Root/kernel/seccomp.c

1/*
2 * linux/kernel/seccomp.c
3 *
4 * Copyright 2004-2005 Andrea Arcangeli <andrea@cpushare.com>
5 *
6 * Copyright (C) 2012 Google, Inc.
7 * Will Drewry <wad@chromium.org>
8 *
9 * This defines a simple but solid secure-computing facility.
10 *
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.
14 */
15
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
22/* #define SECCOMP_DEBUG 1 */
23
24#ifdef CONFIG_SECCOMP_FILTER
25#include <asm/syscall.h>
26#include <linux/filter.h>
27#include <linux/ptrace.h>
28#include <linux/security.h>
29#include <linux/slab.h>
30#include <linux/tracehook.h>
31#include <linux/uaccess.h>
32
33/**
34 * struct seccomp_filter - container for seccomp BPF programs
35 *
36 * @usage: reference count to manage the object lifetime.
37 * get/put helpers should be used when accessing an instance
38 * outside of a lifetime-guarded section. In general, this
39 * is only needed for handling filters shared across tasks.
40 * @prev: points to a previously installed, or inherited, filter
41 * @len: the number of instructions in the program
42 * @insns: the BPF program instructions to evaluate
43 *
44 * seccomp_filter objects are organized in a tree linked via the @prev
45 * pointer. For any task, it appears to be a singly-linked list starting
46 * with current->seccomp.filter, the most recently attached or inherited filter.
47 * However, multiple filters may share a @prev node, by way of fork(), which
48 * results in a unidirectional tree existing in memory. This is similar to
49 * how namespaces work.
50 *
51 * seccomp_filter objects should never be modified after being attached
52 * to a task_struct (other than @usage).
53 */
54struct seccomp_filter {
55    atomic_t usage;
56    struct seccomp_filter *prev;
57    unsigned short len; /* Instruction count */
58    struct sock_filter_int insnsi[];
59};
60
61/* Limit any path through the tree to 256KB worth of instructions. */
62#define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
63
64/*
65 * Endianness is explicitly ignored and left for BPF program authors to manage
66 * as per the specific architecture.
67 */
68static void populate_seccomp_data(struct seccomp_data *sd)
69{
70    struct task_struct *task = current;
71    struct pt_regs *regs = task_pt_regs(task);
72    unsigned long args[6];
73
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);
84}
85
86/**
87 * seccomp_check_filter - verify seccomp filter code
88 * @filter: filter to verify
89 * @flen: length of filter
90 *
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.
95 *
96 * Returns 0 if the rule set is legal or -EINVAL if not.
97 */
98static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
99{
100    int pc;
101    for (pc = 0; pc < flen; pc++) {
102        struct sock_filter *ftest = &filter[pc];
103        u16 code = ftest->code;
104        u32 k = ftest->k;
105
106        switch (code) {
107        case BPF_S_LD_W_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)
111                return -EINVAL;
112            continue;
113        case BPF_S_LD_W_LEN:
114            ftest->code = BPF_LD | BPF_IMM;
115            ftest->k = sizeof(struct seccomp_data);
116            continue;
117        case BPF_S_LDX_W_LEN:
118            ftest->code = BPF_LDX | BPF_IMM;
119            ftest->k = sizeof(struct seccomp_data);
120            continue;
121        /* Explicitly include allowed calls. */
122        case BPF_S_RET_K:
123        case BPF_S_RET_A:
124        case BPF_S_ALU_ADD_K:
125        case BPF_S_ALU_ADD_X:
126        case BPF_S_ALU_SUB_K:
127        case BPF_S_ALU_SUB_X:
128        case BPF_S_ALU_MUL_K:
129        case BPF_S_ALU_MUL_X:
130        case BPF_S_ALU_DIV_X:
131        case BPF_S_ALU_AND_K:
132        case BPF_S_ALU_AND_X:
133        case BPF_S_ALU_OR_K:
134        case BPF_S_ALU_OR_X:
135        case BPF_S_ALU_XOR_K:
136        case BPF_S_ALU_XOR_X:
137        case BPF_S_ALU_LSH_K:
138        case BPF_S_ALU_LSH_X:
139        case BPF_S_ALU_RSH_K:
140        case BPF_S_ALU_RSH_X:
141        case BPF_S_ALU_NEG:
142        case BPF_S_LD_IMM:
143        case BPF_S_LDX_IMM:
144        case BPF_S_MISC_TAX:
145        case BPF_S_MISC_TXA:
146        case BPF_S_ALU_DIV_K:
147        case BPF_S_LD_MEM:
148        case BPF_S_LDX_MEM:
149        case BPF_S_ST:
150        case BPF_S_STX:
151        case BPF_S_JMP_JA:
152        case BPF_S_JMP_JEQ_K:
153        case BPF_S_JMP_JEQ_X:
154        case BPF_S_JMP_JGE_K:
155        case BPF_S_JMP_JGE_X:
156        case BPF_S_JMP_JGT_K:
157        case BPF_S_JMP_JGT_X:
158        case BPF_S_JMP_JSET_K:
159        case BPF_S_JMP_JSET_X:
160            sk_decode_filter(ftest, ftest);
161            continue;
162        default:
163            return -EINVAL;
164        }
165    }
166    return 0;
167}
168
169/**
170 * seccomp_run_filters - evaluates all seccomp filters against @syscall
171 * @syscall: number of the current system call
172 *
173 * Returns valid seccomp BPF response codes.
174 */
175static u32 seccomp_run_filters(int syscall)
176{
177    struct seccomp_filter *f;
178    struct seccomp_data sd;
179    u32 ret = SECCOMP_RET_ALLOW;
180
181    /* Ensure unexpected behavior doesn't result in failing open. */
182    if (WARN_ON(current->seccomp.filter == NULL))
183        return SECCOMP_RET_KILL;
184
185    populate_seccomp_data(&sd);
186
187    /*
188     * All filters in the list are evaluated and the lowest BPF return
189     * value always takes priority (ignoring the DATA).
190     */
191    for (f = current->seccomp.filter; f; f = f->prev) {
192        u32 cur_ret = sk_run_filter_int_seccomp(&sd, f->insnsi);
193        if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
194            ret = cur_ret;
195    }
196    return ret;
197}
198
199/**
200 * seccomp_attach_filter: Attaches a seccomp filter to current.
201 * @fprog: BPF program to install
202 *
203 * Returns 0 on success or an errno on failure.
204 */
205static long seccomp_attach_filter(struct sock_fprog *fprog)
206{
207    struct seccomp_filter *filter;
208    unsigned long fp_size = fprog->len * sizeof(struct sock_filter);
209    unsigned long total_insns = fprog->len;
210    struct sock_filter *fp;
211    int new_len;
212    long ret;
213
214    if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
215        return -EINVAL;
216
217    for (filter = current->seccomp.filter; filter; filter = filter->prev)
218        total_insns += filter->len + 4; /* include a 4 instr penalty */
219    if (total_insns > MAX_INSNS_PER_PATH)
220        return -ENOMEM;
221
222    /*
223     * Installing a seccomp filter requires that the task have
224     * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
225     * This avoids scenarios where unprivileged tasks can affect the
226     * behavior of privileged children.
227     */
228    if (!current->no_new_privs &&
229        security_capable_noaudit(current_cred(), current_user_ns(),
230                     CAP_SYS_ADMIN) != 0)
231        return -EACCES;
232
233    fp = kzalloc(fp_size, GFP_KERNEL|__GFP_NOWARN);
234    if (!fp)
235        return -ENOMEM;
236
237    /* Copy the instructions from fprog. */
238    ret = -EFAULT;
239    if (copy_from_user(fp, fprog->filter, fp_size))
240        goto free_prog;
241
242    /* Check and rewrite the fprog via the skb checker */
243    ret = sk_chk_filter(fp, fprog->len);
244    if (ret)
245        goto free_prog;
246
247    /* Check and rewrite the fprog for seccomp use */
248    ret = seccomp_check_filter(fp, fprog->len);
249    if (ret)
250        goto free_prog;
251
252    /* Convert 'sock_filter' insns to 'sock_filter_int' insns */
253    ret = sk_convert_filter(fp, fprog->len, NULL, &new_len);
254    if (ret)
255        goto free_prog;
256
257    /* Allocate a new seccomp_filter */
258    ret = -ENOMEM;
259    filter = kzalloc(sizeof(struct seccomp_filter) +
260             sizeof(struct sock_filter_int) * new_len,
261             GFP_KERNEL|__GFP_NOWARN);
262    if (!filter)
263        goto free_prog;
264
265    ret = sk_convert_filter(fp, fprog->len, filter->insnsi, &new_len);
266    if (ret)
267        goto free_filter;
268    kfree(fp);
269
270    atomic_set(&filter->usage, 1);
271    filter->len = new_len;
272
273    /*
274     * If there is an existing filter, make it the prev and don't drop its
275     * task reference.
276     */
277    filter->prev = current->seccomp.filter;
278    current->seccomp.filter = filter;
279    return 0;
280
281free_filter:
282    kfree(filter);
283free_prog:
284    kfree(fp);
285    return ret;
286}
287
288/**
289 * seccomp_attach_user_filter - attaches a user-supplied sock_fprog
290 * @user_filter: pointer to the user data containing a sock_fprog.
291 *
292 * Returns 0 on success and non-zero otherwise.
293 */
294static long seccomp_attach_user_filter(char __user *user_filter)
295{
296    struct sock_fprog fprog;
297    long ret = -EFAULT;
298
299#ifdef CONFIG_COMPAT
300    if (is_compat_task()) {
301        struct compat_sock_fprog fprog32;
302        if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
303            goto out;
304        fprog.len = fprog32.len;
305        fprog.filter = compat_ptr(fprog32.filter);
306    } else /* falls through to the if below. */
307#endif
308    if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
309        goto out;
310    ret = seccomp_attach_filter(&fprog);
311out:
312    return ret;
313}
314
315/* get_seccomp_filter - increments the reference count of the filter on @tsk */
316void get_seccomp_filter(struct task_struct *tsk)
317{
318    struct seccomp_filter *orig = tsk->seccomp.filter;
319    if (!orig)
320        return;
321    /* Reference count is bounded by the number of total processes. */
322    atomic_inc(&orig->usage);
323}
324
325/* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
326void put_seccomp_filter(struct task_struct *tsk)
327{
328    struct seccomp_filter *orig = tsk->seccomp.filter;
329    /* Clean up single-reference branches iteratively. */
330    while (orig && atomic_dec_and_test(&orig->usage)) {
331        struct seccomp_filter *freeme = orig;
332        orig = orig->prev;
333        kfree(freeme);
334    }
335}
336
337/**
338 * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
339 * @syscall: syscall number to send to userland
340 * @reason: filter-supplied reason code to send to userland (via si_errno)
341 *
342 * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
343 */
344static void seccomp_send_sigsys(int syscall, int reason)
345{
346    struct siginfo info;
347    memset(&info, 0, sizeof(info));
348    info.si_signo = SIGSYS;
349    info.si_code = SYS_SECCOMP;
350    info.si_call_addr = (void __user *)KSTK_EIP(current);
351    info.si_errno = reason;
352    info.si_arch = syscall_get_arch();
353    info.si_syscall = syscall;
354    force_sig_info(SIGSYS, &info, current);
355}
356#endif /* CONFIG_SECCOMP_FILTER */
357
358/*
359 * Secure computing mode 1 allows only read/write/exit/sigreturn.
360 * To be fully secure this must be combined with rlimit
361 * to limit the stack allocations too.
362 */
363static int mode1_syscalls[] = {
364    __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
365    0, /* null terminated */
366};
367
368#ifdef CONFIG_COMPAT
369static int mode1_syscalls_32[] = {
370    __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
371    0, /* null terminated */
372};
373#endif
374
375int __secure_computing(int this_syscall)
376{
377    int mode = current->seccomp.mode;
378    int exit_sig = 0;
379    int *syscall;
380    u32 ret;
381
382    switch (mode) {
383    case SECCOMP_MODE_STRICT:
384        syscall = mode1_syscalls;
385#ifdef CONFIG_COMPAT
386        if (is_compat_task())
387            syscall = mode1_syscalls_32;
388#endif
389        do {
390            if (*syscall == this_syscall)
391                return 0;
392        } while (*++syscall);
393        exit_sig = SIGKILL;
394        ret = SECCOMP_RET_KILL;
395        break;
396#ifdef CONFIG_SECCOMP_FILTER
397    case SECCOMP_MODE_FILTER: {
398        int data;
399        struct pt_regs *regs = task_pt_regs(current);
400        ret = seccomp_run_filters(this_syscall);
401        data = ret & SECCOMP_RET_DATA;
402        ret &= SECCOMP_RET_ACTION;
403        switch (ret) {
404        case SECCOMP_RET_ERRNO:
405            /* Set the low-order 16-bits as a errno. */
406            syscall_set_return_value(current, regs,
407                         -data, 0);
408            goto skip;
409        case SECCOMP_RET_TRAP:
410            /* Show the handler the original registers. */
411            syscall_rollback(current, regs);
412            /* Let the filter pass back 16 bits of data. */
413            seccomp_send_sigsys(this_syscall, data);
414            goto skip;
415        case SECCOMP_RET_TRACE:
416            /* Skip these calls if there is no tracer. */
417            if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
418                syscall_set_return_value(current, regs,
419                             -ENOSYS, 0);
420                goto skip;
421            }
422            /* Allow the BPF to provide the event message */
423            ptrace_event(PTRACE_EVENT_SECCOMP, data);
424            /*
425             * The delivery of a fatal signal during event
426             * notification may silently skip tracer notification.
427             * Terminating the task now avoids executing a system
428             * call that may not be intended.
429             */
430            if (fatal_signal_pending(current))
431                break;
432            if (syscall_get_nr(current, regs) < 0)
433                goto skip; /* Explicit request to skip. */
434
435            return 0;
436        case SECCOMP_RET_ALLOW:
437            return 0;
438        case SECCOMP_RET_KILL:
439        default:
440            break;
441        }
442        exit_sig = SIGSYS;
443        break;
444    }
445#endif
446    default:
447        BUG();
448    }
449
450#ifdef SECCOMP_DEBUG
451    dump_stack();
452#endif
453    audit_seccomp(this_syscall, exit_sig, ret);
454    do_exit(exit_sig);
455#ifdef CONFIG_SECCOMP_FILTER
456skip:
457    audit_seccomp(this_syscall, exit_sig, ret);
458#endif
459    return -1;
460}
461
462long prctl_get_seccomp(void)
463{
464    return current->seccomp.mode;
465}
466
467/**
468 * prctl_set_seccomp: configures current->seccomp.mode
469 * @seccomp_mode: requested mode to use
470 * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
471 *
472 * This function may be called repeatedly with a @seccomp_mode of
473 * SECCOMP_MODE_FILTER to install additional filters. Every filter
474 * successfully installed will be evaluated (in reverse order) for each system
475 * call the task makes.
476 *
477 * Once current->seccomp.mode is non-zero, it may not be changed.
478 *
479 * Returns 0 on success or -EINVAL on failure.
480 */
481long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
482{
483    long ret = -EINVAL;
484
485    if (current->seccomp.mode &&
486        current->seccomp.mode != seccomp_mode)
487        goto out;
488
489    switch (seccomp_mode) {
490    case SECCOMP_MODE_STRICT:
491        ret = 0;
492#ifdef TIF_NOTSC
493        disable_TSC();
494#endif
495        break;
496#ifdef CONFIG_SECCOMP_FILTER
497    case SECCOMP_MODE_FILTER:
498        ret = seccomp_attach_user_filter(filter);
499        if (ret)
500            goto out;
501        break;
502#endif
503    default:
504        goto out;
505    }
506
507    current->seccomp.mode = seccomp_mode;
508    set_thread_flag(TIF_SECCOMP);
509out:
510    return ret;
511}
512

Archive Download this file



interactive