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  * All rights reserved.
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  * 1. Redistributions of source code must retain the above copyright
  15  *    notice, this list of conditions and the following disclaimer.
  16  * 2. Redistributions in binary form must reproduce the above copyright
  17  *    notice, this list of conditions and the following disclaimer in the
  18  *    documentation and/or other materials provided with the distribution.
  19  * 3. The name of the author may not be used to endorse or promote products
  20  *    derived from this software without specific prior written permission.
  21  *
  22  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  23  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  24  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  25  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  27  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  31  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include "defs.h"
  35 #include "native_defs.h"
  36 #include "nsig.h"
  37 #include <sys/param.h>
  38
  39 /* for struct iovec */
  40 #include <sys/uio.h>
  41
  42 #include "regs.h"
  43 #include "ptrace.h"
  44
  45 #if defined(SPARC64)
  46 # undef PTRACE_GETREGS
  47 # define PTRACE_GETREGS PTRACE_GETREGS64
  48 # undef PTRACE_SETREGS
  49 # define PTRACE_SETREGS PTRACE_SETREGS64
  50 #endif
  51
  52 #if defined SPARC64
  53 # include <asm/psrcompat.h>
  54 #elif defined SPARC
  55 # include <asm/psr.h>
  56 #endif
  57
  58 #ifdef IA64
  59 # include <asm/rse.h>
  60 #endif
  61
  62 #ifndef NT_PRSTATUS
  63 # define NT_PRSTATUS 1
  64 #endif
  65
  66 #include "syscall.h"
  67
  68 /* Define these shorthand notations to simplify the syscallent files. */
  69 #define TD TRACE_DESC
  70 #define TF TRACE_FILE
  71 #define TI TRACE_IPC
  72 #define TN TRACE_NETWORK
  73 #define TP TRACE_PROCESS
  74 #define TS TRACE_SIGNAL
  75 #define TM TRACE_MEMORY
  76 #define NF SYSCALL_NEVER_FAILS
  77 #define MA MAX_ARGS
  78 #define SI STACKTRACE_INVALIDATE_CACHE
  79 #define SE STACKTRACE_CAPTURE_ON_ENTER
  80 #define CST COMPAT_SYSCALL_TYPES
  81
  82 #define SEN(syscall_name) SEN_ ## syscall_name, SYS_FUNC_NAME(sys_ ## syscall_name)
  83
  84 const struct_sysent sysent0[] = {
  85 #include "syscallent.h"
  86 };
  87
  88 #if SUPPORTED_PERSONALITIES > 1
  89 # include PERSONALITY1_INCLUDE_FUNCS
  90 static const struct_sysent sysent1[] = {
  91 # include "syscallent1.h"
  92 };
  93 #endif
  94
  95 #if SUPPORTED_PERSONALITIES > 2
  96 # include PERSONALITY2_INCLUDE_FUNCS
  97 static const struct_sysent sysent2[] = {
  98 # include "syscallent2.h"
  99 };
 100 #endif
 101
 102 /* Now undef them since short defines cause wicked namespace pollution. */
 103 #undef SEN
 104 #undef TD
 105 #undef TF
 106 #undef TI
 107 #undef TN
 108 #undef TP
 109 #undef TS
 110 #undef TM
 111 #undef NF
 112 #undef MA
 113 #undef SI
 114 #undef SE
 115 #undef CST
 116
 117 /*
 118  * `ioctlent[012].h' files are automatically generated by the auxiliary
 119  * program `ioctlsort', such that the list is sorted by the `code' field.
 120  * This has the side-effect of resolving the _IO.. macros into
 121  * plain integers, eliminating the need to include here everything
 122  * in "/usr/include".
 123  */
 124
 125 const char *const errnoent0[] = {
 126 #include "errnoent.h"
 127 };
 128 const char *const signalent0[] = {
 129 #include "signalent.h"
 130 };
 131 const struct_ioctlent ioctlent0[] = {
 132 #include "ioctlent0.h"
 133 };
 134
 135 #if SUPPORTED_PERSONALITIES > 1
 136 static const char *const errnoent1[] = {
 137 # include "errnoent1.h"
 138 };
 139 static const char *const signalent1[] = {
 140 # include "signalent1.h"
 141 };
 142 static const struct_ioctlent ioctlent1[] = {
 143 # include "ioctlent1.h"
 144 };
 145 # include PERSONALITY0_INCLUDE_PRINTERS_DECLS
 146 static const struct_printers printers0 = {
 147 # include PERSONALITY0_INCLUDE_PRINTERS_DEFS
 148 };
 149 # include PERSONALITY1_INCLUDE_PRINTERS_DECLS
 150 static const struct_printers printers1 = {
 151 # include PERSONALITY1_INCLUDE_PRINTERS_DEFS
 152 };
 153 #endif
 154
 155 #if SUPPORTED_PERSONALITIES > 2
 156 static const char *const errnoent2[] = {
 157 # include "errnoent2.h"
 158 };
 159 static const char *const signalent2[] = {
 160 # include "signalent2.h"
 161 };
 162 static const struct_ioctlent ioctlent2[] = {
 163 # include "ioctlent2.h"
 164 };
 165 # include PERSONALITY2_INCLUDE_PRINTERS_DECLS
 166 static const struct_printers printers2 = {
 167 # include PERSONALITY2_INCLUDE_PRINTERS_DEFS
 168 };
 169 #endif
 170
 171 enum {
 172         nsyscalls0 = ARRAY_SIZE(sysent0)
 173 #if SUPPORTED_PERSONALITIES > 1
 174         , nsyscalls1 = ARRAY_SIZE(sysent1)
 175 # if SUPPORTED_PERSONALITIES > 2
 176         , nsyscalls2 = ARRAY_SIZE(sysent2)
 177 # endif
 178 #endif
 179 };
 180
 181 enum {
 182         nerrnos0 = ARRAY_SIZE(errnoent0)
 183 #if SUPPORTED_PERSONALITIES > 1
 184         , nerrnos1 = ARRAY_SIZE(errnoent1)
 185 # if SUPPORTED_PERSONALITIES > 2
 186         , nerrnos2 = ARRAY_SIZE(errnoent2)
 187 # endif
 188 #endif
 189 };
 190
 191 enum {
 192         nsignals0 = ARRAY_SIZE(signalent0)
 193 #if SUPPORTED_PERSONALITIES > 1
 194         , nsignals1 = ARRAY_SIZE(signalent1)
 195 # if SUPPORTED_PERSONALITIES > 2
 196         , nsignals2 = ARRAY_SIZE(signalent2)
 197 # endif
 198 #endif
 199 };
 200
 201 enum {
 202         nioctlents0 = ARRAY_SIZE(ioctlent0)
 203 #if SUPPORTED_PERSONALITIES > 1
 204         , nioctlents1 = ARRAY_SIZE(ioctlent1)
 205 # if SUPPORTED_PERSONALITIES > 2
 206         , nioctlents2 = ARRAY_SIZE(ioctlent2)
 207 # endif
 208 #endif
 209 };
 210
 211 #if SUPPORTED_PERSONALITIES > 1
 212 const struct_sysent *sysent = sysent0;
 213 const char *const *errnoent = errnoent0;
 214 const char *const *signalent = signalent0;
 215 const struct_ioctlent *ioctlent = ioctlent0;
 216 const struct_printers *printers = &printers0;
 217 #endif
 218
 219 unsigned nsyscalls = nsyscalls0;
 220 unsigned nerrnos = nerrnos0;
 221 unsigned nsignals = nsignals0;
 222 unsigned nioctlents = nioctlents0;
 223
 224 const unsigned int nsyscall_vec[SUPPORTED_PERSONALITIES] = {
 225         nsyscalls0,
 226 #if SUPPORTED_PERSONALITIES > 1
 227         nsyscalls1,
 228 #endif
 229 #if SUPPORTED_PERSONALITIES > 2
 230         nsyscalls2,
 231 #endif
 232 };
 233 const struct_sysent *const sysent_vec[SUPPORTED_PERSONALITIES] = {
 234         sysent0,
 235 #if SUPPORTED_PERSONALITIES > 1
 236         sysent1,
 237 #endif
 238 #if SUPPORTED_PERSONALITIES > 2
 239         sysent2,
 240 #endif
 241 };
 242
 243 #if SUPPORTED_PERSONALITIES > 1
 244 unsigned current_personality;
 245
 246 # ifndef current_wordsize
 247 unsigned current_wordsize;
 248 static const int personality_wordsize[SUPPORTED_PERSONALITIES] = {
 249         PERSONALITY0_WORDSIZE,
 250         PERSONALITY1_WORDSIZE,
 251 # if SUPPORTED_PERSONALITIES > 2
 252         PERSONALITY2_WORDSIZE,
 253 # endif
 254 };
 255 # endif
 256
 257 # ifndef current_klongsize
 258 unsigned current_klongsize;
 259 static const int personality_klongsize[SUPPORTED_PERSONALITIES] = {
 260         PERSONALITY0_KLONGSIZE,
 261         PERSONALITY1_KLONGSIZE,
 262 #  if SUPPORTED_PERSONALITIES > 2
 263         PERSONALITY2_KLONGSIZE,
 264 #  endif
 265 };
 266 # endif
 267
 268 void
 269 set_personality(int personality)
 270 {
 271         nsyscalls = nsyscall_vec[personality];
 272         sysent = sysent_vec[personality];
 273
 274         switch (personality) {
 275         case 0:
 276                 errnoent = errnoent0;
 277                 nerrnos = nerrnos0;
 278                 ioctlent = ioctlent0;
 279                 nioctlents = nioctlents0;
 280                 signalent = signalent0;
 281                 nsignals = nsignals0;
 282                 printers = &printers0;
 283                 break;
 284
 285         case 1:
 286                 errnoent = errnoent1;
 287                 nerrnos = nerrnos1;
 288                 ioctlent = ioctlent1;
 289                 nioctlents = nioctlents1;
 290                 signalent = signalent1;
 291                 nsignals = nsignals1;
 292                 printers = &printers1;
 293                 break;
 294
 295 # if SUPPORTED_PERSONALITIES > 2
 296         case 2:
 297                 errnoent = errnoent2;
 298                 nerrnos = nerrnos2;
 299                 ioctlent = ioctlent2;
 300                 nioctlents = nioctlents2;
 301                 signalent = signalent2;
 302                 nsignals = nsignals2;
 303                 printers = &printers2;
 304                 break;
 305 # endif
 306         }
 307
 308         current_personality = personality;
 309 # ifndef current_wordsize
 310         current_wordsize = personality_wordsize[personality];
 311 # endif
 312 # ifndef current_klongsize
 313         current_klongsize = personality_klongsize[personality];
 314 # endif
 315 }
 316
 317 static void
 318 update_personality(struct tcb *tcp, unsigned int personality)
 319 {
 320         if (personality == current_personality)
 321                 return;
 322         set_personality(personality);
 323
 324         if (personality == tcp->currpers)
 325                 return;
 326         tcp->currpers = personality;
 327
 328 # undef PERSONALITY_NAMES
 329 # if defined POWERPC64
 330 #  define PERSONALITY_NAMES {"64 bit", "32 bit"}
 331 # elif defined X86_64
 332 #  define PERSONALITY_NAMES {"64 bit", "32 bit", "x32"}
 333 # elif defined X32
 334 #  define PERSONALITY_NAMES {"x32", "32 bit"}
 335 # elif defined AARCH64
 336 #  define PERSONALITY_NAMES {"64 bit", "32 bit"}
 337 # elif defined TILE
 338 #  define PERSONALITY_NAMES {"64-bit", "32-bit"}
 339 # endif
 340 # ifdef PERSONALITY_NAMES
 341         if (!qflag) {
 342                 static const char *const names[] = PERSONALITY_NAMES;
 343                 error_msg("[ Process PID=%d runs in %s mode. ]",
 344                           tcp->pid, names[personality]);
 345         }
 346 # endif
 347 }
 348 #endif
 349
 350 #ifdef SYS_socket_subcall
 351 static void
 352 decode_socket_subcall(struct tcb *tcp)
 353 {
 354         const int call = tcp->u_arg[0];
 355
 356         if (call < 1 || call >= SYS_socket_nsubcalls)
 357                 return;
 358
 359         const kernel_ulong_t scno = SYS_socket_subcall + call;
 360         const unsigned int nargs = sysent[scno].nargs;
 361         uint64_t buf[nargs];
 362
 363         if (umoven(tcp, tcp->u_arg[1], nargs * current_wordsize, buf) < 0)
 364                 return;
 365
 366         tcp->scno = scno;
 367         tcp->qual_flg = qual_flags(scno);
 368         tcp->s_ent = &sysent[scno];
 369
 370         unsigned int i;
 371         for (i = 0; i < nargs; ++i)
 372                 tcp->u_arg[i] = (sizeof(uint32_t) == current_wordsize)
 373                                 ? ((uint32_t *) (void *) buf)[i] : buf[i];
 374 }
 375 #endif
 376
 377 #ifdef SYS_ipc_subcall
 378 static void
 379 decode_ipc_subcall(struct tcb *tcp)
 380 {
 381         unsigned int call = tcp->u_arg[0];
 382         const unsigned int version = call >> 16;
 383
 384         if (version) {
 385 # if defined S390 || defined S390X
 386                 return;
 387 # else
 388 #  ifdef SPARC64
 389                 if (current_wordsize == 8)
 390                         return;
 391 #  endif
 392                 set_tcb_priv_ulong(tcp, version);
 393                 call &= 0xffff;
 394 # endif
 395         }
 396
 397         switch (call) {
 398                 case  1: case  2: case  3: case  4:
 399                 case 11: case 12: case 13: case 14:
 400                 case 21: case 22: case 23: case 24:
 401                         break;
 402                 default:
 403                         return;
 404         }
 405
 406         tcp->scno = SYS_ipc_subcall + call;
 407         tcp->qual_flg = qual_flags(tcp->scno);
 408         tcp->s_ent = &sysent[tcp->scno];
 409
 410         const unsigned int n = tcp->s_ent->nargs;
 411         unsigned int i;
 412         for (i = 0; i < n; i++)
 413                 tcp->u_arg[i] = tcp->u_arg[i + 1];
 414 }
 415 #endif
 416
 417 #ifdef LINUX_MIPSO32
 418 static void
 419 decode_mips_subcall(struct tcb *tcp)
 420 {
 421         if (!scno_is_valid(tcp->u_arg[0]))
 422                 return;
 423         tcp->scno = tcp->u_arg[0];
 424         tcp->qual_flg = qual_flags(tcp->scno);
 425         tcp->s_ent = &sysent[tcp->scno];
 426         memmove(&tcp->u_arg[0], &tcp->u_arg[1],
 427                 sizeof(tcp->u_arg) - sizeof(tcp->u_arg[0]));
 428         /*
 429          * Fetching the last arg of 7-arg syscalls (fadvise64_64
 430          * and sync_file_range) requires additional code,
 431          * see linux/mips/get_syscall_args.c
 432          */
 433         if (tcp->s_ent->nargs == MAX_ARGS) {
 434                 if (umoven(tcp,
 435                            mips_REG_SP + MAX_ARGS * sizeof(tcp->u_arg[0]),
 436                            sizeof(tcp->u_arg[0]),
 437                            &tcp->u_arg[MAX_ARGS - 1]) < 0)
 438                 tcp->u_arg[MAX_ARGS - 1] = 0;
 439         }
 440 }
 441 #endif /* LINUX_MIPSO32 */
 442
 443 static void
 444 dumpio(struct tcb *tcp)
 445 {
 446         if (syserror(tcp))
 447                 return;
 448
 449         int fd = tcp->u_arg[0];
 450         if (fd < 0)
 451                 return;
 452
 453         if (is_number_in_set(fd, &read_set)) {
 454                 switch (tcp->s_ent->sen) {
 455                 case SEN_read:
 456                 case SEN_pread:
 457                 case SEN_recv:
 458                 case SEN_recvfrom:
 459                 case SEN_mq_timedreceive:
 460                         dumpstr(tcp, tcp->u_arg[1], tcp->u_rval);
 461                         return;
 462                 case SEN_readv:
 463                 case SEN_preadv:
 464                 case SEN_preadv2:
 465                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1],
 466                                      tcp->u_rval);
 467                         return;
 468                 case SEN_recvmsg:
 469                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], tcp->u_rval);
 470                         return;
 471                 case SEN_recvmmsg:
 472                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 473                         return;
 474                 }
 475         }
 476         if (is_number_in_set(fd, &write_set)) {
 477                 switch (tcp->s_ent->sen) {
 478                 case SEN_write:
 479                 case SEN_pwrite:
 480                 case SEN_send:
 481                 case SEN_sendto:
 482                 case SEN_mq_timedsend:
 483                         dumpstr(tcp, tcp->u_arg[1], tcp->u_arg[2]);
 484                         break;
 485                 case SEN_writev:
 486                 case SEN_pwritev:
 487                 case SEN_pwritev2:
 488                 case SEN_vmsplice:
 489                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1], -1);
 490                         break;
 491                 case SEN_sendmsg:
 492                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], -1);
 493                         break;
 494                 case SEN_sendmmsg:
 495                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 496                         break;
 497                 }
 498         }
 499 }
 500
 501 /*
 502  * Shuffle syscall numbers so that we don't have huge gaps in syscall table.
 503  * The shuffling should be an involution: shuffle_scno(shuffle_scno(n)) == n.
 504  */
 505 static kernel_ulong_t
 506 shuffle_scno(kernel_ulong_t scno)
 507 {
 508 #if defined(ARM) || defined(AARCH64) /* So far only 32-bit ARM needs this */
 509         if (scno < ARM_FIRST_SHUFFLED_SYSCALL)
 510                 return scno;
 511
 512         /* __ARM_NR_cmpxchg? Swap with LAST_ORDINARY+1 */
 513         if (scno == ARM_FIRST_SHUFFLED_SYSCALL)
 514                 return 0x000ffff0;
 515         if (scno == 0x000ffff0)
 516                 return ARM_FIRST_SHUFFLED_SYSCALL;
 517
 518 # define ARM_SECOND_SHUFFLED_SYSCALL (ARM_FIRST_SHUFFLED_SYSCALL + 1)
 519         /*
 520          * Is it ARM specific syscall?
 521          * Swap [0x000f0000, 0x000f0000 + LAST_SPECIAL] range
 522          * with [SECOND_SHUFFLED, SECOND_SHUFFLED + LAST_SPECIAL] range.
 523          */
 524         if (scno >= 0x000f0000 &&
 525             scno <= 0x000f0000 + ARM_LAST_SPECIAL_SYSCALL) {
 526                 return scno - 0x000f0000 + ARM_SECOND_SHUFFLED_SYSCALL;
 527         }
 528         if (scno <= ARM_SECOND_SHUFFLED_SYSCALL + ARM_LAST_SPECIAL_SYSCALL) {
 529                 return scno + 0x000f0000 - ARM_SECOND_SHUFFLED_SYSCALL;
 530         }
 531 #endif /* ARM || AARCH64 */
 532
 533         return scno;
 534 }
 535
 536 const char *
 537 err_name(unsigned long err)
 538 {
 539         if ((err < nerrnos) && errnoent[err])
 540                 return errnoent[err];
 541
 542         return NULL;
 543 }
 544
 545 static long get_regs_error;
 546
 547 void
 548 clear_regs(void)
 549 {
 550         get_regs_error = -1;
 551 }
 552
 553 static int get_syscall_args(struct tcb *);
 554 static int get_syscall_result(struct tcb *);
 555 static int arch_get_scno(struct tcb *tcp);
 556 static int arch_set_scno(struct tcb *, kernel_ulong_t);
 557 static void get_error(struct tcb *, const bool);
 558 static int arch_set_error(struct tcb *);
 559
 560 struct fault_opts *fault_vec[SUPPORTED_PERSONALITIES];
 561
 562 static struct fault_opts *
 563 tcb_fault_opts(struct tcb *tcp)
 564 {
 565         return (scno_in_range(tcp->scno) && tcp->fault_vec[current_personality])
 566                ? &tcp->fault_vec[current_personality][tcp->scno] : NULL;
 567 }
 568
 569
 570 static long
 571 inject_syscall_fault_entering(struct tcb *tcp)
 572 {
 573         if (!tcp->fault_vec[current_personality]) {
 574                 tcp->fault_vec[current_personality] =
 575                         xcalloc(nsyscalls, sizeof(**fault_vec));
 576                 memcpy(tcp->fault_vec[current_personality],
 577                        fault_vec[current_personality],
 578                        nsyscalls * sizeof(**fault_vec));
 579         }
 580
 581         struct fault_opts *opts = tcb_fault_opts(tcp);
 582
 583         if (!opts || opts->first == 0)
 584                 return 0;
 585
 586         --opts->first;
 587
 588         if (opts->first != 0)
 589                 return 0;
 590
 591         opts->first = opts->step;
 592
 593         if (!arch_set_scno(tcp, -1))
 594                 tcp->flags |= TCB_FAULT_INJ;
 595
 596         return 0;
 597 }
 598
 599 static long
 600 update_syscall_fault_exiting(struct tcb *tcp)
 601 {
 602         struct fault_opts *opts = tcb_fault_opts(tcp);
 603
 604         if (opts && opts->err && tcp->u_error != opts->err) {
 605                 unsigned long u_error = tcp->u_error;
 606                 tcp->u_error = opts->err;
 607                 if (arch_set_error(tcp))
 608                         tcp->u_error = u_error;
 609         }
 610
 611         return 0;
 612 }
 613
 614 static int
 615 trace_syscall_entering(struct tcb *tcp)
 616 {
 617         int res, scno_good;
 618
 619         scno_good = res = get_scno(tcp);
 620         if (res == 0)
 621                 return res;
 622         if (res == 1)
 623                 res = get_syscall_args(tcp);
 624
 625         if (res != 1) {
 626                 printleader(tcp);
 627                 tprintf("%s(", scno_good == 1 ? tcp->s_ent->sys_name : "????");
 628                 /*
 629                  * " <unavailable>" will be added later by the code which
 630                  * detects ptrace errors.
 631                  */
 632                 goto ret;
 633         }
 634
 635 #ifdef LINUX_MIPSO32
 636         if (SEN_syscall == tcp->s_ent->sen)
 637                 decode_mips_subcall(tcp);
 638 #endif
 639
 640 #if defined(SYS_socket_subcall) || defined(SYS_ipc_subcall)
 641         switch (tcp->s_ent->sen) {
 642 # ifdef SYS_socket_subcall
 643                 case SEN_socketcall:
 644                         decode_socket_subcall(tcp);
 645                         break;
 646 # endif
 647 # ifdef SYS_ipc_subcall
 648                 case SEN_ipc:
 649                         decode_ipc_subcall(tcp);
 650                         break;
 651 # endif
 652         }
 653 #endif
 654
 655         /* Restrain from fault injection while the trace executes strace code. */
 656         if (hide_log(tcp)) {
 657                 tcp->qual_flg &= ~QUAL_FAULT;
 658         }
 659
 660         switch (tcp->s_ent->sen) {
 661                 case SEN_execve:
 662                 case SEN_execveat:
 663 #if defined SPARC || defined SPARC64
 664                 case SEN_execv:
 665 #endif
 666                         tcp->flags &= ~TCB_HIDE_LOG;
 667                         break;
 668         }
 669
 670         if (!(tcp->qual_flg & QUAL_TRACE)
 671          || (tracing_paths && !pathtrace_match(tcp))
 672         ) {
 673                 tcp->flags |= TCB_INSYSCALL | TCB_FILTERED;
 674                 tcp->sys_func_rval = 0;
 675                 return 0;
 676         }
 677
 678         tcp->flags &= ~TCB_FILTERED;
 679
 680         if (hide_log(tcp)) {
 681                 res = 0;
 682                 goto ret;
 683         }
 684
 685         if (tcp->qual_flg & QUAL_FAULT)
 686                 inject_syscall_fault_entering(tcp);
 687
 688         if (cflag == CFLAG_ONLY_STATS) {
 689                 res = 0;
 690                 goto ret;
 691         }
 692
 693 #ifdef USE_LIBUNWIND
 694         if (stack_trace_enabled) {
 695                 if (tcp->s_ent->sys_flags & STACKTRACE_CAPTURE_ON_ENTER)
 696                         unwind_capture_stacktrace(tcp);
 697         }
 698 #endif
 699
 700         printleader(tcp);
 701         tprintf("%s(", tcp->s_ent->sys_name);
 702         if (tcp->qual_flg & QUAL_RAW)
 703                 res = printargs(tcp);
 704         else
 705                 res = tcp->s_ent->sys_func(tcp);
 706
 707         fflush(tcp->outf);
 708  ret:
 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)
 713                 gettimeofday(&tcp->etime, NULL);
 714         return res;
 715 }
 716
 717 static bool
 718 syscall_fault_injected(struct tcb *tcp)
 719 {
 720         return tcp->flags & TCB_FAULT_INJ;
 721 }
 722
 723 static int
 724 trace_syscall_exiting(struct tcb *tcp)
 725 {
 726         int sys_res;
 727         struct timeval tv;
 728         int res;
 729         unsigned long u_error;
 730         const char *u_error_str;
 731
 732         /* Measure the exit time as early as possible to avoid errors. */
 733         if (Tflag || cflag)
 734                 gettimeofday(&tv, NULL);
 735
 736 #ifdef USE_LIBUNWIND
 737         if (stack_trace_enabled) {
 738                 if (tcp->s_ent->sys_flags & STACKTRACE_INVALIDATE_CACHE)
 739                         unwind_cache_invalidate(tcp);
 740         }
 741 #endif
 742
 743 #if SUPPORTED_PERSONALITIES > 1
 744         update_personality(tcp, tcp->currpers);
 745 #endif
 746         res = (get_regs_error ? -1 : get_syscall_result(tcp));
 747         if (filtered(tcp) || hide_log(tcp))
 748                 goto ret;
 749
 750         if (syserror(tcp) && syscall_fault_injected(tcp))
 751                 update_syscall_fault_exiting(tcp);
 752
 753         if (cflag) {
 754                 count_syscall(tcp, &tv);
 755                 if (cflag == CFLAG_ONLY_STATS) {
 756                         goto ret;
 757                 }
 758         }
 759
 760         /* If not in -ff mode, and printing_tcp != tcp,
 761          * then the log currently does not end with output
 762          * of _our syscall entry_, but with something else.
 763          * We need to say which syscall's return is this.
 764          *
 765          * Forced reprinting via TCB_REPRINT is used only by
 766          * "strace -ff -oLOG test/threaded_execve" corner case.
 767          * It's the only case when -ff mode needs reprinting.
 768          */
 769         if ((followfork < 2 && printing_tcp != tcp) || (tcp->flags & TCB_REPRINT)) {
 770                 tcp->flags &= ~TCB_REPRINT;
 771                 printleader(tcp);
 772                 tprintf("<... %s resumed> ", tcp->s_ent->sys_name);
 773         }
 774         printing_tcp = tcp;
 775
 776         tcp->s_prev_ent = NULL;
 777         if (res != 1) {
 778                 /* There was error in one of prior ptrace ops */
 779                 tprints(") ");
 780                 tabto();
 781                 tprints("= ? <unavailable>\n");
 782                 line_ended();
 783                 tcp->flags &= ~(TCB_INSYSCALL | TCB_FAULT_INJ);
 784                 tcp->sys_func_rval = 0;
 785                 free_tcb_priv_data(tcp);
 786                 return res;
 787         }
 788         tcp->s_prev_ent = tcp->s_ent;
 789
 790         sys_res = 0;
 791         if (tcp->qual_flg & QUAL_RAW) {
 792                 /* sys_res = printargs(tcp); - but it's nop on sysexit */
 793         } else {
 794         /* FIXME: not_failing_only (IOW, option -z) is broken:
 795          * failure of syscall is known only after syscall return.
 796          * Thus we end up with something like this on, say, ENOENT:
 797          *     open("doesnt_exist", O_RDONLY <unfinished ...>
 798          *     {next syscall decode}
 799          * whereas the intended result is that open(...) line
 800          * is not shown at all.
 801          */
 802                 if (not_failing_only && tcp->u_error)
 803                         goto ret;       /* ignore failed syscalls */
 804                 if (tcp->sys_func_rval & RVAL_DECODED)
 805                         sys_res = tcp->sys_func_rval;
 806                 else
 807                         sys_res = tcp->s_ent->sys_func(tcp);
 808         }
 809
 810         tprints(") ");
 811         tabto();
 812         u_error = tcp->u_error;
 813
 814         if (tcp->qual_flg & QUAL_RAW) {
 815                 if (u_error) {
 816                         tprintf("= -1 (errno %lu)", u_error);
 817                         if (syscall_fault_injected(tcp))
 818                                 tprints(" (INJECTED)");
 819                 } else {
 820                         tprintf("= %#" PRI_klx, tcp->u_rval);
 821                 }
 822         }
 823         else if (!(sys_res & RVAL_NONE) && u_error) {
 824                 switch (u_error) {
 825                 /* Blocked signals do not interrupt any syscalls.
 826                  * In this case syscalls don't return ERESTARTfoo codes.
 827                  *
 828                  * Deadly signals set to SIG_DFL interrupt syscalls
 829                  * and kill the process regardless of which of the codes below
 830                  * is returned by the interrupted syscall.
 831                  * In some cases, kernel forces a kernel-generated deadly
 832                  * signal to be unblocked and set to SIG_DFL (and thus cause
 833                  * death) if it is blocked or SIG_IGNed: for example, SIGSEGV
 834                  * or SIGILL. (The alternative is to leave process spinning
 835                  * forever on the faulty instruction - not useful).
 836                  *
 837                  * SIG_IGNed signals and non-deadly signals set to SIG_DFL
 838                  * (for example, SIGCHLD, SIGWINCH) interrupt syscalls,
 839                  * but kernel will always restart them.
 840                  */
 841                 case ERESTARTSYS:
 842                         /* Most common type of signal-interrupted syscall exit code.
 843                          * The system call will be restarted with the same arguments
 844                          * if SA_RESTART is set; otherwise, it will fail with EINTR.
 845                          */
 846                         tprints("= ? ERESTARTSYS (To be restarted if SA_RESTART is set)");
 847                         break;
 848                 case ERESTARTNOINTR:
 849                         /* Rare. For example, fork() returns this if interrupted.
 850                          * SA_RESTART is ignored (assumed set): the restart is unconditional.
 851                          */
 852                         tprints("= ? ERESTARTNOINTR (To be restarted)");
 853                         break;
 854                 case ERESTARTNOHAND:
 855                         /* pause(), rt_sigsuspend() etc use this code.
 856                          * SA_RESTART is ignored (assumed not set):
 857                          * syscall won't restart (will return EINTR instead)
 858                          * even after signal with SA_RESTART set. However,
 859                          * after SIG_IGN or SIG_DFL signal it will restart
 860                          * (thus the name "restart only if has no handler").
 861                          */
 862                         tprints("= ? ERESTARTNOHAND (To be restarted if no handler)");
 863                         break;
 864                 case ERESTART_RESTARTBLOCK:
 865                         /* Syscalls like nanosleep(), poll() which can't be
 866                          * restarted with their original arguments use this
 867                          * code. Kernel will execute restart_syscall() instead,
 868                          * which changes arguments before restarting syscall.
 869                          * SA_RESTART is ignored (assumed not set) similarly
 870                          * to ERESTARTNOHAND. (Kernel can't honor SA_RESTART
 871                          * since restart data is saved in "restart block"
 872                          * in task struct, and if signal handler uses a syscall
 873                          * which in turn saves another such restart block,
 874                          * old data is lost and restart becomes impossible)
 875                          */
 876                         tprints("= ? ERESTART_RESTARTBLOCK (Interrupted by signal)");
 877                         break;
 878                 default:
 879                         u_error_str = err_name(u_error);
 880                         if (u_error_str)
 881                                 tprintf("= -1 %s (%s)",
 882                                         u_error_str, strerror(u_error));
 883                         else
 884                                 tprintf("= -1 %lu (%s)",
 885                                         u_error, strerror(u_error));
 886                         break;
 887                 }
 888                 if (syscall_fault_injected(tcp))
 889                         tprintf(" (INJECTED)");
 890                 if ((sys_res & RVAL_STR) && tcp->auxstr)
 891                         tprintf(" (%s)", tcp->auxstr);
 892         }
 893         else {
 894                 if (sys_res & RVAL_NONE)
 895                         tprints("= ?");
 896                 else {
 897                         switch (sys_res & RVAL_MASK) {
 898                         case RVAL_HEX:
 899 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
 900                                 if (current_wordsize < sizeof(tcp->u_rval)) {
 901                                         tprintf("= %#x",
 902                                                 (unsigned int) tcp->u_rval);
 903                                 } else
 904 #endif
 905                                 {
 906                                         tprintf("= %#" PRI_klx, tcp->u_rval);
 907                                 }
 908                                 break;
 909                         case RVAL_OCTAL:
 910                                 tprints("= ");
 911                                 print_numeric_long_umask(tcp->u_rval);
 912                                 break;
 913                         case RVAL_UDECIMAL:
 914 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
 915                                 if (current_wordsize < sizeof(tcp->u_rval)) {
 916                                         tprintf("= %u",
 917                                                 (unsigned int) tcp->u_rval);
 918                                 } else
 919 #endif
 920                                 {
 921                                         tprintf("= %" PRI_klu, tcp->u_rval);
 922                                 }
 923                                 break;
 924                         case RVAL_DECIMAL:
 925                                 tprintf("= %" PRI_kld, tcp->u_rval);
 926                                 break;
 927                         case RVAL_FD:
 928                                 if (show_fd_path) {
 929                                         tprints("= ");
 930                                         printfd(tcp, tcp->u_rval);
 931                                 }
 932                                 else
 933                                         tprintf("= %" PRI_kld, tcp->u_rval);
 934                                 break;
 935                         default:
 936                                 error_msg("invalid rval format");
 937                                 break;
 938                         }
 939                 }
 940                 if ((sys_res & RVAL_STR) && tcp->auxstr)
 941                         tprintf(" (%s)", tcp->auxstr);
 942         }
 943         if (Tflag) {
 944                 tv_sub(&tv, &tv, &tcp->etime);
 945                 tprintf(" <%ld.%06ld>",
 946                         (long) tv.tv_sec, (long) tv.tv_usec);
 947         }
 948         tprints("\n");
 949         dumpio(tcp);
 950         line_ended();
 951
 952 #ifdef USE_LIBUNWIND
 953         if (stack_trace_enabled)
 954                 unwind_print_stacktrace(tcp);
 955 #endif
 956
 957  ret:
 958         tcp->flags &= ~(TCB_INSYSCALL | TCB_FAULT_INJ);
 959         tcp->sys_func_rval = 0;
 960         free_tcb_priv_data(tcp);
 961         return 0;
 962 }
 963
 964 int
 965 trace_syscall(struct tcb *tcp)
 966 {
 967         return exiting(tcp) ?
 968                 trace_syscall_exiting(tcp) : trace_syscall_entering(tcp);
 969 }
 970
 971 bool
 972 is_erestart(struct tcb *tcp)
 973 {
 974         switch (tcp->u_error) {
 975                 case ERESTARTSYS:
 976                 case ERESTARTNOINTR:
 977                 case ERESTARTNOHAND:
 978                 case ERESTART_RESTARTBLOCK:
 979                         return true;
 980                 default:
 981                         return false;
 982         }
 983 }
 984
 985 static unsigned long saved_u_error;
 986
 987 void
 988 temporarily_clear_syserror(struct tcb *tcp)
 989 {
 990         saved_u_error = tcp->u_error;
 991         tcp->u_error = 0;
 992 }
 993
 994 void
 995 restore_cleared_syserror(struct tcb *tcp)
 996 {
 997         tcp->u_error = saved_u_error;
 998 }
 999
1000 /*
1001  * Check the syscall return value register value for whether it is
1002  * a negated errno code indicating an error, or a success return value.
1003  */
1004 static inline bool
1005 is_negated_errno(kernel_ulong_t val)
1006 {
1007         /* Linux kernel defines MAX_ERRNO to 4095. */
1008         kernel_ulong_t max = -(kernel_long_t) 4095;
1009
1010 #ifndef current_klongsize
1011         if (current_klongsize < sizeof(val)) {
1012                 val = (uint32_t) val;
1013                 max = (uint32_t) max;
1014         }
1015 #endif /* !current_klongsize */
1016
1017         return val >= max;
1018 }
1019
1020 #include "arch_regs.c"
1021
1022 #ifdef HAVE_GETRVAL2
1023 # include "arch_getrval2.c"
1024 #endif
1025
1026 void
1027 print_pc(struct tcb *tcp)
1028 {
1029 #if defined ARCH_PC_REG
1030 # define ARCH_GET_PC 0
1031 #elif defined ARCH_PC_PEEK_ADDR
1032         kernel_ulong_t pc;
1033 # define ARCH_PC_REG pc
1034 # define ARCH_GET_PC upeek(tcp->pid, ARCH_PC_PEEK_ADDR, &pc)
1035 #else
1036 # error Neither ARCH_PC_REG nor ARCH_PC_PEEK_ADDR is defined
1037 #endif
1038         if (get_regs_error || ARCH_GET_PC)
1039                 tprints(current_wordsize == 4 ? "[????????] "
1040                                               : "[????????????????] ");
1041         else
1042                 tprintf(current_wordsize == 4
1043                         ? "[%08" PRI_klx "] " : "[%016" PRI_klx "] ",
1044                         (kernel_ulong_t) ARCH_PC_REG);
1045 }
1046
1047 #include "getregs_old.h"
1048
1049 #undef ptrace_getregset_or_getregs
1050 #undef ptrace_setregset_or_setregs
1051 #ifdef ARCH_REGS_FOR_GETREGSET
1052
1053 # define ptrace_getregset_or_getregs ptrace_getregset
1054 static long
1055 ptrace_getregset(pid_t pid)
1056 {
1057 # ifdef ARCH_IOVEC_FOR_GETREGSET
1058         /* variable iovec */
1059         ARCH_IOVEC_FOR_GETREGSET.iov_len = sizeof(ARCH_REGS_FOR_GETREGSET);
1060         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS,
1061                       &ARCH_IOVEC_FOR_GETREGSET);
1062 # else
1063         /* constant iovec */
1064         static struct iovec io = {
1065                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1066                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1067         };
1068         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS, &io);
1069
1070 # endif
1071 }
1072
1073 # ifndef HAVE_GETREGS_OLD
1074 #  define ptrace_setregset_or_setregs ptrace_setregset
1075 static int
1076 ptrace_setregset(pid_t pid)
1077 {
1078 #  ifdef ARCH_IOVEC_FOR_GETREGSET
1079         /* variable iovec */
1080         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS,
1081                       &ARCH_IOVEC_FOR_GETREGSET);
1082 #  else
1083         /* constant iovec */
1084         static struct iovec io = {
1085                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1086                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1087         };
1088         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS, &io);
1089 #  endif
1090 }
1091 # endif /* !HAVE_GETREGS_OLD */
1092
1093 #elif defined ARCH_REGS_FOR_GETREGS
1094
1095 # define ptrace_getregset_or_getregs ptrace_getregs
1096 static long
1097 ptrace_getregs(pid_t pid)
1098 {
1099 # if defined SPARC || defined SPARC64
1100         /* SPARC systems have the meaning of data and addr reversed */
1101         return ptrace(PTRACE_GETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1102 # else
1103         return ptrace(PTRACE_GETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1104 # endif
1105 }
1106
1107 # ifndef HAVE_GETREGS_OLD
1108 #  define ptrace_setregset_or_setregs ptrace_setregs
1109 static int
1110 ptrace_setregs(pid_t pid)
1111 {
1112 #  if defined SPARC || defined SPARC64
1113         /* SPARC systems have the meaning of data and addr reversed */
1114         return ptrace(PTRACE_SETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1115 #  else
1116         return ptrace(PTRACE_SETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1117 #  endif
1118 }
1119 # endif /* !HAVE_GETREGS_OLD */
1120
1121 #endif /* ARCH_REGS_FOR_GETREGSET || ARCH_REGS_FOR_GETREGS */
1122
1123 void
1124 get_regs(pid_t pid)
1125 {
1126 #undef USE_GET_SYSCALL_RESULT_REGS
1127 #ifdef ptrace_getregset_or_getregs
1128
1129 # ifdef HAVE_GETREGS_OLD
1130         /*
1131          * Try PTRACE_GETREGSET/PTRACE_GETREGS first,
1132          * fallback to getregs_old.
1133          */
1134         static int use_getregs_old;
1135         if (use_getregs_old < 0) {
1136                 get_regs_error = ptrace_getregset_or_getregs(pid);
1137                 return;
1138         } else if (use_getregs_old == 0) {
1139                 get_regs_error = ptrace_getregset_or_getregs(pid);
1140                 if (get_regs_error >= 0) {
1141                         use_getregs_old = -1;
1142                         return;
1143                 }
1144                 if (errno == EPERM || errno == ESRCH)
1145                         return;
1146                 use_getregs_old = 1;
1147         }
1148         get_regs_error = getregs_old(pid);
1149 # else /* !HAVE_GETREGS_OLD */
1150         /* Assume that PTRACE_GETREGSET/PTRACE_GETREGS works. */
1151         get_regs_error = ptrace_getregset_or_getregs(pid);
1152 # endif /* !HAVE_GETREGS_OLD */
1153
1154 #else /* !ptrace_getregset_or_getregs */
1155
1156 # define USE_GET_SYSCALL_RESULT_REGS 1
1157 # warning get_regs is not implemented for this architecture yet
1158         get_regs_error = 0;
1159
1160 #endif /* !ptrace_getregset_or_getregs */
1161 }
1162
1163 #ifdef ptrace_setregset_or_setregs
1164 static int
1165 set_regs(pid_t pid)
1166 {
1167         return ptrace_setregset_or_setregs(pid);
1168 }
1169 #endif /* ptrace_setregset_or_setregs */
1170
1171 struct sysent_buf {
1172         struct tcb *tcp;
1173         struct_sysent ent;
1174         char buf[sizeof("syscall_%lu") + sizeof(kernel_ulong_t) * 3];
1175 };
1176
1177 static void
1178 free_sysent_buf(void *ptr)
1179 {
1180         struct sysent_buf *s = ptr;
1181         s->tcp->s_prev_ent = s->tcp->s_ent = NULL;
1182         free(ptr);
1183 }
1184
1185 /*
1186  * Returns:
1187  * 0: "ignore this ptrace stop", bail out of trace_syscall_entering() silently.
1188  * 1: ok, continue in trace_syscall_entering().
1189  * other: error, trace_syscall_entering() should print error indicator
1190  *    ("????" etc) and bail out.
1191  */
1192 int
1193 get_scno(struct tcb *tcp)
1194 {
1195         if (get_regs_error)
1196                 return -1;
1197
1198         int rc = arch_get_scno(tcp);
1199         if (rc != 1)
1200                 return rc;
1201
1202         if (scno_is_valid(tcp->scno)) {
1203                 tcp->s_ent = &sysent[tcp->scno];
1204                 tcp->qual_flg = qual_flags(tcp->scno);
1205         } else {
1206                 struct sysent_buf *s = xcalloc(1, sizeof(*s));
1207
1208                 s->tcp = tcp;
1209                 s->ent.nargs = MAX_ARGS;
1210                 s->ent.sen = SEN_printargs;
1211                 s->ent.sys_func = printargs;
1212                 s->ent.sys_name = s->buf;
1213                 sprintf(s->buf, "syscall_%" PRI_klu, shuffle_scno(tcp->scno));
1214
1215                 tcp->s_ent = &s->ent;
1216                 tcp->qual_flg = QUAL_RAW | DEFAULT_QUAL_FLAGS;
1217
1218                 set_tcb_priv_data(tcp, s, free_sysent_buf);
1219
1220                 if (debug_flag)
1221                         error_msg("pid %d invalid syscall %" PRI_kld,
1222                                   tcp->pid, tcp->scno);
1223         }
1224         return 1;
1225 }
1226
1227 #ifdef USE_GET_SYSCALL_RESULT_REGS
1228 static int get_syscall_result_regs(struct tcb *);
1229 #endif
1230
1231 /* Returns:
1232  * 1: ok, continue in trace_syscall_exiting().
1233  * -1: error, trace_syscall_exiting() should print error indicator
1234  *    ("????" etc) and bail out.
1235  */
1236 static int
1237 get_syscall_result(struct tcb *tcp)
1238 {
1239 #ifdef USE_GET_SYSCALL_RESULT_REGS
1240         if (get_syscall_result_regs(tcp))
1241                 return -1;
1242 #endif
1243         tcp->u_error = 0;
1244         get_error(tcp, !(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS));
1245
1246         return 1;
1247 }
1248
1249 #include "get_scno.c"
1250 #include "set_scno.c"
1251 #include "get_syscall_args.c"
1252 #ifdef USE_GET_SYSCALL_RESULT_REGS
1253 # include "get_syscall_result.c"
1254 #endif
1255 #include "get_error.c"
1256 #include "set_error.c"
1257 #ifdef HAVE_GETREGS_OLD
1258 # include "getregs_old.c"
1259 #endif
1260
1261 const char *
1262 syscall_name(kernel_ulong_t scno)
1263 {
1264 #if defined X32_PERSONALITY_NUMBER && defined __X32_SYSCALL_BIT
1265         if (current_personality == X32_PERSONALITY_NUMBER)
1266                 scno &= ~__X32_SYSCALL_BIT;
1267 #endif
1268         return scno_is_valid(scno) ? sysent[scno].sys_name: NULL;
1269 }