projects/linux·blob v5.10 · refs

2105 lines · 54.2 KiB

1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/kernel/seccomp.c
4 *
5 * Copyright 2004-2005 Andrea Arcangeli <[email protected]>
6 *
7 * Copyright (C) 2012 Google, Inc.
8 * Will Drewry <[email protected]>
9 *
10 * This defines a simple but solid secure-computing facility.
11 *
12 * Mode 1 uses a fixed list of allowed system calls.
13 * Mode 2 allows user-defined system call filters in the form
14 * of Berkeley Packet Filters/Linux Socket Filters.
15 */
16#define pr_fmt(fmt) "seccomp: " fmt
17
18#include <linux/refcount.h>
19#include <linux/audit.h>
20#include <linux/compat.h>
21#include <linux/coredump.h>
22#include <linux/kmemleak.h>
23#include <linux/nospec.h>
24#include <linux/prctl.h>
25#include <linux/sched.h>
26#include <linux/sched/task_stack.h>
27#include <linux/seccomp.h>
28#include <linux/slab.h>
29#include <linux/syscalls.h>
30#include <linux/sysctl.h>
31
32#ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
33#include <asm/syscall.h>
34#endif
35
36#ifdef CONFIG_SECCOMP_FILTER
37#include <linux/file.h>
38#include <linux/filter.h>
39#include <linux/pid.h>
40#include <linux/ptrace.h>
41#include <linux/capability.h>
42#include <linux/tracehook.h>
43#include <linux/uaccess.h>
44#include <linux/anon_inodes.h>
45#include <linux/lockdep.h>
46
47/*
48 * When SECCOMP_IOCTL_NOTIF_ID_VALID was first introduced, it had the
49 * wrong direction flag in the ioctl number. This is the broken one,
50 * which the kernel needs to keep supporting until all userspaces stop
51 * using the wrong command number.
52 */
53#define SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR SECCOMP_IOR(2, __u64)
54
55enum notify_state {
56 SECCOMP_NOTIFY_INIT,
57 SECCOMP_NOTIFY_SENT,
58 SECCOMP_NOTIFY_REPLIED,
59};
60
61struct seccomp_knotif {
62 /* The struct pid of the task whose filter triggered the notification */
63 struct task_struct *task;
64
65 /* The "cookie" for this request; this is unique for this filter. */
66 u64 id;
67
68 /*
69 * The seccomp data. This pointer is valid the entire time this
70 * notification is active, since it comes from __seccomp_filter which
71 * eclipses the entire lifecycle here.
72 */
73 const struct seccomp_data *data;
74
75 /*
76 * Notification states. When SECCOMP_RET_USER_NOTIF is returned, a
77 * struct seccomp_knotif is created and starts out in INIT. Once the
78 * handler reads the notification off of an FD, it transitions to SENT.
79 * If a signal is received the state transitions back to INIT and
80 * another message is sent. When the userspace handler replies, state
81 * transitions to REPLIED.
82 */
83 enum notify_state state;
84
85 /* The return values, only valid when in SECCOMP_NOTIFY_REPLIED */
86 int error;
87 long val;
88 u32 flags;
89
90 /*
91 * Signals when this has changed states, such as the listener
92 * dying, a new seccomp addfd message, or changing to REPLIED
93 */
94 struct completion ready;
95
96 struct list_head list;
97
98 /* outstanding addfd requests */
99 struct list_head addfd;
100};
101
102/**
103 * struct seccomp_kaddfd - container for seccomp_addfd ioctl messages
104 *
105 * @file: A reference to the file to install in the other task
106 * @fd: The fd number to install it at. If the fd number is -1, it means the
107 * installing process should allocate the fd as normal.
108 * @flags: The flags for the new file descriptor. At the moment, only O_CLOEXEC
109 * is allowed.
110 * @ret: The return value of the installing process. It is set to the fd num
111 * upon success (>= 0).
112 * @completion: Indicates that the installing process has completed fd
113 * installation, or gone away (either due to successful
114 * reply, or signal)
115 *
116 */
117struct seccomp_kaddfd {
118 struct file *file;
119 int fd;
120 unsigned int flags;
121
122 /* To only be set on reply */
123 int ret;
124 struct completion completion;
125 struct list_head list;
126};
127
128/**
129 * struct notification - container for seccomp userspace notifications. Since
130 * most seccomp filters will not have notification listeners attached and this
131 * structure is fairly large, we store the notification-specific stuff in a
132 * separate structure.
133 *
134 * @request: A semaphore that users of this notification can wait on for
135 * changes. Actual reads and writes are still controlled with
136 * filter->notify_lock.
137 * @next_id: The id of the next request.
138 * @notifications: A list of struct seccomp_knotif elements.
139 */
140struct notification {
141 struct semaphore request;
142 u64 next_id;
143 struct list_head notifications;
144};
145
146/**
147 * struct seccomp_filter - container for seccomp BPF programs
148 *
149 * @refs: Reference count to manage the object lifetime.
150 * A filter's reference count is incremented for each directly
151 * attached task, once for the dependent filter, and if
152 * requested for the user notifier. When @refs reaches zero,
153 * the filter can be freed.
154 * @users: A filter's @users count is incremented for each directly
155 * attached task (filter installation, fork(), thread_sync),
156 * and once for the dependent filter (tracked in filter->prev).
157 * When it reaches zero it indicates that no direct or indirect
158 * users of that filter exist. No new tasks can get associated with
159 * this filter after reaching 0. The @users count is always smaller
160 * or equal to @refs. Hence, reaching 0 for @users does not mean
161 * the filter can be freed.
162 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
163 * @prev: points to a previously installed, or inherited, filter
164 * @prog: the BPF program to evaluate
165 * @notif: the struct that holds all notification related information
166 * @notify_lock: A lock for all notification-related accesses.
167 * @wqh: A wait queue for poll if a notifier is in use.
168 *
169 * seccomp_filter objects are organized in a tree linked via the @prev
170 * pointer. For any task, it appears to be a singly-linked list starting
171 * with current->seccomp.filter, the most recently attached or inherited filter.
172 * However, multiple filters may share a @prev node, by way of fork(), which
173 * results in a unidirectional tree existing in memory. This is similar to
174 * how namespaces work.
175 *
176 * seccomp_filter objects should never be modified after being attached
177 * to a task_struct (other than @refs).
178 */
179struct seccomp_filter {
180 refcount_t refs;
181 refcount_t users;
182 bool log;
183 struct seccomp_filter *prev;
184 struct bpf_prog *prog;
185 struct notification *notif;
186 struct mutex notify_lock;
187 wait_queue_head_t wqh;
188};
189
190/* Limit any path through the tree to 256KB worth of instructions. */
191#define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
192
193/*
194 * Endianness is explicitly ignored and left for BPF program authors to manage
195 * as per the specific architecture.
196 */
197static void populate_seccomp_data(struct seccomp_data *sd)
198{
199 /*
200 * Instead of using current_pt_reg(), we're already doing the work
201 * to safely fetch "current", so just use "task" everywhere below.
202 */
203 struct task_struct *task = current;
204 struct pt_regs *regs = task_pt_regs(task);
205 unsigned long args[6];
206
207 sd->nr = syscall_get_nr(task, regs);
208 sd->arch = syscall_get_arch(task);
209 syscall_get_arguments(task, regs, args);
210 sd->args[0] = args[0];
211 sd->args[1] = args[1];
212 sd->args[2] = args[2];
213 sd->args[3] = args[3];
214 sd->args[4] = args[4];
215 sd->args[5] = args[5];
216 sd->instruction_pointer = KSTK_EIP(task);
217}
218
219/**
220 * seccomp_check_filter - verify seccomp filter code
221 * @filter: filter to verify
222 * @flen: length of filter
223 *
224 * Takes a previously checked filter (by bpf_check_classic) and
225 * redirects all filter code that loads struct sk_buff data
226 * and related data through seccomp_bpf_load. It also
227 * enforces length and alignment checking of those loads.
228 *
229 * Returns 0 if the rule set is legal or -EINVAL if not.
230 */
231static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
232{
233 int pc;
234 for (pc = 0; pc < flen; pc++) {
235 struct sock_filter *ftest = &filter[pc];
236 u16 code = ftest->code;
237 u32 k = ftest->k;
238
239 switch (code) {
240 case BPF_LD | BPF_W | BPF_ABS:
241 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
242 /* 32-bit aligned and not out of bounds. */
243 if (k >= sizeof(struct seccomp_data) || k & 3)
244 return -EINVAL;
245 continue;
246 case BPF_LD | BPF_W | BPF_LEN:
247 ftest->code = BPF_LD | BPF_IMM;
248 ftest->k = sizeof(struct seccomp_data);
249 continue;
250 case BPF_LDX | BPF_W | BPF_LEN:
251 ftest->code = BPF_LDX | BPF_IMM;
252 ftest->k = sizeof(struct seccomp_data);
253 continue;
254 /* Explicitly include allowed calls. */
255 case BPF_RET | BPF_K:
256 case BPF_RET | BPF_A:
257 case BPF_ALU | BPF_ADD | BPF_K:
258 case BPF_ALU | BPF_ADD | BPF_X:
259 case BPF_ALU | BPF_SUB | BPF_K:
260 case BPF_ALU | BPF_SUB | BPF_X:
261 case BPF_ALU | BPF_MUL | BPF_K:
262 case BPF_ALU | BPF_MUL | BPF_X:
263 case BPF_ALU | BPF_DIV | BPF_K:
264 case BPF_ALU | BPF_DIV | BPF_X:
265 case BPF_ALU | BPF_AND | BPF_K:
266 case BPF_ALU | BPF_AND | BPF_X:
267 case BPF_ALU | BPF_OR | BPF_K:
268 case BPF_ALU | BPF_OR | BPF_X:
269 case BPF_ALU | BPF_XOR | BPF_K:
270 case BPF_ALU | BPF_XOR | BPF_X:
271 case BPF_ALU | BPF_LSH | BPF_K:
272 case BPF_ALU | BPF_LSH | BPF_X:
273 case BPF_ALU | BPF_RSH | BPF_K:
274 case BPF_ALU | BPF_RSH | BPF_X:
275 case BPF_ALU | BPF_NEG:
276 case BPF_LD | BPF_IMM:
277 case BPF_LDX | BPF_IMM:
278 case BPF_MISC | BPF_TAX:
279 case BPF_MISC | BPF_TXA:
280 case BPF_LD | BPF_MEM:
281 case BPF_LDX | BPF_MEM:
282 case BPF_ST:
283 case BPF_STX:
284 case BPF_JMP | BPF_JA:
285 case BPF_JMP | BPF_JEQ | BPF_K:
286 case BPF_JMP | BPF_JEQ | BPF_X:
287 case BPF_JMP | BPF_JGE | BPF_K:
288 case BPF_JMP | BPF_JGE | BPF_X:
289 case BPF_JMP | BPF_JGT | BPF_K:
290 case BPF_JMP | BPF_JGT | BPF_X:
291 case BPF_JMP | BPF_JSET | BPF_K:
292 case BPF_JMP | BPF_JSET | BPF_X:
293 continue;
294 default:
295 return -EINVAL;
296 }
297 }
298 return 0;
299}
300
301/**
302 * seccomp_run_filters - evaluates all seccomp filters against @sd
303 * @sd: optional seccomp data to be passed to filters
304 * @match: stores struct seccomp_filter that resulted in the return value,
305 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
306 * be unchanged.
307 *
308 * Returns valid seccomp BPF response codes.
309 */
310#define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
311static u32 seccomp_run_filters(const struct seccomp_data *sd,
312 struct seccomp_filter **match)
313{
314 u32 ret = SECCOMP_RET_ALLOW;
315 /* Make sure cross-thread synced filter points somewhere sane. */
316 struct seccomp_filter *f =
317 READ_ONCE(current->seccomp.filter);
318
319 /* Ensure unexpected behavior doesn't result in failing open. */
320 if (WARN_ON(f == NULL))
321 return SECCOMP_RET_KILL_PROCESS;
322
323 /*
324 * All filters in the list are evaluated and the lowest BPF return
325 * value always takes priority (ignoring the DATA).
326 */
327 for (; f; f = f->prev) {
328 u32 cur_ret = bpf_prog_run_pin_on_cpu(f->prog, sd);
329
330 if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
331 ret = cur_ret;
332 *match = f;
333 }
334 }
335 return ret;
336}
337#endif /* CONFIG_SECCOMP_FILTER */
338
339static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
340{
341 assert_spin_locked(&current->sighand->siglock);
342
343 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
344 return false;
345
346 return true;
347}
348
349void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
350
351static inline void seccomp_assign_mode(struct task_struct *task,
352 unsigned long seccomp_mode,
353 unsigned long flags)
354{
355 assert_spin_locked(&task->sighand->siglock);
356
357 task->seccomp.mode = seccomp_mode;
358 /*
359 * Make sure TIF_SECCOMP cannot be set before the mode (and
360 * filter) is set.
361 */
362 smp_mb__before_atomic();
363 /* Assume default seccomp processes want spec flaw mitigation. */
364 if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
365 arch_seccomp_spec_mitigate(task);
366 set_tsk_thread_flag(task, TIF_SECCOMP);
367}
368
369#ifdef CONFIG_SECCOMP_FILTER
370/* Returns 1 if the parent is an ancestor of the child. */
371static int is_ancestor(struct seccomp_filter *parent,
372 struct seccomp_filter *child)
373{
374 /* NULL is the root ancestor. */
375 if (parent == NULL)
376 return 1;
377 for (; child; child = child->prev)
378 if (child == parent)
379 return 1;
380 return 0;
381}
382
383/**
384 * seccomp_can_sync_threads: checks if all threads can be synchronized
385 *
386 * Expects sighand and cred_guard_mutex locks to be held.
387 *
388 * Returns 0 on success, -ve on error, or the pid of a thread which was
389 * either not in the correct seccomp mode or did not have an ancestral
390 * seccomp filter.
391 */
392static inline pid_t seccomp_can_sync_threads(void)
393{
394 struct task_struct *thread, *caller;
395
396 BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
397 assert_spin_locked(&current->sighand->siglock);
398
399 /* Validate all threads being eligible for synchronization. */
400 caller = current;
401 for_each_thread(caller, thread) {
402 pid_t failed;
403
404 /* Skip current, since it is initiating the sync. */
405 if (thread == caller)
406 continue;
407
408 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
409 (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
410 is_ancestor(thread->seccomp.filter,
411 caller->seccomp.filter)))
412 continue;
413
414 /* Return the first thread that cannot be synchronized. */
415 failed = task_pid_vnr(thread);
416 /* If the pid cannot be resolved, then return -ESRCH */
417 if (WARN_ON(failed == 0))
418 failed = -ESRCH;
419 return failed;
420 }
421
422 return 0;
423}
424
425static inline void seccomp_filter_free(struct seccomp_filter *filter)
426{
427 if (filter) {
428 bpf_prog_destroy(filter->prog);
429 kfree(filter);
430 }
431}
432
433static void __seccomp_filter_orphan(struct seccomp_filter *orig)
434{
435 while (orig && refcount_dec_and_test(&orig->users)) {
436 if (waitqueue_active(&orig->wqh))
437 wake_up_poll(&orig->wqh, EPOLLHUP);
438 orig = orig->prev;
439 }
440}
441
442static void __put_seccomp_filter(struct seccomp_filter *orig)
443{
444 /* Clean up single-reference branches iteratively. */
445 while (orig && refcount_dec_and_test(&orig->refs)) {
446 struct seccomp_filter *freeme = orig;
447 orig = orig->prev;
448 seccomp_filter_free(freeme);
449 }
450}
451
452static void __seccomp_filter_release(struct seccomp_filter *orig)
453{
454 /* Notify about any unused filters in the task's former filter tree. */
455 __seccomp_filter_orphan(orig);
456 /* Finally drop all references to the task's former tree. */
457 __put_seccomp_filter(orig);
458}
459
460/**
461 * seccomp_filter_release - Detach the task from its filter tree,
462 * drop its reference count, and notify
463 * about unused filters
464 *
465 * This function should only be called when the task is exiting as
466 * it detaches it from its filter tree. As such, READ_ONCE() and
467 * barriers are not needed here, as would normally be needed.
468 */
469void seccomp_filter_release(struct task_struct *tsk)
470{
471 struct seccomp_filter *orig = tsk->seccomp.filter;
472
473 /* Detach task from its filter tree. */
474 tsk->seccomp.filter = NULL;
475 __seccomp_filter_release(orig);
476}
477
478/**
479 * seccomp_sync_threads: sets all threads to use current's filter
480 *
481 * Expects sighand and cred_guard_mutex locks to be held, and for
482 * seccomp_can_sync_threads() to have returned success already
483 * without dropping the locks.
484 *
485 */
486static inline void seccomp_sync_threads(unsigned long flags)
487{
488 struct task_struct *thread, *caller;
489
490 BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
491 assert_spin_locked(&current->sighand->siglock);
492
493 /* Synchronize all threads. */
494 caller = current;
495 for_each_thread(caller, thread) {
496 /* Skip current, since it needs no changes. */
497 if (thread == caller)
498 continue;
499
500 /* Get a task reference for the new leaf node. */
501 get_seccomp_filter(caller);
502
503 /*
504 * Drop the task reference to the shared ancestor since
505 * current's path will hold a reference. (This also
506 * allows a put before the assignment.)
507 */
508 __seccomp_filter_release(thread->seccomp.filter);
509
510 /* Make our new filter tree visible. */
511 smp_store_release(&thread->seccomp.filter,
512 caller->seccomp.filter);
513 atomic_set(&thread->seccomp.filter_count,
514 atomic_read(&thread->seccomp.filter_count));
515
516 /*
517 * Don't let an unprivileged task work around
518 * the no_new_privs restriction by creating
519 * a thread that sets it up, enters seccomp,
520 * then dies.
521 */
522 if (task_no_new_privs(caller))
523 task_set_no_new_privs(thread);
524
525 /*
526 * Opt the other thread into seccomp if needed.
527 * As threads are considered to be trust-realm
528 * equivalent (see ptrace_may_access), it is safe to
529 * allow one thread to transition the other.
530 */
531 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
532 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
533 flags);
534 }
535}
536
537/**
538 * seccomp_prepare_filter: Prepares a seccomp filter for use.
539 * @fprog: BPF program to install
540 *
541 * Returns filter on success or an ERR_PTR on failure.
542 */
543static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
544{
545 struct seccomp_filter *sfilter;
546 int ret;
547 const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
548
549 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
550 return ERR_PTR(-EINVAL);
551
552 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
553
554 /*
555 * Installing a seccomp filter requires that the task has
556 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
557 * This avoids scenarios where unprivileged tasks can affect the
558 * behavior of privileged children.
559 */
560 if (!task_no_new_privs(current) &&
561 !ns_capable_noaudit(current_user_ns(), CAP_SYS_ADMIN))
562 return ERR_PTR(-EACCES);
563
564 /* Allocate a new seccomp_filter */
565 sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
566 if (!sfilter)
567 return ERR_PTR(-ENOMEM);
568
569 mutex_init(&sfilter->notify_lock);
570 ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
571 seccomp_check_filter, save_orig);
572 if (ret < 0) {
573 kfree(sfilter);
574 return ERR_PTR(ret);
575 }
576
577 refcount_set(&sfilter->refs, 1);
578 refcount_set(&sfilter->users, 1);
579 init_waitqueue_head(&sfilter->wqh);
580
581 return sfilter;
582}
583
584/**
585 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
586 * @user_filter: pointer to the user data containing a sock_fprog.
587 *
588 * Returns 0 on success and non-zero otherwise.
589 */
590static struct seccomp_filter *
591seccomp_prepare_user_filter(const char __user *user_filter)
592{
593 struct sock_fprog fprog;
594 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
595
596#ifdef CONFIG_COMPAT
597 if (in_compat_syscall()) {
598 struct compat_sock_fprog fprog32;
599 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
600 goto out;
601 fprog.len = fprog32.len;
602 fprog.filter = compat_ptr(fprog32.filter);
603 } else /* falls through to the if below. */
604#endif
605 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
606 goto out;
607 filter = seccomp_prepare_filter(&fprog);
608out:
609 return filter;
610}
611
612/**
613 * seccomp_attach_filter: validate and attach filter
614 * @flags: flags to change filter behavior
615 * @filter: seccomp filter to add to the current process
616 *
617 * Caller must be holding current->sighand->siglock lock.
618 *
619 * Returns 0 on success, -ve on error, or
620 * - in TSYNC mode: the pid of a thread which was either not in the correct
621 * seccomp mode or did not have an ancestral seccomp filter
622 * - in NEW_LISTENER mode: the fd of the new listener
623 */
624static long seccomp_attach_filter(unsigned int flags,
625 struct seccomp_filter *filter)
626{
627 unsigned long total_insns;
628 struct seccomp_filter *walker;
629
630 assert_spin_locked(&current->sighand->siglock);
631
632 /* Validate resulting filter length. */
633 total_insns = filter->prog->len;
634 for (walker = current->seccomp.filter; walker; walker = walker->prev)
635 total_insns += walker->prog->len + 4; /* 4 instr penalty */
636 if (total_insns > MAX_INSNS_PER_PATH)
637 return -ENOMEM;
638
639 /* If thread sync has been requested, check that it is possible. */
640 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
641 int ret;
642
643 ret = seccomp_can_sync_threads();
644 if (ret) {
645 if (flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH)
646 return -ESRCH;
647 else
648 return ret;
649 }
650 }
651
652 /* Set log flag, if present. */
653 if (flags & SECCOMP_FILTER_FLAG_LOG)
654 filter->log = true;
655
656 /*
657 * If there is an existing filter, make it the prev and don't drop its
658 * task reference.
659 */
660 filter->prev = current->seccomp.filter;
661 current->seccomp.filter = filter;
662 atomic_inc(&current->seccomp.filter_count);
663
664 /* Now that the new filter is in place, synchronize to all threads. */
665 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
666 seccomp_sync_threads(flags);
667
668 return 0;
669}
670
671static void __get_seccomp_filter(struct seccomp_filter *filter)
672{
673 refcount_inc(&filter->refs);
674}
675
676/* get_seccomp_filter - increments the reference count of the filter on @tsk */
677void get_seccomp_filter(struct task_struct *tsk)
678{
679 struct seccomp_filter *orig = tsk->seccomp.filter;
680 if (!orig)
681 return;
682 __get_seccomp_filter(orig);
683 refcount_inc(&orig->users);
684}
685
686static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason)
687{
688 clear_siginfo(info);
689 info->si_signo = SIGSYS;
690 info->si_code = SYS_SECCOMP;
691 info->si_call_addr = (void __user *)KSTK_EIP(current);
692 info->si_errno = reason;
693 info->si_arch = syscall_get_arch(current);
694 info->si_syscall = syscall;
695}
696
697/**
698 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
699 * @syscall: syscall number to send to userland
700 * @reason: filter-supplied reason code to send to userland (via si_errno)
701 *
702 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
703 */
704static void seccomp_send_sigsys(int syscall, int reason)
705{
706 struct kernel_siginfo info;
707 seccomp_init_siginfo(&info, syscall, reason);
708 force_sig_info(&info);
709}
710#endif /* CONFIG_SECCOMP_FILTER */
711
712/* For use with seccomp_actions_logged */
713#define SECCOMP_LOG_KILL_PROCESS (1 << 0)
714#define SECCOMP_LOG_KILL_THREAD (1 << 1)
715#define SECCOMP_LOG_TRAP (1 << 2)
716#define SECCOMP_LOG_ERRNO (1 << 3)
717#define SECCOMP_LOG_TRACE (1 << 4)
718#define SECCOMP_LOG_LOG (1 << 5)
719#define SECCOMP_LOG_ALLOW (1 << 6)
720#define SECCOMP_LOG_USER_NOTIF (1 << 7)
721
722static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
723 SECCOMP_LOG_KILL_THREAD |
724 SECCOMP_LOG_TRAP |
725 SECCOMP_LOG_ERRNO |
726 SECCOMP_LOG_USER_NOTIF |
727 SECCOMP_LOG_TRACE |
728 SECCOMP_LOG_LOG;
729
730static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
731 bool requested)
732{
733 bool log = false;
734
735 switch (action) {
736 case SECCOMP_RET_ALLOW:
737 break;
738 case SECCOMP_RET_TRAP:
739 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
740 break;
741 case SECCOMP_RET_ERRNO:
742 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
743 break;
744 case SECCOMP_RET_TRACE:
745 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
746 break;
747 case SECCOMP_RET_USER_NOTIF:
748 log = requested && seccomp_actions_logged & SECCOMP_LOG_USER_NOTIF;
749 break;
750 case SECCOMP_RET_LOG:
751 log = seccomp_actions_logged & SECCOMP_LOG_LOG;
752 break;
753 case SECCOMP_RET_KILL_THREAD:
754 log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
755 break;
756 case SECCOMP_RET_KILL_PROCESS:
757 default:
758 log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
759 }
760
761 /*
762 * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
763 * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
764 * any action from being logged by removing the action name from the
765 * seccomp_actions_logged sysctl.
766 */
767 if (!log)
768 return;
769
770 audit_seccomp(syscall, signr, action);
771}
772
773/*
774 * Secure computing mode 1 allows only read/write/exit/sigreturn.
775 * To be fully secure this must be combined with rlimit
776 * to limit the stack allocations too.
777 */
778static const int mode1_syscalls[] = {
779 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
780 -1, /* negative terminated */
781};
782
783static void __secure_computing_strict(int this_syscall)
784{
785 const int *allowed_syscalls = mode1_syscalls;
786#ifdef CONFIG_COMPAT
787 if (in_compat_syscall())
788 allowed_syscalls = get_compat_mode1_syscalls();
789#endif
790 do {
791 if (*allowed_syscalls == this_syscall)
792 return;
793 } while (*++allowed_syscalls != -1);
794
795#ifdef SECCOMP_DEBUG
796 dump_stack();
797#endif
798 seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
799 do_exit(SIGKILL);
800}
801
802#ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
803void secure_computing_strict(int this_syscall)
804{
805 int mode = current->seccomp.mode;
806
807 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
808 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
809 return;
810
811 if (mode == SECCOMP_MODE_DISABLED)
812 return;
813 else if (mode == SECCOMP_MODE_STRICT)
814 __secure_computing_strict(this_syscall);
815 else
816 BUG();
817}
818#else
819
820#ifdef CONFIG_SECCOMP_FILTER
821static u64 seccomp_next_notify_id(struct seccomp_filter *filter)
822{
823 /*
824 * Note: overflow is ok here, the id just needs to be unique per
825 * filter.
826 */
827 lockdep_assert_held(&filter->notify_lock);
828 return filter->notif->next_id++;
829}
830
831static void seccomp_handle_addfd(struct seccomp_kaddfd *addfd)
832{
833 /*
834 * Remove the notification, and reset the list pointers, indicating
835 * that it has been handled.
836 */
837 list_del_init(&addfd->list);
838 addfd->ret = receive_fd_replace(addfd->fd, addfd->file, addfd->flags);
839 complete(&addfd->completion);
840}
841
842static int seccomp_do_user_notification(int this_syscall,
843 struct seccomp_filter *match,
844 const struct seccomp_data *sd)
845{
846 int err;
847 u32 flags = 0;
848 long ret = 0;
849 struct seccomp_knotif n = {};
850 struct seccomp_kaddfd *addfd, *tmp;
851
852 mutex_lock(&match->notify_lock);
853 err = -ENOSYS;
854 if (!match->notif)
855 goto out;
856
857 n.task = current;
858 n.state = SECCOMP_NOTIFY_INIT;
859 n.data = sd;
860 n.id = seccomp_next_notify_id(match);
861 init_completion(&n.ready);
862 list_add(&n.list, &match->notif->notifications);
863 INIT_LIST_HEAD(&n.addfd);
864
865 up(&match->notif->request);
866 wake_up_poll(&match->wqh, EPOLLIN | EPOLLRDNORM);
867 mutex_unlock(&match->notify_lock);
868
869 /*
870 * This is where we wait for a reply from userspace.
871 */
872wait:
873 err = wait_for_completion_interruptible(&n.ready);
874 mutex_lock(&match->notify_lock);
875 if (err == 0) {
876 /* Check if we were woken up by a addfd message */
877 addfd = list_first_entry_or_null(&n.addfd,
878 struct seccomp_kaddfd, list);
879 if (addfd && n.state != SECCOMP_NOTIFY_REPLIED) {
880 seccomp_handle_addfd(addfd);
881 mutex_unlock(&match->notify_lock);
882 goto wait;
883 }
884 ret = n.val;
885 err = n.error;
886 flags = n.flags;
887 }
888
889 /* If there were any pending addfd calls, clear them out */
890 list_for_each_entry_safe(addfd, tmp, &n.addfd, list) {
891 /* The process went away before we got a chance to handle it */
892 addfd->ret = -ESRCH;
893 list_del_init(&addfd->list);
894 complete(&addfd->completion);
895 }
896
897 /*
898 * Note that it's possible the listener died in between the time when
899 * we were notified of a response (or a signal) and when we were able to
900 * re-acquire the lock, so only delete from the list if the
901 * notification actually exists.
902 *
903 * Also note that this test is only valid because there's no way to
904 * *reattach* to a notifier right now. If one is added, we'll need to
905 * keep track of the notif itself and make sure they match here.
906 */
907 if (match->notif)
908 list_del(&n.list);
909out:
910 mutex_unlock(&match->notify_lock);
911
912 /* Userspace requests to continue the syscall. */
913 if (flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE)
914 return 0;
915
916 syscall_set_return_value(current, current_pt_regs(),
917 err, ret);
918 return -1;
919}
920
921static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
922 const bool recheck_after_trace)
923{
924 u32 filter_ret, action;
925 struct seccomp_filter *match = NULL;
926 int data;
927 struct seccomp_data sd_local;
928
929 /*
930 * Make sure that any changes to mode from another thread have
931 * been seen after TIF_SECCOMP was seen.
932 */
933 rmb();
934
935 if (!sd) {
936 populate_seccomp_data(&sd_local);
937 sd = &sd_local;
938 }
939
940 filter_ret = seccomp_run_filters(sd, &match);
941 data = filter_ret & SECCOMP_RET_DATA;
942 action = filter_ret & SECCOMP_RET_ACTION_FULL;
943
944 switch (action) {
945 case SECCOMP_RET_ERRNO:
946 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
947 if (data > MAX_ERRNO)
948 data = MAX_ERRNO;
949 syscall_set_return_value(current, current_pt_regs(),
950 -data, 0);
951 goto skip;
952
953 case SECCOMP_RET_TRAP:
954 /* Show the handler the original registers. */
955 syscall_rollback(current, current_pt_regs());
956 /* Let the filter pass back 16 bits of data. */
957 seccomp_send_sigsys(this_syscall, data);
958 goto skip;
959
960 case SECCOMP_RET_TRACE:
961 /* We've been put in this state by the ptracer already. */
962 if (recheck_after_trace)
963 return 0;
964
965 /* ENOSYS these calls if there is no tracer attached. */
966 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
967 syscall_set_return_value(current,
968 current_pt_regs(),
969 -ENOSYS, 0);
970 goto skip;
971 }
972
973 /* Allow the BPF to provide the event message */
974 ptrace_event(PTRACE_EVENT_SECCOMP, data);
975 /*
976 * The delivery of a fatal signal during event
977 * notification may silently skip tracer notification,
978 * which could leave us with a potentially unmodified
979 * syscall that the tracer would have liked to have
980 * changed. Since the process is about to die, we just
981 * force the syscall to be skipped and let the signal
982 * kill the process and correctly handle any tracer exit
983 * notifications.
984 */
985 if (fatal_signal_pending(current))
986 goto skip;
987 /* Check if the tracer forced the syscall to be skipped. */
988 this_syscall = syscall_get_nr(current, current_pt_regs());
989 if (this_syscall < 0)
990 goto skip;
991
992 /*
993 * Recheck the syscall, since it may have changed. This
994 * intentionally uses a NULL struct seccomp_data to force
995 * a reload of all registers. This does not goto skip since
996 * a skip would have already been reported.
997 */
998 if (__seccomp_filter(this_syscall, NULL, true))
999 return -1;
1000
1001 return 0;
1002
1003 case SECCOMP_RET_USER_NOTIF:
1004 if (seccomp_do_user_notification(this_syscall, match, sd))
1005 goto skip;
1006
1007 return 0;
1008
1009 case SECCOMP_RET_LOG:
1010 seccomp_log(this_syscall, 0, action, true);
1011 return 0;
1012
1013 case SECCOMP_RET_ALLOW:
1014 /*
1015 * Note that the "match" filter will always be NULL for
1016 * this action since SECCOMP_RET_ALLOW is the starting
1017 * state in seccomp_run_filters().
1018 */
1019 return 0;
1020
1021 case SECCOMP_RET_KILL_THREAD:
1022 case SECCOMP_RET_KILL_PROCESS:
1023 default:
1024 seccomp_log(this_syscall, SIGSYS, action, true);
1025 /* Dump core only if this is the last remaining thread. */
1026 if (action != SECCOMP_RET_KILL_THREAD ||
1027 get_nr_threads(current) == 1) {
1028 kernel_siginfo_t info;
1029
1030 /* Show the original registers in the dump. */
1031 syscall_rollback(current, current_pt_regs());
1032 /* Trigger a manual coredump since do_exit skips it. */
1033 seccomp_init_siginfo(&info, this_syscall, data);
1034 do_coredump(&info);
1035 }
1036 if (action == SECCOMP_RET_KILL_THREAD)
1037 do_exit(SIGSYS);
1038 else
1039 do_group_exit(SIGSYS);
1040 }
1041
1042 unreachable();
1043
1044skip:
1045 seccomp_log(this_syscall, 0, action, match ? match->log : false);
1046 return -1;
1047}
1048#else
1049static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
1050 const bool recheck_after_trace)
1051{
1052 BUG();
1053}
1054#endif
1055
1056int __secure_computing(const struct seccomp_data *sd)
1057{
1058 int mode = current->seccomp.mode;
1059 int this_syscall;
1060
1061 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
1062 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
1063 return 0;
1064
1065 this_syscall = sd ? sd->nr :
1066 syscall_get_nr(current, current_pt_regs());
1067
1068 switch (mode) {
1069 case SECCOMP_MODE_STRICT:
1070 __secure_computing_strict(this_syscall); /* may call do_exit */
1071 return 0;
1072 case SECCOMP_MODE_FILTER:
1073 return __seccomp_filter(this_syscall, sd, false);
1074 default:
1075 BUG();
1076 }
1077}
1078#endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
1079
1080long prctl_get_seccomp(void)
1081{
1082 return current->seccomp.mode;
1083}
1084
1085/**
1086 * seccomp_set_mode_strict: internal function for setting strict seccomp
1087 *
1088 * Once current->seccomp.mode is non-zero, it may not be changed.
1089 *
1090 * Returns 0 on success or -EINVAL on failure.
1091 */
1092static long seccomp_set_mode_strict(void)
1093{
1094 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
1095 long ret = -EINVAL;
1096
1097 spin_lock_irq(&current->sighand->siglock);
1098
1099 if (!seccomp_may_assign_mode(seccomp_mode))
1100 goto out;
1101
1102#ifdef TIF_NOTSC
1103 disable_TSC();
1104#endif
1105 seccomp_assign_mode(current, seccomp_mode, 0);
1106 ret = 0;
1107
1108out:
1109 spin_unlock_irq(&current->sighand->siglock);
1110
1111 return ret;
1112}
1113
1114#ifdef CONFIG_SECCOMP_FILTER
1115static void seccomp_notify_free(struct seccomp_filter *filter)
1116{
1117 kfree(filter->notif);
1118 filter->notif = NULL;
1119}
1120
1121static void seccomp_notify_detach(struct seccomp_filter *filter)
1122{
1123 struct seccomp_knotif *knotif;
1124
1125 if (!filter)
1126 return;
1127
1128 mutex_lock(&filter->notify_lock);
1129
1130 /*
1131 * If this file is being closed because e.g. the task who owned it
1132 * died, let's wake everyone up who was waiting on us.
1133 */
1134 list_for_each_entry(knotif, &filter->notif->notifications, list) {
1135 if (knotif->state == SECCOMP_NOTIFY_REPLIED)
1136 continue;
1137
1138 knotif->state = SECCOMP_NOTIFY_REPLIED;
1139 knotif->error = -ENOSYS;
1140 knotif->val = 0;
1141
1142 /*
1143 * We do not need to wake up any pending addfd messages, as
1144 * the notifier will do that for us, as this just looks
1145 * like a standard reply.
1146 */
1147 complete(&knotif->ready);
1148 }
1149
1150 seccomp_notify_free(filter);
1151 mutex_unlock(&filter->notify_lock);
1152}
1153
1154static int seccomp_notify_release(struct inode *inode, struct file *file)
1155{
1156 struct seccomp_filter *filter = file->private_data;
1157
1158 seccomp_notify_detach(filter);
1159 __put_seccomp_filter(filter);
1160 return 0;
1161}
1162
1163/* must be called with notif_lock held */
1164static inline struct seccomp_knotif *
1165find_notification(struct seccomp_filter *filter, u64 id)
1166{
1167 struct seccomp_knotif *cur;
1168
1169 lockdep_assert_held(&filter->notify_lock);
1170
1171 list_for_each_entry(cur, &filter->notif->notifications, list) {
1172 if (cur->id == id)
1173 return cur;
1174 }
1175
1176 return NULL;
1177}
1178
1179
1180static long seccomp_notify_recv(struct seccomp_filter *filter,
1181 void __user *buf)
1182{
1183 struct seccomp_knotif *knotif = NULL, *cur;
1184 struct seccomp_notif unotif;
1185 ssize_t ret;
1186
1187 /* Verify that we're not given garbage to keep struct extensible. */
1188 ret = check_zeroed_user(buf, sizeof(unotif));
1189 if (ret < 0)
1190 return ret;
1191 if (!ret)
1192 return -EINVAL;
1193
1194 memset(&unotif, 0, sizeof(unotif));
1195
1196 ret = down_interruptible(&filter->notif->request);
1197 if (ret < 0)
1198 return ret;
1199
1200 mutex_lock(&filter->notify_lock);
1201 list_for_each_entry(cur, &filter->notif->notifications, list) {
1202 if (cur->state == SECCOMP_NOTIFY_INIT) {
1203 knotif = cur;
1204 break;
1205 }
1206 }
1207
1208 /*
1209 * If we didn't find a notification, it could be that the task was
1210 * interrupted by a fatal signal between the time we were woken and
1211 * when we were able to acquire the rw lock.
1212 */
1213 if (!knotif) {
1214 ret = -ENOENT;
1215 goto out;
1216 }
1217
1218 unotif.id = knotif->id;
1219 unotif.pid = task_pid_vnr(knotif->task);
1220 unotif.data = *(knotif->data);
1221
1222 knotif->state = SECCOMP_NOTIFY_SENT;
1223 wake_up_poll(&filter->wqh, EPOLLOUT | EPOLLWRNORM);
1224 ret = 0;
1225out:
1226 mutex_unlock(&filter->notify_lock);
1227
1228 if (ret == 0 && copy_to_user(buf, &unotif, sizeof(unotif))) {
1229 ret = -EFAULT;
1230
1231 /*
1232 * Userspace screwed up. To make sure that we keep this
1233 * notification alive, let's reset it back to INIT. It
1234 * may have died when we released the lock, so we need to make
1235 * sure it's still around.
1236 */
1237 mutex_lock(&filter->notify_lock);
1238 knotif = find_notification(filter, unotif.id);
1239 if (knotif) {
1240 knotif->state = SECCOMP_NOTIFY_INIT;
1241 up(&filter->notif->request);
1242 }
1243 mutex_unlock(&filter->notify_lock);
1244 }
1245
1246 return ret;
1247}
1248
1249static long seccomp_notify_send(struct seccomp_filter *filter,
1250 void __user *buf)
1251{
1252 struct seccomp_notif_resp resp = {};
1253 struct seccomp_knotif *knotif;
1254 long ret;
1255
1256 if (copy_from_user(&resp, buf, sizeof(resp)))
1257 return -EFAULT;
1258
1259 if (resp.flags & ~SECCOMP_USER_NOTIF_FLAG_CONTINUE)
1260 return -EINVAL;
1261
1262 if ((resp.flags & SECCOMP_USER_NOTIF_FLAG_CONTINUE) &&
1263 (resp.error || resp.val))
1264 return -EINVAL;
1265
1266 ret = mutex_lock_interruptible(&filter->notify_lock);
1267 if (ret < 0)
1268 return ret;
1269
1270 knotif = find_notification(filter, resp.id);
1271 if (!knotif) {
1272 ret = -ENOENT;
1273 goto out;
1274 }
1275
1276 /* Allow exactly one reply. */
1277 if (knotif->state != SECCOMP_NOTIFY_SENT) {
1278 ret = -EINPROGRESS;
1279 goto out;
1280 }
1281
1282 ret = 0;
1283 knotif->state = SECCOMP_NOTIFY_REPLIED;
1284 knotif->error = resp.error;
1285 knotif->val = resp.val;
1286 knotif->flags = resp.flags;
1287 complete(&knotif->ready);
1288out:
1289 mutex_unlock(&filter->notify_lock);
1290 return ret;
1291}
1292
1293static long seccomp_notify_id_valid(struct seccomp_filter *filter,
1294 void __user *buf)
1295{
1296 struct seccomp_knotif *knotif;
1297 u64 id;
1298 long ret;
1299
1300 if (copy_from_user(&id, buf, sizeof(id)))
1301 return -EFAULT;
1302
1303 ret = mutex_lock_interruptible(&filter->notify_lock);
1304 if (ret < 0)
1305 return ret;
1306
1307 knotif = find_notification(filter, id);
1308 if (knotif && knotif->state == SECCOMP_NOTIFY_SENT)
1309 ret = 0;
1310 else
1311 ret = -ENOENT;
1312
1313 mutex_unlock(&filter->notify_lock);
1314 return ret;
1315}
1316
1317static long seccomp_notify_addfd(struct seccomp_filter *filter,
1318 struct seccomp_notif_addfd __user *uaddfd,
1319 unsigned int size)
1320{
1321 struct seccomp_notif_addfd addfd;
1322 struct seccomp_knotif *knotif;
1323 struct seccomp_kaddfd kaddfd;
1324 int ret;
1325
1326 BUILD_BUG_ON(sizeof(addfd) < SECCOMP_NOTIFY_ADDFD_SIZE_VER0);
1327 BUILD_BUG_ON(sizeof(addfd) != SECCOMP_NOTIFY_ADDFD_SIZE_LATEST);
1328
1329 if (size < SECCOMP_NOTIFY_ADDFD_SIZE_VER0 || size >= PAGE_SIZE)
1330 return -EINVAL;
1331
1332 ret = copy_struct_from_user(&addfd, sizeof(addfd), uaddfd, size);
1333 if (ret)
1334 return ret;
1335
1336 if (addfd.newfd_flags & ~O_CLOEXEC)
1337 return -EINVAL;
1338
1339 if (addfd.flags & ~SECCOMP_ADDFD_FLAG_SETFD)
1340 return -EINVAL;
1341
1342 if (addfd.newfd && !(addfd.flags & SECCOMP_ADDFD_FLAG_SETFD))
1343 return -EINVAL;
1344
1345 kaddfd.file = fget(addfd.srcfd);
1346 if (!kaddfd.file)
1347 return -EBADF;
1348
1349 kaddfd.flags = addfd.newfd_flags;
1350 kaddfd.fd = (addfd.flags & SECCOMP_ADDFD_FLAG_SETFD) ?
1351 addfd.newfd : -1;
1352 init_completion(&kaddfd.completion);
1353
1354 ret = mutex_lock_interruptible(&filter->notify_lock);
1355 if (ret < 0)
1356 goto out;
1357
1358 knotif = find_notification(filter, addfd.id);
1359 if (!knotif) {
1360 ret = -ENOENT;
1361 goto out_unlock;
1362 }
1363
1364 /*
1365 * We do not want to allow for FD injection to occur before the
1366 * notification has been picked up by a userspace handler, or after
1367 * the notification has been replied to.
1368 */
1369 if (knotif->state != SECCOMP_NOTIFY_SENT) {
1370 ret = -EINPROGRESS;
1371 goto out_unlock;
1372 }
1373
1374 list_add(&kaddfd.list, &knotif->addfd);
1375 complete(&knotif->ready);
1376 mutex_unlock(&filter->notify_lock);
1377
1378 /* Now we wait for it to be processed or be interrupted */
1379 ret = wait_for_completion_interruptible(&kaddfd.completion);
1380 if (ret == 0) {
1381 /*
1382 * We had a successful completion. The other side has already
1383 * removed us from the addfd queue, and
1384 * wait_for_completion_interruptible has a memory barrier upon
1385 * success that lets us read this value directly without
1386 * locking.
1387 */
1388 ret = kaddfd.ret;
1389 goto out;
1390 }
1391
1392 mutex_lock(&filter->notify_lock);
1393 /*
1394 * Even though we were woken up by a signal and not a successful
1395 * completion, a completion may have happened in the mean time.
1396 *
1397 * We need to check again if the addfd request has been handled,
1398 * and if not, we will remove it from the queue.
1399 */
1400 if (list_empty(&kaddfd.list))
1401 ret = kaddfd.ret;
1402 else
1403 list_del(&kaddfd.list);
1404
1405out_unlock:
1406 mutex_unlock(&filter->notify_lock);
1407out:
1408 fput(kaddfd.file);
1409
1410 return ret;
1411}
1412
1413static long seccomp_notify_ioctl(struct file *file, unsigned int cmd,
1414 unsigned long arg)
1415{
1416 struct seccomp_filter *filter = file->private_data;
1417 void __user *buf = (void __user *)arg;
1418
1419 /* Fixed-size ioctls */
1420 switch (cmd) {
1421 case SECCOMP_IOCTL_NOTIF_RECV:
1422 return seccomp_notify_recv(filter, buf);
1423 case SECCOMP_IOCTL_NOTIF_SEND:
1424 return seccomp_notify_send(filter, buf);
1425 case SECCOMP_IOCTL_NOTIF_ID_VALID_WRONG_DIR:
1426 case SECCOMP_IOCTL_NOTIF_ID_VALID:
1427 return seccomp_notify_id_valid(filter, buf);
1428 }
1429
1430 /* Extensible Argument ioctls */
1431#define EA_IOCTL(cmd) ((cmd) & ~(IOC_INOUT | IOCSIZE_MASK))
1432 switch (EA_IOCTL(cmd)) {
1433 case EA_IOCTL(SECCOMP_IOCTL_NOTIF_ADDFD):
1434 return seccomp_notify_addfd(filter, buf, _IOC_SIZE(cmd));
1435 default:
1436 return -EINVAL;
1437 }
1438}
1439
1440static __poll_t seccomp_notify_poll(struct file *file,
1441 struct poll_table_struct *poll_tab)
1442{
1443 struct seccomp_filter *filter = file->private_data;
1444 __poll_t ret = 0;
1445 struct seccomp_knotif *cur;
1446
1447 poll_wait(file, &filter->wqh, poll_tab);
1448
1449 if (mutex_lock_interruptible(&filter->notify_lock) < 0)
1450 return EPOLLERR;
1451
1452 list_for_each_entry(cur, &filter->notif->notifications, list) {
1453 if (cur->state == SECCOMP_NOTIFY_INIT)
1454 ret |= EPOLLIN | EPOLLRDNORM;
1455 if (cur->state == SECCOMP_NOTIFY_SENT)
1456 ret |= EPOLLOUT | EPOLLWRNORM;
1457 if ((ret & EPOLLIN) && (ret & EPOLLOUT))
1458 break;
1459 }
1460
1461 mutex_unlock(&filter->notify_lock);
1462
1463 if (refcount_read(&filter->users) == 0)
1464 ret |= EPOLLHUP;
1465
1466 return ret;
1467}
1468
1469static const struct file_operations seccomp_notify_ops = {
1470 .poll = seccomp_notify_poll,
1471 .release = seccomp_notify_release,
1472 .unlocked_ioctl = seccomp_notify_ioctl,
1473 .compat_ioctl = seccomp_notify_ioctl,
1474};
1475
1476static struct file *init_listener(struct seccomp_filter *filter)
1477{
1478 struct file *ret;
1479
1480 ret = ERR_PTR(-ENOMEM);
1481 filter->notif = kzalloc(sizeof(*(filter->notif)), GFP_KERNEL);
1482 if (!filter->notif)
1483 goto out;
1484
1485 sema_init(&filter->notif->request, 0);
1486 filter->notif->next_id = get_random_u64();
1487 INIT_LIST_HEAD(&filter->notif->notifications);
1488
1489 ret = anon_inode_getfile("seccomp notify", &seccomp_notify_ops,
1490 filter, O_RDWR);
1491 if (IS_ERR(ret))
1492 goto out_notif;
1493
1494 /* The file has a reference to it now */
1495 __get_seccomp_filter(filter);
1496
1497out_notif:
1498 if (IS_ERR(ret))
1499 seccomp_notify_free(filter);
1500out:
1501 return ret;
1502}
1503
1504/*
1505 * Does @new_child have a listener while an ancestor also has a listener?
1506 * If so, we'll want to reject this filter.
1507 * This only has to be tested for the current process, even in the TSYNC case,
1508 * because TSYNC installs @child with the same parent on all threads.
1509 * Note that @new_child is not hooked up to its parent at this point yet, so
1510 * we use current->seccomp.filter.
1511 */
1512static bool has_duplicate_listener(struct seccomp_filter *new_child)
1513{
1514 struct seccomp_filter *cur;
1515
1516 /* must be protected against concurrent TSYNC */
1517 lockdep_assert_held(&current->sighand->siglock);
1518
1519 if (!new_child->notif)
1520 return false;
1521 for (cur = current->seccomp.filter; cur; cur = cur->prev) {
1522 if (cur->notif)
1523 return true;
1524 }
1525
1526 return false;
1527}
1528
1529/**
1530 * seccomp_set_mode_filter: internal function for setting seccomp filter
1531 * @flags: flags to change filter behavior
1532 * @filter: struct sock_fprog containing filter
1533 *
1534 * This function may be called repeatedly to install additional filters.
1535 * Every filter successfully installed will be evaluated (in reverse order)
1536 * for each system call the task makes.
1537 *
1538 * Once current->seccomp.mode is non-zero, it may not be changed.
1539 *
1540 * Returns 0 on success or -EINVAL on failure.
1541 */
1542static long seccomp_set_mode_filter(unsigned int flags,
1543 const char __user *filter)
1544{
1545 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
1546 struct seccomp_filter *prepared = NULL;
1547 long ret = -EINVAL;
1548 int listener = -1;
1549 struct file *listener_f = NULL;
1550
1551 /* Validate flags. */
1552 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
1553 return -EINVAL;
1554
1555 /*
1556 * In the successful case, NEW_LISTENER returns the new listener fd.
1557 * But in the failure case, TSYNC returns the thread that died. If you
1558 * combine these two flags, there's no way to tell whether something
1559 * succeeded or failed. So, let's disallow this combination if the user
1560 * has not explicitly requested no errors from TSYNC.
1561 */
1562 if ((flags & SECCOMP_FILTER_FLAG_TSYNC) &&
1563 (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) &&
1564 ((flags & SECCOMP_FILTER_FLAG_TSYNC_ESRCH) == 0))
1565 return -EINVAL;
1566
1567 /* Prepare the new filter before holding any locks. */
1568 prepared = seccomp_prepare_user_filter(filter);
1569 if (IS_ERR(prepared))
1570 return PTR_ERR(prepared);
1571
1572 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1573 listener = get_unused_fd_flags(O_CLOEXEC);
1574 if (listener < 0) {
1575 ret = listener;
1576 goto out_free;
1577 }
1578
1579 listener_f = init_listener(prepared);
1580 if (IS_ERR(listener_f)) {
1581 put_unused_fd(listener);
1582 ret = PTR_ERR(listener_f);
1583 goto out_free;
1584 }
1585 }
1586
1587 /*
1588 * Make sure we cannot change seccomp or nnp state via TSYNC
1589 * while another thread is in the middle of calling exec.
1590 */
1591 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
1592 mutex_lock_killable(&current->signal->cred_guard_mutex))
1593 goto out_put_fd;
1594
1595 spin_lock_irq(&current->sighand->siglock);
1596
1597 if (!seccomp_may_assign_mode(seccomp_mode))
1598 goto out;
1599
1600 if (has_duplicate_listener(prepared)) {
1601 ret = -EBUSY;
1602 goto out;
1603 }
1604
1605 ret = seccomp_attach_filter(flags, prepared);
1606 if (ret)
1607 goto out;
1608 /* Do not free the successfully attached filter. */
1609 prepared = NULL;
1610
1611 seccomp_assign_mode(current, seccomp_mode, flags);
1612out:
1613 spin_unlock_irq(&current->sighand->siglock);
1614 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
1615 mutex_unlock(&current->signal->cred_guard_mutex);
1616out_put_fd:
1617 if (flags & SECCOMP_FILTER_FLAG_NEW_LISTENER) {
1618 if (ret) {
1619 listener_f->private_data = NULL;
1620 fput(listener_f);
1621 put_unused_fd(listener);
1622 seccomp_notify_detach(prepared);
1623 } else {
1624 fd_install(listener, listener_f);
1625 ret = listener;
1626 }
1627 }
1628out_free:
1629 seccomp_filter_free(prepared);
1630 return ret;
1631}
1632#else
1633static inline long seccomp_set_mode_filter(unsigned int flags,
1634 const char __user *filter)
1635{
1636 return -EINVAL;
1637}
1638#endif
1639
1640static long seccomp_get_action_avail(const char __user *uaction)
1641{
1642 u32 action;
1643
1644 if (copy_from_user(&action, uaction, sizeof(action)))
1645 return -EFAULT;
1646
1647 switch (action) {
1648 case SECCOMP_RET_KILL_PROCESS:
1649 case SECCOMP_RET_KILL_THREAD:
1650 case SECCOMP_RET_TRAP:
1651 case SECCOMP_RET_ERRNO:
1652 case SECCOMP_RET_USER_NOTIF:
1653 case SECCOMP_RET_TRACE:
1654 case SECCOMP_RET_LOG:
1655 case SECCOMP_RET_ALLOW:
1656 break;
1657 default:
1658 return -EOPNOTSUPP;
1659 }
1660
1661 return 0;
1662}
1663
1664static long seccomp_get_notif_sizes(void __user *usizes)
1665{
1666 struct seccomp_notif_sizes sizes = {
1667 .seccomp_notif = sizeof(struct seccomp_notif),
1668 .seccomp_notif_resp = sizeof(struct seccomp_notif_resp),
1669 .seccomp_data = sizeof(struct seccomp_data),
1670 };
1671
1672 if (copy_to_user(usizes, &sizes, sizeof(sizes)))
1673 return -EFAULT;
1674
1675 return 0;
1676}
1677
1678/* Common entry point for both prctl and syscall. */
1679static long do_seccomp(unsigned int op, unsigned int flags,
1680 void __user *uargs)
1681{
1682 switch (op) {
1683 case SECCOMP_SET_MODE_STRICT:
1684 if (flags != 0 || uargs != NULL)
1685 return -EINVAL;
1686 return seccomp_set_mode_strict();
1687 case SECCOMP_SET_MODE_FILTER:
1688 return seccomp_set_mode_filter(flags, uargs);
1689 case SECCOMP_GET_ACTION_AVAIL:
1690 if (flags != 0)
1691 return -EINVAL;
1692
1693 return seccomp_get_action_avail(uargs);
1694 case SECCOMP_GET_NOTIF_SIZES:
1695 if (flags != 0)
1696 return -EINVAL;
1697
1698 return seccomp_get_notif_sizes(uargs);
1699 default:
1700 return -EINVAL;
1701 }
1702}
1703
1704SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
1705 void __user *, uargs)
1706{
1707 return do_seccomp(op, flags, uargs);
1708}
1709
1710/**
1711 * prctl_set_seccomp: configures current->seccomp.mode
1712 * @seccomp_mode: requested mode to use
1713 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
1714 *
1715 * Returns 0 on success or -EINVAL on failure.
1716 */
1717long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
1718{
1719 unsigned int op;
1720 void __user *uargs;
1721
1722 switch (seccomp_mode) {
1723 case SECCOMP_MODE_STRICT:
1724 op = SECCOMP_SET_MODE_STRICT;
1725 /*
1726 * Setting strict mode through prctl always ignored filter,
1727 * so make sure it is always NULL here to pass the internal
1728 * check in do_seccomp().
1729 */
1730 uargs = NULL;
1731 break;
1732 case SECCOMP_MODE_FILTER:
1733 op = SECCOMP_SET_MODE_FILTER;
1734 uargs = filter;
1735 break;
1736 default:
1737 return -EINVAL;
1738 }
1739
1740 /* prctl interface doesn't have flags, so they are always zero. */
1741 return do_seccomp(op, 0, uargs);
1742}
1743
1744#if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
1745static struct seccomp_filter *get_nth_filter(struct task_struct *task,
1746 unsigned long filter_off)
1747{
1748 struct seccomp_filter *orig, *filter;
1749 unsigned long count;
1750
1751 /*
1752 * Note: this is only correct because the caller should be the (ptrace)
1753 * tracer of the task, otherwise lock_task_sighand is needed.
1754 */
1755 spin_lock_irq(&task->sighand->siglock);
1756
1757 if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
1758 spin_unlock_irq(&task->sighand->siglock);
1759 return ERR_PTR(-EINVAL);
1760 }
1761
1762 orig = task->seccomp.filter;
1763 __get_seccomp_filter(orig);
1764 spin_unlock_irq(&task->sighand->siglock);
1765
1766 count = 0;
1767 for (filter = orig; filter; filter = filter->prev)
1768 count++;
1769
1770 if (filter_off >= count) {
1771 filter = ERR_PTR(-ENOENT);
1772 goto out;
1773 }
1774
1775 count -= filter_off;
1776 for (filter = orig; filter && count > 1; filter = filter->prev)
1777 count--;
1778
1779 if (WARN_ON(count != 1 || !filter)) {
1780 filter = ERR_PTR(-ENOENT);
1781 goto out;
1782 }
1783
1784 __get_seccomp_filter(filter);
1785
1786out:
1787 __put_seccomp_filter(orig);
1788 return filter;
1789}
1790
1791long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
1792 void __user *data)
1793{
1794 struct seccomp_filter *filter;
1795 struct sock_fprog_kern *fprog;
1796 long ret;
1797
1798 if (!capable(CAP_SYS_ADMIN) ||
1799 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1800 return -EACCES;
1801 }
1802
1803 filter = get_nth_filter(task, filter_off);
1804 if (IS_ERR(filter))
1805 return PTR_ERR(filter);
1806
1807 fprog = filter->prog->orig_prog;
1808 if (!fprog) {
1809 /* This must be a new non-cBPF filter, since we save
1810 * every cBPF filter's orig_prog above when
1811 * CONFIG_CHECKPOINT_RESTORE is enabled.
1812 */
1813 ret = -EMEDIUMTYPE;
1814 goto out;
1815 }
1816
1817 ret = fprog->len;
1818 if (!data)
1819 goto out;
1820
1821 if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1822 ret = -EFAULT;
1823
1824out:
1825 __put_seccomp_filter(filter);
1826 return ret;
1827}
1828
1829long seccomp_get_metadata(struct task_struct *task,
1830 unsigned long size, void __user *data)
1831{
1832 long ret;
1833 struct seccomp_filter *filter;
1834 struct seccomp_metadata kmd = {};
1835
1836 if (!capable(CAP_SYS_ADMIN) ||
1837 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1838 return -EACCES;
1839 }
1840
1841 size = min_t(unsigned long, size, sizeof(kmd));
1842
1843 if (size < sizeof(kmd.filter_off))
1844 return -EINVAL;
1845
1846 if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
1847 return -EFAULT;
1848
1849 filter = get_nth_filter(task, kmd.filter_off);
1850 if (IS_ERR(filter))
1851 return PTR_ERR(filter);
1852
1853 if (filter->log)
1854 kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
1855
1856 ret = size;
1857 if (copy_to_user(data, &kmd, size))
1858 ret = -EFAULT;
1859
1860 __put_seccomp_filter(filter);
1861 return ret;
1862}
1863#endif
1864
1865#ifdef CONFIG_SYSCTL
1866
1867/* Human readable action names for friendly sysctl interaction */
1868#define SECCOMP_RET_KILL_PROCESS_NAME "kill_process"
1869#define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
1870#define SECCOMP_RET_TRAP_NAME "trap"
1871#define SECCOMP_RET_ERRNO_NAME "errno"
1872#define SECCOMP_RET_USER_NOTIF_NAME "user_notif"
1873#define SECCOMP_RET_TRACE_NAME "trace"
1874#define SECCOMP_RET_LOG_NAME "log"
1875#define SECCOMP_RET_ALLOW_NAME "allow"
1876
1877static const char seccomp_actions_avail[] =
1878 SECCOMP_RET_KILL_PROCESS_NAME " "
1879 SECCOMP_RET_KILL_THREAD_NAME " "
1880 SECCOMP_RET_TRAP_NAME " "
1881 SECCOMP_RET_ERRNO_NAME " "
1882 SECCOMP_RET_USER_NOTIF_NAME " "
1883 SECCOMP_RET_TRACE_NAME " "
1884 SECCOMP_RET_LOG_NAME " "
1885 SECCOMP_RET_ALLOW_NAME;
1886
1887struct seccomp_log_name {
1888 u32 log;
1889 const char *name;
1890};
1891
1892static const struct seccomp_log_name seccomp_log_names[] = {
1893 { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
1894 { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1895 { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1896 { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1897 { SECCOMP_LOG_USER_NOTIF, SECCOMP_RET_USER_NOTIF_NAME },
1898 { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1899 { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1900 { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1901 { }
1902};
1903
1904static bool seccomp_names_from_actions_logged(char *names, size_t size,
1905 u32 actions_logged,
1906 const char *sep)
1907{
1908 const struct seccomp_log_name *cur;
1909 bool append_sep = false;
1910
1911 for (cur = seccomp_log_names; cur->name && size; cur++) {
1912 ssize_t ret;
1913
1914 if (!(actions_logged & cur->log))
1915 continue;
1916
1917 if (append_sep) {
1918 ret = strscpy(names, sep, size);
1919 if (ret < 0)
1920 return false;
1921
1922 names += ret;
1923 size -= ret;
1924 } else
1925 append_sep = true;
1926
1927 ret = strscpy(names, cur->name, size);
1928 if (ret < 0)
1929 return false;
1930
1931 names += ret;
1932 size -= ret;
1933 }
1934
1935 return true;
1936}
1937
1938static bool seccomp_action_logged_from_name(u32 *action_logged,
1939 const char *name)
1940{
1941 const struct seccomp_log_name *cur;
1942
1943 for (cur = seccomp_log_names; cur->name; cur++) {
1944 if (!strcmp(cur->name, name)) {
1945 *action_logged = cur->log;
1946 return true;
1947 }
1948 }
1949
1950 return false;
1951}
1952
1953static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1954{
1955 char *name;
1956
1957 *actions_logged = 0;
1958 while ((name = strsep(&names, " ")) && *name) {
1959 u32 action_logged = 0;
1960
1961 if (!seccomp_action_logged_from_name(&action_logged, name))
1962 return false;
1963
1964 *actions_logged |= action_logged;
1965 }
1966
1967 return true;
1968}
1969
1970static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1971 size_t *lenp, loff_t *ppos)
1972{
1973 char names[sizeof(seccomp_actions_avail)];
1974 struct ctl_table table;
1975
1976 memset(names, 0, sizeof(names));
1977
1978 if (!seccomp_names_from_actions_logged(names, sizeof(names),
1979 seccomp_actions_logged, " "))
1980 return -EINVAL;
1981
1982 table = *ro_table;
1983 table.data = names;
1984 table.maxlen = sizeof(names);
1985 return proc_dostring(&table, 0, buffer, lenp, ppos);
1986}
1987
1988static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1989 size_t *lenp, loff_t *ppos, u32 *actions_logged)
1990{
1991 char names[sizeof(seccomp_actions_avail)];
1992 struct ctl_table table;
1993 int ret;
1994
1995 if (!capable(CAP_SYS_ADMIN))
1996 return -EPERM;
1997
1998 memset(names, 0, sizeof(names));
1999
2000 table = *ro_table;
2001 table.data = names;
2002 table.maxlen = sizeof(names);
2003 ret = proc_dostring(&table, 1, buffer, lenp, ppos);
2004 if (ret)
2005 return ret;
2006
2007 if (!seccomp_actions_logged_from_names(actions_logged, table.data))
2008 return -EINVAL;
2009
2010 if (*actions_logged & SECCOMP_LOG_ALLOW)
2011 return -EINVAL;
2012
2013 seccomp_actions_logged = *actions_logged;
2014 return 0;
2015}
2016
2017static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
2018 int ret)
2019{
2020 char names[sizeof(seccomp_actions_avail)];
2021 char old_names[sizeof(seccomp_actions_avail)];
2022 const char *new = names;
2023 const char *old = old_names;
2024
2025 if (!audit_enabled)
2026 return;
2027
2028 memset(names, 0, sizeof(names));
2029 memset(old_names, 0, sizeof(old_names));
2030
2031 if (ret)
2032 new = "?";
2033 else if (!actions_logged)
2034 new = "(none)";
2035 else if (!seccomp_names_from_actions_logged(names, sizeof(names),
2036 actions_logged, ","))
2037 new = "?";
2038
2039 if (!old_actions_logged)
2040 old = "(none)";
2041 else if (!seccomp_names_from_actions_logged(old_names,
2042 sizeof(old_names),
2043 old_actions_logged, ","))
2044 old = "?";
2045
2046 return audit_seccomp_actions_logged(new, old, !ret);
2047}
2048
2049static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
2050 void *buffer, size_t *lenp,
2051 loff_t *ppos)
2052{
2053 int ret;
2054
2055 if (write) {
2056 u32 actions_logged = 0;
2057 u32 old_actions_logged = seccomp_actions_logged;
2058
2059 ret = write_actions_logged(ro_table, buffer, lenp, ppos,
2060 &actions_logged);
2061 audit_actions_logged(actions_logged, old_actions_logged, ret);
2062 } else
2063 ret = read_actions_logged(ro_table, buffer, lenp, ppos);
2064
2065 return ret;
2066}
2067
2068static struct ctl_path seccomp_sysctl_path[] = {
2069 { .procname = "kernel", },
2070 { .procname = "seccomp", },
2071 { }
2072};
2073
2074static struct ctl_table seccomp_sysctl_table[] = {
2075 {
2076 .procname = "actions_avail",
2077 .data = (void *) &seccomp_actions_avail,
2078 .maxlen = sizeof(seccomp_actions_avail),
2079 .mode = 0444,
2080 .proc_handler = proc_dostring,
2081 },
2082 {
2083 .procname = "actions_logged",
2084 .mode = 0644,
2085 .proc_handler = seccomp_actions_logged_handler,
2086 },
2087 { }
2088};
2089
2090static int __init seccomp_sysctl_init(void)
2091{
2092 struct ctl_table_header *hdr;
2093
2094 hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
2095 if (!hdr)
2096 pr_warn("sysctl registration failed\n");
2097 else
2098 kmemleak_not_leak(hdr);
2099
2100 return 0;
2101}
2102
2103device_initcall(seccomp_sysctl_init)
2104
2105#endif /* CONFIG_SYSCTL */