]> granicus.if.org Git - strace/blob - syscall.c
Update ioctl entries from linux v4.18
[strace] / 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) 1999-2018 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 "mmap_notify.h"
37 #include "native_defs.h"
38 #include "ptrace.h"
39 #include "nsig.h"
40 #include "number_set.h"
41 #include "delay.h"
42 #include "retval.h"
43 #include <limits.h>
44
45 /* for struct iovec */
46 #include <sys/uio.h>
47
48 /* for __X32_SYSCALL_BIT */
49 #include <asm/unistd.h>
50
51 #include "regs.h"
52
53 #if defined(SPARC64)
54 # undef PTRACE_GETREGS
55 # define PTRACE_GETREGS PTRACE_GETREGS64
56 # undef PTRACE_SETREGS
57 # define PTRACE_SETREGS PTRACE_SETREGS64
58 #endif
59
60 #include "syscall.h"
61 #include "xstring.h"
62
63 /* Define these shorthand notations to simplify the syscallent files. */
64 #include "sysent_shorthand_defs.h"
65
66 #define SEN(syscall_name) SEN_ ## syscall_name, SYS_FUNC_NAME(sys_ ## syscall_name)
67
68 const struct_sysent sysent0[] = {
69 #include "syscallent.h"
70 };
71
72 #if SUPPORTED_PERSONALITIES > 1
73 # include PERSONALITY1_INCLUDE_FUNCS
74 static const struct_sysent sysent1[] = {
75 # include "syscallent1.h"
76 };
77 #endif
78
79 #if SUPPORTED_PERSONALITIES > 2
80 # include PERSONALITY2_INCLUDE_FUNCS
81 static const struct_sysent sysent2[] = {
82 # include "syscallent2.h"
83 };
84 #endif
85
86 /* Now undef them since short defines cause wicked namespace pollution. */
87 #include "sysent_shorthand_undefs.h"
88
89 const char *const errnoent[] = {
90 #include "errnoent.h"
91 };
92 const char *const signalent[] = {
93 #include "signalent.h"
94
95 };
96 /*
97  * `ioctlent[012].h' files are automatically generated by the auxiliary
98  * program `ioctlsort', such that the list is sorted by the `code' field.
99  * This has the side-effect of resolving the _IO.. macros into
100  * plain integers, eliminating the need to include here everything
101  * in "/usr/include".
102  */
103
104 const struct_ioctlent ioctlent0[] = {
105 #include "ioctlent0.h"
106 };
107
108 #if SUPPORTED_PERSONALITIES > 1
109 static const struct_ioctlent ioctlent1[] = {
110 # include "ioctlent1.h"
111 };
112 # include PERSONALITY0_INCLUDE_PRINTERS_DECLS
113 static const struct_printers printers0 = {
114 # include PERSONALITY0_INCLUDE_PRINTERS_DEFS
115 };
116 # include PERSONALITY1_INCLUDE_PRINTERS_DECLS
117 static const struct_printers printers1 = {
118 # include PERSONALITY1_INCLUDE_PRINTERS_DEFS
119 };
120 #endif
121
122 #if SUPPORTED_PERSONALITIES > 2
123 static const struct_ioctlent ioctlent2[] = {
124 # include "ioctlent2.h"
125 };
126 # include PERSONALITY2_INCLUDE_PRINTERS_DECLS
127 static const struct_printers printers2 = {
128 # include PERSONALITY2_INCLUDE_PRINTERS_DEFS
129 };
130 #endif
131
132 enum {
133         nsyscalls0 = ARRAY_SIZE(sysent0)
134 #if SUPPORTED_PERSONALITIES > 1
135         , nsyscalls1 = ARRAY_SIZE(sysent1)
136 # if SUPPORTED_PERSONALITIES > 2
137         , nsyscalls2 = ARRAY_SIZE(sysent2)
138 # endif
139 #endif
140 };
141
142 enum {
143         nioctlents0 = ARRAY_SIZE(ioctlent0)
144 #if SUPPORTED_PERSONALITIES > 1
145         , nioctlents1 = ARRAY_SIZE(ioctlent1)
146 # if SUPPORTED_PERSONALITIES > 2
147         , nioctlents2 = ARRAY_SIZE(ioctlent2)
148 # endif
149 #endif
150 };
151
152 #if SUPPORTED_PERSONALITIES > 1
153 const struct_sysent *sysent = sysent0;
154 const struct_ioctlent *ioctlent = ioctlent0;
155 const struct_printers *printers = &printers0;
156 #endif
157
158 const unsigned int nerrnos = ARRAY_SIZE(errnoent);
159 const unsigned int nsignals = ARRAY_SIZE(signalent);
160 unsigned nsyscalls = nsyscalls0;
161 unsigned nioctlents = nioctlents0;
162
163 const unsigned int nsyscall_vec[SUPPORTED_PERSONALITIES] = {
164         nsyscalls0,
165 #if SUPPORTED_PERSONALITIES > 1
166         nsyscalls1,
167 #endif
168 #if SUPPORTED_PERSONALITIES > 2
169         nsyscalls2,
170 #endif
171 };
172 const struct_sysent *const sysent_vec[SUPPORTED_PERSONALITIES] = {
173         sysent0,
174 #if SUPPORTED_PERSONALITIES > 1
175         sysent1,
176 #endif
177 #if SUPPORTED_PERSONALITIES > 2
178         sysent2,
179 #endif
180 };
181
182 const char *const personality_names[] =
183 # if defined X86_64
184         {"64 bit", "32 bit", "x32"}
185 # elif defined X32
186         {"x32", "32 bit"}
187 # elif SUPPORTED_PERSONALITIES == 2
188         {"64 bit", "32 bit"}
189 # else
190         {STRINGIFY_VAL(__WORDSIZE) " bit"}
191 # endif
192         ;
193
194 const char *const personality_designators[] =
195 # if defined X86_64
196         { "64", "32", "x32" }
197 # elif defined X32
198         { "x32", "32" }
199 # elif SUPPORTED_PERSONALITIES == 2
200         { "64", "32" }
201 # else
202         { STRINGIFY_VAL(__WORDSIZE) }
203 # endif
204         ;
205
206 #if SUPPORTED_PERSONALITIES > 1
207
208 unsigned current_personality;
209
210 # ifndef current_wordsize
211 unsigned current_wordsize = PERSONALITY0_WORDSIZE;
212 static const int personality_wordsize[SUPPORTED_PERSONALITIES] = {
213         PERSONALITY0_WORDSIZE,
214         PERSONALITY1_WORDSIZE,
215 # if SUPPORTED_PERSONALITIES > 2
216         PERSONALITY2_WORDSIZE,
217 # endif
218 };
219 # endif
220
221 # ifndef current_klongsize
222 unsigned current_klongsize = PERSONALITY0_KLONGSIZE;
223 static const int personality_klongsize[SUPPORTED_PERSONALITIES] = {
224         PERSONALITY0_KLONGSIZE,
225         PERSONALITY1_KLONGSIZE,
226 #  if SUPPORTED_PERSONALITIES > 2
227         PERSONALITY2_KLONGSIZE,
228 #  endif
229 };
230 # endif
231
232 void
233 set_personality(unsigned int personality)
234 {
235         if (personality == current_personality)
236                 return;
237
238         if (personality >= SUPPORTED_PERSONALITIES)
239                 error_msg_and_die("Requested switch to unsupported personality "
240                                   "%u", personality);
241
242         nsyscalls = nsyscall_vec[personality];
243         sysent = sysent_vec[personality];
244
245         switch (personality) {
246         case 0:
247                 ioctlent = ioctlent0;
248                 nioctlents = nioctlents0;
249                 printers = &printers0;
250                 break;
251
252         case 1:
253                 ioctlent = ioctlent1;
254                 nioctlents = nioctlents1;
255                 printers = &printers1;
256                 break;
257
258 # if SUPPORTED_PERSONALITIES > 2
259         case 2:
260                 ioctlent = ioctlent2;
261                 nioctlents = nioctlents2;
262                 printers = &printers2;
263                 break;
264 # endif
265         }
266
267         current_personality = personality;
268 # ifndef current_wordsize
269         current_wordsize = personality_wordsize[personality];
270 # endif
271 # ifndef current_klongsize
272         current_klongsize = personality_klongsize[personality];
273 # endif
274 }
275
276 static void
277 update_personality(struct tcb *tcp, unsigned int personality)
278 {
279         static bool need_mpers_warning[] =
280                 { false, !HAVE_PERSONALITY_1_MPERS, !HAVE_PERSONALITY_2_MPERS };
281
282         set_personality(personality);
283
284         if (personality == tcp->currpers)
285                 return;
286         tcp->currpers = personality;
287
288         if (!qflag) {
289                 error_msg("[ Process PID=%d runs in %s mode. ]",
290                           tcp->pid, personality_names[personality]);
291         }
292
293         if (need_mpers_warning[personality]) {
294                 error_msg("WARNING: Proper structure decoding for this "
295                           "personality is not supported, please consider "
296                           "building strace with mpers support enabled.");
297                 need_mpers_warning[personality] = false;
298         }
299 }
300 #endif
301
302 #ifdef SYS_socket_subcall
303 static void
304 decode_socket_subcall(struct tcb *tcp)
305 {
306         const int call = tcp->u_arg[0];
307
308         if (call < 1 || call >= SYS_socket_nsubcalls)
309                 return;
310
311         const kernel_ulong_t scno = SYS_socket_subcall + call;
312         const unsigned int nargs = sysent[scno].nargs;
313         uint64_t buf[nargs];
314
315         if (umoven(tcp, tcp->u_arg[1], nargs * current_wordsize, buf) < 0)
316                 return;
317
318         tcp->scno = scno;
319         tcp->qual_flg = qual_flags(scno);
320         tcp->s_ent = &sysent[scno];
321
322         unsigned int i;
323         for (i = 0; i < nargs; ++i)
324                 tcp->u_arg[i] = (sizeof(uint32_t) == current_wordsize)
325                                 ? ((uint32_t *) (void *) buf)[i] : buf[i];
326 }
327 #endif /* SYS_socket_subcall */
328
329 #ifdef SYS_ipc_subcall
330 static void
331 decode_ipc_subcall(struct tcb *tcp)
332 {
333         unsigned int call = tcp->u_arg[0];
334         const unsigned int version = call >> 16;
335
336         if (version) {
337 # if defined S390 || defined S390X
338                 return;
339 # else
340 #  ifdef SPARC64
341                 if (current_wordsize == 8)
342                         return;
343 #  endif
344                 set_tcb_priv_ulong(tcp, version);
345                 call &= 0xffff;
346 # endif
347         }
348
349         switch (call) {
350                 case  1: case  2: case  3: case  4:
351                 case 11: case 12: case 13: case 14:
352                 case 21: case 22: case 23: case 24:
353                         break;
354                 default:
355                         return;
356         }
357
358         tcp->scno = SYS_ipc_subcall + call;
359         tcp->qual_flg = qual_flags(tcp->scno);
360         tcp->s_ent = &sysent[tcp->scno];
361
362         const unsigned int n = tcp->s_ent->nargs;
363         unsigned int i;
364         for (i = 0; i < n; i++)
365                 tcp->u_arg[i] = tcp->u_arg[i + 1];
366 }
367 #endif /* SYS_ipc_subcall */
368
369 #ifdef SYS_syscall_subcall
370 static void
371 decode_syscall_subcall(struct tcb *tcp)
372 {
373         if (!scno_is_valid(tcp->u_arg[0]))
374                 return;
375         tcp->scno = tcp->u_arg[0];
376         tcp->qual_flg = qual_flags(tcp->scno);
377         tcp->s_ent = &sysent[tcp->scno];
378         memmove(&tcp->u_arg[0], &tcp->u_arg[1],
379                 sizeof(tcp->u_arg) - sizeof(tcp->u_arg[0]));
380 # ifdef LINUX_MIPSO32
381         /*
382          * Fetching the last arg of 7-arg syscalls (fadvise64_64
383          * and sync_file_range) requires additional code,
384          * see linux/mips/get_syscall_args.c
385          */
386         if (tcp->s_ent->nargs == MAX_ARGS) {
387                 if (umoven(tcp,
388                            mips_REG_SP + MAX_ARGS * sizeof(tcp->u_arg[0]),
389                            sizeof(tcp->u_arg[0]),
390                            &tcp->u_arg[MAX_ARGS - 1]) < 0)
391                 tcp->u_arg[MAX_ARGS - 1] = 0;
392         }
393 # endif /* LINUX_MIPSO32 */
394 }
395 #endif /* SYS_syscall_subcall */
396
397 static void
398 dumpio(struct tcb *tcp)
399 {
400         int fd = tcp->u_arg[0];
401         if (fd < 0)
402                 return;
403
404         if (is_number_in_set(fd, write_set)) {
405                 switch (tcp->s_ent->sen) {
406                 case SEN_write:
407                 case SEN_pwrite:
408                 case SEN_send:
409                 case SEN_sendto:
410                 case SEN_mq_timedsend:
411                         dumpstr(tcp, tcp->u_arg[1], tcp->u_arg[2]);
412                         break;
413                 case SEN_writev:
414                 case SEN_pwritev:
415                 case SEN_pwritev2:
416                 case SEN_vmsplice:
417                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1], -1);
418                         break;
419                 case SEN_sendmsg:
420                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], -1);
421                         break;
422                 case SEN_sendmmsg:
423                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
424                         break;
425                 }
426         }
427
428         if (syserror(tcp))
429                 return;
430
431         if (is_number_in_set(fd, read_set)) {
432                 switch (tcp->s_ent->sen) {
433                 case SEN_read:
434                 case SEN_pread:
435                 case SEN_recv:
436                 case SEN_recvfrom:
437                 case SEN_mq_timedreceive:
438                         dumpstr(tcp, tcp->u_arg[1], tcp->u_rval);
439                         return;
440                 case SEN_readv:
441                 case SEN_preadv:
442                 case SEN_preadv2:
443                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1],
444                                      tcp->u_rval);
445                         return;
446                 case SEN_recvmsg:
447                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], tcp->u_rval);
448                         return;
449                 case SEN_recvmmsg:
450                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
451                         return;
452                 }
453         }
454 }
455
456 const char *
457 err_name(unsigned long err)
458 {
459         if ((err < nerrnos) && errnoent[err])
460                 return errnoent[err];
461
462         return NULL;
463 }
464
465 static void
466 print_err_ret(kernel_ulong_t ret, unsigned long u_error)
467 {
468         const char *u_error_str = err_name(u_error);
469
470         if (u_error_str)
471                 tprintf("= %" PRI_kld " %s (%s)",
472                         ret, u_error_str, strerror(u_error));
473         else
474                 tprintf("= %" PRI_kld " (errno %lu)", ret, u_error);
475 }
476
477 static long get_regs(struct tcb *);
478 static int get_syscall_args(struct tcb *);
479 static int get_syscall_result(struct tcb *);
480 static int arch_get_scno(struct tcb *tcp);
481 static int arch_set_scno(struct tcb *, kernel_ulong_t);
482 static void get_error(struct tcb *, const bool);
483 static int arch_set_error(struct tcb *);
484 static int arch_set_success(struct tcb *);
485
486 struct inject_opts *inject_vec[SUPPORTED_PERSONALITIES];
487
488 static struct inject_opts *
489 tcb_inject_opts(struct tcb *tcp)
490 {
491         return (scno_in_range(tcp->scno) && tcp->inject_vec[current_personality])
492                ? &tcp->inject_vec[current_personality][tcp->scno] : NULL;
493 }
494
495
496 static long
497 tamper_with_syscall_entering(struct tcb *tcp, unsigned int *signo)
498 {
499         if (!tcp->inject_vec[current_personality]) {
500                 tcp->inject_vec[current_personality] =
501                         xcalloc(nsyscalls, sizeof(**inject_vec));
502                 memcpy(tcp->inject_vec[current_personality],
503                        inject_vec[current_personality],
504                        nsyscalls * sizeof(**inject_vec));
505         }
506
507         struct inject_opts *opts = tcb_inject_opts(tcp);
508
509         if (!opts || opts->first == 0)
510                 return 0;
511
512         --opts->first;
513
514         if (opts->first != 0)
515                 return 0;
516
517         opts->first = opts->step;
518
519         if (!recovering(tcp)) {
520                 if (opts->data.flags & INJECT_F_SIGNAL)
521                         *signo = opts->data.signo;
522                 if (opts->data.flags & (INJECT_F_ERROR | INJECT_F_RETVAL)) {
523                         kernel_long_t scno =
524                                 (opts->data.flags & INJECT_F_SYSCALL)
525                                 ? (kernel_long_t) shuffle_scno(opts->data.scno)
526                                 : -1;
527
528                         if (!arch_set_scno(tcp, scno)) {
529                                 tcp->flags |= TCB_TAMPERED;
530                                 if (scno != -1)
531                                         tcp->flags |= TCB_TAMPERED_NO_FAIL;
532                         }
533                 }
534                 if (opts->data.flags & INJECT_F_DELAY_ENTER)
535                         delay_tcb(tcp, opts->data.delay_idx, true);
536                 if (opts->data.flags & INJECT_F_DELAY_EXIT)
537                         tcp->flags |= TCB_INJECT_DELAY_EXIT;
538         }
539
540         return 0;
541 }
542
543 static long
544 tamper_with_syscall_exiting(struct tcb *tcp)
545 {
546         struct inject_opts *opts = tcb_inject_opts(tcp);
547         if (!opts)
548                 return 0;
549
550         if (inject_delay_exit(tcp))
551                 delay_tcb(tcp, opts->data.delay_idx, false);
552
553         if (!syscall_tampered(tcp))
554                 return 0;
555
556         if (!syserror(tcp) ^ !!syscall_tampered_nofail(tcp)) {
557                 error_msg("Failed to tamper with process %d: unexpectedly got"
558                           " %serror (return value %#" PRI_klx ", error %lu)",
559                           tcp->pid, syscall_tampered_nofail(tcp) ? "" : "no ",
560                           tcp->u_rval, tcp->u_error);
561
562                 return 1;
563         }
564
565         bool update_tcb = false;
566
567         if (opts->data.flags & INJECT_F_RETVAL) {
568                 kernel_long_t inject_rval =
569                         retval_get(opts->data.rval_idx);
570                 kernel_long_t u_rval = tcp->u_rval;
571
572                 tcp->u_rval = inject_rval;
573                 if (arch_set_success(tcp)) {
574                         tcp->u_rval = u_rval;
575                 } else {
576                         update_tcb = true;
577                         tcp->u_error = 0;
578                 }
579         } else {
580                 unsigned long new_error = retval_get(opts->data.rval_idx);
581
582                 if (new_error != tcp->u_error && new_error <= MAX_ERRNO_VALUE) {
583                         unsigned long u_error = tcp->u_error;
584
585                         tcp->u_error = new_error;
586                         if (arch_set_error(tcp)) {
587                                 tcp->u_error = u_error;
588                         } else {
589                                 update_tcb = true;
590                         }
591                 }
592         }
593
594         if (update_tcb) {
595                 tcp->u_error = 0;
596                 get_error(tcp, !(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS));
597         }
598
599         return 0;
600 }
601
602 /*
603  * Returns:
604  * 0: "ignore this ptrace stop", bail out silently.
605  * 1: ok, decoded; call
606  *    syscall_entering_finish(tcp, syscall_entering_trace(tcp, ...)).
607  * other: error; call syscall_entering_finish(tcp, res), where res is the value
608  *    returned.
609  */
610 int
611 syscall_entering_decode(struct tcb *tcp)
612 {
613         int res = get_scno(tcp);
614         if (res == 0)
615                 return res;
616         int scno_good = res;
617         if (res != 1 || (res = get_syscall_args(tcp)) != 1) {
618                 printleader(tcp);
619                 tprintf("%s(", scno_good == 1 ? tcp->s_ent->sys_name : "????");
620                 /*
621                  * " <unavailable>" will be added later by the code which
622                  * detects ptrace errors.
623                  */
624                 return res;
625         }
626
627 #if defined SYS_ipc_subcall     \
628  || defined SYS_socket_subcall  \
629  || defined SYS_syscall_subcall
630         for (;;) {
631                 switch (tcp->s_ent->sen) {
632 # ifdef SYS_ipc_subcall
633                 case SEN_ipc:
634                         decode_ipc_subcall(tcp);
635                         break;
636 # endif
637 # ifdef SYS_socket_subcall
638                 case SEN_socketcall:
639                         decode_socket_subcall(tcp);
640                         break;
641 # endif
642 # ifdef SYS_syscall_subcall
643                 case SEN_syscall:
644                         decode_syscall_subcall(tcp);
645                         if (tcp->s_ent->sen != SEN_syscall)
646                                 continue;
647                         break;
648 # endif
649                 }
650                 break;
651         }
652 #endif
653
654         return 1;
655 }
656
657 int
658 syscall_entering_trace(struct tcb *tcp, unsigned int *sig)
659 {
660         /* Restrain from fault injection while the trace executes strace code. */
661         if (hide_log(tcp)) {
662                 tcp->qual_flg &= ~QUAL_INJECT;
663         }
664
665         switch (tcp->s_ent->sen) {
666                 case SEN_execve:
667                 case SEN_execveat:
668 #if defined SPARC || defined SPARC64
669                 case SEN_execv:
670 #endif
671                         tcp->flags &= ~TCB_HIDE_LOG;
672                         break;
673         }
674
675         if (!traced(tcp) || (tracing_paths && !pathtrace_match(tcp))) {
676                 tcp->flags |= TCB_FILTERED;
677                 return 0;
678         }
679
680         tcp->flags &= ~TCB_FILTERED;
681
682         if (hide_log(tcp)) {
683                 return 0;
684         }
685
686         if (inject(tcp))
687                 tamper_with_syscall_entering(tcp, sig);
688
689         if (cflag == CFLAG_ONLY_STATS) {
690                 return 0;
691         }
692
693 #ifdef ENABLE_STACKTRACE
694         if (stack_trace_enabled) {
695                 if (tcp->s_ent->sys_flags & STACKTRACE_CAPTURE_ON_ENTER)
696                         unwind_tcb_capture(tcp);
697         }
698 #endif
699
700         printleader(tcp);
701         tprintf("%s(", tcp->s_ent->sys_name);
702         int res = raw(tcp) ? printargs(tcp) : tcp->s_ent->sys_func(tcp);
703         fflush(tcp->outf);
704         return res;
705 }
706
707 void
708 syscall_entering_finish(struct tcb *tcp, int res)
709 {
710         tcp->flags |= TCB_INSYSCALL;
711         tcp->sys_func_rval = res;
712         /* Measure the entrance time as late as possible to avoid errors. */
713         if ((Tflag || cflag) && !filtered(tcp))
714                 clock_gettime(CLOCK_MONOTONIC, &tcp->etime);
715 }
716
717 /* Returns:
718  * 0: "bail out".
719  * 1: ok.
720  * -1: error in one of ptrace ops.
721  *
722  * If not 0, call syscall_exiting_trace(tcp, res), where res is the return
723  *    value. Anyway, call syscall_exiting_finish(tcp) then.
724  */
725 int
726 syscall_exiting_decode(struct tcb *tcp, struct timespec *pts)
727 {
728         /* Measure the exit time as early as possible to avoid errors. */
729         if ((Tflag || cflag) && !(filtered(tcp) || hide_log(tcp)))
730                 clock_gettime(CLOCK_MONOTONIC, pts);
731
732         if (tcp->s_ent->sys_flags & MEMORY_MAPPING_CHANGE)
733                 mmap_notify_report(tcp);
734
735         if (filtered(tcp) || hide_log(tcp))
736                 return 0;
737
738 #if SUPPORTED_PERSONALITIES > 1
739         update_personality(tcp, tcp->currpers);
740 #endif
741
742         return get_syscall_result(tcp);
743 }
744
745 int
746 syscall_exiting_trace(struct tcb *tcp, struct timespec *ts, int res)
747 {
748         if (syscall_tampered(tcp) || inject_delay_exit(tcp))
749                 tamper_with_syscall_exiting(tcp);
750
751         if (cflag) {
752                 count_syscall(tcp, ts);
753                 if (cflag == CFLAG_ONLY_STATS) {
754                         return 0;
755                 }
756         }
757
758         /* If not in -ff mode, and printing_tcp != tcp,
759          * then the log currently does not end with output
760          * of _our syscall entry_, but with something else.
761          * We need to say which syscall's return is this.
762          *
763          * Forced reprinting via TCB_REPRINT is used only by
764          * "strace -ff -oLOG test/threaded_execve" corner case.
765          * It's the only case when -ff mode needs reprinting.
766          */
767         if ((followfork < 2 && printing_tcp != tcp) || (tcp->flags & TCB_REPRINT)) {
768                 tcp->flags &= ~TCB_REPRINT;
769                 printleader(tcp);
770                 tprintf("<... %s resumed> ", tcp->s_ent->sys_name);
771         }
772         printing_tcp = tcp;
773
774         tcp->s_prev_ent = NULL;
775         if (res != 1) {
776                 /* There was error in one of prior ptrace ops */
777                 tprints(") ");
778                 tabto();
779                 tprints("= ? <unavailable>\n");
780                 line_ended();
781                 return res;
782         }
783         tcp->s_prev_ent = tcp->s_ent;
784
785         int sys_res = 0;
786         if (raw(tcp)) {
787                 /* sys_res = printargs(tcp); - but it's nop on sysexit */
788         } else {
789         /* FIXME: not_failing_only (IOW, option -z) is broken:
790          * failure of syscall is known only after syscall return.
791          * Thus we end up with something like this on, say, ENOENT:
792          *     open("does_not_exist", O_RDONLY <unfinished ...>
793          *     {next syscall decode}
794          * whereas the intended result is that open(...) line
795          * is not shown at all.
796          */
797                 if (not_failing_only && tcp->u_error)
798                         return 0;       /* ignore failed syscalls */
799                 if (tcp->sys_func_rval & RVAL_DECODED)
800                         sys_res = tcp->sys_func_rval;
801                 else
802                         sys_res = tcp->s_ent->sys_func(tcp);
803         }
804
805         tprints(") ");
806         tabto();
807
808         if (raw(tcp)) {
809                 if (tcp->u_error)
810                         print_err_ret(tcp->u_rval, tcp->u_error);
811                 else
812                         tprintf("= %#" PRI_klx, tcp->u_rval);
813
814                 if (syscall_tampered(tcp))
815                         tprints(" (INJECTED)");
816         } else if (!(sys_res & RVAL_NONE) && tcp->u_error) {
817                 switch (tcp->u_error) {
818                 /* Blocked signals do not interrupt any syscalls.
819                  * In this case syscalls don't return ERESTARTfoo codes.
820                  *
821                  * Deadly signals set to SIG_DFL interrupt syscalls
822                  * and kill the process regardless of which of the codes below
823                  * is returned by the interrupted syscall.
824                  * In some cases, kernel forces a kernel-generated deadly
825                  * signal to be unblocked and set to SIG_DFL (and thus cause
826                  * death) if it is blocked or SIG_IGNed: for example, SIGSEGV
827                  * or SIGILL. (The alternative is to leave process spinning
828                  * forever on the faulty instruction - not useful).
829                  *
830                  * SIG_IGNed signals and non-deadly signals set to SIG_DFL
831                  * (for example, SIGCHLD, SIGWINCH) interrupt syscalls,
832                  * but kernel will always restart them.
833                  */
834                 case ERESTARTSYS:
835                         /* Most common type of signal-interrupted syscall exit code.
836                          * The system call will be restarted with the same arguments
837                          * if SA_RESTART is set; otherwise, it will fail with EINTR.
838                          */
839                         tprints("= ? ERESTARTSYS (To be restarted if SA_RESTART is set)");
840                         break;
841                 case ERESTARTNOINTR:
842                         /* Rare. For example, fork() returns this if interrupted.
843                          * SA_RESTART is ignored (assumed set): the restart is unconditional.
844                          */
845                         tprints("= ? ERESTARTNOINTR (To be restarted)");
846                         break;
847                 case ERESTARTNOHAND:
848                         /* pause(), rt_sigsuspend() etc use this code.
849                          * SA_RESTART is ignored (assumed not set):
850                          * syscall won't restart (will return EINTR instead)
851                          * even after signal with SA_RESTART set. However,
852                          * after SIG_IGN or SIG_DFL signal it will restart
853                          * (thus the name "restart only if has no handler").
854                          */
855                         tprints("= ? ERESTARTNOHAND (To be restarted if no handler)");
856                         break;
857                 case ERESTART_RESTARTBLOCK:
858                         /* Syscalls like nanosleep(), poll() which can't be
859                          * restarted with their original arguments use this
860                          * code. Kernel will execute restart_syscall() instead,
861                          * which changes arguments before restarting syscall.
862                          * SA_RESTART is ignored (assumed not set) similarly
863                          * to ERESTARTNOHAND. (Kernel can't honor SA_RESTART
864                          * since restart data is saved in "restart block"
865                          * in task struct, and if signal handler uses a syscall
866                          * which in turn saves another such restart block,
867                          * old data is lost and restart becomes impossible)
868                          */
869                         tprints("= ? ERESTART_RESTARTBLOCK (Interrupted by signal)");
870                         break;
871                 default:
872                         print_err_ret(tcp->u_rval, tcp->u_error);
873                         break;
874                 }
875                 if (syscall_tampered(tcp))
876                         tprints(" (INJECTED)");
877                 if ((sys_res & RVAL_STR) && tcp->auxstr)
878                         tprintf(" (%s)", tcp->auxstr);
879         } else {
880                 if (sys_res & RVAL_NONE)
881                         tprints("= ?");
882                 else {
883                         switch (sys_res & RVAL_MASK) {
884                         case RVAL_HEX:
885 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
886                                 if (current_klongsize < sizeof(tcp->u_rval)) {
887                                         tprintf("= %#x",
888                                                 (unsigned int) tcp->u_rval);
889                                 } else
890 #endif
891                                 {
892                                         tprintf("= %#" PRI_klx, tcp->u_rval);
893                                 }
894                                 break;
895                         case RVAL_OCTAL:
896                                 tprints("= ");
897                                 print_numeric_long_umask(tcp->u_rval);
898                                 break;
899                         case RVAL_UDECIMAL:
900 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
901                                 if (current_klongsize < sizeof(tcp->u_rval)) {
902                                         tprintf("= %u",
903                                                 (unsigned int) tcp->u_rval);
904                                 } else
905 #endif
906                                 {
907                                         tprintf("= %" PRI_klu, tcp->u_rval);
908                                 }
909                                 break;
910                         case RVAL_FD:
911                                 if (show_fd_path) {
912                                         tprints("= ");
913                                         printfd(tcp, tcp->u_rval);
914                                 } else
915                                         tprintf("= %" PRI_kld, tcp->u_rval);
916                                 break;
917                         default:
918                                 error_msg("invalid rval format");
919                                 break;
920                         }
921                 }
922                 if ((sys_res & RVAL_STR) && tcp->auxstr)
923                         tprintf(" (%s)", tcp->auxstr);
924                 if (syscall_tampered(tcp))
925                         tprints(" (INJECTED)");
926         }
927         if (Tflag) {
928                 ts_sub(ts, ts, &tcp->etime);
929                 tprintf(" <%ld.%06ld>",
930                         (long) ts->tv_sec, (long) ts->tv_nsec / 1000);
931         }
932         tprints("\n");
933         dumpio(tcp);
934         line_ended();
935
936 #ifdef ENABLE_STACKTRACE
937         if (stack_trace_enabled)
938                 unwind_tcb_print(tcp);
939 #endif
940         return 0;
941 }
942
943 void
944 syscall_exiting_finish(struct tcb *tcp)
945 {
946         tcp->flags &= ~(TCB_INSYSCALL | TCB_TAMPERED | TCB_INJECT_DELAY_EXIT);
947         tcp->sys_func_rval = 0;
948         free_tcb_priv_data(tcp);
949 }
950
951 bool
952 is_erestart(struct tcb *tcp)
953 {
954         switch (tcp->u_error) {
955                 case ERESTARTSYS:
956                 case ERESTARTNOINTR:
957                 case ERESTARTNOHAND:
958                 case ERESTART_RESTARTBLOCK:
959                         return true;
960                 default:
961                         return false;
962         }
963 }
964
965 static unsigned long saved_u_error;
966
967 void
968 temporarily_clear_syserror(struct tcb *tcp)
969 {
970         saved_u_error = tcp->u_error;
971         tcp->u_error = 0;
972 }
973
974 void
975 restore_cleared_syserror(struct tcb *tcp)
976 {
977         tcp->u_error = saved_u_error;
978 }
979
980 #define XLAT_MACROS_ONLY
981 # include "xlat/nt_descriptor_types.h"
982 #undef XLAT_MACROS_ONLY
983
984 #include "arch_regs.c"
985
986 #if HAVE_ARCH_GETRVAL2
987 # include "arch_getrval2.c"
988 #endif
989
990 void
991 print_pc(struct tcb *tcp)
992 {
993 #if defined ARCH_PC_REG
994 # define ARCH_GET_PC 0
995 #elif defined ARCH_PC_PEEK_ADDR
996         kernel_ulong_t pc;
997 # define ARCH_PC_REG pc
998 # define ARCH_GET_PC upeek(tcp, ARCH_PC_PEEK_ADDR, &pc)
999 #else
1000 # error Neither ARCH_PC_REG nor ARCH_PC_PEEK_ADDR is defined
1001 #endif
1002         if (get_regs(tcp) < 0 || ARCH_GET_PC)
1003                 tprints(current_wordsize == 4 ? "[????????] "
1004                                               : "[????????????????] ");
1005         else
1006                 tprintf(current_wordsize == 4
1007                         ? "[%08" PRI_klx "] " : "[%016" PRI_klx "] ",
1008                         (kernel_ulong_t) ARCH_PC_REG);
1009 }
1010
1011 #include "getregs_old.h"
1012
1013 #undef ptrace_getregset_or_getregs
1014 #undef ptrace_setregset_or_setregs
1015 #ifdef ARCH_REGS_FOR_GETREGSET
1016
1017 # define ptrace_getregset_or_getregs ptrace_getregset
1018 static long
1019 ptrace_getregset(pid_t pid)
1020 {
1021 # ifdef ARCH_IOVEC_FOR_GETREGSET
1022         /* variable iovec */
1023         ARCH_IOVEC_FOR_GETREGSET.iov_len = sizeof(ARCH_REGS_FOR_GETREGSET);
1024         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS,
1025                       &ARCH_IOVEC_FOR_GETREGSET);
1026 # else
1027         /* constant iovec */
1028         static struct iovec io = {
1029                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1030                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1031         };
1032         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS, &io);
1033
1034 # endif
1035 }
1036
1037 # ifndef HAVE_GETREGS_OLD
1038 #  define ptrace_setregset_or_setregs ptrace_setregset
1039 static int
1040 ptrace_setregset(pid_t pid)
1041 {
1042 #  ifdef ARCH_IOVEC_FOR_GETREGSET
1043         /* variable iovec */
1044         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS,
1045                       &ARCH_IOVEC_FOR_GETREGSET);
1046 #  else
1047         /* constant iovec */
1048         static struct iovec io = {
1049                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1050                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1051         };
1052         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS, &io);
1053 #  endif
1054 }
1055 # endif /* !HAVE_GETREGS_OLD */
1056
1057 #elif defined ARCH_REGS_FOR_GETREGS
1058
1059 # define ptrace_getregset_or_getregs ptrace_getregs
1060 static long
1061 ptrace_getregs(pid_t pid)
1062 {
1063 # if defined SPARC || defined SPARC64
1064         /* SPARC systems have the meaning of data and addr reversed */
1065         return ptrace(PTRACE_GETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1066 # else
1067         return ptrace(PTRACE_GETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1068 # endif
1069 }
1070
1071 # ifndef HAVE_GETREGS_OLD
1072 #  define ptrace_setregset_or_setregs ptrace_setregs
1073 static int
1074 ptrace_setregs(pid_t pid)
1075 {
1076 #  if defined SPARC || defined SPARC64
1077         /* SPARC systems have the meaning of data and addr reversed */
1078         return ptrace(PTRACE_SETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1079 #  else
1080         return ptrace(PTRACE_SETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1081 #  endif
1082 }
1083 # endif /* !HAVE_GETREGS_OLD */
1084
1085 #endif /* ARCH_REGS_FOR_GETREGSET || ARCH_REGS_FOR_GETREGS */
1086
1087 #ifdef ptrace_getregset_or_getregs
1088 static long get_regs_error;
1089 #endif
1090
1091 void
1092 clear_regs(struct tcb *tcp)
1093 {
1094 #ifdef ptrace_getregset_or_getregs
1095         get_regs_error = -1;
1096 #endif
1097 }
1098
1099 static long
1100 get_regs(struct tcb *const tcp)
1101 {
1102 #ifdef ptrace_getregset_or_getregs
1103
1104         if (get_regs_error != -1)
1105                 return get_regs_error;
1106
1107 # ifdef HAVE_GETREGS_OLD
1108         /*
1109          * Try PTRACE_GETREGSET/PTRACE_GETREGS first,
1110          * fallback to getregs_old.
1111          */
1112         static int use_getregs_old;
1113         if (use_getregs_old < 0) {
1114                 return get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1115         } else if (use_getregs_old == 0) {
1116                 get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1117                 if (get_regs_error >= 0) {
1118                         use_getregs_old = -1;
1119                         return get_regs_error;
1120                 }
1121                 if (errno == EPERM || errno == ESRCH)
1122                         return get_regs_error;
1123                 use_getregs_old = 1;
1124         }
1125         return get_regs_error = getregs_old(tcp);
1126 # else /* !HAVE_GETREGS_OLD */
1127         /* Assume that PTRACE_GETREGSET/PTRACE_GETREGS works. */
1128         get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1129
1130 #  if defined ARCH_PERSONALITY_0_IOV_SIZE
1131         if (get_regs_error)
1132                 return get_regs_error;
1133
1134         switch (ARCH_IOVEC_FOR_GETREGSET.iov_len) {
1135         case ARCH_PERSONALITY_0_IOV_SIZE:
1136                 update_personality(tcp, 0);
1137                 break;
1138         case ARCH_PERSONALITY_1_IOV_SIZE:
1139                 update_personality(tcp, 1);
1140                 break;
1141         default: {
1142                 static bool printed = false;
1143
1144                 if (!printed) {
1145                         error_msg("Unsupported regset size returned by "
1146                                   "PTRACE_GETREGSET: %zu",
1147                                   ARCH_IOVEC_FOR_GETREGSET.iov_len);
1148
1149                         printed = true;
1150                 }
1151
1152                 update_personality(tcp, 0);
1153         }
1154         }
1155 #  endif /* ARCH_PERSONALITY_0_IOV_SIZE */
1156
1157         return get_regs_error;
1158
1159 # endif /* !HAVE_GETREGS_OLD */
1160
1161 #else /* !ptrace_getregset_or_getregs */
1162
1163 # warning get_regs is not implemented for this architecture yet
1164         return 0;
1165
1166 #endif /* !ptrace_getregset_or_getregs */
1167 }
1168
1169 #ifdef ptrace_setregset_or_setregs
1170 static int
1171 set_regs(pid_t pid)
1172 {
1173         return ptrace_setregset_or_setregs(pid);
1174 }
1175 #endif /* ptrace_setregset_or_setregs */
1176
1177 struct sysent_buf {
1178         struct tcb *tcp;
1179         struct_sysent ent;
1180         char buf[sizeof("syscall_0x") + sizeof(kernel_ulong_t) * 2];
1181 };
1182
1183 static void
1184 free_sysent_buf(void *ptr)
1185 {
1186         struct sysent_buf *s = ptr;
1187         s->tcp->s_prev_ent = s->tcp->s_ent = NULL;
1188         free(ptr);
1189 }
1190
1191 /*
1192  * Returns:
1193  * 0: "ignore this ptrace stop", syscall_entering_decode() should return a "bail
1194  *    out silently" code.
1195  * 1: ok, continue in syscall_entering_decode().
1196  * other: error, syscall_entering_decode() should print error indicator
1197  *    ("????" etc) and return an appropriate code.
1198  */
1199 int
1200 get_scno(struct tcb *tcp)
1201 {
1202         if (get_regs(tcp) < 0)
1203                 return -1;
1204
1205         int rc = arch_get_scno(tcp);
1206         if (rc != 1)
1207                 return rc;
1208
1209         tcp->scno = shuffle_scno(tcp->scno);
1210
1211         if (scno_is_valid(tcp->scno)) {
1212                 tcp->s_ent = &sysent[tcp->scno];
1213                 tcp->qual_flg = qual_flags(tcp->scno);
1214         } else {
1215                 struct sysent_buf *s = xcalloc(1, sizeof(*s));
1216
1217                 s->tcp = tcp;
1218                 s->ent.nargs = MAX_ARGS;
1219                 s->ent.sen = SEN_printargs;
1220                 s->ent.sys_func = printargs;
1221                 s->ent.sys_name = s->buf;
1222                 xsprintf(s->buf, "syscall_%#" PRI_klx, shuffle_scno(tcp->scno));
1223
1224                 tcp->s_ent = &s->ent;
1225                 tcp->qual_flg = QUAL_RAW | DEFAULT_QUAL_FLAGS;
1226
1227                 set_tcb_priv_data(tcp, s, free_sysent_buf);
1228
1229                 debug_msg("pid %d invalid syscall %#" PRI_klx,
1230                           tcp->pid, shuffle_scno(tcp->scno));
1231         }
1232
1233         /*
1234          * We refrain from argument decoding during recovering
1235          * as tracee memory mappings has changed and the registers
1236          * are very likely pointing to garbage already.
1237          */
1238         if (recovering(tcp))
1239                 tcp->qual_flg |= QUAL_RAW;
1240
1241         return 1;
1242 }
1243
1244 #ifdef ptrace_getregset_or_getregs
1245 # define get_syscall_result_regs get_regs
1246 #else
1247 static int get_syscall_result_regs(struct tcb *);
1248 #endif
1249
1250 /* Returns:
1251  * 1: ok, continue in syscall_exiting_trace().
1252  * -1: error, syscall_exiting_trace() should print error indicator
1253  *    ("????" etc) and bail out.
1254  */
1255 static int
1256 get_syscall_result(struct tcb *tcp)
1257 {
1258         if (get_syscall_result_regs(tcp) < 0)
1259                 return -1;
1260         tcp->u_error = 0;
1261         get_error(tcp,
1262                   (!(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS)
1263                         || syscall_tampered(tcp))
1264                   && !syscall_tampered_nofail(tcp));
1265
1266         return 1;
1267 }
1268
1269 #include "get_scno.c"
1270 #include "set_scno.c"
1271 #include "get_syscall_args.c"
1272 #ifndef ptrace_getregset_or_getregs
1273 # include "get_syscall_result.c"
1274 #endif
1275 #include "get_error.c"
1276 #include "set_error.c"
1277 #ifdef HAVE_GETREGS_OLD
1278 # include "getregs_old.c"
1279 #endif
1280 #include "shuffle_scno.c"
1281
1282 const char *
1283 syscall_name(kernel_ulong_t scno)
1284 {
1285         return scno_is_valid(scno) ? sysent[scno].sys_name : NULL;
1286 }