granicus.if.org Git - strace/blob - syscall.c

   1 /*
   2  * Copyright (c) 1991, 1992 Paul Kranenburg <pk@cs.few.eur.nl>
   3  * Copyright (c) 1993 Branko Lankester <branko@hacktic.nl>
   4  * Copyright (c) 1993, 1994, 1995, 1996 Rick Sladkey <jrs@world.std.com>
   5  * Copyright (c) 1996-1999 Wichert Akkerman <wichert@cistron.nl>
   6  * Copyright (c) 1999 IBM Deutschland Entwicklung GmbH, IBM Corporation
   7  *                     Linux for s390 port by D.J. Barrow
   8  *                    <barrow_dj@mail.yahoo.com,djbarrow@de.ibm.com>
   9  * Copyright (c) 1999-2017 The strace developers.
  10  * All rights reserved.
  11  *
  12  * Redistribution and use in source and binary forms, with or without
  13  * modification, are permitted provided that the following conditions
  14  * are met:
  15  * 1. Redistributions of source code must retain the above copyright
  16  *    notice, this list of conditions and the following disclaimer.
  17  * 2. Redistributions in binary form must reproduce the above copyright
  18  *    notice, this list of conditions and the following disclaimer in the
  19  *    documentation and/or other materials provided with the distribution.
  20  * 3. The name of the author may not be used to endorse or promote products
  21  *    derived from this software without specific prior written permission.
  22  *
  23  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  24  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  25  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  26  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  27  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  28  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  29  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  30  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  31  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  32  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  33  */
  34
  35 #include "defs.h"
  36 #include "native_defs.h"
  37 #include "nsig.h"
  38 #include "number_set.h"
  39 #include <sys/param.h>
  40
  41 /* for struct iovec */
  42 #include <sys/uio.h>
  43
  44 /* for __X32_SYSCALL_BIT */
  45 #include <asm/unistd.h>
  46
  47 #include "regs.h"
  48 #include "ptrace.h"
  49
  50 #if defined(SPARC64)
  51 # undef PTRACE_GETREGS
  52 # define PTRACE_GETREGS PTRACE_GETREGS64
  53 # undef PTRACE_SETREGS
  54 # define PTRACE_SETREGS PTRACE_SETREGS64
  55 #endif
  56
  57 #ifndef NT_PRSTATUS
  58 # define NT_PRSTATUS 1
  59 #endif
  60
  61 #include "syscall.h"
  62 #include "xstring.h"
  63
  64 /* Define these shorthand notations to simplify the syscallent files. */
  65 #include "sysent_shorthand_defs.h"
  66
  67 #define SEN(syscall_name) SEN_ ## syscall_name, SYS_FUNC_NAME(sys_ ## syscall_name)
  68
  69 const struct_sysent sysent0[] = {
  70 #include "syscallent.h"
  71 };
  72
  73 #if SUPPORTED_PERSONALITIES > 1
  74 # include PERSONALITY1_INCLUDE_FUNCS
  75 static const struct_sysent sysent1[] = {
  76 # include "syscallent1.h"
  77 };
  78 #endif
  79
  80 #if SUPPORTED_PERSONALITIES > 2
  81 # include PERSONALITY2_INCLUDE_FUNCS
  82 static const struct_sysent sysent2[] = {
  83 # include "syscallent2.h"
  84 };
  85 #endif
  86
  87 /* Now undef them since short defines cause wicked namespace pollution. */
  88 #include "sysent_shorthand_undefs.h"
  89
  90 /*
  91  * `ioctlent[012].h' files are automatically generated by the auxiliary
  92  * program `ioctlsort', such that the list is sorted by the `code' field.
  93  * This has the side-effect of resolving the _IO.. macros into
  94  * plain integers, eliminating the need to include here everything
  95  * in "/usr/include".
  96  */
  97
  98 const char *const errnoent0[] = {
  99 #include "errnoent.h"
 100 };
 101 const char *const signalent0[] = {
 102 #include "signalent.h"
 103 };
 104 const struct_ioctlent ioctlent0[] = {
 105 #include "ioctlent0.h"
 106 };
 107
 108 #if SUPPORTED_PERSONALITIES > 1
 109 static const char *const errnoent1[] = {
 110 # include "errnoent1.h"
 111 };
 112 static const char *const signalent1[] = {
 113 # include "signalent1.h"
 114 };
 115 static const struct_ioctlent ioctlent1[] = {
 116 # include "ioctlent1.h"
 117 };
 118 # include PERSONALITY0_INCLUDE_PRINTERS_DECLS
 119 static const struct_printers printers0 = {
 120 # include PERSONALITY0_INCLUDE_PRINTERS_DEFS
 121 };
 122 # include PERSONALITY1_INCLUDE_PRINTERS_DECLS
 123 static const struct_printers printers1 = {
 124 # include PERSONALITY1_INCLUDE_PRINTERS_DEFS
 125 };
 126 #endif
 127
 128 #if SUPPORTED_PERSONALITIES > 2
 129 static const char *const errnoent2[] = {
 130 # include "errnoent2.h"
 131 };
 132 static const char *const signalent2[] = {
 133 # include "signalent2.h"
 134 };
 135 static const struct_ioctlent ioctlent2[] = {
 136 # include "ioctlent2.h"
 137 };
 138 # include PERSONALITY2_INCLUDE_PRINTERS_DECLS
 139 static const struct_printers printers2 = {
 140 # include PERSONALITY2_INCLUDE_PRINTERS_DEFS
 141 };
 142 #endif
 143
 144 enum {
 145         nsyscalls0 = ARRAY_SIZE(sysent0)
 146 #if SUPPORTED_PERSONALITIES > 1
 147         , nsyscalls1 = ARRAY_SIZE(sysent1)
 148 # if SUPPORTED_PERSONALITIES > 2
 149         , nsyscalls2 = ARRAY_SIZE(sysent2)
 150 # endif
 151 #endif
 152 };
 153
 154 enum {
 155         nerrnos0 = ARRAY_SIZE(errnoent0)
 156 #if SUPPORTED_PERSONALITIES > 1
 157         , nerrnos1 = ARRAY_SIZE(errnoent1)
 158 # if SUPPORTED_PERSONALITIES > 2
 159         , nerrnos2 = ARRAY_SIZE(errnoent2)
 160 # endif
 161 #endif
 162 };
 163
 164 enum {
 165         nsignals0 = ARRAY_SIZE(signalent0)
 166 #if SUPPORTED_PERSONALITIES > 1
 167         , nsignals1 = ARRAY_SIZE(signalent1)
 168 # if SUPPORTED_PERSONALITIES > 2
 169         , nsignals2 = ARRAY_SIZE(signalent2)
 170 # endif
 171 #endif
 172 };
 173
 174 enum {
 175         nioctlents0 = ARRAY_SIZE(ioctlent0)
 176 #if SUPPORTED_PERSONALITIES > 1
 177         , nioctlents1 = ARRAY_SIZE(ioctlent1)
 178 # if SUPPORTED_PERSONALITIES > 2
 179         , nioctlents2 = ARRAY_SIZE(ioctlent2)
 180 # endif
 181 #endif
 182 };
 183
 184 #if SUPPORTED_PERSONALITIES > 1
 185 const struct_sysent *sysent = sysent0;
 186 const char *const *errnoent = errnoent0;
 187 const char *const *signalent = signalent0;
 188 const struct_ioctlent *ioctlent = ioctlent0;
 189 const struct_printers *printers = &printers0;
 190 #endif
 191
 192 unsigned nsyscalls = nsyscalls0;
 193 unsigned nerrnos = nerrnos0;
 194 unsigned nsignals = nsignals0;
 195 unsigned nioctlents = nioctlents0;
 196
 197 const unsigned int nsyscall_vec[SUPPORTED_PERSONALITIES] = {
 198         nsyscalls0,
 199 #if SUPPORTED_PERSONALITIES > 1
 200         nsyscalls1,
 201 #endif
 202 #if SUPPORTED_PERSONALITIES > 2
 203         nsyscalls2,
 204 #endif
 205 };
 206 const struct_sysent *const sysent_vec[SUPPORTED_PERSONALITIES] = {
 207         sysent0,
 208 #if SUPPORTED_PERSONALITIES > 1
 209         sysent1,
 210 #endif
 211 #if SUPPORTED_PERSONALITIES > 2
 212         sysent2,
 213 #endif
 214 };
 215
 216 const char *const personality_names[] =
 217 # if defined X86_64
 218         {"64 bit", "32 bit", "x32"}
 219 # elif defined X32
 220         {"x32", "32 bit"}
 221 # elif SUPPORTED_PERSONALITIES == 2
 222         {"64 bit", "32 bit"}
 223 # else
 224         {STRINGIFY_VAL(__WORDSIZE) " bit"}
 225 # endif
 226         ;
 227
 228 #if SUPPORTED_PERSONALITIES > 1
 229
 230 unsigned current_personality;
 231
 232 # ifndef current_wordsize
 233 unsigned current_wordsize;
 234 static const int personality_wordsize[SUPPORTED_PERSONALITIES] = {
 235         PERSONALITY0_WORDSIZE,
 236         PERSONALITY1_WORDSIZE,
 237 # if SUPPORTED_PERSONALITIES > 2
 238         PERSONALITY2_WORDSIZE,
 239 # endif
 240 };
 241 # endif
 242
 243 # ifndef current_klongsize
 244 unsigned current_klongsize;
 245 static const int personality_klongsize[SUPPORTED_PERSONALITIES] = {
 246         PERSONALITY0_KLONGSIZE,
 247         PERSONALITY1_KLONGSIZE,
 248 #  if SUPPORTED_PERSONALITIES > 2
 249         PERSONALITY2_KLONGSIZE,
 250 #  endif
 251 };
 252 # endif
 253
 254 void
 255 set_personality(int personality)
 256 {
 257         nsyscalls = nsyscall_vec[personality];
 258         sysent = sysent_vec[personality];
 259
 260         switch (personality) {
 261         case 0:
 262                 errnoent = errnoent0;
 263                 nerrnos = nerrnos0;
 264                 ioctlent = ioctlent0;
 265                 nioctlents = nioctlents0;
 266                 signalent = signalent0;
 267                 nsignals = nsignals0;
 268                 printers = &printers0;
 269                 break;
 270
 271         case 1:
 272                 errnoent = errnoent1;
 273                 nerrnos = nerrnos1;
 274                 ioctlent = ioctlent1;
 275                 nioctlents = nioctlents1;
 276                 signalent = signalent1;
 277                 nsignals = nsignals1;
 278                 printers = &printers1;
 279                 break;
 280
 281 # if SUPPORTED_PERSONALITIES > 2
 282         case 2:
 283                 errnoent = errnoent2;
 284                 nerrnos = nerrnos2;
 285                 ioctlent = ioctlent2;
 286                 nioctlents = nioctlents2;
 287                 signalent = signalent2;
 288                 nsignals = nsignals2;
 289                 printers = &printers2;
 290                 break;
 291 # endif
 292         }
 293
 294         current_personality = personality;
 295 # ifndef current_wordsize
 296         current_wordsize = personality_wordsize[personality];
 297 # endif
 298 # ifndef current_klongsize
 299         current_klongsize = personality_klongsize[personality];
 300 # endif
 301 }
 302
 303 static void
 304 update_personality(struct tcb *tcp, unsigned int personality)
 305 {
 306         if (personality == current_personality)
 307                 return;
 308         set_personality(personality);
 309
 310         if (personality == tcp->currpers)
 311                 return;
 312         tcp->currpers = personality;
 313
 314         if (!qflag) {
 315                 error_msg("[ Process PID=%d runs in %s mode. ]",
 316                           tcp->pid, personality_names[personality]);
 317         }
 318 }
 319 #endif
 320
 321 #ifdef SYS_socket_subcall
 322 static void
 323 decode_socket_subcall(struct tcb *tcp)
 324 {
 325         const int call = tcp->u_arg[0];
 326
 327         if (call < 1 || call >= SYS_socket_nsubcalls)
 328                 return;
 329
 330         const kernel_ulong_t scno = SYS_socket_subcall + call;
 331         const unsigned int nargs = sysent[scno].nargs;
 332         uint64_t buf[nargs];
 333
 334         if (umoven(tcp, tcp->u_arg[1], nargs * current_wordsize, buf) < 0)
 335                 return;
 336
 337         tcp->scno = scno;
 338         tcp->qual_flg = qual_flags(scno);
 339         tcp->s_ent = &sysent[scno];
 340
 341         unsigned int i;
 342         for (i = 0; i < nargs; ++i)
 343                 tcp->u_arg[i] = (sizeof(uint32_t) == current_wordsize)
 344                                 ? ((uint32_t *) (void *) buf)[i] : buf[i];
 345 }
 346 #endif /* SYS_socket_subcall */
 347
 348 #ifdef SYS_ipc_subcall
 349 static void
 350 decode_ipc_subcall(struct tcb *tcp)
 351 {
 352         unsigned int call = tcp->u_arg[0];
 353         const unsigned int version = call >> 16;
 354
 355         if (version) {
 356 # if defined S390 || defined S390X
 357                 return;
 358 # else
 359 #  ifdef SPARC64
 360                 if (current_wordsize == 8)
 361                         return;
 362 #  endif
 363                 set_tcb_priv_ulong(tcp, version);
 364                 call &= 0xffff;
 365 # endif
 366         }
 367
 368         switch (call) {
 369                 case  1: case  2: case  3: case  4:
 370                 case 11: case 12: case 13: case 14:
 371                 case 21: case 22: case 23: case 24:
 372                         break;
 373                 default:
 374                         return;
 375         }
 376
 377         tcp->scno = SYS_ipc_subcall + call;
 378         tcp->qual_flg = qual_flags(tcp->scno);
 379         tcp->s_ent = &sysent[tcp->scno];
 380
 381         const unsigned int n = tcp->s_ent->nargs;
 382         unsigned int i;
 383         for (i = 0; i < n; i++)
 384                 tcp->u_arg[i] = tcp->u_arg[i + 1];
 385 }
 386 #endif /* SYS_ipc_subcall */
 387
 388 #ifdef SYS_syscall_subcall
 389 static void
 390 decode_syscall_subcall(struct tcb *tcp)
 391 {
 392         if (!scno_is_valid(tcp->u_arg[0]))
 393                 return;
 394         tcp->scno = tcp->u_arg[0];
 395         tcp->qual_flg = qual_flags(tcp->scno);
 396         tcp->s_ent = &sysent[tcp->scno];
 397         memmove(&tcp->u_arg[0], &tcp->u_arg[1],
 398                 sizeof(tcp->u_arg) - sizeof(tcp->u_arg[0]));
 399 # ifdef LINUX_MIPSO32
 400         /*
 401          * Fetching the last arg of 7-arg syscalls (fadvise64_64
 402          * and sync_file_range) requires additional code,
 403          * see linux/mips/get_syscall_args.c
 404          */
 405         if (tcp->s_ent->nargs == MAX_ARGS) {
 406                 if (umoven(tcp,
 407                            mips_REG_SP + MAX_ARGS * sizeof(tcp->u_arg[0]),
 408                            sizeof(tcp->u_arg[0]),
 409                            &tcp->u_arg[MAX_ARGS - 1]) < 0)
 410                 tcp->u_arg[MAX_ARGS - 1] = 0;
 411         }
 412 # endif /* LINUX_MIPSO32 */
 413 }
 414 #endif /* SYS_syscall_subcall */
 415
 416 static void
 417 dumpio(struct tcb *tcp)
 418 {
 419         if (syserror(tcp))
 420                 return;
 421
 422         int fd = tcp->u_arg[0];
 423         if (fd < 0)
 424                 return;
 425
 426         if (is_number_in_set(fd, read_set)) {
 427                 switch (tcp->s_ent->sen) {
 428                 case SEN_read:
 429                 case SEN_pread:
 430                 case SEN_recv:
 431                 case SEN_recvfrom:
 432                 case SEN_mq_timedreceive:
 433                         dumpstr(tcp, tcp->u_arg[1], tcp->u_rval);
 434                         return;
 435                 case SEN_readv:
 436                 case SEN_preadv:
 437                 case SEN_preadv2:
 438                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1],
 439                                      tcp->u_rval);
 440                         return;
 441                 case SEN_recvmsg:
 442                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], tcp->u_rval);
 443                         return;
 444                 case SEN_recvmmsg:
 445                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 446                         return;
 447                 }
 448         }
 449         if (is_number_in_set(fd, write_set)) {
 450                 switch (tcp->s_ent->sen) {
 451                 case SEN_write:
 452                 case SEN_pwrite:
 453                 case SEN_send:
 454                 case SEN_sendto:
 455                 case SEN_mq_timedsend:
 456                         dumpstr(tcp, tcp->u_arg[1], tcp->u_arg[2]);
 457                         break;
 458                 case SEN_writev:
 459                 case SEN_pwritev:
 460                 case SEN_pwritev2:
 461                 case SEN_vmsplice:
 462                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1], -1);
 463                         break;
 464                 case SEN_sendmsg:
 465                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], -1);
 466                         break;
 467                 case SEN_sendmmsg:
 468                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 469                         break;
 470                 }
 471         }
 472 }
 473
 474 /*
 475  * Shuffle syscall numbers so that we don't have huge gaps in syscall table.
 476  * The shuffling should be an involution: shuffle_scno(shuffle_scno(n)) == n.
 477  */
 478 static kernel_ulong_t
 479 shuffle_scno(kernel_ulong_t scno)
 480 {
 481 #ifdef ARM_FIRST_SHUFFLED_SYSCALL       /* So far only 32-bit ARM needs this */
 482         if (scno < ARM_FIRST_SHUFFLED_SYSCALL)
 483                 return scno;
 484
 485         /* __ARM_NR_cmpxchg? Swap with LAST_ORDINARY+1 */
 486         if (scno == ARM_FIRST_SHUFFLED_SYSCALL)
 487                 return 0x000ffff0;
 488         if (scno == 0x000ffff0)
 489                 return ARM_FIRST_SHUFFLED_SYSCALL;
 490
 491 # define ARM_SECOND_SHUFFLED_SYSCALL (ARM_FIRST_SHUFFLED_SYSCALL + 1)
 492         /*
 493          * Is it ARM specific syscall?
 494          * Swap [0x000f0000, 0x000f0000 + LAST_SPECIAL] range
 495          * with [SECOND_SHUFFLED, SECOND_SHUFFLED + LAST_SPECIAL] range.
 496          */
 497         if (scno >= 0x000f0000 &&
 498             scno <= 0x000f0000 + ARM_LAST_SPECIAL_SYSCALL) {
 499                 return scno - 0x000f0000 + ARM_SECOND_SHUFFLED_SYSCALL;
 500         }
 501         if (scno <= ARM_SECOND_SHUFFLED_SYSCALL + ARM_LAST_SPECIAL_SYSCALL) {
 502                 return scno + 0x000f0000 - ARM_SECOND_SHUFFLED_SYSCALL;
 503         }
 504 #endif /* ARM_FIRST_SHUFFLED_SYSCALL */
 505
 506         return scno;
 507 }
 508
 509 const char *
 510 err_name(unsigned long err)
 511 {
 512         if ((err < nerrnos) && errnoent[err])
 513                 return errnoent[err];
 514
 515         return NULL;
 516 }
 517
 518 static long get_regs(struct tcb *);
 519 static int get_syscall_args(struct tcb *);
 520 static int get_syscall_result(struct tcb *);
 521 static int arch_get_scno(struct tcb *tcp);
 522 static int arch_set_scno(struct tcb *, kernel_ulong_t);
 523 static void get_error(struct tcb *, const bool);
 524 static int arch_set_error(struct tcb *);
 525 static int arch_set_success(struct tcb *);
 526
 527 struct inject_opts *inject_vec[SUPPORTED_PERSONALITIES];
 528
 529 static struct inject_opts *
 530 tcb_inject_opts(struct tcb *tcp)
 531 {
 532         return (scno_in_range(tcp->scno) && tcp->inject_vec[current_personality])
 533                ? &tcp->inject_vec[current_personality][tcp->scno] : NULL;
 534 }
 535
 536
 537 static long
 538 tamper_with_syscall_entering(struct tcb *tcp, unsigned int *signo)
 539 {
 540         if (!tcp->inject_vec[current_personality]) {
 541                 tcp->inject_vec[current_personality] =
 542                         xcalloc(nsyscalls, sizeof(**inject_vec));
 543                 memcpy(tcp->inject_vec[current_personality],
 544                        inject_vec[current_personality],
 545                        nsyscalls * sizeof(**inject_vec));
 546         }
 547
 548         struct inject_opts *opts = tcb_inject_opts(tcp);
 549
 550         if (!opts || opts->first == 0)
 551                 return 0;
 552
 553         --opts->first;
 554
 555         if (opts->first != 0)
 556                 return 0;
 557
 558         opts->first = opts->step;
 559
 560         if (opts->data.flags & INJECT_F_SIGNAL)
 561                 *signo = opts->data.signo;
 562         if (opts->data.flags & INJECT_F_RETVAL && !arch_set_scno(tcp, -1))
 563                 tcp->flags |= TCB_TAMPERED;
 564
 565         return 0;
 566 }
 567
 568 static long
 569 tamper_with_syscall_exiting(struct tcb *tcp)
 570 {
 571         struct inject_opts *opts = tcb_inject_opts(tcp);
 572
 573         if (!opts)
 574                 return 0;
 575
 576         if (opts->data.rval >= 0) {
 577                 kernel_long_t u_rval = tcp->u_rval;
 578
 579                 tcp->u_rval = opts->data.rval;
 580                 if (arch_set_success(tcp)) {
 581                         tcp->u_rval = u_rval;
 582                 } else {
 583                         tcp->u_error = 0;
 584                 }
 585         } else {
 586                 unsigned long new_error = -opts->data.rval;
 587
 588                 if (new_error != tcp->u_error && new_error <= MAX_ERRNO_VALUE) {
 589                         unsigned long u_error = tcp->u_error;
 590
 591                         tcp->u_error = new_error;
 592                         if (arch_set_error(tcp)) {
 593                                 tcp->u_error = u_error;
 594                         }
 595                 }
 596         }
 597
 598         return 0;
 599 }
 600
 601 /*
 602  * Returns:
 603  * 0: "ignore this ptrace stop", bail out silently.
 604  * 1: ok, decoded; call
 605  *    syscall_entering_finish(tcp, syscall_entering_trace(tcp, ...)).
 606  * other: error; call syscall_entering_finish(tcp, res), where res is the value
 607  *    returned.
 608  */
 609 int
 610 syscall_entering_decode(struct tcb *tcp)
 611 {
 612         int res = get_scno(tcp);
 613         if (res == 0)
 614                 return res;
 615         int scno_good = res;
 616         if (res != 1 || (res = get_syscall_args(tcp)) != 1) {
 617                 printleader(tcp);
 618                 tprintf("%s(", scno_good == 1 ? tcp->s_ent->sys_name : "????");
 619                 /*
 620                  * " <unavailable>" will be added later by the code which
 621                  * detects ptrace errors.
 622                  */
 623                 return res;
 624         }
 625
 626 #if defined SYS_ipc_subcall     \
 627  || defined SYS_socket_subcall  \
 628  || defined SYS_syscall_subcall
 629         for (;;) {
 630                 switch (tcp->s_ent->sen) {
 631 # ifdef SYS_ipc_subcall
 632                 case SEN_ipc:
 633                         decode_ipc_subcall(tcp);
 634                         break;
 635 # endif
 636 # ifdef SYS_socket_subcall
 637                 case SEN_socketcall:
 638                         decode_socket_subcall(tcp);
 639                         break;
 640 # endif
 641 # ifdef SYS_syscall_subcall
 642                 case SEN_syscall:
 643                         decode_syscall_subcall(tcp);
 644                         if (tcp->s_ent->sen != SEN_syscall)
 645                                 continue;
 646                         break;
 647 # endif
 648                 }
 649                 break;
 650         }
 651 #endif
 652
 653         return 1;
 654 }
 655
 656 int
 657 syscall_entering_trace(struct tcb *tcp, unsigned int *sig)
 658 {
 659         /* Restrain from fault injection while the trace executes strace code. */
 660         if (hide_log(tcp)) {
 661                 tcp->qual_flg &= ~QUAL_INJECT;
 662         }
 663
 664         switch (tcp->s_ent->sen) {
 665                 case SEN_execve:
 666                 case SEN_execveat:
 667 #if defined SPARC || defined SPARC64
 668                 case SEN_execv:
 669 #endif
 670                         tcp->flags &= ~TCB_HIDE_LOG;
 671                         break;
 672         }
 673
 674         if (!traced(tcp) || (tracing_paths && !pathtrace_match(tcp))) {
 675                 tcp->flags |= TCB_FILTERED;
 676                 return 0;
 677         }
 678
 679         tcp->flags &= ~TCB_FILTERED;
 680
 681         if (hide_log(tcp)) {
 682                 return 0;
 683         }
 684
 685         if (inject(tcp))
 686                 tamper_with_syscall_entering(tcp, sig);
 687
 688         if (cflag == CFLAG_ONLY_STATS) {
 689                 return 0;
 690         }
 691
 692 #ifdef USE_LIBUNWIND
 693         if (stack_trace_enabled) {
 694                 if (tcp->s_ent->sys_flags & STACKTRACE_CAPTURE_ON_ENTER)
 695                         unwind_capture_stacktrace(tcp);
 696         }
 697 #endif
 698
 699         printleader(tcp);
 700         tprintf("%s(", tcp->s_ent->sys_name);
 701         int res = raw(tcp) ? printargs(tcp) : tcp->s_ent->sys_func(tcp);
 702         fflush(tcp->outf);
 703         return res;
 704 }
 705
 706 void
 707 syscall_entering_finish(struct tcb *tcp, int res)
 708 {
 709         tcp->flags |= TCB_INSYSCALL;
 710         tcp->sys_func_rval = res;
 711         /* Measure the entrance time as late as possible to avoid errors. */
 712         if ((Tflag || cflag) && !filtered(tcp))
 713                 gettimeofday(&tcp->etime, NULL);
 714 }
 715
 716 static bool
 717 syscall_tampered(struct tcb *tcp)
 718 {
 719         return tcp->flags & TCB_TAMPERED;
 720 }
 721
 722 /* Returns:
 723  * 0: "bail out".
 724  * 1: ok.
 725  * -1: error in one of ptrace ops.
 726  *
 727  * If not 0, call syscall_exiting_trace(tcp, res), where res is the return
 728  *    value. Anyway, call syscall_exiting_finish(tcp) then.
 729  */
 730 int
 731 syscall_exiting_decode(struct tcb *tcp, struct timeval *ptv)
 732 {
 733         /* Measure the exit time as early as possible to avoid errors. */
 734         if ((Tflag || cflag) && !(filtered(tcp) || hide_log(tcp)))
 735                 gettimeofday(ptv, NULL);
 736
 737 #ifdef USE_LIBUNWIND
 738         if (stack_trace_enabled) {
 739                 if (tcp->s_ent->sys_flags & STACKTRACE_INVALIDATE_CACHE)
 740                         unwind_cache_invalidate(tcp);
 741         }
 742 #endif
 743
 744         if (filtered(tcp) || hide_log(tcp))
 745                 return 0;
 746
 747 #if SUPPORTED_PERSONALITIES > 1
 748         update_personality(tcp, tcp->currpers);
 749 #endif
 750
 751         return get_syscall_result(tcp);
 752 }
 753
 754 int
 755 syscall_exiting_trace(struct tcb *tcp, struct timeval tv, int res)
 756 {
 757         if (syserror(tcp) && syscall_tampered(tcp))
 758                 tamper_with_syscall_exiting(tcp);
 759
 760         if (cflag) {
 761                 count_syscall(tcp, &tv);
 762                 if (cflag == CFLAG_ONLY_STATS) {
 763                         return 0;
 764                 }
 765         }
 766
 767         /* If not in -ff mode, and printing_tcp != tcp,
 768          * then the log currently does not end with output
 769          * of _our syscall entry_, but with something else.
 770          * We need to say which syscall's return is this.
 771          *
 772          * Forced reprinting via TCB_REPRINT is used only by
 773          * "strace -ff -oLOG test/threaded_execve" corner case.
 774          * It's the only case when -ff mode needs reprinting.
 775          */
 776         if ((followfork < 2 && printing_tcp != tcp) || (tcp->flags & TCB_REPRINT)) {
 777                 tcp->flags &= ~TCB_REPRINT;
 778                 printleader(tcp);
 779                 tprintf("<... %s resumed> ", tcp->s_ent->sys_name);
 780         }
 781         printing_tcp = tcp;
 782
 783         tcp->s_prev_ent = NULL;
 784         if (res != 1) {
 785                 /* There was error in one of prior ptrace ops */
 786                 tprints(") ");
 787                 tabto();
 788                 tprints("= ? <unavailable>\n");
 789                 line_ended();
 790                 return res;
 791         }
 792         tcp->s_prev_ent = tcp->s_ent;
 793
 794         int sys_res = 0;
 795         if (raw(tcp)) {
 796                 /* sys_res = printargs(tcp); - but it's nop on sysexit */
 797         } else {
 798         /* FIXME: not_failing_only (IOW, option -z) is broken:
 799          * failure of syscall is known only after syscall return.
 800          * Thus we end up with something like this on, say, ENOENT:
 801          *     open("does_not_exist", O_RDONLY <unfinished ...>
 802          *     {next syscall decode}
 803          * whereas the intended result is that open(...) line
 804          * is not shown at all.
 805          */
 806                 if (not_failing_only && tcp->u_error)
 807                         return 0;       /* ignore failed syscalls */
 808                 if (tcp->sys_func_rval & RVAL_DECODED)
 809                         sys_res = tcp->sys_func_rval;
 810                 else
 811                         sys_res = tcp->s_ent->sys_func(tcp);
 812         }
 813
 814         tprints(") ");
 815         tabto();
 816         unsigned long u_error = tcp->u_error;
 817
 818         if (raw(tcp)) {
 819                 if (u_error) {
 820                         tprintf("= -1 (errno %lu)", u_error);
 821                 } else {
 822                         tprintf("= %#" PRI_klx, tcp->u_rval);
 823                 }
 824                 if (syscall_tampered(tcp))
 825                         tprints(" (INJECTED)");
 826         } else if (!(sys_res & RVAL_NONE) && u_error) {
 827                 const char *u_error_str;
 828
 829                 switch (u_error) {
 830                 /* Blocked signals do not interrupt any syscalls.
 831                  * In this case syscalls don't return ERESTARTfoo codes.
 832                  *
 833                  * Deadly signals set to SIG_DFL interrupt syscalls
 834                  * and kill the process regardless of which of the codes below
 835                  * is returned by the interrupted syscall.
 836                  * In some cases, kernel forces a kernel-generated deadly
 837                  * signal to be unblocked and set to SIG_DFL (and thus cause
 838                  * death) if it is blocked or SIG_IGNed: for example, SIGSEGV
 839                  * or SIGILL. (The alternative is to leave process spinning
 840                  * forever on the faulty instruction - not useful).
 841                  *
 842                  * SIG_IGNed signals and non-deadly signals set to SIG_DFL
 843                  * (for example, SIGCHLD, SIGWINCH) interrupt syscalls,
 844                  * but kernel will always restart them.
 845                  */
 846                 case ERESTARTSYS:
 847                         /* Most common type of signal-interrupted syscall exit code.
 848                          * The system call will be restarted with the same arguments
 849                          * if SA_RESTART is set; otherwise, it will fail with EINTR.
 850                          */
 851                         tprints("= ? ERESTARTSYS (To be restarted if SA_RESTART is set)");
 852                         break;
 853                 case ERESTARTNOINTR:
 854                         /* Rare. For example, fork() returns this if interrupted.
 855                          * SA_RESTART is ignored (assumed set): the restart is unconditional.
 856                          */
 857                         tprints("= ? ERESTARTNOINTR (To be restarted)");
 858                         break;
 859                 case ERESTARTNOHAND:
 860                         /* pause(), rt_sigsuspend() etc use this code.
 861                          * SA_RESTART is ignored (assumed not set):
 862                          * syscall won't restart (will return EINTR instead)
 863                          * even after signal with SA_RESTART set. However,
 864                          * after SIG_IGN or SIG_DFL signal it will restart
 865                          * (thus the name "restart only if has no handler").
 866                          */
 867                         tprints("= ? ERESTARTNOHAND (To be restarted if no handler)");
 868                         break;
 869                 case ERESTART_RESTARTBLOCK:
 870                         /* Syscalls like nanosleep(), poll() which can't be
 871                          * restarted with their original arguments use this
 872                          * code. Kernel will execute restart_syscall() instead,
 873                          * which changes arguments before restarting syscall.
 874                          * SA_RESTART is ignored (assumed not set) similarly
 875                          * to ERESTARTNOHAND. (Kernel can't honor SA_RESTART
 876                          * since restart data is saved in "restart block"
 877                          * in task struct, and if signal handler uses a syscall
 878                          * which in turn saves another such restart block,
 879                          * old data is lost and restart becomes impossible)
 880                          */
 881                         tprints("= ? ERESTART_RESTARTBLOCK (Interrupted by signal)");
 882                         break;
 883                 default:
 884                         u_error_str = err_name(u_error);
 885                         if (u_error_str)
 886                                 tprintf("= -1 %s (%s)",
 887                                         u_error_str, strerror(u_error));
 888                         else
 889                                 tprintf("= -1 %lu (%s)",
 890                                         u_error, strerror(u_error));
 891                         break;
 892                 }
 893                 if (syscall_tampered(tcp))
 894                         tprints(" (INJECTED)");
 895                 if ((sys_res & RVAL_STR) && tcp->auxstr)
 896                         tprintf(" (%s)", tcp->auxstr);
 897         } else {
 898                 if (sys_res & RVAL_NONE)
 899                         tprints("= ?");
 900                 else {
 901                         switch (sys_res & RVAL_MASK) {
 902                         case RVAL_HEX:
 903 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
 904                                 if (current_wordsize < sizeof(tcp->u_rval)) {
 905                                         tprintf("= %#x",
 906                                                 (unsigned int) tcp->u_rval);
 907                                 } else
 908 #endif
 909                                 {
 910                                         tprintf("= %#" PRI_klx, tcp->u_rval);
 911                                 }
 912                                 break;
 913                         case RVAL_OCTAL:
 914                                 tprints("= ");
 915                                 print_numeric_long_umask(tcp->u_rval);
 916                                 break;
 917                         case RVAL_UDECIMAL:
 918 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
 919                                 if (current_wordsize < sizeof(tcp->u_rval)) {
 920                                         tprintf("= %u",
 921                                                 (unsigned int) tcp->u_rval);
 922                                 } else
 923 #endif
 924                                 {
 925                                         tprintf("= %" PRI_klu, tcp->u_rval);
 926                                 }
 927                                 break;
 928                         case RVAL_DECIMAL:
 929                                 tprintf("= %" PRI_kld, tcp->u_rval);
 930                                 break;
 931                         case RVAL_FD:
 932                                 if (show_fd_path) {
 933                                         tprints("= ");
 934                                         printfd(tcp, tcp->u_rval);
 935                                 } else
 936                                         tprintf("= %" PRI_kld, tcp->u_rval);
 937                                 break;
 938                         default:
 939                                 error_msg("invalid rval format");
 940                                 break;
 941                         }
 942                 }
 943                 if ((sys_res & RVAL_STR) && tcp->auxstr)
 944                         tprintf(" (%s)", tcp->auxstr);
 945                 if (syscall_tampered(tcp))
 946                         tprints(" (INJECTED)");
 947         }
 948         if (Tflag) {
 949                 tv_sub(&tv, &tv, &tcp->etime);
 950                 tprintf(" <%ld.%06ld>",
 951                         (long) tv.tv_sec, (long) tv.tv_usec);
 952         }
 953         tprints("\n");
 954         dumpio(tcp);
 955         line_ended();
 956
 957 #ifdef USE_LIBUNWIND
 958         if (stack_trace_enabled)
 959                 unwind_print_stacktrace(tcp);
 960 #endif
 961         return 0;
 962 }
 963
 964 void
 965 syscall_exiting_finish(struct tcb *tcp)
 966 {
 967         tcp->flags &= ~(TCB_INSYSCALL | TCB_TAMPERED);
 968         tcp->sys_func_rval = 0;
 969         free_tcb_priv_data(tcp);
 970 }
 971
 972 bool
 973 is_erestart(struct tcb *tcp)
 974 {
 975         switch (tcp->u_error) {
 976                 case ERESTARTSYS:
 977                 case ERESTARTNOINTR:
 978                 case ERESTARTNOHAND:
 979                 case ERESTART_RESTARTBLOCK:
 980                         return true;
 981                 default:
 982                         return false;
 983         }
 984 }
 985
 986 static unsigned long saved_u_error;
 987
 988 void
 989 temporarily_clear_syserror(struct tcb *tcp)
 990 {
 991         saved_u_error = tcp->u_error;
 992         tcp->u_error = 0;
 993 }
 994
 995 void
 996 restore_cleared_syserror(struct tcb *tcp)
 997 {
 998         tcp->u_error = saved_u_error;
 999 }
1000
1001 #include "arch_regs.c"
1002
1003 #ifdef HAVE_GETRVAL2
1004 # include "arch_getrval2.c"
1005 #endif
1006
1007 void
1008 print_pc(struct tcb *tcp)
1009 {
1010 #if defined ARCH_PC_REG
1011 # define ARCH_GET_PC 0
1012 #elif defined ARCH_PC_PEEK_ADDR
1013         kernel_ulong_t pc;
1014 # define ARCH_PC_REG pc
1015 # define ARCH_GET_PC upeek(tcp, ARCH_PC_PEEK_ADDR, &pc)
1016 #else
1017 # error Neither ARCH_PC_REG nor ARCH_PC_PEEK_ADDR is defined
1018 #endif
1019         if (get_regs(tcp) < 0 || ARCH_GET_PC)
1020                 tprints(current_wordsize == 4 ? "[????????] "
1021                                               : "[????????????????] ");
1022         else
1023                 tprintf(current_wordsize == 4
1024                         ? "[%08" PRI_klx "] " : "[%016" PRI_klx "] ",
1025                         (kernel_ulong_t) ARCH_PC_REG);
1026 }
1027
1028 #include "getregs_old.h"
1029
1030 #undef ptrace_getregset_or_getregs
1031 #undef ptrace_setregset_or_setregs
1032 #ifdef ARCH_REGS_FOR_GETREGSET
1033
1034 # define ptrace_getregset_or_getregs ptrace_getregset
1035 static long
1036 ptrace_getregset(pid_t pid)
1037 {
1038 # ifdef ARCH_IOVEC_FOR_GETREGSET
1039         /* variable iovec */
1040         ARCH_IOVEC_FOR_GETREGSET.iov_len = sizeof(ARCH_REGS_FOR_GETREGSET);
1041         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS,
1042                       &ARCH_IOVEC_FOR_GETREGSET);
1043 # else
1044         /* constant iovec */
1045         static struct iovec io = {
1046                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1047                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1048         };
1049         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS, &io);
1050
1051 # endif
1052 }
1053
1054 # ifndef HAVE_GETREGS_OLD
1055 #  define ptrace_setregset_or_setregs ptrace_setregset
1056 static int
1057 ptrace_setregset(pid_t pid)
1058 {
1059 #  ifdef ARCH_IOVEC_FOR_GETREGSET
1060         /* variable iovec */
1061         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS,
1062                       &ARCH_IOVEC_FOR_GETREGSET);
1063 #  else
1064         /* constant iovec */
1065         static struct iovec io = {
1066                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1067                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1068         };
1069         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS, &io);
1070 #  endif
1071 }
1072 # endif /* !HAVE_GETREGS_OLD */
1073
1074 #elif defined ARCH_REGS_FOR_GETREGS
1075
1076 # define ptrace_getregset_or_getregs ptrace_getregs
1077 static long
1078 ptrace_getregs(pid_t pid)
1079 {
1080 # if defined SPARC || defined SPARC64
1081         /* SPARC systems have the meaning of data and addr reversed */
1082         return ptrace(PTRACE_GETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1083 # else
1084         return ptrace(PTRACE_GETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1085 # endif
1086 }
1087
1088 # ifndef HAVE_GETREGS_OLD
1089 #  define ptrace_setregset_or_setregs ptrace_setregs
1090 static int
1091 ptrace_setregs(pid_t pid)
1092 {
1093 #  if defined SPARC || defined SPARC64
1094         /* SPARC systems have the meaning of data and addr reversed */
1095         return ptrace(PTRACE_SETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1096 #  else
1097         return ptrace(PTRACE_SETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1098 #  endif
1099 }
1100 # endif /* !HAVE_GETREGS_OLD */
1101
1102 #endif /* ARCH_REGS_FOR_GETREGSET || ARCH_REGS_FOR_GETREGS */
1103
1104 #ifdef ptrace_getregset_or_getregs
1105 static long get_regs_error;
1106 #endif
1107
1108 void
1109 clear_regs(void)
1110 {
1111 #ifdef ptrace_getregset_or_getregs
1112         get_regs_error = -1;
1113 #endif
1114 }
1115
1116 static long
1117 get_regs(struct tcb *const tcp)
1118 {
1119 #ifdef ptrace_getregset_or_getregs
1120
1121         if (get_regs_error != -1)
1122                 return get_regs_error;
1123
1124 # ifdef HAVE_GETREGS_OLD
1125         /*
1126          * Try PTRACE_GETREGSET/PTRACE_GETREGS first,
1127          * fallback to getregs_old.
1128          */
1129         static int use_getregs_old;
1130         if (use_getregs_old < 0) {
1131                 return get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1132         } else if (use_getregs_old == 0) {
1133                 get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1134                 if (get_regs_error >= 0) {
1135                         use_getregs_old = -1;
1136                         return get_regs_error;
1137                 }
1138                 if (errno == EPERM || errno == ESRCH)
1139                         return get_regs_error;
1140                 use_getregs_old = 1;
1141         }
1142         return get_regs_error = getregs_old(tcp);
1143 # else /* !HAVE_GETREGS_OLD */
1144         /* Assume that PTRACE_GETREGSET/PTRACE_GETREGS works. */
1145         return get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1146 # endif /* !HAVE_GETREGS_OLD */
1147
1148 #else /* !ptrace_getregset_or_getregs */
1149
1150 # warning get_regs is not implemented for this architecture yet
1151         return 0;
1152
1153 #endif /* !ptrace_getregset_or_getregs */
1154 }
1155
1156 #ifdef ptrace_setregset_or_setregs
1157 static int
1158 set_regs(pid_t pid)
1159 {
1160         return ptrace_setregset_or_setregs(pid);
1161 }
1162 #endif /* ptrace_setregset_or_setregs */
1163
1164 struct sysent_buf {
1165         struct tcb *tcp;
1166         struct_sysent ent;
1167         char buf[sizeof("syscall_%lu") + sizeof(kernel_ulong_t) * 3];
1168 };
1169
1170 static void
1171 free_sysent_buf(void *ptr)
1172 {
1173         struct sysent_buf *s = ptr;
1174         s->tcp->s_prev_ent = s->tcp->s_ent = NULL;
1175         free(ptr);
1176 }
1177
1178 /*
1179  * Returns:
1180  * 0: "ignore this ptrace stop", syscall_entering_decode() should return a "bail
1181  *    out silently" code.
1182  * 1: ok, continue in syscall_entering_decode().
1183  * other: error, syscall_entering_decode() should print error indicator
1184  *    ("????" etc) and return an appropriate code.
1185  */
1186 int
1187 get_scno(struct tcb *tcp)
1188 {
1189         if (get_regs(tcp) < 0)
1190                 return -1;
1191
1192         int rc = arch_get_scno(tcp);
1193         if (rc != 1)
1194                 return rc;
1195
1196         if (scno_is_valid(tcp->scno)) {
1197                 tcp->s_ent = &sysent[tcp->scno];
1198                 tcp->qual_flg = qual_flags(tcp->scno);
1199         } else {
1200                 struct sysent_buf *s = xcalloc(1, sizeof(*s));
1201
1202                 s->tcp = tcp;
1203                 s->ent.nargs = MAX_ARGS;
1204                 s->ent.sen = SEN_printargs;
1205                 s->ent.sys_func = printargs;
1206                 s->ent.sys_name = s->buf;
1207                 xsprintf(s->buf, "syscall_%" PRI_klu, shuffle_scno(tcp->scno));
1208
1209                 tcp->s_ent = &s->ent;
1210                 tcp->qual_flg = QUAL_RAW | DEFAULT_QUAL_FLAGS;
1211
1212                 set_tcb_priv_data(tcp, s, free_sysent_buf);
1213
1214                 debug_msg("pid %d invalid syscall %" PRI_kld,
1215                           tcp->pid, tcp->scno);
1216         }
1217         return 1;
1218 }
1219
1220 #ifdef ptrace_getregset_or_getregs
1221 # define get_syscall_result_regs get_regs
1222 #else
1223 static int get_syscall_result_regs(struct tcb *);
1224 #endif
1225
1226 /* Returns:
1227  * 1: ok, continue in syscall_exiting_trace().
1228  * -1: error, syscall_exiting_trace() should print error indicator
1229  *    ("????" etc) and bail out.
1230  */
1231 static int
1232 get_syscall_result(struct tcb *tcp)
1233 {
1234         if (get_syscall_result_regs(tcp) < 0)
1235                 return -1;
1236         tcp->u_error = 0;
1237         get_error(tcp, !(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS));
1238
1239         return 1;
1240 }
1241
1242 #include "get_scno.c"
1243 #include "set_scno.c"
1244 #include "get_syscall_args.c"
1245 #ifndef ptrace_getregset_or_getregs
1246 # include "get_syscall_result.c"
1247 #endif
1248 #include "get_error.c"
1249 #include "set_error.c"
1250 #ifdef HAVE_GETREGS_OLD
1251 # include "getregs_old.c"
1252 #endif
1253
1254 const char *
1255 syscall_name(kernel_ulong_t scno)
1256 {
1257 #if defined X32_PERSONALITY_NUMBER && defined __X32_SYSCALL_BIT
1258         if (current_personality == X32_PERSONALITY_NUMBER)
1259                 scno &= ~__X32_SYSCALL_BIT;
1260 #endif
1261         return scno_is_valid(scno) ? sysent[scno].sys_name : NULL;
1262 }