 | return sfilter; |
e2cfabdf WD |
467 | } |
468 | ||
469 | /** | |
c8bee430 | 470 | * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog |
e2cfabdf WD |
471 | * @user_filter: pointer to the user data containing a sock_fprog. |
472 | * | |
473 | * Returns 0 on success and non-zero otherwise. | |
474 | */ | |
c8bee430 KC |
475 | static struct seccomp_filter * |
476 | seccomp_prepare_user_filter(const char __user *user_filter) | |
e2cfabdf WD |
477 | { |
478 | struct sock_fprog fprog; | |
c8bee430 | 479 | struct seccomp_filter *filter = ERR_PTR(-EFAULT); |
e2cfabdf WD |
480 | |
481 | #ifdef CONFIG_COMPAT | |
5c38065e | 482 | if (in_compat_syscall()) { |
e2cfabdf WD |
483 | struct compat_sock_fprog fprog32; |
484 | if (copy_from_user(&fprog32, user_filter, sizeof(fprog32))) | |
485 | goto out; | |
486 | fprog.len = fprog32.len; | |
487 | fprog.filter = compat_ptr(fprog32.filter); | |
488 | } else /* falls through to the if below. */ | |
489 | #endif | |
490 | if (copy_from_user(&fprog, user_filter, sizeof(fprog))) | |
491 | goto out; | |
c8bee430 | 492 | filter = seccomp_prepare_filter(&fprog); |
e2cfabdf | 493 | out: |
c8bee430 KC |
494 | return filter; |
495 | } | |
496 | ||
497 | /** | |
498 | * seccomp_attach_filter: validate and attach filter | |
499 | * @flags: flags to change filter behavior | |
500 | * @filter: seccomp filter to add to the current process | |
501 | * | |
dbd95212 KC |
502 | * Caller must be holding current->sighand->siglock lock. |
503 | * | |
7a0df7fb TA |
504 | * Returns 0 on success, -ve on error, or |
505 | * - in TSYNC mode: the pid of a thread which was either not in the correct | |
506 | * seccomp mode or did not have an ancestral seccomp filter | |
507 | * - in NEW_LISTENER mode: the fd of the new listener | |
c8bee430 KC |
508 | */ |
509 | static long seccomp_attach_filter(unsigned int flags, | |
510 | struct seccomp_filter *filter) | |
511 | { | |
512 | unsigned long total_insns; | |
513 | struct seccomp_filter *walker; | |
514 | ||
69f6a34b | 515 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 516 | |
c8bee430 KC |
517 | /* Validate resulting filter length. */ |
518 | total_insns = filter->prog->len; | |
519 | for (walker = current->seccomp.filter; walker; walker = walker->prev) | |
520 | total_insns += walker->prog->len + 4; /* 4 instr penalty */ | |
521 | if (total_insns > MAX_INSNS_PER_PATH) | |
522 | return -ENOMEM; | |
523 | ||
c2e1f2e3 KC |
524 | /* If thread sync has been requested, check that it is possible. */ |
525 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) { | |
526 | int ret; | |
527 | ||
528 | ret = seccomp_can_sync_threads(); | |
529 | if (ret) | |
530 | return ret; | |
531 | } | |
532 | ||
e66a3997 TH |
533 | /* Set log flag, if present. */ |
534 | if (flags & SECCOMP_FILTER_FLAG_LOG) | |
535 | filter->log = true; | |
536 | ||
c8bee430 KC |
537 | /* |
538 | * If there is an existing filter, make it the prev and don't drop its | |
539 | * task reference. | |
540 | */ | |
541 | filter->prev = current->seccomp.filter; | |
542 | current->seccomp.filter = filter; | |
543 | ||
c2e1f2e3 KC |
544 | /* Now that the new filter is in place, synchronize to all threads. */ |
545 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) | |
00a02d0c | 546 | seccomp_sync_threads(flags); |
c2e1f2e3 | 547 | |
c8bee430 | 548 | return 0; |
e2cfabdf WD |
549 | } |
550 | ||
084f5601 | 551 | static void __get_seccomp_filter(struct seccomp_filter *filter) |
66a733ea | 552 | { |
66a733ea ON |
553 | refcount_inc(&filter->usage); |
554 | } | |
555 | ||
e2cfabdf WD |
556 | /* get_seccomp_filter - increments the reference count of the filter on @tsk */ |
557 | void get_seccomp_filter(struct task_struct *tsk) | |
558 | { | |
559 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
560 | if (!orig) | |
561 | return; | |
66a733ea | 562 | __get_seccomp_filter(orig); |
e2cfabdf WD |
563 | } |
564 | ||
c8bee430 KC |
565 | static inline void seccomp_filter_free(struct seccomp_filter *filter) |
566 | { | |
567 | if (filter) { | |
bab18991 | 568 | bpf_prog_destroy(filter->prog); |
c8bee430 KC |
569 | kfree(filter); |
570 | } | |
571 | } | |
572 | ||
66a733ea | 573 | static void __put_seccomp_filter(struct seccomp_filter *orig) |
e2cfabdf | 574 | { |
e2cfabdf | 575 | /* Clean up single-reference branches iteratively. */ |
0b5fa229 | 576 | while (orig && refcount_dec_and_test(&orig->usage)) { |
e2cfabdf WD |
577 | struct seccomp_filter *freeme = orig; |
578 | orig = orig->prev; | |
c8bee430 | 579 | seccomp_filter_free(freeme); |
e2cfabdf WD |
580 | } |
581 | } | |
bb6ea430 | 582 | |
66a733ea ON |
583 | /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */ |
584 | void put_seccomp_filter(struct task_struct *tsk) | |
585 | { | |
586 | __put_seccomp_filter(tsk->seccomp.filter); | |
587 | } | |
588 | ||
ae7795bc | 589 | static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason) |
b25e6716 | 590 | { |
3b10db2b | 591 | clear_siginfo(info); |
b25e6716 MF |
592 | info->si_signo = SIGSYS; |
593 | info->si_code = SYS_SECCOMP; | |
594 | info->si_call_addr = (void __user *)KSTK_EIP(current); | |
595 | info->si_errno = reason; | |
16add411 | 596 | info->si_arch = syscall_get_arch(current); |
b25e6716 MF |
597 | info->si_syscall = syscall; |
598 | } | |
599 | ||
bb6ea430 WD |
600 | /** |
601 | * seccomp_send_sigsys - signals the task to allow in-process syscall emulation | |
602 | * @syscall: syscall number to send to userland | |
603 | * @reason: filter-supplied reason code to send to userland (via si_errno) | |
604 | * | |
605 | * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info. | |
606 | */ | |
607 | static void seccomp_send_sigsys(int syscall, int reason) | |
608 | { | |
ae7795bc | 609 | struct kernel_siginfo info; |
b25e6716 | 610 | seccomp_init_siginfo(&info, syscall, reason); |
a89e9b8a | 611 | force_sig_info(&info); |
bb6ea430 | 612 | } |
e2cfabdf | 613 | #endif /* CONFIG_SECCOMP_FILTER */ |
1da177e4 | 614 | |
0ddec0fc | 615 | /* For use with seccomp_actions_logged */ |
4d3b0b05 KC |
616 | #define SECCOMP_LOG_KILL_PROCESS (1 << 0) |
617 | #define SECCOMP_LOG_KILL_THREAD (1 << 1) | |
0ddec0fc TH |
618 | #define SECCOMP_LOG_TRAP (1 << 2) |
619 | #define SECCOMP_LOG_ERRNO (1 << 3) | |
620 | #define SECCOMP_LOG_TRACE (1 << 4) | |
59f5cf44 TH |
621 | #define SECCOMP_LOG_LOG (1 << 5) |
622 | #define SECCOMP_LOG_ALLOW (1 << 6) | |
6a21cc50 | 623 | #define SECCOMP_LOG_USER_NOTIF (1 << 7) |
0ddec0fc | 624 | |
4d3b0b05 KC |
625 | static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS | |
626 | SECCOMP_LOG_KILL_THREAD | | |
fd76875c KC |
627 | SECCOMP_LOG_TRAP | |
628 | SECCOMP_LOG_ERRNO | | |
6a21cc50 | 629 | SECCOMP_LOG_USER_NOTIF | |
fd76875c | 630 | SECCOMP_LOG_TRACE | |
59f5cf44 | 631 | SECCOMP_LOG_LOG; |
0ddec0fc | 632 | |
e66a3997 TH |
633 | static inline void seccomp_log(unsigned long syscall, long signr, u32 action, |
634 | bool requested) | |
0ddec0fc TH |
635 | { |
636 | bool log = false; | |
637 | ||
638 | switch (action) { | |
639 | case SECCOMP_RET_ALLOW: | |
e66a3997 | 640 | break; |
0ddec0fc | 641 | case SECCOMP_RET_TRAP: |
e66a3997 TH |
642 | log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP; |
643 | break; | |
0ddec0fc | 644 | case SECCOMP_RET_ERRNO: |
e66a3997 TH |
645 | log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO; |
646 | break; | |
0ddec0fc | 647 | case SECCOMP_RET_TRACE: |
e66a3997 | 648 | log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE; |
0ddec0fc | 649 | break; |
6a21cc50 TA |
650 | case SECCOMP_RET_USER_NOTIF: |
651 | log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF; | |
652 | break; | |
59f5cf44 TH |
653 | case SECCOMP_RET_LOG: |
654 | log = seccomp_actions_logged & SECCOMP_LOG_LOG; | |
655 | break; | |
fd76875c | 656 | case SECCOMP_RET_KILL_THREAD: |
fd76875c | 657 | log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD; |
4d3b0b05 KC |
658 | break; |
659 | case SECCOMP_RET_KILL_PROCESS: | |
660 | default: | |
661 | log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS; | |
0ddec0fc TH |
662 | } |
663 | ||
664 | /* | |
326bee02 TH |
665 | * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the |
666 | * FILTER_FLAG_LOG bit was set. The admin has the ability to silence | |
667 | * any action from being logged by removing the action name from the | |
668 | * seccomp_actions_logged sysctl. | |
0ddec0fc | 669 | */ |
326bee02 TH |
670 | if (!log) |
671 | return; | |
0ddec0fc | 672 | |
326bee02 | 673 | audit_seccomp(syscall, signr, action); |
0ddec0fc TH |
674 | } |
675 | ||
1da177e4 LT |
676 | /* |
677 | * Secure computing mode 1 allows only read/write/exit/sigreturn. | |
678 | * To be fully secure this must be combined with rlimit | |
679 | * to limit the stack allocations too. | |
680 | */ | |
cb4253aa | 681 | static const int mode1_syscalls[] = { |
1da177e4 LT |
682 | __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, |
683 | 0, /* null terminated */ | |
684 | }; | |
685 | ||
a4412fc9 | 686 | static void __secure_computing_strict(int this_syscall) |
1da177e4 | 687 | { |
cb4253aa | 688 | const int *syscall_whitelist = mode1_syscalls; |
a4412fc9 | 689 | #ifdef CONFIG_COMPAT |
5c38065e | 690 | if (in_compat_syscall()) |
c983f0e8 | 691 | syscall_whitelist = get_compat_mode1_syscalls(); |
a4412fc9 AL |
692 | #endif |
693 | do { | |
694 | if (*syscall_whitelist == this_syscall) | |
695 | return; | |
696 | } while (*++syscall_whitelist); | |
697 | ||
698 | #ifdef SECCOMP_DEBUG | |
699 | dump_stack(); | |
700 | #endif | |
fd76875c | 701 | seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true); |
a4412fc9 AL |
702 | do_exit(SIGKILL); |
703 | } | |
704 | ||
705 | #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER | |
706 | void secure_computing_strict(int this_syscall) | |
707 | { | |
708 | int mode = current->seccomp.mode; | |
709 | ||
97f2645f | 710 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 TA |
711 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
712 | return; | |
713 | ||
221272f9 | 714 | if (mode == SECCOMP_MODE_DISABLED) |
a4412fc9 AL |
715 | return; |
716 | else if (mode == SECCOMP_MODE_STRICT) | |
717 | __secure_computing_strict(this_syscall); | |
718 | else | |
719 | BUG(); | |
720 | } | |
721 | #else | |
13aa72f0 AL |
722 | |
723 | #ifdef CONFIG_SECCOMP_FILTER | |
6a21cc50 TA |
724 | static u64 seccomp_next_notify_id(struct seccomp_filter *filter) |
725 | { | |
726 | /* | |
727 | * Note: overflow is ok here, the id just needs to be unique per | |
728 | * filter. | |
729 | */ | |
730 | lockdep_assert_held(&filter->notify_lock); | |
731 | return filter->notif->next_id++; | |
732 | } | |
733 | ||
fb3c5386 CB |
734 | static int seccomp_do_user_notification(int this_syscall, |
735 | struct seccomp_filter *match, | |
736 | const struct seccomp_data *sd) | |
6a21cc50 TA |
737 | { |
738 | int err; | |
fb3c5386 | 739 | u32 flags = 0; |
6a21cc50 TA |
740 | long ret = 0; |
741 | struct seccomp_knotif n = {}; | |
742 | ||
743 | mutex_lock(&match->notify_lock); | |
744 | err = -ENOSYS; | |
745 | if (!match->notif) | |
746 | goto out; | |
747 | ||
748 | n.task = current; | |
749 | n.state = SECCOMP_NOTIFY_INIT; | |
750 | n.data = sd; | |
751 | n.id = seccomp_next_notify_id(match); | |
752 | init_completion(&n.ready); | |
753 | list_add(&n.list, &match->notif->notifications); | |
754 | ||
755 | up(&match->notif->request); | |
756 | wake_up_poll(&match->notif->wqh, EPOLLIN | EPOLLRDNORM); | |
757 | mutex_unlock(&match->notify_lock); | |
758 | ||
759 | /* | |
760 | * This is where we wait for a reply from userspace. | |
761 | */ | |
762 | err = wait_for_completion_interruptible(&n.ready); | |
763 | mutex_lock(&match->notify_lock); | |
764 | if (err == 0) { | |
765 | ret = n.val; | |
766 | err = n.error; | |
fb3c5386 | 767 | flags = n.flags; |
6a21cc50 TA |
768 | } |
769 | ||
770 | /* | |
771 | * Note that it's possible the listener died in between the time when | |
772 | * we were notified of a respons (or a signal) and when we were able to | |
773 | * re-acquire the lock, so only delete from the list if the | |
774 | * notification actually exists. | |
775 | * | |
776 | * Also note that this test is only valid because there's no way to | |
777 | * *reattach* to a notifier right now. If one is added, we'll need to | |
778 | * keep track of the notif itself and make sure they match here. | |
779 | */ | |
780 | if (match->notif) | |
781 | list_del(&n.list); | |
782 | out: | |
783 | mutex_unlock(&match->notify_lock); | |
fb3c5386 CB |
784 | |
785 | /* Userspace requests to continue the syscall. */ | |
786 | if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) | |
787 | return 0; | |
788 | ||
6a21cc50 TA |
789 | syscall_set_return_value(current, task_pt_regs(current), |
790 | err, ret); | |
fb3c5386 | 791 | return -1; |
6a21cc50 TA |
792 | } |
793 | ||
ce6526e8 KC |
794 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
795 | const bool recheck_after_trace) | |
13aa72f0 AL |
796 | { |
797 | u32 filter_ret, action; | |
deb4de8b | 798 | struct seccomp_filter *match = NULL; |
13aa72f0 | 799 | int data; |
db511391 | 800 | struct seccomp_data sd_local; |
1da177e4 | 801 | |
3ba2530c KC |
802 | /* |
803 | * Make sure that any changes to mode from another thread have | |
804 | * been seen after TIF_SECCOMP was seen. | |
805 | */ | |
806 | rmb(); | |
807 | ||
db511391 TA |
808 | if (!sd) { |
809 | populate_seccomp_data(&sd_local); | |
810 | sd = &sd_local; | |
811 | } | |
812 | ||
deb4de8b | 813 | filter_ret = seccomp_run_filters(sd, &match); |
13aa72f0 | 814 | data = filter_ret & SECCOMP_RET_DATA; |
0466bdb9 | 815 | action = filter_ret & SECCOMP_RET_ACTION_FULL; |
13aa72f0 AL |
816 | |
817 | switch (action) { | |
818 | case SECCOMP_RET_ERRNO: | |
580c57f1 KC |
819 | /* Set low-order bits as an errno, capped at MAX_ERRNO. */ |
820 | if (data > MAX_ERRNO) | |
821 | data = MAX_ERRNO; | |
d39bd00d | 822 | syscall_set_return_value(current, task_pt_regs(current), |
13aa72f0 AL |
823 | -data, 0); |
824 | goto skip; | |
825 | ||
826 | case SECCOMP_RET_TRAP: | |
827 | /* Show the handler the original registers. */ | |
d39bd00d | 828 | syscall_rollback(current, task_pt_regs(current)); |
13aa72f0 AL |
829 | /* Let the filter pass back 16 bits of data. */ |
830 | seccomp_send_sigsys(this_syscall, data); | |
831 | goto skip; | |
832 | ||
833 | case SECCOMP_RET_TRACE: | |
ce6526e8 KC |
834 | /* We've been put in this state by the ptracer already. */ |
835 | if (recheck_after_trace) | |
836 | return 0; | |
837 | ||
8112c4f1 KC |
838 | /* ENOSYS these calls if there is no tracer attached. */ |
839 | if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { | |
840 | syscall_set_return_value(current, | |
841 | task_pt_regs(current), | |
842 | -ENOSYS, 0); | |
843 | goto skip; | |
844 | } | |
845 | ||
846 | /* Allow the BPF to provide the event message */ | |
847 | ptrace_event(PTRACE_EVENT_SECCOMP, data); | |
848 | /* | |
849 | * The delivery of a fatal signal during event | |
485a252a KC |
850 | * notification may silently skip tracer notification, |
851 | * which could leave us with a potentially unmodified | |
852 | * syscall that the tracer would have liked to have | |
853 | * changed. Since the process is about to die, we just | |
854 | * force the syscall to be skipped and let the signal | |
855 | * kill the process and correctly handle any tracer exit | |
856 | * notifications. | |
8112c4f1 KC |
857 | */ |
858 | if (fatal_signal_pending(current)) | |
485a252a | 859 | goto skip; |
8112c4f1 KC |
860 | /* Check if the tracer forced the syscall to be skipped. */ |
861 | this_syscall = syscall_get_nr(current, task_pt_regs(current)); | |
862 | if (this_syscall < 0) | |
863 | goto skip; | |
864 | ||
ce6526e8 KC |
865 | /* |
866 | * Recheck the syscall, since it may have changed. This | |
867 | * intentionally uses a NULL struct seccomp_data to force | |
868 | * a reload of all registers. This does not goto skip since | |
869 | * a skip would have already been reported. | |
870 | */ | |
871 | if (__seccomp_filter(this_syscall, NULL, true)) | |
872 | return -1; | |
873 | ||
8112c4f1 | 874 | return 0; |
13aa72f0 | 875 | |
6a21cc50 | 876 | case SECCOMP_RET_USER_NOTIF: |
fb3c5386 CB |
877 | if (seccomp_do_user_notification(this_syscall, match, sd)) |
878 | goto skip; | |
879 | ||
880 | return 0; | |
6a21cc50 | 881 | |
59f5cf44 TH |
882 | case SECCOMP_RET_LOG: |
883 | seccomp_log(this_syscall, 0, action, true); | |
884 | return 0; | |
885 | ||
13aa72f0 | 886 | case SECCOMP_RET_ALLOW: |
deb4de8b KC |
887 | /* |
888 | * Note that the "match" filter will always be NULL for | |
889 | * this action since SECCOMP_RET_ALLOW is the starting | |
890 | * state in seccomp_run_filters(). | |
891 | */ | |
8112c4f1 | 892 | return 0; |
13aa72f0 | 893 | |
fd76875c | 894 | case SECCOMP_RET_KILL_THREAD: |
4d3b0b05 | 895 | case SECCOMP_RET_KILL_PROCESS: |
131b6351 | 896 | default: |
e66a3997 | 897 | seccomp_log(this_syscall, SIGSYS, action, true); |
d7276e32 | 898 | /* Dump core only if this is the last remaining thread. */ |
4d3b0b05 KC |
899 | if (action == SECCOMP_RET_KILL_PROCESS || |
900 | get_nr_threads(current) == 1) { | |
ae7795bc | 901 | kernel_siginfo_t info; |
131b6351 | 902 | |
d7276e32 KC |
903 | /* Show the original registers in the dump. */ |
904 | syscall_rollback(current, task_pt_regs(current)); | |
905 | /* Trigger a manual coredump since do_exit skips it. */ | |
906 | seccomp_init_siginfo(&info, this_syscall, data); | |
907 | do_coredump(&info); | |
908 | } | |
4d3b0b05 KC |
909 | if (action == SECCOMP_RET_KILL_PROCESS) |
910 | do_group_exit(SIGSYS); | |
911 | else | |
912 | do_exit(SIGSYS); | |
13aa72f0 AL |
913 | } |
914 | ||
915 | unreachable(); | |
916 | ||
917 | skip: | |
e66a3997 | 918 | seccomp_log(this_syscall, 0, action, match ? match->log : false); |
8112c4f1 KC |
919 | return -1; |
920 | } | |
921 | #else | |
ce6526e8 KC |
922 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
923 | const bool recheck_after_trace) | |
8112c4f1 KC |
924 | { |
925 | BUG(); | |
13aa72f0 | 926 | } |
1da177e4 | 927 | #endif |
13aa72f0 | 928 | |
8112c4f1 | 929 | int __secure_computing(const struct seccomp_data *sd) |
13aa72f0 AL |
930 | { |
931 | int mode = current->seccomp.mode; | |
8112c4f1 | 932 | int this_syscall; |
13aa72f0 | 933 | |
97f2645f | 934 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 | 935 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
8112c4f1 KC |
936 | return 0; |
937 | ||
938 | this_syscall = sd ? sd->nr : | |
939 | syscall_get_nr(current, task_pt_regs(current)); | |
13c4a901 | 940 | |
13aa72f0 | 941 | switch (mode) { |
e2cfabdf | 942 | case SECCOMP_MODE_STRICT: |
13aa72f0 | 943 | __secure_computing_strict(this_syscall); /* may call do_exit */ |
8112c4f1 | 944 | return 0; |
13aa72f0 | 945 | case SECCOMP_MODE_FILTER: |
ce6526e8 | 946 | return __seccomp_filter(this_syscall, sd, false); |
1da177e4 LT |
947 | default: |
948 | BUG(); | |
949 | } | |
13aa72f0 | 950 | } |
a4412fc9 | 951 | #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ |
1d9d02fe AA |
952 | |
953 | long prctl_get_seccomp(void) | |
954 | { | |
955 | return current->seccomp.mode; | |
956 | } | |
957 | ||
e2cfabdf | 958 | /** |
3b23dd12 | 959 | * seccomp_set_mode_strict: internal function for setting strict seccomp |
e2cfabdf WD |
960 | * |
961 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
962 | * | |
963 | * Returns 0 on success or -EINVAL on failure. | |
964 | */ | |
3b23dd12 | 965 | static long seccomp_set_mode_strict(void) |
1d9d02fe | 966 | { |
3b23dd12 | 967 | const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; |
e2cfabdf | 968 | long ret = -EINVAL; |
1d9d02fe | 969 | |
dbd95212 KC |
970 | spin_lock_irq(¤t->sighand->siglock); |
971 | ||
1f41b450 | 972 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1d9d02fe AA |
973 | goto out; |
974 | ||
cf99abac | 975 | #ifdef TIF_NOTSC |
3b23dd12 | 976 | disable_TSC(); |
cf99abac | 977 | #endif |
00a02d0c | 978 | seccomp_assign_mode(current, seccomp_mode, 0); |
3b23dd12 KC |
979 | ret = 0; |
980 | ||
981 | out: | |
dbd95212 | 982 | spin_unlock_irq(¤t->sighand->siglock); |
3b23dd12 KC |
983 | |
984 | return ret; | |
985 | } | |
986 | ||
e2cfabdf | 987 | #ifdef CONFIG_SECCOMP_FILTER |
6a21cc50 TA |
988 | static int seccomp_notify_release(struct inode *inode, struct file *file) |
989 | { | |
990 | struct seccomp_filter *filter = file->private_data; | |
991 | struct seccomp_knotif *knotif; | |
992 | ||
a811dc61 TA |
993 | if (!filter) |
994 | return 0; | |
995 | ||
6a21cc50 TA |
996 | mutex_lock(&filter->notify_lock); |
997 | ||
998 | /* | |
999 | * If this file is being closed because e.g. the task who owned it | |
1000 | * died, let's wake everyone up who was waiting on us. | |
1001 | */ | |
1002 | list_for_each_entry(knotif, &filter->notif->notifications, list) { | |
1003 | if (knotif->state == SECCOMP_NOTIFY_REPLIED) | |
1004 | continue; | |
1005 | ||
1006 | knotif->state = SECCOMP_NOTIFY_REPLIED; | |
1007 | knotif->error = -ENOSYS; | |
1008 | knotif->val = 0; | |
1009 | ||
1010 | complete(&knotif->ready); | |
1011 | } | |
1012 | ||
1013 | kfree(filter->notif); | |
1014 | filter->notif = NULL; | |
1015 | mutex_unlock(&filter->notify_lock); | |
1016 | __put_seccomp_filter(filter); | |
1017 | return 0; | |
1018 | } | |
1019 | ||
1020 | static long seccomp_notify_recv(struct seccomp_filter *filter, | |
1021 | void __user *buf) | |
1022 | { | |
1023 | struct seccomp_knotif *knotif = NULL, *cur; | |
1024 | struct seccomp_notif unotif; | |
1025 | ssize_t ret; | |
1026 | ||
2882d53c SD |
1027 | /* Verify that we're not given garbage to keep struct extensible. */ |
1028 | ret = check_zeroed_user(buf, sizeof(unotif)); | |
1029 | if (ret < 0) | |
1030 | return ret; | |
1031 | if (!ret) | |
1032 | return -EINVAL; | |
1033 | ||
6a21cc50 TA |
1034 | memset(&unotif, 0, sizeof(unotif)); |
1035 | ||
1036 | ret = down_interruptible(&filter->notif->request); | |
1037 | if (ret < 0) | |
1038 | return ret; | |
1039 | ||
1040 | mutex_lock(&filter->notify_lock); | |
1041 | list_for_each_entry(cur, &filter->notif->notifications, list) { | |
1042 | if (cur->state == SECCOMP_NOTIFY_INIT) { | |
1043 | knotif = cur; | |
1044 | break; | |
1045 | } | |
1046 | } | |
1047 | ||
1048 | /* | |
1049 | * If we didn't find a notification, it could be that the task was | |
1050 | * interrupted by a fatal signal between the time we were woken and | |
1051 | * when we were able to acquire the rw lock. | |
1052 | */ | |
1053 | if (!knotif) { | |
1054 | ret = -ENOENT; | |
1055 | goto out; | |
1056 | } | |
1057 | ||
1058 | unotif.id = knotif->id; | |
1059 | unotif.pid = task_pid_vnr(knotif->task); | |
1060 | unotif.data = *(knotif->data); | |
1061 | ||
1062 | knotif->state = SECCOMP_NOTIFY_SENT; | |
1063 | wake_up_poll(&filter->notif->wqh, EPOLLOUT | EPOLLWRNORM); | |
1064 | ret = 0; | |
1065 | out: | |
1066 | mutex_unlock(&filter->notify_lock); | |
1067 | ||
1068 | if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) { | |
1069 | ret = -EFAULT; | |
1070 | ||
1071 | /* | |
1072 | * Userspace screwed up. To make sure that we keep this | |
1073 | * notification alive, let's reset it back to INIT. It | |
1074 | * may have died when we released the lock, so we need to make | |
1075 | * sure it's still around. | |
1076 | */ | |
1077 | knotif = NULL; | |
1078 | mutex_lock(&filter->notify_lock); | |
1079 | list_for_each_entry(cur, &filter->notif->notifications, list) { | |
1080 | if (cur->id == unotif.id) { | |
1081 | knotif = cur; | |
1082 | break; | |
1083 | } | |
1084 | } | |
1085 | ||
1086 | if (knotif) { | |
1087 | knotif->state = SECCOMP_NOTIFY_INIT; | |
1088 | up(&filter->notif->request); | |
1089 | } | |
1090 | mutex_unlock(&filter->notify_lock); | |
1091 | } | |
1092 | ||
1093 | return ret; | |
1094 | } | |
1095 | ||
1096 | static long seccomp_notify_send(struct seccomp_filter *filter, | |
1097 | void __user *buf) | |
1098 | { | |
1099 | struct seccomp_notif_resp resp = {}; | |
1100 | struct seccomp_knotif *knotif = NULL, *cur; | |
1101 | long ret; | |
1102 | ||
1103 | if (copy_from_user(&resp, buf, sizeof(resp))) | |
1104 | return -EFAULT; | |
1105 | ||
fb3c5386 CB |
1106 | if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE) |
1107 | return -EINVAL; | |
1108 | ||
1109 | if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) && | |
1110 | (resp.error || resp.val)) | |
6a21cc50 TA |
1111 | return -EINVAL; |
1112 | ||
1113 | ret = mutex_lock_interruptible(&filter->notify_lock); | |
1114 | if (ret < 0) | |
1115 | return ret; | |
1116 | ||
1117 | list_for_each_entry(cur, &filter->notif->notifications, list) { | |
1118 | if (cur->id == resp.id) { | |
1119 | knotif = cur; | |
1120 | break; | |
1121 | } | |
1122 | } | |
1123 | ||
1124 | if (!knotif) { | |
1125 | ret = -ENOENT; | |
1126 | goto out; | |
1127 | } | |
1128 | ||
1129 | /* Allow exactly one reply. */ | |
1130 | if (knotif->state != SECCOMP_NOTIFY_SENT) { | |
1131 | ret = -EINPROGRESS; | |
1132 | goto out; | |
1133 | } | |
1134 | ||
1135 | ret = 0; | |
1136 | knotif->state = SECCOMP_NOTIFY_REPLIED; | |
1137 | knotif->error = resp.error; | |
1138 | knotif->val = resp.val; | |
fb3c5386 | 1139 | knotif->flags = resp.flags; |
6a21cc50 TA |
1140 | complete(&knotif->ready); |
1141 | out: | |
1142 | mutex_unlock(&filter->notify_lock); | |
1143 | return ret; | |
1144 | } | |
1145 | ||
1146 | static long seccomp_notify_id_valid(struct seccomp_filter *filter, | |
1147 | void __user *buf) | |
1148 | { | |
1149 | struct seccomp_knotif *knotif = NULL; | |
1150 | u64 id; | |
1151 | long ret; | |
1152 | ||
1153 | if (copy_from_user(&id, buf, sizeof(id))) | |
1154 | return -EFAULT; | |
1155 | ||
1156 | ret = mutex_lock_interruptible(&filter->notify_lock); | |
1157 | if (ret < 0) | |
1158 | return ret; | |
1159 | ||
1160 | ret = -ENOENT; | |
1161 | list_for_each_entry(knotif, &filter->notif->notifications, list) { | |
1162 | if (knotif->id == id) { | |
1163 | if (knotif->state == SECCOMP_NOTIFY_SENT) | |
1164 | ret = 0; | |
1165 | goto out; | |
1166 | } | |
1167 | } | |
1168 | ||
1169 | out: | |
1170 | mutex_unlock(&filter->notify_lock); | |
1171 | return ret; | |
1172 | } | |
1173 | ||
1174 | static long seccomp_notify_ioctl(struct file *file, unsigned int cmd, | |
1175 | unsigned long arg) | |
1176 | { | |
1177 | struct seccomp_filter *filter = file->private_data; | |
1178 | void __user *buf = (void __user *)arg; | |
1179 | ||
1180 | switch (cmd) { | |
1181 | case SECCOMP_IOCTL_NOTIF_RECV: | |
1182 | return seccomp_notify_recv(filter, buf); | |
1183 | case SECCOMP_IOCTL_NOTIF_SEND: | |
1184 | return seccomp_notify_send(filter, buf); | |
1185 | case SECCOMP_IOCTL_NOTIF_ID_VALID: | |
1186 | return seccomp_notify_id_valid(filter, buf); | |
1187 | default: | |
1188 | return -EINVAL; | |
1189 | } | |
1190 | } | |
1191 | ||
1192 | static __poll_t seccomp_notify_poll(struct file *file, | |
1193 | struct poll_table_struct *poll_tab) | |
1194 | { | |
1195 | struct seccomp_filter *filter = file->private_data; | |
1196 | __poll_t ret = 0; | |
1197 | struct seccomp_knotif *cur; | |
1198 | ||
1199 | poll_wait(file, &filter->notif->wqh, poll_tab); | |
1200 | ||
319deec7 | 1201 | if (mutex_lock_interruptible(&filter->notify_lock) < 0) |
6a21cc50 TA |
1202 | return EPOLLERR; |
1203 | ||
1204 | list_for_each_entry(cur, &filter->notif->notifications, list) { | |
1205 | if (cur->state == SECCOMP_NOTIFY_INIT) | |
1206 | ret |= EPOLLIN | EPOLLRDNORM; | |
1207 | if (cur->state == SECCOMP_NOTIFY_SENT) | |
1208 | ret |= EPOLLOUT | EPOLLWRNORM; | |
1209 | if ((ret & EPOLLIN) && (ret & EPOLLOUT)) | |
1210 | break; | |
1211 | } | |
1212 | ||
1213 | mutex_unlock(&filter->notify_lock); | |
1214 | ||
1215 | return ret; | |
1216 | } | |
1217 | ||
1218 | static const struct file_operations seccomp_notify_ops = { | |
1219 | .poll = seccomp_notify_poll, | |
1220 | .release = seccomp_notify_release, | |
1221 | .unlocked_ioctl = seccomp_notify_ioctl, | |
1222 | }; | |
1223 | ||
1224 | static struct file *init_listener(struct seccomp_filter *filter) | |
1225 | { | |
1226 | struct file *ret = ERR_PTR(-EBUSY); | |
1227 | struct seccomp_filter *cur; | |
1228 | ||
1229 | for (cur = current->seccomp.filter; cur; cur = cur->prev) { | |
1230 | if (cur->notif) | |
1231 | goto out; | |
1232 | } | |
1233 | ||
1234 | ret = ERR_PTR(-ENOMEM); | |
1235 | filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL); | |
1236 | if (!filter->notif) | |
1237 | goto out; | |
1238 | ||
1239 | sema_init(&filter->notif->request, 0); | |
1240 | filter->notif->next_id = get_random_u64(); | |
1241 | INIT_LIST_HEAD(&filter->notif->notifications); | |
1242 | init_waitqueue_head(&filter->notif->wqh); | |
1243 | ||
1244 | ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops, | |
1245 | filter, O_RDWR); | |
1246 | if (IS_ERR(ret)) | |
1247 | goto out_notif; | |
1248 | ||
1249 | /* The file has a reference to it now */ | |
1250 | __get_seccomp_filter(filter); | |
1251 | ||
1252 | out_notif: | |
1253 | if (IS_ERR(ret)) | |
1254 | kfree(filter->notif); | |
1255 | out: | |
1256 | return ret; | |
1257 | } | |
1258 | ||
3b23dd12 KC |
1259 | /** |
1260 | * seccomp_set_mode_filter: internal function for setting seccomp filter | |
48dc92b9 | 1261 | * @flags: flags to change filter behavior |
3b23dd12 KC |
1262 | * @filter: struct sock_fprog containing filter |
1263 | * | |
1264 | * This function may be called repeatedly to install additional filters. | |
1265 | * Every filter successfully installed will be evaluated (in reverse order) | |
1266 | * for each system call the task makes. | |
1267 | * | |
1268 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
1269 | * | |
1270 | * Returns 0 on success or -EINVAL on failure. | |
1271 | */ | |
48dc92b9 KC |
1272 | static long seccomp_set_mode_filter(unsigned int flags, |
1273 | const char __user *filter) | |
3b23dd12 KC |
1274 | { |
1275 | const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; | |
c8bee430 | 1276 | struct seccomp_filter *prepared = NULL; |
3b23dd12 | 1277 | long ret = -EINVAL; |
6a21cc50 TA |
1278 | int listener = -1; |
1279 | struct file *listener_f = NULL; | |
3b23dd12 | 1280 | |
48dc92b9 | 1281 | /* Validate flags. */ |
c2e1f2e3 | 1282 | if (flags & ~SECCOMP_FILTER_FLAG_MASK) |
dbd95212 | 1283 | return -EINVAL; |
48dc92b9 | 1284 | |
7a0df7fb TA |
1285 | /* |
1286 | * In the successful case, NEW_LISTENER returns the new listener fd. | |
1287 | * But in the failure case, TSYNC returns the thread that died. If you | |
1288 | * combine these two flags, there's no way to tell whether something | |
1289 | * succeeded or failed. So, let's disallow this combination. | |
1290 | */ | |
1291 | if ((flags & SECCOMP_FILTER_FLAG_TSYNC) && | |
1292 | (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER)) | |
1293 | return -EINVAL; | |
1294 | ||
c8bee430 KC |
1295 | /* Prepare the new filter before holding any locks. */ |
1296 | prepared = seccomp_prepare_user_filter(filter); | |
1297 | if (IS_ERR(prepared)) | |
1298 | return PTR_ERR(prepared); | |
1299 | ||
6a21cc50 TA |
1300 | if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) { |
1301 | listener = get_unused_fd_flags(O_CLOEXEC); | |
1302 | if (listener < 0) { | |
1303 | ret = listener; | |
1304 | goto out_free; | |
1305 | } | |
1306 | ||
1307 | listener_f = init_listener(prepared); | |
1308 | if (IS_ERR(listener_f)) { | |
1309 | put_unused_fd(listener); | |
1310 | ret = PTR_ERR(listener_f); | |
1311 | goto out_free; | |
1312 | } | |
1313 | } | |
1314 | ||
c2e1f2e3 KC |
1315 | /* |
1316 | * Make sure we cannot change seccomp or nnp state via TSYNC | |
1317 | * while another thread is in the middle of calling exec. | |
1318 | */ | |
1319 | if (flags & SECCOMP_FILTER_FLAG_TSYNC && | |
1320 | mutex_lock_killable(¤t->signal->cred_guard_mutex)) | |
6a21cc50 | 1321 | goto out_put_fd; |
c2e1f2e3 | 1322 | |
dbd95212 KC |
1323 | spin_lock_irq(¤t->sighand->siglock); |
1324 | ||
3b23dd12 KC |
1325 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1326 | goto out; | |
1327 | ||
c8bee430 | 1328 | ret = seccomp_attach_filter(flags, prepared); |
3b23dd12 | 1329 | if (ret) |
e2cfabdf | 1330 | goto out; |
c8bee430 KC |
1331 | /* Do not free the successfully attached filter. */ |
1332 | prepared = NULL; | |
1d9d02fe | 1333 | |
00a02d0c | 1334 | seccomp_assign_mode(current, seccomp_mode, flags); |
e2cfabdf | 1335 | out: |
dbd95212 | 1336 | spin_unlock_irq(¤t->sighand->siglock); |
c2e1f2e3 KC |
1337 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
1338 | mutex_unlock(¤t->signal->cred_guard_mutex); | |
6a21cc50 TA |
1339 | out_put_fd: |
1340 | if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) { | |
7a0df7fb | 1341 | if (ret) { |
a811dc61 | 1342 | listener_f->private_data = NULL; |
6a21cc50 TA |
1343 | fput(listener_f); |
1344 | put_unused_fd(listener); | |
1345 | } else { | |
1346 | fd_install(listener, listener_f); | |
1347 | ret = listener; | |
1348 | } | |
1349 | } | |
c2e1f2e3 | 1350 | out_free: |
c8bee430 | 1351 | seccomp_filter_free(prepared); |
1d9d02fe AA |
1352 | return ret; |
1353 | } | |
3b23dd12 | 1354 | #else |
48dc92b9 KC |
1355 | static inline long seccomp_set_mode_filter(unsigned int flags, |
1356 | const char __user *filter) | |
3b23dd12 KC |
1357 | { |
1358 | return -EINVAL; | |
1359 | } | |
1360 | #endif | |
d78ab02c | 1361 | |
d612b1fd TH |
1362 | static long seccomp_get_action_avail(const char __user *uaction) |
1363 | { | |
1364 | u32 action; | |
1365 | ||
1366 | if (copy_from_user(&action, uaction, sizeof(action))) | |
1367 | return -EFAULT; | |
1368 | ||
1369 | switch (action) { | |
0466bdb9 | 1370 | case SECCOMP_RET_KILL_PROCESS: |
fd76875c | 1371 | case SECCOMP_RET_KILL_THREAD: |
d612b1fd TH |
1372 | case SECCOMP_RET_TRAP: |
1373 | case SECCOMP_RET_ERRNO: | |
6a21cc50 | 1374 | case SECCOMP_RET_USER_NOTIF: |
d612b1fd | 1375 | case SECCOMP_RET_TRACE: |
59f5cf44 | 1376 | case SECCOMP_RET_LOG: |
d612b1fd TH |
1377 | case SECCOMP_RET_ALLOW: |
1378 | break; | |
1379 | default: | |
1380 | return -EOPNOTSUPP; | |
1381 | } | |
1382 | ||
1383 | return 0; | |
1384 | } | |
1385 | ||
6a21cc50 TA |
1386 | static long seccomp_get_notif_sizes(void __user *usizes) |
1387 | { | |
1388 | struct seccomp_notif_sizes sizes = { | |
1389 | .seccomp_notif = sizeof(struct seccomp_notif), | |
1390 | .seccomp_notif_resp = sizeof(struct seccomp_notif_resp), | |
1391 | .seccomp_data = sizeof(struct seccomp_data), | |
1392 | }; | |
1393 | ||
1394 | if (copy_to_user(usizes, &sizes, sizeof(sizes))) | |
1395 | return -EFAULT; | |
1396 | ||
1397 | return 0; | |
1398 | } | |
1399 | ||
48dc92b9 KC |
1400 | /* Common entry point for both prctl and syscall. */ |
1401 | static long do_seccomp(unsigned int op, unsigned int flags, | |
a5662e4d | 1402 | void __user *uargs) |
48dc92b9 KC |
1403 | { |
1404 | switch (op) { | |
1405 | case SECCOMP_SET_MODE_STRICT: | |
1406 | if (flags != 0 || uargs != NULL) | |
1407 | return -EINVAL; | |
1408 | return seccomp_set_mode_strict(); | |
1409 | case SECCOMP_SET_MODE_FILTER: | |
1410 | return seccomp_set_mode_filter(flags, uargs); | |
d612b1fd TH |
1411 | case SECCOMP_GET_ACTION_AVAIL: |
1412 | if (flags != 0) | |
1413 | return -EINVAL; | |
1414 | ||
1415 | return seccomp_get_action_avail(uargs); | |
6a21cc50 TA |
1416 | case SECCOMP_GET_NOTIF_SIZES: |
1417 | if (flags != 0) | |
1418 | return -EINVAL; | |
1419 | ||
1420 | return seccomp_get_notif_sizes(uargs); | |
48dc92b9 KC |
1421 | default: |
1422 | return -EINVAL; | |
1423 | } | |
1424 | } | |
1425 | ||
1426 | SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, | |
a5662e4d | 1427 | void __user *, uargs) |
48dc92b9 KC |
1428 | { |
1429 | return do_seccomp(op, flags, uargs); | |
1430 | } | |
1431 | ||
d78ab02c KC |
1432 | /** |
1433 | * prctl_set_seccomp: configures current->seccomp.mode | |
1434 | * @seccomp_mode: requested mode to use | |
1435 | * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER | |
1436 | * | |
1437 | * Returns 0 on success or -EINVAL on failure. | |
1438 | */ | |
a5662e4d | 1439 | long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter) |
d78ab02c | 1440 | { |
48dc92b9 | 1441 | unsigned int op; |
a5662e4d | 1442 | void __user *uargs; |
48dc92b9 | 1443 | |
3b23dd12 KC |
1444 | switch (seccomp_mode) { |
1445 | case SECCOMP_MODE_STRICT: | |
48dc92b9 KC |
1446 | op = SECCOMP_SET_MODE_STRICT; |
1447 | /* | |
1448 | * Setting strict mode through prctl always ignored filter, | |
1449 | * so make sure it is always NULL here to pass the internal | |
1450 | * check in do_seccomp(). | |
1451 | */ | |
1452 | uargs = NULL; | |
1453 | break; | |
3b23dd12 | 1454 | case SECCOMP_MODE_FILTER: |
48dc92b9 KC |
1455 | op = SECCOMP_SET_MODE_FILTER; |
1456 | uargs = filter; | |
1457 | break; | |
3b23dd12 KC |
1458 | default: |
1459 | return -EINVAL; | |
1460 | } | |
48dc92b9 KC |
1461 | |
1462 | /* prctl interface doesn't have flags, so they are always zero. */ | |
1463 | return do_seccomp(op, 0, uargs); | |
d78ab02c | 1464 | } |
f8e529ed TA |
1465 | |
1466 | #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE) | |
f06eae83 TA |
1467 | static struct seccomp_filter *get_nth_filter(struct task_struct *task, |
1468 | unsigned long filter_off) | |
f8e529ed | 1469 | { |
f06eae83 TA |
1470 | struct seccomp_filter *orig, *filter; |
1471 | unsigned long count; | |
f8e529ed | 1472 | |
f06eae83 TA |
1473 | /* |
1474 | * Note: this is only correct because the caller should be the (ptrace) | |
1475 | * tracer of the task, otherwise lock_task_sighand is needed. | |
1476 | */ | |
f8e529ed | 1477 | spin_lock_irq(&task->sighand->siglock); |
f06eae83 | 1478 | |
f8e529ed | 1479 | if (task->seccomp.mode != SECCOMP_MODE_FILTER) { |
f06eae83 TA |
1480 | spin_unlock_irq(&task->sighand->siglock); |
1481 | return ERR_PTR(-EINVAL); | |
f8e529ed TA |
1482 | } |
1483 | ||
f06eae83 TA |
1484 | orig = task->seccomp.filter; |
1485 | __get_seccomp_filter(orig); | |
1486 | spin_unlock_irq(&task->sighand->siglock); | |
1487 | ||
1488 | count = 0; | |
1489 | for (filter = orig; filter; filter = filter->prev) | |
f8e529ed | 1490 | count++; |
f8e529ed TA |
1491 | |
1492 | if (filter_off >= count) { | |
f06eae83 | 1493 | filter = ERR_PTR(-ENOENT); |
f8e529ed TA |
1494 | goto out; |
1495 | } | |
f8e529ed | 1496 | |
f06eae83 TA |
1497 | count -= filter_off; |
1498 | for (filter = orig; filter && count > 1; filter = filter->prev) | |
f8e529ed | 1499 | count--; |
f8e529ed TA |
1500 | |
1501 | if (WARN_ON(count != 1 || !filter)) { | |
f06eae83 | 1502 | filter = ERR_PTR(-ENOENT); |
f8e529ed TA |
1503 | goto out; |
1504 | } | |
1505 | ||
f06eae83 TA |
1506 | __get_seccomp_filter(filter); |
1507 | ||
1508 | out: | |
1509 | __put_seccomp_filter(orig); | |
1510 | return filter; | |
1511 | } | |
1512 | ||
1513 | long seccomp_get_filter(struct task_struct *task, unsigned long filter_off, | |
1514 | void __user *data) | |
1515 | { | |
1516 | struct seccomp_filter *filter; | |
1517 | struct sock_fprog_kern *fprog; | |
1518 | long ret; | |
1519 | ||
1520 | if (!capable(CAP_SYS_ADMIN) || | |
1521 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { | |
1522 | return -EACCES; | |
1523 | } | |
1524 | ||
1525 | filter = get_nth_filter(task, filter_off); | |
1526 | if (IS_ERR(filter)) | |
1527 | return PTR_ERR(filter); | |
1528 | ||
f8e529ed TA |
1529 | fprog = filter->prog->orig_prog; |
1530 | if (!fprog) { | |
470bf1f2 | 1531 | /* This must be a new non-cBPF filter, since we save |
f8e529ed TA |
1532 | * every cBPF filter's orig_prog above when |
1533 | * CONFIG_CHECKPOINT_RESTORE is enabled. | |
1534 | */ | |
1535 | ret = -EMEDIUMTYPE; | |
1536 | goto out; | |
1537 | } | |
1538 | ||
1539 | ret = fprog->len; | |
1540 | if (!data) | |
1541 | goto out; | |
1542 | ||
f8e529ed TA |
1543 | if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog))) |
1544 | ret = -EFAULT; | |
1545 | ||
f8e529ed | 1546 | out: |
66a733ea | 1547 | __put_seccomp_filter(filter); |
f8e529ed | 1548 | return ret; |
f8e529ed | 1549 | } |
f8e529ed | 1550 | |
26500475 TA |
1551 | long seccomp_get_metadata(struct task_struct *task, |
1552 | unsigned long size, void __user *data) | |
1553 | { | |
1554 | long ret; | |
1555 | struct seccomp_filter *filter; | |
1556 | struct seccomp_metadata kmd = {}; | |
1557 | ||
1558 | if (!capable(CAP_SYS_ADMIN) || | |
1559 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { | |
1560 | return -EACCES; | |
1561 | } | |
1562 | ||
1563 | size = min_t(unsigned long, size, sizeof(kmd)); | |
1564 | ||
63bb0045 TA |
1565 | if (size < sizeof(kmd.filter_off)) |
1566 | return -EINVAL; | |
1567 | ||
1568 | if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off))) | |
26500475 TA |
1569 | return -EFAULT; |
1570 | ||
1571 | filter = get_nth_filter(task, kmd.filter_off); | |
1572 | if (IS_ERR(filter)) | |
1573 | return PTR_ERR(filter); | |
1574 | ||
26500475 TA |
1575 | if (filter->log) |
1576 | kmd.flags |= SECCOMP_FILTER_FLAG_LOG; | |
1577 | ||
1578 | ret = size; | |
1579 | if (copy_to_user(data, &kmd, size)) | |
1580 | ret = -EFAULT; | |
1581 | ||
1582 | __put_seccomp_filter(filter); | |
f8e529ed TA |
1583 | return ret; |
1584 | } | |
1585 | #endif | |
8e5f1ad1 TH |
1586 | |
1587 | #ifdef CONFIG_SYSCTL | |
1588 | ||
1589 | /* Human readable action names for friendly sysctl interaction */ | |
0466bdb9 | 1590 | #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process" |
fd76875c | 1591 | #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread" |
8e5f1ad1 TH |
1592 | #define SECCOMP_RET_TRAP_NAME "trap" |
1593 | #define SECCOMP_RET_ERRNO_NAME "errno" | |
6a21cc50 | 1594 | #define SECCOMP_RET_USER_NOTIF_NAME "user_notif" |
8e5f1ad1 | 1595 | #define SECCOMP_RET_TRACE_NAME "trace" |
59f5cf44 | 1596 | #define SECCOMP_RET_LOG_NAME "log" |
8e5f1ad1 TH |
1597 | #define SECCOMP_RET_ALLOW_NAME "allow" |
1598 | ||
fd76875c | 1599 | static const char seccomp_actions_avail[] = |
0466bdb9 | 1600 | SECCOMP_RET_KILL_PROCESS_NAME " " |
fd76875c KC |
1601 | SECCOMP_RET_KILL_THREAD_NAME " " |
1602 | SECCOMP_RET_TRAP_NAME " " | |
1603 | SECCOMP_RET_ERRNO_NAME " " | |
6a21cc50 | 1604 | SECCOMP_RET_USER_NOTIF_NAME " " |
fd76875c KC |
1605 | SECCOMP_RET_TRACE_NAME " " |
1606 | SECCOMP_RET_LOG_NAME " " | |
1607 | SECCOMP_RET_ALLOW_NAME; | |
8e5f1ad1 | 1608 | |
0ddec0fc TH |
1609 | struct seccomp_log_name { |
1610 | u32 log; | |
1611 | const char *name; | |
1612 | }; | |
1613 | ||
1614 | static const struct seccomp_log_name seccomp_log_names[] = { | |
0466bdb9 | 1615 | { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME }, |
fd76875c | 1616 | { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME }, |
0ddec0fc TH |
1617 | { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME }, |
1618 | { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME }, | |
6a21cc50 | 1619 | { SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME }, |
0ddec0fc | 1620 | { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME }, |
59f5cf44 | 1621 | { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME }, |
0ddec0fc TH |
1622 | { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME }, |
1623 | { } | |
1624 | }; | |
1625 | ||
1626 | static bool seccomp_names_from_actions_logged(char *names, size_t size, | |
beb44aca TH |
1627 | u32 actions_logged, |
1628 | const char *sep) | |
0ddec0fc TH |
1629 | { |
1630 | const struct seccomp_log_name *cur; | |
beb44aca | 1631 | bool append_sep = false; |
0ddec0fc TH |
1632 | |
1633 | for (cur = seccomp_log_names; cur->name && size; cur++) { | |
1634 | ssize_t ret; | |
1635 | ||
1636 | if (!(actions_logged & cur->log)) | |
1637 | continue; | |
1638 | ||
beb44aca TH |
1639 | if (append_sep) { |
1640 | ret = strscpy(names, sep, size); | |
0ddec0fc TH |
1641 | if (ret < 0) |
1642 | return false; | |
1643 | ||
1644 | names += ret; | |
1645 | size -= ret; | |
1646 | } else | |
beb44aca | 1647 | append_sep = true; |
0ddec0fc TH |
1648 | |
1649 | ret = strscpy(names, cur->name, size); | |
1650 | if (ret < 0) | |
1651 | return false; | |
1652 | ||
1653 | names += ret; | |
1654 | size -= ret; | |
1655 | } | |
1656 | ||
1657 | return true; | |
1658 | } | |
1659 | ||
1660 | static bool seccomp_action_logged_from_name(u32 *action_logged, | |
1661 | const char *name) | |
1662 | { | |
1663 | const struct seccomp_log_name *cur; | |
1664 | ||
1665 | for (cur = seccomp_log_names; cur->name; cur++) { | |
1666 | if (!strcmp(cur->name, name)) { | |
1667 | *action_logged = cur->log; | |
1668 | return true; | |
1669 | } | |
1670 | } | |
1671 | ||
1672 | return false; | |
1673 | } | |
1674 | ||
1675 | static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names) | |
1676 | { | |
1677 | char *name; | |
1678 | ||
1679 | *actions_logged = 0; | |
1680 | while ((name = strsep(&names, " ")) && *name) { | |
1681 | u32 action_logged = 0; | |
1682 | ||
1683 | if (!seccomp_action_logged_from_name(&action_logged, name)) | |
1684 | return false; | |
1685 | ||
1686 | *actions_logged |= action_logged; | |
1687 | } | |
1688 | ||
1689 | return true; | |
1690 | } | |
1691 | ||
d013db02 TH |
1692 | static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer, |
1693 | size_t *lenp, loff_t *ppos) | |
0ddec0fc TH |
1694 | { |
1695 | char names[sizeof(seccomp_actions_avail)]; | |
1696 | struct ctl_table table; | |
d013db02 TH |
1697 | |
1698 | memset(names, 0, sizeof(names)); | |
1699 | ||
1700 | if (!seccomp_names_from_actions_logged(names, sizeof(names), | |
beb44aca | 1701 | seccomp_actions_logged, " ")) |
d013db02 TH |
1702 | return -EINVAL; |
1703 | ||
1704 | table = *ro_table; | |
1705 | table.data = names; | |
1706 | table.maxlen = sizeof(names); | |
1707 | return proc_dostring(&table, 0, buffer, lenp, ppos); | |
1708 | } | |
1709 | ||
1710 | static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer, | |
ea6eca77 | 1711 | size_t *lenp, loff_t *ppos, u32 *actions_logged) |
0ddec0fc TH |
1712 | { |
1713 | char names[sizeof(seccomp_actions_avail)]; | |
1714 | struct ctl_table table; | |
1715 | int ret; | |
1716 | ||
d013db02 | 1717 | if (!capable(CAP_SYS_ADMIN)) |
0ddec0fc TH |
1718 | return -EPERM; |
1719 | ||
1720 | memset(names, 0, sizeof(names)); | |
1721 | ||
0ddec0fc TH |
1722 | table = *ro_table; |
1723 | table.data = names; | |
1724 | table.maxlen = sizeof(names); | |
d013db02 | 1725 | ret = proc_dostring(&table, 1, buffer, lenp, ppos); |
0ddec0fc TH |
1726 | if (ret) |
1727 | return ret; | |
1728 | ||
ea6eca77 | 1729 | if (!seccomp_actions_logged_from_names(actions_logged, table.data)) |
d013db02 | 1730 | return -EINVAL; |
0ddec0fc | 1731 | |
ea6eca77 | 1732 | if (*actions_logged & SECCOMP_LOG_ALLOW) |
d013db02 | 1733 | return -EINVAL; |
0ddec0fc | 1734 | |
ea6eca77 | 1735 | seccomp_actions_logged = *actions_logged; |
0ddec0fc TH |
1736 | return 0; |
1737 | } | |
0ddec0fc | 1738 | |
ea6eca77 TH |
1739 | static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged, |
1740 | int ret) | |
1741 | { | |
1742 | char names[sizeof(seccomp_actions_avail)]; | |
1743 | char old_names[sizeof(seccomp_actions_avail)]; | |
1744 | const char *new = names; | |
1745 | const char *old = old_names; | |
0ddec0fc | 1746 | |
ea6eca77 TH |
1747 | if (!audit_enabled) |
1748 | return; | |
1749 | ||
1750 | memset(names, 0, sizeof(names)); | |
1751 | memset(old_names, 0, sizeof(old_names)); | |
1752 | ||
1753 | if (ret) | |
1754 | new = "?"; | |
1755 | else if (!actions_logged) | |
1756 | new = "(none)"; | |
1757 | else if (!seccomp_names_from_actions_logged(names, sizeof(names), | |
1758 | actions_logged, ",")) | |
1759 | new = "?"; | |
1760 | ||
1761 | if (!old_actions_logged) | |
1762 | old = "(none)"; | |
1763 | else if (!seccomp_names_from_actions_logged(old_names, | |
1764 | sizeof(old_names), | |
1765 | old_actions_logged, ",")) | |
1766 | old = "?"; | |
1767 | ||
1768 | return audit_seccomp_actions_logged(new, old, !ret); | |
1769 | } | |
1770 | ||
d013db02 TH |
1771 | static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write, |
1772 | void __user *buffer, size_t *lenp, | |
1773 | loff_t *ppos) | |
1774 | { | |
ea6eca77 TH |
1775 | int ret; |
1776 | ||
1777 | if (write) { | |
1778 | u32 actions_logged = 0; | |
1779 | u32 old_actions_logged = seccomp_actions_logged; | |
1780 | ||
1781 | ret = write_actions_logged(ro_table, buffer, lenp, ppos, | |
1782 | &actions_logged); | |
1783 | audit_actions_logged(actions_logged, old_actions_logged, ret); | |
1784 | } else | |
1785 | ret = read_actions_logged(ro_table, buffer, lenp, ppos); | |
1786 | ||
1787 | return ret; | |
0ddec0fc TH |
1788 | } |
1789 | ||
8e5f1ad1 TH |
1790 | static struct ctl_path seccomp_sysctl_path[] = { |
1791 | { .procname = "kernel", }, | |
1792 | { .procname = "seccomp", }, | |
1793 | { } | |
1794 | }; | |
1795 | ||
1796 | static struct ctl_table seccomp_sysctl_table[] = { | |
1797 | { | |
1798 | .procname = "actions_avail", | |
1799 | .data = (void *) &seccomp_actions_avail, | |
1800 | .maxlen = sizeof(seccomp_actions_avail), | |
1801 | .mode = 0444, | |
1802 | .proc_handler = proc_dostring, | |
1803 | }, | |
0ddec0fc TH |
1804 | { |
1805 | .procname = "actions_logged", | |
1806 | .mode = 0644, | |
1807 | .proc_handler = seccomp_actions_logged_handler, | |
1808 | }, | |
8e5f1ad1 TH |
1809 | { } |
1810 | }; | |
1811 | ||
1812 | static int __init seccomp_sysctl_init(void) | |
1813 | { | |
1814 | struct ctl_table_header *hdr; | |
1815 | ||
1816 | hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table); | |
1817 | if (!hdr) | |
1818 | pr_warn("seccomp: sysctl registration failed\n"); | |
1819 | else | |
1820 | kmemleak_not_leak(hdr); | |
1821 | ||
1822 | return 0; | |
1823 | } | |
1824 | ||
1825 | device_initcall(seccomp_sysctl_init) | |
1826 | ||
1827 | #endif /* CONFIG_SYSCTL */ |