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