1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * linux/kernel/seccomp.c
   4 *
   5 * Copyright 2004-2005  Andrea Arcangeli <andrea@cpushare.com>
   6 *
   7 * Copyright (C) 2012 Google, Inc.
   8 * Will Drewry <wad@chromium.org>
   9 *
  10 * This defines a simple but solid secure-computing facility.
  11 *
  12 * Mode 1 uses a fixed list of allowed system calls.
  13 * Mode 2 allows user-defined system call filters in the form
  14 *        of Berkeley Packet Filters/Linux Socket Filters.
  15 */
  16
  17#include <linux/refcount.h>
  18#include <linux/audit.h>
  19#include <linux/compat.h>
  20#include <linux/coredump.h>
  21#include <linux/kmemleak.h>
  22#include <linux/nospec.h>
  23#include <linux/prctl.h>
  24#include <linux/sched.h>
  25#include <linux/sched/task_stack.h>
  26#include <linux/seccomp.h>
  27#include <linux/slab.h>
  28#include <linux/syscalls.h>
  29#include <linux/sysctl.h>
  30
  31#ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
  32#include <asm/syscall.h>
  33#endif
  34
  35#ifdef CONFIG_SECCOMP_FILTER
  36#include <linux/filter.h>
  37#include <linux/pid.h>
  38#include <linux/ptrace.h>
  39#include <linux/security.h>
  40#include <linux/tracehook.h>
  41#include <linux/uaccess.h>
  42
  43/**
  44 * struct seccomp_filter - container for seccomp BPF programs
  45 *
  46 * @usage: reference count to manage the object lifetime.
  47 *         get/put helpers should be used when accessing an instance
  48 *         outside of a lifetime-guarded section.  In general, this
  49 *         is only needed for handling filters shared across tasks.
  50 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
  51 * @prev: points to a previously installed, or inherited, filter
  52 * @prog: the BPF program to evaluate
  53 *
  54 * seccomp_filter objects are organized in a tree linked via the @prev
  55 * pointer.  For any task, it appears to be a singly-linked list starting
  56 * with current->seccomp.filter, the most recently attached or inherited filter.
  57 * However, multiple filters may share a @prev node, by way of fork(), which
  58 * results in a unidirectional tree existing in memory.  This is similar to
  59 * how namespaces work.
  60 *
  61 * seccomp_filter objects should never be modified after being attached
  62 * to a task_struct (other than @usage).
  63 */
  64struct seccomp_filter {
  65	refcount_t usage;
  66	bool log;
  67	struct seccomp_filter *prev;
  68	struct bpf_prog *prog;
  69};
  70
  71/* Limit any path through the tree to 256KB worth of instructions. */
  72#define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
  73
  74/*
  75 * Endianness is explicitly ignored and left for BPF program authors to manage
  76 * as per the specific architecture.
  77 */
  78static void populate_seccomp_data(struct seccomp_data *sd)
  79{
  80	struct task_struct *task = current;
  81	struct pt_regs *regs = task_pt_regs(task);
  82	unsigned long args[6];
  83
  84	sd->nr = syscall_get_nr(task, regs);
  85	sd->arch = syscall_get_arch();
  86	syscall_get_arguments(task, regs, 0, 6, args);
  87	sd->args[0] = args[0];
  88	sd->args[1] = args[1];
  89	sd->args[2] = args[2];
  90	sd->args[3] = args[3];
  91	sd->args[4] = args[4];
  92	sd->args[5] = args[5];
  93	sd->instruction_pointer = KSTK_EIP(task);
  94}
  95
  96/**
  97 *	seccomp_check_filter - verify seccomp filter code
  98 *	@filter: filter to verify
  99 *	@flen: length of filter
 100 *
 101 * Takes a previously checked filter (by bpf_check_classic) and
 102 * redirects all filter code that loads struct sk_buff data
 103 * and related data through seccomp_bpf_load.  It also
 104 * enforces length and alignment checking of those loads.
 105 *
 106 * Returns 0 if the rule set is legal or -EINVAL if not.
 107 */
 108static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
 109{
 110	int pc;
 111	for (pc = 0; pc < flen; pc++) {
 112		struct sock_filter *ftest = &filter[pc];
 113		u16 code = ftest->code;
 114		u32 k = ftest->k;
 115
 116		switch (code) {
 117		case BPF_LD | BPF_W | BPF_ABS:
 118			ftest->code = BPF_LDX | BPF_W | BPF_ABS;
 119			/* 32-bit aligned and not out of bounds. */
 120			if (k >= sizeof(struct seccomp_data) || k & 3)
 121				return -EINVAL;
 122			continue;
 123		case BPF_LD | BPF_W | BPF_LEN:
 124			ftest->code = BPF_LD | BPF_IMM;
 125			ftest->k = sizeof(struct seccomp_data);
 126			continue;
 127		case BPF_LDX | BPF_W | BPF_LEN:
 128			ftest->code = BPF_LDX | BPF_IMM;
 129			ftest->k = sizeof(struct seccomp_data);
 130			continue;
 131		/* Explicitly include allowed calls. */
 132		case BPF_RET | BPF_K:
 133		case BPF_RET | BPF_A:
 134		case BPF_ALU | BPF_ADD | BPF_K:
 135		case BPF_ALU | BPF_ADD | BPF_X:
 136		case BPF_ALU | BPF_SUB | BPF_K:
 137		case BPF_ALU | BPF_SUB | BPF_X:
 138		case BPF_ALU | BPF_MUL | BPF_K:
 139		case BPF_ALU | BPF_MUL | BPF_X:
 140		case BPF_ALU | BPF_DIV | BPF_K:
 141		case BPF_ALU | BPF_DIV | BPF_X:
 142		case BPF_ALU | BPF_AND | BPF_K:
 143		case BPF_ALU | BPF_AND | BPF_X:
 144		case BPF_ALU | BPF_OR | BPF_K:
 145		case BPF_ALU | BPF_OR | BPF_X:
 146		case BPF_ALU | BPF_XOR | BPF_K:
 147		case BPF_ALU | BPF_XOR | BPF_X:
 148		case BPF_ALU | BPF_LSH | BPF_K:
 149		case BPF_ALU | BPF_LSH | BPF_X:
 150		case BPF_ALU | BPF_RSH | BPF_K:
 151		case BPF_ALU | BPF_RSH | BPF_X:
 152		case BPF_ALU | BPF_NEG:
 153		case BPF_LD | BPF_IMM:
 154		case BPF_LDX | BPF_IMM:
 155		case BPF_MISC | BPF_TAX:
 156		case BPF_MISC | BPF_TXA:
 157		case BPF_LD | BPF_MEM:
 158		case BPF_LDX | BPF_MEM:
 159		case BPF_ST:
 160		case BPF_STX:
 161		case BPF_JMP | BPF_JA:
 162		case BPF_JMP | BPF_JEQ | BPF_K:
 163		case BPF_JMP | BPF_JEQ | BPF_X:
 164		case BPF_JMP | BPF_JGE | BPF_K:
 165		case BPF_JMP | BPF_JGE | BPF_X:
 166		case BPF_JMP | BPF_JGT | BPF_K:
 167		case BPF_JMP | BPF_JGT | BPF_X:
 168		case BPF_JMP | BPF_JSET | BPF_K:
 169		case BPF_JMP | BPF_JSET | BPF_X:
 170			continue;
 171		default:
 172			return -EINVAL;
 173		}
 174	}
 175	return 0;
 176}
 177
 178/**
 179 * seccomp_run_filters - evaluates all seccomp filters against @sd
 180 * @sd: optional seccomp data to be passed to filters
 181 * @match: stores struct seccomp_filter that resulted in the return value,
 182 *         unless filter returned SECCOMP_RET_ALLOW, in which case it will
 183 *         be unchanged.
 184 *
 185 * Returns valid seccomp BPF response codes.
 186 */
 187#define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
 188static u32 seccomp_run_filters(const struct seccomp_data *sd,
 189			       struct seccomp_filter **match)
 190{
 191	struct seccomp_data sd_local;
 192	u32 ret = SECCOMP_RET_ALLOW;
 193	/* Make sure cross-thread synced filter points somewhere sane. */
 194	struct seccomp_filter *f =
 195			READ_ONCE(current->seccomp.filter);
 196
 197	/* Ensure unexpected behavior doesn't result in failing open. */
 198	if (unlikely(WARN_ON(f == NULL)))
 199		return SECCOMP_RET_KILL_PROCESS;
 200
 201	if (!sd) {
 202		populate_seccomp_data(&sd_local);
 203		sd = &sd_local;
 204	}
 205
 206	/*
 207	 * All filters in the list are evaluated and the lowest BPF return
 208	 * value always takes priority (ignoring the DATA).
 209	 */
 210	for (; f; f = f->prev) {
 211		u32 cur_ret = BPF_PROG_RUN(f->prog, sd);
 212
 213		if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
 214			ret = cur_ret;
 215			*match = f;
 216		}
 217	}
 218	return ret;
 219}
 220#endif /* CONFIG_SECCOMP_FILTER */
 221
 222static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
 223{
 224	assert_spin_locked(&current->sighand->siglock);
 225
 226	if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
 227		return false;
 228
 229	return true;
 230}
 231
 232void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
 233
 234static inline void seccomp_assign_mode(struct task_struct *task,
 235				       unsigned long seccomp_mode,
 236				       unsigned long flags)
 237{
 238	assert_spin_locked(&task->sighand->siglock);
 239
 240	task->seccomp.mode = seccomp_mode;
 241	/*
 242	 * Make sure TIF_SECCOMP cannot be set before the mode (and
 243	 * filter) is set.
 244	 */
 245	smp_mb__before_atomic();
 246	/* Assume default seccomp processes want spec flaw mitigation. */
 247	if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
 248		arch_seccomp_spec_mitigate(task);
 249	set_tsk_thread_flag(task, TIF_SECCOMP);
 250}
 251
 252#ifdef CONFIG_SECCOMP_FILTER
 253/* Returns 1 if the parent is an ancestor of the child. */
 254static int is_ancestor(struct seccomp_filter *parent,
 255		       struct seccomp_filter *child)
 256{
 257	/* NULL is the root ancestor. */
 258	if (parent == NULL)
 259		return 1;
 260	for (; child; child = child->prev)
 261		if (child == parent)
 262			return 1;
 263	return 0;
 264}
 265
 266/**
 267 * seccomp_can_sync_threads: checks if all threads can be synchronized
 268 *
 269 * Expects sighand and cred_guard_mutex locks to be held.
 270 *
 271 * Returns 0 on success, -ve on error, or the pid of a thread which was
 272 * either not in the correct seccomp mode or it did not have an ancestral
 273 * seccomp filter.
 274 */
 275static inline pid_t seccomp_can_sync_threads(void)
 276{
 277	struct task_struct *thread, *caller;
 278
 279	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
 280	assert_spin_locked(&current->sighand->siglock);
 281
 282	/* Validate all threads being eligible for synchronization. */
 283	caller = current;
 284	for_each_thread(caller, thread) {
 285		pid_t failed;
 286
 287		/* Skip current, since it is initiating the sync. */
 288		if (thread == caller)
 289			continue;
 290
 291		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
 292		    (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
 293		     is_ancestor(thread->seccomp.filter,
 294				 caller->seccomp.filter)))
 295			continue;
 296
 297		/* Return the first thread that cannot be synchronized. */
 298		failed = task_pid_vnr(thread);
 299		/* If the pid cannot be resolved, then return -ESRCH */
 300		if (unlikely(WARN_ON(failed == 0)))
 301			failed = -ESRCH;
 302		return failed;
 303	}
 304
 305	return 0;
 306}
 307
 308/**
 309 * seccomp_sync_threads: sets all threads to use current's filter
 310 *
 311 * Expects sighand and cred_guard_mutex locks to be held, and for
 312 * seccomp_can_sync_threads() to have returned success already
 313 * without dropping the locks.
 314 *
 315 */
 316static inline void seccomp_sync_threads(unsigned long flags)
 317{
 318	struct task_struct *thread, *caller;
 319
 320	BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
 321	assert_spin_locked(&current->sighand->siglock);
 322
 323	/* Synchronize all threads. */
 324	caller = current;
 325	for_each_thread(caller, thread) {
 326		/* Skip current, since it needs no changes. */
 327		if (thread == caller)
 328			continue;
 329
 330		/* Get a task reference for the new leaf node. */
 331		get_seccomp_filter(caller);
 332		/*
 333		 * Drop the task reference to the shared ancestor since
 334		 * current's path will hold a reference.  (This also
 335		 * allows a put before the assignment.)
 336		 */
 337		put_seccomp_filter(thread);
 338		smp_store_release(&thread->seccomp.filter,
 339				  caller->seccomp.filter);
 340
 341		/*
 342		 * Don't let an unprivileged task work around
 343		 * the no_new_privs restriction by creating
 344		 * a thread that sets it up, enters seccomp,
 345		 * then dies.
 346		 */
 347		if (task_no_new_privs(caller))
 348			task_set_no_new_privs(thread);
 349
 350		/*
 351		 * Opt the other thread into seccomp if needed.
 352		 * As threads are considered to be trust-realm
 353		 * equivalent (see ptrace_may_access), it is safe to
 354		 * allow one thread to transition the other.
 355		 */
 356		if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
 357			seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
 358					    flags);
 359	}
 360}
 361
 362/**
 363 * seccomp_prepare_filter: Prepares a seccomp filter for use.
 364 * @fprog: BPF program to install
 365 *
 366 * Returns filter on success or an ERR_PTR on failure.
 367 */
 368static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
 369{
 370	struct seccomp_filter *sfilter;
 371	int ret;
 372	const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
 373
 374	if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
 375		return ERR_PTR(-EINVAL);
 376
 377	BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
 378
 379	/*
 380	 * Installing a seccomp filter requires that the task has
 381	 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
 382	 * This avoids scenarios where unprivileged tasks can affect the
 383	 * behavior of privileged children.
 384	 */
 385	if (!task_no_new_privs(current) &&
 386	    security_capable_noaudit(current_cred(), current_user_ns(),
 387				     CAP_SYS_ADMIN) != 0)
 388		return ERR_PTR(-EACCES);
 389
 390	/* Allocate a new seccomp_filter */
 391	sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
 392	if (!sfilter)
 393		return ERR_PTR(-ENOMEM);
 394
 395	ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
 396					seccomp_check_filter, save_orig);
 397	if (ret < 0) {
 398		kfree(sfilter);
 399		return ERR_PTR(ret);
 400	}
 401
 402	refcount_set(&sfilter->usage, 1);
 403
 404	return sfilter;
 405}
 406
 407/**
 408 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
 409 * @user_filter: pointer to the user data containing a sock_fprog.
 410 *
 411 * Returns 0 on success and non-zero otherwise.
 412 */
 413static struct seccomp_filter *
 414seccomp_prepare_user_filter(const char __user *user_filter)
 415{
 416	struct sock_fprog fprog;
 417	struct seccomp_filter *filter = ERR_PTR(-EFAULT);
 418
 419#ifdef CONFIG_COMPAT
 420	if (in_compat_syscall()) {
 421		struct compat_sock_fprog fprog32;
 422		if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
 423			goto out;
 424		fprog.len = fprog32.len;
 425		fprog.filter = compat_ptr(fprog32.filter);
 426	} else /* falls through to the if below. */
 427#endif
 428	if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
 429		goto out;
 430	filter = seccomp_prepare_filter(&fprog);
 431out:
 432	return filter;
 433}
 434
 435/**
 436 * seccomp_attach_filter: validate and attach filter
 437 * @flags:  flags to change filter behavior
 438 * @filter: seccomp filter to add to the current process
 439 *
 440 * Caller must be holding current->sighand->siglock lock.
 441 *
 442 * Returns 0 on success, -ve on error.
 443 */
 444static long seccomp_attach_filter(unsigned int flags,
 445				  struct seccomp_filter *filter)
 446{
 447	unsigned long total_insns;
 448	struct seccomp_filter *walker;
 449
 450	assert_spin_locked(&current->sighand->siglock);
 451
 452	/* Validate resulting filter length. */
 453	total_insns = filter->prog->len;
 454	for (walker = current->seccomp.filter; walker; walker = walker->prev)
 455		total_insns += walker->prog->len + 4;  /* 4 instr penalty */
 456	if (total_insns > MAX_INSNS_PER_PATH)
 457		return -ENOMEM;
 458
 459	/* If thread sync has been requested, check that it is possible. */
 460	if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
 461		int ret;
 462
 463		ret = seccomp_can_sync_threads();
 464		if (ret)
 465			return ret;
 466	}
 467
 468	/* Set log flag, if present. */
 469	if (flags & SECCOMP_FILTER_FLAG_LOG)
 470		filter->log = true;
 471
 472	/*
 473	 * If there is an existing filter, make it the prev and don't drop its
 474	 * task reference.
 475	 */
 476	filter->prev = current->seccomp.filter;
 477	current->seccomp.filter = filter;
 478
 479	/* Now that the new filter is in place, synchronize to all threads. */
 480	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
 481		seccomp_sync_threads(flags);
 482
 483	return 0;
 484}
 485
 486static void __get_seccomp_filter(struct seccomp_filter *filter)
 487{
 488	/* Reference count is bounded by the number of total processes. */
 489	refcount_inc(&filter->usage);
 490}
 491
 492/* get_seccomp_filter - increments the reference count of the filter on @tsk */
 493void get_seccomp_filter(struct task_struct *tsk)
 494{
 495	struct seccomp_filter *orig = tsk->seccomp.filter;
 496	if (!orig)
 497		return;
 498	__get_seccomp_filter(orig);
 499}
 500
 501static inline void seccomp_filter_free(struct seccomp_filter *filter)
 502{
 503	if (filter) {
 504		bpf_prog_destroy(filter->prog);
 505		kfree(filter);
 506	}
 507}
 508
 509static void __put_seccomp_filter(struct seccomp_filter *orig)
 510{
 511	/* Clean up single-reference branches iteratively. */
 512	while (orig && refcount_dec_and_test(&orig->usage)) {
 513		struct seccomp_filter *freeme = orig;
 514		orig = orig->prev;
 515		seccomp_filter_free(freeme);
 516	}
 517}
 518
 519/* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
 520void put_seccomp_filter(struct task_struct *tsk)
 521{
 522	__put_seccomp_filter(tsk->seccomp.filter);
 523}
 524
 525static void seccomp_init_siginfo(siginfo_t *info, int syscall, int reason)
 526{
 527	clear_siginfo(info);
 528	info->si_signo = SIGSYS;
 529	info->si_code = SYS_SECCOMP;
 530	info->si_call_addr = (void __user *)KSTK_EIP(current);
 531	info->si_errno = reason;
 532	info->si_arch = syscall_get_arch();
 533	info->si_syscall = syscall;
 534}
 535
 536/**
 537 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
 538 * @syscall: syscall number to send to userland
 539 * @reason: filter-supplied reason code to send to userland (via si_errno)
 540 *
 541 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
 542 */
 543static void seccomp_send_sigsys(int syscall, int reason)
 544{
 545	struct siginfo info;
 546	seccomp_init_siginfo(&info, syscall, reason);
 547	force_sig_info(SIGSYS, &info, current);
 548}
 549#endif	/* CONFIG_SECCOMP_FILTER */
 550
 551/* For use with seccomp_actions_logged */
 552#define SECCOMP_LOG_KILL_PROCESS	(1 << 0)
 553#define SECCOMP_LOG_KILL_THREAD		(1 << 1)
 554#define SECCOMP_LOG_TRAP		(1 << 2)
 555#define SECCOMP_LOG_ERRNO		(1 << 3)
 556#define SECCOMP_LOG_TRACE		(1 << 4)
 557#define SECCOMP_LOG_LOG			(1 << 5)
 558#define SECCOMP_LOG_ALLOW		(1 << 6)
 559
 560static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
 561				    SECCOMP_LOG_KILL_THREAD  |
 562				    SECCOMP_LOG_TRAP  |
 563				    SECCOMP_LOG_ERRNO |
 564				    SECCOMP_LOG_TRACE |
 565				    SECCOMP_LOG_LOG;
 566
 567static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
 568			       bool requested)
 569{
 570	bool log = false;
 571
 572	switch (action) {
 573	case SECCOMP_RET_ALLOW:
 574		break;
 575	case SECCOMP_RET_TRAP:
 576		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
 577		break;
 578	case SECCOMP_RET_ERRNO:
 579		log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
 580		break;
 581	case SECCOMP_RET_TRACE:
 582		log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
 583		break;
 584	case SECCOMP_RET_LOG:
 585		log = seccomp_actions_logged & SECCOMP_LOG_LOG;
 586		break;
 587	case SECCOMP_RET_KILL_THREAD:
 588		log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
 589		break;
 590	case SECCOMP_RET_KILL_PROCESS:
 591	default:
 592		log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
 593	}
 594
 595	/*
 596	 * Force an audit message to be emitted when the action is RET_KILL_*,
 597	 * RET_LOG, or the FILTER_FLAG_LOG bit was set and the action is
 598	 * allowed to be logged by the admin.
 599	 */
 600	if (log)
 601		return __audit_seccomp(syscall, signr, action);
 602
 603	/*
 604	 * Let the audit subsystem decide if the action should be audited based
 605	 * on whether the current task itself is being audited.
 606	 */
 607	return audit_seccomp(syscall, signr, action);
 608}
 609
 610/*
 611 * Secure computing mode 1 allows only read/write/exit/sigreturn.
 612 * To be fully secure this must be combined with rlimit
 613 * to limit the stack allocations too.
 614 */
 615static const int mode1_syscalls[] = {
 616	__NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
 617	0, /* null terminated */
 618};
 619
 620static void __secure_computing_strict(int this_syscall)
 621{
 622	const int *syscall_whitelist = mode1_syscalls;
 623#ifdef CONFIG_COMPAT
 624	if (in_compat_syscall())
 625		syscall_whitelist = get_compat_mode1_syscalls();
 626#endif
 627	do {
 628		if (*syscall_whitelist == this_syscall)
 629			return;
 630	} while (*++syscall_whitelist);
 631
 632#ifdef SECCOMP_DEBUG
 633	dump_stack();
 634#endif
 635	seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
 636	do_exit(SIGKILL);
 637}
 638
 639#ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
 640void secure_computing_strict(int this_syscall)
 641{
 642	int mode = current->seccomp.mode;
 643
 644	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
 645	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
 646		return;
 647
 648	if (mode == SECCOMP_MODE_DISABLED)
 649		return;
 650	else if (mode == SECCOMP_MODE_STRICT)
 651		__secure_computing_strict(this_syscall);
 652	else
 653		BUG();
 654}
 655#else
 656
 657#ifdef CONFIG_SECCOMP_FILTER
 658static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
 659			    const bool recheck_after_trace)
 660{
 661	u32 filter_ret, action;
 662	struct seccomp_filter *match = NULL;
 663	int data;
 664
 665	/*
 666	 * Make sure that any changes to mode from another thread have
 667	 * been seen after TIF_SECCOMP was seen.
 668	 */
 669	rmb();
 670
 671	filter_ret = seccomp_run_filters(sd, &match);
 672	data = filter_ret & SECCOMP_RET_DATA;
 673	action = filter_ret & SECCOMP_RET_ACTION_FULL;
 674
 675	switch (action) {
 676	case SECCOMP_RET_ERRNO:
 677		/* Set low-order bits as an errno, capped at MAX_ERRNO. */
 678		if (data > MAX_ERRNO)
 679			data = MAX_ERRNO;
 680		syscall_set_return_value(current, task_pt_regs(current),
 681					 -data, 0);
 682		goto skip;
 683
 684	case SECCOMP_RET_TRAP:
 685		/* Show the handler the original registers. */
 686		syscall_rollback(current, task_pt_regs(current));
 687		/* Let the filter pass back 16 bits of data. */
 688		seccomp_send_sigsys(this_syscall, data);
 689		goto skip;
 690
 691	case SECCOMP_RET_TRACE:
 692		/* We've been put in this state by the ptracer already. */
 693		if (recheck_after_trace)
 694			return 0;
 695
 696		/* ENOSYS these calls if there is no tracer attached. */
 697		if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
 698			syscall_set_return_value(current,
 699						 task_pt_regs(current),
 700						 -ENOSYS, 0);
 701			goto skip;
 702		}
 703
 704		/* Allow the BPF to provide the event message */
 705		ptrace_event(PTRACE_EVENT_SECCOMP, data);
 706		/*
 707		 * The delivery of a fatal signal during event
 708		 * notification may silently skip tracer notification,
 709		 * which could leave us with a potentially unmodified
 710		 * syscall that the tracer would have liked to have
 711		 * changed. Since the process is about to die, we just
 712		 * force the syscall to be skipped and let the signal
 713		 * kill the process and correctly handle any tracer exit
 714		 * notifications.
 715		 */
 716		if (fatal_signal_pending(current))
 717			goto skip;
 718		/* Check if the tracer forced the syscall to be skipped. */
 719		this_syscall = syscall_get_nr(current, task_pt_regs(current));
 720		if (this_syscall < 0)
 721			goto skip;
 722
 723		/*
 724		 * Recheck the syscall, since it may have changed. This
 725		 * intentionally uses a NULL struct seccomp_data to force
 726		 * a reload of all registers. This does not goto skip since
 727		 * a skip would have already been reported.
 728		 */
 729		if (__seccomp_filter(this_syscall, NULL, true))
 730			return -1;
 731
 732		return 0;
 733
 734	case SECCOMP_RET_LOG:
 735		seccomp_log(this_syscall, 0, action, true);
 736		return 0;
 737
 738	case SECCOMP_RET_ALLOW:
 739		/*
 740		 * Note that the "match" filter will always be NULL for
 741		 * this action since SECCOMP_RET_ALLOW is the starting
 742		 * state in seccomp_run_filters().
 743		 */
 744		return 0;
 745
 746	case SECCOMP_RET_KILL_THREAD:
 747	case SECCOMP_RET_KILL_PROCESS:
 748	default:
 749		seccomp_log(this_syscall, SIGSYS, action, true);
 750		/* Dump core only if this is the last remaining thread. */
 751		if (action == SECCOMP_RET_KILL_PROCESS ||
 752		    get_nr_threads(current) == 1) {
 753			siginfo_t info;
 754
 755			/* Show the original registers in the dump. */
 756			syscall_rollback(current, task_pt_regs(current));
 757			/* Trigger a manual coredump since do_exit skips it. */
 758			seccomp_init_siginfo(&info, this_syscall, data);
 759			do_coredump(&info);
 760		}
 761		if (action == SECCOMP_RET_KILL_PROCESS)
 762			do_group_exit(SIGSYS);
 763		else
 764			do_exit(SIGSYS);
 765	}
 766
 767	unreachable();
 768
 769skip:
 770	seccomp_log(this_syscall, 0, action, match ? match->log : false);
 771	return -1;
 772}
 773#else
 774static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
 775			    const bool recheck_after_trace)
 776{
 777	BUG();
 778}
 779#endif
 780
 781int __secure_computing(const struct seccomp_data *sd)
 782{
 783	int mode = current->seccomp.mode;
 784	int this_syscall;
 785
 786	if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
 787	    unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
 788		return 0;
 789
 790	this_syscall = sd ? sd->nr :
 791		syscall_get_nr(current, task_pt_regs(current));
 792
 793	switch (mode) {
 794	case SECCOMP_MODE_STRICT:
 795		__secure_computing_strict(this_syscall);  /* may call do_exit */
 796		return 0;
 797	case SECCOMP_MODE_FILTER:
 798		return __seccomp_filter(this_syscall, sd, false);
 
 
 
 
 
 
 799	default:
 800		BUG();
 801	}
 
 
 
 
 
 802}
 803#endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
 804
 805long prctl_get_seccomp(void)
 806{
 807	return current->seccomp.mode;
 808}
 809
 810/**
 811 * seccomp_set_mode_strict: internal function for setting strict seccomp
 812 *
 813 * Once current->seccomp.mode is non-zero, it may not be changed.
 814 *
 815 * Returns 0 on success or -EINVAL on failure.
 816 */
 817static long seccomp_set_mode_strict(void)
 818{
 819	const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
 820	long ret = -EINVAL;
 821
 822	spin_lock_irq(&current->sighand->siglock);
 823
 824	if (!seccomp_may_assign_mode(seccomp_mode))
 
 
 825		goto out;
 826
 
 
 
 
 827#ifdef TIF_NOTSC
 828	disable_TSC();
 829#endif
 830	seccomp_assign_mode(current, seccomp_mode, 0);
 831	ret = 0;
 832
 833out:
 834	spin_unlock_irq(&current->sighand->siglock);
 835
 836	return ret;
 837}
 838
 839#ifdef CONFIG_SECCOMP_FILTER
 840/**
 841 * seccomp_set_mode_filter: internal function for setting seccomp filter
 842 * @flags:  flags to change filter behavior
 843 * @filter: struct sock_fprog containing filter
 844 *
 845 * This function may be called repeatedly to install additional filters.
 846 * Every filter successfully installed will be evaluated (in reverse order)
 847 * for each system call the task makes.
 848 *
 849 * Once current->seccomp.mode is non-zero, it may not be changed.
 850 *
 851 * Returns 0 on success or -EINVAL on failure.
 852 */
 853static long seccomp_set_mode_filter(unsigned int flags,
 854				    const char __user *filter)
 855{
 856	const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
 857	struct seccomp_filter *prepared = NULL;
 858	long ret = -EINVAL;
 859
 860	/* Validate flags. */
 861	if (flags & ~SECCOMP_FILTER_FLAG_MASK)
 862		return -EINVAL;
 863
 864	/* Prepare the new filter before holding any locks. */
 865	prepared = seccomp_prepare_user_filter(filter);
 866	if (IS_ERR(prepared))
 867		return PTR_ERR(prepared);
 868
 869	/*
 870	 * Make sure we cannot change seccomp or nnp state via TSYNC
 871	 * while another thread is in the middle of calling exec.
 872	 */
 873	if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
 874	    mutex_lock_killable(&current->signal->cred_guard_mutex))
 875		goto out_free;
 876
 877	spin_lock_irq(&current->sighand->siglock);
 878
 879	if (!seccomp_may_assign_mode(seccomp_mode))
 880		goto out;
 881
 882	ret = seccomp_attach_filter(flags, prepared);
 883	if (ret)
 884		goto out;
 885	/* Do not free the successfully attached filter. */
 886	prepared = NULL;
 887
 888	seccomp_assign_mode(current, seccomp_mode, flags);
 889out:
 890	spin_unlock_irq(&current->sighand->siglock);
 891	if (flags & SECCOMP_FILTER_FLAG_TSYNC)
 892		mutex_unlock(&current->signal->cred_guard_mutex);
 893out_free:
 894	seccomp_filter_free(prepared);
 895	return ret;
 896}
 897#else
 898static inline long seccomp_set_mode_filter(unsigned int flags,
 899					   const char __user *filter)
 900{
 901	return -EINVAL;
 902}
 903#endif
 904
 905static long seccomp_get_action_avail(const char __user *uaction)
 906{
 907	u32 action;
 908
 909	if (copy_from_user(&action, uaction, sizeof(action)))
 910		return -EFAULT;
 911
 912	switch (action) {
 913	case SECCOMP_RET_KILL_PROCESS:
 914	case SECCOMP_RET_KILL_THREAD:
 915	case SECCOMP_RET_TRAP:
 916	case SECCOMP_RET_ERRNO:
 917	case SECCOMP_RET_TRACE:
 918	case SECCOMP_RET_LOG:
 919	case SECCOMP_RET_ALLOW:
 920		break;
 921	default:
 922		return -EOPNOTSUPP;
 923	}
 924
 925	return 0;
 926}
 927
 928/* Common entry point for both prctl and syscall. */
 929static long do_seccomp(unsigned int op, unsigned int flags,
 930		       const char __user *uargs)
 931{
 932	switch (op) {
 933	case SECCOMP_SET_MODE_STRICT:
 934		if (flags != 0 || uargs != NULL)
 935			return -EINVAL;
 936		return seccomp_set_mode_strict();
 937	case SECCOMP_SET_MODE_FILTER:
 938		return seccomp_set_mode_filter(flags, uargs);
 939	case SECCOMP_GET_ACTION_AVAIL:
 940		if (flags != 0)
 941			return -EINVAL;
 942
 943		return seccomp_get_action_avail(uargs);
 944	default:
 945		return -EINVAL;
 946	}
 947}
 948
 949SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
 950			 const char __user *, uargs)
 951{
 952	return do_seccomp(op, flags, uargs);
 953}
 954
 955/**
 956 * prctl_set_seccomp: configures current->seccomp.mode
 957 * @seccomp_mode: requested mode to use
 958 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
 959 *
 960 * Returns 0 on success or -EINVAL on failure.
 961 */
 962long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
 963{
 964	unsigned int op;
 965	char __user *uargs;
 966
 967	switch (seccomp_mode) {
 968	case SECCOMP_MODE_STRICT:
 969		op = SECCOMP_SET_MODE_STRICT;
 970		/*
 971		 * Setting strict mode through prctl always ignored filter,
 972		 * so make sure it is always NULL here to pass the internal
 973		 * check in do_seccomp().
 974		 */
 975		uargs = NULL;
 976		break;
 977	case SECCOMP_MODE_FILTER:
 978		op = SECCOMP_SET_MODE_FILTER;
 979		uargs = filter;
 980		break;
 981	default:
 982		return -EINVAL;
 983	}
 984
 985	/* prctl interface doesn't have flags, so they are always zero. */
 986	return do_seccomp(op, 0, uargs);
 987}
 988
 989#if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
 990static struct seccomp_filter *get_nth_filter(struct task_struct *task,
 991					     unsigned long filter_off)
 992{
 993	struct seccomp_filter *orig, *filter;
 994	unsigned long count;
 995
 996	/*
 997	 * Note: this is only correct because the caller should be the (ptrace)
 998	 * tracer of the task, otherwise lock_task_sighand is needed.
 999	 */
1000	spin_lock_irq(&task->sighand->siglock);
1001
1002	if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
1003		spin_unlock_irq(&task->sighand->siglock);
1004		return ERR_PTR(-EINVAL);
1005	}
1006
1007	orig = task->seccomp.filter;
1008	__get_seccomp_filter(orig);
1009	spin_unlock_irq(&task->sighand->siglock);
1010
1011	count = 0;
1012	for (filter = orig; filter; filter = filter->prev)
1013		count++;
1014
1015	if (filter_off >= count) {
1016		filter = ERR_PTR(-ENOENT);
1017		goto out;
1018	}
1019
1020	count -= filter_off;
1021	for (filter = orig; filter && count > 1; filter = filter->prev)
1022		count--;
1023
1024	if (WARN_ON(count != 1 || !filter)) {
1025		filter = ERR_PTR(-ENOENT);
1026		goto out;
1027	}
1028
1029	__get_seccomp_filter(filter);
1030
1031out:
1032	__put_seccomp_filter(orig);
1033	return filter;
1034}
1035
1036long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
1037			void __user *data)
1038{
1039	struct seccomp_filter *filter;
1040	struct sock_fprog_kern *fprog;
1041	long ret;
1042
1043	if (!capable(CAP_SYS_ADMIN) ||
1044	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1045		return -EACCES;
1046	}
1047
1048	filter = get_nth_filter(task, filter_off);
1049	if (IS_ERR(filter))
1050		return PTR_ERR(filter);
1051
1052	fprog = filter->prog->orig_prog;
1053	if (!fprog) {
1054		/* This must be a new non-cBPF filter, since we save
1055		 * every cBPF filter's orig_prog above when
1056		 * CONFIG_CHECKPOINT_RESTORE is enabled.
1057		 */
1058		ret = -EMEDIUMTYPE;
1059		goto out;
1060	}
1061
1062	ret = fprog->len;
1063	if (!data)
1064		goto out;
1065
1066	if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1067		ret = -EFAULT;
1068
1069out:
1070	__put_seccomp_filter(filter);
1071	return ret;
1072}
1073
1074long seccomp_get_metadata(struct task_struct *task,
1075			  unsigned long size, void __user *data)
1076{
1077	long ret;
1078	struct seccomp_filter *filter;
1079	struct seccomp_metadata kmd = {};
1080
1081	if (!capable(CAP_SYS_ADMIN) ||
1082	    current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1083		return -EACCES;
1084	}
1085
1086	size = min_t(unsigned long, size, sizeof(kmd));
1087
1088	if (size < sizeof(kmd.filter_off))
1089		return -EINVAL;
1090
1091	if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
1092		return -EFAULT;
1093
1094	filter = get_nth_filter(task, kmd.filter_off);
1095	if (IS_ERR(filter))
1096		return PTR_ERR(filter);
1097
1098	if (filter->log)
1099		kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
1100
1101	ret = size;
1102	if (copy_to_user(data, &kmd, size))
1103		ret = -EFAULT;
1104
1105	__put_seccomp_filter(filter);
1106	return ret;
1107}
1108#endif
1109
1110#ifdef CONFIG_SYSCTL
1111
1112/* Human readable action names for friendly sysctl interaction */
1113#define SECCOMP_RET_KILL_PROCESS_NAME	"kill_process"
1114#define SECCOMP_RET_KILL_THREAD_NAME	"kill_thread"
1115#define SECCOMP_RET_TRAP_NAME		"trap"
1116#define SECCOMP_RET_ERRNO_NAME		"errno"
1117#define SECCOMP_RET_TRACE_NAME		"trace"
1118#define SECCOMP_RET_LOG_NAME		"log"
1119#define SECCOMP_RET_ALLOW_NAME		"allow"
1120
1121static const char seccomp_actions_avail[] =
1122				SECCOMP_RET_KILL_PROCESS_NAME	" "
1123				SECCOMP_RET_KILL_THREAD_NAME	" "
1124				SECCOMP_RET_TRAP_NAME		" "
1125				SECCOMP_RET_ERRNO_NAME		" "
1126				SECCOMP_RET_TRACE_NAME		" "
1127				SECCOMP_RET_LOG_NAME		" "
1128				SECCOMP_RET_ALLOW_NAME;
1129
1130struct seccomp_log_name {
1131	u32		log;
1132	const char	*name;
1133};
1134
1135static const struct seccomp_log_name seccomp_log_names[] = {
1136	{ SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
1137	{ SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1138	{ SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1139	{ SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1140	{ SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1141	{ SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1142	{ SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1143	{ }
1144};
1145
1146static bool seccomp_names_from_actions_logged(char *names, size_t size,
1147					      u32 actions_logged)
1148{
1149	const struct seccomp_log_name *cur;
1150	bool append_space = false;
1151
1152	for (cur = seccomp_log_names; cur->name && size; cur++) {
1153		ssize_t ret;
1154
1155		if (!(actions_logged & cur->log))
1156			continue;
1157
1158		if (append_space) {
1159			ret = strscpy(names, " ", size);
1160			if (ret < 0)
1161				return false;
1162
1163			names += ret;
1164			size -= ret;
1165		} else
1166			append_space = true;
1167
1168		ret = strscpy(names, cur->name, size);
1169		if (ret < 0)
1170			return false;
1171
1172		names += ret;
1173		size -= ret;
1174	}
1175
1176	return true;
1177}
1178
1179static bool seccomp_action_logged_from_name(u32 *action_logged,
1180					    const char *name)
1181{
1182	const struct seccomp_log_name *cur;
1183
1184	for (cur = seccomp_log_names; cur->name; cur++) {
1185		if (!strcmp(cur->name, name)) {
1186			*action_logged = cur->log;
1187			return true;
1188		}
1189	}
1190
1191	return false;
1192}
1193
1194static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1195{
1196	char *name;
1197
1198	*actions_logged = 0;
1199	while ((name = strsep(&names, " ")) && *name) {
1200		u32 action_logged = 0;
1201
1202		if (!seccomp_action_logged_from_name(&action_logged, name))
1203			return false;
1204
1205		*actions_logged |= action_logged;
1206	}
1207
1208	return true;
1209}
1210
1211static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
1212					  void __user *buffer, size_t *lenp,
1213					  loff_t *ppos)
1214{
1215	char names[sizeof(seccomp_actions_avail)];
1216	struct ctl_table table;
1217	int ret;
1218
1219	if (write && !capable(CAP_SYS_ADMIN))
1220		return -EPERM;
1221
1222	memset(names, 0, sizeof(names));
1223
1224	if (!write) {
1225		if (!seccomp_names_from_actions_logged(names, sizeof(names),
1226						       seccomp_actions_logged))
1227			return -EINVAL;
1228	}
1229
1230	table = *ro_table;
1231	table.data = names;
1232	table.maxlen = sizeof(names);
1233	ret = proc_dostring(&table, write, buffer, lenp, ppos);
1234	if (ret)
1235		return ret;
1236
1237	if (write) {
1238		u32 actions_logged;
1239
1240		if (!seccomp_actions_logged_from_names(&actions_logged,
1241						       table.data))
1242			return -EINVAL;
1243
1244		if (actions_logged & SECCOMP_LOG_ALLOW)
1245			return -EINVAL;
1246
1247		seccomp_actions_logged = actions_logged;
1248	}
1249
1250	return 0;
1251}
1252
1253static struct ctl_path seccomp_sysctl_path[] = {
1254	{ .procname = "kernel", },
1255	{ .procname = "seccomp", },
1256	{ }
1257};
1258
1259static struct ctl_table seccomp_sysctl_table[] = {
1260	{
1261		.procname	= "actions_avail",
1262		.data		= (void *) &seccomp_actions_avail,
1263		.maxlen		= sizeof(seccomp_actions_avail),
1264		.mode		= 0444,
1265		.proc_handler	= proc_dostring,
1266	},
1267	{
1268		.procname	= "actions_logged",
1269		.mode		= 0644,
1270		.proc_handler	= seccomp_actions_logged_handler,
1271	},
1272	{ }
1273};
1274
1275static int __init seccomp_sysctl_init(void)
1276{
1277	struct ctl_table_header *hdr;
1278
1279	hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
1280	if (!hdr)
1281		pr_warn("seccomp: sysctl registration failed\n");
1282	else
1283		kmemleak_not_leak(hdr);
1284
1285	return 0;
1286}
1287
1288device_initcall(seccomp_sysctl_init)
1289
1290#endif /* CONFIG_SYSCTL */