xref: /openbmc/linux/arch/um/os-Linux/start_up.c (revision 36e45463)
1 /*
2  * Copyright (C) 2000, 2001, 2002 Jeff Dike (jdike@karaya.com)
3  * Licensed under the GPL
4  */
5 
6 #include <pty.h>
7 #include <stdio.h>
8 #include <stddef.h>
9 #include <stdarg.h>
10 #include <stdlib.h>
11 #include <string.h>
12 #include <unistd.h>
13 #include <signal.h>
14 #include <sched.h>
15 #include <fcntl.h>
16 #include <errno.h>
17 #include <sys/time.h>
18 #include <sys/wait.h>
19 #include <sys/mman.h>
20 #include <sys/resource.h>
21 #include <asm/unistd.h>
22 #include <asm/page.h>
23 #include <sys/types.h>
24 #include "kern_util.h"
25 #include "user.h"
26 #include "signal_kern.h"
27 #include "sysdep/ptrace.h"
28 #include "sysdep/sigcontext.h"
29 #include "irq_user.h"
30 #include "ptrace_user.h"
31 #include "mem_user.h"
32 #include "init.h"
33 #include "os.h"
34 #include "uml-config.h"
35 #include "choose-mode.h"
36 #include "mode.h"
37 #include "tempfile.h"
38 #include "kern_constants.h"
39 
40 #ifdef UML_CONFIG_MODE_SKAS
41 #include "skas.h"
42 #include "skas_ptrace.h"
43 #include "registers.h"
44 #endif
45 
46 static int ptrace_child(void *arg)
47 {
48 	int ret;
49 	int pid = os_getpid(), ppid = getppid();
50 	int sc_result;
51 
52 	change_sig(SIGWINCH, 0);
53 	if(ptrace(PTRACE_TRACEME, 0, 0, 0) < 0){
54 		perror("ptrace");
55 		os_kill_process(pid, 0);
56 	}
57 	kill(pid, SIGSTOP);
58 
59 	/*This syscall will be intercepted by the parent. Don't call more than
60 	 * once, please.*/
61 	sc_result = os_getpid();
62 
63 	if (sc_result == pid)
64 		ret = 1; /*Nothing modified by the parent, we are running
65 			   normally.*/
66 	else if (sc_result == ppid)
67 		ret = 0; /*Expected in check_ptrace and check_sysemu when they
68 			   succeed in modifying the stack frame*/
69 	else
70 		ret = 2; /*Serious trouble! This could be caused by a bug in
71 			   host 2.6 SKAS3/2.6 patch before release -V6, together
72 			   with a bug in the UML code itself.*/
73 	_exit(ret);
74 }
75 
76 static void fatal_perror(char *str)
77 {
78 	perror(str);
79 	exit(1);
80 }
81 
82 static void fatal(char *fmt, ...)
83 {
84 	va_list list;
85 
86 	va_start(list, fmt);
87 	vprintf(fmt, list);
88 	va_end(list);
89 	fflush(stdout);
90 
91 	exit(1);
92 }
93 
94 static void non_fatal(char *fmt, ...)
95 {
96 	va_list list;
97 
98 	va_start(list, fmt);
99 	vprintf(fmt, list);
100 	va_end(list);
101 	fflush(stdout);
102 }
103 
104 static int start_ptraced_child(void **stack_out)
105 {
106 	void *stack;
107 	unsigned long sp;
108 	int pid, n, status;
109 
110 	stack = mmap(NULL, PAGE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
111 		     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
112 	if(stack == MAP_FAILED)
113 		fatal_perror("check_ptrace : mmap failed");
114 	sp = (unsigned long) stack + PAGE_SIZE - sizeof(void *);
115 	pid = clone(ptrace_child, (void *) sp, SIGCHLD, NULL);
116 	if(pid < 0)
117 		fatal_perror("start_ptraced_child : clone failed");
118 	CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
119 	if(n < 0)
120 		fatal_perror("check_ptrace : clone failed");
121 	if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
122 		fatal("check_ptrace : expected SIGSTOP, got status = %d",
123 		      status);
124 
125 	*stack_out = stack;
126 	return pid;
127 }
128 
129 /* When testing for SYSEMU support, if it is one of the broken versions, we
130  * must just avoid using sysemu, not panic, but only if SYSEMU features are
131  * broken.
132  * So only for SYSEMU features we test mustpanic, while normal host features
133  * must work anyway!
134  */
135 static int stop_ptraced_child(int pid, void *stack, int exitcode,
136 			      int mustexit)
137 {
138 	int status, n, ret = 0;
139 
140 	if(ptrace(PTRACE_CONT, pid, 0, 0) < 0)
141 		fatal_perror("stop_ptraced_child : ptrace failed");
142 	CATCH_EINTR(n = waitpid(pid, &status, 0));
143 	if(!WIFEXITED(status) || (WEXITSTATUS(status) != exitcode)) {
144 		int exit_with = WEXITSTATUS(status);
145 		if (exit_with == 2)
146 			non_fatal("check_ptrace : child exited with status 2. "
147 				  "Serious trouble happening! Try updating "
148 				  "your host skas patch!\nDisabling SYSEMU "
149 				  "support.");
150 		non_fatal("check_ptrace : child exited with exitcode %d, while "
151 			  "expecting %d; status 0x%x\n", exit_with,
152 			  exitcode, status);
153 		if (mustexit)
154 			exit(1);
155 		ret = -1;
156 	}
157 
158 	if(munmap(stack, PAGE_SIZE) < 0)
159 		fatal_perror("check_ptrace : munmap failed");
160 	return ret;
161 }
162 
163 /* Changed only during early boot */
164 int ptrace_faultinfo = 1;
165 int ptrace_ldt = 1;
166 int proc_mm = 1;
167 int skas_needs_stub = 0;
168 
169 static int __init skas0_cmd_param(char *str, int* add)
170 {
171 	ptrace_faultinfo = proc_mm = 0;
172 	return 0;
173 }
174 
175 /* The two __uml_setup would conflict, without this stupid alias. */
176 
177 static int __init mode_skas0_cmd_param(char *str, int* add)
178 	__attribute__((alias("skas0_cmd_param")));
179 
180 __uml_setup("skas0", skas0_cmd_param,
181 		"skas0\n"
182 		"    Disables SKAS3 usage, so that SKAS0 is used, unless \n"
183 	        "    you specify mode=tt.\n\n");
184 
185 __uml_setup("mode=skas0", mode_skas0_cmd_param,
186 		"mode=skas0\n"
187 		"    Disables SKAS3 usage, so that SKAS0 is used, unless you \n"
188 		"    specify mode=tt. Note that this was recently added - on \n"
189 		"    older kernels you must use simply \"skas0\".\n\n");
190 
191 /* Changed only during early boot */
192 static int force_sysemu_disabled = 0;
193 
194 static int __init nosysemu_cmd_param(char *str, int* add)
195 {
196 	force_sysemu_disabled = 1;
197 	return 0;
198 }
199 
200 __uml_setup("nosysemu", nosysemu_cmd_param,
201 "nosysemu\n"
202 "    Turns off syscall emulation patch for ptrace (SYSEMU) on.\n"
203 "    SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
204 "    behaviour of ptrace() and helps reducing host context switch rate.\n"
205 "    To make it working, you need a kernel patch for your host, too.\n"
206 "    See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
207 "    information.\n\n");
208 
209 static void __init check_sysemu(void)
210 {
211 	void *stack;
212 	int pid, n, status, count=0;
213 
214 	non_fatal("Checking syscall emulation patch for ptrace...");
215 	sysemu_supported = 0;
216 	pid = start_ptraced_child(&stack);
217 
218 	if(ptrace(PTRACE_SYSEMU, pid, 0, 0) < 0)
219 		goto fail;
220 
221 	CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
222 	if (n < 0)
223 		fatal_perror("check_sysemu : wait failed");
224 	if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGTRAP))
225 		fatal("check_sysemu : expected SIGTRAP, got status = %d",
226 		      status);
227 
228 	n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
229 		   os_getpid());
230 	if(n < 0)
231 		fatal_perror("check_sysemu : failed to modify system call "
232 			     "return");
233 
234 	if (stop_ptraced_child(pid, stack, 0, 0) < 0)
235 		goto fail_stopped;
236 
237 	sysemu_supported = 1;
238 	non_fatal("OK\n");
239 	set_using_sysemu(!force_sysemu_disabled);
240 
241 	non_fatal("Checking advanced syscall emulation patch for ptrace...");
242 	pid = start_ptraced_child(&stack);
243 
244 	if((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
245 		   (void *) PTRACE_O_TRACESYSGOOD) < 0))
246 		fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
247 
248 	while(1){
249 		count++;
250 		if(ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
251 			goto fail;
252 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
253 		if(n < 0)
254 			fatal_perror("check_ptrace : wait failed");
255 
256 		if(WIFSTOPPED(status) && (WSTOPSIG(status) == (SIGTRAP|0x80))){
257 			if (!count)
258 				fatal("check_ptrace : SYSEMU_SINGLESTEP "
259 				      "doesn't singlestep");
260 			n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
261 				   os_getpid());
262 			if(n < 0)
263 				fatal_perror("check_sysemu : failed to modify "
264 					     "system call return");
265 			break;
266 		}
267 		else if(WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
268 			count++;
269 		else
270 			fatal("check_ptrace : expected SIGTRAP or "
271 			      "(SIGTRAP | 0x80), got status = %d", status);
272 	}
273 	if (stop_ptraced_child(pid, stack, 0, 0) < 0)
274 		goto fail_stopped;
275 
276 	sysemu_supported = 2;
277 	non_fatal("OK\n");
278 
279 	if ( !force_sysemu_disabled )
280 		set_using_sysemu(sysemu_supported);
281 	return;
282 
283 fail:
284 	stop_ptraced_child(pid, stack, 1, 0);
285 fail_stopped:
286 	non_fatal("missing\n");
287 }
288 
289 static void __init check_ptrace(void)
290 {
291 	void *stack;
292 	int pid, syscall, n, status;
293 
294 	non_fatal("Checking that ptrace can change system call numbers...");
295 	pid = start_ptraced_child(&stack);
296 
297 	if((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
298 		   (void *) PTRACE_O_TRACESYSGOOD) < 0))
299 		fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
300 
301 	while(1){
302 		if(ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
303 			fatal_perror("check_ptrace : ptrace failed");
304 
305 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
306 		if(n < 0)
307 			fatal_perror("check_ptrace : wait failed");
308 
309 		if(!WIFSTOPPED(status) ||
310 		   (WSTOPSIG(status) != (SIGTRAP | 0x80)))
311 			fatal("check_ptrace : expected (SIGTRAP|0x80), "
312 			       "got status = %d", status);
313 
314 		syscall = ptrace(PTRACE_PEEKUSR, pid, PT_SYSCALL_NR_OFFSET,
315 				 0);
316 		if(syscall == __NR_getpid){
317 			n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
318 				   __NR_getppid);
319 			if(n < 0)
320 				fatal_perror("check_ptrace : failed to modify "
321 					     "system call");
322 			break;
323 		}
324 	}
325 	stop_ptraced_child(pid, stack, 0, 1);
326 	non_fatal("OK\n");
327 	check_sysemu();
328 }
329 
330 extern void check_tmpexec(void);
331 
332 static void __init check_coredump_limit(void)
333 {
334 	struct rlimit lim;
335 	int err = getrlimit(RLIMIT_CORE, &lim);
336 
337 	if(err){
338 		perror("Getting core dump limit");
339 		return;
340 	}
341 
342 	printf("Core dump limits :\n\tsoft - ");
343 	if(lim.rlim_cur == RLIM_INFINITY)
344 		printf("NONE\n");
345 	else printf("%lu\n", lim.rlim_cur);
346 
347 	printf("\thard - ");
348 	if(lim.rlim_max == RLIM_INFINITY)
349 		printf("NONE\n");
350 	else printf("%lu\n", lim.rlim_max);
351 }
352 
353 void __init os_early_checks(void)
354 {
355 	/* Print out the core dump limits early */
356 	check_coredump_limit();
357 
358 	check_ptrace();
359 
360 	/* Need to check this early because mmapping happens before the
361 	 * kernel is running.
362 	 */
363 	check_tmpexec();
364 }
365 
366 static int __init noprocmm_cmd_param(char *str, int* add)
367 {
368 	proc_mm = 0;
369 	return 0;
370 }
371 
372 __uml_setup("noprocmm", noprocmm_cmd_param,
373 "noprocmm\n"
374 "    Turns off usage of /proc/mm, even if host supports it.\n"
375 "    To support /proc/mm, the host needs to be patched using\n"
376 "    the current skas3 patch.\n\n");
377 
378 static int __init noptracefaultinfo_cmd_param(char *str, int* add)
379 {
380 	ptrace_faultinfo = 0;
381 	return 0;
382 }
383 
384 __uml_setup("noptracefaultinfo", noptracefaultinfo_cmd_param,
385 "noptracefaultinfo\n"
386 "    Turns off usage of PTRACE_FAULTINFO, even if host supports\n"
387 "    it. To support PTRACE_FAULTINFO, the host needs to be patched\n"
388 "    using the current skas3 patch.\n\n");
389 
390 static int __init noptraceldt_cmd_param(char *str, int* add)
391 {
392 	ptrace_ldt = 0;
393 	return 0;
394 }
395 
396 __uml_setup("noptraceldt", noptraceldt_cmd_param,
397 "noptraceldt\n"
398 "    Turns off usage of PTRACE_LDT, even if host supports it.\n"
399 "    To support PTRACE_LDT, the host needs to be patched using\n"
400 "    the current skas3 patch.\n\n");
401 
402 #ifdef UML_CONFIG_MODE_SKAS
403 static inline void check_skas3_ptrace_faultinfo(void)
404 {
405 	struct ptrace_faultinfo fi;
406 	void *stack;
407 	int pid, n;
408 
409 	non_fatal("  - PTRACE_FAULTINFO...");
410 	pid = start_ptraced_child(&stack);
411 
412 	n = ptrace(PTRACE_FAULTINFO, pid, 0, &fi);
413 	if (n < 0) {
414 		ptrace_faultinfo = 0;
415 		if(errno == EIO)
416 			non_fatal("not found\n");
417 		else
418 			perror("not found");
419 	}
420 	else {
421 		if (!ptrace_faultinfo)
422 			non_fatal("found but disabled on command line\n");
423 		else
424 			non_fatal("found\n");
425 	}
426 
427 	init_registers(pid);
428 	stop_ptraced_child(pid, stack, 1, 1);
429 }
430 
431 static inline void check_skas3_ptrace_ldt(void)
432 {
433 #ifdef PTRACE_LDT
434 	void *stack;
435 	int pid, n;
436 	unsigned char ldtbuf[40];
437 	struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
438 		.func = 2, /* read default ldt */
439 		.ptr = ldtbuf,
440 		.bytecount = sizeof(ldtbuf)};
441 
442 	non_fatal("  - PTRACE_LDT...");
443 	pid = start_ptraced_child(&stack);
444 
445 	n = ptrace(PTRACE_LDT, pid, 0, (unsigned long) &ldt_op);
446 	if (n < 0) {
447 		if(errno == EIO)
448 			non_fatal("not found\n");
449 		else {
450 			perror("not found");
451 		}
452 		ptrace_ldt = 0;
453 	}
454 	else {
455 		if(ptrace_ldt)
456 			non_fatal("found\n");
457 		else
458 			non_fatal("found, but use is disabled\n");
459 	}
460 
461 	stop_ptraced_child(pid, stack, 1, 1);
462 #else
463 	/* PTRACE_LDT might be disabled via cmdline option.
464 	 * We want to override this, else we might use the stub
465 	 * without real need
466 	 */
467 	ptrace_ldt = 1;
468 #endif
469 }
470 
471 static inline void check_skas3_proc_mm(void)
472 {
473 	non_fatal("  - /proc/mm...");
474 	if (access("/proc/mm", W_OK) < 0) {
475 		proc_mm = 0;
476 		perror("not found");
477 	}
478 	else {
479 		if (!proc_mm)
480 			non_fatal("found but disabled on command line\n");
481 		else
482 			non_fatal("found\n");
483 	}
484 }
485 
486 int can_do_skas(void)
487 {
488 	non_fatal("Checking for the skas3 patch in the host:\n");
489 
490 	check_skas3_proc_mm();
491 	check_skas3_ptrace_faultinfo();
492 	check_skas3_ptrace_ldt();
493 
494 	if(!proc_mm || !ptrace_faultinfo || !ptrace_ldt)
495 		skas_needs_stub = 1;
496 
497 	return 1;
498 }
499 #else
500 int can_do_skas(void)
501 {
502 	return 0;
503 }
504 #endif
505 
506 int __init parse_iomem(char *str, int *add)
507 {
508 	struct iomem_region *new;
509 	struct stat64 buf;
510 	char *file, *driver;
511 	int fd, size;
512 
513 	driver = str;
514 	file = strchr(str,',');
515 	if(file == NULL){
516 		printf("parse_iomem : failed to parse iomem\n");
517 		goto out;
518 	}
519 	*file = '\0';
520 	file++;
521 	fd = open(file, O_RDWR, 0);
522 	if(fd < 0){
523 		os_print_error(fd, "parse_iomem - Couldn't open io file");
524 		goto out;
525 	}
526 
527 	if(fstat64(fd, &buf) < 0){
528 		perror("parse_iomem - cannot stat_fd file");
529 		goto out_close;
530 	}
531 
532 	new = malloc(sizeof(*new));
533 	if(new == NULL){
534 		perror("Couldn't allocate iomem_region struct");
535 		goto out_close;
536 	}
537 
538 	size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
539 
540 	*new = ((struct iomem_region) { .next		= iomem_regions,
541 					.driver		= driver,
542 					.fd		= fd,
543 					.size		= size,
544 					.phys		= 0,
545 					.virt		= 0 });
546 	iomem_regions = new;
547 	iomem_size += new->size + UM_KERN_PAGE_SIZE;
548 
549 	return 0;
550  out_close:
551 	close(fd);
552  out:
553 	return 1;
554 }
555