2 * linux/kernel/seccomp.c
4 * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com>
6 * Copyright (C) 2012 Google, Inc.
7 * Will Drewry <wad@chromium.org>
9 * This defines a simple but solid secure-computing facility.
11 * Mode 1 uses a fixed list of allowed system calls.
12 * Mode 2 allows user-defined system call filters in the form
13 * of Berkeley Packet Filters/Linux Socket Filters.
16 #include <linux/refcount.h>
17 #include <linux/audit.h>
18 #include <linux/compat.h>
19 #include <linux/coredump.h>
20 #include <linux/kmemleak.h>
21 #include <linux/sched.h>
22 #include <linux/sched/task_stack.h>
23 #include <linux/seccomp.h>
24 #include <linux/slab.h>
25 #include <linux/syscalls.h>
26 #include <linux/sysctl.h>
28 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
29 #include <asm/syscall.h>
32 #ifdef CONFIG_SECCOMP_FILTER
33 #include <linux/filter.h>
34 #include <linux/pid.h>
35 #include <linux/ptrace.h>
36 #include <linux/security.h>
37 #include <linux/tracehook.h>
38 #include <linux/uaccess.h>
41 * struct seccomp_filter - container for seccomp BPF programs
43 * @usage: reference count to manage the object lifetime.
44 * get/put helpers should be used when accessing an instance
45 * outside of a lifetime-guarded section. In general, this
46 * is only needed for handling filters shared across tasks.
47 * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
48 * @prev: points to a previously installed, or inherited, filter
49 * @prog: the BPF program to evaluate
51 * seccomp_filter objects are organized in a tree linked via the @prev
52 * pointer. For any task, it appears to be a singly-linked list starting
53 * with current->seccomp.filter, the most recently attached or inherited filter.
54 * However, multiple filters may share a @prev node, by way of fork(), which
55 * results in a unidirectional tree existing in memory. This is similar to
56 * how namespaces work.
58 * seccomp_filter objects should never be modified after being attached
59 * to a task_struct (other than @usage).
61 struct seccomp_filter {
64 struct seccomp_filter *prev;
65 struct bpf_prog *prog;
68 /* Limit any path through the tree to 256KB worth of instructions. */
69 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
72 * Endianness is explicitly ignored and left for BPF program authors to manage
73 * as per the specific architecture.
75 static void populate_seccomp_data(struct seccomp_data *sd)
77 struct task_struct *task = current;
78 struct pt_regs *regs = task_pt_regs(task);
79 unsigned long args[6];
81 sd->nr = syscall_get_nr(task, regs);
82 sd->arch = syscall_get_arch();
83 syscall_get_arguments(task, regs, 0, 6, args);
84 sd->args[0] = args[0];
85 sd->args[1] = args[1];
86 sd->args[2] = args[2];
87 sd->args[3] = args[3];
88 sd->args[4] = args[4];
89 sd->args[5] = args[5];
90 sd->instruction_pointer = KSTK_EIP(task);
94 * seccomp_check_filter - verify seccomp filter code
95 * @filter: filter to verify
96 * @flen: length of filter
98 * Takes a previously checked filter (by bpf_check_classic) and
99 * redirects all filter code that loads struct sk_buff data
100 * and related data through seccomp_bpf_load. It also
101 * enforces length and alignment checking of those loads.
103 * Returns 0 if the rule set is legal or -EINVAL if not.
105 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
108 for (pc = 0; pc < flen; pc++) {
109 struct sock_filter *ftest = &filter[pc];
110 u16 code = ftest->code;
114 case BPF_LD | BPF_W | BPF_ABS:
115 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
116 /* 32-bit aligned and not out of bounds. */
117 if (k >= sizeof(struct seccomp_data) || k & 3)
120 case BPF_LD | BPF_W | BPF_LEN:
121 ftest->code = BPF_LD | BPF_IMM;
122 ftest->k = sizeof(struct seccomp_data);
124 case BPF_LDX | BPF_W | BPF_LEN:
125 ftest->code = BPF_LDX | BPF_IMM;
126 ftest->k = sizeof(struct seccomp_data);
128 /* Explicitly include allowed calls. */
129 case BPF_RET | BPF_K:
130 case BPF_RET | BPF_A:
131 case BPF_ALU | BPF_ADD | BPF_K:
132 case BPF_ALU | BPF_ADD | BPF_X:
133 case BPF_ALU | BPF_SUB | BPF_K:
134 case BPF_ALU | BPF_SUB | BPF_X:
135 case BPF_ALU | BPF_MUL | BPF_K:
136 case BPF_ALU | BPF_MUL | BPF_X:
137 case BPF_ALU | BPF_DIV | BPF_K:
138 case BPF_ALU | BPF_DIV | BPF_X:
139 case BPF_ALU | BPF_AND | BPF_K:
140 case BPF_ALU | BPF_AND | BPF_X:
141 case BPF_ALU | BPF_OR | BPF_K:
142 case BPF_ALU | BPF_OR | BPF_X:
143 case BPF_ALU | BPF_XOR | BPF_K:
144 case BPF_ALU | BPF_XOR | BPF_X:
145 case BPF_ALU | BPF_LSH | BPF_K:
146 case BPF_ALU | BPF_LSH | BPF_X:
147 case BPF_ALU | BPF_RSH | BPF_K:
148 case BPF_ALU | BPF_RSH | BPF_X:
149 case BPF_ALU | BPF_NEG:
150 case BPF_LD | BPF_IMM:
151 case BPF_LDX | BPF_IMM:
152 case BPF_MISC | BPF_TAX:
153 case BPF_MISC | BPF_TXA:
154 case BPF_LD | BPF_MEM:
155 case BPF_LDX | BPF_MEM:
158 case BPF_JMP | BPF_JA:
159 case BPF_JMP | BPF_JEQ | BPF_K:
160 case BPF_JMP | BPF_JEQ | BPF_X:
161 case BPF_JMP | BPF_JGE | BPF_K:
162 case BPF_JMP | BPF_JGE | BPF_X:
163 case BPF_JMP | BPF_JGT | BPF_K:
164 case BPF_JMP | BPF_JGT | BPF_X:
165 case BPF_JMP | BPF_JSET | BPF_K:
166 case BPF_JMP | BPF_JSET | BPF_X:
176 * seccomp_run_filters - evaluates all seccomp filters against @sd
177 * @sd: optional seccomp data to be passed to filters
178 * @match: stores struct seccomp_filter that resulted in the return value,
179 * unless filter returned SECCOMP_RET_ALLOW, in which case it will
182 * Returns valid seccomp BPF response codes.
184 static u32 seccomp_run_filters(const struct seccomp_data *sd,
185 struct seccomp_filter **match)
187 struct seccomp_data sd_local;
188 u32 ret = SECCOMP_RET_ALLOW;
189 /* Make sure cross-thread synced filter points somewhere sane. */
190 struct seccomp_filter *f =
191 lockless_dereference(current->seccomp.filter);
193 /* Ensure unexpected behavior doesn't result in failing open. */
194 if (unlikely(WARN_ON(f == NULL)))
195 return SECCOMP_RET_KILL_PROCESS;
198 populate_seccomp_data(&sd_local);
203 * All filters in the list are evaluated and the lowest BPF return
204 * value always takes priority (ignoring the DATA).
206 for (; f; f = f->prev) {
207 u32 cur_ret = BPF_PROG_RUN(f->prog, sd);
209 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION)) {
216 #endif /* CONFIG_SECCOMP_FILTER */
218 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
220 assert_spin_locked(¤t->sighand->siglock);
222 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
228 static inline void seccomp_assign_mode(struct task_struct *task,
229 unsigned long seccomp_mode)
231 assert_spin_locked(&task->sighand->siglock);
233 task->seccomp.mode = seccomp_mode;
235 * Make sure TIF_SECCOMP cannot be set before the mode (and
238 smp_mb__before_atomic();
239 set_tsk_thread_flag(task, TIF_SECCOMP);
242 #ifdef CONFIG_SECCOMP_FILTER
243 /* Returns 1 if the parent is an ancestor of the child. */
244 static int is_ancestor(struct seccomp_filter *parent,
245 struct seccomp_filter *child)
247 /* NULL is the root ancestor. */
250 for (; child; child = child->prev)
257 * seccomp_can_sync_threads: checks if all threads can be synchronized
259 * Expects sighand and cred_guard_mutex locks to be held.
261 * Returns 0 on success, -ve on error, or the pid of a thread which was
262 * either not in the correct seccomp mode or it did not have an ancestral
265 static inline pid_t seccomp_can_sync_threads(void)
267 struct task_struct *thread, *caller;
269 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
270 assert_spin_locked(¤t->sighand->siglock);
272 /* Validate all threads being eligible for synchronization. */
274 for_each_thread(caller, thread) {
277 /* Skip current, since it is initiating the sync. */
278 if (thread == caller)
281 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
282 (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
283 is_ancestor(thread->seccomp.filter,
284 caller->seccomp.filter)))
287 /* Return the first thread that cannot be synchronized. */
288 failed = task_pid_vnr(thread);
289 /* If the pid cannot be resolved, then return -ESRCH */
290 if (unlikely(WARN_ON(failed == 0)))
299 * seccomp_sync_threads: sets all threads to use current's filter
301 * Expects sighand and cred_guard_mutex locks to be held, and for
302 * seccomp_can_sync_threads() to have returned success already
303 * without dropping the locks.
306 static inline void seccomp_sync_threads(void)
308 struct task_struct *thread, *caller;
310 BUG_ON(!mutex_is_locked(¤t->signal->cred_guard_mutex));
311 assert_spin_locked(¤t->sighand->siglock);
313 /* Synchronize all threads. */
315 for_each_thread(caller, thread) {
316 /* Skip current, since it needs no changes. */
317 if (thread == caller)
320 /* Get a task reference for the new leaf node. */
321 get_seccomp_filter(caller);
323 * Drop the task reference to the shared ancestor since
324 * current's path will hold a reference. (This also
325 * allows a put before the assignment.)
327 put_seccomp_filter(thread);
328 smp_store_release(&thread->seccomp.filter,
329 caller->seccomp.filter);
332 * Don't let an unprivileged task work around
333 * the no_new_privs restriction by creating
334 * a thread that sets it up, enters seccomp,
337 if (task_no_new_privs(caller))
338 task_set_no_new_privs(thread);
341 * Opt the other thread into seccomp if needed.
342 * As threads are considered to be trust-realm
343 * equivalent (see ptrace_may_access), it is safe to
344 * allow one thread to transition the other.
346 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
347 seccomp_assign_mode(thread, SECCOMP_MODE_FILTER);
352 * seccomp_prepare_filter: Prepares a seccomp filter for use.
353 * @fprog: BPF program to install
355 * Returns filter on success or an ERR_PTR on failure.
357 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
359 struct seccomp_filter *sfilter;
361 const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
363 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
364 return ERR_PTR(-EINVAL);
366 BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
369 * Installing a seccomp filter requires that the task has
370 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
371 * This avoids scenarios where unprivileged tasks can affect the
372 * behavior of privileged children.
374 if (!task_no_new_privs(current) &&
375 security_capable_noaudit(current_cred(), current_user_ns(),
377 return ERR_PTR(-EACCES);
379 /* Allocate a new seccomp_filter */
380 sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
382 return ERR_PTR(-ENOMEM);
384 ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
385 seccomp_check_filter, save_orig);
391 refcount_set(&sfilter->usage, 1);
397 * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
398 * @user_filter: pointer to the user data containing a sock_fprog.
400 * Returns 0 on success and non-zero otherwise.
402 static struct seccomp_filter *
403 seccomp_prepare_user_filter(const char __user *user_filter)
405 struct sock_fprog fprog;
406 struct seccomp_filter *filter = ERR_PTR(-EFAULT);
409 if (in_compat_syscall()) {
410 struct compat_sock_fprog fprog32;
411 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
413 fprog.len = fprog32.len;
414 fprog.filter = compat_ptr(fprog32.filter);
415 } else /* falls through to the if below. */
417 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
419 filter = seccomp_prepare_filter(&fprog);
425 * seccomp_attach_filter: validate and attach filter
426 * @flags: flags to change filter behavior
427 * @filter: seccomp filter to add to the current process
429 * Caller must be holding current->sighand->siglock lock.
431 * Returns 0 on success, -ve on error.
433 static long seccomp_attach_filter(unsigned int flags,
434 struct seccomp_filter *filter)
436 unsigned long total_insns;
437 struct seccomp_filter *walker;
439 assert_spin_locked(¤t->sighand->siglock);
441 /* Validate resulting filter length. */
442 total_insns = filter->prog->len;
443 for (walker = current->seccomp.filter; walker; walker = walker->prev)
444 total_insns += walker->prog->len + 4; /* 4 instr penalty */
445 if (total_insns > MAX_INSNS_PER_PATH)
448 /* If thread sync has been requested, check that it is possible. */
449 if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
452 ret = seccomp_can_sync_threads();
457 /* Set log flag, if present. */
458 if (flags & SECCOMP_FILTER_FLAG_LOG)
462 * If there is an existing filter, make it the prev and don't drop its
465 filter->prev = current->seccomp.filter;
466 current->seccomp.filter = filter;
468 /* Now that the new filter is in place, synchronize to all threads. */
469 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
470 seccomp_sync_threads();
475 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
476 void get_seccomp_filter(struct task_struct *tsk)
478 struct seccomp_filter *orig = tsk->seccomp.filter;
481 /* Reference count is bounded by the number of total processes. */
482 refcount_inc(&orig->usage);
485 static inline void seccomp_filter_free(struct seccomp_filter *filter)
488 bpf_prog_destroy(filter->prog);
493 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
494 void put_seccomp_filter(struct task_struct *tsk)
496 struct seccomp_filter *orig = tsk->seccomp.filter;
497 /* Clean up single-reference branches iteratively. */
498 while (orig && refcount_dec_and_test(&orig->usage)) {
499 struct seccomp_filter *freeme = orig;
501 seccomp_filter_free(freeme);
505 static void seccomp_init_siginfo(siginfo_t *info, int syscall, int reason)
507 memset(info, 0, sizeof(*info));
508 info->si_signo = SIGSYS;
509 info->si_code = SYS_SECCOMP;
510 info->si_call_addr = (void __user *)KSTK_EIP(current);
511 info->si_errno = reason;
512 info->si_arch = syscall_get_arch();
513 info->si_syscall = syscall;
517 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
518 * @syscall: syscall number to send to userland
519 * @reason: filter-supplied reason code to send to userland (via si_errno)
521 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
523 static void seccomp_send_sigsys(int syscall, int reason)
526 seccomp_init_siginfo(&info, syscall, reason);
527 force_sig_info(SIGSYS, &info, current);
529 #endif /* CONFIG_SECCOMP_FILTER */
531 /* For use with seccomp_actions_logged */
532 #define SECCOMP_LOG_KILL_PROCESS (1 << 0)
533 #define SECCOMP_LOG_KILL_THREAD (1 << 1)
534 #define SECCOMP_LOG_TRAP (1 << 2)
535 #define SECCOMP_LOG_ERRNO (1 << 3)
536 #define SECCOMP_LOG_TRACE (1 << 4)
537 #define SECCOMP_LOG_LOG (1 << 5)
538 #define SECCOMP_LOG_ALLOW (1 << 6)
540 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
541 SECCOMP_LOG_KILL_THREAD |
547 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
553 case SECCOMP_RET_ALLOW:
555 case SECCOMP_RET_TRAP:
556 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
558 case SECCOMP_RET_ERRNO:
559 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
561 case SECCOMP_RET_TRACE:
562 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
564 case SECCOMP_RET_LOG:
565 log = seccomp_actions_logged & SECCOMP_LOG_LOG;
567 case SECCOMP_RET_KILL_THREAD:
568 log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
570 case SECCOMP_RET_KILL_PROCESS:
572 log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
576 * Force an audit message to be emitted when the action is RET_KILL_*,
577 * RET_LOG, or the FILTER_FLAG_LOG bit was set and the action is
578 * allowed to be logged by the admin.
581 return __audit_seccomp(syscall, signr, action);
584 * Let the audit subsystem decide if the action should be audited based
585 * on whether the current task itself is being audited.
587 return audit_seccomp(syscall, signr, action);
591 * Secure computing mode 1 allows only read/write/exit/sigreturn.
592 * To be fully secure this must be combined with rlimit
593 * to limit the stack allocations too.
595 static const int mode1_syscalls[] = {
596 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
597 0, /* null terminated */
600 static void __secure_computing_strict(int this_syscall)
602 const int *syscall_whitelist = mode1_syscalls;
604 if (in_compat_syscall())
605 syscall_whitelist = get_compat_mode1_syscalls();
608 if (*syscall_whitelist == this_syscall)
610 } while (*++syscall_whitelist);
615 seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
619 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
620 void secure_computing_strict(int this_syscall)
622 int mode = current->seccomp.mode;
624 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
625 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
628 if (mode == SECCOMP_MODE_DISABLED)
630 else if (mode == SECCOMP_MODE_STRICT)
631 __secure_computing_strict(this_syscall);
637 #ifdef CONFIG_SECCOMP_FILTER
638 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
639 const bool recheck_after_trace)
641 u32 filter_ret, action;
642 struct seccomp_filter *match = NULL;
646 * Make sure that any changes to mode from another thread have
647 * been seen after TIF_SECCOMP was seen.
651 filter_ret = seccomp_run_filters(sd, &match);
652 data = filter_ret & SECCOMP_RET_DATA;
653 action = filter_ret & SECCOMP_RET_ACTION;
656 case SECCOMP_RET_ERRNO:
657 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
658 if (data > MAX_ERRNO)
660 syscall_set_return_value(current, task_pt_regs(current),
664 case SECCOMP_RET_TRAP:
665 /* Show the handler the original registers. */
666 syscall_rollback(current, task_pt_regs(current));
667 /* Let the filter pass back 16 bits of data. */
668 seccomp_send_sigsys(this_syscall, data);
671 case SECCOMP_RET_TRACE:
672 /* We've been put in this state by the ptracer already. */
673 if (recheck_after_trace)
676 /* ENOSYS these calls if there is no tracer attached. */
677 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
678 syscall_set_return_value(current,
679 task_pt_regs(current),
684 /* Allow the BPF to provide the event message */
685 ptrace_event(PTRACE_EVENT_SECCOMP, data);
687 * The delivery of a fatal signal during event
688 * notification may silently skip tracer notification,
689 * which could leave us with a potentially unmodified
690 * syscall that the tracer would have liked to have
691 * changed. Since the process is about to die, we just
692 * force the syscall to be skipped and let the signal
693 * kill the process and correctly handle any tracer exit
696 if (fatal_signal_pending(current))
698 /* Check if the tracer forced the syscall to be skipped. */
699 this_syscall = syscall_get_nr(current, task_pt_regs(current));
700 if (this_syscall < 0)
704 * Recheck the syscall, since it may have changed. This
705 * intentionally uses a NULL struct seccomp_data to force
706 * a reload of all registers. This does not goto skip since
707 * a skip would have already been reported.
709 if (__seccomp_filter(this_syscall, NULL, true))
714 case SECCOMP_RET_LOG:
715 seccomp_log(this_syscall, 0, action, true);
718 case SECCOMP_RET_ALLOW:
720 * Note that the "match" filter will always be NULL for
721 * this action since SECCOMP_RET_ALLOW is the starting
722 * state in seccomp_run_filters().
726 case SECCOMP_RET_KILL_THREAD:
727 case SECCOMP_RET_KILL_PROCESS:
729 seccomp_log(this_syscall, SIGSYS, action, true);
730 /* Dump core only if this is the last remaining thread. */
731 if (action == SECCOMP_RET_KILL_PROCESS ||
732 get_nr_threads(current) == 1) {
735 /* Show the original registers in the dump. */
736 syscall_rollback(current, task_pt_regs(current));
737 /* Trigger a manual coredump since do_exit skips it. */
738 seccomp_init_siginfo(&info, this_syscall, data);
741 if (action == SECCOMP_RET_KILL_PROCESS)
742 do_group_exit(SIGSYS);
750 seccomp_log(this_syscall, 0, action, match ? match->log : false);
754 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
755 const bool recheck_after_trace)
761 int __secure_computing(const struct seccomp_data *sd)
763 int mode = current->seccomp.mode;
766 if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
767 unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
770 this_syscall = sd ? sd->nr :
771 syscall_get_nr(current, task_pt_regs(current));
774 case SECCOMP_MODE_STRICT:
775 __secure_computing_strict(this_syscall); /* may call do_exit */
777 case SECCOMP_MODE_FILTER:
778 return __seccomp_filter(this_syscall, sd, false);
783 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
785 long prctl_get_seccomp(void)
787 return current->seccomp.mode;
791 * seccomp_set_mode_strict: internal function for setting strict seccomp
793 * Once current->seccomp.mode is non-zero, it may not be changed.
795 * Returns 0 on success or -EINVAL on failure.
797 static long seccomp_set_mode_strict(void)
799 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
802 spin_lock_irq(¤t->sighand->siglock);
804 if (!seccomp_may_assign_mode(seccomp_mode))
810 seccomp_assign_mode(current, seccomp_mode);
814 spin_unlock_irq(¤t->sighand->siglock);
819 #ifdef CONFIG_SECCOMP_FILTER
821 * seccomp_set_mode_filter: internal function for setting seccomp filter
822 * @flags: flags to change filter behavior
823 * @filter: struct sock_fprog containing filter
825 * This function may be called repeatedly to install additional filters.
826 * Every filter successfully installed will be evaluated (in reverse order)
827 * for each system call the task makes.
829 * Once current->seccomp.mode is non-zero, it may not be changed.
831 * Returns 0 on success or -EINVAL on failure.
833 static long seccomp_set_mode_filter(unsigned int flags,
834 const char __user *filter)
836 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
837 struct seccomp_filter *prepared = NULL;
840 /* Validate flags. */
841 if (flags & ~SECCOMP_FILTER_FLAG_MASK)
844 /* Prepare the new filter before holding any locks. */
845 prepared = seccomp_prepare_user_filter(filter);
846 if (IS_ERR(prepared))
847 return PTR_ERR(prepared);
850 * Make sure we cannot change seccomp or nnp state via TSYNC
851 * while another thread is in the middle of calling exec.
853 if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
854 mutex_lock_killable(¤t->signal->cred_guard_mutex))
857 spin_lock_irq(¤t->sighand->siglock);
859 if (!seccomp_may_assign_mode(seccomp_mode))
862 ret = seccomp_attach_filter(flags, prepared);
865 /* Do not free the successfully attached filter. */
868 seccomp_assign_mode(current, seccomp_mode);
870 spin_unlock_irq(¤t->sighand->siglock);
871 if (flags & SECCOMP_FILTER_FLAG_TSYNC)
872 mutex_unlock(¤t->signal->cred_guard_mutex);
874 seccomp_filter_free(prepared);
878 static inline long seccomp_set_mode_filter(unsigned int flags,
879 const char __user *filter)
885 static long seccomp_get_action_avail(const char __user *uaction)
889 if (copy_from_user(&action, uaction, sizeof(action)))
893 case SECCOMP_RET_KILL_THREAD:
894 case SECCOMP_RET_TRAP:
895 case SECCOMP_RET_ERRNO:
896 case SECCOMP_RET_TRACE:
897 case SECCOMP_RET_LOG:
898 case SECCOMP_RET_ALLOW:
907 /* Common entry point for both prctl and syscall. */
908 static long do_seccomp(unsigned int op, unsigned int flags,
909 const char __user *uargs)
912 case SECCOMP_SET_MODE_STRICT:
913 if (flags != 0 || uargs != NULL)
915 return seccomp_set_mode_strict();
916 case SECCOMP_SET_MODE_FILTER:
917 return seccomp_set_mode_filter(flags, uargs);
918 case SECCOMP_GET_ACTION_AVAIL:
922 return seccomp_get_action_avail(uargs);
928 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
929 const char __user *, uargs)
931 return do_seccomp(op, flags, uargs);
935 * prctl_set_seccomp: configures current->seccomp.mode
936 * @seccomp_mode: requested mode to use
937 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
939 * Returns 0 on success or -EINVAL on failure.
941 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
946 switch (seccomp_mode) {
947 case SECCOMP_MODE_STRICT:
948 op = SECCOMP_SET_MODE_STRICT;
950 * Setting strict mode through prctl always ignored filter,
951 * so make sure it is always NULL here to pass the internal
952 * check in do_seccomp().
956 case SECCOMP_MODE_FILTER:
957 op = SECCOMP_SET_MODE_FILTER;
964 /* prctl interface doesn't have flags, so they are always zero. */
965 return do_seccomp(op, 0, uargs);
968 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
969 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
972 struct seccomp_filter *filter;
973 struct sock_fprog_kern *fprog;
975 unsigned long count = 0;
977 if (!capable(CAP_SYS_ADMIN) ||
978 current->seccomp.mode != SECCOMP_MODE_DISABLED) {
982 spin_lock_irq(&task->sighand->siglock);
983 if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
988 filter = task->seccomp.filter;
990 filter = filter->prev;
994 if (filter_off >= count) {
1000 filter = task->seccomp.filter;
1001 while (filter && count > 1) {
1002 filter = filter->prev;
1006 if (WARN_ON(count != 1 || !filter)) {
1007 /* The filter tree shouldn't shrink while we're using it. */
1012 fprog = filter->prog->orig_prog;
1014 /* This must be a new non-cBPF filter, since we save
1015 * every cBPF filter's orig_prog above when
1016 * CONFIG_CHECKPOINT_RESTORE is enabled.
1026 get_seccomp_filter(task);
1027 spin_unlock_irq(&task->sighand->siglock);
1029 if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1032 put_seccomp_filter(task);
1036 spin_unlock_irq(&task->sighand->siglock);
1041 #ifdef CONFIG_SYSCTL
1043 /* Human readable action names for friendly sysctl interaction */
1044 #define SECCOMP_RET_KILL_THREAD_NAME "kill_thread"
1045 #define SECCOMP_RET_TRAP_NAME "trap"
1046 #define SECCOMP_RET_ERRNO_NAME "errno"
1047 #define SECCOMP_RET_TRACE_NAME "trace"
1048 #define SECCOMP_RET_LOG_NAME "log"
1049 #define SECCOMP_RET_ALLOW_NAME "allow"
1051 static const char seccomp_actions_avail[] =
1052 SECCOMP_RET_KILL_THREAD_NAME " "
1053 SECCOMP_RET_TRAP_NAME " "
1054 SECCOMP_RET_ERRNO_NAME " "
1055 SECCOMP_RET_TRACE_NAME " "
1056 SECCOMP_RET_LOG_NAME " "
1057 SECCOMP_RET_ALLOW_NAME;
1059 struct seccomp_log_name {
1064 static const struct seccomp_log_name seccomp_log_names[] = {
1065 { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1066 { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1067 { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1068 { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1069 { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1070 { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1074 static bool seccomp_names_from_actions_logged(char *names, size_t size,
1077 const struct seccomp_log_name *cur;
1078 bool append_space = false;
1080 for (cur = seccomp_log_names; cur->name && size; cur++) {
1083 if (!(actions_logged & cur->log))
1087 ret = strscpy(names, " ", size);
1094 append_space = true;
1096 ret = strscpy(names, cur->name, size);
1107 static bool seccomp_action_logged_from_name(u32 *action_logged,
1110 const struct seccomp_log_name *cur;
1112 for (cur = seccomp_log_names; cur->name; cur++) {
1113 if (!strcmp(cur->name, name)) {
1114 *action_logged = cur->log;
1122 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1126 *actions_logged = 0;
1127 while ((name = strsep(&names, " ")) && *name) {
1128 u32 action_logged = 0;
1130 if (!seccomp_action_logged_from_name(&action_logged, name))
1133 *actions_logged |= action_logged;
1139 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
1140 void __user *buffer, size_t *lenp,
1143 char names[sizeof(seccomp_actions_avail)];
1144 struct ctl_table table;
1147 if (write && !capable(CAP_SYS_ADMIN))
1150 memset(names, 0, sizeof(names));
1153 if (!seccomp_names_from_actions_logged(names, sizeof(names),
1154 seccomp_actions_logged))
1160 table.maxlen = sizeof(names);
1161 ret = proc_dostring(&table, write, buffer, lenp, ppos);
1168 if (!seccomp_actions_logged_from_names(&actions_logged,
1172 if (actions_logged & SECCOMP_LOG_ALLOW)
1175 seccomp_actions_logged = actions_logged;
1181 static struct ctl_path seccomp_sysctl_path[] = {
1182 { .procname = "kernel", },
1183 { .procname = "seccomp", },
1187 static struct ctl_table seccomp_sysctl_table[] = {
1189 .procname = "actions_avail",
1190 .data = (void *) &seccomp_actions_avail,
1191 .maxlen = sizeof(seccomp_actions_avail),
1193 .proc_handler = proc_dostring,
1196 .procname = "actions_logged",
1198 .proc_handler = seccomp_actions_logged_handler,
1203 static int __init seccomp_sysctl_init(void)
1205 struct ctl_table_header *hdr;
1207 hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
1209 pr_warn("seccomp: sysctl registration failed\n");
1211 kmemleak_not_leak(hdr);
1216 device_initcall(seccomp_sysctl_init)
1218 #endif /* CONFIG_SYSCTL */