]> granicus.if.org Git - strace/blob - syscall.c
mem: handle potential error from sysconf(_SC_PAGESIZE)
[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_cache.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 #ifndef NT_PRSTATUS
61 # define NT_PRSTATUS 1
62 #endif
63
64 #include "syscall.h"
65 #include "xstring.h"
66
67 /* Define these shorthand notations to simplify the syscallent files. */
68 #include "sysent_shorthand_defs.h"
69
70 #define SEN(syscall_name) SEN_ ## syscall_name, SYS_FUNC_NAME(sys_ ## syscall_name)
71
72 const struct_sysent sysent0[] = {
73 #include "syscallent.h"
74 };
75
76 #if SUPPORTED_PERSONALITIES > 1
77 # include PERSONALITY1_INCLUDE_FUNCS
78 static const struct_sysent sysent1[] = {
79 # include "syscallent1.h"
80 };
81 #endif
82
83 #if SUPPORTED_PERSONALITIES > 2
84 # include PERSONALITY2_INCLUDE_FUNCS
85 static const struct_sysent sysent2[] = {
86 # include "syscallent2.h"
87 };
88 #endif
89
90 /* Now undef them since short defines cause wicked namespace pollution. */
91 #include "sysent_shorthand_undefs.h"
92
93 /*
94  * `ioctlent[012].h' files are automatically generated by the auxiliary
95  * program `ioctlsort', such that the list is sorted by the `code' field.
96  * This has the side-effect of resolving the _IO.. macros into
97  * plain integers, eliminating the need to include here everything
98  * in "/usr/include".
99  */
100
101 const char *const errnoent0[] = {
102 #include "errnoent.h"
103 };
104 const char *const signalent0[] = {
105 #include "signalent.h"
106 };
107 const struct_ioctlent ioctlent0[] = {
108 #include "ioctlent0.h"
109 };
110
111 #if SUPPORTED_PERSONALITIES > 1
112 static const char *const errnoent1[] = {
113 # include "errnoent1.h"
114 };
115 static const char *const signalent1[] = {
116 # include "signalent1.h"
117 };
118 static const struct_ioctlent ioctlent1[] = {
119 # include "ioctlent1.h"
120 };
121 # include PERSONALITY0_INCLUDE_PRINTERS_DECLS
122 static const struct_printers printers0 = {
123 # include PERSONALITY0_INCLUDE_PRINTERS_DEFS
124 };
125 # include PERSONALITY1_INCLUDE_PRINTERS_DECLS
126 static const struct_printers printers1 = {
127 # include PERSONALITY1_INCLUDE_PRINTERS_DEFS
128 };
129 #endif
130
131 #if SUPPORTED_PERSONALITIES > 2
132 static const char *const errnoent2[] = {
133 # include "errnoent2.h"
134 };
135 static const char *const signalent2[] = {
136 # include "signalent2.h"
137 };
138 static const struct_ioctlent ioctlent2[] = {
139 # include "ioctlent2.h"
140 };
141 # include PERSONALITY2_INCLUDE_PRINTERS_DECLS
142 static const struct_printers printers2 = {
143 # include PERSONALITY2_INCLUDE_PRINTERS_DEFS
144 };
145 #endif
146
147 enum {
148         nsyscalls0 = ARRAY_SIZE(sysent0)
149 #if SUPPORTED_PERSONALITIES > 1
150         , nsyscalls1 = ARRAY_SIZE(sysent1)
151 # if SUPPORTED_PERSONALITIES > 2
152         , nsyscalls2 = ARRAY_SIZE(sysent2)
153 # endif
154 #endif
155 };
156
157 enum {
158         nerrnos0 = ARRAY_SIZE(errnoent0)
159 #if SUPPORTED_PERSONALITIES > 1
160         , nerrnos1 = ARRAY_SIZE(errnoent1)
161 # if SUPPORTED_PERSONALITIES > 2
162         , nerrnos2 = ARRAY_SIZE(errnoent2)
163 # endif
164 #endif
165 };
166
167 enum {
168         nsignals0 = ARRAY_SIZE(signalent0)
169 #if SUPPORTED_PERSONALITIES > 1
170         , nsignals1 = ARRAY_SIZE(signalent1)
171 # if SUPPORTED_PERSONALITIES > 2
172         , nsignals2 = ARRAY_SIZE(signalent2)
173 # endif
174 #endif
175 };
176
177 enum {
178         nioctlents0 = ARRAY_SIZE(ioctlent0)
179 #if SUPPORTED_PERSONALITIES > 1
180         , nioctlents1 = ARRAY_SIZE(ioctlent1)
181 # if SUPPORTED_PERSONALITIES > 2
182         , nioctlents2 = ARRAY_SIZE(ioctlent2)
183 # endif
184 #endif
185 };
186
187 #if SUPPORTED_PERSONALITIES > 1
188 const struct_sysent *sysent = sysent0;
189 const char *const *errnoent = errnoent0;
190 const char *const *signalent = signalent0;
191 const struct_ioctlent *ioctlent = ioctlent0;
192 const struct_printers *printers = &printers0;
193 #endif
194
195 unsigned nsyscalls = nsyscalls0;
196 unsigned nerrnos = nerrnos0;
197 unsigned nsignals = nsignals0;
198 unsigned nioctlents = nioctlents0;
199
200 const unsigned int nsyscall_vec[SUPPORTED_PERSONALITIES] = {
201         nsyscalls0,
202 #if SUPPORTED_PERSONALITIES > 1
203         nsyscalls1,
204 #endif
205 #if SUPPORTED_PERSONALITIES > 2
206         nsyscalls2,
207 #endif
208 };
209 const struct_sysent *const sysent_vec[SUPPORTED_PERSONALITIES] = {
210         sysent0,
211 #if SUPPORTED_PERSONALITIES > 1
212         sysent1,
213 #endif
214 #if SUPPORTED_PERSONALITIES > 2
215         sysent2,
216 #endif
217 };
218
219 const char *const personality_names[] =
220 # if defined X86_64
221         {"64 bit", "32 bit", "x32"}
222 # elif defined X32
223         {"x32", "32 bit"}
224 # elif SUPPORTED_PERSONALITIES == 2
225         {"64 bit", "32 bit"}
226 # else
227         {STRINGIFY_VAL(__WORDSIZE) " bit"}
228 # endif
229         ;
230
231 #if SUPPORTED_PERSONALITIES > 1
232
233 unsigned current_personality;
234
235 # ifndef current_wordsize
236 unsigned current_wordsize = PERSONALITY0_WORDSIZE;
237 static const int personality_wordsize[SUPPORTED_PERSONALITIES] = {
238         PERSONALITY0_WORDSIZE,
239         PERSONALITY1_WORDSIZE,
240 # if SUPPORTED_PERSONALITIES > 2
241         PERSONALITY2_WORDSIZE,
242 # endif
243 };
244 # endif
245
246 # ifndef current_klongsize
247 unsigned current_klongsize = PERSONALITY0_KLONGSIZE;
248 static const int personality_klongsize[SUPPORTED_PERSONALITIES] = {
249         PERSONALITY0_KLONGSIZE,
250         PERSONALITY1_KLONGSIZE,
251 #  if SUPPORTED_PERSONALITIES > 2
252         PERSONALITY2_KLONGSIZE,
253 #  endif
254 };
255 # endif
256
257 void
258 set_personality(unsigned int personality)
259 {
260         if (personality == current_personality)
261                 return;
262
263         if (personality >= SUPPORTED_PERSONALITIES)
264                 error_msg_and_die("Requested switch to unsupported personality "
265                                   "%u", personality);
266
267         nsyscalls = nsyscall_vec[personality];
268         sysent = sysent_vec[personality];
269
270         switch (personality) {
271         case 0:
272                 errnoent = errnoent0;
273                 nerrnos = nerrnos0;
274                 ioctlent = ioctlent0;
275                 nioctlents = nioctlents0;
276                 signalent = signalent0;
277                 nsignals = nsignals0;
278                 printers = &printers0;
279                 break;
280
281         case 1:
282                 errnoent = errnoent1;
283                 nerrnos = nerrnos1;
284                 ioctlent = ioctlent1;
285                 nioctlents = nioctlents1;
286                 signalent = signalent1;
287                 nsignals = nsignals1;
288                 printers = &printers1;
289                 break;
290
291 # if SUPPORTED_PERSONALITIES > 2
292         case 2:
293                 errnoent = errnoent2;
294                 nerrnos = nerrnos2;
295                 ioctlent = ioctlent2;
296                 nioctlents = nioctlents2;
297                 signalent = signalent2;
298                 nsignals = nsignals2;
299                 printers = &printers2;
300                 break;
301 # endif
302         }
303
304         current_personality = personality;
305 # ifndef current_wordsize
306         current_wordsize = personality_wordsize[personality];
307 # endif
308 # ifndef current_klongsize
309         current_klongsize = personality_klongsize[personality];
310 # endif
311 }
312
313 static void
314 update_personality(struct tcb *tcp, unsigned int personality)
315 {
316         static bool need_mpers_warning[] =
317                 { false, !HAVE_PERSONALITY_1_MPERS, !HAVE_PERSONALITY_2_MPERS };
318
319         set_personality(personality);
320
321         if (personality == tcp->currpers)
322                 return;
323         tcp->currpers = personality;
324
325         if (!qflag) {
326                 error_msg("[ Process PID=%d runs in %s mode. ]",
327                           tcp->pid, personality_names[personality]);
328         }
329
330         if (need_mpers_warning[personality]) {
331                 error_msg("WARNING: Proper structure decoding for this "
332                           "personality is not supported, please consider "
333                           "building strace with mpers support enabled.");
334                 need_mpers_warning[personality] = false;
335         }
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 kernel_ulong_t 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 /* SYS_socket_subcall */
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 /* SYS_ipc_subcall */
405
406 #ifdef SYS_syscall_subcall
407 static void
408 decode_syscall_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 # ifdef LINUX_MIPSO32
418         /*
419          * Fetching the last arg of 7-arg syscalls (fadvise64_64
420          * and sync_file_range) requires additional code,
421          * see linux/mips/get_syscall_args.c
422          */
423         if (tcp->s_ent->nargs == MAX_ARGS) {
424                 if (umoven(tcp,
425                            mips_REG_SP + MAX_ARGS * sizeof(tcp->u_arg[0]),
426                            sizeof(tcp->u_arg[0]),
427                            &tcp->u_arg[MAX_ARGS - 1]) < 0)
428                 tcp->u_arg[MAX_ARGS - 1] = 0;
429         }
430 # endif /* LINUX_MIPSO32 */
431 }
432 #endif /* SYS_syscall_subcall */
433
434 static void
435 dumpio(struct tcb *tcp)
436 {
437         int fd = tcp->u_arg[0];
438         if (fd < 0)
439                 return;
440
441         if (is_number_in_set(fd, write_set)) {
442                 switch (tcp->s_ent->sen) {
443                 case SEN_write:
444                 case SEN_pwrite:
445                 case SEN_send:
446                 case SEN_sendto:
447                 case SEN_mq_timedsend:
448                         dumpstr(tcp, tcp->u_arg[1], tcp->u_arg[2]);
449                         break;
450                 case SEN_writev:
451                 case SEN_pwritev:
452                 case SEN_pwritev2:
453                 case SEN_vmsplice:
454                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1], -1);
455                         break;
456                 case SEN_sendmsg:
457                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], -1);
458                         break;
459                 case SEN_sendmmsg:
460                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
461                         break;
462                 }
463         }
464
465         if (syserror(tcp))
466                 return;
467
468         if (is_number_in_set(fd, read_set)) {
469                 switch (tcp->s_ent->sen) {
470                 case SEN_read:
471                 case SEN_pread:
472                 case SEN_recv:
473                 case SEN_recvfrom:
474                 case SEN_mq_timedreceive:
475                         dumpstr(tcp, tcp->u_arg[1], tcp->u_rval);
476                         return;
477                 case SEN_readv:
478                 case SEN_preadv:
479                 case SEN_preadv2:
480                         dumpiov_upto(tcp, tcp->u_arg[2], tcp->u_arg[1],
481                                      tcp->u_rval);
482                         return;
483                 case SEN_recvmsg:
484                         dumpiov_in_msghdr(tcp, tcp->u_arg[1], tcp->u_rval);
485                         return;
486                 case SEN_recvmmsg:
487                         dumpiov_in_mmsghdr(tcp, tcp->u_arg[1]);
488                         return;
489                 }
490         }
491 }
492
493 const char *
494 err_name(unsigned long err)
495 {
496         if ((err < nerrnos) && errnoent[err])
497                 return errnoent[err];
498
499         return NULL;
500 }
501
502 static long get_regs(struct tcb *);
503 static int get_syscall_args(struct tcb *);
504 static int get_syscall_result(struct tcb *);
505 static int arch_get_scno(struct tcb *tcp);
506 static int arch_set_scno(struct tcb *, kernel_ulong_t);
507 static void get_error(struct tcb *, const bool);
508 static int arch_set_error(struct tcb *);
509 static int arch_set_success(struct tcb *);
510
511 struct inject_opts *inject_vec[SUPPORTED_PERSONALITIES];
512
513 static struct inject_opts *
514 tcb_inject_opts(struct tcb *tcp)
515 {
516         return (scno_in_range(tcp->scno) && tcp->inject_vec[current_personality])
517                ? &tcp->inject_vec[current_personality][tcp->scno] : NULL;
518 }
519
520
521 static long
522 tamper_with_syscall_entering(struct tcb *tcp, unsigned int *signo)
523 {
524         if (!tcp->inject_vec[current_personality]) {
525                 tcp->inject_vec[current_personality] =
526                         xcalloc(nsyscalls, sizeof(**inject_vec));
527                 memcpy(tcp->inject_vec[current_personality],
528                        inject_vec[current_personality],
529                        nsyscalls * sizeof(**inject_vec));
530         }
531
532         struct inject_opts *opts = tcb_inject_opts(tcp);
533
534         if (!opts || opts->first == 0)
535                 return 0;
536
537         --opts->first;
538
539         if (opts->first != 0)
540                 return 0;
541
542         opts->first = opts->step;
543
544         if (!recovering(tcp)) {
545                 if (opts->data.flags & INJECT_F_SIGNAL)
546                         *signo = opts->data.signo;
547                 if (opts->data.flags & (INJECT_F_ERROR | INJECT_F_RETVAL) &&
548                     !arch_set_scno(tcp, -1))
549                         tcp->flags |= TCB_TAMPERED;
550                 if (opts->data.flags & INJECT_F_DELAY_ENTER)
551                         delay_tcb(tcp, opts->data.delay_idx, true);
552                 if (opts->data.flags & INJECT_F_DELAY_EXIT)
553                         tcp->flags |= TCB_INJECT_DELAY_EXIT;
554         }
555
556         return 0;
557 }
558
559 static long
560 tamper_with_syscall_exiting(struct tcb *tcp)
561 {
562         struct inject_opts *opts = tcb_inject_opts(tcp);
563         if (!opts)
564                 return 0;
565
566         if (inject_delay_exit(tcp))
567                 delay_tcb(tcp, opts->data.delay_idx, false);
568
569         if (!syscall_tampered(tcp))
570                 return 0;
571
572         if (!syserror(tcp)) {
573                 error_msg("Failed to tamper with process %d: got no error "
574                           "(return value %#" PRI_klx ")",
575                           tcp->pid, tcp->u_rval);
576
577                 return 1;
578         }
579
580         bool update_tcb = false;
581
582         if (opts->data.flags & INJECT_F_RETVAL) {
583                 kernel_long_t inject_rval =
584                         retval_get(opts->data.rval_idx);
585                 kernel_long_t u_rval = tcp->u_rval;
586
587                 tcp->u_rval = inject_rval;
588                 if (arch_set_success(tcp)) {
589                         tcp->u_rval = u_rval;
590                 } else {
591                         update_tcb = true;
592                         tcp->u_error = 0;
593                 }
594         } else {
595                 unsigned long new_error = retval_get(opts->data.rval_idx);
596
597                 if (new_error != tcp->u_error && new_error <= MAX_ERRNO_VALUE) {
598                         unsigned long u_error = tcp->u_error;
599
600                         tcp->u_error = new_error;
601                         if (arch_set_error(tcp)) {
602                                 tcp->u_error = u_error;
603                         } else {
604                                 update_tcb = true;
605                         }
606                 }
607         }
608
609         if (update_tcb) {
610                 tcp->u_error = 0;
611                 get_error(tcp, !(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS));
612         }
613
614         return 0;
615 }
616
617 /*
618  * Returns:
619  * 0: "ignore this ptrace stop", bail out silently.
620  * 1: ok, decoded; call
621  *    syscall_entering_finish(tcp, syscall_entering_trace(tcp, ...)).
622  * other: error; call syscall_entering_finish(tcp, res), where res is the value
623  *    returned.
624  */
625 int
626 syscall_entering_decode(struct tcb *tcp)
627 {
628         int res = get_scno(tcp);
629         if (res == 0)
630                 return res;
631         int scno_good = res;
632         if (res != 1 || (res = get_syscall_args(tcp)) != 1) {
633                 printleader(tcp);
634                 tprintf("%s(", scno_good == 1 ? tcp->s_ent->sys_name : "????");
635                 /*
636                  * " <unavailable>" will be added later by the code which
637                  * detects ptrace errors.
638                  */
639                 return res;
640         }
641
642 #if defined SYS_ipc_subcall     \
643  || defined SYS_socket_subcall  \
644  || defined SYS_syscall_subcall
645         for (;;) {
646                 switch (tcp->s_ent->sen) {
647 # ifdef SYS_ipc_subcall
648                 case SEN_ipc:
649                         decode_ipc_subcall(tcp);
650                         break;
651 # endif
652 # ifdef SYS_socket_subcall
653                 case SEN_socketcall:
654                         decode_socket_subcall(tcp);
655                         break;
656 # endif
657 # ifdef SYS_syscall_subcall
658                 case SEN_syscall:
659                         decode_syscall_subcall(tcp);
660                         if (tcp->s_ent->sen != SEN_syscall)
661                                 continue;
662                         break;
663 # endif
664                 }
665                 break;
666         }
667 #endif
668
669         return 1;
670 }
671
672 int
673 syscall_entering_trace(struct tcb *tcp, unsigned int *sig)
674 {
675         /* Restrain from fault injection while the trace executes strace code. */
676         if (hide_log(tcp)) {
677                 tcp->qual_flg &= ~QUAL_INJECT;
678         }
679
680         switch (tcp->s_ent->sen) {
681                 case SEN_execve:
682                 case SEN_execveat:
683 #if defined SPARC || defined SPARC64
684                 case SEN_execv:
685 #endif
686                         tcp->flags &= ~TCB_HIDE_LOG;
687                         break;
688         }
689
690         if (!traced(tcp) || (tracing_paths && !pathtrace_match(tcp))) {
691                 tcp->flags |= TCB_FILTERED;
692                 return 0;
693         }
694
695         tcp->flags &= ~TCB_FILTERED;
696
697         if (hide_log(tcp)) {
698                 return 0;
699         }
700
701         if (inject(tcp))
702                 tamper_with_syscall_entering(tcp, sig);
703
704         if (cflag == CFLAG_ONLY_STATS) {
705                 return 0;
706         }
707
708 #ifdef ENABLE_STACKTRACE
709         if (stack_trace_enabled) {
710                 if (tcp->s_ent->sys_flags & STACKTRACE_CAPTURE_ON_ENTER)
711                         unwind_tcb_capture(tcp);
712         }
713 #endif
714
715         printleader(tcp);
716         tprintf("%s(", tcp->s_ent->sys_name);
717         int res = raw(tcp) ? printargs(tcp) : tcp->s_ent->sys_func(tcp);
718         fflush(tcp->outf);
719         return res;
720 }
721
722 void
723 syscall_entering_finish(struct tcb *tcp, int res)
724 {
725         tcp->flags |= TCB_INSYSCALL;
726         tcp->sys_func_rval = res;
727         /* Measure the entrance time as late as possible to avoid errors. */
728         if ((Tflag || cflag) && !filtered(tcp))
729                 clock_gettime(CLOCK_MONOTONIC, &tcp->etime);
730 }
731
732 /* Returns:
733  * 0: "bail out".
734  * 1: ok.
735  * -1: error in one of ptrace ops.
736  *
737  * If not 0, call syscall_exiting_trace(tcp, res), where res is the return
738  *    value. Anyway, call syscall_exiting_finish(tcp) then.
739  */
740 int
741 syscall_exiting_decode(struct tcb *tcp, struct timespec *pts)
742 {
743         /* Measure the exit time as early as possible to avoid errors. */
744         if ((Tflag || cflag) && !(filtered(tcp) || hide_log(tcp)))
745                 clock_gettime(CLOCK_MONOTONIC, pts);
746
747         if (mmap_cache_is_enabled()) {
748                 if (tcp->s_ent->sys_flags & STACKTRACE_INVALIDATE_CACHE)
749                         mmap_cache_invalidate(tcp);
750         }
751
752         if (filtered(tcp) || hide_log(tcp))
753                 return 0;
754
755 #if SUPPORTED_PERSONALITIES > 1
756         update_personality(tcp, tcp->currpers);
757 #endif
758
759         return get_syscall_result(tcp);
760 }
761
762 int
763 syscall_exiting_trace(struct tcb *tcp, struct timespec *ts, int res)
764 {
765         if (syscall_tampered(tcp) || inject_delay_exit(tcp))
766                 tamper_with_syscall_exiting(tcp);
767
768         if (cflag) {
769                 count_syscall(tcp, ts);
770                 if (cflag == CFLAG_ONLY_STATS) {
771                         return 0;
772                 }
773         }
774
775         /* If not in -ff mode, and printing_tcp != tcp,
776          * then the log currently does not end with output
777          * of _our syscall entry_, but with something else.
778          * We need to say which syscall's return is this.
779          *
780          * Forced reprinting via TCB_REPRINT is used only by
781          * "strace -ff -oLOG test/threaded_execve" corner case.
782          * It's the only case when -ff mode needs reprinting.
783          */
784         if ((followfork < 2 && printing_tcp != tcp) || (tcp->flags & TCB_REPRINT)) {
785                 tcp->flags &= ~TCB_REPRINT;
786                 printleader(tcp);
787                 tprintf("<... %s resumed> ", tcp->s_ent->sys_name);
788         }
789         printing_tcp = tcp;
790
791         tcp->s_prev_ent = NULL;
792         if (res != 1) {
793                 /* There was error in one of prior ptrace ops */
794                 tprints(") ");
795                 tabto();
796                 tprints("= ? <unavailable>\n");
797                 line_ended();
798                 return res;
799         }
800         tcp->s_prev_ent = tcp->s_ent;
801
802         int sys_res = 0;
803         if (raw(tcp)) {
804                 /* sys_res = printargs(tcp); - but it's nop on sysexit */
805         } else {
806         /* FIXME: not_failing_only (IOW, option -z) is broken:
807          * failure of syscall is known only after syscall return.
808          * Thus we end up with something like this on, say, ENOENT:
809          *     open("does_not_exist", O_RDONLY <unfinished ...>
810          *     {next syscall decode}
811          * whereas the intended result is that open(...) line
812          * is not shown at all.
813          */
814                 if (not_failing_only && tcp->u_error)
815                         return 0;       /* ignore failed syscalls */
816                 if (tcp->sys_func_rval & RVAL_DECODED)
817                         sys_res = tcp->sys_func_rval;
818                 else
819                         sys_res = tcp->s_ent->sys_func(tcp);
820         }
821
822         tprints(") ");
823         tabto();
824         unsigned long u_error = tcp->u_error;
825
826         if (raw(tcp)) {
827                 if (u_error) {
828                         tprintf("= -1 (errno %lu)", u_error);
829                 } else {
830                         tprintf("= %#" PRI_klx, tcp->u_rval);
831                 }
832                 if (syscall_tampered(tcp))
833                         tprints(" (INJECTED)");
834         } else if (!(sys_res & RVAL_NONE) && u_error) {
835                 const char *u_error_str;
836
837                 switch (u_error) {
838                 /* Blocked signals do not interrupt any syscalls.
839                  * In this case syscalls don't return ERESTARTfoo codes.
840                  *
841                  * Deadly signals set to SIG_DFL interrupt syscalls
842                  * and kill the process regardless of which of the codes below
843                  * is returned by the interrupted syscall.
844                  * In some cases, kernel forces a kernel-generated deadly
845                  * signal to be unblocked and set to SIG_DFL (and thus cause
846                  * death) if it is blocked or SIG_IGNed: for example, SIGSEGV
847                  * or SIGILL. (The alternative is to leave process spinning
848                  * forever on the faulty instruction - not useful).
849                  *
850                  * SIG_IGNed signals and non-deadly signals set to SIG_DFL
851                  * (for example, SIGCHLD, SIGWINCH) interrupt syscalls,
852                  * but kernel will always restart them.
853                  */
854                 case ERESTARTSYS:
855                         /* Most common type of signal-interrupted syscall exit code.
856                          * The system call will be restarted with the same arguments
857                          * if SA_RESTART is set; otherwise, it will fail with EINTR.
858                          */
859                         tprints("= ? ERESTARTSYS (To be restarted if SA_RESTART is set)");
860                         break;
861                 case ERESTARTNOINTR:
862                         /* Rare. For example, fork() returns this if interrupted.
863                          * SA_RESTART is ignored (assumed set): the restart is unconditional.
864                          */
865                         tprints("= ? ERESTARTNOINTR (To be restarted)");
866                         break;
867                 case ERESTARTNOHAND:
868                         /* pause(), rt_sigsuspend() etc use this code.
869                          * SA_RESTART is ignored (assumed not set):
870                          * syscall won't restart (will return EINTR instead)
871                          * even after signal with SA_RESTART set. However,
872                          * after SIG_IGN or SIG_DFL signal it will restart
873                          * (thus the name "restart only if has no handler").
874                          */
875                         tprints("= ? ERESTARTNOHAND (To be restarted if no handler)");
876                         break;
877                 case ERESTART_RESTARTBLOCK:
878                         /* Syscalls like nanosleep(), poll() which can't be
879                          * restarted with their original arguments use this
880                          * code. Kernel will execute restart_syscall() instead,
881                          * which changes arguments before restarting syscall.
882                          * SA_RESTART is ignored (assumed not set) similarly
883                          * to ERESTARTNOHAND. (Kernel can't honor SA_RESTART
884                          * since restart data is saved in "restart block"
885                          * in task struct, and if signal handler uses a syscall
886                          * which in turn saves another such restart block,
887                          * old data is lost and restart becomes impossible)
888                          */
889                         tprints("= ? ERESTART_RESTARTBLOCK (Interrupted by signal)");
890                         break;
891                 default:
892                         u_error_str = err_name(u_error);
893                         if (u_error_str)
894                                 tprintf("= %" PRI_kld " %s (%s)", tcp->u_rval,
895                                         u_error_str, strerror(u_error));
896                         else
897                                 tprintf("= %" PRI_kld " %lu (%s)", tcp->u_rval,
898                                         u_error, strerror(u_error));
899                         break;
900                 }
901                 if (syscall_tampered(tcp))
902                         tprints(" (INJECTED)");
903                 if ((sys_res & RVAL_STR) && tcp->auxstr)
904                         tprintf(" (%s)", tcp->auxstr);
905         } else {
906                 if (sys_res & RVAL_NONE)
907                         tprints("= ?");
908                 else {
909                         switch (sys_res & RVAL_MASK) {
910                         case RVAL_HEX:
911 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
912                                 if (current_klongsize < sizeof(tcp->u_rval)) {
913                                         tprintf("= %#x",
914                                                 (unsigned int) tcp->u_rval);
915                                 } else
916 #endif
917                                 {
918                                         tprintf("= %#" PRI_klx, tcp->u_rval);
919                                 }
920                                 break;
921                         case RVAL_OCTAL:
922                                 tprints("= ");
923                                 print_numeric_long_umask(tcp->u_rval);
924                                 break;
925                         case RVAL_UDECIMAL:
926 #if ANY_WORDSIZE_LESS_THAN_KERNEL_LONG
927                                 if (current_klongsize < sizeof(tcp->u_rval)) {
928                                         tprintf("= %u",
929                                                 (unsigned int) tcp->u_rval);
930                                 } else
931 #endif
932                                 {
933                                         tprintf("= %" PRI_klu, tcp->u_rval);
934                                 }
935                                 break;
936                         case RVAL_FD:
937                                 if (show_fd_path) {
938                                         tprints("= ");
939                                         printfd(tcp, tcp->u_rval);
940                                 } else
941                                         tprintf("= %" PRI_kld, tcp->u_rval);
942                                 break;
943                         default:
944                                 error_msg("invalid rval format");
945                                 break;
946                         }
947                 }
948                 if ((sys_res & RVAL_STR) && tcp->auxstr)
949                         tprintf(" (%s)", tcp->auxstr);
950                 if (syscall_tampered(tcp))
951                         tprints(" (INJECTED)");
952         }
953         if (Tflag) {
954                 ts_sub(ts, ts, &tcp->etime);
955                 tprintf(" <%ld.%06ld>",
956                         (long) ts->tv_sec, (long) ts->tv_nsec / 1000);
957         }
958         tprints("\n");
959         dumpio(tcp);
960         line_ended();
961
962 #ifdef ENABLE_STACKTRACE
963         if (stack_trace_enabled)
964                 unwind_tcb_print(tcp);
965 #endif
966         return 0;
967 }
968
969 void
970 syscall_exiting_finish(struct tcb *tcp)
971 {
972         tcp->flags &= ~(TCB_INSYSCALL | TCB_TAMPERED | TCB_INJECT_DELAY_EXIT);
973         tcp->sys_func_rval = 0;
974         free_tcb_priv_data(tcp);
975 }
976
977 bool
978 is_erestart(struct tcb *tcp)
979 {
980         switch (tcp->u_error) {
981                 case ERESTARTSYS:
982                 case ERESTARTNOINTR:
983                 case ERESTARTNOHAND:
984                 case ERESTART_RESTARTBLOCK:
985                         return true;
986                 default:
987                         return false;
988         }
989 }
990
991 static unsigned long saved_u_error;
992
993 void
994 temporarily_clear_syserror(struct tcb *tcp)
995 {
996         saved_u_error = tcp->u_error;
997         tcp->u_error = 0;
998 }
999
1000 void
1001 restore_cleared_syserror(struct tcb *tcp)
1002 {
1003         tcp->u_error = saved_u_error;
1004 }
1005
1006 #include "arch_regs.c"
1007
1008 #if HAVE_ARCH_GETRVAL2
1009 # include "arch_getrval2.c"
1010 #endif
1011
1012 void
1013 print_pc(struct tcb *tcp)
1014 {
1015 #if defined ARCH_PC_REG
1016 # define ARCH_GET_PC 0
1017 #elif defined ARCH_PC_PEEK_ADDR
1018         kernel_ulong_t pc;
1019 # define ARCH_PC_REG pc
1020 # define ARCH_GET_PC upeek(tcp, ARCH_PC_PEEK_ADDR, &pc)
1021 #else
1022 # error Neither ARCH_PC_REG nor ARCH_PC_PEEK_ADDR is defined
1023 #endif
1024         if (get_regs(tcp) < 0 || ARCH_GET_PC)
1025                 tprints(current_wordsize == 4 ? "[????????] "
1026                                               : "[????????????????] ");
1027         else
1028                 tprintf(current_wordsize == 4
1029                         ? "[%08" PRI_klx "] " : "[%016" PRI_klx "] ",
1030                         (kernel_ulong_t) ARCH_PC_REG);
1031 }
1032
1033 #include "getregs_old.h"
1034
1035 #undef ptrace_getregset_or_getregs
1036 #undef ptrace_setregset_or_setregs
1037 #ifdef ARCH_REGS_FOR_GETREGSET
1038
1039 # define ptrace_getregset_or_getregs ptrace_getregset
1040 static long
1041 ptrace_getregset(pid_t pid)
1042 {
1043 # ifdef ARCH_IOVEC_FOR_GETREGSET
1044         /* variable iovec */
1045         ARCH_IOVEC_FOR_GETREGSET.iov_len = sizeof(ARCH_REGS_FOR_GETREGSET);
1046         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS,
1047                       &ARCH_IOVEC_FOR_GETREGSET);
1048 # else
1049         /* constant iovec */
1050         static struct iovec io = {
1051                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1052                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1053         };
1054         return ptrace(PTRACE_GETREGSET, pid, NT_PRSTATUS, &io);
1055
1056 # endif
1057 }
1058
1059 # ifndef HAVE_GETREGS_OLD
1060 #  define ptrace_setregset_or_setregs ptrace_setregset
1061 static int
1062 ptrace_setregset(pid_t pid)
1063 {
1064 #  ifdef ARCH_IOVEC_FOR_GETREGSET
1065         /* variable iovec */
1066         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS,
1067                       &ARCH_IOVEC_FOR_GETREGSET);
1068 #  else
1069         /* constant iovec */
1070         static struct iovec io = {
1071                 .iov_base = &ARCH_REGS_FOR_GETREGSET,
1072                 .iov_len = sizeof(ARCH_REGS_FOR_GETREGSET)
1073         };
1074         return ptrace(PTRACE_SETREGSET, pid, NT_PRSTATUS, &io);
1075 #  endif
1076 }
1077 # endif /* !HAVE_GETREGS_OLD */
1078
1079 #elif defined ARCH_REGS_FOR_GETREGS
1080
1081 # define ptrace_getregset_or_getregs ptrace_getregs
1082 static long
1083 ptrace_getregs(pid_t pid)
1084 {
1085 # if defined SPARC || defined SPARC64
1086         /* SPARC systems have the meaning of data and addr reversed */
1087         return ptrace(PTRACE_GETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1088 # else
1089         return ptrace(PTRACE_GETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1090 # endif
1091 }
1092
1093 # ifndef HAVE_GETREGS_OLD
1094 #  define ptrace_setregset_or_setregs ptrace_setregs
1095 static int
1096 ptrace_setregs(pid_t pid)
1097 {
1098 #  if defined SPARC || defined SPARC64
1099         /* SPARC systems have the meaning of data and addr reversed */
1100         return ptrace(PTRACE_SETREGS, pid, (void *) &ARCH_REGS_FOR_GETREGS, 0);
1101 #  else
1102         return ptrace(PTRACE_SETREGS, pid, NULL, &ARCH_REGS_FOR_GETREGS);
1103 #  endif
1104 }
1105 # endif /* !HAVE_GETREGS_OLD */
1106
1107 #endif /* ARCH_REGS_FOR_GETREGSET || ARCH_REGS_FOR_GETREGS */
1108
1109 #ifdef ptrace_getregset_or_getregs
1110 static long get_regs_error;
1111 #endif
1112
1113 void
1114 clear_regs(struct tcb *tcp)
1115 {
1116 #ifdef ptrace_getregset_or_getregs
1117         get_regs_error = -1;
1118 #endif
1119 }
1120
1121 static long
1122 get_regs(struct tcb *const tcp)
1123 {
1124 #ifdef ptrace_getregset_or_getregs
1125
1126         if (get_regs_error != -1)
1127                 return get_regs_error;
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                 return get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1137         } else if (use_getregs_old == 0) {
1138                 get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1139                 if (get_regs_error >= 0) {
1140                         use_getregs_old = -1;
1141                         return get_regs_error;
1142                 }
1143                 if (errno == EPERM || errno == ESRCH)
1144                         return get_regs_error;
1145                 use_getregs_old = 1;
1146         }
1147         return get_regs_error = getregs_old(tcp);
1148 # else /* !HAVE_GETREGS_OLD */
1149         /* Assume that PTRACE_GETREGSET/PTRACE_GETREGS works. */
1150         get_regs_error = ptrace_getregset_or_getregs(tcp->pid);
1151
1152 #  if defined ARCH_PERSONALITY_0_IOV_SIZE
1153         if (get_regs_error)
1154                 return get_regs_error;
1155
1156         switch (ARCH_IOVEC_FOR_GETREGSET.iov_len) {
1157         case ARCH_PERSONALITY_0_IOV_SIZE:
1158                 update_personality(tcp, 0);
1159                 break;
1160         case ARCH_PERSONALITY_1_IOV_SIZE:
1161                 update_personality(tcp, 1);
1162                 break;
1163         default: {
1164                 static bool printed = false;
1165
1166                 if (!printed) {
1167                         error_msg("Unsupported regset size returned by "
1168                                   "PTRACE_GETREGSET: %zu",
1169                                   ARCH_IOVEC_FOR_GETREGSET.iov_len);
1170
1171                         printed = true;
1172                 }
1173
1174                 update_personality(tcp, 0);
1175         }
1176         }
1177 #  endif /* ARCH_PERSONALITY_0_IOV_SIZE */
1178
1179         return get_regs_error;
1180
1181 # endif /* !HAVE_GETREGS_OLD */
1182
1183 #else /* !ptrace_getregset_or_getregs */
1184
1185 # warning get_regs is not implemented for this architecture yet
1186         return 0;
1187
1188 #endif /* !ptrace_getregset_or_getregs */
1189 }
1190
1191 #ifdef ptrace_setregset_or_setregs
1192 static int
1193 set_regs(pid_t pid)
1194 {
1195         return ptrace_setregset_or_setregs(pid);
1196 }
1197 #endif /* ptrace_setregset_or_setregs */
1198
1199 struct sysent_buf {
1200         struct tcb *tcp;
1201         struct_sysent ent;
1202         char buf[sizeof("syscall_0x") + sizeof(kernel_ulong_t) * 2];
1203 };
1204
1205 static void
1206 free_sysent_buf(void *ptr)
1207 {
1208         struct sysent_buf *s = ptr;
1209         s->tcp->s_prev_ent = s->tcp->s_ent = NULL;
1210         free(ptr);
1211 }
1212
1213 /*
1214  * Returns:
1215  * 0: "ignore this ptrace stop", syscall_entering_decode() should return a "bail
1216  *    out silently" code.
1217  * 1: ok, continue in syscall_entering_decode().
1218  * other: error, syscall_entering_decode() should print error indicator
1219  *    ("????" etc) and return an appropriate code.
1220  */
1221 int
1222 get_scno(struct tcb *tcp)
1223 {
1224         if (get_regs(tcp) < 0)
1225                 return -1;
1226
1227         int rc = arch_get_scno(tcp);
1228         if (rc != 1)
1229                 return rc;
1230
1231         tcp->scno = shuffle_scno(tcp->scno);
1232
1233         if (scno_is_valid(tcp->scno)) {
1234                 tcp->s_ent = &sysent[tcp->scno];
1235                 tcp->qual_flg = qual_flags(tcp->scno);
1236         } else {
1237                 struct sysent_buf *s = xcalloc(1, sizeof(*s));
1238
1239                 s->tcp = tcp;
1240                 s->ent.nargs = MAX_ARGS;
1241                 s->ent.sen = SEN_printargs;
1242                 s->ent.sys_func = printargs;
1243                 s->ent.sys_name = s->buf;
1244                 xsprintf(s->buf, "syscall_%#" PRI_klx, shuffle_scno(tcp->scno));
1245
1246                 tcp->s_ent = &s->ent;
1247                 tcp->qual_flg = QUAL_RAW | DEFAULT_QUAL_FLAGS;
1248
1249                 set_tcb_priv_data(tcp, s, free_sysent_buf);
1250
1251                 debug_msg("pid %d invalid syscall %#" PRI_klx,
1252                           tcp->pid, shuffle_scno(tcp->scno));
1253         }
1254
1255         /*
1256          * We refrain from argument decoding during recovering
1257          * as tracee memory mappings has changed and the registers
1258          * are very likely pointing to garbage already.
1259          */
1260         if (recovering(tcp))
1261                 tcp->qual_flg |= QUAL_RAW;
1262
1263         return 1;
1264 }
1265
1266 #ifdef ptrace_getregset_or_getregs
1267 # define get_syscall_result_regs get_regs
1268 #else
1269 static int get_syscall_result_regs(struct tcb *);
1270 #endif
1271
1272 /* Returns:
1273  * 1: ok, continue in syscall_exiting_trace().
1274  * -1: error, syscall_exiting_trace() should print error indicator
1275  *    ("????" etc) and bail out.
1276  */
1277 static int
1278 get_syscall_result(struct tcb *tcp)
1279 {
1280         if (get_syscall_result_regs(tcp) < 0)
1281                 return -1;
1282         tcp->u_error = 0;
1283         get_error(tcp,
1284                   !(tcp->s_ent->sys_flags & SYSCALL_NEVER_FAILS)
1285                         || syscall_tampered(tcp));
1286
1287         return 1;
1288 }
1289
1290 #include "get_scno.c"
1291 #include "set_scno.c"
1292 #include "get_syscall_args.c"
1293 #ifndef ptrace_getregset_or_getregs
1294 # include "get_syscall_result.c"
1295 #endif
1296 #include "get_error.c"
1297 #include "set_error.c"
1298 #ifdef HAVE_GETREGS_OLD
1299 # include "getregs_old.c"
1300 #endif
1301 #include "shuffle_scno.c"
1302
1303 const char *
1304 syscall_name(kernel_ulong_t scno)
1305 {
1306         return scno_is_valid(scno) ? sysent[scno].sys_name : NULL;
1307 }