393e029f778ab0df3dde07e4957cef2291a62c7b
[sfrench/cifs-2.6.git] / kernel / seccomp.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * linux/kernel/seccomp.c
4  *
5  * Copyright 2004-2005  Andrea Arcangeli <andrea@cpushare.com>
6  *
7  * Copyright (C) 2012 Google, Inc.
8  * Will Drewry <wad@chromium.org>
9  *
10  * This defines a simple but solid secure-computing facility.
11  *
12  * Mode 1 uses a fixed list of allowed system calls.
13  * Mode 2 allows user-defined system call filters in the form
14  *        of Berkeley Packet Filters/Linux Socket Filters.
15  */
16
17 #include <linux/refcount.h>
18 #include <linux/audit.h>
19 #include <linux/compat.h>
20 #include <linux/coredump.h>
21 #include <linux/kmemleak.h>
22 #include <linux/nospec.h>
23 #include <linux/prctl.h>
24 #include <linux/sched.h>
25 #include <linux/sched/task_stack.h>
26 #include <linux/seccomp.h>
27 #include <linux/slab.h>
28 #include <linux/syscalls.h>
29 #include <linux/sysctl.h>
30
31 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
32 #include <asm/syscall.h>
33 #endif
34
35 #ifdef CONFIG_SECCOMP_FILTER
36 #include <linux/filter.h>
37 #include <linux/pid.h>
38 #include <linux/ptrace.h>
39 #include <linux/security.h>
40 #include <linux/tracehook.h>
41 #include <linux/uaccess.h>
42
43 /**
44  * struct seccomp_filter - container for seccomp BPF programs
45  *
46  * @usage: reference count to manage the object lifetime.
47  *         get/put helpers should be used when accessing an instance
48  *         outside of a lifetime-guarded section.  In general, this
49  *         is only needed for handling filters shared across tasks.
50  * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
51  * @prev: points to a previously installed, or inherited, filter
52  * @prog: the BPF program to evaluate
53  *
54  * seccomp_filter objects are organized in a tree linked via the @prev
55  * pointer.  For any task, it appears to be a singly-linked list starting
56  * with current->seccomp.filter, the most recently attached or inherited filter.
57  * However, multiple filters may share a @prev node, by way of fork(), which
58  * results in a unidirectional tree existing in memory.  This is similar to
59  * how namespaces work.
60  *
61  * seccomp_filter objects should never be modified after being attached
62  * to a task_struct (other than @usage).
63  */
64 struct seccomp_filter {
65         refcount_t usage;
66         bool log;
67         struct seccomp_filter *prev;
68         struct bpf_prog *prog;
69 };
70
71 /* Limit any path through the tree to 256KB worth of instructions. */
72 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
73
74 /*
75  * Endianness is explicitly ignored and left for BPF program authors to manage
76  * as per the specific architecture.
77  */
78 static void populate_seccomp_data(struct seccomp_data *sd)
79 {
80         struct task_struct *task = current;
81         struct pt_regs *regs = task_pt_regs(task);
82         unsigned long args[6];
83
84         sd->nr = syscall_get_nr(task, regs);
85         sd->arch = syscall_get_arch();
86         syscall_get_arguments(task, regs, 0, 6, args);
87         sd->args[0] = args[0];
88         sd->args[1] = args[1];
89         sd->args[2] = args[2];
90         sd->args[3] = args[3];
91         sd->args[4] = args[4];
92         sd->args[5] = args[5];
93         sd->instruction_pointer = KSTK_EIP(task);
94 }
95
96 /**
97  *      seccomp_check_filter - verify seccomp filter code
98  *      @filter: filter to verify
99  *      @flen: length of filter
100  *
101  * Takes a previously checked filter (by bpf_check_classic) and
102  * redirects all filter code that loads struct sk_buff data
103  * and related data through seccomp_bpf_load.  It also
104  * enforces length and alignment checking of those loads.
105  *
106  * Returns 0 if the rule set is legal or -EINVAL if not.
107  */
108 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
109 {
110         int pc;
111         for (pc = 0; pc < flen; pc++) {
112                 struct sock_filter *ftest = &filter[pc];
113                 u16 code = ftest->code;
114                 u32 k = ftest->k;
115
116                 switch (code) {
117                 case BPF_LD | BPF_W | BPF_ABS:
118                         ftest->code = BPF_LDX | BPF_W | BPF_ABS;
119                         /* 32-bit aligned and not out of bounds. */
120                         if (k >= sizeof(struct seccomp_data) || k & 3)
121                                 return -EINVAL;
122                         continue;
123                 case BPF_LD | BPF_W | BPF_LEN:
124                         ftest->code = BPF_LD | BPF_IMM;
125                         ftest->k = sizeof(struct seccomp_data);
126                         continue;
127                 case BPF_LDX | BPF_W | BPF_LEN:
128                         ftest->code = BPF_LDX | BPF_IMM;
129                         ftest->k = sizeof(struct seccomp_data);
130                         continue;
131                 /* Explicitly include allowed calls. */
132                 case BPF_RET | BPF_K:
133                 case BPF_RET | BPF_A:
134                 case BPF_ALU | BPF_ADD | BPF_K:
135                 case BPF_ALU | BPF_ADD | BPF_X:
136                 case BPF_ALU | BPF_SUB | BPF_K:
137                 case BPF_ALU | BPF_SUB | BPF_X:
138                 case BPF_ALU | BPF_MUL | BPF_K:
139                 case BPF_ALU | BPF_MUL | BPF_X:
140                 case BPF_ALU | BPF_DIV | BPF_K:
141                 case BPF_ALU | BPF_DIV | BPF_X:
142                 case BPF_ALU | BPF_AND | BPF_K:
143                 case BPF_ALU | BPF_AND | BPF_X:
144                 case BPF_ALU | BPF_OR | BPF_K:
145                 case BPF_ALU | BPF_OR | BPF_X:
146                 case BPF_ALU | BPF_XOR | BPF_K:
147                 case BPF_ALU | BPF_XOR | BPF_X:
148                 case BPF_ALU | BPF_LSH | BPF_K:
149                 case BPF_ALU | BPF_LSH | BPF_X:
150                 case BPF_ALU | BPF_RSH | BPF_K:
151                 case BPF_ALU | BPF_RSH | BPF_X:
152                 case BPF_ALU | BPF_NEG:
153                 case BPF_LD | BPF_IMM:
154                 case BPF_LDX | BPF_IMM:
155                 case BPF_MISC | BPF_TAX:
156                 case BPF_MISC | BPF_TXA:
157                 case BPF_LD | BPF_MEM:
158                 case BPF_LDX | BPF_MEM:
159                 case BPF_ST:
160                 case BPF_STX:
161                 case BPF_JMP | BPF_JA:
162                 case BPF_JMP | BPF_JEQ | BPF_K:
163                 case BPF_JMP | BPF_JEQ | BPF_X:
164                 case BPF_JMP | BPF_JGE | BPF_K:
165                 case BPF_JMP | BPF_JGE | BPF_X:
166                 case BPF_JMP | BPF_JGT | BPF_K:
167                 case BPF_JMP | BPF_JGT | BPF_X:
168                 case BPF_JMP | BPF_JSET | BPF_K:
169                 case BPF_JMP | BPF_JSET | BPF_X:
170                         continue;
171                 default:
172                         return -EINVAL;
173                 }
174         }
175         return 0;
176 }
177
178 /**
179  * seccomp_run_filters - evaluates all seccomp filters against @sd
180  * @sd: optional seccomp data to be passed to filters
181  * @match: stores struct seccomp_filter that resulted in the return value,
182  *         unless filter returned SECCOMP_RET_ALLOW, in which case it will
183  *         be unchanged.
184  *
185  * Returns valid seccomp BPF response codes.
186  */
187 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
188 static u32 seccomp_run_filters(const struct seccomp_data *sd,
189                                struct seccomp_filter **match)
190 {
191         u32 ret = SECCOMP_RET_ALLOW;
192         /* Make sure cross-thread synced filter points somewhere sane. */
193         struct seccomp_filter *f =
194                         READ_ONCE(current->seccomp.filter);
195
196         /* Ensure unexpected behavior doesn't result in failing open. */
197         if (WARN_ON(f == NULL))
198                 return SECCOMP_RET_KILL_PROCESS;
199
200         /*
201          * All filters in the list are evaluated and the lowest BPF return
202          * value always takes priority (ignoring the DATA).
203          */
204         for (; f; f = f->prev) {
205                 u32 cur_ret = BPF_PROG_RUN(f->prog, sd);
206
207                 if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
208                         ret = cur_ret;
209                         *match = f;
210                 }
211         }
212         return ret;
213 }
214 #endif /* CONFIG_SECCOMP_FILTER */
215
216 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
217 {
218         assert_spin_locked(&current->sighand->siglock);
219
220         if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
221                 return false;
222
223         return true;
224 }
225
226 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
227
228 static inline void seccomp_assign_mode(struct task_struct *task,
229                                        unsigned long seccomp_mode,
230                                        unsigned long flags)
231 {
232         assert_spin_locked(&task->sighand->siglock);
233
234         task->seccomp.mode = seccomp_mode;
235         /*
236          * Make sure TIF_SECCOMP cannot be set before the mode (and
237          * filter) is set.
238          */
239         smp_mb__before_atomic();
240         /* Assume default seccomp processes want spec flaw mitigation. */
241         if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
242                 arch_seccomp_spec_mitigate(task);
243         set_tsk_thread_flag(task, TIF_SECCOMP);
244 }
245
246 #ifdef CONFIG_SECCOMP_FILTER
247 /* Returns 1 if the parent is an ancestor of the child. */
248 static int is_ancestor(struct seccomp_filter *parent,
249                        struct seccomp_filter *child)
250 {
251         /* NULL is the root ancestor. */
252         if (parent == NULL)
253                 return 1;
254         for (; child; child = child->prev)
255                 if (child == parent)
256                         return 1;
257         return 0;
258 }
259
260 /**
261  * seccomp_can_sync_threads: checks if all threads can be synchronized
262  *
263  * Expects sighand and cred_guard_mutex locks to be held.
264  *
265  * Returns 0 on success, -ve on error, or the pid of a thread which was
266  * either not in the correct seccomp mode or it did not have an ancestral
267  * seccomp filter.
268  */
269 static inline pid_t seccomp_can_sync_threads(void)
270 {
271         struct task_struct *thread, *caller;
272
273         BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
274         assert_spin_locked(&current->sighand->siglock);
275
276         /* Validate all threads being eligible for synchronization. */
277         caller = current;
278         for_each_thread(caller, thread) {
279                 pid_t failed;
280
281                 /* Skip current, since it is initiating the sync. */
282                 if (thread == caller)
283                         continue;
284
285                 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
286                     (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
287                      is_ancestor(thread->seccomp.filter,
288                                  caller->seccomp.filter)))
289                         continue;
290
291                 /* Return the first thread that cannot be synchronized. */
292                 failed = task_pid_vnr(thread);
293                 /* If the pid cannot be resolved, then return -ESRCH */
294                 if (WARN_ON(failed == 0))
295                         failed = -ESRCH;
296                 return failed;
297         }
298
299         return 0;
300 }
301
302 /**
303  * seccomp_sync_threads: sets all threads to use current's filter
304  *
305  * Expects sighand and cred_guard_mutex locks to be held, and for
306  * seccomp_can_sync_threads() to have returned success already
307  * without dropping the locks.
308  *
309  */
310 static inline void seccomp_sync_threads(unsigned long flags)
311 {
312         struct task_struct *thread, *caller;
313
314         BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
315         assert_spin_locked(&current->sighand->siglock);
316
317         /* Synchronize all threads. */
318         caller = current;
319         for_each_thread(caller, thread) {
320                 /* Skip current, since it needs no changes. */
321                 if (thread == caller)
322                         continue;
323
324                 /* Get a task reference for the new leaf node. */
325                 get_seccomp_filter(caller);
326                 /*
327                  * Drop the task reference to the shared ancestor since
328                  * current's path will hold a reference.  (This also
329                  * allows a put before the assignment.)
330                  */
331                 put_seccomp_filter(thread);
332                 smp_store_release(&thread->seccomp.filter,
333                                   caller->seccomp.filter);
334
335                 /*
336                  * Don't let an unprivileged task work around
337                  * the no_new_privs restriction by creating
338                  * a thread that sets it up, enters seccomp,
339                  * then dies.
340                  */
341                 if (task_no_new_privs(caller))
342                         task_set_no_new_privs(thread);
343
344                 /*
345                  * Opt the other thread into seccomp if needed.
346                  * As threads are considered to be trust-realm
347                  * equivalent (see ptrace_may_access), it is safe to
348                  * allow one thread to transition the other.
349                  */
350                 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
351                         seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
352                                             flags);
353         }
354 }
355
356 /**
357  * seccomp_prepare_filter: Prepares a seccomp filter for use.
358  * @fprog: BPF program to install
359  *
360  * Returns filter on success or an ERR_PTR on failure.
361  */
362 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
363 {
364         struct seccomp_filter *sfilter;
365         int ret;
366         const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
367
368         if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
369                 return ERR_PTR(-EINVAL);
370
371         BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
372
373         /*
374          * Installing a seccomp filter requires that the task has
375          * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
376          * This avoids scenarios where unprivileged tasks can affect the
377          * behavior of privileged children.
378          */
379         if (!task_no_new_privs(current) &&
380             security_capable_noaudit(current_cred(), current_user_ns(),
381                                      CAP_SYS_ADMIN) != 0)
382                 return ERR_PTR(-EACCES);
383
384         /* Allocate a new seccomp_filter */
385         sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
386         if (!sfilter)
387                 return ERR_PTR(-ENOMEM);
388
389         ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
390                                         seccomp_check_filter, save_orig);
391         if (ret < 0) {
392                 kfree(sfilter);
393                 return ERR_PTR(ret);
394         }
395
396         refcount_set(&sfilter->usage, 1);
397
398         return sfilter;
399 }
400
401 /**
402  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
403  * @user_filter: pointer to the user data containing a sock_fprog.
404  *
405  * Returns 0 on success and non-zero otherwise.
406  */
407 static struct seccomp_filter *
408 seccomp_prepare_user_filter(const char __user *user_filter)
409 {
410         struct sock_fprog fprog;
411         struct seccomp_filter *filter = ERR_PTR(-EFAULT);
412
413 #ifdef CONFIG_COMPAT
414         if (in_compat_syscall()) {
415                 struct compat_sock_fprog fprog32;
416                 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
417                         goto out;
418                 fprog.len = fprog32.len;
419                 fprog.filter = compat_ptr(fprog32.filter);
420         } else /* falls through to the if below. */
421 #endif
422         if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
423                 goto out;
424         filter = seccomp_prepare_filter(&fprog);
425 out:
426         return filter;
427 }
428
429 /**
430  * seccomp_attach_filter: validate and attach filter
431  * @flags:  flags to change filter behavior
432  * @filter: seccomp filter to add to the current process
433  *
434  * Caller must be holding current->sighand->siglock lock.
435  *
436  * Returns 0 on success, -ve on error.
437  */
438 static long seccomp_attach_filter(unsigned int flags,
439                                   struct seccomp_filter *filter)
440 {
441         unsigned long total_insns;
442         struct seccomp_filter *walker;
443
444         assert_spin_locked(&current->sighand->siglock);
445
446         /* Validate resulting filter length. */
447         total_insns = filter->prog->len;
448         for (walker = current->seccomp.filter; walker; walker = walker->prev)
449                 total_insns += walker->prog->len + 4;  /* 4 instr penalty */
450         if (total_insns > MAX_INSNS_PER_PATH)
451                 return -ENOMEM;
452
453         /* If thread sync has been requested, check that it is possible. */
454         if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
455                 int ret;
456
457                 ret = seccomp_can_sync_threads();
458                 if (ret)
459                         return ret;
460         }
461
462         /* Set log flag, if present. */
463         if (flags & SECCOMP_FILTER_FLAG_LOG)
464                 filter->log = true;
465
466         /*
467          * If there is an existing filter, make it the prev and don't drop its
468          * task reference.
469          */
470         filter->prev = current->seccomp.filter;
471         current->seccomp.filter = filter;
472
473         /* Now that the new filter is in place, synchronize to all threads. */
474         if (flags & SECCOMP_FILTER_FLAG_TSYNC)
475                 seccomp_sync_threads(flags);
476
477         return 0;
478 }
479
480 static void __get_seccomp_filter(struct seccomp_filter *filter)
481 {
482         /* Reference count is bounded by the number of total processes. */
483         refcount_inc(&filter->usage);
484 }
485
486 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
487 void get_seccomp_filter(struct task_struct *tsk)
488 {
489         struct seccomp_filter *orig = tsk->seccomp.filter;
490         if (!orig)
491                 return;
492         __get_seccomp_filter(orig);
493 }
494
495 static inline void seccomp_filter_free(struct seccomp_filter *filter)
496 {
497         if (filter) {
498                 bpf_prog_destroy(filter->prog);
499                 kfree(filter);
500         }
501 }
502
503 static void __put_seccomp_filter(struct seccomp_filter *orig)
504 {
505         /* Clean up single-reference branches iteratively. */
506         while (orig && refcount_dec_and_test(&orig->usage)) {
507                 struct seccomp_filter *freeme = orig;
508                 orig = orig->prev;
509                 seccomp_filter_free(freeme);
510         }
511 }
512
513 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
514 void put_seccomp_filter(struct task_struct *tsk)
515 {
516         __put_seccomp_filter(tsk->seccomp.filter);
517 }
518
519 static void seccomp_init_siginfo(kernel_siginfo_t *info, int syscall, int reason)
520 {
521         clear_siginfo(info);
522         info->si_signo = SIGSYS;
523         info->si_code = SYS_SECCOMP;
524         info->si_call_addr = (void __user *)KSTK_EIP(current);
525         info->si_errno = reason;
526         info->si_arch = syscall_get_arch();
527         info->si_syscall = syscall;
528 }
529
530 /**
531  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
532  * @syscall: syscall number to send to userland
533  * @reason: filter-supplied reason code to send to userland (via si_errno)
534  *
535  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
536  */
537 static void seccomp_send_sigsys(int syscall, int reason)
538 {
539         struct kernel_siginfo info;
540         seccomp_init_siginfo(&info, syscall, reason);
541         force_sig_info(SIGSYS, &info, current);
542 }
543 #endif  /* CONFIG_SECCOMP_FILTER */
544
545 /* For use with seccomp_actions_logged */
546 #define SECCOMP_LOG_KILL_PROCESS        (1 << 0)
547 #define SECCOMP_LOG_KILL_THREAD         (1 << 1)
548 #define SECCOMP_LOG_TRAP                (1 << 2)
549 #define SECCOMP_LOG_ERRNO               (1 << 3)
550 #define SECCOMP_LOG_TRACE               (1 << 4)
551 #define SECCOMP_LOG_LOG                 (1 << 5)
552 #define SECCOMP_LOG_ALLOW               (1 << 6)
553
554 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
555                                     SECCOMP_LOG_KILL_THREAD  |
556                                     SECCOMP_LOG_TRAP  |
557                                     SECCOMP_LOG_ERRNO |
558                                     SECCOMP_LOG_TRACE |
559                                     SECCOMP_LOG_LOG;
560
561 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
562                                bool requested)
563 {
564         bool log = false;
565
566         switch (action) {
567         case SECCOMP_RET_ALLOW:
568                 break;
569         case SECCOMP_RET_TRAP:
570                 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
571                 break;
572         case SECCOMP_RET_ERRNO:
573                 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
574                 break;
575         case SECCOMP_RET_TRACE:
576                 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
577                 break;
578         case SECCOMP_RET_LOG:
579                 log = seccomp_actions_logged & SECCOMP_LOG_LOG;
580                 break;
581         case SECCOMP_RET_KILL_THREAD:
582                 log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
583                 break;
584         case SECCOMP_RET_KILL_PROCESS:
585         default:
586                 log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
587         }
588
589         /*
590          * Emit an audit message when the action is RET_KILL_*, RET_LOG, or the
591          * FILTER_FLAG_LOG bit was set. The admin has the ability to silence
592          * any action from being logged by removing the action name from the
593          * seccomp_actions_logged sysctl.
594          */
595         if (!log)
596                 return;
597
598         audit_seccomp(syscall, signr, action);
599 }
600
601 /*
602  * Secure computing mode 1 allows only read/write/exit/sigreturn.
603  * To be fully secure this must be combined with rlimit
604  * to limit the stack allocations too.
605  */
606 static const int mode1_syscalls[] = {
607         __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
608         0, /* null terminated */
609 };
610
611 static void __secure_computing_strict(int this_syscall)
612 {
613         const int *syscall_whitelist = mode1_syscalls;
614 #ifdef CONFIG_COMPAT
615         if (in_compat_syscall())
616                 syscall_whitelist = get_compat_mode1_syscalls();
617 #endif
618         do {
619                 if (*syscall_whitelist == this_syscall)
620                         return;
621         } while (*++syscall_whitelist);
622
623 #ifdef SECCOMP_DEBUG
624         dump_stack();
625 #endif
626         seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
627         do_exit(SIGKILL);
628 }
629
630 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
631 void secure_computing_strict(int this_syscall)
632 {
633         int mode = current->seccomp.mode;
634
635         if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
636             unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
637                 return;
638
639         if (mode == SECCOMP_MODE_DISABLED)
640                 return;
641         else if (mode == SECCOMP_MODE_STRICT)
642                 __secure_computing_strict(this_syscall);
643         else
644                 BUG();
645 }
646 #else
647
648 #ifdef CONFIG_SECCOMP_FILTER
649 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
650                             const bool recheck_after_trace)
651 {
652         u32 filter_ret, action;
653         struct seccomp_filter *match = NULL;
654         int data;
655         struct seccomp_data sd_local;
656
657         /*
658          * Make sure that any changes to mode from another thread have
659          * been seen after TIF_SECCOMP was seen.
660          */
661         rmb();
662
663         if (!sd) {
664                 populate_seccomp_data(&sd_local);
665                 sd = &sd_local;
666         }
667
668         filter_ret = seccomp_run_filters(sd, &match);
669         data = filter_ret & SECCOMP_RET_DATA;
670         action = filter_ret & SECCOMP_RET_ACTION_FULL;
671
672         switch (action) {
673         case SECCOMP_RET_ERRNO:
674                 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
675                 if (data > MAX_ERRNO)
676                         data = MAX_ERRNO;
677                 syscall_set_return_value(current, task_pt_regs(current),
678                                          -data, 0);
679                 goto skip;
680
681         case SECCOMP_RET_TRAP:
682                 /* Show the handler the original registers. */
683                 syscall_rollback(current, task_pt_regs(current));
684                 /* Let the filter pass back 16 bits of data. */
685                 seccomp_send_sigsys(this_syscall, data);
686                 goto skip;
687
688         case SECCOMP_RET_TRACE:
689                 /* We've been put in this state by the ptracer already. */
690                 if (recheck_after_trace)
691                         return 0;
692
693                 /* ENOSYS these calls if there is no tracer attached. */
694                 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
695                         syscall_set_return_value(current,
696                                                  task_pt_regs(current),
697                                                  -ENOSYS, 0);
698                         goto skip;
699                 }
700
701                 /* Allow the BPF to provide the event message */
702                 ptrace_event(PTRACE_EVENT_SECCOMP, data);
703                 /*
704                  * The delivery of a fatal signal during event
705                  * notification may silently skip tracer notification,
706                  * which could leave us with a potentially unmodified
707                  * syscall that the tracer would have liked to have
708                  * changed. Since the process is about to die, we just
709                  * force the syscall to be skipped and let the signal
710                  * kill the process and correctly handle any tracer exit
711                  * notifications.
712                  */
713                 if (fatal_signal_pending(current))
714                         goto skip;
715                 /* Check if the tracer forced the syscall to be skipped. */
716                 this_syscall = syscall_get_nr(current, task_pt_regs(current));
717                 if (this_syscall < 0)
718                         goto skip;
719
720                 /*
721                  * Recheck the syscall, since it may have changed. This
722                  * intentionally uses a NULL struct seccomp_data to force
723                  * a reload of all registers. This does not goto skip since
724                  * a skip would have already been reported.
725                  */
726                 if (__seccomp_filter(this_syscall, NULL, true))
727                         return -1;
728
729                 return 0;
730
731         case SECCOMP_RET_LOG:
732                 seccomp_log(this_syscall, 0, action, true);
733                 return 0;
734
735         case SECCOMP_RET_ALLOW:
736                 /*
737                  * Note that the "match" filter will always be NULL for
738                  * this action since SECCOMP_RET_ALLOW is the starting
739                  * state in seccomp_run_filters().
740                  */
741                 return 0;
742
743         case SECCOMP_RET_KILL_THREAD:
744         case SECCOMP_RET_KILL_PROCESS:
745         default:
746                 seccomp_log(this_syscall, SIGSYS, action, true);
747                 /* Dump core only if this is the last remaining thread. */
748                 if (action == SECCOMP_RET_KILL_PROCESS ||
749                     get_nr_threads(current) == 1) {
750                         kernel_siginfo_t info;
751
752                         /* Show the original registers in the dump. */
753                         syscall_rollback(current, task_pt_regs(current));
754                         /* Trigger a manual coredump since do_exit skips it. */
755                         seccomp_init_siginfo(&info, this_syscall, data);
756                         do_coredump(&info);
757                 }
758                 if (action == SECCOMP_RET_KILL_PROCESS)
759                         do_group_exit(SIGSYS);
760                 else
761                         do_exit(SIGSYS);
762         }
763
764         unreachable();
765
766 skip:
767         seccomp_log(this_syscall, 0, action, match ? match->log : false);
768         return -1;
769 }
770 #else
771 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
772                             const bool recheck_after_trace)
773 {
774         BUG();
775 }
776 #endif
777
778 int __secure_computing(const struct seccomp_data *sd)
779 {
780         int mode = current->seccomp.mode;
781         int this_syscall;
782
783         if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
784             unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
785                 return 0;
786
787         this_syscall = sd ? sd->nr :
788                 syscall_get_nr(current, task_pt_regs(current));
789
790         switch (mode) {
791         case SECCOMP_MODE_STRICT:
792                 __secure_computing_strict(this_syscall);  /* may call do_exit */
793                 return 0;
794         case SECCOMP_MODE_FILTER:
795                 return __seccomp_filter(this_syscall, sd, false);
796         default:
797                 BUG();
798         }
799 }
800 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
801
802 long prctl_get_seccomp(void)
803 {
804         return current->seccomp.mode;
805 }
806
807 /**
808  * seccomp_set_mode_strict: internal function for setting strict seccomp
809  *
810  * Once current->seccomp.mode is non-zero, it may not be changed.
811  *
812  * Returns 0 on success or -EINVAL on failure.
813  */
814 static long seccomp_set_mode_strict(void)
815 {
816         const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
817         long ret = -EINVAL;
818
819         spin_lock_irq(&current->sighand->siglock);
820
821         if (!seccomp_may_assign_mode(seccomp_mode))
822                 goto out;
823
824 #ifdef TIF_NOTSC
825         disable_TSC();
826 #endif
827         seccomp_assign_mode(current, seccomp_mode, 0);
828         ret = 0;
829
830 out:
831         spin_unlock_irq(&current->sighand->siglock);
832
833         return ret;
834 }
835
836 #ifdef CONFIG_SECCOMP_FILTER
837 /**
838  * seccomp_set_mode_filter: internal function for setting seccomp filter
839  * @flags:  flags to change filter behavior
840  * @filter: struct sock_fprog containing filter
841  *
842  * This function may be called repeatedly to install additional filters.
843  * Every filter successfully installed will be evaluated (in reverse order)
844  * for each system call the task makes.
845  *
846  * Once current->seccomp.mode is non-zero, it may not be changed.
847  *
848  * Returns 0 on success or -EINVAL on failure.
849  */
850 static long seccomp_set_mode_filter(unsigned int flags,
851                                     const char __user *filter)
852 {
853         const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
854         struct seccomp_filter *prepared = NULL;
855         long ret = -EINVAL;
856
857         /* Validate flags. */
858         if (flags & ~SECCOMP_FILTER_FLAG_MASK)
859                 return -EINVAL;
860
861         /* Prepare the new filter before holding any locks. */
862         prepared = seccomp_prepare_user_filter(filter);
863         if (IS_ERR(prepared))
864                 return PTR_ERR(prepared);
865
866         /*
867          * Make sure we cannot change seccomp or nnp state via TSYNC
868          * while another thread is in the middle of calling exec.
869          */
870         if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
871             mutex_lock_killable(&current->signal->cred_guard_mutex))
872                 goto out_free;
873
874         spin_lock_irq(&current->sighand->siglock);
875
876         if (!seccomp_may_assign_mode(seccomp_mode))
877                 goto out;
878
879         ret = seccomp_attach_filter(flags, prepared);
880         if (ret)
881                 goto out;
882         /* Do not free the successfully attached filter. */
883         prepared = NULL;
884
885         seccomp_assign_mode(current, seccomp_mode, flags);
886 out:
887         spin_unlock_irq(&current->sighand->siglock);
888         if (flags & SECCOMP_FILTER_FLAG_TSYNC)
889                 mutex_unlock(&current->signal->cred_guard_mutex);
890 out_free:
891         seccomp_filter_free(prepared);
892         return ret;
893 }
894 #else
895 static inline long seccomp_set_mode_filter(unsigned int flags,
896                                            const char __user *filter)
897 {
898         return -EINVAL;
899 }
900 #endif
901
902 static long seccomp_get_action_avail(const char __user *uaction)
903 {
904         u32 action;
905
906         if (copy_from_user(&action, uaction, sizeof(action)))
907                 return -EFAULT;
908
909         switch (action) {
910         case SECCOMP_RET_KILL_PROCESS:
911         case SECCOMP_RET_KILL_THREAD:
912         case SECCOMP_RET_TRAP:
913         case SECCOMP_RET_ERRNO:
914         case SECCOMP_RET_TRACE:
915         case SECCOMP_RET_LOG:
916         case SECCOMP_RET_ALLOW:
917                 break;
918         default:
919                 return -EOPNOTSUPP;
920         }
921
922         return 0;
923 }
924
925 /* Common entry point for both prctl and syscall. */
926 static long do_seccomp(unsigned int op, unsigned int flags,
927                        void __user *uargs)
928 {
929         switch (op) {
930         case SECCOMP_SET_MODE_STRICT:
931                 if (flags != 0 || uargs != NULL)
932                         return -EINVAL;
933                 return seccomp_set_mode_strict();
934         case SECCOMP_SET_MODE_FILTER:
935                 return seccomp_set_mode_filter(flags, uargs);
936         case SECCOMP_GET_ACTION_AVAIL:
937                 if (flags != 0)
938                         return -EINVAL;
939
940                 return seccomp_get_action_avail(uargs);
941         default:
942                 return -EINVAL;
943         }
944 }
945
946 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
947                          void __user *, uargs)
948 {
949         return do_seccomp(op, flags, uargs);
950 }
951
952 /**
953  * prctl_set_seccomp: configures current->seccomp.mode
954  * @seccomp_mode: requested mode to use
955  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
956  *
957  * Returns 0 on success or -EINVAL on failure.
958  */
959 long prctl_set_seccomp(unsigned long seccomp_mode, void __user *filter)
960 {
961         unsigned int op;
962         void __user *uargs;
963
964         switch (seccomp_mode) {
965         case SECCOMP_MODE_STRICT:
966                 op = SECCOMP_SET_MODE_STRICT;
967                 /*
968                  * Setting strict mode through prctl always ignored filter,
969                  * so make sure it is always NULL here to pass the internal
970                  * check in do_seccomp().
971                  */
972                 uargs = NULL;
973                 break;
974         case SECCOMP_MODE_FILTER:
975                 op = SECCOMP_SET_MODE_FILTER;
976                 uargs = filter;
977                 break;
978         default:
979                 return -EINVAL;
980         }
981
982         /* prctl interface doesn't have flags, so they are always zero. */
983         return do_seccomp(op, 0, uargs);
984 }
985
986 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
987 static struct seccomp_filter *get_nth_filter(struct task_struct *task,
988                                              unsigned long filter_off)
989 {
990         struct seccomp_filter *orig, *filter;
991         unsigned long count;
992
993         /*
994          * Note: this is only correct because the caller should be the (ptrace)
995          * tracer of the task, otherwise lock_task_sighand is needed.
996          */
997         spin_lock_irq(&task->sighand->siglock);
998
999         if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
1000                 spin_unlock_irq(&task->sighand->siglock);
1001                 return ERR_PTR(-EINVAL);
1002         }
1003
1004         orig = task->seccomp.filter;
1005         __get_seccomp_filter(orig);
1006         spin_unlock_irq(&task->sighand->siglock);
1007
1008         count = 0;
1009         for (filter = orig; filter; filter = filter->prev)
1010                 count++;
1011
1012         if (filter_off >= count) {
1013                 filter = ERR_PTR(-ENOENT);
1014                 goto out;
1015         }
1016
1017         count -= filter_off;
1018         for (filter = orig; filter && count > 1; filter = filter->prev)
1019                 count--;
1020
1021         if (WARN_ON(count != 1 || !filter)) {
1022                 filter = ERR_PTR(-ENOENT);
1023                 goto out;
1024         }
1025
1026         __get_seccomp_filter(filter);
1027
1028 out:
1029         __put_seccomp_filter(orig);
1030         return filter;
1031 }
1032
1033 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
1034                         void __user *data)
1035 {
1036         struct seccomp_filter *filter;
1037         struct sock_fprog_kern *fprog;
1038         long ret;
1039
1040         if (!capable(CAP_SYS_ADMIN) ||
1041             current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1042                 return -EACCES;
1043         }
1044
1045         filter = get_nth_filter(task, filter_off);
1046         if (IS_ERR(filter))
1047                 return PTR_ERR(filter);
1048
1049         fprog = filter->prog->orig_prog;
1050         if (!fprog) {
1051                 /* This must be a new non-cBPF filter, since we save
1052                  * every cBPF filter's orig_prog above when
1053                  * CONFIG_CHECKPOINT_RESTORE is enabled.
1054                  */
1055                 ret = -EMEDIUMTYPE;
1056                 goto out;
1057         }
1058
1059         ret = fprog->len;
1060         if (!data)
1061                 goto out;
1062
1063         if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1064                 ret = -EFAULT;
1065
1066 out:
1067         __put_seccomp_filter(filter);
1068         return ret;
1069 }
1070
1071 long seccomp_get_metadata(struct task_struct *task,
1072                           unsigned long size, void __user *data)
1073 {
1074         long ret;
1075         struct seccomp_filter *filter;
1076         struct seccomp_metadata kmd = {};
1077
1078         if (!capable(CAP_SYS_ADMIN) ||
1079             current->seccomp.mode != SECCOMP_MODE_DISABLED) {
1080                 return -EACCES;
1081         }
1082
1083         size = min_t(unsigned long, size, sizeof(kmd));
1084
1085         if (size < sizeof(kmd.filter_off))
1086                 return -EINVAL;
1087
1088         if (copy_from_user(&kmd.filter_off, data, sizeof(kmd.filter_off)))
1089                 return -EFAULT;
1090
1091         filter = get_nth_filter(task, kmd.filter_off);
1092         if (IS_ERR(filter))
1093                 return PTR_ERR(filter);
1094
1095         if (filter->log)
1096                 kmd.flags |= SECCOMP_FILTER_FLAG_LOG;
1097
1098         ret = size;
1099         if (copy_to_user(data, &kmd, size))
1100                 ret = -EFAULT;
1101
1102         __put_seccomp_filter(filter);
1103         return ret;
1104 }
1105 #endif
1106
1107 #ifdef CONFIG_SYSCTL
1108
1109 /* Human readable action names for friendly sysctl interaction */
1110 #define SECCOMP_RET_KILL_PROCESS_NAME   "kill_process"
1111 #define SECCOMP_RET_KILL_THREAD_NAME    "kill_thread"
1112 #define SECCOMP_RET_TRAP_NAME           "trap"
1113 #define SECCOMP_RET_ERRNO_NAME          "errno"
1114 #define SECCOMP_RET_TRACE_NAME          "trace"
1115 #define SECCOMP_RET_LOG_NAME            "log"
1116 #define SECCOMP_RET_ALLOW_NAME          "allow"
1117
1118 static const char seccomp_actions_avail[] =
1119                                 SECCOMP_RET_KILL_PROCESS_NAME   " "
1120                                 SECCOMP_RET_KILL_THREAD_NAME    " "
1121                                 SECCOMP_RET_TRAP_NAME           " "
1122                                 SECCOMP_RET_ERRNO_NAME          " "
1123                                 SECCOMP_RET_TRACE_NAME          " "
1124                                 SECCOMP_RET_LOG_NAME            " "
1125                                 SECCOMP_RET_ALLOW_NAME;
1126
1127 struct seccomp_log_name {
1128         u32             log;
1129         const char      *name;
1130 };
1131
1132 static const struct seccomp_log_name seccomp_log_names[] = {
1133         { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
1134         { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1135         { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1136         { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1137         { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1138         { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1139         { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1140         { }
1141 };
1142
1143 static bool seccomp_names_from_actions_logged(char *names, size_t size,
1144                                               u32 actions_logged,
1145                                               const char *sep)
1146 {
1147         const struct seccomp_log_name *cur;
1148         bool append_sep = false;
1149
1150         for (cur = seccomp_log_names; cur->name && size; cur++) {
1151                 ssize_t ret;
1152
1153                 if (!(actions_logged & cur->log))
1154                         continue;
1155
1156                 if (append_sep) {
1157                         ret = strscpy(names, sep, size);
1158                         if (ret < 0)
1159                                 return false;
1160
1161                         names += ret;
1162                         size -= ret;
1163                 } else
1164                         append_sep = true;
1165
1166                 ret = strscpy(names, cur->name, size);
1167                 if (ret < 0)
1168                         return false;
1169
1170                 names += ret;
1171                 size -= ret;
1172         }
1173
1174         return true;
1175 }
1176
1177 static bool seccomp_action_logged_from_name(u32 *action_logged,
1178                                             const char *name)
1179 {
1180         const struct seccomp_log_name *cur;
1181
1182         for (cur = seccomp_log_names; cur->name; cur++) {
1183                 if (!strcmp(cur->name, name)) {
1184                         *action_logged = cur->log;
1185                         return true;
1186                 }
1187         }
1188
1189         return false;
1190 }
1191
1192 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1193 {
1194         char *name;
1195
1196         *actions_logged = 0;
1197         while ((name = strsep(&names, " ")) && *name) {
1198                 u32 action_logged = 0;
1199
1200                 if (!seccomp_action_logged_from_name(&action_logged, name))
1201                         return false;
1202
1203                 *actions_logged |= action_logged;
1204         }
1205
1206         return true;
1207 }
1208
1209 static int read_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1210                                size_t *lenp, loff_t *ppos)
1211 {
1212         char names[sizeof(seccomp_actions_avail)];
1213         struct ctl_table table;
1214
1215         memset(names, 0, sizeof(names));
1216
1217         if (!seccomp_names_from_actions_logged(names, sizeof(names),
1218                                                seccomp_actions_logged, " "))
1219                 return -EINVAL;
1220
1221         table = *ro_table;
1222         table.data = names;
1223         table.maxlen = sizeof(names);
1224         return proc_dostring(&table, 0, buffer, lenp, ppos);
1225 }
1226
1227 static int write_actions_logged(struct ctl_table *ro_table, void __user *buffer,
1228                                 size_t *lenp, loff_t *ppos, u32 *actions_logged)
1229 {
1230         char names[sizeof(seccomp_actions_avail)];
1231         struct ctl_table table;
1232         int ret;
1233
1234         if (!capable(CAP_SYS_ADMIN))
1235                 return -EPERM;
1236
1237         memset(names, 0, sizeof(names));
1238
1239         table = *ro_table;
1240         table.data = names;
1241         table.maxlen = sizeof(names);
1242         ret = proc_dostring(&table, 1, buffer, lenp, ppos);
1243         if (ret)
1244                 return ret;
1245
1246         if (!seccomp_actions_logged_from_names(actions_logged, table.data))
1247                 return -EINVAL;
1248
1249         if (*actions_logged & SECCOMP_LOG_ALLOW)
1250                 return -EINVAL;
1251
1252         seccomp_actions_logged = *actions_logged;
1253         return 0;
1254 }
1255
1256 static void audit_actions_logged(u32 actions_logged, u32 old_actions_logged,
1257                                  int ret)
1258 {
1259         char names[sizeof(seccomp_actions_avail)];
1260         char old_names[sizeof(seccomp_actions_avail)];
1261         const char *new = names;
1262         const char *old = old_names;
1263
1264         if (!audit_enabled)
1265                 return;
1266
1267         memset(names, 0, sizeof(names));
1268         memset(old_names, 0, sizeof(old_names));
1269
1270         if (ret)
1271                 new = "?";
1272         else if (!actions_logged)
1273                 new = "(none)";
1274         else if (!seccomp_names_from_actions_logged(names, sizeof(names),
1275                                                     actions_logged, ","))
1276                 new = "?";
1277
1278         if (!old_actions_logged)
1279                 old = "(none)";
1280         else if (!seccomp_names_from_actions_logged(old_names,
1281                                                     sizeof(old_names),
1282                                                     old_actions_logged, ","))
1283                 old = "?";
1284
1285         return audit_seccomp_actions_logged(new, old, !ret);
1286 }
1287
1288 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
1289                                           void __user *buffer, size_t *lenp,
1290                                           loff_t *ppos)
1291 {
1292         int ret;
1293
1294         if (write) {
1295                 u32 actions_logged = 0;
1296                 u32 old_actions_logged = seccomp_actions_logged;
1297
1298                 ret = write_actions_logged(ro_table, buffer, lenp, ppos,
1299                                            &actions_logged);
1300                 audit_actions_logged(actions_logged, old_actions_logged, ret);
1301         } else
1302                 ret = read_actions_logged(ro_table, buffer, lenp, ppos);
1303
1304         return ret;
1305 }
1306
1307 static struct ctl_path seccomp_sysctl_path[] = {
1308         { .procname = "kernel", },
1309         { .procname = "seccomp", },
1310         { }
1311 };
1312
1313 static struct ctl_table seccomp_sysctl_table[] = {
1314         {
1315                 .procname       = "actions_avail",
1316                 .data           = (void *) &seccomp_actions_avail,
1317                 .maxlen         = sizeof(seccomp_actions_avail),
1318                 .mode           = 0444,
1319                 .proc_handler   = proc_dostring,
1320         },
1321         {
1322                 .procname       = "actions_logged",
1323                 .mode           = 0644,
1324                 .proc_handler   = seccomp_actions_logged_handler,
1325         },
1326         { }
1327 };
1328
1329 static int __init seccomp_sysctl_init(void)
1330 {
1331         struct ctl_table_header *hdr;
1332
1333         hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
1334         if (!hdr)
1335                 pr_warn("seccomp: sysctl registration failed\n");
1336         else
1337                 kmemleak_not_leak(hdr);
1338
1339         return 0;
1340 }
1341
1342 device_initcall(seccomp_sysctl_init)
1343
1344 #endif /* CONFIG_SYSCTL */