1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/kernel/seccomp.c 4 * 5 * Copyright 2004-2005 Andrea Arcangeli <[email protected]> 6 * 7 * Copyright (C) 2012 Google, Inc. 8 * Will Drewry <[email protected]> 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 #define pr_fmt(fmt) "seccomp: " fmt 17 18 #include <linux/refcount.h> 19 #include <linux/audit.h> 20 #include <linux/compat.h> 21 #include <linux/coredump.h> 22 #include <linux/kmemleak.h> 23 #include <linux/nospec.h> 24 #include <linux/prctl.h> 25 #include <linux/sched.h> 26 #include <linux/sched/task_stack.h> 27 #include <linux/seccomp.h> 28 #include <linux/slab.h> 29 #include <linux/syscalls.h> 30 #include <linux/sysctl.h> 31 32 #include <asm/syscall.h> 33 34 /* Not exposed in headers: strictly internal use only. */ 35 #define SECCOMP_MODE_DEAD (SECCOMP_MODE_FILTER + 1) 36 37 #ifdef CONFIG_SECCOMP_FILTER 38 #include <linux/file.h> 39 #include <linux/filter.h> 40 #include <linux/pid.h> 41 #include <linux/ptrace.h> 42 #include <linux/capability.h> 43 #include <linux/uaccess.h> 44 #include <linux/anon_inodes.h> 45 #include <linux/lockdep.h> 46 47 /* 48 * When SECCOMP_IOCTL_NOTIF_ID_VALID was first introduced, it had the 49 * wrong direction flag in the ioctl number. This is the broken one, 50 * which the kernel needs to keep supporting until all userspaces stop 51 * using the wrong command number. 52 */ 53 #define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR SECCOMP_IOR(2, __u64) 54 55 enum notify_state { 56 SECCOMP_NOTIFY_INIT, 57 SECCOMP_NOTIFY_SENT, 58 SECCOMP_NOTIFY_REPLIED, 59 }; 60 61 struct seccomp_knotif { 62 /* The struct pid of the task whose filter triggered the notification */ 63 struct task_struct *task; 64 65 /* The "cookie" for this request; this is unique for this filter. */ 66 u64 id; 67 68 /* 69 * The seccomp data. This pointer is valid the entire time this 70 * notification is active, since it comes from __seccomp_filter which 71 * eclipses the entire lifecycle here. 72 */ 73 const struct seccomp_data *data; 74 75 /* 76 * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a 77 * struct seccomp_knotif is created and starts out in INIT. Once the 78 * handler reads the notification off of an FD, it transitions to SENT. 79 * If a signal is received the state transitions back to INIT and 80 * another message is sent. When the userspace handler replies, state 81 * transitions to REPLIED. 82 */ 83 enum notify_state state; 84 85 /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */ 86 int error; 87 long val; 88 u32 flags; 89 90 /* 91 * Signals when this has changed states, such as the listener 92 * dying, a new seccomp addfd message, or changing to REPLIED 93 */ 94 struct completion ready; 95 96 struct list_head list; 97 98 /* outstanding addfd requests */ 99 struct list_head addfd; 100 }; 101 102 /** 103 * struct seccomp_kaddfd - container for seccomp_addfd ioctl messages 104 * 105 * @file: A reference to the file to install in the other task 106 * @fd: The fd number to install it at. If the fd number is -1, it means the 107 * installing process should allocate the fd as normal. 108 * @flags: The flags for the new file descriptor. At the moment, only O_CLOEXEC 109 * is allowed. 110 * @ioctl_flags: The flags used for the seccomp_addfd ioctl. 111 * @setfd: whether or not SECCOMP_ADDFD_FLAG_SETFD was set during notify_addfd 112 * @ret: The return value of the installing process. It is set to the fd num 113 * upon success (>= 0). 114 * @completion: Indicates that the installing process has completed fd 115 * installation, or gone away (either due to successful 116 * reply, or signal) 117 * @list: list_head for chaining seccomp_kaddfd together. 118 * 119 */ 120 struct seccomp_kaddfd { 121 struct file *file; 122 int fd; 123 unsigned int flags; 124 __u32 ioctl_flags; 125 126 union { 127 bool setfd; 128 /* To only be set on reply */ 129 int ret; 130 }; 131 struct completion completion; 132 struct list_head list; 133 }; 134 135 /** 136 * struct notification - container for seccomp userspace notifications. Since 137 * most seccomp filters will not have notification listeners attached and this 138 * structure is fairly large, we store the notification-specific stuff in a 139 * separate structure. 140 * 141 * @requests: A semaphore that users of this notification can wait on for 142 * changes. Actual reads and writes are still controlled with 143 * filter->notify_lock. 144 * @flags: A set of SECCOMP_USER_NOTIF_FD_* flags. 145 * @next_id: The id of the next request. 146 * @notifications: A list of struct seccomp_knotif elements. 147 */ 148 149 struct notification { 150 atomic_t requests; 151 u32 flags; 152 u64 next_id; 153 struct list_head notifications; 154 }; 155 156 #ifdef SECCOMP_ARCH_NATIVE 157 /** 158 * struct action_cache - per-filter cache of seccomp actions per 159 * arch/syscall pair 160 * 161 * @allow_native: A bitmap where each bit represents whether the 162 * filter will always allow the syscall, for the 163 * native architecture. 164 * @allow_compat: A bitmap where each bit represents whether the 165 * filter will always allow the syscall, for the 166 * compat architecture. 167 */ 168 struct action_cache { 169 DECLARE_BITMAP(allow_native, SECCOMP_ARCH_NATIVE_NR); 170 #ifdef SECCOMP_ARCH_COMPAT 171 DECLARE_BITMAP(allow_compat, SECCOMP_ARCH_COMPAT_NR); 172 #endif 173 }; 174 #else 175 struct action_cache { }; 176 177 static inline bool seccomp_cache_check_allow(const struct seccomp_filter *sfilter, 178 const struct seccomp_data *sd) 179 { 180 return false; 181 } 182 183 static inline void seccomp_cache_prepare(struct seccomp_filter *sfilter) 184 { 185 } 186 #endif /* SECCOMP_ARCH_NATIVE */ 187 188 /** 189 * struct seccomp_filter - container for seccomp BPF programs 190 * 191 * @refs: Reference count to manage the object lifetime. 192 * A filter's reference count is incremented for each directly 193 * attached task, once for the dependent filter, and if 194 * requested for the user notifier. When @refs reaches zero, 195 * the filter can be freed. 196 * @users: A filter's @users count is incremented for each directly 197 * attached task (filter installation, fork(), thread_sync), 198 * and once for the dependent filter (tracked in filter->prev). 199 * When it reaches zero it indicates that no direct or indirect 200 * users of that filter exist. No new tasks can get associated with 201 * this filter after reaching 0. The @users count is always smaller 202 * or equal to @refs. Hence, reaching 0 for @users does not mean 203 * the filter can be freed. 204 * @cache: cache of arch/syscall mappings to actions 205 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged 206 * @wait_killable_recv: Put notifying process in killable state once the 207 * notification is received by the userspace listener. 208 * @prev: points to a previously installed, or inherited, filter 209 * @prog: the BPF program to evaluate 210 * @notif: the struct that holds all notification related information 211 * @notify_lock: A lock for all notification-related accesses. 212 * @wqh: A wait queue for poll if a notifier is in use. 213 * 214 * seccomp_filter objects are organized in a tree linked via the @prev 215 * pointer. For any task, it appears to be a singly-linked list starting 216 * with current->seccomp.filter, the most recently attached or inherited filter. 217 * However, multiple filters may share a @prev node, by way of fork(), which 218 * results in a unidirectional tree existing in memory. This is similar to 219 * how namespaces work. 220 * 221 * seccomp_filter objects should never be modified after being attached 222 * to a task_struct (other than @refs). 223 */ 224 struct seccomp_filter { 225 refcount_t refs; 226 refcount_t users; 227 bool log; 228 bool wait_killable_recv; 229 struct action_cache cache; 230 struct seccomp_filter *prev; 231 struct bpf_prog *prog; 232 struct notification *notif; 233 struct mutex notify_lock; 234 wait_queue_head_t wqh; 235 }; 236 237 /* Limit any path through the tree to 256KB worth of instructions. */ 238 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter)) 239 240 /* 241 * Endianness is explicitly ignored and left for BPF program authors to manage 242 * as per the specific architecture. 243 */ 244 static void populate_seccomp_data(struct seccomp_data *sd) 245 { 246 /* 247 * Instead of using current_pt_reg(), we're already doing the work 248 * to safely fetch "current", so just use "task" everywhere below. 249 */ 250 struct task_struct *task = current; 251 struct pt_regs *regs = task_pt_regs(task); 252 unsigned long args[6]; 253 254 sd->nr = syscall_get_nr(task, regs); 255 sd->arch = syscall_get_arch(task); 256 syscall_get_arguments(task, regs, args); 257 sd->args[0] = args[0]; 258 sd->args[1] = args[1]; 259 sd->args[2] = args[2]; 260 sd->args[3] = args[3]; 261 sd->args[4] = args[4]; 262 sd->args[5] = args[5]; 263 sd->instruction_pointer = KSTK_EIP(task); 264 } 265 266 /** 267 * seccomp_check_filter - verify seccomp filter code 268 * @filter: filter to verify 269 * @flen: length of filter 270 * 271 * Takes a previously checked filter (by bpf_check_classic) and 272 * redirects all filter code that loads struct sk_buff data 273 * and related data through seccomp_bpf_load. It also 274 * enforces length and alignment checking of those loads. 275 * 276 * Returns 0 if the rule set is legal or -EINVAL if not. 277 */ 278 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen) 279 { 280 int pc; 281 for (pc = 0; pc < flen; pc++) { 282 struct sock_filter *ftest = &filter[pc]; 283 u16 code = ftest->code; 284 u32 k = ftest->k; 285 286 switch (code) { 287 case BPF_LD | BPF_W | BPF_ABS: 288 ftest->code = BPF_LDX | BPF_W | BPF_ABS; 289 /* 32-bit aligned and not out of bounds. */ 290 if (k >= sizeof(struct seccomp_data) || k & 3) 291 return -EINVAL; 292 continue; 293 case BPF_LD | BPF_W | BPF_LEN: 294 ftest->code = BPF_LD | BPF_IMM; 295 ftest->k = sizeof(struct seccomp_data); 296 continue; 297 case BPF_LDX | BPF_W | BPF_LEN: 298 ftest->code = BPF_LDX | BPF_IMM; 299 ftest->k = sizeof(struct seccomp_data); 300 continue; 301 /* Explicitly include allowed calls. */ 302 case BPF_RET | BPF_K: 303 case BPF_RET | BPF_A: 304 case BPF_ALU | BPF_ADD | BPF_K: 305 case BPF_ALU | BPF_ADD | BPF_X: 306 case BPF_ALU | BPF_SUB | BPF_K: 307 case BPF_ALU | BPF_SUB | BPF_X: 308 case BPF_ALU | BPF_MUL | BPF_K: 309 case BPF_ALU | BPF_MUL | BPF_X: 310 case BPF_ALU | BPF_DIV | BPF_K: 311 case BPF_ALU | BPF_DIV | BPF_X: 312 case BPF_ALU | BPF_AND | BPF_K: 313 case BPF_ALU | BPF_AND | BPF_X: 314 case BPF_ALU | BPF_OR | BPF_K: 315 case BPF_ALU | BPF_OR | BPF_X: 316 case BPF_ALU | BPF_XOR | BPF_K: 317 case BPF_ALU | BPF_XOR | BPF_X: 318 case BPF_ALU | BPF_LSH | BPF_K: 319 case BPF_ALU | BPF_LSH | BPF_X: 320 case BPF_ALU | BPF_RSH | BPF_K: 321 case BPF_ALU | BPF_RSH | BPF_X: 322 case BPF_ALU | BPF_NEG: 323 case BPF_LD | BPF_IMM: 324 case BPF_LDX | BPF_IMM: 325 case BPF_MISC | BPF_TAX: 326 case BPF_MISC | BPF_TXA: 327 case BPF_LD | BPF_MEM: 328 case BPF_LDX | BPF_MEM: 329 case BPF_ST: 330 case BPF_STX: 331 case BPF_JMP | BPF_JA: 332 case BPF_JMP | BPF_JEQ | BPF_K: 333 case BPF_JMP | BPF_JEQ | BPF_X: 334 case BPF_JMP | BPF_JGE | BPF_K: 335 case BPF_JMP | BPF_JGE | BPF_X: 336 case BPF_JMP | BPF_JGT | BPF_K: 337 case BPF_JMP | BPF_JGT | BPF_X: 338 case BPF_JMP | BPF_JSET | BPF_K: 339 case BPF_JMP | BPF_JSET | BPF_X: 340 continue; 341 default: 342 return -EINVAL; 343 } 344 } 345 return 0; 346 } 347 348 #ifdef SECCOMP_ARCH_NATIVE 349 static inline bool seccomp_cache_check_allow_bitmap(const void *bitmap, 350 size_t bitmap_size, 351 int syscall_nr) 352 { 353 if (unlikely(syscall_nr < 0 || syscall_nr >= bitmap_size)) 354 return false; 355 syscall_nr = array_index_nospec(syscall_nr, bitmap_size); 356 357 return test_bit(syscall_nr, bitmap); 358 } 359 360 /** 361 * seccomp_cache_check_allow - lookup seccomp cache 362 * @sfilter: The seccomp filter 363 * @sd: The seccomp data to lookup the cache with 364 * 365 * Returns true if the seccomp_data is cached and allowed. 366 */ 367 static inline bool seccomp_cache_check_allow(const struct seccomp_filter *sfilter, 368 const struct seccomp_data *sd) 369 { 370 int syscall_nr = sd->nr; 371 const struct action_cache *cache = &sfilter->cache; 372 373 #ifndef SECCOMP_ARCH_COMPAT 374 /* A native-only architecture doesn't need to check sd->arch. */ 375 return seccomp_cache_check_allow_bitmap(cache->allow_native, 376 SECCOMP_ARCH_NATIVE_NR, 377 syscall_nr); 378 #else 379 if (likely(sd->arch == SECCOMP_ARCH_NATIVE)) 380 return seccomp_cache_check_allow_bitmap(cache->allow_native, 381 SECCOMP_ARCH_NATIVE_NR, 382 syscall_nr); 383 if (likely(sd->arch == SECCOMP_ARCH_COMPAT)) 384 return seccomp_cache_check_allow_bitmap(cache->allow_compat, 385 SECCOMP_ARCH_COMPAT_NR, 386 syscall_nr); 387 #endif /* SECCOMP_ARCH_COMPAT */ 388 389 WARN_ON_ONCE(true); 390 return false; 391 } 392 #endif /* SECCOMP_ARCH_NATIVE */ 393 394 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL))) 395 /** 396 * seccomp_run_filters - evaluates all seccomp filters against @sd 397 * @sd: optional seccomp data to be passed to filters 398 * @match: stores struct seccomp_filter that resulted in the return value, 399 * unless filter returned SECCOMP_RET_ALLOW, in which case it will 400 * be unchanged. 401 * 402 * Returns valid seccomp BPF response codes. 403 */ 404 static u32 seccomp_run_filters(const struct seccomp_data *sd, 405 struct seccomp_filter **match) 406 { 407 u32 ret = SECCOMP_RET_ALLOW; 408 /* Make sure cross-thread synced filter points somewhere sane. */ 409 struct seccomp_filter *f = 410 READ_ONCE(current->seccomp.filter); 411 412 /* Ensure unexpected behavior doesn't result in failing open. */ 413 if (WARN_ON(f == NULL)) 414 return SECCOMP_RET_KILL_PROCESS; 415 416 if (seccomp_cache_check_allow(f, sd)) 417 return SECCOMP_RET_ALLOW; 418 419 /* 420 * All filters in the list are evaluated and the lowest BPF return 421 * value always takes priority (ignoring the DATA). 422 */ 423 for (; f; f = f->prev) { 424 u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd); 425 426 if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) { 427 ret = cur_ret; 428 *match = f; 429 } 430 } 431 return ret; 432 } 433 #endif /* CONFIG_SECCOMP_FILTER */ 434 435 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode) 436 { 437 assert_spin_locked(¤t->sighand->siglock); 438 439 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode) 440 return false; 441 442 return true; 443 } 444 445 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { } 446 447 static inline void seccomp_assign_mode(struct task_struct *task, 448 unsigned long seccomp_mode, 449 unsigned long flags) 450 { 451 assert_spin_locked(&task->sighand->siglock); 452 453 task->seccomp.mode = seccomp_mode; 454 /* 455 * Make sure SYSCALL_WORK_SECCOMP cannot be set before the mode (and 456 * filter) is set. 457 */ 458 smp_mb__before_atomic(); 459 /* Assume default seccomp processes want spec flaw mitigation. */ 460 if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0) 461 arch_seccomp_spec_mitigate(task); 462 set_task_syscall_work(task, SECCOMP); 463 } 464 465 #ifdef CONFIG_SECCOMP_FILTER 466 /* Returns 1 if the parent is an ancestor of the child. */ 467 static int is_ancestor(struct seccomp_filter *parent, 468 struct seccomp_filter *child) 469 { 470 /* NULL is the root ancestor. */ 471 if (parent == NULL) 472 return 1; 473 for (; child; child = child->prev) 474 if (child == parent) 475 return 1; 476 return 0; 477 } 478 479 /** 480 * seccomp_can_sync_threads: checks if all threads can be synchronized 481 * 482 * Expects sighand and cred_guard_mutex locks to be held. 483 * 484 * Returns 0 on success, -ve on error, or the pid of a thread which was 485 * either not in the correct seccomp mode or did not have an ancestral 486 * seccomp filter. 487 */ 488 static inline pid_t seccomp_can_sync_threads(void) 489 { 490 struct task_struct *thread, *caller; 491 492 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); 493 assert_spin_locked(¤t->sighand->siglock); 494 495 /* Validate all threads being eligible for synchronization. */ 496 caller = current; 497 for_each_thread(caller, thread) { 498 pid_t failed; 499 500 /* Skip current, since it is initiating the sync. */ 501 if (thread == caller) 502 continue; 503 /* Skip exited threads. */ 504 if (thread->flags & PF_EXITING) 505 continue; 506 507 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED || 508 (thread->seccomp.mode == SECCOMP_MODE_FILTER && 509 is_ancestor(thread->seccomp.filter, 510 caller->seccomp.filter))) 511 continue; 512 513 /* Return the first thread that cannot be synchronized. */ 514 failed = task_pid_vnr(thread); 515 /* If the pid cannot be resolved, then return -ESRCH */ 516 if (WARN_ON(failed == 0)) 517 failed = -ESRCH; 518 return failed; 519 } 520 521 return 0; 522 } 523 524 static inline void seccomp_filter_free(struct seccomp_filter *filter) 525 { 526 if (filter) { 527 bpf_prog_destroy(filter->prog); 528 kfree(filter); 529 } 530 } 531 532 static void __seccomp_filter_orphan(struct seccomp_filter *orig) 533 { 534 while (orig && refcount_dec_and_test(&orig->users)) { 535 if (waitqueue_active(&orig->wqh)) 536 wake_up_poll(&orig->wqh, EPOLLHUP); 537 orig = orig->prev; 538 } 539 } 540 541 static void __put_seccomp_filter(struct seccomp_filter *orig) 542 { 543 /* Clean up single-reference branches iteratively. */ 544 while (orig && refcount_dec_and_test(&orig->refs)) { 545 struct seccomp_filter *freeme = orig; 546 orig = orig->prev; 547 seccomp_filter_free(freeme); 548 } 549 } 550 551 static void __seccomp_filter_release(struct seccomp_filter *orig) 552 { 553 /* Notify about any unused filters in the task's former filter tree. */ 554 __seccomp_filter_orphan(orig); 555 /* Finally drop all references to the task's former tree. */ 556 __put_seccomp_filter(orig); 557 } 558 559 /** 560 * seccomp_filter_release - Detach the task from its filter tree, 561 * drop its reference count, and notify 562 * about unused filters 563 * 564 * @tsk: task the filter should be released from. 565 * 566 * This function should only be called when the task is exiting as 567 * it detaches it from its filter tree. PF_EXITING has to be set 568 * for the task. 569 */ 570 void seccomp_filter_release(struct task_struct *tsk) 571 { 572 struct seccomp_filter *orig; 573 574 if (WARN_ON((tsk->flags & PF_EXITING) == 0)) 575 return; 576 577 spin_lock_irq(&tsk->sighand->siglock); 578 orig = tsk->seccomp.filter; 579 /* Detach task from its filter tree. */ 580 tsk->seccomp.filter = NULL; 581 spin_unlock_irq(&tsk->sighand->siglock); 582 __seccomp_filter_release(orig); 583 } 584 585 /** 586 * seccomp_sync_threads: sets all threads to use current's filter 587 * 588 * @flags: SECCOMP_FILTER_FLAG_* flags to set during sync. 589 * 590 * Expects sighand and cred_guard_mutex locks to be held, and for 591 * seccomp_can_sync_threads() to have returned success already 592 * without dropping the locks. 593 * 594 */ 595 static inline void seccomp_sync_threads(unsigned long flags) 596 { 597 struct task_struct *thread, *caller; 598 599 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex)); 600 assert_spin_locked(¤t->sighand->siglock); 601 602 /* Synchronize all threads. */ 603 caller = current; 604 for_each_thread(caller, thread) { 605 /* Skip current, since it needs no changes. */ 606 if (thread == caller) 607 continue; 608 609 /* 610 * Skip exited threads. seccomp_filter_release could have 611 * been already called for this task. 612 */ 613 if (thread->flags & PF_EXITING) 614 continue; 615 616 /* Get a task reference for the new leaf node. */ 617 get_seccomp_filter(caller); 618 619 /* 620 * Drop the task reference to the shared ancestor since 621 * current's path will hold a reference. (This also 622 * allows a put before the assignment.) 623 */ 624 __seccomp_filter_release(thread->seccomp.filter); 625 626 /* Make our new filter tree visible. */ 627 smp_store_release(&thread->seccomp.filter, 628 caller->seccomp.filter); 629 atomic_set(&thread->seccomp.filter_count, 630 atomic_read(&caller->seccomp.filter_count)); 631 632 /* 633 * Don't let an unprivileged task work around 634 * the no_new_privs restriction by creating 635 * a thread that sets it up, enters seccomp, 636 * then dies. 637 */ 638 if (task_no_new_privs(caller)) 639 task_set_no_new_privs(thread); 640 641 /* 642 * Opt the other thread into seccomp if needed. 643 * As threads are considered to be trust-realm 644 * equivalent (see ptrace_may_access), it is safe to 645 * allow one thread to transition the other. 646 */ 647 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) 648 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER, 649 flags); 650 } 651 } 652 653 /** 654 * seccomp_prepare_filter: Prepares a seccomp filter for use. 655 * @fprog: BPF program to install 656 * 657 * Returns filter on success or an ERR_PTR on failure. 658 */ 659 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog) 660 { 661 struct seccomp_filter *sfilter; 662 int ret; 663 const bool save_orig = 664 #if defined(CONFIG_CHECKPOINT_RESTORE) || defined(SECCOMP_ARCH_NATIVE) 665 true; 666 #else 667 false; 668 #endif 669 670 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS) 671 return ERR_PTR(-EINVAL); 672 673 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter)); 674 675 /* 676 * Installing a seccomp filter requires that the task has 677 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs. 678 * This avoids scenarios where unprivileged tasks can affect the 679 * behavior of privileged children. 680 */ 681 if (!task_no_new_privs(current) && 682 !ns_capable_noaudit(current_user_ns(), CAP_SYS_ADMIN)) 683 return ERR_PTR(-EACCES); 684 685 /* Allocate a new seccomp_filter */ 686 sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN); 687 if (!sfilter) 688 return ERR_PTR(-ENOMEM); 689 690 mutex_init(&sfilter->notify_lock); 691 ret = bpf_prog_create_from_user(&sfilter->prog, fprog, 692 seccomp_check_filter, save_orig); 693 if (ret < 0) { 694 kfree(sfilter); 695 return ERR_PTR(ret); 696 } 697 698 refcount_set(&sfilter->refs, 1); 699 refcount_set(&sfilter->users, 1); 700 init_waitqueue_head(&sfilter->wqh); 701 702 return sfilter; 703 } 704 705 /** 706 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog 707 * @user_filter: pointer to the user data containing a sock_fprog. 708 * 709 * Returns 0 on success and non-zero otherwise. 710 */ 711 static struct seccomp_filter * 712 seccomp_prepare_user_filter(const char __user *user_filter) 713 { 714 struct sock_fprog fprog; 715 struct seccomp_filter *filter = ERR_PTR(-EFAULT); 716 717 #ifdef CONFIG_COMPAT 718 if (in_compat_syscall()) { 719 struct compat_sock_fprog fprog32; 720 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32))) 721 goto out; 722 fprog.len = fprog32.len; 723 fprog.filter = compat_ptr(fprog32.filter); 724 } else /* falls through to the if below. */ 725 #endif 726 if (copy_from_user(&fprog, user_filter, sizeof(fprog))) 727 goto out; 728 filter = seccomp_prepare_filter(&fprog); 729 out: 730 return filter; 731 } 732 733 #ifdef SECCOMP_ARCH_NATIVE 734 /** 735 * seccomp_is_const_allow - check if filter is constant allow with given data 736 * @fprog: The BPF programs 737 * @sd: The seccomp data to check against, only syscall number and arch 738 * number are considered constant. 739 */ 740 static bool seccomp_is_const_allow(struct sock_fprog_kern *fprog, 741 struct seccomp_data *sd) 742 { 743 unsigned int reg_value = 0; 744 unsigned int pc; 745 bool op_res; 746 747 if (WARN_ON_ONCE(!fprog)) 748 return false; 749 750 /* Our single exception to filtering. */ 751 #ifdef __NR_uretprobe 752 #ifdef SECCOMP_ARCH_COMPAT 753 if (sd->arch == SECCOMP_ARCH_NATIVE) 754 #endif 755 if (sd->nr == __NR_uretprobe) 756 return true; 757 #endif 758 759 for (pc = 0; pc < fprog->len; pc++) { 760 struct sock_filter *insn = &fprog->filter[pc]; 761 u16 code = insn->code; 762 u32 k = insn->k; 763 764 switch (code) { 765 case BPF_LD | BPF_W | BPF_ABS: 766 switch (k) { 767 case offsetof(struct seccomp_data, nr): 768 reg_value = sd->nr; 769 break; 770 case offsetof(struct seccomp_data, arch): 771 reg_value = sd->arch; 772 break; 773 default: 774 /* can't optimize (non-constant value load) */ 775 return false; 776 } 777 break; 778 case BPF_RET | BPF_K: 779 /* reached return with constant values only, check allow */ 780 return k == SECCOMP_RET_ALLOW; 781 case BPF_JMP | BPF_JA: 782 pc += insn->k; 783 break; 784 case BPF_JMP | BPF_JEQ | BPF_K: 785 case BPF_JMP | BPF_JGE | BPF_K: 786 case BPF_JMP | BPF_JGT | BPF_K: 787 case BPF_JMP | BPF_JSET | BPF_K: 788 switch (BPF_OP(code)) { 789 case BPF_JEQ: 790 op_res = reg_value == k; 791 break; 792 case BPF_JGE: 793 op_res = reg_value >= k; 794 break; 795 case BPF_JGT: 796 op_res = reg_value > k; 797 break; 798 case BPF_JSET: 799 op_res = !!(reg_value & k); 800 break; 801 default: 802 /* can't optimize (unknown jump) */ 803 return false; 804 } 805 806 pc += op_res ? insn->jt : insn->jf; 807 break; 808 case BPF_ALU | BPF_AND | BPF_K: 809 reg_value &= k; 810 break; 811 default: 812 /* can't optimize (unknown insn) */ 813 return false; 814 } 815 } 816 817 /* ran off the end of the filter?! */ 818 WARN_ON(1); 819 return false; 820 } 821 822 static void seccomp_cache_prepare_bitmap(struct seccomp_filter *sfilter, 823 void *bitmap, const void *bitmap_prev, 824 size_t bitmap_size, int arch) 825 { 826 struct sock_fprog_kern *fprog = sfilter->prog->orig_prog; 827 struct seccomp_data sd; 828 int nr; 829 830 if (bitmap_prev) { 831 /* The new filter must be as restrictive as the last. */ 832 bitmap_copy(bitmap, bitmap_prev, bitmap_size); 833 } else { 834 /* Before any filters, all syscalls are always allowed. */ 835 bitmap_fill(bitmap, bitmap_size); 836 } 837 838 for (nr = 0; nr < bitmap_size; nr++) { 839 /* No bitmap change: not a cacheable action. */ 840 if (!test_bit(nr, bitmap)) 841 continue; 842 843 sd.nr = nr; 844 sd.arch = arch; 845 846 /* No bitmap change: continue to always allow. */ 847 if (seccomp_is_const_allow(fprog, &sd)) 848 continue; 849 850 /* 851 * Not a cacheable action: always run filters. 852 * atomic clear_bit() not needed, filter not visible yet. 853 */ 854 __clear_bit(nr, bitmap); 855 } 856 } 857 858 /** 859 * seccomp_cache_prepare - emulate the filter to find cacheable syscalls 860 * @sfilter: The seccomp filter 861 * 862 * Returns 0 if successful or -errno if error occurred. 863 */ 864 static void seccomp_cache_prepare(struct seccomp_filter *sfilter) 865 { 866 struct action_cache *cache = &sfilter->cache; 867 const struct action_cache *cache_prev = 868 sfilter->prev ? &sfilter->prev->cache : NULL; 869 870 seccomp_cache_prepare_bitmap(sfilter, cache->allow_native, 871 cache_prev ? cache_prev->allow_native : NULL, 872 SECCOMP_ARCH_NATIVE_NR, 873 SECCOMP_ARCH_NATIVE); 874 875 #ifdef SECCOMP_ARCH_COMPAT 876 seccomp_cache_prepare_bitmap(sfilter, cache->allow_compat, 877 cache_prev ? cache_prev->allow_compat : NULL, 878 SECCOMP_ARCH_COMPAT_NR, 879 SECCOMP_ARCH_COMPAT); 880 #endif /* SECCOMP_ARCH_COMPAT */ 881 } 882 #endif /* SECCOMP_ARCH_NATIVE */ 883 884 /** 885 * seccomp_attach_filter: validate and attach filter 886 * @flags: flags to change filter behavior 887 * @filter: seccomp filter to add to the current process 888 * 889 * Caller must be holding current->sighand->siglock lock. 890 * 891 * Returns 0 on success, -ve on error, or 892 * - in TSYNC mode: the pid of a thread which was either not in the correct 893 * seccomp mode or did not have an ancestral seccomp filter 894 * - in NEW_LISTENER mode: the fd of the new listener 895 */ 896 static long seccomp_attach_filter(unsigned int flags, 897 struct seccomp_filter *filter) 898 { 899 unsigned long total_insns; 900 struct seccomp_filter *walker; 901 902 assert_spin_locked(¤t->sighand->siglock); 903 904 /* Validate resulting filter length. */ 905 total_insns = filter->prog->len; 906 for (walker = current->seccomp.filter; walker; walker = walker->prev) 907 total_insns += walker->prog->len + 4; /* 4 instr penalty */ 908 if (total_insns > MAX_INSNS_PER_PATH) 909 return -ENOMEM; 910 911 /* If thread sync has been requested, check that it is possible. */ 912 if (flags & SECCOMP_FILTER_FLAG_TSYNC) { 913 int ret; 914 915 ret = seccomp_can_sync_threads(); 916 if (ret) { 917 if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) 918 return -ESRCH; 919 else 920 return ret; 921 } 922 } 923 924 /* Set log flag, if present. */ 925 if (flags & SECCOMP_FILTER_FLAG_LOG) 926 filter->log = true; 927 928 /* Set wait killable flag, if present. */ 929 if (flags & SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV) 930 filter->wait_killable_recv = true; 931 932 /* 933 * If there is an existing filter, make it the prev and don't drop its 934 * task reference. 935 */ 936 filter->prev = current->seccomp.filter; 937 seccomp_cache_prepare(filter); 938 current->seccomp.filter = filter; 939 atomic_inc(¤t->seccomp.filter_count); 940 941 /* Now that the new filter is in place, synchronize to all threads. */ 942 if (flags & SECCOMP_FILTER_FLAG_TSYNC) 943 seccomp_sync_threads(flags); 944 945 return 0; 946 } 947 948 static void __get_seccomp_filter(struct seccomp_filter *filter) 949 { 950 refcount_inc(&filter->refs); 951 } 952 953 /* get_seccomp_filter - increments the reference count of the filter on @tsk */ 954 void get_seccomp_filter(struct task_struct *tsk) 955 { 956 struct seccomp_filter *orig = tsk->seccomp.filter; 957 if (!orig) 958 return; 959 __get_seccomp_filter(orig); 960 refcount_inc(&orig->users); 961 } 962 963 #endif /* CONFIG_SECCOMP_FILTER */ 964 965 /* For use with seccomp_actions_logged */ 966 #define SECCOMP_LOG_KILL_PROCESS (1 << 0) 967 #define SECCOMP_LOG_KILL_THREAD (1 << 1) 968 #define SECCOMP_LOG_TRAP (1 << 2) 969 #define SECCOMP_LOG_ERRNO (1 << 3) 970 #define SECCOMP_LOG_TRACE (1 << 4) 971 #define SECCOMP_LOG_LOG (1 << 5) 972 #define SECCOMP_LOG_ALLOW (1 << 6) 973 #define SECCOMP_LOG_USER_NOTIF (1 << 7) 974 975 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS | 976 SECCOMP_LOG_KILL_THREAD | 977 SECCOMP_LOG_TRAP | 978 SECCOMP_LOG_ERRNO | 979 SECCOMP_LOG_USER_NOTIF | 980 SECCOMP_LOG_TRACE | 981 SECCOMP_LOG_LOG; 982 983 static inline void seccomp_log(unsigned long syscall, long signr, u32 action, 984 bool requested) 985 { 986 bool log = false; 987 988 switch (action) { 989 case SECCOMP_RET_ALLOW: 990 break; 991 case SECCOMP_RET_TRAP: 992 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP; 993 break; 994 case SECCOMP_RET_ERRNO: 995 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO; 996 break; 997 case SECCOMP_RET_TRACE: 998 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE; 999 break; 1000 case SECCOMP_RET_USER_NOTIF: 1001 log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF; 1002 break; 1003 case SECCOMP_RET_LOG: 1004 log = seccomp_actions_logged & SECCOMP_LOG_LOG; 1005 break; 1006 case SECCOMP_RET_KILL_THREAD: 1007 log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD; 1008 break; 1009 case SECCOMP_RET_KILL_PROCESS: 1010 default: 1011 log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS; 1012 } 1013 1014 /* 1015 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the 1016 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence 1017 * any action from being logged by removing the action name from the 1018 * seccomp_actions_logged sysctl. 1019 */ 1020 if (!log) 1021 return; 1022 1023 audit_seccomp(syscall, signr, action); 1024 } 1025 1026 /* 1027 * Secure computing mode 1 allows only read/write/exit/sigreturn. 1028 * To be fully secure this must be combined with rlimit 1029 * to limit the stack allocations too. 1030 */ 1031 static const int mode1_syscalls[] = { 1032 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn, 1033 #ifdef __NR_uretprobe 1034 __NR_uretprobe, 1035 #endif 1036 -1, /* negative terminated */ 1037 }; 1038 1039 static void __secure_computing_strict(int this_syscall) 1040 { 1041 const int *allowed_syscalls = mode1_syscalls; 1042 #ifdef CONFIG_COMPAT 1043 if (in_compat_syscall()) 1044 allowed_syscalls = get_compat_mode1_syscalls(); 1045 #endif 1046 do { 1047 if (*allowed_syscalls == this_syscall) 1048 return; 1049 } while (*++allowed_syscalls != -1); 1050 1051 #ifdef SECCOMP_DEBUG 1052 dump_stack(); 1053 #endif 1054 current->seccomp.mode = SECCOMP_MODE_DEAD; 1055 seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true); 1056 do_exit(SIGKILL); 1057 } 1058 1059 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER 1060 void secure_computing_strict(int this_syscall) 1061 { 1062 int mode = current->seccomp.mode; 1063 1064 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && 1065 unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) 1066 return; 1067 1068 if (mode == SECCOMP_MODE_DISABLED) 1069 return; 1070 else if (mode == SECCOMP_MODE_STRICT) 1071 __secure_computing_strict(this_syscall); 1072 else 1073 BUG(); 1074 } 1075 int __secure_computing(const struct seccomp_data *sd) 1076 { 1077 int this_syscall = sd ? sd->nr : 1078 syscall_get_nr(current, current_pt_regs()); 1079 1080 secure_computing_strict(this_syscall); 1081 return 0; 1082 } 1083 #else 1084 1085 #ifdef CONFIG_SECCOMP_FILTER 1086 static u64 seccomp_next_notify_id(struct seccomp_filter *filter) 1087 { 1088 /* 1089 * Note: overflow is ok here, the id just needs to be unique per 1090 * filter. 1091 */ 1092 lockdep_assert_held(&filter->notify_lock); 1093 return filter->notif->next_id++; 1094 } 1095 1096 static void seccomp_handle_addfd(struct seccomp_kaddfd *addfd, struct seccomp_knotif *n) 1097 { 1098 int fd; 1099 1100 /* 1101 * Remove the notification, and reset the list pointers, indicating 1102 * that it has been handled. 1103 */ 1104 list_del_init(&addfd->list); 1105 if (!addfd->setfd) 1106 fd = receive_fd(addfd->file, NULL, addfd->flags); 1107 else 1108 fd = receive_fd_replace(addfd->fd, addfd->file, addfd->flags); 1109 addfd->ret = fd; 1110 1111 if (addfd->ioctl_flags & SECCOMP_ADDFD_FLAG_SEND) { 1112 /* If we fail reset and return an error to the notifier */ 1113 if (fd < 0) { 1114 n->state = SECCOMP_NOTIFY_SENT; 1115 } else { 1116 /* Return the FD we just added */ 1117 n->flags = 0; 1118 n->error = 0; 1119 n->val = fd; 1120 } 1121 } 1122 1123 /* 1124 * Mark the notification as completed. From this point, addfd mem 1125 * might be invalidated and we can't safely read it anymore. 1126 */ 1127 complete(&addfd->completion); 1128 } 1129 1130 static bool should_sleep_killable(struct seccomp_filter *match, 1131 struct seccomp_knotif *n) 1132 { 1133 return match->wait_killable_recv && n->state == SECCOMP_NOTIFY_SENT; 1134 } 1135 1136 static int seccomp_do_user_notification(int this_syscall, 1137 struct seccomp_filter *match, 1138 const struct seccomp_data *sd) 1139 { 1140 int err; 1141 u32 flags = 0; 1142 long ret = 0; 1143 struct seccomp_knotif n = {}; 1144 struct seccomp_kaddfd *addfd, *tmp; 1145 1146 mutex_lock(&match->notify_lock); 1147 err = -ENOSYS; 1148 if (!match->notif) 1149 goto out; 1150 1151 n.task = current; 1152 n.state = SECCOMP_NOTIFY_INIT; 1153 n.data = sd; 1154 n.id = seccomp_next_notify_id(match); 1155 init_completion(&n.ready); 1156 list_add_tail(&n.list, &match->notif->notifications); 1157 INIT_LIST_HEAD(&n.addfd); 1158 1159 atomic_inc(&match->notif->requests); 1160 if (match->notif->flags & SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP) 1161 wake_up_poll_on_current_cpu(&match->wqh, EPOLLIN | EPOLLRDNORM); 1162 else 1163 wake_up_poll(&match->wqh, EPOLLIN | EPOLLRDNORM); 1164 1165 /* 1166 * This is where we wait for a reply from userspace. 1167 */ 1168 do { 1169 bool wait_killable = should_sleep_killable(match, &n); 1170 1171 mutex_unlock(&match->notify_lock); 1172 if (wait_killable) 1173 err = wait_for_completion_killable(&n.ready); 1174 else 1175 err = wait_for_completion_interruptible(&n.ready); 1176 mutex_lock(&match->notify_lock); 1177 1178 if (err != 0) { 1179 /* 1180 * Check to see if the notifcation got picked up and 1181 * whether we should switch to wait killable. 1182 */ 1183 if (!wait_killable && should_sleep_killable(match, &n)) 1184 continue; 1185 1186 goto interrupted; 1187 } 1188 1189 addfd = list_first_entry_or_null(&n.addfd, 1190 struct seccomp_kaddfd, list); 1191 /* Check if we were woken up by a addfd message */ 1192 if (addfd) 1193 seccomp_handle_addfd(addfd, &n); 1194 1195 } while (n.state != SECCOMP_NOTIFY_REPLIED); 1196 1197 ret = n.val; 1198 err = n.error; 1199 flags = n.flags; 1200 1201 interrupted: 1202 /* If there were any pending addfd calls, clear them out */ 1203 list_for_each_entry_safe(addfd, tmp, &n.addfd, list) { 1204 /* The process went away before we got a chance to handle it */ 1205 addfd->ret = -ESRCH; 1206 list_del_init(&addfd->list); 1207 complete(&addfd->completion); 1208 } 1209 1210 /* 1211 * Note that it's possible the listener died in between the time when 1212 * we were notified of a response (or a signal) and when we were able to 1213 * re-acquire the lock, so only delete from the list if the 1214 * notification actually exists. 1215 * 1216 * Also note that this test is only valid because there's no way to 1217 * *reattach* to a notifier right now. If one is added, we'll need to 1218 * keep track of the notif itself and make sure they match here. 1219 */ 1220 if (match->notif) 1221 list_del(&n.list); 1222 out: 1223 mutex_unlock(&match->notify_lock); 1224 1225 /* Userspace requests to continue the syscall. */ 1226 if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) 1227 return 0; 1228 1229 syscall_set_return_value(current, current_pt_regs(), 1230 err, ret); 1231 return -1; 1232 } 1233 1234 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, 1235 const bool recheck_after_trace) 1236 { 1237 u32 filter_ret, action; 1238 struct seccomp_filter *match = NULL; 1239 int data; 1240 struct seccomp_data sd_local; 1241 1242 /* 1243 * Make sure that any changes to mode from another thread have 1244 * been seen after SYSCALL_WORK_SECCOMP was seen. 1245 */ 1246 smp_rmb(); 1247 1248 if (!sd) { 1249 populate_seccomp_data(&sd_local); 1250 sd = &sd_local; 1251 } 1252 1253 filter_ret = seccomp_run_filters(sd, &match); 1254 data = filter_ret & SECCOMP_RET_DATA; 1255 action = filter_ret & SECCOMP_RET_ACTION_FULL; 1256 1257 switch (action) { 1258 case SECCOMP_RET_ERRNO: 1259 /* Set low-order bits as an errno, capped at MAX_ERRNO. */ 1260 if (data > MAX_ERRNO) 1261 data = MAX_ERRNO; 1262 syscall_set_return_value(current, current_pt_regs(), 1263 -data, 0); 1264 goto skip; 1265 1266 case SECCOMP_RET_TRAP: 1267 /* Show the handler the original registers. */ 1268 syscall_rollback(current, current_pt_regs()); 1269 /* Let the filter pass back 16 bits of data. */ 1270 force_sig_seccomp(this_syscall, data, false); 1271 goto skip; 1272 1273 case SECCOMP_RET_TRACE: 1274 /* We've been put in this state by the ptracer already. */ 1275 if (recheck_after_trace) 1276 return 0; 1277 1278 /* ENOSYS these calls if there is no tracer attached. */ 1279 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) { 1280 syscall_set_return_value(current, 1281 current_pt_regs(), 1282 -ENOSYS, 0); 1283 goto skip; 1284 } 1285 1286 /* Allow the BPF to provide the event message */ 1287 ptrace_event(PTRACE_EVENT_SECCOMP, data); 1288 /* 1289 * The delivery of a fatal signal during event 1290 * notification may silently skip tracer notification, 1291 * which could leave us with a potentially unmodified 1292 * syscall that the tracer would have liked to have 1293 * changed. Since the process is about to die, we just 1294 * force the syscall to be skipped and let the signal 1295 * kill the process and correctly handle any tracer exit 1296 * notifications. 1297 */ 1298 if (fatal_signal_pending(current)) 1299 goto skip; 1300 /* Check if the tracer forced the syscall to be skipped. */ 1301 this_syscall = syscall_get_nr(current, current_pt_regs()); 1302 if (this_syscall < 0) 1303 goto skip; 1304 1305 /* 1306 * Recheck the syscall, since it may have changed. This 1307 * intentionally uses a NULL struct seccomp_data to force 1308 * a reload of all registers. This does not goto skip since 1309 * a skip would have already been reported. 1310 */ 1311 if (__seccomp_filter(this_syscall, NULL, true)) 1312 return -1; 1313 1314 return 0; 1315 1316 case SECCOMP_RET_USER_NOTIF: 1317 if (seccomp_do_user_notification(this_syscall, match, sd)) 1318 goto skip; 1319 1320 return 0; 1321 1322 case SECCOMP_RET_LOG: 1323 seccomp_log(this_syscall, 0, action, true); 1324 return 0; 1325 1326 case SECCOMP_RET_ALLOW: 1327 /* 1328 * Note that the "match" filter will always be NULL for 1329 * this action since SECCOMP_RET_ALLOW is the starting 1330 * state in seccomp_run_filters(). 1331 */ 1332 return 0; 1333 1334 case SECCOMP_RET_KILL_THREAD: 1335 case SECCOMP_RET_KILL_PROCESS: 1336 default: 1337 current->seccomp.mode = SECCOMP_MODE_DEAD; 1338 seccomp_log(this_syscall, SIGSYS, action, true); 1339 /* Dump core only if this is the last remaining thread. */ 1340 if (action != SECCOMP_RET_KILL_THREAD || 1341 (atomic_read(¤t->signal->live) == 1)) { 1342 /* Show the original registers in the dump. */ 1343 syscall_rollback(current, current_pt_regs()); 1344 /* Trigger a coredump with SIGSYS */ 1345 force_sig_seccomp(this_syscall, data, true); 1346 } else { 1347 do_exit(SIGSYS); 1348 } 1349 return -1; /* skip the syscall go directly to signal handling */ 1350 } 1351 1352 unreachable(); 1353 1354 skip: 1355 seccomp_log(this_syscall, 0, action, match ? match->log : false); 1356 return -1; 1357 } 1358 #else 1359 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd, 1360 const bool recheck_after_trace) 1361 { 1362 BUG(); 1363 1364 return -1; 1365 } 1366 #endif 1367 1368 int __secure_computing(const struct seccomp_data *sd) 1369 { 1370 int mode = current->seccomp.mode; 1371 int this_syscall; 1372 1373 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) && 1374 unlikely(current->ptrace & PT_SUSPEND_SECCOMP)) 1375 return 0; 1376 1377 this_syscall = sd ? sd->nr : 1378 syscall_get_nr(current, current_pt_regs()); 1379 1380 switch (mode) { 1381 case SECCOMP_MODE_STRICT: 1382 __secure_computing_strict(this_syscall); /* may call do_exit */ 1383 return 0; 1384 case SECCOMP_MODE_FILTER: 1385 return __seccomp_filter(this_syscall, sd, false); 1386 /* Surviving SECCOMP_RET_KILL_* must be proactively impossible. */ 1387 case SECCOMP_MODE_DEAD: 1388 WARN_ON_ONCE(1); 1389 do_exit(SIGKILL); 1390 return -1; 1391 default: 1392 BUG(); 1393 } 1394 } 1395 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */ 1396 1397 long prctl_get_seccomp(void) 1398 { 1399 return current->seccomp.mode; 1400 } 1401 1402 /** 1403 * seccomp_set_mode_strict: internal function for setting strict seccomp 1404 * 1405 * Once current->seccomp.mode is non-zero, it may not be changed. 1406 * 1407 * Returns 0 on success or -EINVAL on failure. 1408 */ 1409 static long seccomp_set_mode_strict(void) 1410 { 1411 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT; 1412 long ret = -EINVAL; 1413 1414 spin_lock_irq(¤t->sighand->siglock); 1415 1416 if (!seccomp_may_assign_mode(seccomp_mode)) 1417 goto out; 1418 1419 #ifdef TIF_NOTSC 1420 disable_TSC(); 1421 #endif 1422 seccomp_assign_mode(current, seccomp_mode, 0); 1423 ret = 0; 1424 1425 out: 1426 spin_unlock_irq(¤t->sighand->siglock); 1427 1428 return ret; 1429 } 1430 1431 #ifdef CONFIG_SECCOMP_FILTER 1432 static void seccomp_notify_free(struct seccomp_filter *filter) 1433 { 1434 kfree(filter->notif); 1435 filter->notif = NULL; 1436 } 1437 1438 static void seccomp_notify_detach(struct seccomp_filter *filter) 1439 { 1440 struct seccomp_knotif *knotif; 1441 1442 if (!filter) 1443 return; 1444 1445 mutex_lock(&filter->notify_lock); 1446 1447 /* 1448 * If this file is being closed because e.g. the task who owned it 1449 * died, let's wake everyone up who was waiting on us. 1450 */ 1451 list_for_each_entry(knotif, &filter->notif->notifications, list) { 1452 if (knotif->state == SECCOMP_NOTIFY_REPLIED) 1453 continue; 1454 1455 knotif->state = SECCOMP_NOTIFY_REPLIED; 1456 knotif->error = -ENOSYS; 1457 knotif->val = 0; 1458 1459 /* 1460 * We do not need to wake up any pending addfd messages, as 1461 * the notifier will do that for us, as this just looks 1462 * like a standard reply. 1463 */ 1464 complete(&knotif->ready); 1465 } 1466 1467 seccomp_notify_free(filter); 1468 mutex_unlock(&filter->notify_lock); 1469 } 1470 1471 static int seccomp_notify_release(struct inode *inode, struct file *file) 1472 { 1473 struct seccomp_filter *filter = file->private_data; 1474 1475 seccomp_notify_detach(filter); 1476 __put_seccomp_filter(filter); 1477 return 0; 1478 } 1479 1480 /* must be called with notif_lock held */ 1481 static inline struct seccomp_knotif * 1482 find_notification(struct seccomp_filter *filter, u64 id) 1483 { 1484 struct seccomp_knotif *cur; 1485 1486 lockdep_assert_held(&filter->notify_lock); 1487 1488 list_for_each_entry(cur, &filter->notif->notifications, list) { 1489 if (cur->id == id) 1490 return cur; 1491 } 1492 1493 return NULL; 1494 } 1495 1496 static int recv_wake_function(wait_queue_entry_t *wait, unsigned int mode, int sync, 1497 void *key) 1498 { 1499 /* Avoid a wakeup if event not interesting for us. */ 1500 if (key && !(key_to_poll(key) & (EPOLLIN | EPOLLERR | EPOLLHUP))) 1501 return 0; 1502 return autoremove_wake_function(wait, mode, sync, key); 1503 } 1504 1505 static int recv_wait_event(struct seccomp_filter *filter) 1506 { 1507 DEFINE_WAIT_FUNC(wait, recv_wake_function); 1508 int ret; 1509 1510 if (refcount_read(&filter->users) == 0) 1511 return 0; 1512 1513 if (atomic_dec_if_positive(&filter->notif->requests) >= 0) 1514 return 0; 1515 1516 for (;;) { 1517 ret = prepare_to_wait_event(&filter->wqh, &wait, TASK_INTERRUPTIBLE); 1518 1519 if (atomic_dec_if_positive(&filter->notif->requests) >= 0) 1520 break; 1521 if (refcount_read(&filter->users) == 0) 1522 break; 1523 1524 if (ret) 1525 return ret; 1526 1527 schedule(); 1528 } 1529 finish_wait(&filter->wqh, &wait); 1530 return 0; 1531 } 1532 1533 static long seccomp_notify_recv(struct seccomp_filter *filter, 1534 void __user *buf) 1535 { 1536 struct seccomp_knotif *knotif = NULL, *cur; 1537 struct seccomp_notif unotif; 1538 ssize_t ret; 1539 1540 /* Verify that we're not given garbage to keep struct extensible. */ 1541 ret = check_zeroed_user(buf, sizeof(unotif)); 1542 if (ret < 0) 1543 return ret; 1544 if (!ret) 1545 return -EINVAL; 1546 1547 memset(&unotif, 0, sizeof(unotif)); 1548 1549 ret = recv_wait_event(filter); 1550 if (ret < 0) 1551 return ret; 1552 1553 mutex_lock(&filter->notify_lock); 1554 list_for_each_entry(cur, &filter->notif->notifications, list) { 1555 if (cur->state == SECCOMP_NOTIFY_INIT) { 1556 knotif = cur; 1557 break; 1558 } 1559 } 1560 1561 /* 1562 * If we didn't find a notification, it could be that the task was 1563 * interrupted by a fatal signal between the time we were woken and 1564 * when we were able to acquire the rw lock. 1565 */ 1566 if (!knotif) { 1567 ret = -ENOENT; 1568 goto out; 1569 } 1570 1571 unotif.id = knotif->id; 1572 unotif.pid = task_pid_vnr(knotif->task); 1573 unotif.data = *(knotif->data); 1574 1575 knotif->state = SECCOMP_NOTIFY_SENT; 1576 wake_up_poll(&filter->wqh, EPOLLOUT | EPOLLWRNORM); 1577 ret = 0; 1578 out: 1579 mutex_unlock(&filter->notify_lock); 1580 1581 if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) { 1582 ret = -EFAULT; 1583 1584 /* 1585 * Userspace screwed up. To make sure that we keep this 1586 * notification alive, let's reset it back to INIT. It 1587 * may have died when we released the lock, so we need to make 1588 * sure it's still around. 1589 */ 1590 mutex_lock(&filter->notify_lock); 1591 knotif = find_notification(filter, unotif.id); 1592 if (knotif) { 1593 /* Reset the process to make sure it's not stuck */ 1594 if (should_sleep_killable(filter, knotif)) 1595 complete(&knotif->ready); 1596 knotif->state = SECCOMP_NOTIFY_INIT; 1597 atomic_inc(&filter->notif->requests); 1598 wake_up_poll(&filter->wqh, EPOLLIN | EPOLLRDNORM); 1599 } 1600 mutex_unlock(&filter->notify_lock); 1601 } 1602 1603 return ret; 1604 } 1605 1606 static long seccomp_notify_send(struct seccomp_filter *filter, 1607 void __user *buf) 1608 { 1609 struct seccomp_notif_resp resp = {}; 1610 struct seccomp_knotif *knotif; 1611 long ret; 1612 1613 if (copy_from_user(&resp, buf, sizeof(resp))) 1614 return -EFAULT; 1615 1616 if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE) 1617 return -EINVAL; 1618 1619 if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) && 1620 (resp.error || resp.val)) 1621 return -EINVAL; 1622 1623 ret = mutex_lock_interruptible(&filter->notify_lock); 1624 if (ret < 0) 1625 return ret; 1626 1627 knotif = find_notification(filter, resp.id); 1628 if (!knotif) { 1629 ret = -ENOENT; 1630 goto out; 1631 } 1632 1633 /* Allow exactly one reply. */ 1634 if (knotif->state != SECCOMP_NOTIFY_SENT) { 1635 ret = -EINPROGRESS; 1636 goto out; 1637 } 1638 1639 ret = 0; 1640 knotif->state = SECCOMP_NOTIFY_REPLIED; 1641 knotif->error = resp.error; 1642 knotif->val = resp.val; 1643 knotif->flags = resp.flags; 1644 if (filter->notif->flags & SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP) 1645 complete_on_current_cpu(&knotif->ready); 1646 else 1647 complete(&knotif->ready); 1648 out: 1649 mutex_unlock(&filter->notify_lock); 1650 return ret; 1651 } 1652 1653 static long seccomp_notify_id_valid(struct seccomp_filter *filter, 1654 void __user *buf) 1655 { 1656 struct seccomp_knotif *knotif; 1657 u64 id; 1658 long ret; 1659 1660 if (copy_from_user(&id, buf, sizeof(id))) 1661 return -EFAULT; 1662 1663 ret = mutex_lock_interruptible(&filter->notify_lock); 1664 if (ret < 0) 1665 return ret; 1666 1667 knotif = find_notification(filter, id); 1668 if (knotif && knotif->state == SECCOMP_NOTIFY_SENT) 1669 ret = 0; 1670 else 1671 ret = -ENOENT; 1672 1673 mutex_unlock(&filter->notify_lock); 1674 return ret; 1675 } 1676 1677 static long seccomp_notify_set_flags(struct seccomp_filter *filter, 1678 unsigned long flags) 1679 { 1680 long ret; 1681 1682 if (flags & ~SECCOMP_USER_NOTIF_FD_SYNC_WAKE_UP) 1683 return -EINVAL; 1684 1685 ret = mutex_lock_interruptible(&filter->notify_lock); 1686 if (ret < 0) 1687 return ret; 1688 filter->notif->flags = flags; 1689 mutex_unlock(&filter->notify_lock); 1690 return 0; 1691 } 1692 1693 static long seccomp_notify_addfd(struct seccomp_filter *filter, 1694 struct seccomp_notif_addfd __user *uaddfd, 1695 unsigned int size) 1696 { 1697 struct seccomp_notif_addfd addfd; 1698 struct seccomp_knotif *knotif; 1699 struct seccomp_kaddfd kaddfd; 1700 int ret; 1701 1702 BUILD_BUG_ON(sizeof(addfd) < SECCOMP_NOTIFY_ADDFD_SIZE_VER0); 1703 BUILD_BUG_ON(sizeof(addfd) != SECCOMP_NOTIFY_ADDFD_SIZE_LATEST); 1704 1705 if (size < SECCOMP_NOTIFY_ADDFD_SIZE_VER0 || size >= PAGE_SIZE) 1706 return -EINVAL; 1707 1708 ret = copy_struct_from_user(&addfd, sizeof(addfd), uaddfd, size); 1709 if (ret) 1710 return ret; 1711 1712 if (addfd.newfd_flags & ~O_CLOEXEC) 1713 return -EINVAL; 1714 1715 if (addfd.flags & ~(SECCOMP_ADDFD_FLAG_SETFD | SECCOMP_ADDFD_FLAG_SEND)) 1716 return -EINVAL; 1717 1718 if (addfd.newfd && !(addfd.flags & SECCOMP_ADDFD_FLAG_SETFD)) 1719 return -EINVAL; 1720 1721 kaddfd.file = fget(addfd.srcfd); 1722 if (!kaddfd.file) 1723 return -EBADF; 1724 1725 kaddfd.ioctl_flags = addfd.flags; 1726 kaddfd.flags = addfd.newfd_flags; 1727 kaddfd.setfd = addfd.flags & SECCOMP_ADDFD_FLAG_SETFD; 1728 kaddfd.fd = addfd.newfd; 1729 init_completion(&kaddfd.completion); 1730 1731 ret = mutex_lock_interruptible(&filter->notify_lock); 1732 if (ret < 0) 1733 goto out; 1734 1735 knotif = find_notification(filter, addfd.id); 1736 if (!knotif) { 1737 ret = -ENOENT; 1738 goto out_unlock; 1739 } 1740 1741 /* 1742 * We do not want to allow for FD injection to occur before the 1743 * notification has been picked up by a userspace handler, or after 1744 * the notification has been replied to. 1745 */ 1746 if (knotif->state != SECCOMP_NOTIFY_SENT) { 1747 ret = -EINPROGRESS; 1748 goto out_unlock; 1749 } 1750 1751 if (addfd.flags & SECCOMP_ADDFD_FLAG_SEND) { 1752 /* 1753 * Disallow queuing an atomic addfd + send reply while there are 1754 * some addfd requests still to process. 1755 * 1756 * There is no clear reason to support it and allows us to keep 1757 * the loop on the other side straight-forward. 1758 */ 1759 if (!list_empty(&knotif->addfd)) { 1760 ret = -EBUSY; 1761 goto out_unlock; 1762 } 1763 1764 /* Allow exactly only one reply */ 1765 knotif->state = SECCOMP_NOTIFY_REPLIED; 1766 } 1767 1768 list_add(&kaddfd.list, &knotif->addfd); 1769 complete(&knotif->ready); 1770 mutex_unlock(&filter->notify_lock); 1771 1772 /* Now we wait for it to be processed or be interrupted */ 1773 ret = wait_for_completion_interruptible(&kaddfd.completion); 1774 if (ret == 0) { 1775 /* 1776 * We had a successful completion. The other side has already 1777 * removed us from the addfd queue, and 1778 * wait_for_completion_interruptible has a memory barrier upon 1779 * success that lets us read this value directly without 1780 * locking. 1781 */ 1782 ret = kaddfd.ret; 1783 goto out; 1784 } 1785 1786 mutex_lock(&filter->notify_lock); 1787 /* 1788 * Even though we were woken up by a signal and not a successful 1789 * completion, a completion may have happened in the mean time. 1790 * 1791 * We need to check again if the addfd request has been handled, 1792 * and if not, we will remove it from the queue. 1793 */ 1794 if (list_empty(&kaddfd.list)) 1795 ret = kaddfd.ret; 1796 else 1797 list_del(&kaddfd.list); 1798 1799 out_unlock: 1800 mutex_unlock(&filter->notify_lock); 1801 out: 1802 fput(kaddfd.file); 1803 1804 return ret; 1805 } 1806 1807 static long seccomp_notify_ioctl(struct file *file, unsigned int cmd, 1808 unsigned long arg) 1809 { 1810 struct seccomp_filter *filter = file->private_data; 1811 void __user *buf = (void __user *)arg; 1812 1813 /* Fixed-size ioctls */ 1814 switch (cmd) { 1815 case SECCOMP_IOCTL_NOTIF_RECV: 1816 return seccomp_notify_recv(filter, buf); 1817 case SECCOMP_IOCTL_NOTIF_SEND: 1818 return seccomp_notify_send(filter, buf); 1819 case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR: 1820 case SECCOMP_IOCTL_NOTIF_ID_VALID: 1821 return seccomp_notify_id_valid(filter, buf); 1822 case SECCOMP_IOCTL_NOTIF_SET_FLAGS: 1823 return seccomp_notify_set_flags(filter, arg); 1824 } 1825 1826 /* Extensible Argument ioctls */ 1827 #define EA_IOCTL(cmd) ((cmd) & ~(IOC_INOUT | IOCSIZE_MASK)) 1828 switch (EA_IOCTL(cmd)) { 1829 case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADDFD): 1830 return seccomp_notify_addfd(filter, buf, _IOC_SIZE(cmd)); 1831 default: 1832 return -EINVAL; 1833 } 1834 } 1835 1836 static __poll_t seccomp_notify_poll(struct file *file, 1837 struct poll_table_struct *poll_tab) 1838 { 1839 struct seccomp_filter *filter = file->private_data; 1840 __poll_t ret = 0; 1841 struct seccomp_knotif *cur; 1842 1843 poll_wait(file, &filter->wqh, poll_tab); 1844 1845 if (mutex_lock_interruptible(&filter->notify_lock) < 0) 1846 return EPOLLERR; 1847 1848 list_for_each_entry(cur, &filter->notif->notifications, list) { 1849 if (cur->state == SECCOMP_NOTIFY_INIT) 1850 ret |= EPOLLIN | EPOLLRDNORM; 1851 if (cur->state == SECCOMP_NOTIFY_SENT) 1852 ret |= EPOLLOUT | EPOLLWRNORM; 1853 if ((ret & EPOLLIN) && (ret & EPOLLOUT)) 1854 break; 1855 } 1856 1857 mutex_unlock(&filter->notify_lock); 1858 1859 if (refcount_read(&filter->users) == 0) 1860 ret |= EPOLLHUP; 1861 1862 return ret; 1863 } 1864 1865 static const struct file_operations seccomp_notify_ops = { 1866 .poll = seccomp_notify_poll, 1867 .release = seccomp_notify_release, 1868 .unlocked_ioctl = seccomp_notify_ioctl, 1869 .compat_ioctl = seccomp_notify_ioctl, 1870 }; 1871 1872 static struct file *init_listener(struct seccomp_filter *filter) 1873 { 1874 struct file *ret; 1875 1876 ret = ERR_PTR(-ENOMEM); 1877 filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL); 1878 if (!filter->notif) 1879 goto out; 1880 1881 filter->notif->next_id = get_random_u64(); 1882 INIT_LIST_HEAD(&filter->notif->notifications); 1883 1884 ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops, 1885 filter, O_RDWR); 1886 if (IS_ERR(ret)) 1887 goto out_notif; 1888 1889 /* The file has a reference to it now */ 1890 __get_seccomp_filter(filter); 1891 1892 out_notif: 1893 if (IS_ERR(ret)) 1894 seccomp_notify_free(filter); 1895 out: 1896 return ret; 1897 } 1898 1899 /* 1900 * Does @new_child have a listener while an ancestor also has a listener? 1901 * If so, we'll want to reject this filter. 1902 * This only has to be tested for the current process, even in the TSYNC case, 1903 * because TSYNC installs @child with the same parent on all threads. 1904 * Note that @new_child is not hooked up to its parent at this point yet, so 1905 * we use current->seccomp.filter. 1906 */ 1907 static bool has_duplicate_listener(struct seccomp_filter *new_child) 1908 { 1909 struct seccomp_filter *cur; 1910 1911 /* must be protected against concurrent TSYNC */ 1912 lockdep_assert_held(¤t->sighand->siglock); 1913 1914 if (!new_child->notif) 1915 return false; 1916 for (cur = current->seccomp.filter; cur; cur = cur->prev) { 1917 if (cur->notif) 1918 return true; 1919 } 1920 1921 return false; 1922 } 1923 1924 /** 1925 * seccomp_set_mode_filter: internal function for setting seccomp filter 1926 * @flags: flags to change filter behavior 1927 * @filter: struct sock_fprog containing filter 1928 * 1929 * This function may be called repeatedly to install additional filters. 1930 * Every filter successfully installed will be evaluated (in reverse order) 1931 * for each system call the task makes. 1932 * 1933 * Once current->seccomp.mode is non-zero, it may not be changed. 1934 * 1935 * Returns 0 on success or -EINVAL on failure. 1936 */ 1937 static long seccomp_set_mode_filter(unsigned int flags, 1938 const char __user *filter) 1939 { 1940 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER; 1941 struct seccomp_filter *prepared = NULL; 1942 long ret = -EINVAL; 1943 int listener = -1; 1944 struct file *listener_f = NULL; 1945 1946 /* Validate flags. */ 1947 if (flags & ~SECCOMP_FILTER_FLAG_MASK) 1948 return -EINVAL; 1949 1950 /* 1951 * In the successful case, NEW_LISTENER returns the new listener fd. 1952 * But in the failure case, TSYNC returns the thread that died. If you 1953 * combine these two flags, there's no way to tell whether something 1954 * succeeded or failed. So, let's disallow this combination if the user 1955 * has not explicitly requested no errors from TSYNC. 1956 */ 1957 if ((flags & SECCOMP_FILTER_FLAG_TSYNC) && 1958 (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) && 1959 ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0)) 1960 return -EINVAL; 1961 1962 /* 1963 * The SECCOMP_FILTER_FLAG_WAIT_KILLABLE_SENT flag doesn't make sense 1964 * without the SECCOMP_FILTER_FLAG_NEW_LISTENER flag. 1965 */ 1966 if ((flags & SECCOMP_FILTER_FLAG_WAIT_KILLABLE_RECV) && 1967 ((flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) == 0)) 1968 return -EINVAL; 1969 1970 /* Prepare the new filter before holding any locks. */ 1971 prepared = seccomp_prepare_user_filter(filter); 1972 if (IS_ERR(prepared)) 1973 return PTR_ERR(prepared); 1974 1975 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) { 1976 listener = get_unused_fd_flags(O_CLOEXEC); 1977 if (listener < 0) { 1978 ret = listener; 1979 goto out_free; 1980 } 1981 1982 listener_f = init_listener(prepared); 1983 if (IS_ERR(listener_f)) { 1984 put_unused_fd(listener); 1985 ret = PTR_ERR(listener_f); 1986 goto out_free; 1987 } 1988 } 1989 1990 /* 1991 * Make sure we cannot change seccomp or nnp state via TSYNC 1992 * while another thread is in the middle of calling exec. 1993 */ 1994 if (flags & SECCOMP_FILTER_FLAG_TSYNC && 1995 mutex_lock_killable(¤t->signal->cred_guard_mutex)) 1996 goto out_put_fd; 1997 1998 spin_lock_irq(¤t->sighand->siglock); 1999 2000 if (!seccomp_may_assign_mode(seccomp_mode)) 2001 goto out; 2002 2003 if (has_duplicate_listener(prepared)) { 2004 ret = -EBUSY; 2005 goto out; 2006 } 2007 2008 ret = seccomp_attach_filter(flags, prepared); 2009 if (ret) 2010 goto out; 2011 /* Do not free the successfully attached filter. */ 2012 prepared = NULL; 2013 2014 seccomp_assign_mode(current, seccomp_mode, flags); 2015 out: 2016 spin_unlock_irq(¤t->sighand->siglock); 2017 if (flags & SECCOMP_FILTER_FLAG_TSYNC) 2018 mutex_unlock(¤t->signal->cred_guard_mutex); 2019 out_put_fd: 2020 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) { 2021 if (ret) { 2022 listener_f->private_data = NULL; 2023 fput(listener_f); 2024 put_unused_fd(listener); 2025 seccomp_notify_detach(prepared); 2026 } else { 2027 fd_install(listener, listener_f); 2028 ret = listener; 2029 } 2030 } 2031 out_free: 2032 seccomp_filter_free(prepared); 2033 return ret; 2034 } 2035 #else 2036 static inline long seccomp_set_mode_filter(unsigned int flags, 2037 const char __user *filter) 2038 { 2039 return -EINVAL; 2040 } 2041 #endif 2042 2043 static long seccomp_get_action_avail(const char __user *uaction) 2044 { 2045 u32 action; 2046 2047 if (copy_from_user(&action, uaction, sizeof(action))) 2048 return -EFAULT; 2049 2050 switch (action) { 2051 case SECCOMP_RET_KILL_PROCESS: 2052 case SECCOMP_RET_KILL_THREAD: 2053 case SECCOMP_RET_TRAP: 2054 case SECCOMP_RET_ERRNO: 2055 case SECCOMP_RET_USER_NOTIF: 2056 case SECCOMP_RET_TRACE: 2057 case SECCOMP_RET_LOG: 2058 case SECCOMP_RET_ALLOW: 2059 break; 2060 default: 2061 return -EOPNOTSUPP; 2062 } 2063 2064 return 0; 2065 } 2066 2067 static long seccomp_get_notif_sizes(void __user *usizes) 2068 { 2069 struct seccomp_notif_sizes sizes = { 2070 .seccomp_notif = sizeof(struct seccomp_notif), 2071 .seccomp_notif_resp = sizeof(struct seccomp_notif_resp), 2072 .seccomp_data = sizeof(struct seccomp_data), 2073 }; 2074 2075 if (copy_to_user(usizes, &sizes, sizeof(sizes))) 2076 return -EFAULT; 2077 2078 return 0; 2079 } 2080 2081 /* Common entry point for both prctl and syscall. */ 2082 static long do_seccomp(unsigned int op, unsigned int flags, 2083 void __user *uargs) 2084 { 2085 switch (op) { 2086 case SECCOMP_SET_MODE_STRICT: 2087 if (flags != 0 || uargs != NULL) 2088 return -EINVAL; 2089 return seccomp_set_mode_strict(); 2090 case SECCOMP_SET_MODE_FILTER: 2091 return seccomp_set_mode_filter(flags, uargs); 2092 case SECCOMP_GET_ACTION_AVAIL: 2093 if (flags != 0) 2094 return -EINVAL; 2095 2096 return seccomp_get_action_avail(uargs); 2097 case SECCOMP_GET_NOTIF_SIZES: 2098 if (flags != 0) 2099 return -EINVAL; 2100 2101 return seccomp_get_notif_sizes(uargs); 2102 default: 2103 return -EINVAL; 2104 } 2105 } 2106 2107 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags, 2108 void __user *, uargs) 2109 { 2110 return do_seccomp(op, flags, uargs); 2111 } 2112 2113 /** 2114 * prctl_set_seccomp: configures current->seccomp.mode 2115 * @seccomp_mode: requested mode to use 2116 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER 2117 * 2118 * Returns 0 on success or -EINVAL on failure. 2119 */ 2120 long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter) 2121 { 2122 unsigned int op; 2123 void __user *uargs; 2124 2125 switch (seccomp_mode) { 2126 case SECCOMP_MODE_STRICT: 2127 op = SECCOMP_SET_MODE_STRICT; 2128 /* 2129 * Setting strict mode through prctl always ignored filter, 2130 * so make sure it is always NULL here to pass the internal 2131 * check in do_seccomp(). 2132 */ 2133 uargs = NULL; 2134 break; 2135 case SECCOMP_MODE_FILTER: 2136 op = SECCOMP_SET_MODE_FILTER; 2137 uargs = filter; 2138 break; 2139 default: 2140 return -EINVAL; 2141 } 2142 2143 /* prctl interface doesn't have flags, so they are always zero. */ 2144 return do_seccomp(op, 0, uargs); 2145 } 2146 2147 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE) 2148 static struct seccomp_filter *get_nth_filter(struct task_struct *task, 2149 unsigned long filter_off) 2150 { 2151 struct seccomp_filter *orig, *filter; 2152 unsigned long count; 2153 2154 /* 2155 * Note: this is only correct because the caller should be the (ptrace) 2156 * tracer of the task, otherwise lock_task_sighand is needed. 2157 */ 2158 spin_lock_irq(&task->sighand->siglock); 2159 2160 if (task->seccomp.mode != SECCOMP_MODE_FILTER) { 2161 spin_unlock_irq(&task->sighand->siglock); 2162 return ERR_PTR(-EINVAL); 2163 } 2164 2165 orig = task->seccomp.filter; 2166 __get_seccomp_filter(orig); 2167 spin_unlock_irq(&task->sighand->siglock); 2168 2169 count = 0; 2170 for (filter = orig; filter; filter = filter->prev) 2171 count++; 2172 2173 if (filter_off >= count) { 2174 filter = ERR_PTR(-ENOENT); 2175 goto out; 2176 } 2177 2178 count -= filter_off; 2179 for (filter = orig; filter && count > 1; filter = filter->prev) 2180 count--; 2181 2182 if (WARN_ON(count != 1 || !filter)) { 2183 filter = ERR_PTR(-ENOENT); 2184 goto out; 2185 } 2186 2187 __get_seccomp_filter(filter); 2188 2189 out: 2190 __put_seccomp_filter(orig); 2191 return filter; 2192 } 2193 2194 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off, 2195 void __user *data) 2196 { 2197 struct seccomp_filter *filter; 2198 struct sock_fprog_kern *fprog; 2199 long ret; 2200 2201 if (!capable(CAP_SYS_ADMIN) || 2202 current->seccomp.mode != SECCOMP_MODE_DISABLED) { 2203 return -EACCES; 2204 } 2205 2206 filter = get_nth_filter(task, filter_off); 2207 if (IS_ERR(filter)) 2208 return PTR_ERR(filter); 2209 2210 fprog = filter->prog->orig_prog; 2211 if (!fprog) { 2212 /* This must be a new non-cBPF filter, since we save 2213 * every cBPF filter's orig_prog above when 2214 * CONFIG_CHECKPOINT_RESTORE is enabled. 2215 */ 2216 ret = -EMEDIUMTYPE; 2217 goto out; 2218 } 2219 2220 ret = fprog->len; 2221 if (!data) 2222 goto out; 2223 2224 if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog))) 2225 ret = -EFAULT; 2226 2227 out: 2228 __put_seccomp_filter(filter); 2229 return ret; 2230 } 2231 2232 long seccomp_get_metadata(struct task_struct *task, 2233 unsigned long size, void __user *data) 2234 { 2235 long ret; 2236 struct seccomp_filter *filter; 2237 struct seccomp_metadata kmd = {}; 2238 2239 if (!capable(CAP_SYS_ADMIN) || 2240 current->seccomp.mode != SECCOMP_MODE_DISABLED) { 2241 return -EACCES; 2242 } 2243 2244 size = min_t(unsigned long, size, sizeof(kmd)); 2245 2246 if (size < sizeof(kmd.filter_off)) 2247 return -EINVAL; 2248 2249 if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off))) 2250 return -EFAULT; 2251 2252 filter = get_nth_filter(task, kmd.filter_off); 2253 if (IS_ERR(filter)) 2254 return PTR_ERR(filter); 2255 2256 if (filter->log) 2257 kmd.flags |= SECCOMP_FILTER_FLAG_LOG; 2258 2259 ret = size; 2260 if (copy_to_user(data, &kmd, size)) 2261 ret = -EFAULT; 2262 2263 __put_seccomp_filter(filter); 2264 return ret; 2265 } 2266 #endif 2267 2268 #ifdef CONFIG_SYSCTL 2269 2270 /* Human readable action names for friendly sysctl interaction */ 2271 #define SECCOMP_RET_KILL_PROCESS_NAME "kill_process" 2272 #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread" 2273 #define SECCOMP_RET_TRAP_NAME "trap" 2274 #define SECCOMP_RET_ERRNO_NAME "errno" 2275 #define SECCOMP_RET_USER_NOTIF_NAME "user_notif" 2276 #define SECCOMP_RET_TRACE_NAME "trace" 2277 #define SECCOMP_RET_LOG_NAME "log" 2278 #define SECCOMP_RET_ALLOW_NAME "allow" 2279 2280 static const char seccomp_actions_avail[] = 2281 SECCOMP_RET_KILL_PROCESS_NAME " " 2282 SECCOMP_RET_KILL_THREAD_NAME " " 2283 SECCOMP_RET_TRAP_NAME " " 2284 SECCOMP_RET_ERRNO_NAME " " 2285 SECCOMP_RET_USER_NOTIF_NAME " " 2286 SECCOMP_RET_TRACE_NAME " " 2287 SECCOMP_RET_LOG_NAME " " 2288 SECCOMP_RET_ALLOW_NAME; 2289 2290 struct seccomp_log_name { 2291 u32 log; 2292 const char *name; 2293 }; 2294 2295 static const struct seccomp_log_name seccomp_log_names[] = { 2296 { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME }, 2297 { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME }, 2298 { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME }, 2299 { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME }, 2300 { SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME }, 2301 { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME }, 2302 { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME }, 2303 { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME }, 2304 { } 2305 }; 2306 2307 static bool seccomp_names_from_actions_logged(char *names, size_t size, 2308 u32 actions_logged, 2309 const char *sep) 2310 { 2311 const struct seccomp_log_name *cur; 2312 bool append_sep = false; 2313 2314 for (cur = seccomp_log_names; cur->name && size; cur++) { 2315 ssize_t ret; 2316 2317 if (!(actions_logged & cur->log)) 2318 continue; 2319 2320 if (append_sep) { 2321 ret = strscpy(names, sep, size); 2322 if (ret < 0) 2323 return false; 2324 2325 names += ret; 2326 size -= ret; 2327 } else 2328 append_sep = true; 2329 2330 ret = strscpy(names, cur->name, size); 2331 if (ret < 0) 2332 return false; 2333 2334 names += ret; 2335 size -= ret; 2336 } 2337 2338 return true; 2339 } 2340 2341 static bool seccomp_action_logged_from_name(u32 *action_logged, 2342 const char *name) 2343 { 2344 const struct seccomp_log_name *cur; 2345 2346 for (cur = seccomp_log_names; cur->name; cur++) { 2347 if (!strcmp(cur->name, name)) { 2348 *action_logged = cur->log; 2349 return true; 2350 } 2351 } 2352 2353 return false; 2354 } 2355 2356 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names) 2357 { 2358 char *name; 2359 2360 *actions_logged = 0; 2361 while ((name = strsep(&names, " ")) && *name) { 2362 u32 action_logged = 0; 2363 2364 if (!seccomp_action_logged_from_name(&action_logged, name)) 2365 return false; 2366 2367 *actions_logged |= action_logged; 2368 } 2369 2370 return true; 2371 } 2372 2373 static int read_actions_logged(const struct ctl_table *ro_table, void *buffer, 2374 size_t *lenp, loff_t *ppos) 2375 { 2376 char names[sizeof(seccomp_actions_avail)]; 2377 struct ctl_table table; 2378 2379 memset(names, 0, sizeof(names)); 2380 2381 if (!seccomp_names_from_actions_logged(names, sizeof(names), 2382 seccomp_actions_logged, " ")) 2383 return -EINVAL; 2384 2385 table = *ro_table; 2386 table.data = names; 2387 table.maxlen = sizeof(names); 2388 return proc_dostring(&table, 0, buffer, lenp, ppos); 2389 } 2390 2391 static int write_actions_logged(const struct ctl_table *ro_table, void *buffer, 2392 size_t *lenp, loff_t *ppos, u32 *actions_logged) 2393 { 2394 char names[sizeof(seccomp_actions_avail)]; 2395 struct ctl_table table; 2396 int ret; 2397 2398 if (!capable(CAP_SYS_ADMIN)) 2399 return -EPERM; 2400 2401 memset(names, 0, sizeof(names)); 2402 2403 table = *ro_table; 2404 table.data = names; 2405 table.maxlen = sizeof(names); 2406 ret = proc_dostring(&table, 1, buffer, lenp, ppos); 2407 if (ret) 2408 return ret; 2409 2410 if (!seccomp_actions_logged_from_names(actions_logged, table.data)) 2411 return -EINVAL; 2412 2413 if (*actions_logged & SECCOMP_LOG_ALLOW) 2414 return -EINVAL; 2415 2416 seccomp_actions_logged = *actions_logged; 2417 return 0; 2418 } 2419 2420 static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged, 2421 int ret) 2422 { 2423 char names[sizeof(seccomp_actions_avail)]; 2424 char old_names[sizeof(seccomp_actions_avail)]; 2425 const char *new = names; 2426 const char *old = old_names; 2427 2428 if (!audit_enabled) 2429 return; 2430 2431 memset(names, 0, sizeof(names)); 2432 memset(old_names, 0, sizeof(old_names)); 2433 2434 if (ret) 2435 new = "?"; 2436 else if (!actions_logged) 2437 new = "(none)"; 2438 else if (!seccomp_names_from_actions_logged(names, sizeof(names), 2439 actions_logged, ",")) 2440 new = "?"; 2441 2442 if (!old_actions_logged) 2443 old = "(none)"; 2444 else if (!seccomp_names_from_actions_logged(old_names, 2445 sizeof(old_names), 2446 old_actions_logged, ",")) 2447 old = "?"; 2448 2449 return audit_seccomp_actions_logged(new, old, !ret); 2450 } 2451 2452 static int seccomp_actions_logged_handler(const struct ctl_table *ro_table, int write, 2453 void *buffer, size_t *lenp, 2454 loff_t *ppos) 2455 { 2456 int ret; 2457 2458 if (write) { 2459 u32 actions_logged = 0; 2460 u32 old_actions_logged = seccomp_actions_logged; 2461 2462 ret = write_actions_logged(ro_table, buffer, lenp, ppos, 2463 &actions_logged); 2464 audit_actions_logged(actions_logged, old_actions_logged, ret); 2465 } else 2466 ret = read_actions_logged(ro_table, buffer, lenp, ppos); 2467 2468 return ret; 2469 } 2470 2471 static const struct ctl_table seccomp_sysctl_table[] = { 2472 { 2473 .procname = "actions_avail", 2474 .data = (void *) &seccomp_actions_avail, 2475 .maxlen = sizeof(seccomp_actions_avail), 2476 .mode = 0444, 2477 .proc_handler = proc_dostring, 2478 }, 2479 { 2480 .procname = "actions_logged", 2481 .mode = 0644, 2482 .proc_handler = seccomp_actions_logged_handler, 2483 }, 2484 }; 2485 2486 static int __init seccomp_sysctl_init(void) 2487 { 2488 register_sysctl_init("kernel/seccomp", seccomp_sysctl_table); 2489 return 0; 2490 } 2491 2492 device_initcall(seccomp_sysctl_init) 2493 2494 #endif /* CONFIG_SYSCTL */ 2495 2496 #ifdef CONFIG_SECCOMP_CACHE_DEBUG 2497 /* Currently CONFIG_SECCOMP_CACHE_DEBUG implies SECCOMP_ARCH_NATIVE */ 2498 static void proc_pid_seccomp_cache_arch(struct seq_file *m, const char *name, 2499 const void *bitmap, size_t bitmap_size) 2500 { 2501 int nr; 2502 2503 for (nr = 0; nr < bitmap_size; nr++) { 2504 bool cached = test_bit(nr, bitmap); 2505 char *status = cached ? "ALLOW" : "FILTER"; 2506 2507 seq_printf(m, "%s %d %s\n", name, nr, status); 2508 } 2509 } 2510 2511 int proc_pid_seccomp_cache(struct seq_file *m, struct pid_namespace *ns, 2512 struct pid *pid, struct task_struct *task) 2513 { 2514 struct seccomp_filter *f; 2515 unsigned long flags; 2516 2517 /* 2518 * We don't want some sandboxed process to know what their seccomp 2519 * filters consist of. 2520 */ 2521 if (!file_ns_capable(m->file, &init_user_ns, CAP_SYS_ADMIN)) 2522 return -EACCES; 2523 2524 if (!lock_task_sighand(task, &flags)) 2525 return -ESRCH; 2526 2527 f = READ_ONCE(task->seccomp.filter); 2528 if (!f) { 2529 unlock_task_sighand(task, &flags); 2530 return 0; 2531 } 2532 2533 /* prevent filter from being freed while we are printing it */ 2534 __get_seccomp_filter(f); 2535 unlock_task_sighand(task, &flags); 2536 2537 proc_pid_seccomp_cache_arch(m, SECCOMP_ARCH_NATIVE_NAME, 2538 f->cache.allow_native, 2539 SECCOMP_ARCH_NATIVE_NR); 2540 2541 #ifdef SECCOMP_ARCH_COMPAT 2542 proc_pid_seccomp_cache_arch(m, SECCOMP_ARCH_COMPAT_NAME, 2543 f->cache.allow_compat, 2544 SECCOMP_ARCH_COMPAT_NR); 2545 #endif /* SECCOMP_ARCH_COMPAT */ 2546 2547 __put_seccomp_filter(f); 2548 return 0; 2549 } 2550 #endif /* CONFIG_SECCOMP_CACHE_DEBUG */ 2551