vserver 2.0 rc7
[linux-2.6.git] / arch / um / kernel / tt / tracer.c
1 /* 
2  * Copyright (C) 2002 Jeff Dike (jdike@karaya.com)
3  * Licensed under the GPL
4  */
5
6 #include <stdio.h>
7 #include <stdlib.h>
8 #include <stdarg.h>
9 #include <unistd.h>
10 #include <signal.h>
11 #include <errno.h>
12 #include <sched.h>
13 #include <string.h>
14 #include <sys/mman.h>
15 #include <sys/time.h>
16 #include <sys/wait.h>
17 #include "user.h"
18 #include "sysdep/ptrace.h"
19 #include "sigcontext.h"
20 #include "sysdep/sigcontext.h"
21 #include "os.h"
22 #include "signal_user.h"
23 #include "user_util.h"
24 #include "mem_user.h"
25 #include "process.h"
26 #include "kern_util.h"
27 #include "chan_user.h"
28 #include "ptrace_user.h"
29 #include "irq_user.h"
30 #include "mode.h"
31 #include "tt.h"
32
33 static int tracer_winch[2];
34
35 int is_tracer_winch(int pid, int fd, void *data)
36 {
37         if(pid != os_getpgrp())
38                 return(0);
39
40         register_winch_irq(tracer_winch[0], fd, -1, data);
41         return(1);
42 }
43
44 static void tracer_winch_handler(int sig)
45 {
46         int n;
47         char c = 1;
48
49         n = os_write_file(tracer_winch[1], &c, sizeof(c));
50         if(n != sizeof(c))
51                 printk("tracer_winch_handler - write failed, err = %d\n", -n);
52 }
53
54 /* Called only by the tracing thread during initialization */
55
56 static void setup_tracer_winch(void)
57 {
58         int err;
59
60         err = os_pipe(tracer_winch, 1, 1);
61         if(err < 0){
62                 printk("setup_tracer_winch : os_pipe failed, err = %d\n", -err);
63                 return;
64         }
65         signal(SIGWINCH, tracer_winch_handler);
66 }
67
68 void attach_process(int pid)
69 {
70         if((ptrace(PTRACE_ATTACH, pid, 0, 0) < 0) ||
71            (ptrace(PTRACE_CONT, pid, 0, 0) < 0))
72                 tracer_panic("OP_FORK failed to attach pid");
73         wait_for_stop(pid, SIGSTOP, PTRACE_CONT, NULL);
74         if (ptrace(PTRACE_OLDSETOPTIONS, pid, 0, (void *)PTRACE_O_TRACESYSGOOD) < 0)
75                 tracer_panic("OP_FORK: PTRACE_SETOPTIONS failed, errno = %d", errno);
76         if(ptrace(PTRACE_CONT, pid, 0, 0) < 0)
77                 tracer_panic("OP_FORK failed to continue process");
78 }
79
80 void tracer_panic(char *format, ...)
81 {
82         va_list ap;
83
84         va_start(ap, format);
85         vprintf(format, ap);
86         va_end(ap);
87         printf("\n");
88         while(1) pause();
89 }
90
91 static void tracer_segv(int sig, struct sigcontext sc)
92 {
93         struct faultinfo fi;
94         GET_FAULTINFO_FROM_SC(fi, &sc);
95         printf("Tracing thread segfault at address 0x%lx, ip 0x%lx\n",
96                FAULT_ADDRESS(fi), SC_IP(&sc));
97         while(1)
98                 pause();
99 }
100
101 /* Changed early in boot, and then only read */
102 int debug = 0;
103 int debug_stop = 1;
104 int debug_parent = 0;
105 int honeypot = 0;
106
107 static int signal_tramp(void *arg)
108 {
109         int (*proc)(void *);
110
111         if(honeypot && munmap((void *) (host_task_size - 0x10000000),
112                               0x10000000)) 
113                 panic("Unmapping stack failed");
114         if(ptrace(PTRACE_TRACEME, 0, 0, 0) < 0)
115                 panic("ptrace PTRACE_TRACEME failed");
116         os_stop_process(os_getpid());
117         change_sig(SIGWINCH, 0);
118         signal(SIGUSR1, SIG_IGN);
119         change_sig(SIGCHLD, 0);
120         signal(SIGSEGV, (__sighandler_t) sig_handler);
121         set_cmdline("(idle thread)");
122         set_init_pid(os_getpid());
123         init_irq_signals(0);
124         proc = arg;
125         return((*proc)(NULL));
126 }
127
128 static void sleeping_process_signal(int pid, int sig)
129 {
130         switch(sig){
131         /* These two result from UML being ^Z-ed and bg-ed.  PTRACE_CONT is
132          * right because the process must be in the kernel already.
133          */
134         case SIGCONT:
135         case SIGTSTP:
136                 if(ptrace(PTRACE_CONT, pid, 0, sig) < 0)
137                         tracer_panic("sleeping_process_signal : Failed to "
138                                      "continue pid %d, signal = %d, "
139                                      "errno = %d\n", pid, sig, errno);
140                 break;
141
142         /* This happens when the debugger (e.g. strace) is doing system call 
143          * tracing on the kernel.  During a context switch, the current task
144          * will be set to the incoming process and the outgoing process will
145          * hop into write and then read.  Since it's not the current process
146          * any more, the trace of those will land here.  So, we need to just 
147          * PTRACE_SYSCALL it.
148          */
149         case (SIGTRAP + 0x80):
150                 if(ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
151                         tracer_panic("sleeping_process_signal : Failed to "
152                                      "PTRACE_SYSCALL pid %d, errno = %d\n",
153                                      pid, errno);
154                 break;
155         case SIGSTOP:
156                 break;
157         default:
158                 tracer_panic("sleeping process %d got unexpected "
159                              "signal : %d\n", pid, sig);
160                 break;
161         }
162 }
163
164 /* Accessed only by the tracing thread */
165 int debugger_pid = -1;
166 int debugger_parent = -1;
167 int debugger_fd = -1;
168 int gdb_pid = -1;
169
170 struct {
171         int pid;
172         int signal;
173         unsigned long addr;
174         struct timeval time;
175 } signal_record[1024][32];
176
177 int signal_index[32];
178 int nsignals = 0;
179 int debug_trace = 0;
180 extern int io_nsignals, io_count, intr_count;
181
182 extern void signal_usr1(int sig);
183
184 int tracing_pid = -1;
185
186 int tracer(int (*init_proc)(void *), void *sp)
187 {
188         void *task = NULL;
189         int status, pid = 0, sig = 0, cont_type, tracing = 0, op = 0;
190         int proc_id = 0, n, err, old_tracing = 0, strace = 0;
191         int local_using_sysemu = 0;
192 #ifdef UML_CONFIG_SYSCALL_DEBUG
193         unsigned long eip = 0;
194         int last_index;
195 #endif
196         signal(SIGPIPE, SIG_IGN);
197         setup_tracer_winch();
198         tracing_pid = os_getpid();
199         printf("tracing thread pid = %d\n", tracing_pid);
200
201         pid = clone(signal_tramp, sp, CLONE_FILES | SIGCHLD, init_proc);
202         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
203         if(n < 0){
204                 printf("waitpid on idle thread failed, errno = %d\n", errno);
205                 exit(1);
206         }
207         if (ptrace(PTRACE_OLDSETOPTIONS, pid, 0, (void *)PTRACE_O_TRACESYSGOOD) < 0) {
208                 printf("Failed to PTRACE_SETOPTIONS for idle thread, errno = %d\n", errno);
209                 exit(1);
210         }
211         if((ptrace(PTRACE_CONT, pid, 0, 0) < 0)){
212                 printf("Failed to continue idle thread, errno = %d\n", errno);
213                 exit(1);
214         }
215
216         signal(SIGSEGV, (sighandler_t) tracer_segv);
217         signal(SIGUSR1, signal_usr1);
218         if(debug_trace){
219                 printf("Tracing thread pausing to be attached\n");
220                 stop();
221         }
222         if(debug){
223                 if(gdb_pid != -1) 
224                         debugger_pid = attach_debugger(pid, gdb_pid, 1);
225                 else debugger_pid = init_ptrace_proxy(pid, 1, debug_stop);
226                 if(debug_parent){
227                         debugger_parent = os_process_parent(debugger_pid);
228                         init_parent_proxy(debugger_parent);
229                         err = attach(debugger_parent);
230                         if(err){
231                                 printf("Failed to attach debugger parent %d, "
232                                        "errno = %d\n", debugger_parent, -err);
233                                 debugger_parent = -1;
234                         }
235                         else {
236                                 if(ptrace(PTRACE_SYSCALL, debugger_parent, 
237                                           0, 0) < 0){
238                                         printf("Failed to continue debugger "
239                                                "parent, errno = %d\n", errno);
240                                         debugger_parent = -1;
241                                 }
242                         }
243                 }
244         }
245         set_cmdline("(tracing thread)");
246         while(1){
247                 CATCH_EINTR(pid = waitpid(-1, &status, WUNTRACED));
248                 if(pid <= 0){
249                         if(errno != ECHILD){
250                                 printf("wait failed - errno = %d\n", errno);
251                         }
252                         continue;
253                 }
254                 if(pid == debugger_pid){
255                         int cont = 0;
256
257                         if(WIFEXITED(status) || WIFSIGNALED(status))
258                                 debugger_pid = -1;
259                         /* XXX Figure out how to deal with gdb and SMP */
260                         else cont = debugger_signal(status, cpu_tasks[0].pid);
261                         if(cont == PTRACE_SYSCALL) strace = 1;
262                         continue;
263                 }
264                 else if(pid == debugger_parent){
265                         debugger_parent_signal(status, pid);
266                         continue;
267                 }
268                 nsignals++;
269                 if(WIFEXITED(status)) ;
270 #ifdef notdef
271                 {
272                         printf("Child %d exited with status %d\n", pid, 
273                                WEXITSTATUS(status));
274                 }
275 #endif
276                 else if(WIFSIGNALED(status)){
277                         sig = WTERMSIG(status);
278                         if(sig != 9){
279                                 printf("Child %d exited with signal %d\n", pid,
280                                        sig);
281                         }
282                 }
283                 else if(WIFSTOPPED(status)){
284                         proc_id = pid_to_processor_id(pid);
285                         sig = WSTOPSIG(status);
286 #ifdef UML_CONFIG_SYSCALL_DEBUG
287                         if(signal_index[proc_id] == 1024){
288                                 signal_index[proc_id] = 0;
289                                 last_index = 1023;
290                         }
291                         else last_index = signal_index[proc_id] - 1;
292                         if(((sig == SIGPROF) || (sig == SIGVTALRM) ||
293                             (sig == SIGALRM)) &&
294                            (signal_record[proc_id][last_index].signal == sig)&&
295                            (signal_record[proc_id][last_index].pid == pid))
296                                 signal_index[proc_id] = last_index;
297                         signal_record[proc_id][signal_index[proc_id]].pid = pid;
298                         gettimeofday(&signal_record[proc_id][signal_index[proc_id]].time, NULL);
299                         eip = ptrace(PTRACE_PEEKUSR, pid, PT_IP_OFFSET, 0);
300                         signal_record[proc_id][signal_index[proc_id]].addr = eip;
301                         signal_record[proc_id][signal_index[proc_id]++].signal = sig;
302 #endif
303                         if(proc_id == -1){
304                                 sleeping_process_signal(pid, sig);
305                                 continue;
306                         }
307
308                         task = cpu_tasks[proc_id].task;
309                         tracing = is_tracing(task);
310                         old_tracing = tracing;
311
312                         /* Assume: no syscall, when coming from user */
313                         if ( tracing )
314                                 do_sigtrap(task);
315
316                         switch(sig){
317                         case SIGUSR1:
318                                 sig = 0;
319                                 op = do_proc_op(task, proc_id);
320                                 switch(op){
321                                 /*
322                                  * This is called when entering user mode; after
323                                  * this, we start intercepting syscalls.
324                                  *
325                                  * In fact, a process is started in kernel mode,
326                                  * so with is_tracing() == 0 (and that is reset
327                                  * when executing syscalls, since UML kernel has
328                                  * the right to do syscalls);
329                                  */
330                                 case OP_TRACE_ON:
331                                         arch_leave_kernel(task, pid);
332                                         tracing = 1;
333                                         break;
334                                 case OP_REBOOT:
335                                 case OP_HALT:
336                                         unmap_physmem();
337                                         kmalloc_ok = 0;
338                                         os_kill_ptraced_process(pid, 0);
339                                         /* Now let's reap remaining zombies */
340                                         errno = 0;
341                                         do {
342                                                 waitpid(-1, &status,
343                                                         WUNTRACED);
344                                         } while (errno != ECHILD);
345                                         return(op == OP_REBOOT);
346                                 case OP_NONE:
347                                         printf("Detaching pid %d\n", pid);
348                                         detach(pid, SIGSTOP);
349                                         continue;
350                                 default:
351                                         break;
352                                 }
353                                 /* OP_EXEC switches host processes on us,
354                                  * we want to continue the new one.
355                                  */
356                                 pid = cpu_tasks[proc_id].pid;
357                                 break;
358                         case (SIGTRAP + 0x80):
359                                 if(!tracing && (debugger_pid != -1)){
360                                         child_signal(pid, status & 0x7fff);
361                                         continue;
362                                 }
363                                 tracing = 0;
364                                 /* local_using_sysemu has been already set
365                                  * below, since if we are here, is_tracing() on
366                                  * the traced task was 1, i.e. the process had
367                                  * already run through one iteration of the
368                                  * loop which executed a OP_TRACE_ON request.*/
369                                 do_syscall(task, pid, local_using_sysemu);
370                                 sig = SIGUSR2;
371                                 break;
372                         case SIGTRAP:
373                                 if(!tracing && (debugger_pid != -1)){
374                                         child_signal(pid, status);
375                                         continue;
376                                 }
377                                 tracing = 0;
378                                 break;
379                         case SIGPROF:
380                                 if(tracing) sig = 0;
381                                 break;
382                         case SIGCHLD:
383                         case SIGHUP:
384                                 sig = 0;
385                                 break;
386                         case SIGSEGV:
387                         case SIGIO:
388                         case SIGALRM:
389                         case SIGVTALRM:
390                         case SIGFPE:
391                         case SIGBUS:
392                         case SIGILL:
393                         case SIGWINCH:
394
395                         default:
396                                 tracing = 0;
397                                 break;
398                         }
399                         set_tracing(task, tracing);
400
401                         if(!tracing && old_tracing)
402                                 arch_enter_kernel(task, pid);
403
404                         if(!tracing && (debugger_pid != -1) && (sig != 0) &&
405                                 (sig != SIGALRM) && (sig != SIGVTALRM) &&
406                                 (sig != SIGSEGV) && (sig != SIGTRAP) &&
407                                 (sig != SIGUSR2) && (sig != SIGIO) &&
408                                 (sig != SIGFPE)){
409                                 child_signal(pid, status);
410                                 continue;
411                         }
412
413                         local_using_sysemu = get_using_sysemu();
414
415                         if(tracing)
416                                 cont_type = SELECT_PTRACE_OPERATION(local_using_sysemu,
417                                                                     singlestepping(task));
418                         else if((debugger_pid != -1) && strace)
419                                 cont_type = PTRACE_SYSCALL;
420                         else
421                                 cont_type = PTRACE_CONT;
422
423                         if(ptrace(cont_type, pid, 0, sig) != 0){
424                                 tracer_panic("ptrace failed to continue "
425                                              "process - errno = %d\n", 
426                                              errno);
427                         }
428                 }
429         }
430         return(0);
431 }
432
433 static int __init uml_debug_setup(char *line, int *add)
434 {
435         char *next;
436
437         debug = 1;
438         *add = 0;
439         if(*line != '=') return(0);
440         line++;
441
442         while(line != NULL){
443                 next = strchr(line, ',');
444                 if(next) *next++ = '\0';
445                 
446                 if(!strcmp(line, "go")) debug_stop = 0;
447                 else if(!strcmp(line, "parent")) debug_parent = 1;
448                 else printf("Unknown debug option : '%s'\n", line);
449
450                 line = next;
451         }
452         return(0);
453 }
454
455 __uml_setup("debug", uml_debug_setup,
456 "debug\n"
457 "    Starts up the kernel under the control of gdb. See the \n"
458 "    kernel debugging tutorial and the debugging session pages\n"
459 "    at http://user-mode-linux.sourceforge.net/ for more information.\n\n"
460 );
461
462 static int __init uml_debugtrace_setup(char *line, int *add)
463 {
464         debug_trace = 1;
465         return 0;
466 }
467 __uml_setup("debugtrace", uml_debugtrace_setup,
468 "debugtrace\n"
469 "    Causes the tracing thread to pause until it is attached by a\n"
470 "    debugger and continued.  This is mostly for debugging crashes\n"
471 "    early during boot, and should be pretty much obsoleted by\n"
472 "    the debug switch.\n\n"
473 );
474
475 /*
476  * Overrides for Emacs so that we follow Linus's tabbing style.
477  * Emacs will notice this stuff at the end of the file and automatically
478  * adjust the settings for this buffer only.  This must remain at the end
479  * of the file.
480  * ---------------------------------------------------------------------------
481  * Local variables:
482  * c-file-style: "linux"
483  * End:
484  */