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 * @insns: 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 unsigned short len; /* Instruction count */
59 struct sock_filter insns[];
62 /* Limit any path through the tree to 256KB worth of instructions. */
63 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
66 * get_u32 - returns a u32 offset into data
67 * @data: a unsigned 64 bit value
68 * @index: 0 or 1 to return the first or second 32-bits
70 * This inline exists to hide the length of unsigned long. If a 32-bit
71 * unsigned long is passed in, it will be extended and the top 32-bits will be
72 * 0. If it is a 64-bit unsigned long, then whatever data is resident will be
75 * Endianness is explicitly ignored and left for BPF program authors to manage
76 * as per the specific architecture.
78 static inline u32 get_u32(u64 data, int index)
80 return ((u32 *)&data)[index];
83 /* Helper for bpf_load below. */
84 #define BPF_DATA(_name) offsetof(struct seccomp_data, _name)
86 * bpf_load: checks and returns a pointer to the requested offset
87 * @off: offset into struct seccomp_data to load from
89 * Returns the requested 32-bits of data.
90 * seccomp_check_filter() should assure that @off is 32-bit aligned
91 * and not out of bounds. Failure to do so is a BUG.
93 u32 seccomp_bpf_load(int off)
95 struct pt_regs *regs = task_pt_regs(current);
96 if (off == BPF_DATA(nr))
97 return syscall_get_nr(current, regs);
98 if (off == BPF_DATA(arch))
99 return syscall_get_arch();
100 if (off >= BPF_DATA(args[0]) && off < BPF_DATA(args[6])) {
102 int arg = (off - BPF_DATA(args[0])) / sizeof(u64);
103 int index = !!(off % sizeof(u64));
104 syscall_get_arguments(current, regs, arg, 1, &value);
105 return get_u32(value, index);
107 if (off == BPF_DATA(instruction_pointer))
108 return get_u32(KSTK_EIP(current), 0);
109 if (off == BPF_DATA(instruction_pointer) + sizeof(u32))
110 return get_u32(KSTK_EIP(current), 1);
111 /* seccomp_check_filter should make this impossible. */
116 * seccomp_check_filter - verify seccomp filter code
117 * @filter: filter to verify
118 * @flen: length of filter
120 * Takes a previously checked filter (by sk_chk_filter) and
121 * redirects all filter code that loads struct sk_buff data
122 * and related data through seccomp_bpf_load. It also
123 * enforces length and alignment checking of those loads.
125 * Returns 0 if the rule set is legal or -EINVAL if not.
127 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
130 for (pc = 0; pc < flen; pc++) {
131 struct sock_filter *ftest = &filter[pc];
132 u16 code = ftest->code;
137 ftest->code = BPF_S_ANC_SECCOMP_LD_W;
138 /* 32-bit aligned and not out of bounds. */
139 if (k >= sizeof(struct seccomp_data) || k & 3)
143 ftest->code = BPF_S_LD_IMM;
144 ftest->k = sizeof(struct seccomp_data);
146 case BPF_S_LDX_W_LEN:
147 ftest->code = BPF_S_LDX_IMM;
148 ftest->k = sizeof(struct seccomp_data);
150 /* Explicitly include allowed calls. */
153 case BPF_S_ALU_ADD_K:
154 case BPF_S_ALU_ADD_X:
155 case BPF_S_ALU_SUB_K:
156 case BPF_S_ALU_SUB_X:
157 case BPF_S_ALU_MUL_K:
158 case BPF_S_ALU_MUL_X:
159 case BPF_S_ALU_DIV_X:
160 case BPF_S_ALU_AND_K:
161 case BPF_S_ALU_AND_X:
164 case BPF_S_ALU_XOR_K:
165 case BPF_S_ALU_XOR_X:
166 case BPF_S_ALU_LSH_K:
167 case BPF_S_ALU_LSH_X:
168 case BPF_S_ALU_RSH_K:
169 case BPF_S_ALU_RSH_X:
175 case BPF_S_ALU_DIV_K:
181 case BPF_S_JMP_JEQ_K:
182 case BPF_S_JMP_JEQ_X:
183 case BPF_S_JMP_JGE_K:
184 case BPF_S_JMP_JGE_X:
185 case BPF_S_JMP_JGT_K:
186 case BPF_S_JMP_JGT_X:
187 case BPF_S_JMP_JSET_K:
188 case BPF_S_JMP_JSET_X:
198 * seccomp_run_filters - evaluates all seccomp filters against @syscall
199 * @syscall: number of the current system call
201 * Returns valid seccomp BPF response codes.
203 static u32 seccomp_run_filters(int syscall)
205 struct seccomp_filter *f;
206 u32 ret = SECCOMP_RET_ALLOW;
208 /* Ensure unexpected behavior doesn't result in failing open. */
209 if (WARN_ON(current->seccomp.filter == NULL))
210 return SECCOMP_RET_KILL;
213 * All filters in the list are evaluated and the lowest BPF return
214 * value always takes priority (ignoring the DATA).
216 for (f = current->seccomp.filter; f; f = f->prev) {
217 u32 cur_ret = sk_run_filter(NULL, f->insns);
218 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
223 #endif /* CONFIG_SECCOMP_FILTER */
225 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
227 if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
233 static inline void seccomp_assign_mode(unsigned long seccomp_mode)
235 current->seccomp.mode = seccomp_mode;
236 set_tsk_thread_flag(current, TIF_SECCOMP);
239 #ifdef CONFIG_SECCOMP_FILTER
241 * seccomp_attach_filter: Attaches a seccomp filter to current.
242 * @fprog: BPF program to install
244 * Returns 0 on success or an errno on failure.
246 static long seccomp_attach_filter(struct sock_fprog *fprog)
248 struct seccomp_filter *filter;
249 unsigned long fp_size = fprog->len * sizeof(struct sock_filter);
250 unsigned long total_insns = fprog->len;
253 if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
256 for (filter = current->seccomp.filter; filter; filter = filter->prev)
257 total_insns += filter->len + 4; /* include a 4 instr penalty */
258 if (total_insns > MAX_INSNS_PER_PATH)
262 * Installing a seccomp filter requires that the task have
263 * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
264 * This avoids scenarios where unprivileged tasks can affect the
265 * behavior of privileged children.
267 if (!current->no_new_privs &&
268 security_capable_noaudit(current_cred(), current_user_ns(),
272 /* Allocate a new seccomp_filter */
273 filter = kzalloc(sizeof(struct seccomp_filter) + fp_size,
274 GFP_KERNEL|__GFP_NOWARN);
277 atomic_set(&filter->usage, 1);
278 filter->len = fprog->len;
280 /* Copy the instructions from fprog. */
282 if (copy_from_user(filter->insns, fprog->filter, fp_size))
285 /* Check and rewrite the fprog via the skb checker */
286 ret = sk_chk_filter(filter->insns, filter->len);
290 /* Check and rewrite the fprog for seccomp use */
291 ret = seccomp_check_filter(filter->insns, filter->len);
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;
308 * seccomp_attach_user_filter - attaches a user-supplied sock_fprog
309 * @user_filter: pointer to the user data containing a sock_fprog.
311 * Returns 0 on success and non-zero otherwise.
313 static long seccomp_attach_user_filter(const char __user *user_filter)
315 struct sock_fprog fprog;
319 if (is_compat_task()) {
320 struct compat_sock_fprog fprog32;
321 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
323 fprog.len = fprog32.len;
324 fprog.filter = compat_ptr(fprog32.filter);
325 } else /* falls through to the if below. */
327 if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
329 ret = seccomp_attach_filter(&fprog);
334 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
335 void get_seccomp_filter(struct task_struct *tsk)
337 struct seccomp_filter *orig = tsk->seccomp.filter;
340 /* Reference count is bounded by the number of total processes. */
341 atomic_inc(&orig->usage);
344 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
345 void put_seccomp_filter(struct task_struct *tsk)
347 struct seccomp_filter *orig = tsk->seccomp.filter;
348 /* Clean up single-reference branches iteratively. */
349 while (orig && atomic_dec_and_test(&orig->usage)) {
350 struct seccomp_filter *freeme = orig;
357 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
358 * @syscall: syscall number to send to userland
359 * @reason: filter-supplied reason code to send to userland (via si_errno)
361 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
363 static void seccomp_send_sigsys(int syscall, int reason)
366 memset(&info, 0, sizeof(info));
367 info.si_signo = SIGSYS;
368 info.si_code = SYS_SECCOMP;
369 info.si_call_addr = (void __user *)KSTK_EIP(current);
370 info.si_errno = reason;
371 info.si_arch = syscall_get_arch();
372 info.si_syscall = syscall;
373 force_sig_info(SIGSYS, &info, current);
375 #endif /* CONFIG_SECCOMP_FILTER */
378 * Secure computing mode 1 allows only read/write/exit/sigreturn.
379 * To be fully secure this must be combined with rlimit
380 * to limit the stack allocations too.
382 static int mode1_syscalls[] = {
383 __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
384 0, /* null terminated */
388 static int mode1_syscalls_32[] = {
389 __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
390 0, /* null terminated */
394 int __secure_computing(int this_syscall)
396 int mode = current->seccomp.mode;
402 case SECCOMP_MODE_STRICT:
403 syscall = mode1_syscalls;
405 if (is_compat_task())
406 syscall = mode1_syscalls_32;
409 if (*syscall == this_syscall)
411 } while (*++syscall);
413 ret = SECCOMP_RET_KILL;
415 #ifdef CONFIG_SECCOMP_FILTER
416 case SECCOMP_MODE_FILTER: {
418 struct pt_regs *regs = task_pt_regs(current);
419 ret = seccomp_run_filters(this_syscall);
420 data = ret & SECCOMP_RET_DATA;
421 ret &= SECCOMP_RET_ACTION;
423 case SECCOMP_RET_ERRNO:
424 /* Set the low-order 16-bits as a errno. */
425 syscall_set_return_value(current, regs,
428 case SECCOMP_RET_TRAP:
429 /* Show the handler the original registers. */
430 syscall_rollback(current, regs);
431 /* Let the filter pass back 16 bits of data. */
432 seccomp_send_sigsys(this_syscall, data);
434 case SECCOMP_RET_TRACE:
435 /* Skip these calls if there is no tracer. */
436 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
437 syscall_set_return_value(current, regs,
441 /* Allow the BPF to provide the event message */
442 ptrace_event(PTRACE_EVENT_SECCOMP, data);
444 * The delivery of a fatal signal during event
445 * notification may silently skip tracer notification.
446 * Terminating the task now avoids executing a system
447 * call that may not be intended.
449 if (fatal_signal_pending(current))
451 if (syscall_get_nr(current, regs) < 0)
452 goto skip; /* Explicit request to skip. */
455 case SECCOMP_RET_ALLOW:
457 case SECCOMP_RET_KILL:
472 audit_seccomp(this_syscall, exit_sig, ret);
474 #ifdef CONFIG_SECCOMP_FILTER
476 audit_seccomp(this_syscall, exit_sig, ret);
481 long prctl_get_seccomp(void)
483 return current->seccomp.mode;
487 * seccomp_set_mode_strict: internal function for setting strict seccomp
489 * Once current->seccomp.mode is non-zero, it may not be changed.
491 * Returns 0 on success or -EINVAL on failure.
493 static long seccomp_set_mode_strict(void)
495 const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
498 if (!seccomp_may_assign_mode(seccomp_mode))
504 seccomp_assign_mode(seccomp_mode);
512 #ifdef CONFIG_SECCOMP_FILTER
514 * seccomp_set_mode_filter: internal function for setting seccomp filter
515 * @flags: flags to change filter behavior
516 * @filter: struct sock_fprog containing filter
518 * This function may be called repeatedly to install additional filters.
519 * Every filter successfully installed will be evaluated (in reverse order)
520 * for each system call the task makes.
522 * Once current->seccomp.mode is non-zero, it may not be changed.
524 * Returns 0 on success or -EINVAL on failure.
526 static long seccomp_set_mode_filter(unsigned int flags,
527 const char __user *filter)
529 const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
532 /* Validate flags. */
536 if (!seccomp_may_assign_mode(seccomp_mode))
539 ret = seccomp_attach_user_filter(filter);
543 seccomp_assign_mode(seccomp_mode);
548 static inline long seccomp_set_mode_filter(unsigned int flags,
549 const char __user *filter)
555 /* Common entry point for both prctl and syscall. */
556 static long do_seccomp(unsigned int op, unsigned int flags,
557 const char __user *uargs)
560 case SECCOMP_SET_MODE_STRICT:
561 if (flags != 0 || uargs != NULL)
563 return seccomp_set_mode_strict();
564 case SECCOMP_SET_MODE_FILTER:
565 return seccomp_set_mode_filter(flags, uargs);
571 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
572 const char __user *, uargs)
574 return do_seccomp(op, flags, uargs);
578 * prctl_set_seccomp: configures current->seccomp.mode
579 * @seccomp_mode: requested mode to use
580 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
582 * Returns 0 on success or -EINVAL on failure.
584 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
589 switch (seccomp_mode) {
590 case SECCOMP_MODE_STRICT:
591 op = SECCOMP_SET_MODE_STRICT;
593 * Setting strict mode through prctl always ignored filter,
594 * so make sure it is always NULL here to pass the internal
595 * check in do_seccomp().
599 case SECCOMP_MODE_FILTER:
600 op = SECCOMP_SET_MODE_FILTER;
607 /* prctl interface doesn't have flags, so they are always zero. */
608 return do_seccomp(op, 0, uargs);