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