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