xref: /openbmc/linux/arch/um/os-Linux/start_up.c (revision 8fdff1dc)
1 /*
2  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.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 <errno.h>
11 #include <fcntl.h>
12 #include <sched.h>
13 #include <signal.h>
14 #include <string.h>
15 #include <sys/mman.h>
16 #include <sys/stat.h>
17 #include <sys/wait.h>
18 #include <asm/unistd.h>
19 #include <init.h>
20 #include <os.h>
21 #include <mem_user.h>
22 #include <ptrace_user.h>
23 #include <registers.h>
24 #include <skas.h>
25 #include <skas_ptrace.h>
26 
27 static void ptrace_child(void)
28 {
29 	int ret;
30 	/* Calling os_getpid because some libcs cached getpid incorrectly */
31 	int pid = os_getpid(), ppid = getppid();
32 	int sc_result;
33 
34 	if (change_sig(SIGWINCH, 0) < 0 ||
35 	    ptrace(PTRACE_TRACEME, 0, 0, 0) < 0) {
36 		perror("ptrace");
37 		kill(pid, SIGKILL);
38 	}
39 	kill(pid, SIGSTOP);
40 
41 	/*
42 	 * This syscall will be intercepted by the parent. Don't call more than
43 	 * once, please.
44 	 */
45 	sc_result = os_getpid();
46 
47 	if (sc_result == pid)
48 		/* Nothing modified by the parent, we are running normally. */
49 		ret = 1;
50 	else if (sc_result == ppid)
51 		/*
52 		 * Expected in check_ptrace and check_sysemu when they succeed
53 		 * in modifying the stack frame
54 		 */
55 		ret = 0;
56 	else
57 		/* Serious trouble! This could be caused by a bug in host 2.6
58 		 * SKAS3/2.6 patch before release -V6, together with a bug in
59 		 * the UML code itself.
60 		 */
61 		ret = 2;
62 
63 	exit(ret);
64 }
65 
66 static void fatal_perror(const char *str)
67 {
68 	perror(str);
69 	exit(1);
70 }
71 
72 static void fatal(char *fmt, ...)
73 {
74 	va_list list;
75 
76 	va_start(list, fmt);
77 	vfprintf(stderr, fmt, list);
78 	va_end(list);
79 
80 	exit(1);
81 }
82 
83 static void non_fatal(char *fmt, ...)
84 {
85 	va_list list;
86 
87 	va_start(list, fmt);
88 	vfprintf(stderr, fmt, list);
89 	va_end(list);
90 }
91 
92 static int start_ptraced_child(void)
93 {
94 	int pid, n, status;
95 
96 	pid = fork();
97 	if (pid == 0)
98 		ptrace_child();
99 	else if (pid < 0)
100 		fatal_perror("start_ptraced_child : fork failed");
101 
102 	CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
103 	if (n < 0)
104 		fatal_perror("check_ptrace : waitpid failed");
105 	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
106 		fatal("check_ptrace : expected SIGSTOP, got status = %d",
107 		      status);
108 
109 	return pid;
110 }
111 
112 /* When testing for SYSEMU support, if it is one of the broken versions, we
113  * must just avoid using sysemu, not panic, but only if SYSEMU features are
114  * broken.
115  * So only for SYSEMU features we test mustpanic, while normal host features
116  * must work anyway!
117  */
118 static int stop_ptraced_child(int pid, int exitcode, int mustexit)
119 {
120 	int status, n, ret = 0;
121 
122 	if (ptrace(PTRACE_CONT, pid, 0, 0) < 0) {
123 		perror("stop_ptraced_child : ptrace failed");
124 		return -1;
125 	}
126 	CATCH_EINTR(n = waitpid(pid, &status, 0));
127 	if (!WIFEXITED(status) || (WEXITSTATUS(status) != exitcode)) {
128 		int exit_with = WEXITSTATUS(status);
129 		if (exit_with == 2)
130 			non_fatal("check_ptrace : child exited with status 2. "
131 				  "\nDisabling SYSEMU support.\n");
132 		non_fatal("check_ptrace : child exited with exitcode %d, while "
133 			  "expecting %d; status 0x%x\n", exit_with,
134 			  exitcode, status);
135 		if (mustexit)
136 			exit(1);
137 		ret = -1;
138 	}
139 
140 	return ret;
141 }
142 
143 /* Changed only during early boot */
144 int ptrace_faultinfo;
145 static int disable_ptrace_faultinfo;
146 
147 int ptrace_ldt;
148 static int disable_ptrace_ldt;
149 
150 int proc_mm;
151 static int disable_proc_mm;
152 
153 int have_switch_mm;
154 static int disable_switch_mm;
155 
156 int skas_needs_stub;
157 
158 static int __init skas0_cmd_param(char *str, int* add)
159 {
160 	disable_ptrace_faultinfo = 1;
161 	disable_ptrace_ldt = 1;
162 	disable_proc_mm = 1;
163 	disable_switch_mm = 1;
164 
165 	return 0;
166 }
167 
168 /* The two __uml_setup would conflict, without this stupid alias. */
169 
170 static int __init mode_skas0_cmd_param(char *str, int* add)
171 	__attribute__((alias("skas0_cmd_param")));
172 
173 __uml_setup("skas0", skas0_cmd_param,
174 "skas0\n"
175 "    Disables SKAS3 and SKAS4 usage, so that SKAS0 is used\n\n");
176 
177 __uml_setup("mode=skas0", mode_skas0_cmd_param,
178 "mode=skas0\n"
179 "    Disables SKAS3 and SKAS4 usage, so that SKAS0 is used.\n\n");
180 
181 /* Changed only during early boot */
182 static int force_sysemu_disabled = 0;
183 
184 static int __init nosysemu_cmd_param(char *str, int* add)
185 {
186 	force_sysemu_disabled = 1;
187 	return 0;
188 }
189 
190 __uml_setup("nosysemu", nosysemu_cmd_param,
191 "nosysemu\n"
192 "    Turns off syscall emulation patch for ptrace (SYSEMU) on.\n"
193 "    SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
194 "    behaviour of ptrace() and helps reducing host context switch rate.\n"
195 "    To make it working, you need a kernel patch for your host, too.\n"
196 "    See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
197 "    information.\n\n");
198 
199 static void __init check_sysemu(void)
200 {
201 	unsigned long regs[MAX_REG_NR];
202 	int pid, n, status, count=0;
203 
204 	non_fatal("Checking syscall emulation patch for ptrace...");
205 	sysemu_supported = 0;
206 	pid = start_ptraced_child();
207 
208 	if (ptrace(PTRACE_SYSEMU, pid, 0, 0) < 0)
209 		goto fail;
210 
211 	CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
212 	if (n < 0)
213 		fatal_perror("check_sysemu : wait failed");
214 	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGTRAP))
215 		fatal("check_sysemu : expected SIGTRAP, got status = %d\n",
216 		      status);
217 
218 	if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
219 		fatal_perror("check_sysemu : PTRACE_GETREGS failed");
220 	if (PT_SYSCALL_NR(regs) != __NR_getpid) {
221 		non_fatal("check_sysemu got system call number %d, "
222 			  "expected %d...", PT_SYSCALL_NR(regs), __NR_getpid);
223 		goto fail;
224 	}
225 
226 	n = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_RET_OFFSET, os_getpid());
227 	if (n < 0) {
228 		non_fatal("check_sysemu : failed to modify system call "
229 			  "return");
230 		goto fail;
231 	}
232 
233 	if (stop_ptraced_child(pid, 0, 0) < 0)
234 		goto fail_stopped;
235 
236 	sysemu_supported = 1;
237 	non_fatal("OK\n");
238 	set_using_sysemu(!force_sysemu_disabled);
239 
240 	non_fatal("Checking advanced syscall emulation patch for ptrace...");
241 	pid = start_ptraced_child();
242 
243 	if ((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
244 		   (void *) PTRACE_O_TRACESYSGOOD) < 0))
245 		fatal_perror("check_sysemu: PTRACE_OLDSETOPTIONS failed");
246 
247 	while (1) {
248 		count++;
249 		if (ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
250 			goto fail;
251 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
252 		if (n < 0)
253 			fatal_perror("check_sysemu: wait failed");
254 
255 		if (WIFSTOPPED(status) &&
256 		    (WSTOPSIG(status) == (SIGTRAP|0x80))) {
257 			if (!count) {
258 				non_fatal("check_sysemu: SYSEMU_SINGLESTEP "
259 					  "doesn't singlestep");
260 				goto fail;
261 			}
262 			n = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_RET_OFFSET,
263 				   os_getpid());
264 			if (n < 0)
265 				fatal_perror("check_sysemu : failed to modify "
266 					     "system call return");
267 			break;
268 		}
269 		else if (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
270 			count++;
271 		else {
272 			non_fatal("check_sysemu: expected SIGTRAP or "
273 				  "(SIGTRAP | 0x80), got status = %d\n",
274 				  status);
275 			goto fail;
276 		}
277 	}
278 	if (stop_ptraced_child(pid, 0, 0) < 0)
279 		goto fail_stopped;
280 
281 	sysemu_supported = 2;
282 	non_fatal("OK\n");
283 
284 	if (!force_sysemu_disabled)
285 		set_using_sysemu(sysemu_supported);
286 	return;
287 
288 fail:
289 	stop_ptraced_child(pid, 1, 0);
290 fail_stopped:
291 	non_fatal("missing\n");
292 }
293 
294 static void __init check_ptrace(void)
295 {
296 	int pid, syscall, n, status;
297 
298 	non_fatal("Checking that ptrace can change system call numbers...");
299 	pid = start_ptraced_child();
300 
301 	if ((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
302 		   (void *) PTRACE_O_TRACESYSGOOD) < 0))
303 		fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
304 
305 	while (1) {
306 		if (ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
307 			fatal_perror("check_ptrace : ptrace failed");
308 
309 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
310 		if (n < 0)
311 			fatal_perror("check_ptrace : wait failed");
312 
313 		if (!WIFSTOPPED(status) ||
314 		   (WSTOPSIG(status) != (SIGTRAP | 0x80)))
315 			fatal("check_ptrace : expected (SIGTRAP|0x80), "
316 			       "got status = %d", status);
317 
318 		syscall = ptrace(PTRACE_PEEKUSER, pid, PT_SYSCALL_NR_OFFSET,
319 				 0);
320 		if (syscall == __NR_getpid) {
321 			n = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
322 				   __NR_getppid);
323 			if (n < 0)
324 				fatal_perror("check_ptrace : failed to modify "
325 					     "system call");
326 			break;
327 		}
328 	}
329 	stop_ptraced_child(pid, 0, 1);
330 	non_fatal("OK\n");
331 	check_sysemu();
332 }
333 
334 extern void check_tmpexec(void);
335 
336 static void __init check_coredump_limit(void)
337 {
338 	struct rlimit lim;
339 	int err = getrlimit(RLIMIT_CORE, &lim);
340 
341 	if (err) {
342 		perror("Getting core dump limit");
343 		return;
344 	}
345 
346 	printf("Core dump limits :\n\tsoft - ");
347 	if (lim.rlim_cur == RLIM_INFINITY)
348 		printf("NONE\n");
349 	else printf("%lu\n", lim.rlim_cur);
350 
351 	printf("\thard - ");
352 	if (lim.rlim_max == RLIM_INFINITY)
353 		printf("NONE\n");
354 	else printf("%lu\n", lim.rlim_max);
355 }
356 
357 void __init os_early_checks(void)
358 {
359 	int pid;
360 
361 	/* Print out the core dump limits early */
362 	check_coredump_limit();
363 
364 	check_ptrace();
365 
366 	/* Need to check this early because mmapping happens before the
367 	 * kernel is running.
368 	 */
369 	check_tmpexec();
370 
371 	pid = start_ptraced_child();
372 	if (init_registers(pid))
373 		fatal("Failed to initialize default registers");
374 	stop_ptraced_child(pid, 1, 1);
375 }
376 
377 static int __init noprocmm_cmd_param(char *str, int* add)
378 {
379 	disable_proc_mm = 1;
380 	return 0;
381 }
382 
383 __uml_setup("noprocmm", noprocmm_cmd_param,
384 "noprocmm\n"
385 "    Turns off usage of /proc/mm, even if host supports it.\n"
386 "    To support /proc/mm, the host needs to be patched using\n"
387 "    the current skas3 patch.\n\n");
388 
389 static int __init noptracefaultinfo_cmd_param(char *str, int* add)
390 {
391 	disable_ptrace_faultinfo = 1;
392 	return 0;
393 }
394 
395 __uml_setup("noptracefaultinfo", noptracefaultinfo_cmd_param,
396 "noptracefaultinfo\n"
397 "    Turns off usage of PTRACE_FAULTINFO, even if host supports\n"
398 "    it. To support PTRACE_FAULTINFO, the host needs to be patched\n"
399 "    using the current skas3 patch.\n\n");
400 
401 static int __init noptraceldt_cmd_param(char *str, int* add)
402 {
403 	disable_ptrace_ldt = 1;
404 	return 0;
405 }
406 
407 __uml_setup("noptraceldt", noptraceldt_cmd_param,
408 "noptraceldt\n"
409 "    Turns off usage of PTRACE_LDT, even if host supports it.\n"
410 "    To support PTRACE_LDT, the host needs to be patched using\n"
411 "    the current skas3 patch.\n\n");
412 
413 static inline void check_skas3_ptrace_faultinfo(void)
414 {
415 	struct ptrace_faultinfo fi;
416 	int pid, n;
417 
418 	non_fatal("  - PTRACE_FAULTINFO...");
419 	pid = start_ptraced_child();
420 
421 	n = ptrace(PTRACE_FAULTINFO, pid, 0, &fi);
422 	if (n < 0) {
423 		if (errno == EIO)
424 			non_fatal("not found\n");
425 		else
426 			perror("not found");
427 	} else if (disable_ptrace_faultinfo)
428 		non_fatal("found but disabled on command line\n");
429 	else {
430 		ptrace_faultinfo = 1;
431 		non_fatal("found\n");
432 	}
433 
434 	stop_ptraced_child(pid, 1, 1);
435 }
436 
437 static inline void check_skas3_ptrace_ldt(void)
438 {
439 #ifdef PTRACE_LDT
440 	int pid, n;
441 	unsigned char ldtbuf[40];
442 	struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
443 		.func = 2, /* read default ldt */
444 		.ptr = ldtbuf,
445 		.bytecount = sizeof(ldtbuf)};
446 
447 	non_fatal("  - PTRACE_LDT...");
448 	pid = start_ptraced_child();
449 
450 	n = ptrace(PTRACE_LDT, pid, 0, (unsigned long) &ldt_op);
451 	if (n < 0) {
452 		if (errno == EIO)
453 			non_fatal("not found\n");
454 		else
455 			perror("not found");
456 	} else if (disable_ptrace_ldt)
457 		non_fatal("found, but use is disabled\n");
458 	else {
459 		ptrace_ldt = 1;
460 		non_fatal("found\n");
461 	}
462 
463 	stop_ptraced_child(pid, 1, 1);
464 #endif
465 }
466 
467 static inline void check_skas3_proc_mm(void)
468 {
469 	non_fatal("  - /proc/mm...");
470 	if (access("/proc/mm", W_OK) < 0)
471 		perror("not found");
472 	else if (disable_proc_mm)
473 		non_fatal("found but disabled on command line\n");
474 	else {
475 		proc_mm = 1;
476 		non_fatal("found\n");
477 	}
478 }
479 
480 void can_do_skas(void)
481 {
482 	non_fatal("Checking for the skas3 patch in the host:\n");
483 
484 	check_skas3_proc_mm();
485 	check_skas3_ptrace_faultinfo();
486 	check_skas3_ptrace_ldt();
487 
488 	if (!proc_mm || !ptrace_faultinfo || !ptrace_ldt)
489 		skas_needs_stub = 1;
490 }
491 
492 int __init parse_iomem(char *str, int *add)
493 {
494 	struct iomem_region *new;
495 	struct stat64 buf;
496 	char *file, *driver;
497 	int fd, size;
498 
499 	driver = str;
500 	file = strchr(str,',');
501 	if (file == NULL) {
502 		fprintf(stderr, "parse_iomem : failed to parse iomem\n");
503 		goto out;
504 	}
505 	*file = '\0';
506 	file++;
507 	fd = open(file, O_RDWR, 0);
508 	if (fd < 0) {
509 		perror("parse_iomem - Couldn't open io file");
510 		goto out;
511 	}
512 
513 	if (fstat64(fd, &buf) < 0) {
514 		perror("parse_iomem - cannot stat_fd file");
515 		goto out_close;
516 	}
517 
518 	new = malloc(sizeof(*new));
519 	if (new == NULL) {
520 		perror("Couldn't allocate iomem_region struct");
521 		goto out_close;
522 	}
523 
524 	size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
525 
526 	*new = ((struct iomem_region) { .next		= iomem_regions,
527 					.driver		= driver,
528 					.fd		= fd,
529 					.size		= size,
530 					.phys		= 0,
531 					.virt		= 0 });
532 	iomem_regions = new;
533 	iomem_size += new->size + UM_KERN_PAGE_SIZE;
534 
535 	return 0;
536  out_close:
537 	close(fd);
538  out:
539 	return 1;
540 }
541