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 <sys/param.h>
  37 #include <signal.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 #ifndef NSIG
  67 # warning: NSIG is not defined, using 32
  68 # define NSIG 32
  69 #endif
  70
  71 #include "syscall.h"
  72
  73 /* Define these shorthand notations to simplify the syscallent files. */
  74 #define TD TRACE_DESC
  75 #define TF TRACE_FILE
  76 #define TI TRACE_IPC
  77 #define TN TRACE_NETWORK
  78 #define TP TRACE_PROCESS
  79 #define TS TRACE_SIGNAL
  80 #define TM TRACE_MEMORY
  81 #define NF SYSCALL_NEVER_FAILS
  82 #define MA MAX_ARGS
  83 #define SI STACKTRACE_INVALIDATE_CACHE
  84 #define SE STACKTRACE_CAPTURE_ON_ENTER
  85
  86 #define SEN(syscall_name) SEN_ ## syscall_name, SYS_FUNC_NAME(sys_ ## syscall_name)
  87
  88 const struct_sysent sysent0[] = {
  89 #include "syscallent.h"
  90 };
  91
  92 #if SUPPORTED_PERSONALITIES > 1
  93 # include PERSONALITY1_INCLUDE_FUNCS
  94 static const struct_sysent sysent1[] = {
  95 # include "syscallent1.h"
  96 };
  97 #endif
  98
  99 #if SUPPORTED_PERSONALITIES > 2
 100 # include PERSONALITY2_INCLUDE_FUNCS
 101 static const struct_sysent sysent2[] = {
 102 # include "syscallent2.h"
 103 };
 104 #endif
 105
 106 /* Now undef them since short defines cause wicked namespace pollution. */
 107 #undef SEN
 108 #undef TD
 109 #undef TF
 110 #undef TI
 111 #undef TN
 112 #undef TP
 113 #undef TS
 114 #undef TM
 115 #undef NF
 116 #undef MA
 117 #undef SI
 118 #undef SE
 119
 120 /*
 121  * `ioctlent[012].h' files are automatically generated by the auxiliary
 122  * program `ioctlsort', such that the list is sorted by the `code' field.
 123  * This has the side-effect of resolving the _IO.. macros into
 124  * plain integers, eliminating the need to include here everything
 125  * in "/usr/include".
 126  */
 127
 128 const char *const errnoent0[] = {
 129 #include "errnoent.h"
 130 };
 131 const char *const signalent0[] = {
 132 #include "signalent.h"
 133 };
 134 const struct_ioctlent ioctlent0[] = {
 135 #include "ioctlent0.h"
 136 };
 137
 138 #if SUPPORTED_PERSONALITIES > 1
 139 static const char *const errnoent1[] = {
 140 # include "errnoent1.h"
 141 };
 142 static const char *const signalent1[] = {
 143 # include "signalent1.h"
 144 };
 145 static const struct_ioctlent ioctlent1[] = {
 146 # include "ioctlent1.h"
 147 };
 148 # include PERSONALITY0_INCLUDE_PRINTERS_DECLS
 149 static const struct_printers printers0 = {
 150 # include PERSONALITY0_INCLUDE_PRINTERS_DEFS
 151 };
 152 # include PERSONALITY1_INCLUDE_PRINTERS_DECLS
 153 static const struct_printers printers1 = {
 154 # include PERSONALITY1_INCLUDE_PRINTERS_DEFS
 155 };
 156 #endif
 157
 158 #if SUPPORTED_PERSONALITIES > 2
 159 static const char *const errnoent2[] = {
 160 # include "errnoent2.h"
 161 };
 162 static const char *const signalent2[] = {
 163 # include "signalent2.h"
 164 };
 165 static const struct_ioctlent ioctlent2[] = {
 166 # include "ioctlent2.h"
 167 };
 168 # include PERSONALITY2_INCLUDE_PRINTERS_DECLS
 169 static const struct_printers printers2 = {
 170 # include PERSONALITY2_INCLUDE_PRINTERS_DEFS
 171 };
 172 #endif
 173
 174 enum {
 175         nsyscalls0 = ARRAY_SIZE(sysent0)
 176 #if SUPPORTED_PERSONALITIES > 1
 177         , nsyscalls1 = ARRAY_SIZE(sysent1)
 178 # if SUPPORTED_PERSONALITIES > 2
 179         , nsyscalls2 = ARRAY_SIZE(sysent2)
 180 # endif
 181 #endif
 182 };
 183
 184 enum {
 185         nerrnos0 = ARRAY_SIZE(errnoent0)
 186 #if SUPPORTED_PERSONALITIES > 1
 187         , nerrnos1 = ARRAY_SIZE(errnoent1)
 188 # if SUPPORTED_PERSONALITIES > 2
 189         , nerrnos2 = ARRAY_SIZE(errnoent2)
 190 # endif
 191 #endif
 192 };
 193
 194 enum {
 195         nsignals0 = ARRAY_SIZE(signalent0)
 196 #if SUPPORTED_PERSONALITIES > 1
 197         , nsignals1 = ARRAY_SIZE(signalent1)
 198 # if SUPPORTED_PERSONALITIES > 2
 199         , nsignals2 = ARRAY_SIZE(signalent2)
 200 # endif
 201 #endif
 202 };
 203
 204 enum {
 205         nioctlents0 = ARRAY_SIZE(ioctlent0)
 206 #if SUPPORTED_PERSONALITIES > 1
 207         , nioctlents1 = ARRAY_SIZE(ioctlent1)
 208 # if SUPPORTED_PERSONALITIES > 2
 209         , nioctlents2 = ARRAY_SIZE(ioctlent2)
 210 # endif
 211 #endif
 212 };
 213
 214 #if SUPPORTED_PERSONALITIES > 1
 215 const struct_sysent *sysent = sysent0;
 216 const char *const *errnoent = errnoent0;
 217 const char *const *signalent = signalent0;
 218 const struct_ioctlent *ioctlent = ioctlent0;
 219 const struct_printers *printers = &printers0;
 220 #endif
 221
 222 unsigned nsyscalls = nsyscalls0;
 223 unsigned nerrnos = nerrnos0;
 224 unsigned nsignals = nsignals0;
 225 unsigned nioctlents = nioctlents0;
 226
 227 const unsigned int nsyscall_vec[SUPPORTED_PERSONALITIES] = {
 228         nsyscalls0,
 229 #if SUPPORTED_PERSONALITIES > 1
 230         nsyscalls1,
 231 #endif
 232 #if SUPPORTED_PERSONALITIES > 2
 233         nsyscalls2,
 234 #endif
 235 };
 236 const struct_sysent *const sysent_vec[SUPPORTED_PERSONALITIES] = {
 237         sysent0,
 238 #if SUPPORTED_PERSONALITIES > 1
 239         sysent1,
 240 #endif
 241 #if SUPPORTED_PERSONALITIES > 2
 242         sysent2,
 243 #endif
 244 };
 245
 246 #if SUPPORTED_PERSONALITIES > 1
 247 unsigned current_personality;
 248
 249 # ifndef current_wordsize
 250 unsigned current_wordsize;
 251 static const int personality_wordsize[SUPPORTED_PERSONALITIES] = {
 252         PERSONALITY0_WORDSIZE,
 253         PERSONALITY1_WORDSIZE,
 254 # if SUPPORTED_PERSONALITIES > 2
 255         PERSONALITY2_WORDSIZE,
 256 # endif
 257 };
 258 # endif
 259
 260 void
 261 set_personality(int personality)
 262 {
 263         nsyscalls = nsyscall_vec[personality];
 264         sysent = sysent_vec[personality];
 265
 266         switch (personality) {
 267         case 0:
 268                 errnoent = errnoent0;
 269                 nerrnos = nerrnos0;
 270                 ioctlent = ioctlent0;
 271                 nioctlents = nioctlents0;
 272                 signalent = signalent0;
 273                 nsignals = nsignals0;
 274                 printers = &printers0;
 275                 break;
 276
 277         case 1:
 278                 errnoent = errnoent1;
 279                 nerrnos = nerrnos1;
 280                 ioctlent = ioctlent1;
 281                 nioctlents = nioctlents1;
 282                 signalent = signalent1;
 283                 nsignals = nsignals1;
 284                 printers = &printers1;
 285                 break;
 286
 287 # if SUPPORTED_PERSONALITIES > 2
 288         case 2:
 289                 errnoent = errnoent2;
 290                 nerrnos = nerrnos2;
 291                 ioctlent = ioctlent2;
 292                 nioctlents = nioctlents2;
 293                 signalent = signalent2;
 294                 nsignals = nsignals2;
 295                 printers = &printers2;
 296                 break;
 297 # endif
 298         }
 299
 300         current_personality = personality;
 301 # ifndef current_wordsize
 302         current_wordsize = personality_wordsize[personality];
 303 # endif
 304 }
 305
 306 static void
 307 update_personality(struct tcb *tcp, unsigned int personality)
 308 {
 309         if (personality == current_personality)
 310                 return;
 311         set_personality(personality);
 312
 313         if (personality == tcp->currpers)
 314                 return;
 315         tcp->currpers = personality;
 316
 317 # undef PERSONALITY_NAMES
 318 # if defined POWERPC64
 319 #  define PERSONALITY_NAMES {"64 bit", "32 bit"}
 320 # elif defined X86_64
 321 #  define PERSONALITY_NAMES {"64 bit", "32 bit", "x32"}
 322 # elif defined X32
 323 #  define PERSONALITY_NAMES {"x32", "32 bit"}
 324 # elif defined AARCH64
 325 #  define PERSONALITY_NAMES {"64 bit", "32 bit"}
 326 # elif defined TILE
 327 #  define PERSONALITY_NAMES {"64-bit", "32-bit"}
 328 # endif
 329 # ifdef PERSONALITY_NAMES
 330         if (!qflag) {
 331                 static const char *const names[] = PERSONALITY_NAMES;
 332                 error_msg("[ Process PID=%d runs in %s mode. ]",
 333                           tcp->pid, names[personality]);
 334         }
 335 # endif
 336 }
 337 #endif
 338
 339 #ifdef SYS_socket_subcall
 340 static void
 341 decode_socket_subcall(struct tcb *tcp)
 342 {
 343         const int call = tcp->u_arg[0];
 344
 345         if (call < 1 || call >= SYS_socket_nsubcalls)
 346                 return;
 347
 348         const unsigned long scno = SYS_socket_subcall + call;
 349         const unsigned int nargs = sysent[scno].nargs;
 350         uint64_t buf[nargs];
 351
 352         if (umoven(tcp, tcp->u_arg[1], nargs * current_wordsize, buf) < 0)
 353                 return;
 354
 355         tcp->scno = scno;
 356         tcp->qual_flg = qual_flags(scno);
 357         tcp->s_ent = &sysent[scno];
 358
 359         unsigned int i;
 360         for (i = 0; i < nargs; ++i)
 361                 tcp->u_arg[i] = (sizeof(uint32_t) == current_wordsize)
 362                                 ? ((uint32_t *) (void *) buf)[i] : buf[i];
 363 }
 364 #endif
 365
 366 #ifdef SYS_ipc_subcall
 367 static void
 368 decode_ipc_subcall(struct tcb *tcp)
 369 {
 370         unsigned int call = tcp->u_arg[0];
 371         const unsigned int version = call >> 16;
 372
 373         if (version) {
 374 # if defined S390 || defined S390X
 375                 return;
 376 # else
 377 #  ifdef SPARC64
 378                 if (current_wordsize == 8)
 379                         return;
 380 #  endif
 381                 set_tcb_priv_ulong(tcp, version);
 382                 call &= 0xffff;
 383 # endif
 384         }
 385
 386         switch (call) {
 387                 case  1: case  2: case  3: case  4:
 388                 case 11: case 12: case 13: case 14:
 389                 case 21: case 22: case 23: case 24:
 390                         break;
 391                 default:
 392                         return;
 393         }
 394
 395         tcp->scno = SYS_ipc_subcall + call;
 396         tcp->qual_flg = qual_flags(tcp->scno);
 397         tcp->s_ent = &sysent[tcp->scno];
 398
 399         const unsigned int n = tcp->s_ent->nargs;
 400         unsigned int i;
 401         for (i = 0; i < n; i++)
 402                 tcp->u_arg[i] = tcp->u_arg[i + 1];
 403 }
 404 #endif
 405
 406 #ifdef LINUX_MIPSO32
 407 static void
 408 decode_mips_subcall(struct tcb *tcp)
 409 {
 410         if (!SCNO_IS_VALID(tcp->u_arg[0]))
 411                 return;
 412         tcp->scno = tcp->u_arg[0];
 413         tcp->qual_flg = qual_flags(tcp->scno);
 414         tcp->s_ent = &sysent[tcp->scno];
 415         memmove(&tcp->u_arg[0], &tcp->u_arg[1],
 416                 sizeof(tcp->u_arg) - sizeof(tcp->u_arg[0]));
 417         /*
 418          * Fetching the last arg of 7-arg syscalls (fadvise64_64
 419          * and sync_file_range) requires additional code,
 420          * see linux/mips/get_syscall_args.c
 421          */
 422         if (tcp->s_ent->nargs == MAX_ARGS) {
 423                 if (umoven(tcp,
 424                            mips_REG_SP + MAX_ARGS * sizeof(tcp->u_arg[0]),
 425                            sizeof(tcp->u_arg[0]),
 426                            &tcp->u_arg[MAX_ARGS - 1]) < 0)
 427                 tcp->u_arg[MAX_ARGS - 1] = 0;
 428         }
 429 }
 430 #endif /* LINUX_MIPSO32 */
 431
 432 static void
 433 dumpio(struct tcb *tcp)
 434 {
 435         if (syserror(tcp))
 436                 return;
 437
 438         int fd = tcp->u_arg[0];
 439         if (fd < 0)
 440                 return;
 441
 442         if (is_number_in_set(fd, &read_set)) {
 443                 switch (tcp->s_ent->sen) {
 444                 case SEN_read:
 445                 case SEN_pread:
 446                 case SEN_recv:
 447                 case SEN_recvfrom:
 448                 case SEN_mq_timedreceive:
 449                         dumpstr(tcp, tcp->u_arg[1], tcp->u_rval);
 450                         return;
 451                 case SEN_readv:
 452                 case SEN_preadv:
 453                 case SEN_preadv2:
 454                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1],
 455                                      tcp->u_rval);
 456                         return;
 457                 case SEN_recvmsg:
 458                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], tcp->u_rval);
 459                         return;
 460                 case SEN_recvmmsg:
 461                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 462                         return;
 463                 }
 464         }
 465         if (is_number_in_set(fd, &write_set)) {
 466                 switch (tcp->s_ent->sen) {
 467                 case SEN_write:
 468                 case SEN_pwrite:
 469                 case SEN_send:
 470                 case SEN_sendto:
 471                 case SEN_mq_timedsend:
 472                         dumpstr(tcp, tcp->u_arg[1], tcp->u_arg[2]);
 473                         break;
 474                 case SEN_writev:
 475                 case SEN_pwritev:
 476                 case SEN_pwritev2:
 477                 case SEN_vmsplice:
 478                         dumpiov(tcp, tcp->u_arg[2], tcp->u_arg[1]);
 479                         break;
 480                 case SEN_sendmsg:
 481                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], -1UL);
 482                         break;
 483                 case SEN_sendmmsg:
 484                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
 485                         break;
 486                 }
 487         }
 488 }
 489
 490 /*
 491  * Shuffle syscall numbers so that we don't have huge gaps in syscall table.
 492  * The shuffling should be an involution: shuffle_scno(shuffle_scno(n)) == n.
 493  */
 494 #if defined(ARM) || defined(AARCH64) /* So far only 32-bit ARM needs this */
 495 static long
 496 shuffle_scno(unsigned long scno)
 497 {
 498         if (scno < ARM_FIRST_SHUFFLED_SYSCALL)
 499                 return scno;
 500
 501         /* __ARM_NR_cmpxchg? Swap with LAST_ORDINARY+1 */
 502         if (scno == ARM_FIRST_SHUFFLED_SYSCALL)
 503                 return 0x000ffff0;
 504         if (scno == 0x000ffff0)
 505                 return ARM_FIRST_SHUFFLED_SYSCALL;
 506
 507 #define ARM_SECOND_SHUFFLED_SYSCALL (ARM_FIRST_SHUFFLED_SYSCALL + 1)
 508         /*
 509          * Is it ARM specific syscall?
 510          * Swap [0x000f0000, 0x000f0000 + LAST_SPECIAL] range
 511          * with [SECOND_SHUFFLED, SECOND_SHUFFLED + LAST_SPECIAL] range.
 512          */
 513         if (scno >= 0x000f0000 &&
 514             scno <= 0x000f0000 + ARM_LAST_SPECIAL_SYSCALL) {
 515                 return scno - 0x000f0000 + ARM_SECOND_SHUFFLED_SYSCALL;
 516         }
 517         if (scno <= ARM_SECOND_SHUFFLED_SYSCALL + ARM_LAST_SPECIAL_SYSCALL) {
 518                 return scno + 0x000f0000 - ARM_SECOND_SHUFFLED_SYSCALL;
 519         }
 520
 521         return scno;
 522 }
 523 #else
 524 # define shuffle_scno(scno) ((long)(scno))
 525 #endif
 526
 527 const char *
 528 err_name(unsigned long err)
 529 {
 530         if ((err < nerrnos) && errnoent[err])
 531                 return errnoent[err];
 532
 533         return NULL;
 534 }
 535
 536 static long get_regs_error;
 537
 538 void
 539 clear_regs(void)
 540 {
 541         get_regs_error = -1;
 542 }
 543
 544 static int get_syscall_args(struct tcb *);
 545 static int get_syscall_result(struct tcb *);
 546 static int arch_get_scno(struct tcb *tcp);
 547 static int arch_set_scno(struct tcb *, long);
 548 static void get_error(struct tcb *, const bool);
 549 static int arch_set_error(struct tcb *);
 550
 551 struct fault_opts *fault_vec[SUPPORTED_PERSONALITIES];
 552
 553 static struct fault_opts *
 554 tcb_fault_opts(struct tcb *tcp)
 555 {
 556         return (SCNO_IN_RANGE(tcp->scno) && tcp->fault_vec[current_personality])
 557                ? &tcp->fault_vec[current_personality][tcp->scno] : NULL;
 558 }
 559
 560
 561 static long
 562 inject_syscall_fault_entering(struct tcb *tcp)
 563 {
 564         if (!tcp->fault_vec[current_personality]) {
 565                 tcp->fault_vec[current_personality] =
 566                         xcalloc(nsyscalls, sizeof(**fault_vec));
 567                 memcpy(tcp->fault_vec[current_personality],
 568                        fault_vec[current_personality],
 569                        nsyscalls * sizeof(**fault_vec));
 570         }
 571
 572         struct fault_opts *opts = tcb_fault_opts(tcp);
 573
 574         if (!opts || opts->first == 0)
 575                 return 0;
 576
 577         --opts->first;
 578
 579         if (opts->first != 0)
 580                 return 0;
 581
 582         opts->first = opts->step;
 583
 584         if (!arch_set_scno(tcp, -1))
 585                 tcp->flags |= TCB_FAULT_INJ;
 586
 587         return 0;
 588 }
 589
 590 static long
 591 update_syscall_fault_exiting(struct tcb *tcp)
 592 {
 593         struct fault_opts *opts = tcb_fault_opts(tcp);
 594
 595         if (opts && opts->err && tcp->u_error != opts->err) {
 596                 unsigned long u_error = tcp->u_error;
 597                 tcp->u_error = opts->err;
 598                 if (arch_set_error(tcp))
 599                         tcp->u_error = u_error;
 600         }
 601
 602         return 0;
 603 }
 604
 605 static int
 606 trace_syscall_entering(struct tcb *tcp)
 607 {
 608         int res, scno_good;
 609
 610         scno_good = res = get_scno(tcp);
 611         if (res == 0)
 612                 return res;
 613         if (res == 1)
 614                 res = get_syscall_args(tcp);
 615
 616         if (res != 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                 goto ret;
 624         }
 625
 626 #ifdef LINUX_MIPSO32
 627         if (SEN_syscall == tcp->s_ent->sen)
 628                 decode_mips_subcall(tcp);
 629 #endif
 630
 631 #if defined(SYS_socket_subcall) || defined(SYS_ipc_subcall)
 632         switch (tcp->s_ent->sen) {
 633 # ifdef SYS_socket_subcall
 634                 case SEN_socketcall:
 635                         decode_socket_subcall(tcp);
 636                         break;
 637 # endif
 638 # ifdef SYS_ipc_subcall
 639                 case SEN_ipc:
 640                         decode_ipc_subcall(tcp);
 641                         break;
 642 # endif
 643         }
 644 #endif
 645
 646         /* Restrain from fault injection while the trace executes strace code. */
 647         if (hide_log(tcp)) {
 648                 tcp->qual_flg &= ~QUAL_FAULT;
 649         }
 650
 651         switch (tcp->s_ent->sen) {
 652                 case SEN_execve:
 653                 case SEN_execveat:
 654 #if defined SPARC || defined SPARC64
 655                 case SEN_execv:
 656 #endif
 657                         tcp->flags &= ~TCB_HIDE_LOG;
 658                         break;
 659         }
 660
 661         if (!(tcp->qual_flg & QUAL_TRACE)
 662          || (tracing_paths && !pathtrace_match(tcp))
 663         ) {
 664                 tcp->flags |= TCB_INSYSCALL | TCB_FILTERED;
 665                 tcp->sys_func_rval = 0;
 666                 return 0;
 667         }
 668
 669         tcp->flags &= ~TCB_FILTERED;
 670
 671         if (hide_log(tcp)) {
 672                 res = 0;
 673                 goto ret;
 674         }
 675
 676         if (tcp->qual_flg & QUAL_FAULT)
 677                 inject_syscall_fault_entering(tcp);
 678
 679         if (cflag == CFLAG_ONLY_STATS) {
 680                 res = 0;
 681                 goto ret;
 682         }
 683
 684 #ifdef USE_LIBUNWIND
 685         if (stack_trace_enabled) {
 686                 if (tcp->s_ent->sys_flags & STACKTRACE_CAPTURE_ON_ENTER)
 687                         unwind_capture_stacktrace(tcp);
 688         }
 689 #endif
 690
 691         printleader(tcp);
 692         tprintf("%s(", tcp->s_ent->sys_name);
 693         if (tcp->qual_flg & QUAL_RAW)
 694                 res = printargs(tcp);
 695         else
 696                 res = tcp->s_ent->sys_func(tcp);
 697
 698         fflush(tcp->outf);
 699  ret:
 700         tcp->flags |= TCB_INSYSCALL;
 701         tcp->sys_func_rval = res;
 702         /* Measure the entrance time as late as possible to avoid errors. */
 703         if (Tflag || cflag)
 704                 gettimeofday(&tcp->etime, NULL);
 705         return res;
 706 }
 707
 708 static bool
 709 syscall_fault_injected(struct tcb *tcp)
 710 {
 711         return tcp->flags & TCB_FAULT_INJ;
 712 }
 713
 714 static int
 715 trace_syscall_exiting(struct tcb *tcp)
 716 {
 717         int sys_res;
 718         struct timeval tv;
 719         int res;
 720         unsigned long u_error;
 721         const char *u_error_str;
 722
 723         /* Measure the exit time as early as possible to avoid errors. */
 724         if (Tflag || cflag)
 725                 gettimeofday(&tv, NULL);
 726
 727 #ifdef USE_LIBUNWIND
 728         if (stack_trace_enabled) {
 729                 if (tcp->s_ent->sys_flags & STACKTRACE_INVALIDATE_CACHE)
 730                         unwind_cache_invalidate(tcp);
 731         }
 732 #endif
 733
 734 #if SUPPORTED_PERSONALITIES > 1
 735         update_personality(tcp, tcp->currpers);
 736 #endif
 737         res = (get_regs_error ? -1 : get_syscall_result(tcp));
 738         if (filtered(tcp) || hide_log(tcp))
 739                 goto ret;
 740
 741         if (syserror(tcp) && syscall_fault_injected(tcp))
 742                 update_syscall_fault_exiting(tcp);
 743
 744         if (cflag) {
 745                 count_syscall(tcp, &tv);
 746                 if (cflag == CFLAG_ONLY_STATS) {
 747                         goto ret;
 748                 }
 749         }
 750
 751         /* If not in -ff mode, and printing_tcp != tcp,
 752          * then the log currently does not end with output
 753          * of _our syscall entry_, but with something else.
 754          * We need to say which syscall's return is this.
 755          *
 756          * Forced reprinting via TCB_REPRINT is used only by
 757          * "strace -ff -oLOG test/threaded_execve" corner case.
 758          * It's the only case when -ff mode needs reprinting.
 759          */
 760         if ((followfork < 2 && printing_tcp != tcp) || (tcp->flags & TCB_REPRINT)) {
 761                 tcp->flags &= ~TCB_REPRINT;
 762                 printleader(tcp);
 763                 tprintf("<... %s resumed> ", tcp->s_ent->sys_name);
 764         }
 765         printing_tcp = tcp;
 766
 767         tcp->s_prev_ent = NULL;
 768         if (res != 1) {
 769                 /* There was error in one of prior ptrace ops */
 770                 tprints(") ");
 771                 tabto();
 772                 tprints("= ? <unavailable>\n");
 773                 line_ended();
 774                 tcp->flags &= ~(TCB_INSYSCALL | TCB_FAULT_INJ);
 775                 tcp->sys_func_rval = 0;
 776                 free_tcb_priv_data(tcp);
 777                 return res;
 778         }
 779         tcp->s_prev_ent = tcp->s_ent;
 780
 781         sys_res = 0;
 782         if (tcp->qual_flg & QUAL_RAW) {
 783                 /* sys_res = printargs(tcp); - but it's nop on sysexit */
 784         } else {
 785         /* FIXME: not_failing_only (IOW, option -z) is broken:
 786          * failure of syscall is known only after syscall return.
 787          * Thus we end up with something like this on, say, ENOENT:
 788          *     open("doesnt_exist", O_RDONLY <unfinished ...>
 789          *     {next syscall decode}
 790          * whereas the intended result is that open(...) line
 791          * is not shown at all.
 792          */
 793                 if (not_failing_only && tcp->u_error)
 794                         goto ret;       /* ignore failed syscalls */
 795                 if (tcp->sys_func_rval & RVAL_DECODED)
 796                         sys_res = tcp->sys_func_rval;
 797                 else
 798                         sys_res = tcp->s_ent->sys_func(tcp);
 799         }
 800
 801         tprints(") ");
 802         tabto();
 803         u_error = tcp->u_error;
 804
 805         if (tcp->qual_flg & QUAL_RAW) {
 806                 if (u_error) {
 807                         tprintf("= -1 (errno %lu)", u_error);
 808                         if (syscall_fault_injected(tcp))
 809                                 tprints(" (INJECTED)");
 810                 } else {
 811                         tprintf("= %#lx", tcp->u_rval);
 812                 }
 813         }
 814         else if (!(sys_res & RVAL_NONE) && u_error) {
 815                 switch (u_error) {
 816                 /* Blocked signals do not interrupt any syscalls.
 817                  * In this case syscalls don't return ERESTARTfoo codes.
 818                  *
 819                  * Deadly signals set to SIG_DFL interrupt syscalls
 820                  * and kill the process regardless of which of the codes below
 821                  * is returned by the interrupted syscall.
 822                  * In some cases, kernel forces a kernel-generated deadly
 823                  * signal to be unblocked and set to SIG_DFL (and thus cause
 824                  * death) if it is blocked or SIG_IGNed: for example, SIGSEGV
 825                  * or SIGILL. (The alternative is to leave process spinning
 826                  * forever on the faulty instruction - not useful).
 827                  *
 828                  * SIG_IGNed signals and non-deadly signals set to SIG_DFL
 829                  * (for example, SIGCHLD, SIGWINCH) interrupt syscalls,
 830                  * but kernel will always restart them.
 831                  */
 832                 case ERESTARTSYS:
 833                         /* Most common type of signal-interrupted syscall exit code.
 834                          * The system call will be restarted with the same arguments
 835                          * if SA_RESTART is set; otherwise, it will fail with EINTR.
 836                          */
 837                         tprints("= ? ERESTARTSYS (To be restarted if SA_RESTART is set)");
 838                         break;
 839                 case ERESTARTNOINTR:
 840                         /* Rare. For example, fork() returns this if interrupted.
 841                          * SA_RESTART is ignored (assumed set): the restart is unconditional.
 842                          */
 843                         tprints("= ? ERESTARTNOINTR (To be restarted)");
 844                         break;
 845                 case ERESTARTNOHAND:
 846                         /* pause(), rt_sigsuspend() etc use this code.
 847                          * SA_RESTART is ignored (assumed not set):
 848                          * syscall won't restart (will return EINTR instead)
 849                          * even after signal with SA_RESTART set. However,
 850                          * after SIG_IGN or SIG_DFL signal it will restart
 851                          * (thus the name "restart only if has no handler").
 852                          */
 853                         tprints("= ? ERESTARTNOHAND (To be restarted if no handler)");
 854                         break;
 855                 case ERESTART_RESTARTBLOCK:
 856                         /* Syscalls like nanosleep(), poll() which can't be
 857                          * restarted with their original arguments use this
 858                          * code. Kernel will execute restart_syscall() instead,
 859                          * which changes arguments before restarting syscall.
 860                          * SA_RESTART is ignored (assumed not set) similarly
 861                          * to ERESTARTNOHAND. (Kernel can't honor SA_RESTART
 862                          * since restart data is saved in "restart block"
 863                          * in task struct, and if signal handler uses a syscall
 864                          * which in turn saves another such restart block,
 865                          * old data is lost and restart becomes impossible)
 866                          */
 867                         tprints("= ? ERESTART_RESTARTBLOCK (Interrupted by signal)");
 868                         break;
 869                 default:
 870                         u_error_str = err_name(u_error);
 871                         if (u_error_str)
 872                                 tprintf("= -1 %s (%s)",
 873                                         u_error_str, strerror(u_error));
 874                         else
 875                                 tprintf("= -1 %lu (%s)",
 876                                         u_error, strerror(u_error));
 877                         break;
 878                 }
 879                 if (syscall_fault_injected(tcp))
 880                         tprintf(" (INJECTED)");
 881                 if ((sys_res & RVAL_STR) && tcp->auxstr)
 882                         tprintf(" (%s)", tcp->auxstr);
 883         }
 884         else {
 885                 if (sys_res & RVAL_NONE)
 886                         tprints("= ?");
 887                 else {
 888                         switch (sys_res & RVAL_MASK) {
 889                         case RVAL_HEX:
 890 #if SUPPORTED_PERSONALITIES > 1
 891                                 if (current_wordsize < sizeof(long))
 892                                         tprintf("= %#x",
 893                                                 (unsigned int) tcp->u_rval);
 894                                 else
 895 #endif
 896                                         tprintf("= %#lx", tcp->u_rval);
 897                                 break;
 898                         case RVAL_OCTAL:
 899                                 tprints("= ");
 900                                 print_numeric_long_umask(tcp->u_rval);
 901                                 break;
 902                         case RVAL_UDECIMAL:
 903 #if SUPPORTED_PERSONALITIES > 1
 904                                 if (current_wordsize < sizeof(long))
 905                                         tprintf("= %u",
 906                                                 (unsigned int) tcp->u_rval);
 907                                 else
 908 #endif
 909                                         tprintf("= %lu", tcp->u_rval);
 910                                 break;
 911                         case RVAL_DECIMAL:
 912                                 tprintf("= %ld", tcp->u_rval);
 913                                 break;
 914                         case RVAL_FD:
 915                                 if (show_fd_path) {
 916                                         tprints("= ");
 917                                         printfd(tcp, tcp->u_rval);
 918                                 }
 919                                 else
 920                                         tprintf("= %ld", tcp->u_rval);
 921                                 break;
 922 #if HAVE_STRUCT_TCB_EXT_ARG
 923                         /*
 924                         case RVAL_LHEX:
 925                                 tprintf("= %#llx", tcp->u_lrval);
 926                                 break;
 927                         case RVAL_LOCTAL:
 928                                 tprintf("= %#llo", tcp->u_lrval);
 929                                 break;
 930                         */
 931                         case RVAL_LUDECIMAL:
 932                                 tprintf("= %llu", tcp->u_lrval);
 933                                 break;
 934                         /*
 935                         case RVAL_LDECIMAL:
 936                                 tprintf("= %lld", tcp->u_lrval);
 937                                 break;
 938                         */
 939 #endif /* HAVE_STRUCT_TCB_EXT_ARG */
 940                         default:
 941                                 error_msg("invalid rval format");
 942                                 break;
 943                         }
 944                 }
 945                 if ((sys_res & RVAL_STR) && tcp->auxstr)
 946                         tprintf(" (%s)", tcp->auxstr);
 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
 962  ret:
 963         tcp->flags &= ~(TCB_INSYSCALL | TCB_FAULT_INJ);
 964         tcp->sys_func_rval = 0;
 965         free_tcb_priv_data(tcp);
 966         return 0;
 967 }
 968
 969 int
 970 trace_syscall(struct tcb *tcp)
 971 {
 972         return exiting(tcp) ?
 973                 trace_syscall_exiting(tcp) : trace_syscall_entering(tcp);
 974 }
 975
 976 bool
 977 is_erestart(struct tcb *tcp)
 978 {
 979         switch (tcp->u_error) {
 980                 case ERESTARTSYS:
 981                 case ERESTARTNOINTR:
 982                 case ERESTARTNOHAND:
 983                 case ERESTART_RESTARTBLOCK:
 984                         return true;
 985                 default:
 986                         return false;
 987         }
 988 }
 989
 990 static unsigned long saved_u_error;
 991
 992 void
 993 temporarily_clear_syserror(struct tcb *tcp)
 994 {
 995         saved_u_error = tcp->u_error;
 996         tcp->u_error = 0;
 997 }
 998
 999 void
1000 restore_cleared_syserror(struct tcb *tcp)
1001 {
1002         tcp->u_error = saved_u_error;
1003 }
1004
1005 /*
1006  * Check the syscall return value register value for whether it is
1007  * a negated errno code indicating an error, or a success return value.
1008  */
1009 static inline bool
1010 is_negated_errno(kernel_ulong_t val)
1011 {
1012         /* Linux kernel defines MAX_ERRNO to 4095. */
1013         kernel_ulong_t max = -(kernel_long_t) 4095;
1014
1015 #if defined X86_64 || defined X32
1016         /*
1017          * current_wordsize is 4 for x32 personality
1018          * but truncation _must not_ be done in it, so
1019          * check current_personality instead.
1020          */
1021         if (current_personality == 1) {
1022                 val = (uint32_t) val;
1023                 max = (uint32_t) max;
1024         }
1025 #elif SUPPORTED_PERSONALITIES > 1 && SIZEOF_LONG > 4
1026         if (current_wordsize < sizeof(val)) {
1027                 val = (uint32_t) val;
1028                 max = (uint32_t) max;
1029         }
1030 #endif
1031
1032         return val >= max;
1033 }
1034
1035 #include "arch_regs.c"
1036
1037 #ifdef HAVE_GETRVAL2
1038 # include "arch_getrval2.c"
1039 #endif
1040
1041 void
1042 print_pc(struct tcb *tcp)
1043 {
1044 #if defined ARCH_PC_REG
1045 # define ARCH_GET_PC 0
1046 #elif defined ARCH_PC_PEEK_ADDR
1047         long pc;
1048 # define ARCH_PC_REG pc
1049 # define ARCH_GET_PC upeek(tcp->pid, ARCH_PC_PEEK_ADDR, &pc)
1050 #else
1051 # error Neither ARCH_PC_REG nor ARCH_PC_PEEK_ADDR is defined
1052 #endif
1053         if (get_regs_error || ARCH_GET_PC)
1054                 tprints(current_wordsize == 4 ? "[????????] "
1055                                               : "[????????????????] ");
1056         else
1057                 tprintf(current_wordsize == 4 ? "[%08lx] " : "[%016lx] ",
1058                         (unsigned long) ARCH_PC_REG);
1059 }
1060
1061 #include "getregs_old.h"
1062
1063 #undef ptrace_getregset_or_getregs
1064 #undef ptrace_setregset_or_setregs
1065 #ifdef ARCH_REGS_FOR_GETREGSET
1066
1067 # define ptrace_getregset_or_getregs ptrace_getregset
1068 static long
1069 ptrace_getregset(pid_t pid)
1070 {
1071 # ifdef ARCH_IOVEC_FOR_GETREGSET
1072         /* variable iovec */
1073         ARCH_IOVEC_FOR_GETREGSET.iov_len = sizeof(ARCH_REGS_FOR_GETREGSET);
1074         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS,
1075                       &ARCH_IOVEC_FOR_GETREGSET);
1076 # else
1077         /* constant iovec */
1078         static struct iovec io = {
1079                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1080                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1081         };
1082         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS, &io);
1083
1084 # endif
1085 }
1086
1087 # ifndef HAVE_GETREGS_OLD
1088 #  define ptrace_setregset_or_setregs ptrace_setregset
1089 static int
1090 ptrace_setregset(pid_t pid)
1091 {
1092 #  ifdef ARCH_IOVEC_FOR_GETREGSET
1093         /* variable iovec */
1094         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS,
1095                       &ARCH_IOVEC_FOR_GETREGSET);
1096 #  else
1097         /* constant iovec */
1098         static struct iovec io = {
1099                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1100                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1101         };
1102         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS, &io);
1103 #  endif
1104 }
1105 # endif /* !HAVE_GETREGS_OLD */
1106
1107 #elif defined ARCH_REGS_FOR_GETREGS
1108
1109 # define ptrace_getregset_or_getregs ptrace_getregs
1110 static long
1111 ptrace_getregs(pid_t pid)
1112 {
1113 # if defined SPARC || defined SPARC64
1114         /* SPARC systems have the meaning of data and addr reversed */
1115         return ptrace(PTRACE_GETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1116 # else
1117         return ptrace(PTRACE_GETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1118 # endif
1119 }
1120
1121 # ifndef HAVE_GETREGS_OLD
1122 #  define ptrace_setregset_or_setregs ptrace_setregs
1123 static int
1124 ptrace_setregs(pid_t pid)
1125 {
1126 #  if defined SPARC || defined SPARC64
1127         /* SPARC systems have the meaning of data and addr reversed */
1128         return ptrace(PTRACE_SETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1129 #  else
1130         return ptrace(PTRACE_SETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1131 #  endif
1132 }
1133 # endif /* !HAVE_GETREGS_OLD */
1134
1135 #endif /* ARCH_REGS_FOR_GETREGSET || ARCH_REGS_FOR_GETREGS */
1136
1137 void
1138 get_regs(pid_t pid)
1139 {
1140 #undef USE_GET_SYSCALL_RESULT_REGS
1141 #ifdef ptrace_getregset_or_getregs
1142
1143 # ifdef HAVE_GETREGS_OLD
1144         /*
1145          * Try PTRACE_GETREGSET/PTRACE_GETREGS first,
1146          * fallback to getregs_old.
1147          */
1148         static int use_getregs_old;
1149         if (use_getregs_old < 0) {
1150                 get_regs_error = ptrace_getregset_or_getregs(pid);
1151                 return;
1152         } else if (use_getregs_old == 0) {
1153                 get_regs_error = ptrace_getregset_or_getregs(pid);
1154                 if (get_regs_error >= 0) {
1155                         use_getregs_old = -1;
1156                         return;
1157                 }
1158                 if (errno == EPERM || errno == ESRCH)
1159                         return;
1160                 use_getregs_old = 1;
1161         }
1162         get_regs_error = getregs_old(pid);
1163 # else /* !HAVE_GETREGS_OLD */
1164         /* Assume that PTRACE_GETREGSET/PTRACE_GETREGS works. */
1165         get_regs_error = ptrace_getregset_or_getregs(pid);
1166 # endif /* !HAVE_GETREGS_OLD */
1167
1168 #else /* !ptrace_getregset_or_getregs */
1169
1170 # define USE_GET_SYSCALL_RESULT_REGS 1
1171 # warning get_regs is not implemented for this architecture yet
1172         get_regs_error = 0;
1173
1174 #endif /* !ptrace_getregset_or_getregs */
1175 }
1176
1177 #ifdef ptrace_setregset_or_setregs
1178 static int
1179 set_regs(pid_t pid)
1180 {
1181         return ptrace_setregset_or_setregs(pid);
1182 }
1183 #endif /* ptrace_setregset_or_setregs */
1184
1185 struct sysent_buf {
1186         struct tcb *tcp;
1187         struct_sysent ent;
1188         char buf[sizeof("syscall_%lu") + sizeof(long) * 3];
1189 };
1190
1191 static void
1192 free_sysent_buf(void *ptr)
1193 {
1194         struct sysent_buf *s = ptr;
1195         s->tcp->s_prev_ent = s->tcp->s_ent = NULL;
1196         free(ptr);
1197 }
1198
1199 /*
1200  * Returns:
1201  * 0: "ignore this ptrace stop", bail out of trace_syscall_entering() silently.
1202  * 1: ok, continue in trace_syscall_entering().
1203  * other: error, trace_syscall_entering() should print error indicator
1204  *    ("????" etc) and bail out.
1205  */
1206 int
1207 get_scno(struct tcb *tcp)
1208 {
1209         if (get_regs_error)
1210                 return -1;
1211
1212         int rc = arch_get_scno(tcp);
1213         if (rc != 1)
1214                 return rc;
1215
1216         if (SCNO_IS_VALID(tcp->scno)) {
1217                 tcp->s_ent = &sysent[tcp->scno];
1218                 tcp->qual_flg = qual_flags(tcp->scno);
1219         } else {
1220                 struct sysent_buf *s = xcalloc(1, sizeof(*s));
1221
1222                 s->tcp = tcp;
1223                 s->ent.nargs = MAX_ARGS;
1224                 s->ent.sen = SEN_printargs;
1225                 s->ent.sys_func = printargs;
1226                 s->ent.sys_name = s->buf;
1227                 sprintf(s->buf, "syscall_%lu", shuffle_scno(tcp->scno));
1228
1229                 tcp->s_ent = &s->ent;
1230                 tcp->qual_flg = QUAL_RAW | DEFAULT_QUAL_FLAGS;
1231
1232                 set_tcb_priv_data(tcp, s, free_sysent_buf);
1233
1234                 if (debug_flag)
1235                         error_msg("pid %d invalid syscall %ld", tcp->pid, tcp->scno);
1236         }
1237         return 1;
1238 }
1239
1240 #ifdef USE_GET_SYSCALL_RESULT_REGS
1241 static int get_syscall_result_regs(struct tcb *);
1242 #endif
1243
1244 /* Returns:
1245  * 1: ok, continue in trace_syscall_exiting().
1246  * -1: error, trace_syscall_exiting() should print error indicator
1247  *    ("????" etc) and bail out.
1248  */
1249 static int
1250 get_syscall_result(struct tcb *tcp)
1251 {
1252 #ifdef USE_GET_SYSCALL_RESULT_REGS
1253         if (get_syscall_result_regs(tcp))
1254                 return -1;
1255 #endif
1256         tcp->u_error = 0;
1257         get_error(tcp, !(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS));
1258
1259         return 1;
1260 }
1261
1262 #include "get_scno.c"
1263 #include "set_scno.c"
1264 #include "get_syscall_args.c"
1265 #ifdef USE_GET_SYSCALL_RESULT_REGS
1266 # include "get_syscall_result.c"
1267 #endif
1268 #include "get_error.c"
1269 #include "set_error.c"
1270 #ifdef HAVE_GETREGS_OLD
1271 # include "getregs_old.c"
1272 #endif
1273
1274 const char *
1275 syscall_name(long scno)
1276 {
1277 #if defined X32_PERSONALITY_NUMBER && defined __X32_SYSCALL_BIT
1278         if (current_personality == X32_PERSONALITY_NUMBER)
1279                 scno &= ~__X32_SYSCALL_BIT;
1280 #endif
1281         return SCNO_IS_VALID(scno) ? sysent[scno].sys_name: NULL;
1282 }