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 | 43 | #include <linux/anon_inodes.h> |
9f87dcf1 | 44 | #include <linux/lockdep.h> |
6a21cc50 TA |
45 | |
46 | enum notify_state { | |
47 | SECCOMP_NOTIFY_INIT, | |
48 | SECCOMP_NOTIFY_SENT, | |
49 | SECCOMP_NOTIFY_REPLIED, | |
50 | }; | |
51 | ||
52 | struct seccomp_knotif { | |
53 | /* The struct pid of the task whose filter triggered the notification */ | |
54 | struct task_struct *task; | |
55 | ||
56 | /* The "cookie" for this request; this is unique for this filter. */ | |
57 | u64 id; | |
58 | ||
59 | /* | |
60 | * The seccomp data. This pointer is valid the entire time this | |
61 | * notification is active, since it comes from __seccomp_filter which | |
62 | * eclipses the entire lifecycle here. | |
63 | */ | |
64 | const struct seccomp_data *data; | |
65 | ||
66 | /* | |
67 | * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a | |
68 | * struct seccomp_knotif is created and starts out in INIT. Once the | |
69 | * handler reads the notification off of an FD, it transitions to SENT. | |
70 | * If a signal is received the state transitions back to INIT and | |
71 | * another message is sent. When the userspace handler replies, state | |
72 | * transitions to REPLIED. | |
73 | */ | |
74 | enum notify_state state; | |
75 | ||
76 | /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */ | |
77 | int error; | |
78 | long val; | |
fb3c5386 | 79 | u32 flags; |
6a21cc50 TA |
80 | |
81 | /* Signals when this has entered SECCOMP_NOTIFY_REPLIED */ | |
82 | struct completion ready; | |
83 | ||
84 | struct list_head list; | |
85 | }; | |
86 | ||
87 | /** | |
88 | * struct notification - container for seccomp userspace notifications. Since | |
89 | * most seccomp filters will not have notification listeners attached and this | |
90 | * structure is fairly large, we store the notification-specific stuff in a | |
91 | * separate structure. | |
92 | * | |
93 | * @request: A semaphore that users of this notification can wait on for | |
94 | * changes. Actual reads and writes are still controlled with | |
95 | * filter->notify_lock. | |
96 | * @next_id: The id of the next request. | |
97 | * @notifications: A list of struct seccomp_knotif elements. | |
98 | * @wqh: A wait queue for poll. | |
99 | */ | |
100 | struct notification { | |
101 | struct semaphore request; | |
102 | u64 next_id; | |
103 | struct list_head notifications; | |
104 | wait_queue_head_t wqh; | |
105 | }; | |
e2cfabdf WD |
106 | |
107 | /** | |
108 | * struct seccomp_filter - container for seccomp BPF programs | |
109 | * | |
110 | * @usage: reference count to manage the object lifetime. | |
111 | * get/put helpers should be used when accessing an instance | |
112 | * outside of a lifetime-guarded section. In general, this | |
113 | * is only needed for handling filters shared across tasks. | |
e66a3997 | 114 | * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged |
e2cfabdf | 115 | * @prev: points to a previously installed, or inherited, filter |
285fdfc5 | 116 | * @prog: the BPF program to evaluate |
6a21cc50 TA |
117 | * @notif: the struct that holds all notification related information |
118 | * @notify_lock: A lock for all notification-related accesses. | |
e2cfabdf WD |
119 | * |
120 | * seccomp_filter objects are organized in a tree linked via the @prev | |
121 | * pointer. For any task, it appears to be a singly-linked list starting | |
122 | * with current->seccomp.filter, the most recently attached or inherited filter. | |
123 | * However, multiple filters may share a @prev node, by way of fork(), which | |
124 | * results in a unidirectional tree existing in memory. This is similar to | |
125 | * how namespaces work. | |
126 | * | |
127 | * seccomp_filter objects should never be modified after being attached | |
128 | * to a task_struct (other than @usage). | |
129 | */ | |
130 | struct seccomp_filter { | |
0b5fa229 | 131 | refcount_t usage; |
e66a3997 | 132 | bool log; |
e2cfabdf | 133 | struct seccomp_filter *prev; |
7ae457c1 | 134 | struct bpf_prog *prog; |
6a21cc50 TA |
135 | struct notification *notif; |
136 | struct mutex notify_lock; | |
e2cfabdf WD |
137 | }; |
138 | ||
139 | /* Limit any path through the tree to 256KB worth of instructions. */ | |
140 | #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter)) | |
141 | ||
bd4cf0ed | 142 | /* |
e2cfabdf WD |
143 | * Endianness is explicitly ignored and left for BPF program authors to manage |
144 | * as per the specific architecture. | |
145 | */ | |
bd4cf0ed | 146 | static void populate_seccomp_data(struct seccomp_data *sd) |
e2cfabdf | 147 | { |
bd4cf0ed AS |
148 | struct task_struct *task = current; |
149 | struct pt_regs *regs = task_pt_regs(task); | |
2eac7648 | 150 | unsigned long args[6]; |
e2cfabdf | 151 | |
bd4cf0ed | 152 | sd->nr = syscall_get_nr(task, regs); |
16add411 | 153 | sd->arch = syscall_get_arch(task); |
b35f549d | 154 | syscall_get_arguments(task, regs, args); |
2eac7648 DB |
155 | sd->args[0] = args[0]; |
156 | sd->args[1] = args[1]; | |
157 | sd->args[2] = args[2]; | |
158 | sd->args[3] = args[3]; | |
159 | sd->args[4] = args[4]; | |
160 | sd->args[5] = args[5]; | |
bd4cf0ed | 161 | sd->instruction_pointer = KSTK_EIP(task); |
e2cfabdf WD |
162 | } |
163 | ||
164 | /** | |
165 | * seccomp_check_filter - verify seccomp filter code | |
166 | * @filter: filter to verify | |
167 | * @flen: length of filter | |
168 | * | |
4df95ff4 | 169 | * Takes a previously checked filter (by bpf_check_classic) and |
e2cfabdf WD |
170 | * redirects all filter code that loads struct sk_buff data |
171 | * and related data through seccomp_bpf_load. It also | |
172 | * enforces length and alignment checking of those loads. | |
173 | * | |
174 | * Returns 0 if the rule set is legal or -EINVAL if not. | |
175 | */ | |
176 | static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen) | |
177 | { | |
178 | int pc; | |
179 | for (pc = 0; pc < flen; pc++) { | |
180 | struct sock_filter *ftest = &filter[pc]; | |
181 | u16 code = ftest->code; | |
182 | u32 k = ftest->k; | |
183 | ||
184 | switch (code) { | |
34805931 | 185 | case BPF_LD | BPF_W | BPF_ABS: |
bd4cf0ed | 186 | ftest->code = BPF_LDX | BPF_W | BPF_ABS; |
e2cfabdf WD |
187 | /* 32-bit aligned and not out of bounds. */ |
188 | if (k >= sizeof(struct seccomp_data) || k & 3) | |
189 | return -EINVAL; | |
190 | continue; | |
34805931 | 191 | case BPF_LD | BPF_W | BPF_LEN: |
bd4cf0ed | 192 | ftest->code = BPF_LD | BPF_IMM; |
e2cfabdf WD |
193 | ftest->k = sizeof(struct seccomp_data); |
194 | continue; | |
34805931 | 195 | case BPF_LDX | BPF_W | BPF_LEN: |
bd4cf0ed | 196 | ftest->code = BPF_LDX | BPF_IMM; |
e2cfabdf WD |
197 | ftest->k = sizeof(struct seccomp_data); |
198 | continue; | |
199 | /* Explicitly include allowed calls. */ | |
34805931 DB |
200 | case BPF_RET | BPF_K: |
201 | case BPF_RET | BPF_A: | |
202 | case BPF_ALU | BPF_ADD | BPF_K: | |
203 | case BPF_ALU | BPF_ADD | BPF_X: | |
204 | case BPF_ALU | BPF_SUB | BPF_K: | |
205 | case BPF_ALU | BPF_SUB | BPF_X: | |
206 | case BPF_ALU | BPF_MUL | BPF_K: | |
207 | case BPF_ALU | BPF_MUL | BPF_X: | |
208 | case BPF_ALU | BPF_DIV | BPF_K: | |
209 | case BPF_ALU | BPF_DIV | BPF_X: | |
210 | case BPF_ALU | BPF_AND | BPF_K: | |
211 | case BPF_ALU | BPF_AND | BPF_X: | |
212 | case BPF_ALU | BPF_OR | BPF_K: | |
213 | case BPF_ALU | BPF_OR | BPF_X: | |
214 | case BPF_ALU | BPF_XOR | BPF_K: | |
215 | case BPF_ALU | BPF_XOR | BPF_X: | |
216 | case BPF_ALU | BPF_LSH | BPF_K: | |
217 | case BPF_ALU | BPF_LSH | BPF_X: | |
218 | case BPF_ALU | BPF_RSH | BPF_K: | |
219 | case BPF_ALU | BPF_RSH | BPF_X: | |
220 | case BPF_ALU | BPF_NEG: | |
221 | case BPF_LD | BPF_IMM: | |
222 | case BPF_LDX | BPF_IMM: | |
223 | case BPF_MISC | BPF_TAX: | |
224 | case BPF_MISC | BPF_TXA: | |
225 | case BPF_LD | BPF_MEM: | |
226 | case BPF_LDX | BPF_MEM: | |
227 | case BPF_ST: | |
228 | case BPF_STX: | |
229 | case BPF_JMP | BPF_JA: | |
230 | case BPF_JMP | BPF_JEQ | BPF_K: | |
231 | case BPF_JMP | BPF_JEQ | BPF_X: | |
232 | case BPF_JMP | BPF_JGE | BPF_K: | |
233 | case BPF_JMP | BPF_JGE | BPF_X: | |
234 | case BPF_JMP | BPF_JGT | BPF_K: | |
235 | case BPF_JMP | BPF_JGT | BPF_X: | |
236 | case BPF_JMP | BPF_JSET | BPF_K: | |
237 | case BPF_JMP | BPF_JSET | BPF_X: | |
e2cfabdf WD |
238 | continue; |
239 | default: | |
240 | return -EINVAL; | |
241 | } | |
242 | } | |
243 | return 0; | |
244 | } | |
245 | ||
246 | /** | |
285fdfc5 MS |
247 | * seccomp_run_filters - evaluates all seccomp filters against @sd |
248 | * @sd: optional seccomp data to be passed to filters | |
deb4de8b KC |
249 | * @match: stores struct seccomp_filter that resulted in the return value, |
250 | * unless filter returned SECCOMP_RET_ALLOW, in which case it will | |
251 | * be unchanged. | |
e2cfabdf WD |
252 | * |
253 | * Returns valid seccomp BPF response codes. | |
254 | */ | |
0466bdb9 | 255 | #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL))) |
deb4de8b KC |
256 | static u32 seccomp_run_filters(const struct seccomp_data *sd, |
257 | struct seccomp_filter **match) | |
e2cfabdf | 258 | { |
acf3b2c7 | 259 | u32 ret = SECCOMP_RET_ALLOW; |
8225d385 PK |
260 | /* Make sure cross-thread synced filter points somewhere sane. */ |
261 | struct seccomp_filter *f = | |
506458ef | 262 | READ_ONCE(current->seccomp.filter); |
acf3b2c7 WD |
263 | |
264 | /* Ensure unexpected behavior doesn't result in failing open. */ | |
0d42d73a | 265 | if (WARN_ON(f == NULL)) |
4d3b0b05 | 266 | return SECCOMP_RET_KILL_PROCESS; |
acf3b2c7 | 267 | |
e2cfabdf WD |
268 | /* |
269 | * All filters in the list are evaluated and the lowest BPF return | |
acf3b2c7 | 270 | * value always takes priority (ignoring the DATA). |
e2cfabdf | 271 | */ |
3ba2530c | 272 | for (; f; f = f->prev) { |
3d9f773c | 273 | u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd); |
8f577cad | 274 | |
0466bdb9 | 275 | if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) { |
acf3b2c7 | 276 | ret = cur_ret; |
deb4de8b KC |
277 | *match = f; |
278 | } | |
e2cfabdf WD |
279 | } |
280 | return ret; | |
281 | } | |
1f41b450 | 282 | #endif /* CONFIG_SECCOMP_FILTER */ |
e2cfabdf | 283 | |
1f41b450 KC |
284 | static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) |
285 | { | |
69f6a34b | 286 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 287 | |
1f41b450 KC |
288 | if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) |
289 | return false; | |
290 | ||
291 | return true; | |
292 | } | |
293 | ||
8bf37d8c | 294 | void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { } |
5c307089 | 295 | |
3ba2530c | 296 | static inline void seccomp_assign_mode(struct task_struct *task, |
00a02d0c KC |
297 | unsigned long seccomp_mode, |
298 | unsigned long flags) | |
1f41b450 | 299 | { |
69f6a34b | 300 | assert_spin_locked(&task->sighand->siglock); |
dbd95212 | 301 | |
3ba2530c KC |
302 | task->seccomp.mode = seccomp_mode; |
303 | /* | |
304 | * Make sure TIF_SECCOMP cannot be set before the mode (and | |
305 | * filter) is set. | |
306 | */ | |
307 | smp_mb__before_atomic(); | |
00a02d0c KC |
308 | /* Assume default seccomp processes want spec flaw mitigation. */ |
309 | if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0) | |
8bf37d8c | 310 | arch_seccomp_spec_mitigate(task); |
3ba2530c | 311 | set_tsk_thread_flag(task, TIF_SECCOMP); |
1f41b450 KC |
312 | } |
313 | ||
314 | #ifdef CONFIG_SECCOMP_FILTER | |
c2e1f2e3 KC |
315 | /* Returns 1 if the parent is an ancestor of the child. */ |
316 | static int is_ancestor(struct seccomp_filter *parent, | |
317 | struct seccomp_filter *child) | |
318 | { | |
319 | /* NULL is the root ancestor. */ | |
320 | if (parent == NULL) | |
321 | return 1; | |
322 | for (; child; child = child->prev) | |
323 | if (child == parent) | |
324 | return 1; | |
325 | return 0; | |
326 | } | |
327 | ||
328 | /** | |
329 | * seccomp_can_sync_threads: checks if all threads can be synchronized | |
330 | * | |
331 | * Expects sighand and cred_guard_mutex locks to be held. | |
332 | * | |
333 | * Returns 0 on success, -ve on error, or the pid of a thread which was | |
6beff00b | 334 | * either not in the correct seccomp mode or did not have an ancestral |
c2e1f2e3 KC |
335 | * seccomp filter. |
336 | */ | |
337 | static inline pid_t seccomp_can_sync_threads(void) | |
338 | { | |
339 | struct task_struct *thread, *caller; | |
340 | ||
341 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 342 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
343 | |
344 | /* Validate all threads being eligible for synchronization. */ | |
345 | caller = current; | |
346 | for_each_thread(caller, thread) { | |
347 | pid_t failed; | |
348 | ||
349 | /* Skip current, since it is initiating the sync. */ | |
350 | if (thread == caller) | |
351 | continue; | |
352 | ||
353 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || | |
354 | (thread->seccomp.mode == SECCOMP_MODE_FILTER && | |
355 | is_ancestor(thread->seccomp.filter, | |
356 | caller->seccomp.filter))) | |
357 | continue; | |
358 | ||
359 | /* Return the first thread that cannot be synchronized. */ | |
360 | failed = task_pid_vnr(thread); | |
361 | /* If the pid cannot be resolved, then return -ESRCH */ | |
0d42d73a | 362 | if (WARN_ON(failed == 0)) |
c2e1f2e3 KC |
363 | failed = -ESRCH; |
364 | return failed; | |
365 | } | |
366 | ||
367 | return 0; | |
368 | } | |
369 | ||
370 | /** | |
371 | * seccomp_sync_threads: sets all threads to use current's filter | |
372 | * | |
373 | * Expects sighand and cred_guard_mutex locks to be held, and for | |
374 | * seccomp_can_sync_threads() to have returned success already | |
375 | * without dropping the locks. | |
376 | * | |
377 | */ | |
00a02d0c | 378 | static inline void seccomp_sync_threads(unsigned long flags) |
c2e1f2e3 KC |
379 | { |
380 | struct task_struct *thread, *caller; | |
381 | ||
382 | BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); | |
69f6a34b | 383 | assert_spin_locked(¤t->sighand->siglock); |
c2e1f2e3 KC |
384 | |
385 | /* Synchronize all threads. */ | |
386 | caller = current; | |
387 | for_each_thread(caller, thread) { | |
388 | /* Skip current, since it needs no changes. */ | |
389 | if (thread == caller) | |
390 | continue; | |
391 | ||
392 | /* Get a task reference for the new leaf node. */ | |
393 | get_seccomp_filter(caller); | |
394 | /* | |
395 | * Drop the task reference to the shared ancestor since | |
396 | * current's path will hold a reference. (This also | |
397 | * allows a put before the assignment.) | |
398 | */ | |
399 | put_seccomp_filter(thread); | |
400 | smp_store_release(&thread->seccomp.filter, | |
401 | caller->seccomp.filter); | |
c818c03b KC |
402 | atomic_set(&thread->seccomp.filter_count, |
403 | atomic_read(&thread->seccomp.filter_count)); | |
103502a3 JH |
404 | |
405 | /* | |
406 | * Don't let an unprivileged task work around | |
407 | * the no_new_privs restriction by creating | |
408 | * a thread that sets it up, enters seccomp, | |
409 | * then dies. | |
410 | */ | |
411 | if (task_no_new_privs(caller)) | |
412 | task_set_no_new_privs(thread); | |
413 | ||
c2e1f2e3 KC |
414 | /* |
415 | * Opt the other thread into seccomp if needed. | |
416 | * As threads are considered to be trust-realm | |
417 | * equivalent (see ptrace_may_access), it is safe to | |
418 | * allow one thread to transition the other. | |
419 | */ | |
103502a3 | 420 | if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) |
00a02d0c KC |
421 | seccomp_assign_mode(thread, SECCOMP_MODE_FILTER, |
422 | flags); | |
c2e1f2e3 KC |
423 | } |
424 | } | |
425 | ||
e2cfabdf | 426 | /** |
c8bee430 | 427 | * seccomp_prepare_filter: Prepares a seccomp filter for use. |
e2cfabdf WD |
428 | * @fprog: BPF program to install |
429 | * | |
c8bee430 | 430 | * Returns filter on success or an ERR_PTR on failure. |
e2cfabdf | 431 | */ |
c8bee430 | 432 | static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog) |
e2cfabdf | 433 | { |
ac67eb2c DB |
434 | struct seccomp_filter *sfilter; |
435 | int ret; | |
97f2645f | 436 | const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE); |
e2cfabdf WD |
437 | |
438 | if (fprog->len == 0 || fprog->len > BPF_MAXINSNS) | |
c8bee430 | 439 | return ERR_PTR(-EINVAL); |
d9e12f42 | 440 | |
c8bee430 | 441 | BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter)); |
e2cfabdf WD |
442 | |
443 | /* | |
119ce5c8 | 444 | * Installing a seccomp filter requires that the task has |
e2cfabdf WD |
445 | * CAP_SYS_ADMIN in its namespace or be running with no_new_privs. |
446 | * This avoids scenarios where unprivileged tasks can affect the | |
447 | * behavior of privileged children. | |
448 | */ | |
1d4457f9 | 449 | if (!task_no_new_privs(current) && |
c1a85a00 MM |
450 | security_capable(current_cred(), current_user_ns(), |
451 | CAP_SYS_ADMIN, CAP_OPT_NOAUDIT) != 0) | |
c8bee430 | 452 | return ERR_PTR(-EACCES); |
e2cfabdf | 453 | |
bd4cf0ed | 454 | /* Allocate a new seccomp_filter */ |
ac67eb2c DB |
455 | sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN); |
456 | if (!sfilter) | |
d9e12f42 | 457 | return ERR_PTR(-ENOMEM); |
ac67eb2c | 458 | |
6a21cc50 | 459 | mutex_init(&sfilter->notify_lock); |
ac67eb2c | 460 | ret = bpf_prog_create_from_user(&sfilter->prog, fprog, |
f8e529ed | 461 | seccomp_check_filter, save_orig); |
ac67eb2c DB |
462 | if (ret < 0) { |
463 | kfree(sfilter); | |
464 | return ERR_PTR(ret); | |
d9e12f42 | 465 | } |
bd4cf0ed | 466 | |
0b5fa229 | 467 | refcount_set(&sfilter->usage, 1); |
e2cfabdf | 468 | |
ac67eb2c | 469 | return sfilter; |
e2cfabdf WD |
470 | } |
471 | ||
472 | /** | |
c8bee430 | 473 | * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog |
e2cfabdf WD |
474 | * @user_filter: pointer to the user data containing a sock_fprog. |
475 | * | |
476 | * Returns 0 on success and non-zero otherwise. | |
477 | */ | |
c8bee430 KC |
478 | static struct seccomp_filter * |
479 | seccomp_prepare_user_filter(const char __user *user_filter) | |
e2cfabdf WD |
480 | { |
481 | struct sock_fprog fprog; | |
c8bee430 | 482 | struct seccomp_filter *filter = ERR_PTR(-EFAULT); |
e2cfabdf WD |
483 | |
484 | #ifdef CONFIG_COMPAT | |
5c38065e | 485 | if (in_compat_syscall()) { |
e2cfabdf WD |
486 | struct compat_sock_fprog fprog32; |
487 | if (copy_from_user(&fprog32, user_filter, sizeof(fprog32))) | |
488 | goto out; | |
489 | fprog.len = fprog32.len; | |
490 | fprog.filter = compat_ptr(fprog32.filter); | |
491 | } else /* falls through to the if below. */ | |
492 | #endif | |
493 | if (copy_from_user(&fprog, user_filter, sizeof(fprog))) | |
494 | goto out; | |
c8bee430 | 495 | filter = seccomp_prepare_filter(&fprog); |
e2cfabdf | 496 | out: |
c8bee430 KC |
497 | return filter; |
498 | } | |
499 | ||
500 | /** | |
501 | * seccomp_attach_filter: validate and attach filter | |
502 | * @flags: flags to change filter behavior | |
503 | * @filter: seccomp filter to add to the current process | |
504 | * | |
dbd95212 KC |
505 | * Caller must be holding current->sighand->siglock lock. |
506 | * | |
7a0df7fb TA |
507 | * Returns 0 on success, -ve on error, or |
508 | * - in TSYNC mode: the pid of a thread which was either not in the correct | |
509 | * seccomp mode or did not have an ancestral seccomp filter | |
510 | * - in NEW_LISTENER mode: the fd of the new listener | |
c8bee430 KC |
511 | */ |
512 | static long seccomp_attach_filter(unsigned int flags, | |
513 | struct seccomp_filter *filter) | |
514 | { | |
515 | unsigned long total_insns; | |
516 | struct seccomp_filter *walker; | |
517 | ||
69f6a34b | 518 | assert_spin_locked(¤t->sighand->siglock); |
dbd95212 | 519 | |
c8bee430 KC |
520 | /* Validate resulting filter length. */ |
521 | total_insns = filter->prog->len; | |
522 | for (walker = current->seccomp.filter; walker; walker = walker->prev) | |
523 | total_insns += walker->prog->len + 4; /* 4 instr penalty */ | |
524 | if (total_insns > MAX_INSNS_PER_PATH) | |
525 | return -ENOMEM; | |
526 | ||
c2e1f2e3 KC |
527 | /* If thread sync has been requested, check that it is possible. */ |
528 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) { | |
529 | int ret; | |
530 | ||
531 | ret = seccomp_can_sync_threads(); | |
51891498 TA |
532 | if (ret) { |
533 | if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) | |
534 | return -ESRCH; | |
535 | else | |
536 | return ret; | |
537 | } | |
c2e1f2e3 KC |
538 | } |
539 | ||
e66a3997 TH |
540 | /* Set log flag, if present. */ |
541 | if (flags & SECCOMP_FILTER_FLAG_LOG) | |
542 | filter->log = true; | |
543 | ||
c8bee430 KC |
544 | /* |
545 | * If there is an existing filter, make it the prev and don't drop its | |
546 | * task reference. | |
547 | */ | |
548 | filter->prev = current->seccomp.filter; | |
549 | current->seccomp.filter = filter; | |
c818c03b | 550 | atomic_inc(¤t->seccomp.filter_count); |
c8bee430 | 551 | |
c2e1f2e3 KC |
552 | /* Now that the new filter is in place, synchronize to all threads. */ |
553 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) | |
00a02d0c | 554 | seccomp_sync_threads(flags); |
c2e1f2e3 | 555 | |
c8bee430 | 556 | return 0; |
e2cfabdf WD |
557 | } |
558 | ||
084f5601 | 559 | static void __get_seccomp_filter(struct seccomp_filter *filter) |
66a733ea | 560 | { |
66a733ea ON |
561 | refcount_inc(&filter->usage); |
562 | } | |
563 | ||
e2cfabdf WD |
564 | /* get_seccomp_filter - increments the reference count of the filter on @tsk */ |
565 | void get_seccomp_filter(struct task_struct *tsk) | |
566 | { | |
567 | struct seccomp_filter *orig = tsk->seccomp.filter; | |
568 | if (!orig) | |
569 | return; | |
66a733ea | 570 | __get_seccomp_filter(orig); |
e2cfabdf WD |
571 | } |
572 | ||
c8bee430 KC |
573 | static inline void seccomp_filter_free(struct seccomp_filter *filter) |
574 | { | |
575 | if (filter) { | |
bab18991 | 576 | bpf_prog_destroy(filter->prog); |
c8bee430 KC |
577 | kfree(filter); |
578 | } | |
579 | } | |
580 | ||
66a733ea | 581 | static void __put_seccomp_filter(struct seccomp_filter *orig) |
e2cfabdf | 582 | { |
e2cfabdf | 583 | /* Clean up single-reference branches iteratively. */ |
0b5fa229 | 584 | while (orig && refcount_dec_and_test(&orig->usage)) { |
e2cfabdf WD |
585 | struct seccomp_filter *freeme = orig; |
586 | orig = orig->prev; | |
c8bee430 | 587 | seccomp_filter_free(freeme); |
e2cfabdf WD |
588 | } |
589 | } | |
bb6ea430 | 590 | |
66a733ea ON |
591 | /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */ |
592 | void put_seccomp_filter(struct task_struct *tsk) | |
593 | { | |
594 | __put_seccomp_filter(tsk->seccomp.filter); | |
595 | } | |
596 | ||
ae7795bc | 597 | static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason) |
b25e6716 | 598 | { |
3b10db2b | 599 | clear_siginfo(info); |
b25e6716 MF |
600 | info->si_signo = SIGSYS; |
601 | info->si_code = SYS_SECCOMP; | |
602 | info->si_call_addr = (void __user *)KSTK_EIP(current); | |
603 | info->si_errno = reason; | |
16add411 | 604 | info->si_arch = syscall_get_arch(current); |
b25e6716 MF |
605 | info->si_syscall = syscall; |
606 | } | |
607 | ||
bb6ea430 WD |
608 | /** |
609 | * seccomp_send_sigsys - signals the task to allow in-process syscall emulation | |
610 | * @syscall: syscall number to send to userland | |
611 | * @reason: filter-supplied reason code to send to userland (via si_errno) | |
612 | * | |
613 | * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info. | |
614 | */ | |
615 | static void seccomp_send_sigsys(int syscall, int reason) | |
616 | { | |
ae7795bc | 617 | struct kernel_siginfo info; |
b25e6716 | 618 | seccomp_init_siginfo(&info, syscall, reason); |
a89e9b8a | 619 | force_sig_info(&info); |
bb6ea430 | 620 | } |
e2cfabdf | 621 | #endif /* CONFIG_SECCOMP_FILTER */ |
1da177e4 | 622 | |
0ddec0fc | 623 | /* For use with seccomp_actions_logged */ |
4d3b0b05 KC |
624 | #define SECCOMP_LOG_KILL_PROCESS (1 << 0) |
625 | #define SECCOMP_LOG_KILL_THREAD (1 << 1) | |
0ddec0fc TH |
626 | #define SECCOMP_LOG_TRAP (1 << 2) |
627 | #define SECCOMP_LOG_ERRNO (1 << 3) | |
628 | #define SECCOMP_LOG_TRACE (1 << 4) | |
59f5cf44 TH |
629 | #define SECCOMP_LOG_LOG (1 << 5) |
630 | #define SECCOMP_LOG_ALLOW (1 << 6) | |
6a21cc50 | 631 | #define SECCOMP_LOG_USER_NOTIF (1 << 7) |
0ddec0fc | 632 | |
4d3b0b05 KC |
633 | static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS | |
634 | SECCOMP_LOG_KILL_THREAD | | |
fd76875c KC |
635 | SECCOMP_LOG_TRAP | |
636 | SECCOMP_LOG_ERRNO | | |
6a21cc50 | 637 | SECCOMP_LOG_USER_NOTIF | |
fd76875c | 638 | SECCOMP_LOG_TRACE | |
59f5cf44 | 639 | SECCOMP_LOG_LOG; |
0ddec0fc | 640 | |
e66a3997 TH |
641 | static inline void seccomp_log(unsigned long syscall, long signr, u32 action, |
642 | bool requested) | |
0ddec0fc TH |
643 | { |
644 | bool log = false; | |
645 | ||
646 | switch (action) { | |
647 | case SECCOMP_RET_ALLOW: | |
e66a3997 | 648 | break; |
0ddec0fc | 649 | case SECCOMP_RET_TRAP: |
e66a3997 TH |
650 | log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP; |
651 | break; | |
0ddec0fc | 652 | case SECCOMP_RET_ERRNO: |
e66a3997 TH |
653 | log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO; |
654 | break; | |
0ddec0fc | 655 | case SECCOMP_RET_TRACE: |
e66a3997 | 656 | log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE; |
0ddec0fc | 657 | break; |
6a21cc50 TA |
658 | case SECCOMP_RET_USER_NOTIF: |
659 | log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF; | |
660 | break; | |
59f5cf44 TH |
661 | case SECCOMP_RET_LOG: |
662 | log = seccomp_actions_logged & SECCOMP_LOG_LOG; | |
663 | break; | |
fd76875c | 664 | case SECCOMP_RET_KILL_THREAD: |
fd76875c | 665 | log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD; |
4d3b0b05 KC |
666 | break; |
667 | case SECCOMP_RET_KILL_PROCESS: | |
668 | default: | |
669 | log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS; | |
0ddec0fc TH |
670 | } |
671 | ||
672 | /* | |
326bee02 TH |
673 | * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the |
674 | * FILTER_FLAG_LOG bit was set. The admin has the ability to silence | |
675 | * any action from being logged by removing the action name from the | |
676 | * seccomp_actions_logged sysctl. | |
0ddec0fc | 677 | */ |
326bee02 TH |
678 | if (!log) |
679 | return; | |
0ddec0fc | 680 | |
326bee02 | 681 | audit_seccomp(syscall, signr, action); |
0ddec0fc TH |
682 | } |
683 | ||
1da177e4 LT |
684 | /* |
685 | * Secure computing mode 1 allows only read/write/exit/sigreturn. | |
686 | * To be fully secure this must be combined with rlimit | |
687 | * to limit the stack allocations too. | |
688 | */ | |
cb4253aa | 689 | static const int mode1_syscalls[] = { |
1da177e4 LT |
690 | __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, |
691 | 0, /* null terminated */ | |
692 | }; | |
693 | ||
a4412fc9 | 694 | static void __secure_computing_strict(int this_syscall) |
1da177e4 | 695 | { |
cb4253aa | 696 | const int *syscall_whitelist = mode1_syscalls; |
a4412fc9 | 697 | #ifdef CONFIG_COMPAT |
5c38065e | 698 | if (in_compat_syscall()) |
c983f0e8 | 699 | syscall_whitelist = get_compat_mode1_syscalls(); |
a4412fc9 AL |
700 | #endif |
701 | do { | |
702 | if (*syscall_whitelist == this_syscall) | |
703 | return; | |
704 | } while (*++syscall_whitelist); | |
705 | ||
706 | #ifdef SECCOMP_DEBUG | |
707 | dump_stack(); | |
708 | #endif | |
fd76875c | 709 | seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true); |
a4412fc9 AL |
710 | do_exit(SIGKILL); |
711 | } | |
712 | ||
713 | #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER | |
714 | void secure_computing_strict(int this_syscall) | |
715 | { | |
716 | int mode = current->seccomp.mode; | |
717 | ||
97f2645f | 718 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 TA |
719 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
720 | return; | |
721 | ||
221272f9 | 722 | if (mode == SECCOMP_MODE_DISABLED) |
a4412fc9 AL |
723 | return; |
724 | else if (mode == SECCOMP_MODE_STRICT) | |
725 | __secure_computing_strict(this_syscall); | |
726 | else | |
727 | BUG(); | |
728 | } | |
729 | #else | |
13aa72f0 AL |
730 | |
731 | #ifdef CONFIG_SECCOMP_FILTER | |
6a21cc50 TA |
732 | static u64 seccomp_next_notify_id(struct seccomp_filter *filter) |
733 | { | |
734 | /* | |
735 | * Note: overflow is ok here, the id just needs to be unique per | |
736 | * filter. | |
737 | */ | |
738 | lockdep_assert_held(&filter->notify_lock); | |
739 | return filter->notif->next_id++; | |
740 | } | |
741 | ||
fb3c5386 CB |
742 | static int seccomp_do_user_notification(int this_syscall, |
743 | struct seccomp_filter *match, | |
744 | const struct seccomp_data *sd) | |
6a21cc50 TA |
745 | { |
746 | int err; | |
fb3c5386 | 747 | u32 flags = 0; |
6a21cc50 TA |
748 | long ret = 0; |
749 | struct seccomp_knotif n = {}; | |
750 | ||
751 | mutex_lock(&match->notify_lock); | |
752 | err = -ENOSYS; | |
753 | if (!match->notif) | |
754 | goto out; | |
755 | ||
756 | n.task = current; | |
757 | n.state = SECCOMP_NOTIFY_INIT; | |
758 | n.data = sd; | |
759 | n.id = seccomp_next_notify_id(match); | |
760 | init_completion(&n.ready); | |
761 | list_add(&n.list, &match->notif->notifications); | |
762 | ||
763 | up(&match->notif->request); | |
764 | wake_up_poll(&match->notif->wqh, EPOLLIN | EPOLLRDNORM); | |
765 | mutex_unlock(&match->notify_lock); | |
766 | ||
767 | /* | |
768 | * This is where we wait for a reply from userspace. | |
769 | */ | |
770 | err = wait_for_completion_interruptible(&n.ready); | |
771 | mutex_lock(&match->notify_lock); | |
772 | if (err == 0) { | |
773 | ret = n.val; | |
774 | err = n.error; | |
fb3c5386 | 775 | flags = n.flags; |
6a21cc50 TA |
776 | } |
777 | ||
778 | /* | |
779 | * Note that it's possible the listener died in between the time when | |
780 | * we were notified of a respons (or a signal) and when we were able to | |
781 | * re-acquire the lock, so only delete from the list if the | |
782 | * notification actually exists. | |
783 | * | |
784 | * Also note that this test is only valid because there's no way to | |
785 | * *reattach* to a notifier right now. If one is added, we'll need to | |
786 | * keep track of the notif itself and make sure they match here. | |
787 | */ | |
788 | if (match->notif) | |
789 | list_del(&n.list); | |
790 | out: | |
791 | mutex_unlock(&match->notify_lock); | |
fb3c5386 CB |
792 | |
793 | /* Userspace requests to continue the syscall. */ | |
794 | if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) | |
795 | return 0; | |
796 | ||
6a21cc50 TA |
797 | syscall_set_return_value(current, task_pt_regs(current), |
798 | err, ret); | |
fb3c5386 | 799 | return -1; |
6a21cc50 TA |
800 | } |
801 | ||
ce6526e8 KC |
802 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
803 | const bool recheck_after_trace) | |
13aa72f0 AL |
804 | { |
805 | u32 filter_ret, action; | |
deb4de8b | 806 | struct seccomp_filter *match = NULL; |
13aa72f0 | 807 | int data; |
db511391 | 808 | struct seccomp_data sd_local; |
1da177e4 | 809 | |
3ba2530c KC |
810 | /* |
811 | * Make sure that any changes to mode from another thread have | |
812 | * been seen after TIF_SECCOMP was seen. | |
813 | */ | |
814 | rmb(); | |
815 | ||
db511391 TA |
816 | if (!sd) { |
817 | populate_seccomp_data(&sd_local); | |
818 | sd = &sd_local; | |
819 | } | |
820 | ||
deb4de8b | 821 | filter_ret = seccomp_run_filters(sd, &match); |
13aa72f0 | 822 | data = filter_ret & SECCOMP_RET_DATA; |
0466bdb9 | 823 | action = filter_ret & SECCOMP_RET_ACTION_FULL; |
13aa72f0 AL |
824 | |
825 | switch (action) { | |
826 | case SECCOMP_RET_ERRNO: | |
580c57f1 KC |
827 | /* Set low-order bits as an errno, capped at MAX_ERRNO. */ |
828 | if (data > MAX_ERRNO) | |
829 | data = MAX_ERRNO; | |
d39bd00d | 830 | syscall_set_return_value(current, task_pt_regs(current), |
13aa72f0 AL |
831 | -data, 0); |
832 | goto skip; | |
833 | ||
834 | case SECCOMP_RET_TRAP: | |
835 | /* Show the handler the original registers. */ | |
d39bd00d | 836 | syscall_rollback(current, task_pt_regs(current)); |
13aa72f0 AL |
837 | /* Let the filter pass back 16 bits of data. */ |
838 | seccomp_send_sigsys(this_syscall, data); | |
839 | goto skip; | |
840 | ||
841 | case SECCOMP_RET_TRACE: | |
ce6526e8 KC |
842 | /* We've been put in this state by the ptracer already. */ |
843 | if (recheck_after_trace) | |
844 | return 0; | |
845 | ||
8112c4f1 KC |
846 | /* ENOSYS these calls if there is no tracer attached. */ |
847 | if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { | |
848 | syscall_set_return_value(current, | |
849 | task_pt_regs(current), | |
850 | -ENOSYS, 0); | |
851 | goto skip; | |
852 | } | |
853 | ||
854 | /* Allow the BPF to provide the event message */ | |
855 | ptrace_event(PTRACE_EVENT_SECCOMP, data); | |
856 | /* | |
857 | * The delivery of a fatal signal during event | |
485a252a KC |
858 | * notification may silently skip tracer notification, |
859 | * which could leave us with a potentially unmodified | |
860 | * syscall that the tracer would have liked to have | |
861 | * changed. Since the process is about to die, we just | |
862 | * force the syscall to be skipped and let the signal | |
863 | * kill the process and correctly handle any tracer exit | |
864 | * notifications. | |
8112c4f1 KC |
865 | */ |
866 | if (fatal_signal_pending(current)) | |
485a252a | 867 | goto skip; |
8112c4f1 KC |
868 | /* Check if the tracer forced the syscall to be skipped. */ |
869 | this_syscall = syscall_get_nr(current, task_pt_regs(current)); | |
870 | if (this_syscall < 0) | |
871 | goto skip; | |
872 | ||
ce6526e8 KC |
873 | /* |
874 | * Recheck the syscall, since it may have changed. This | |
875 | * intentionally uses a NULL struct seccomp_data to force | |
876 | * a reload of all registers. This does not goto skip since | |
877 | * a skip would have already been reported. | |
878 | */ | |
879 | if (__seccomp_filter(this_syscall, NULL, true)) | |
880 | return -1; | |
881 | ||
8112c4f1 | 882 | return 0; |
13aa72f0 | 883 | |
6a21cc50 | 884 | case SECCOMP_RET_USER_NOTIF: |
fb3c5386 CB |
885 | if (seccomp_do_user_notification(this_syscall, match, sd)) |
886 | goto skip; | |
887 | ||
888 | return 0; | |
6a21cc50 | 889 | |
59f5cf44 TH |
890 | case SECCOMP_RET_LOG: |
891 | seccomp_log(this_syscall, 0, action, true); | |
892 | return 0; | |
893 | ||
13aa72f0 | 894 | case SECCOMP_RET_ALLOW: |
deb4de8b KC |
895 | /* |
896 | * Note that the "match" filter will always be NULL for | |
897 | * this action since SECCOMP_RET_ALLOW is the starting | |
898 | * state in seccomp_run_filters(). | |
899 | */ | |
8112c4f1 | 900 | return 0; |
13aa72f0 | 901 | |
fd76875c | 902 | case SECCOMP_RET_KILL_THREAD: |
4d3b0b05 | 903 | case SECCOMP_RET_KILL_PROCESS: |
131b6351 | 904 | default: |
e66a3997 | 905 | seccomp_log(this_syscall, SIGSYS, action, true); |
d7276e32 | 906 | /* Dump core only if this is the last remaining thread. */ |
4d3b0b05 KC |
907 | if (action == SECCOMP_RET_KILL_PROCESS || |
908 | get_nr_threads(current) == 1) { | |
ae7795bc | 909 | kernel_siginfo_t info; |
131b6351 | 910 | |
d7276e32 KC |
911 | /* Show the original registers in the dump. */ |
912 | syscall_rollback(current, task_pt_regs(current)); | |
913 | /* Trigger a manual coredump since do_exit skips it. */ | |
914 | seccomp_init_siginfo(&info, this_syscall, data); | |
915 | do_coredump(&info); | |
916 | } | |
4d3b0b05 KC |
917 | if (action == SECCOMP_RET_KILL_PROCESS) |
918 | do_group_exit(SIGSYS); | |
919 | else | |
920 | do_exit(SIGSYS); | |
13aa72f0 AL |
921 | } |
922 | ||
923 | unreachable(); | |
924 | ||
925 | skip: | |
e66a3997 | 926 | seccomp_log(this_syscall, 0, action, match ? match->log : false); |
8112c4f1 KC |
927 | return -1; |
928 | } | |
929 | #else | |
ce6526e8 KC |
930 | static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, |
931 | const bool recheck_after_trace) | |
8112c4f1 KC |
932 | { |
933 | BUG(); | |
13aa72f0 | 934 | } |
1da177e4 | 935 | #endif |
13aa72f0 | 936 | |
8112c4f1 | 937 | int __secure_computing(const struct seccomp_data *sd) |
13aa72f0 AL |
938 | { |
939 | int mode = current->seccomp.mode; | |
8112c4f1 | 940 | int this_syscall; |
13aa72f0 | 941 | |
97f2645f | 942 | if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && |
13c4a901 | 943 | unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) |
8112c4f1 KC |
944 | return 0; |
945 | ||
946 | this_syscall = sd ? sd->nr : | |
947 | syscall_get_nr(current, task_pt_regs(current)); | |
13c4a901 | 948 | |
13aa72f0 | 949 | switch (mode) { |
e2cfabdf | 950 | case SECCOMP_MODE_STRICT: |
13aa72f0 | 951 | __secure_computing_strict(this_syscall); /* may call do_exit */ |
8112c4f1 | 952 | return 0; |
13aa72f0 | 953 | case SECCOMP_MODE_FILTER: |
ce6526e8 | 954 | return __seccomp_filter(this_syscall, sd, false); |
1da177e4 LT |
955 | default: |
956 | BUG(); | |
957 | } | |
13aa72f0 | 958 | } |
a4412fc9 | 959 | #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ |
1d9d02fe AA |
960 | |
961 | long prctl_get_seccomp(void) | |
962 | { | |
963 | return current->seccomp.mode; | |
964 | } | |
965 | ||
e2cfabdf | 966 | /** |
3b23dd12 | 967 | * seccomp_set_mode_strict: internal function for setting strict seccomp |
e2cfabdf WD |
968 | * |
969 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
970 | * | |
971 | * Returns 0 on success or -EINVAL on failure. | |
972 | */ | |
3b23dd12 | 973 | static long seccomp_set_mode_strict(void) |
1d9d02fe | 974 | { |
3b23dd12 | 975 | const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; |
e2cfabdf | 976 | long ret = -EINVAL; |
1d9d02fe | 977 | |
dbd95212 KC |
978 | spin_lock_irq(¤t->sighand->siglock); |
979 | ||
1f41b450 | 980 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1d9d02fe AA |
981 | goto out; |
982 | ||
cf99abac | 983 | #ifdef TIF_NOTSC |
3b23dd12 | 984 | disable_TSC(); |
cf99abac | 985 | #endif |
00a02d0c | 986 | seccomp_assign_mode(current, seccomp_mode, 0); |
3b23dd12 KC |
987 | ret = 0; |
988 | ||
989 | out: | |
dbd95212 | 990 | spin_unlock_irq(¤t->sighand->siglock); |
3b23dd12 KC |
991 | |
992 | return ret; | |
993 | } | |
994 | ||
e2cfabdf | 995 | #ifdef CONFIG_SECCOMP_FILTER |
6a21cc50 TA |
996 | static int seccomp_notify_release(struct inode *inode, struct file *file) |
997 | { | |
998 | struct seccomp_filter *filter = file->private_data; | |
999 | struct seccomp_knotif *knotif; | |
1000 | ||
a811dc61 TA |
1001 | if (!filter) |
1002 | return 0; | |
1003 | ||
6a21cc50 TA |
1004 | mutex_lock(&filter->notify_lock); |
1005 | ||
1006 | /* | |
1007 | * If this file is being closed because e.g. the task who owned it | |
1008 | * died, let's wake everyone up who was waiting on us. | |
1009 | */ | |
1010 | list_for_each_entry(knotif, &filter->notif->notifications, list) { | |
1011 | if (knotif->state == SECCOMP_NOTIFY_REPLIED) | |
1012 | continue; | |
1013 | ||
1014 | knotif->state = SECCOMP_NOTIFY_REPLIED; | |
1015 | knotif->error = -ENOSYS; | |
1016 | knotif->val = 0; | |
1017 | ||
1018 | complete(&knotif->ready); | |
1019 | } | |
1020 | ||
1021 | kfree(filter->notif); | |
1022 | filter->notif = NULL; | |
1023 | mutex_unlock(&filter->notify_lock); | |
1024 | __put_seccomp_filter(filter); | |
1025 | return 0; | |
1026 | } | |
1027 | ||
9f87dcf1 SD |
1028 | /* must be called with notif_lock held */ |
1029 | static inline struct seccomp_knotif * | |
1030 | find_notification(struct seccomp_filter *filter, u64 id) | |
1031 | { | |
1032 | struct seccomp_knotif *cur; | |
1033 | ||
1034 | lockdep_assert_held(&filter->notify_lock); | |
1035 | ||
1036 | list_for_each_entry(cur, &filter->notif->notifications, list) { | |
1037 | if (cur->id == id) | |
1038 | return cur; | |
1039 | } | |
1040 | ||
1041 | return NULL; | |
1042 | } | |
1043 | ||
1044 | ||
6a21cc50 TA |
1045 | static long seccomp_notify_recv(struct seccomp_filter *filter, |
1046 | void __user *buf) | |
1047 | { | |
1048 | struct seccomp_knotif *knotif = NULL, *cur; | |
1049 | struct seccomp_notif unotif; | |
1050 | ssize_t ret; | |
1051 | ||
2882d53c SD |
1052 | /* Verify that we're not given garbage to keep struct extensible. */ |
1053 | ret = check_zeroed_user(buf, sizeof(unotif)); | |
1054 | if (ret < 0) | |
1055 | return ret; | |
1056 | if (!ret) | |
1057 | return -EINVAL; | |
1058 | ||
6a21cc50 TA |
1059 | memset(&unotif, 0, sizeof(unotif)); |
1060 | ||
1061 | ret = down_interruptible(&filter->notif->request); | |
1062 | if (ret < 0) | |
1063 | return ret; | |
1064 | ||
1065 | mutex_lock(&filter->notify_lock); | |
1066 | list_for_each_entry(cur, &filter->notif->notifications, list) { | |
1067 | if (cur->state == SECCOMP_NOTIFY_INIT) { | |
1068 | knotif = cur; | |
1069 | break; | |
1070 | } | |
1071 | } | |
1072 | ||
1073 | /* | |
1074 | * If we didn't find a notification, it could be that the task was | |
1075 | * interrupted by a fatal signal between the time we were woken and | |
1076 | * when we were able to acquire the rw lock. | |
1077 | */ | |
1078 | if (!knotif) { | |
1079 | ret = -ENOENT; | |
1080 | goto out; | |
1081 | } | |
1082 | ||
1083 | unotif.id = knotif->id; | |
1084 | unotif.pid = task_pid_vnr(knotif->task); | |
1085 | unotif.data = *(knotif->data); | |
1086 | ||
1087 | knotif->state = SECCOMP_NOTIFY_SENT; | |
1088 | wake_up_poll(&filter->notif->wqh, EPOLLOUT | EPOLLWRNORM); | |
1089 | ret = 0; | |
1090 | out: | |
1091 | mutex_unlock(&filter->notify_lock); | |
1092 | ||
1093 | if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) { | |
1094 | ret = -EFAULT; | |
1095 | ||
1096 | /* | |
1097 | * Userspace screwed up. To make sure that we keep this | |
1098 | * notification alive, let's reset it back to INIT. It | |
1099 | * may have died when we released the lock, so we need to make | |
1100 | * sure it's still around. | |
1101 | */ | |
6a21cc50 | 1102 | mutex_lock(&filter->notify_lock); |
9f87dcf1 | 1103 | knotif = find_notification(filter, unotif.id); |
6a21cc50 TA |
1104 | if (knotif) { |
1105 | knotif->state = SECCOMP_NOTIFY_INIT; | |
1106 | up(&filter->notif->request); | |
1107 | } | |
1108 | mutex_unlock(&filter->notify_lock); | |
1109 | } | |
1110 | ||
1111 | return ret; | |
1112 | } | |
1113 | ||
1114 | static long seccomp_notify_send(struct seccomp_filter *filter, | |
1115 | void __user *buf) | |
1116 | { | |
1117 | struct seccomp_notif_resp resp = {}; | |
9f87dcf1 | 1118 | struct seccomp_knotif *knotif; |
6a21cc50 TA |
1119 | long ret; |
1120 | ||
1121 | if (copy_from_user(&resp, buf, sizeof(resp))) | |
1122 | return -EFAULT; | |
1123 | ||
fb3c5386 CB |
1124 | if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE) |
1125 | return -EINVAL; | |
1126 | ||
1127 | if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) && | |
1128 | (resp.error || resp.val)) | |
6a21cc50 TA |
1129 | return -EINVAL; |
1130 | ||
1131 | ret = mutex_lock_interruptible(&filter->notify_lock); | |
1132 | if (ret < 0) | |
1133 | return ret; | |
1134 | ||
9f87dcf1 | 1135 | knotif = find_notification(filter, resp.id); |
6a21cc50 TA |
1136 | if (!knotif) { |
1137 | ret = -ENOENT; | |
1138 | goto out; | |
1139 | } | |
1140 | ||
1141 | /* Allow exactly one reply. */ | |
1142 | if (knotif->state != SECCOMP_NOTIFY_SENT) { | |
1143 | ret = -EINPROGRESS; | |
1144 | goto out; | |
1145 | } | |
1146 | ||
1147 | ret = 0; | |
1148 | knotif->state = SECCOMP_NOTIFY_REPLIED; | |
1149 | knotif->error = resp.error; | |
1150 | knotif->val = resp.val; | |
fb3c5386 | 1151 | knotif->flags = resp.flags; |
6a21cc50 TA |
1152 | complete(&knotif->ready); |
1153 | out: | |
1154 | mutex_unlock(&filter->notify_lock); | |
1155 | return ret; | |
1156 | } | |
1157 | ||
1158 | static long seccomp_notify_id_valid(struct seccomp_filter *filter, | |
1159 | void __user *buf) | |
1160 | { | |
9f87dcf1 | 1161 | struct seccomp_knotif *knotif; |
6a21cc50 TA |
1162 | u64 id; |
1163 | long ret; | |
1164 | ||
1165 | if (copy_from_user(&id, buf, sizeof(id))) | |
1166 | return -EFAULT; | |
1167 | ||
1168 | ret = mutex_lock_interruptible(&filter->notify_lock); | |
1169 | if (ret < 0) | |
1170 | return ret; | |
1171 | ||
9f87dcf1 SD |
1172 | knotif = find_notification(filter, id); |
1173 | if (knotif && knotif->state == SECCOMP_NOTIFY_SENT) | |
1174 | ret = 0; | |
1175 | else | |
1176 | ret = -ENOENT; | |
6a21cc50 | 1177 | |
6a21cc50 TA |
1178 | mutex_unlock(&filter->notify_lock); |
1179 | return ret; | |
1180 | } | |
1181 | ||
1182 | static long seccomp_notify_ioctl(struct file *file, unsigned int cmd, | |
1183 | unsigned long arg) | |
1184 | { | |
1185 | struct seccomp_filter *filter = file->private_data; | |
1186 | void __user *buf = (void __user *)arg; | |
1187 | ||
1188 | switch (cmd) { | |
1189 | case SECCOMP_IOCTL_NOTIF_RECV: | |
1190 | return seccomp_notify_recv(filter, buf); | |
1191 | case SECCOMP_IOCTL_NOTIF_SEND: | |
1192 | return seccomp_notify_send(filter, buf); | |
1193 | case SECCOMP_IOCTL_NOTIF_ID_VALID: | |
1194 | return seccomp_notify_id_valid(filter, buf); | |
1195 | default: | |
1196 | return -EINVAL; | |
1197 | } | |
1198 | } | |
1199 | ||
1200 | static __poll_t seccomp_notify_poll(struct file *file, | |
1201 | struct poll_table_struct *poll_tab) | |
1202 | { | |
1203 | struct seccomp_filter *filter = file->private_data; | |
1204 | __poll_t ret = 0; | |
1205 | struct seccomp_knotif *cur; | |
1206 | ||
1207 | poll_wait(file, &filter->notif->wqh, poll_tab); | |
1208 | ||
319deec7 | 1209 | if (mutex_lock_interruptible(&filter->notify_lock) < 0) |
6a21cc50 TA |
1210 | return EPOLLERR; |
1211 | ||
1212 | list_for_each_entry(cur, &filter->notif->notifications, list) { | |
1213 | if (cur->state == SECCOMP_NOTIFY_INIT) | |
1214 | ret |= EPOLLIN | EPOLLRDNORM; | |
1215 | if (cur->state == SECCOMP_NOTIFY_SENT) | |
1216 | ret |= EPOLLOUT | EPOLLWRNORM; | |
1217 | if ((ret & EPOLLIN) && (ret & EPOLLOUT)) | |
1218 | break; | |
1219 | } | |
1220 | ||
1221 | mutex_unlock(&filter->notify_lock); | |
1222 | ||
1223 | return ret; | |
1224 | } | |
1225 | ||
1226 | static const struct file_operations seccomp_notify_ops = { | |
1227 | .poll = seccomp_notify_poll, | |
1228 | .release = seccomp_notify_release, | |
1229 | .unlocked_ioctl = seccomp_notify_ioctl, | |
3db81afd | 1230 | .compat_ioctl = seccomp_notify_ioctl, |
6a21cc50 TA |
1231 | }; |
1232 | ||
1233 | static struct file *init_listener(struct seccomp_filter *filter) | |
1234 | { | |
1235 | struct file *ret = ERR_PTR(-EBUSY); | |
1236 | struct seccomp_filter *cur; | |
1237 | ||
1238 | for (cur = current->seccomp.filter; cur; cur = cur->prev) { | |
1239 | if (cur->notif) | |
1240 | goto out; | |
1241 | } | |
1242 | ||
1243 | ret = ERR_PTR(-ENOMEM); | |
1244 | filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL); | |
1245 | if (!filter->notif) | |
1246 | goto out; | |
1247 | ||
1248 | sema_init(&filter->notif->request, 0); | |
1249 | filter->notif->next_id = get_random_u64(); | |
1250 | INIT_LIST_HEAD(&filter->notif->notifications); | |
1251 | init_waitqueue_head(&filter->notif->wqh); | |
1252 | ||
1253 | ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops, | |
1254 | filter, O_RDWR); | |
1255 | if (IS_ERR(ret)) | |
1256 | goto out_notif; | |
1257 | ||
1258 | /* The file has a reference to it now */ | |
1259 | __get_seccomp_filter(filter); | |
1260 | ||
1261 | out_notif: | |
1262 | if (IS_ERR(ret)) | |
1263 | kfree(filter->notif); | |
1264 | out: | |
1265 | return ret; | |
1266 | } | |
1267 | ||
3b23dd12 KC |
1268 | /** |
1269 | * seccomp_set_mode_filter: internal function for setting seccomp filter | |
48dc92b9 | 1270 | * @flags: flags to change filter behavior |
3b23dd12 KC |
1271 | * @filter: struct sock_fprog containing filter |
1272 | * | |
1273 | * This function may be called repeatedly to install additional filters. | |
1274 | * Every filter successfully installed will be evaluated (in reverse order) | |
1275 | * for each system call the task makes. | |
1276 | * | |
1277 | * Once current->seccomp.mode is non-zero, it may not be changed. | |
1278 | * | |
1279 | * Returns 0 on success or -EINVAL on failure. | |
1280 | */ | |
48dc92b9 KC |
1281 | static long seccomp_set_mode_filter(unsigned int flags, |
1282 | const char __user *filter) | |
3b23dd12 KC |
1283 | { |
1284 | const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; | |
c8bee430 | 1285 | struct seccomp_filter *prepared = NULL; |
3b23dd12 | 1286 | long ret = -EINVAL; |
6a21cc50 TA |
1287 | int listener = -1; |
1288 | struct file *listener_f = NULL; | |
3b23dd12 | 1289 | |
48dc92b9 | 1290 | /* Validate flags. */ |
c2e1f2e3 | 1291 | if (flags & ~SECCOMP_FILTER_FLAG_MASK) |
dbd95212 | 1292 | return -EINVAL; |
48dc92b9 | 1293 | |
7a0df7fb TA |
1294 | /* |
1295 | * In the successful case, NEW_LISTENER returns the new listener fd. | |
1296 | * But in the failure case, TSYNC returns the thread that died. If you | |
1297 | * combine these two flags, there's no way to tell whether something | |
51891498 TA |
1298 | * succeeded or failed. So, let's disallow this combination if the user |
1299 | * has not explicitly requested no errors from TSYNC. | |
7a0df7fb TA |
1300 | */ |
1301 | if ((flags & SECCOMP_FILTER_FLAG_TSYNC) && | |
51891498 TA |
1302 | (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) && |
1303 | ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0)) | |
7a0df7fb TA |
1304 | return -EINVAL; |
1305 | ||
c8bee430 KC |
1306 | /* Prepare the new filter before holding any locks. */ |
1307 | prepared = seccomp_prepare_user_filter(filter); | |
1308 | if (IS_ERR(prepared)) | |
1309 | return PTR_ERR(prepared); | |
1310 | ||
6a21cc50 TA |
1311 | if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) { |
1312 | listener = get_unused_fd_flags(O_CLOEXEC); | |
1313 | if (listener < 0) { | |
1314 | ret = listener; | |
1315 | goto out_free; | |
1316 | } | |
1317 | ||
1318 | listener_f = init_listener(prepared); | |
1319 | if (IS_ERR(listener_f)) { | |
1320 | put_unused_fd(listener); | |
1321 | ret = PTR_ERR(listener_f); | |
1322 | goto out_free; | |
1323 | } | |
1324 | } | |
1325 | ||
c2e1f2e3 KC |
1326 | /* |
1327 | * Make sure we cannot change seccomp or nnp state via TSYNC | |
1328 | * while another thread is in the middle of calling exec. | |
1329 | */ | |
1330 | if (flags & SECCOMP_FILTER_FLAG_TSYNC && | |
1331 | mutex_lock_killable(¤t->signal->cred_guard_mutex)) | |
6a21cc50 | 1332 | goto out_put_fd; |
c2e1f2e3 | 1333 | |
dbd95212 KC |
1334 | spin_lock_irq(¤t->sighand->siglock); |
1335 | ||
3b23dd12 KC |
1336 | if (!seccomp_may_assign_mode(seccomp_mode)) |
1337 | goto out; | |
1338 | ||
c8bee430 | 1339 | ret = seccomp_attach_filter(flags, prepared); |
3b23dd12 | 1340 | if (ret) |
e2cfabdf | 1341 | goto out; |
c8bee430 KC |
1342 | /* Do not free the successfully attached filter. */ |
1343 | prepared = NULL; | |
1d9d02fe | 1344 | |
00a02d0c | 1345 | seccomp_assign_mode(current, seccomp_mode, flags); |
e2cfabdf | 1346 | out: |
dbd95212 | 1347 | spin_unlock_irq(¤t->sighand->siglock); |
c2e1f2e3 KC |
1348 | if (flags & SECCOMP_FILTER_FLAG_TSYNC) |
1349 | mutex_unlock(¤t->signal->cred_guard_mutex); | |
6a21cc50 TA |
1350 | out_put_fd: |
1351 | if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) { | |
7a0df7fb | 1352 | if (ret) { |
a811dc61 | 1353 | listener_f->private_data = NULL; |
6a21cc50 TA |
1354 | fput(listener_f); |
1355 | put_unused_fd(listener); | |
1356 | } else { | |
1357 | fd_install(listener, listener_f); | |
1358 | ret = listener; | |
1359 | } | |
1360 | } | |
c2e1f2e3 | 1361 | out_free: |
c8bee430 | 1362 | seccomp_filter_free(prepared); |
1d9d02fe AA |
1363 | return ret; |
1364 | } | |
3b23dd12 | 1365 | #else |
48dc92b9 KC |
1366 | static inline long seccomp_set_mode_filter(unsigned int flags, |
1367 | const char __user *filter) | |
3b23dd12 KC |
1368 | { |
1369 | return -EINVAL; | |
1370 | } | |
1371 | #endif | |
d78ab02c | 1372 | |
d612b1fd TH |
1373 | static long seccomp_get_action_avail(const char __user *uaction) |
1374 | { | |
1375 | u32 action; | |
1376 | ||
1377 | if (copy_from_user(&action, uaction, sizeof(action))) | |
1378 | return -EFAULT; | |
1379 | ||
1380 | switch (action) { | |
0466bdb9 | 1381 | case SECCOMP_RET_KILL_PROCESS: |
fd76875c | 1382 | case SECCOMP_RET_KILL_THREAD: |
d612b1fd TH |
1383 | case SECCOMP_RET_TRAP: |
1384 | case SECCOMP_RET_ERRNO: | |
6a21cc50 | 1385 | case SECCOMP_RET_USER_NOTIF: |
d612b1fd | 1386 | case SECCOMP_RET_TRACE: |
59f5cf44 | 1387 | case SECCOMP_RET_LOG: |
d612b1fd TH |
1388 | case SECCOMP_RET_ALLOW: |
1389 | break; | |
1390 | default: | |
1391 | return -EOPNOTSUPP; | |
1392 | } | |
1393 | ||
1394 | return 0; | |
1395 | } | |
1396 | ||
6a21cc50 TA |
1397 | static long seccomp_get_notif_sizes(void __user *usizes) |
1398 | { | |
1399 | struct seccomp_notif_sizes sizes = { | |
1400 | .seccomp_notif = sizeof(struct seccomp_notif), | |
1401 | .seccomp_notif_resp = sizeof(struct seccomp_notif_resp), | |
1402 | .seccomp_data = sizeof(struct seccomp_data), | |
1403 | }; | |
1404 | ||
1405 | if (copy_to_user(usizes, &sizes, sizeof(sizes))) | |
1406 | return -EFAULT; | |
1407 | ||
1408 | return 0; | |
1409 | } | |
1410 | ||
48dc92b9 KC |
1411 | /* Common entry point for both prctl and syscall. */ |
1412 | static long do_seccomp(unsigned int op, unsigned int flags, | |
a5662e4d | 1413 | void __user *uargs) |
48dc92b9 KC |
1414 | { |
1415 | switch (op) { | |
1416 | case SECCOMP_SET_MODE_STRICT: | |
1417 | if (flags != 0 || uargs != NULL) | |
1418 | return -EINVAL; | |
1419 | return seccomp_set_mode_strict(); | |
1420 | case SECCOMP_SET_MODE_FILTER: | |
1421 | return seccomp_set_mode_filter(flags, uargs); | |
d612b1fd TH |
1422 | case SECCOMP_GET_ACTION_AVAIL: |
1423 | if (flags != 0) | |
1424 | return -EINVAL; | |
1425 | ||
1426 | return seccomp_get_action_avail(uargs); | |
6a21cc50 TA |
1427 | case SECCOMP_GET_NOTIF_SIZES: |
1428 | if (flags != 0) | |
1429 | return -EINVAL; | |
1430 | ||
1431 | return seccomp_get_notif_sizes(uargs); | |
48dc92b9 KC |
1432 | default: |
1433 | return -EINVAL; | |
1434 | } | |
1435 | } | |
1436 | ||
1437 | SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, | |
a5662e4d | 1438 | void __user *, uargs) |
48dc92b9 KC |
1439 | { |
1440 | return do_seccomp(op, flags, uargs); | |
1441 | } | |
1442 | ||
d78ab02c KC |
1443 | /** |
1444 | * prctl_set_seccomp: configures current->seccomp.mode | |
1445 | * @seccomp_mode: requested mode to use | |
1446 | * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER | |
1447 | * | |
1448 | * Returns 0 on success or -EINVAL on failure. | |
1449 | */ | |
a5662e4d | 1450 | long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter) |
d78ab02c | 1451 | { |
48dc92b9 | 1452 | unsigned int op; |
a5662e4d | 1453 | void __user *uargs; |
48dc92b9 | 1454 | |
3b23dd12 KC |
1455 | switch (seccomp_mode) { |
1456 | case SECCOMP_MODE_STRICT: | |
48dc92b9 KC |
1457 | op = SECCOMP_SET_MODE_STRICT; |
1458 | /* | |
1459 | * Setting strict mode through prctl always ignored filter, | |
1460 | * so make sure it is always NULL here to pass the internal | |
1461 | * check in do_seccomp(). | |
1462 | */ | |
1463 | uargs = NULL; | |
1464 | break; | |
3b23dd12 | 1465 | case SECCOMP_MODE_FILTER: |
48dc92b9 KC |
1466 | op = SECCOMP_SET_MODE_FILTER; |
1467 | uargs = filter; | |
1468 | break; | |
3b23dd12 KC |
1469 | default: |
1470 | return -EINVAL; | |
1471 | } | |
48dc92b9 KC |
1472 | |
1473 | /* prctl interface doesn't have flags, so they are always zero. */ | |
1474 | return do_seccomp(op, 0, uargs); | |
d78ab02c | 1475 | } |
f8e529ed TA |
1476 | |
1477 | #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE) | |
f06eae83 TA |
1478 | static struct seccomp_filter *get_nth_filter(struct task_struct *task, |
1479 | unsigned long filter_off) | |
f8e529ed | 1480 | { |
f06eae83 TA |
1481 | struct seccomp_filter *orig, *filter; |
1482 | unsigned long count; | |
f8e529ed | 1483 | |
f06eae83 TA |
1484 | /* |
1485 | * Note: this is only correct because the caller should be the (ptrace) | |
1486 | * tracer of the task, otherwise lock_task_sighand is needed. | |
1487 | */ | |
f8e529ed | 1488 | spin_lock_irq(&task->sighand->siglock); |
f06eae83 | 1489 | |
f8e529ed | 1490 | if (task->seccomp.mode != SECCOMP_MODE_FILTER) { |
f06eae83 TA |
1491 | spin_unlock_irq(&task->sighand->siglock); |
1492 | return ERR_PTR(-EINVAL); | |
f8e529ed TA |
1493 | } |
1494 | ||
f06eae83 TA |
1495 | orig = task->seccomp.filter; |
1496 | __get_seccomp_filter(orig); | |
1497 | spin_unlock_irq(&task->sighand->siglock); | |
1498 | ||
1499 | count = 0; | |
1500 | for (filter = orig; filter; filter = filter->prev) | |
f8e529ed | 1501 | count++; |
f8e529ed TA |
1502 | |
1503 | if (filter_off >= count) { | |
f06eae83 | 1504 | filter = ERR_PTR(-ENOENT); |
f8e529ed TA |
1505 | goto out; |
1506 | } | |
f8e529ed | 1507 | |
f06eae83 TA |
1508 | count -= filter_off; |
1509 | for (filter = orig; filter && count > 1; filter = filter->prev) | |
f8e529ed | 1510 | count--; |
f8e529ed TA |
1511 | |
1512 | if (WARN_ON(count != 1 || !filter)) { | |
f06eae83 | 1513 | filter = ERR_PTR(-ENOENT); |
f8e529ed TA |
1514 | goto out; |
1515 | } | |
1516 | ||
f06eae83 TA |
1517 | __get_seccomp_filter(filter); |
1518 | ||
1519 | out: | |
1520 | __put_seccomp_filter(orig); | |
1521 | return filter; | |
1522 | } | |
1523 | ||
1524 | long seccomp_get_filter(struct task_struct *task, unsigned long filter_off, | |
1525 | void __user *data) | |
1526 | { | |
1527 | struct seccomp_filter *filter; | |
1528 | struct sock_fprog_kern *fprog; | |
1529 | long ret; | |
1530 | ||
1531 | if (!capable(CAP_SYS_ADMIN) || | |
1532 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { | |
1533 | return -EACCES; | |
1534 | } | |
1535 | ||
1536 | filter = get_nth_filter(task, filter_off); | |
1537 | if (IS_ERR(filter)) | |
1538 | return PTR_ERR(filter); | |
1539 | ||
f8e529ed TA |
1540 | fprog = filter->prog->orig_prog; |
1541 | if (!fprog) { | |
470bf1f2 | 1542 | /* This must be a new non-cBPF filter, since we save |
f8e529ed TA |
1543 | * every cBPF filter's orig_prog above when |
1544 | * CONFIG_CHECKPOINT_RESTORE is enabled. | |
1545 | */ | |
1546 | ret = -EMEDIUMTYPE; | |
1547 | goto out; | |
1548 | } | |
1549 | ||
1550 | ret = fprog->len; | |
1551 | if (!data) | |
1552 | goto out; | |
1553 | ||
f8e529ed TA |
1554 | if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog))) |
1555 | ret = -EFAULT; | |
1556 | ||
f8e529ed | 1557 | out: |
66a733ea | 1558 | __put_seccomp_filter(filter); |
f8e529ed | 1559 | return ret; |
f8e529ed | 1560 | } |
f8e529ed | 1561 | |
26500475 TA |
1562 | long seccomp_get_metadata(struct task_struct *task, |
1563 | unsigned long size, void __user *data) | |
1564 | { | |
1565 | long ret; | |
1566 | struct seccomp_filter *filter; | |
1567 | struct seccomp_metadata kmd = {}; | |
1568 | ||
1569 | if (!capable(CAP_SYS_ADMIN) || | |
1570 | current->seccomp.mode != SECCOMP_MODE_DISABLED) { | |
1571 | return -EACCES; | |
1572 | } | |
1573 | ||
1574 | size = min_t(unsigned long, size, sizeof(kmd)); | |
1575 | ||
63bb0045 TA |
1576 | if (size < sizeof(kmd.filter_off)) |
1577 | return -EINVAL; | |
1578 | ||
1579 | if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off))) | |
26500475 TA |
1580 | return -EFAULT; |
1581 | ||
1582 | filter = get_nth_filter(task, kmd.filter_off); | |
1583 | if (IS_ERR(filter)) | |
1584 | return PTR_ERR(filter); | |
1585 | ||
26500475 TA |
1586 | if (filter->log) |
1587 | kmd.flags |= SECCOMP_FILTER_FLAG_LOG; | |
1588 | ||
1589 | ret = size; | |
1590 | if (copy_to_user(data, &kmd, size)) | |
1591 | ret = -EFAULT; | |
1592 | ||
1593 | __put_seccomp_filter(filter); | |
f8e529ed TA |
1594 | return ret; |
1595 | } | |
1596 | #endif | |
8e5f1ad1 TH |
1597 | |
1598 | #ifdef CONFIG_SYSCTL | |
1599 | ||
1600 | /* Human readable action names for friendly sysctl interaction */ | |
0466bdb9 | 1601 | #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process" |
fd76875c | 1602 | #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread" |
8e5f1ad1 TH |
1603 | #define SECCOMP_RET_TRAP_NAME "trap" |
1604 | #define SECCOMP_RET_ERRNO_NAME "errno" | |
6a21cc50 | 1605 | #define SECCOMP_RET_USER_NOTIF_NAME "user_notif" |
8e5f1ad1 | 1606 | #define SECCOMP_RET_TRACE_NAME "trace" |
59f5cf44 | 1607 | #define SECCOMP_RET_LOG_NAME "log" |
8e5f1ad1 TH |
1608 | #define SECCOMP_RET_ALLOW_NAME "allow" |
1609 | ||
fd76875c | 1610 | static const char seccomp_actions_avail[] = |
0466bdb9 | 1611 | SECCOMP_RET_KILL_PROCESS_NAME " " |
fd76875c KC |
1612 | SECCOMP_RET_KILL_THREAD_NAME " " |
1613 | SECCOMP_RET_TRAP_NAME " " | |
1614 | SECCOMP_RET_ERRNO_NAME " " | |
6a21cc50 | 1615 | SECCOMP_RET_USER_NOTIF_NAME " " |
fd76875c KC |
1616 | SECCOMP_RET_TRACE_NAME " " |
1617 | SECCOMP_RET_LOG_NAME " " | |
1618 | SECCOMP_RET_ALLOW_NAME; | |
8e5f1ad1 | 1619 | |
0ddec0fc TH |
1620 | struct seccomp_log_name { |
1621 | u32 log; | |
1622 | const char *name; | |
1623 | }; | |
1624 | ||
1625 | static const struct seccomp_log_name seccomp_log_names[] = { | |
0466bdb9 | 1626 | { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME }, |
fd76875c | 1627 | { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME }, |
0ddec0fc TH |
1628 | { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME }, |
1629 | { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME }, | |
6a21cc50 | 1630 | { SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME }, |
0ddec0fc | 1631 | { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME }, |
59f5cf44 | 1632 | { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME }, |
0ddec0fc TH |
1633 | { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME }, |
1634 | { } | |
1635 | }; | |
1636 | ||
1637 | static bool seccomp_names_from_actions_logged(char *names, size_t size, | |
beb44aca TH |
1638 | u32 actions_logged, |
1639 | const char *sep) | |
0ddec0fc TH |
1640 | { |
1641 | const struct seccomp_log_name *cur; | |
beb44aca | 1642 | bool append_sep = false; |
0ddec0fc TH |
1643 | |
1644 | for (cur = seccomp_log_names; cur->name && size; cur++) { | |
1645 | ssize_t ret; | |
1646 | ||
1647 | if (!(actions_logged & cur->log)) | |
1648 | continue; | |
1649 | ||
beb44aca TH |
1650 | if (append_sep) { |
1651 | ret = strscpy(names, sep, size); | |
0ddec0fc TH |
1652 | if (ret < 0) |
1653 | return false; | |
1654 | ||
1655 | names += ret; | |
1656 | size -= ret; | |
1657 | } else | |
beb44aca | 1658 | append_sep = true; |
0ddec0fc TH |
1659 | |
1660 | ret = strscpy(names, cur->name, size); | |
1661 | if (ret < 0) | |
1662 | return false; | |
1663 | ||
1664 | names += ret; | |
1665 | size -= ret; | |
1666 | } | |
1667 | ||
1668 | return true; | |
1669 | } | |
1670 | ||
1671 | static bool seccomp_action_logged_from_name(u32 *action_logged, | |
1672 | const char *name) | |
1673 | { | |
1674 | const struct seccomp_log_name *cur; | |
1675 | ||
1676 | for (cur = seccomp_log_names; cur->name; cur++) { | |
1677 | if (!strcmp(cur->name, name)) { | |
1678 | *action_logged = cur->log; | |
1679 | return true; | |
1680 | } | |
1681 | } | |
1682 | ||
1683 | return false; | |
1684 | } | |
1685 | ||
1686 | static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names) | |
1687 | { | |
1688 | char *name; | |
1689 | ||
1690 | *actions_logged = 0; | |
1691 | while ((name = strsep(&names, " ")) && *name) { | |
1692 | u32 action_logged = 0; | |
1693 | ||
1694 | if (!seccomp_action_logged_from_name(&action_logged, name)) | |
1695 | return false; | |
1696 | ||
1697 | *actions_logged |= action_logged; | |
1698 | } | |
1699 | ||
1700 | return true; | |
1701 | } | |
1702 | ||
d013db02 TH |
1703 | static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer, |
1704 | size_t *lenp, loff_t *ppos) | |
0ddec0fc TH |
1705 | { |
1706 | char names[sizeof(seccomp_actions_avail)]; | |
1707 | struct ctl_table table; | |
d013db02 TH |
1708 | |
1709 | memset(names, 0, sizeof(names)); | |
1710 | ||
1711 | if (!seccomp_names_from_actions_logged(names, sizeof(names), | |
beb44aca | 1712 | seccomp_actions_logged, " ")) |
d013db02 TH |
1713 | return -EINVAL; |
1714 | ||
1715 | table = *ro_table; | |
1716 | table.data = names; | |
1717 | table.maxlen = sizeof(names); | |
1718 | return proc_dostring(&table, 0, buffer, lenp, ppos); | |
1719 | } | |
1720 | ||
1721 | static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer, | |
ea6eca77 | 1722 | size_t *lenp, loff_t *ppos, u32 *actions_logged) |
0ddec0fc TH |
1723 | { |
1724 | char names[sizeof(seccomp_actions_avail)]; | |
1725 | struct ctl_table table; | |
1726 | int ret; | |
1727 | ||
d013db02 | 1728 | if (!capable(CAP_SYS_ADMIN)) |
0ddec0fc TH |
1729 | return -EPERM; |
1730 | ||
1731 | memset(names, 0, sizeof(names)); | |
1732 | ||
0ddec0fc TH |
1733 | table = *ro_table; |
1734 | table.data = names; | |
1735 | table.maxlen = sizeof(names); | |
d013db02 | 1736 | ret = proc_dostring(&table, 1, buffer, lenp, ppos); |
0ddec0fc TH |
1737 | if (ret) |
1738 | return ret; | |
1739 | ||
ea6eca77 | 1740 | if (!seccomp_actions_logged_from_names(actions_logged, table.data)) |
d013db02 | 1741 | return -EINVAL; |
0ddec0fc | 1742 | |
ea6eca77 | 1743 | if (*actions_logged & SECCOMP_LOG_ALLOW) |
d013db02 | 1744 | return -EINVAL; |
0ddec0fc | 1745 | |
ea6eca77 | 1746 | seccomp_actions_logged = *actions_logged; |
0ddec0fc TH |
1747 | return 0; |
1748 | } | |
0ddec0fc | 1749 | |
ea6eca77 TH |
1750 | static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged, |
1751 | int ret) | |
1752 | { | |
1753 | char names[sizeof(seccomp_actions_avail)]; | |
1754 | char old_names[sizeof(seccomp_actions_avail)]; | |
1755 | const char *new = names; | |
1756 | const char *old = old_names; | |
0ddec0fc | 1757 | |
ea6eca77 TH |
1758 | if (!audit_enabled) |
1759 | return; | |
1760 | ||
1761 | memset(names, 0, sizeof(names)); | |
1762 | memset(old_names, 0, sizeof(old_names)); | |
1763 | ||
1764 | if (ret) | |
1765 | new = "?"; | |
1766 | else if (!actions_logged) | |
1767 | new = "(none)"; | |
1768 | else if (!seccomp_names_from_actions_logged(names, sizeof(names), | |
1769 | actions_logged, ",")) | |
1770 | new = "?"; | |
1771 | ||
1772 | if (!old_actions_logged) | |
1773 | old = "(none)"; | |
1774 | else if (!seccomp_names_from_actions_logged(old_names, | |
1775 | sizeof(old_names), | |
1776 | old_actions_logged, ",")) | |
1777 | old = "?"; | |
1778 | ||
1779 | return audit_seccomp_actions_logged(new, old, !ret); | |
1780 | } | |
1781 | ||
d013db02 | 1782 | static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write, |
32927393 | 1783 | void *buffer, size_t *lenp, |
d013db02 TH |
1784 | loff_t *ppos) |
1785 | { | |
ea6eca77 TH |
1786 | int ret; |
1787 | ||
1788 | if (write) { | |
1789 | u32 actions_logged = 0; | |
1790 | u32 old_actions_logged = seccomp_actions_logged; | |
1791 | ||
1792 | ret = write_actions_logged(ro_table, buffer, lenp, ppos, | |
1793 | &actions_logged); | |
1794 | audit_actions_logged(actions_logged, old_actions_logged, ret); | |
1795 | } else | |
1796 | ret = read_actions_logged(ro_table, buffer, lenp, ppos); | |
1797 | ||
1798 | return ret; | |
0ddec0fc TH |
1799 | } |
1800 | ||
8e5f1ad1 TH |
1801 | static struct ctl_path seccomp_sysctl_path[] = { |
1802 | { .procname = "kernel", }, | |
1803 | { .procname = "seccomp", }, | |
1804 | { } | |
1805 | }; | |
1806 | ||
1807 | static struct ctl_table seccomp_sysctl_table[] = { | |
1808 | { | |
1809 | .procname = "actions_avail", | |
1810 | .data = (void *) &seccomp_actions_avail, | |
1811 | .maxlen = sizeof(seccomp_actions_avail), | |
1812 | .mode = 0444, | |
1813 | .proc_handler = proc_dostring, | |
1814 | }, | |
0ddec0fc TH |
1815 | { |
1816 | .procname = "actions_logged", | |
1817 | .mode = 0644, | |
1818 | .proc_handler = seccomp_actions_logged_handler, | |
1819 | }, | |
8e5f1ad1 TH |
1820 | { } |
1821 | }; | |
1822 | ||
1823 | static int __init seccomp_sysctl_init(void) | |
1824 | { | |
1825 | struct ctl_table_header *hdr; | |
1826 | ||
1827 | hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table); | |
1828 | if (!hdr) | |
1829 | pr_warn("seccomp: sysctl registration failed\n"); | |
1830 | else | |
1831 | kmemleak_not_leak(hdr); | |
1832 | ||
1833 | return 0; | |
1834 | } | |
1835 | ||
1836 | device_initcall(seccomp_sysctl_init) | |
1837 | ||
1838 | #endif /* CONFIG_SYSCTL */ |