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/atomic.h>
17 #include <linux/audit.h>
18 #include <linux/compat.h>
19 #include <linux/sched.h>
20 #include <linux/seccomp.h>
21 #include <linux/syscalls.h>
23 /* #define SECCOMP_DEBUG 1 */
25 #ifdef CONFIG_SECCOMP_FILTER
26 #include <asm/syscall.h>
27 #include <linux/filter.h>
28 #include <linux/ptrace.h>
29 #include <linux/security.h>
30 #include <linux/slab.h>
31 #include <linux/tracehook.h>
32 #include <linux/uaccess.h>
35 * struct seccomp_filter - container for seccomp BPF programs
37 * @usage: reference count to manage the object lifetime.
38 * get/put helpers should be used when accessing an instance
39 * outside of a lifetime-guarded section. In general, this
40 * is only needed for handling filters shared across tasks.
41 * @prev: points to a previously installed, or inherited, filter
42 * @len: the number of instructions in the program
43 * @insnsi: the BPF program instructions to evaluate
45 * seccomp_filter objects are organized in a tree linked via the @prev
46 * pointer. For any task, it appears to be a singly-linked list starting
47 * with current->seccomp.filter, the most recently attached or inherited filter.
48 * However, multiple filters may share a @prev node, by way of fork(), which
49 * results in a unidirectional tree existing in memory. This is similar to
50 * how namespaces work.
52 * seccomp_filter objects should never be modified after being attached
53 * to a task_struct (other than @usage).
55 struct seccomp_filter {
57 struct seccomp_filter *prev;
58 struct sk_filter *prog;
61 /* Limit any path through the tree to 256KB worth of instructions. */
62 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
65 * Endianness is explicitly ignored and left for BPF program authors to manage
66 * as per the specific architecture.
68 static void populate_seccomp_data(struct seccomp_data *sd)
70 struct task_struct *task = current;
71 struct pt_regs *regs = task_pt_regs(task);
72 unsigned long args[6];
74 sd->nr = syscall_get_nr(task, regs);
75 sd->arch = syscall_get_arch();
76 syscall_get_arguments(task, regs, 0, 6, args);
77 sd->args[0] = args[0];
78 sd->args[1] = args[1];
79 sd->args[2] = args[2];
80 sd->args[3] = args[3];
81 sd->args[4] = args[4];
82 sd->args[5] = args[5];
83 sd->instruction_pointer = KSTK_EIP(task);
87 * seccomp_check_filter - verify seccomp filter code
88 * @filter: filter to verify
89 * @flen: length of filter
91 * Takes a previously checked filter (by sk_chk_filter) and
92 * redirects all filter code that loads struct sk_buff data
93 * and related data through seccomp_bpf_load. It also
94 * enforces length and alignment checking of those loads.
96 * Returns 0 if the rule set is legal or -EINVAL if not.
98 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
101 for (pc = 0; pc < flen; pc++) {
102 struct sock_filter *ftest = &filter[pc];
103 u16 code = ftest->code;
107 case BPF_LD | BPF_W | BPF_ABS:
108 ftest->code = BPF_LDX | BPF_W | BPF_ABS;
109 /* 32-bit aligned and not out of bounds. */
110 if (k >= sizeof(struct seccomp_data) || k & 3)
113 case BPF_LD | BPF_W | BPF_LEN:
114 ftest->code = BPF_LD | BPF_IMM;
115 ftest->k = sizeof(struct seccomp_data);
117 case BPF_LDX | BPF_W | BPF_LEN:
118 ftest->code = BPF_LDX | BPF_IMM;
119 ftest->k = sizeof(struct seccomp_data);
121 /* Explicitly include allowed calls. */
122 case BPF_RET | BPF_K:
123 case BPF_RET | BPF_A:
124 case BPF_ALU | BPF_ADD | BPF_K:
125 case BPF_ALU | BPF_ADD | BPF_X:
126 case BPF_ALU | BPF_SUB | BPF_K:
127 case BPF_ALU | BPF_SUB | BPF_X:
128 case BPF_ALU | BPF_MUL | BPF_K:
129 case BPF_ALU | BPF_MUL | BPF_X:
130 case BPF_ALU | BPF_DIV | BPF_K:
131 case BPF_ALU | BPF_DIV | BPF_X:
132 case BPF_ALU | BPF_AND | BPF_K:
133 case BPF_ALU | BPF_AND | BPF_X:
134 case BPF_ALU | BPF_OR | BPF_K:
135 case BPF_ALU | BPF_OR | BPF_X:
136 case BPF_ALU | BPF_XOR | BPF_K:
137 case BPF_ALU | BPF_XOR | BPF_X:
138 case BPF_ALU | BPF_LSH | BPF_K:
139 case BPF_ALU | BPF_LSH | BPF_X:
140 case BPF_ALU | BPF_RSH | BPF_K:
141 case BPF_ALU | BPF_RSH | BPF_X:
142 case BPF_ALU | BPF_NEG:
143 case BPF_LD | BPF_IMM:
144 case BPF_LDX | BPF_IMM:
145 case BPF_MISC | BPF_TAX:
146 case BPF_MISC | BPF_TXA:
147 case BPF_LD | BPF_MEM:
148 case BPF_LDX | BPF_MEM:
151 case BPF_JMP | BPF_JA:
152 case BPF_JMP | BPF_JEQ | BPF_K:
153 case BPF_JMP | BPF_JEQ | BPF_X:
154 case BPF_JMP | BPF_JGE | BPF_K:
155 case BPF_JMP | BPF_JGE | BPF_X:
156 case BPF_JMP | BPF_JGT | BPF_K:
157 case BPF_JMP | BPF_JGT | BPF_X:
158 case BPF_JMP | BPF_JSET | BPF_K:
159 case BPF_JMP | BPF_JSET | BPF_X:
169 * seccomp_run_filters - evaluates all seccomp filters against @syscall
170 * @syscall: number of the current system call
172 * Returns valid seccomp BPF response codes.
174 static u32 seccomp_run_filters(int syscall)
176 struct seccomp_filter *f;
177 struct seccomp_data sd;
178 u32 ret = SECCOMP_RET_ALLOW;
180 /* Ensure unexpected behavior doesn't result in failing open. */
181 if (WARN_ON(current->seccomp.filter == NULL))
182 return SECCOMP_RET_KILL;
184 populate_seccomp_data(&sd);
187 * All filters in the list are evaluated and the lowest BPF return
188 * value always takes priority (ignoring the DATA).
190 for (f = current->seccomp.filter; f; f = f->prev) {
191 u32 cur_ret = SK_RUN_FILTER(f->prog, (void *)&sd);
193 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
198 #endif /* CONFIG_SECCOMP_FILTER */
200 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
202 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
208 static inline void seccomp_assign_mode(unsigned long seccomp_mode)
210 current->seccomp.mode = seccomp_mode;
211 set_tsk_thread_flag(current, TIF_SECCOMP);
214 #ifdef CONFIG_SECCOMP_FILTER
216 * seccomp_attach_filter: Attaches a seccomp filter to current.
217 * @fprog: BPF program to install
219 * Returns 0 on success or an errno on failure.
221 static long seccomp_attach_filter(struct sock_fprog *fprog)
223 struct seccomp_filter *filter;
224 unsigned long fp_size = fprog->len * sizeof(struct sock_filter);
225 unsigned long total_insns = fprog->len;
226 struct sock_filter *fp;
230 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
233 for (filter = current->seccomp.filter; filter; filter = filter->prev)
234 total_insns += filter->prog->len + 4; /* include a 4 instr penalty */
235 if (total_insns > MAX_INSNS_PER_PATH)
239 * Installing a seccomp filter requires that the task has
240 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
241 * This avoids scenarios where unprivileged tasks can affect the
242 * behavior of privileged children.
244 if (!task_no_new_privs(current) &&
245 security_capable_noaudit(current_cred(), current_user_ns(),
249 fp = kzalloc(fp_size, GFP_KERNEL|__GFP_NOWARN);
253 /* Copy the instructions from fprog. */
255 if (copy_from_user(fp, fprog->filter, fp_size))
258 /* Check and rewrite the fprog via the skb checker */
259 ret = sk_chk_filter(fp, fprog->len);
263 /* Check and rewrite the fprog for seccomp use */
264 ret = seccomp_check_filter(fp, fprog->len);
268 /* Convert 'sock_filter' insns to 'sock_filter_int' insns */
269 ret = sk_convert_filter(fp, fprog->len, NULL, &new_len);
273 /* Allocate a new seccomp_filter */
275 filter = kzalloc(sizeof(struct seccomp_filter),
276 GFP_KERNEL|__GFP_NOWARN);
280 filter->prog = kzalloc(sk_filter_size(new_len),
281 GFP_KERNEL|__GFP_NOWARN);
285 ret = sk_convert_filter(fp, fprog->len, filter->prog->insnsi, &new_len);
287 goto free_filter_prog;
290 atomic_set(&filter->usage, 1);
291 filter->prog->len = new_len;
293 sk_filter_select_runtime(filter->prog);
296 * If there is an existing filter, make it the prev and don't drop its
299 filter->prev = current->seccomp.filter;
300 current->seccomp.filter = filter;
313 * seccomp_attach_user_filter - attaches a user-supplied sock_fprog
314 * @user_filter: pointer to the user data containing a sock_fprog.
316 * Returns 0 on success and non-zero otherwise.
318 static long seccomp_attach_user_filter(const char __user *user_filter)
320 struct sock_fprog fprog;
324 if (is_compat_task()) {
325 struct compat_sock_fprog fprog32;
326 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
328 fprog.len = fprog32.len;
329 fprog.filter = compat_ptr(fprog32.filter);
330 } else /* falls through to the if below. */
332 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
334 ret = seccomp_attach_filter(&fprog);
339 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
340 void get_seccomp_filter(struct task_struct *tsk)
342 struct seccomp_filter *orig = tsk->seccomp.filter;
345 /* Reference count is bounded by the number of total processes. */
346 atomic_inc(&orig->usage);
349 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
350 void put_seccomp_filter(struct task_struct *tsk)
352 struct seccomp_filter *orig = tsk->seccomp.filter;
353 /* Clean up single-reference branches iteratively. */
354 while (orig && atomic_dec_and_test(&orig->usage)) {
355 struct seccomp_filter *freeme = orig;
357 sk_filter_free(freeme->prog);
363 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
364 * @syscall: syscall number to send to userland
365 * @reason: filter-supplied reason code to send to userland (via si_errno)
367 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
369 static void seccomp_send_sigsys(int syscall, int reason)
372 memset(&info, 0, sizeof(info));
373 info.si_signo = SIGSYS;
374 info.si_code = SYS_SECCOMP;
375 info.si_call_addr = (void __user *)KSTK_EIP(current);
376 info.si_errno = reason;
377 info.si_arch = syscall_get_arch();
378 info.si_syscall = syscall;
379 force_sig_info(SIGSYS, &info, current);
381 #endif /* CONFIG_SECCOMP_FILTER */
384 * Secure computing mode 1 allows only read/write/exit/sigreturn.
385 * To be fully secure this must be combined with rlimit
386 * to limit the stack allocations too.
388 static int mode1_syscalls[] = {
389 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
390 0, /* null terminated */
394 static int mode1_syscalls_32[] = {
395 __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
396 0, /* null terminated */
400 int __secure_computing(int this_syscall)
402 int mode = current->seccomp.mode;
408 case SECCOMP_MODE_STRICT:
409 syscall = mode1_syscalls;
411 if (is_compat_task())
412 syscall = mode1_syscalls_32;
415 if (*syscall == this_syscall)
417 } while (*++syscall);
419 ret = SECCOMP_RET_KILL;
421 #ifdef CONFIG_SECCOMP_FILTER
422 case SECCOMP_MODE_FILTER: {
424 struct pt_regs *regs = task_pt_regs(current);
425 ret = seccomp_run_filters(this_syscall);
426 data = ret & SECCOMP_RET_DATA;
427 ret &= SECCOMP_RET_ACTION;
429 case SECCOMP_RET_ERRNO:
430 /* Set the low-order 16-bits as a errno. */
431 syscall_set_return_value(current, regs,
434 case SECCOMP_RET_TRAP:
435 /* Show the handler the original registers. */
436 syscall_rollback(current, regs);
437 /* Let the filter pass back 16 bits of data. */
438 seccomp_send_sigsys(this_syscall, data);
440 case SECCOMP_RET_TRACE:
441 /* Skip these calls if there is no tracer. */
442 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
443 syscall_set_return_value(current, regs,
447 /* Allow the BPF to provide the event message */
448 ptrace_event(PTRACE_EVENT_SECCOMP, data);
450 * The delivery of a fatal signal during event
451 * notification may silently skip tracer notification.
452 * Terminating the task now avoids executing a system
453 * call that may not be intended.
455 if (fatal_signal_pending(current))
457 if (syscall_get_nr(current, regs) < 0)
458 goto skip; /* Explicit request to skip. */
461 case SECCOMP_RET_ALLOW:
463 case SECCOMP_RET_KILL:
478 audit_seccomp(this_syscall, exit_sig, ret);
480 #ifdef CONFIG_SECCOMP_FILTER
482 audit_seccomp(this_syscall, exit_sig, ret);
487 long prctl_get_seccomp(void)
489 return current->seccomp.mode;
493 * seccomp_set_mode_strict: internal function for setting strict seccomp
495 * Once current->seccomp.mode is non-zero, it may not be changed.
497 * Returns 0 on success or -EINVAL on failure.
499 static long seccomp_set_mode_strict(void)
501 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
504 if (!seccomp_may_assign_mode(seccomp_mode))
510 seccomp_assign_mode(seccomp_mode);
518 #ifdef CONFIG_SECCOMP_FILTER
520 * seccomp_set_mode_filter: internal function for setting seccomp filter
521 * @flags: flags to change filter behavior
522 * @filter: struct sock_fprog containing filter
524 * This function may be called repeatedly to install additional filters.
525 * Every filter successfully installed will be evaluated (in reverse order)
526 * for each system call the task makes.
528 * Once current->seccomp.mode is non-zero, it may not be changed.
530 * Returns 0 on success or -EINVAL on failure.
532 static long seccomp_set_mode_filter(unsigned int flags,
533 const char __user *filter)
535 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
538 /* Validate flags. */
542 if (!seccomp_may_assign_mode(seccomp_mode))
545 ret = seccomp_attach_user_filter(filter);
549 seccomp_assign_mode(seccomp_mode);
554 static inline long seccomp_set_mode_filter(unsigned int flags,
555 const char __user *filter)
561 /* Common entry point for both prctl and syscall. */
562 static long do_seccomp(unsigned int op, unsigned int flags,
563 const char __user *uargs)
566 case SECCOMP_SET_MODE_STRICT:
567 if (flags != 0 || uargs != NULL)
569 return seccomp_set_mode_strict();
570 case SECCOMP_SET_MODE_FILTER:
571 return seccomp_set_mode_filter(flags, uargs);
577 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
578 const char __user *, uargs)
580 return do_seccomp(op, flags, uargs);
584 * prctl_set_seccomp: configures current->seccomp.mode
585 * @seccomp_mode: requested mode to use
586 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
588 * Returns 0 on success or -EINVAL on failure.
590 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
595 switch (seccomp_mode) {
596 case SECCOMP_MODE_STRICT:
597 op = SECCOMP_SET_MODE_STRICT;
599 * Setting strict mode through prctl always ignored filter,
600 * so make sure it is always NULL here to pass the internal
601 * check in do_seccomp().
605 case SECCOMP_MODE_FILTER:
606 op = SECCOMP_SET_MODE_FILTER;
613 /* prctl interface doesn't have flags, so they are always zero. */
614 return do_seccomp(op, 0, uargs);