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 #include <linux/slab.h>
  22 #include <linux/syscalls.h>
  23
  24 /* #define SECCOMP_DEBUG 1 */
  25
  26 #ifdef CONFIG_SECCOMP_FILTER
  27 #include <asm/syscall.h>
  28 #include <linux/filter.h>
  29 #include <linux/ptrace.h>
  30 #include <linux/security.h>
  31 #include <linux/tracehook.h>
  32 #include <linux/uaccess.h>
  33
  34 /**
  35  * struct seccomp_filter - container for seccomp BPF programs
  36  *
  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
  44  *
  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.
  51  *
  52  * seccomp_filter objects should never be modified after being attached
  53  * to a task_struct (other than @usage).
  54  */
  55 struct seccomp_filter {
  56         atomic_t usage;
  57         struct seccomp_filter *prev;
  58         struct sk_filter *prog;
  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  */
  68 static 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  */
  98 static 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_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)
 111                                 return -EINVAL;
 112                         continue;
 113                 case BPF_LD | BPF_W | BPF_LEN:
 114                         ftest->code = BPF_LD | BPF_IMM;
 115                         ftest->k = sizeof(struct seccomp_data);
 116                         continue;
 117                 case BPF_LDX | BPF_W | BPF_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_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:
 149                 case BPF_ST:
 150                 case BPF_STX:
 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:
 160                         continue;
 161                 default:
 162                         return -EINVAL;
 163                 }
 164         }
 165         return 0;
 166 }
 167
 168 /**
 169  * seccomp_run_filters - evaluates all seccomp filters against @syscall
 170  * @syscall: number of the current system call
 171  *
 172  * Returns valid seccomp BPF response codes.
 173  */
 174 static u32 seccomp_run_filters(int syscall)
 175 {
 176         struct seccomp_filter *f;
 177         struct seccomp_data sd;
 178         u32 ret = SECCOMP_RET_ALLOW;
 179
 180         /* Ensure unexpected behavior doesn't result in failing open. */
 181         if (WARN_ON(current->seccomp.filter == NULL))
 182                 return SECCOMP_RET_KILL;
 183
 184         populate_seccomp_data(&sd);
 185
 186         /*
 187          * All filters in the list are evaluated and the lowest BPF return
 188          * value always takes priority (ignoring the DATA).
 189          */
 190         for (f = current->seccomp.filter; f; f = f->prev) {
 191                 u32 cur_ret = SK_RUN_FILTER(f->prog, (void *)&sd);
 192
 193                 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
 194                         ret = cur_ret;
 195         }
 196         return ret;
 197 }
 198 #endif /* CONFIG_SECCOMP_FILTER */
 199
 200 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
 201 {
 202         BUG_ON(!spin_is_locked(&current->sighand->siglock));
 203
 204         if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
 205                 return false;
 206
 207         return true;
 208 }
 209
 210 static inline void seccomp_assign_mode(unsigned long seccomp_mode)
 211 {
 212         BUG_ON(!spin_is_locked(&current->sighand->siglock));
 213
 214         current->seccomp.mode = seccomp_mode;
 215         set_tsk_thread_flag(current, TIF_SECCOMP);
 216 }
 217
 218 #ifdef CONFIG_SECCOMP_FILTER
 219 /**
 220  * seccomp_prepare_filter: Prepares a seccomp filter for use.
 221  * @fprog: BPF program to install
 222  *
 223  * Returns filter on success or an ERR_PTR on failure.
 224  */
 225 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
 226 {
 227         struct seccomp_filter *filter;
 228         unsigned long fp_size;
 229         struct sock_filter *fp;
 230         int new_len;
 231         long ret;
 232
 233         if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
 234                 return ERR_PTR(-EINVAL);
 235         BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
 236         fp_size = fprog->len * sizeof(struct sock_filter);
 237
 238         /*
 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.
 243          */
 244         if (!task_no_new_privs(current) &&
 245             security_capable_noaudit(current_cred(), current_user_ns(),
 246                                      CAP_SYS_ADMIN) != 0)
 247                 return ERR_PTR(-EACCES);
 248
 249         fp = kzalloc(fp_size, GFP_KERNEL|__GFP_NOWARN);
 250         if (!fp)
 251                 return ERR_PTR(-ENOMEM);
 252
 253         /* Copy the instructions from fprog. */
 254         ret = -EFAULT;
 255         if (copy_from_user(fp, fprog->filter, fp_size))
 256                 goto free_prog;
 257
 258         /* Check and rewrite the fprog via the skb checker */
 259         ret = sk_chk_filter(fp, fprog->len);
 260         if (ret)
 261                 goto free_prog;
 262
 263         /* Check and rewrite the fprog for seccomp use */
 264         ret = seccomp_check_filter(fp, fprog->len);
 265         if (ret)
 266                 goto free_prog;
 267
 268         /* Convert 'sock_filter' insns to 'sock_filter_int' insns */
 269         ret = sk_convert_filter(fp, fprog->len, NULL, &new_len);
 270         if (ret)
 271                 goto free_prog;
 272
 273         /* Allocate a new seccomp_filter */
 274         ret = -ENOMEM;
 275         filter = kzalloc(sizeof(struct seccomp_filter),
 276                          GFP_KERNEL|__GFP_NOWARN);
 277         if (!filter)
 278                 goto free_prog;
 279
 280         filter->prog = kzalloc(sk_filter_size(new_len),
 281                                GFP_KERNEL|__GFP_NOWARN);
 282         if (!filter->prog)
 283                 goto free_filter;
 284
 285         ret = sk_convert_filter(fp, fprog->len, filter->prog->insnsi, &new_len);
 286         if (ret)
 287                 goto free_filter_prog;
 288         kfree(fp);
 289
 290         atomic_set(&filter->usage, 1);
 291         filter->prog->len = new_len;
 292
 293         sk_filter_select_runtime(filter->prog);
 294
 295         return filter;
 296
 297 free_filter_prog:
 298         kfree(filter->prog);
 299 free_filter:
 300         kfree(filter);
 301 free_prog:
 302         kfree(fp);
 303         return ERR_PTR(ret);
 304 }
 305
 306 /**
 307  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
 308  * @user_filter: pointer to the user data containing a sock_fprog.
 309  *
 310  * Returns 0 on success and non-zero otherwise.
 311  */
 312 static struct seccomp_filter *
 313 seccomp_prepare_user_filter(const char __user *user_filter)
 314 {
 315         struct sock_fprog fprog;
 316         struct seccomp_filter *filter = ERR_PTR(-EFAULT);
 317
 318 #ifdef CONFIG_COMPAT
 319         if (is_compat_task()) {
 320                 struct compat_sock_fprog fprog32;
 321                 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
 322                         goto out;
 323                 fprog.len = fprog32.len;
 324                 fprog.filter = compat_ptr(fprog32.filter);
 325         } else /* falls through to the if below. */
 326 #endif
 327         if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
 328                 goto out;
 329         filter = seccomp_prepare_filter(&fprog);
 330 out:
 331         return filter;
 332 }
 333
 334 /**
 335  * seccomp_attach_filter: validate and attach filter
 336  * @flags:  flags to change filter behavior
 337  * @filter: seccomp filter to add to the current process
 338  *
 339  * Caller must be holding current->sighand->siglock lock.
 340  *
 341  * Returns 0 on success, -ve on error.
 342  */
 343 static long seccomp_attach_filter(unsigned int flags,
 344                                   struct seccomp_filter *filter)
 345 {
 346         unsigned long total_insns;
 347         struct seccomp_filter *walker;
 348
 349         BUG_ON(!spin_is_locked(&current->sighand->siglock));
 350
 351         /* Validate resulting filter length. */
 352         total_insns = filter->prog->len;
 353         for (walker = current->seccomp.filter; walker; walker = walker->prev)
 354                 total_insns += walker->prog->len + 4;  /* 4 instr penalty */
 355         if (total_insns > MAX_INSNS_PER_PATH)
 356                 return -ENOMEM;
 357
 358         /*
 359          * If there is an existing filter, make it the prev and don't drop its
 360          * task reference.
 361          */
 362         filter->prev = current->seccomp.filter;
 363         current->seccomp.filter = filter;
 364
 365         return 0;
 366 }
 367
 368 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
 369 void get_seccomp_filter(struct task_struct *tsk)
 370 {
 371         struct seccomp_filter *orig = tsk->seccomp.filter;
 372         if (!orig)
 373                 return;
 374         /* Reference count is bounded by the number of total processes. */
 375         atomic_inc(&orig->usage);
 376 }
 377
 378 static inline void seccomp_filter_free(struct seccomp_filter *filter)
 379 {
 380         if (filter) {
 381                 sk_filter_free(filter->prog);
 382                 kfree(filter);
 383         }
 384 }
 385
 386 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
 387 void put_seccomp_filter(struct task_struct *tsk)
 388 {
 389         struct seccomp_filter *orig = tsk->seccomp.filter;
 390         /* Clean up single-reference branches iteratively. */
 391         while (orig && atomic_dec_and_test(&orig->usage)) {
 392                 struct seccomp_filter *freeme = orig;
 393                 orig = orig->prev;
 394                 seccomp_filter_free(freeme);
 395         }
 396 }
 397
 398 /**
 399  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
 400  * @syscall: syscall number to send to userland
 401  * @reason: filter-supplied reason code to send to userland (via si_errno)
 402  *
 403  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
 404  */
 405 static void seccomp_send_sigsys(int syscall, int reason)
 406 {
 407         struct siginfo info;
 408         memset(&info, 0, sizeof(info));
 409         info.si_signo = SIGSYS;
 410         info.si_code = SYS_SECCOMP;
 411         info.si_call_addr = (void __user *)KSTK_EIP(current);
 412         info.si_errno = reason;
 413         info.si_arch = syscall_get_arch();
 414         info.si_syscall = syscall;
 415         force_sig_info(SIGSYS, &info, current);
 416 }
 417 #endif  /* CONFIG_SECCOMP_FILTER */
 418
 419 /*
 420  * Secure computing mode 1 allows only read/write/exit/sigreturn.
 421  * To be fully secure this must be combined with rlimit
 422  * to limit the stack allocations too.
 423  */
 424 static int mode1_syscalls[] = {
 425         __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
 426         0, /* null terminated */
 427 };
 428
 429 #ifdef CONFIG_COMPAT
 430 static int mode1_syscalls_32[] = {
 431         __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
 432         0, /* null terminated */
 433 };
 434 #endif
 435
 436 int __secure_computing(int this_syscall)
 437 {
 438         int mode = current->seccomp.mode;
 439         int exit_sig = 0;
 440         int *syscall;
 441         u32 ret;
 442
 443         switch (mode) {
 444         case SECCOMP_MODE_STRICT:
 445                 syscall = mode1_syscalls;
 446 #ifdef CONFIG_COMPAT
 447                 if (is_compat_task())
 448                         syscall = mode1_syscalls_32;
 449 #endif
 450                 do {
 451                         if (*syscall == this_syscall)
 452                                 return 0;
 453                 } while (*++syscall);
 454                 exit_sig = SIGKILL;
 455                 ret = SECCOMP_RET_KILL;
 456                 break;
 457 #ifdef CONFIG_SECCOMP_FILTER
 458         case SECCOMP_MODE_FILTER: {
 459                 int data;
 460                 struct pt_regs *regs = task_pt_regs(current);
 461                 ret = seccomp_run_filters(this_syscall);
 462                 data = ret & SECCOMP_RET_DATA;
 463                 ret &= SECCOMP_RET_ACTION;
 464                 switch (ret) {
 465                 case SECCOMP_RET_ERRNO:
 466                         /* Set the low-order 16-bits as a errno. */
 467                         syscall_set_return_value(current, regs,
 468                                                  -data, 0);
 469                         goto skip;
 470                 case SECCOMP_RET_TRAP:
 471                         /* Show the handler the original registers. */
 472                         syscall_rollback(current, regs);
 473                         /* Let the filter pass back 16 bits of data. */
 474                         seccomp_send_sigsys(this_syscall, data);
 475                         goto skip;
 476                 case SECCOMP_RET_TRACE:
 477                         /* Skip these calls if there is no tracer. */
 478                         if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
 479                                 syscall_set_return_value(current, regs,
 480                                                          -ENOSYS, 0);
 481                                 goto skip;
 482                         }
 483                         /* Allow the BPF to provide the event message */
 484                         ptrace_event(PTRACE_EVENT_SECCOMP, data);
 485                         /*
 486                          * The delivery of a fatal signal during event
 487                          * notification may silently skip tracer notification.
 488                          * Terminating the task now avoids executing a system
 489                          * call that may not be intended.
 490                          */
 491                         if (fatal_signal_pending(current))
 492                                 break;
 493                         if (syscall_get_nr(current, regs) < 0)
 494                                 goto skip;  /* Explicit request to skip. */
 495
 496                         return 0;
 497                 case SECCOMP_RET_ALLOW:
 498                         return 0;
 499                 case SECCOMP_RET_KILL:
 500                 default:
 501                         break;
 502                 }
 503                 exit_sig = SIGSYS;
 504                 break;
 505         }
 506 #endif
 507         default:
 508                 BUG();
 509         }
 510
 511 #ifdef SECCOMP_DEBUG
 512         dump_stack();
 513 #endif
 514         audit_seccomp(this_syscall, exit_sig, ret);
 515         do_exit(exit_sig);
 516 #ifdef CONFIG_SECCOMP_FILTER
 517 skip:
 518         audit_seccomp(this_syscall, exit_sig, ret);
 519 #endif
 520         return -1;
 521 }
 522
 523 long prctl_get_seccomp(void)
 524 {
 525         return current->seccomp.mode;
 526 }
 527
 528 /**
 529  * seccomp_set_mode_strict: internal function for setting strict seccomp
 530  *
 531  * Once current->seccomp.mode is non-zero, it may not be changed.
 532  *
 533  * Returns 0 on success or -EINVAL on failure.
 534  */
 535 static long seccomp_set_mode_strict(void)
 536 {
 537         const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
 538         long ret = -EINVAL;
 539
 540         spin_lock_irq(&current->sighand->siglock);
 541
 542         if (!seccomp_may_assign_mode(seccomp_mode))
 543                 goto out;
 544
 545 #ifdef TIF_NOTSC
 546         disable_TSC();
 547 #endif
 548         seccomp_assign_mode(seccomp_mode);
 549         ret = 0;
 550
 551 out:
 552         spin_unlock_irq(&current->sighand->siglock);
 553
 554         return ret;
 555 }
 556
 557 #ifdef CONFIG_SECCOMP_FILTER
 558 /**
 559  * seccomp_set_mode_filter: internal function for setting seccomp filter
 560  * @flags:  flags to change filter behavior
 561  * @filter: struct sock_fprog containing filter
 562  *
 563  * This function may be called repeatedly to install additional filters.
 564  * Every filter successfully installed will be evaluated (in reverse order)
 565  * for each system call the task makes.
 566  *
 567  * Once current->seccomp.mode is non-zero, it may not be changed.
 568  *
 569  * Returns 0 on success or -EINVAL on failure.
 570  */
 571 static long seccomp_set_mode_filter(unsigned int flags,
 572                                     const char __user *filter)
 573 {
 574         const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
 575         struct seccomp_filter *prepared = NULL;
 576         long ret = -EINVAL;
 577
 578         /* Validate flags. */
 579         if (flags != 0)
 580                 return -EINVAL;
 581
 582         /* Prepare the new filter before holding any locks. */
 583         prepared = seccomp_prepare_user_filter(filter);
 584         if (IS_ERR(prepared))
 585                 return PTR_ERR(prepared);
 586
 587         spin_lock_irq(&current->sighand->siglock);
 588
 589         if (!seccomp_may_assign_mode(seccomp_mode))
 590                 goto out;
 591
 592         ret = seccomp_attach_filter(flags, prepared);
 593         if (ret)
 594                 goto out;
 595         /* Do not free the successfully attached filter. */
 596         prepared = NULL;
 597
 598         seccomp_assign_mode(seccomp_mode);
 599 out:
 600         spin_unlock_irq(&current->sighand->siglock);
 601         seccomp_filter_free(prepared);
 602         return ret;
 603 }
 604 #else
 605 static inline long seccomp_set_mode_filter(unsigned int flags,
 606                                            const char __user *filter)
 607 {
 608         return -EINVAL;
 609 }
 610 #endif
 611
 612 /* Common entry point for both prctl and syscall. */
 613 static long do_seccomp(unsigned int op, unsigned int flags,
 614                        const char __user *uargs)
 615 {
 616         switch (op) {
 617         case SECCOMP_SET_MODE_STRICT:
 618                 if (flags != 0 || uargs != NULL)
 619                         return -EINVAL;
 620                 return seccomp_set_mode_strict();
 621         case SECCOMP_SET_MODE_FILTER:
 622                 return seccomp_set_mode_filter(flags, uargs);
 623         default:
 624                 return -EINVAL;
 625         }
 626 }
 627
 628 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
 629                          const char __user *, uargs)
 630 {
 631         return do_seccomp(op, flags, uargs);
 632 }
 633
 634 /**
 635  * prctl_set_seccomp: configures current->seccomp.mode
 636  * @seccomp_mode: requested mode to use
 637  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
 638  *
 639  * Returns 0 on success or -EINVAL on failure.
 640  */
 641 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
 642 {
 643         unsigned int op;
 644         char __user *uargs;
 645
 646         switch (seccomp_mode) {
 647         case SECCOMP_MODE_STRICT:
 648                 op = SECCOMP_SET_MODE_STRICT;
 649                 /*
 650                  * Setting strict mode through prctl always ignored filter,
 651                  * so make sure it is always NULL here to pass the internal
 652                  * check in do_seccomp().
 653                  */
 654                 uargs = NULL;
 655                 break;
 656         case SECCOMP_MODE_FILTER:
 657                 op = SECCOMP_SET_MODE_FILTER;
 658                 uargs = filter;
 659                 break;
 660         default:
 661                 return -EINVAL;
 662         }
 663
 664         /* prctl interface doesn't have flags, so they are always zero. */
 665         return do_seccomp(op, 0, uargs);
 666 }