xref: /linux/arch/um/os-Linux/start_up.c (revision 93d90ad708b8da6efc0e487b66111aa9db7f70c7)
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 <sys/time.h>
19 #include <sys/resource.h>
20 #include <asm/unistd.h>
21 #include <init.h>
22 #include <os.h>
23 #include <mem_user.h>
24 #include <ptrace_user.h>
25 #include <registers.h>
26 #include <skas.h>
27 #include <skas_ptrace.h>
28 
29 static void ptrace_child(void)
30 {
31 	int ret;
32 	/* Calling os_getpid because some libcs cached getpid incorrectly */
33 	int pid = os_getpid(), ppid = getppid();
34 	int sc_result;
35 
36 	if (change_sig(SIGWINCH, 0) < 0 ||
37 	    ptrace(PTRACE_TRACEME, 0, 0, 0) < 0) {
38 		perror("ptrace");
39 		kill(pid, SIGKILL);
40 	}
41 	kill(pid, SIGSTOP);
42 
43 	/*
44 	 * This syscall will be intercepted by the parent. Don't call more than
45 	 * once, please.
46 	 */
47 	sc_result = os_getpid();
48 
49 	if (sc_result == pid)
50 		/* Nothing modified by the parent, we are running normally. */
51 		ret = 1;
52 	else if (sc_result == ppid)
53 		/*
54 		 * Expected in check_ptrace and check_sysemu when they succeed
55 		 * in modifying the stack frame
56 		 */
57 		ret = 0;
58 	else
59 		/* Serious trouble! This could be caused by a bug in host 2.6
60 		 * SKAS3/2.6 patch before release -V6, together with a bug in
61 		 * the UML code itself.
62 		 */
63 		ret = 2;
64 
65 	exit(ret);
66 }
67 
68 static void fatal_perror(const char *str)
69 {
70 	perror(str);
71 	exit(1);
72 }
73 
74 static void fatal(char *fmt, ...)
75 {
76 	va_list list;
77 
78 	va_start(list, fmt);
79 	vfprintf(stderr, fmt, list);
80 	va_end(list);
81 
82 	exit(1);
83 }
84 
85 static void non_fatal(char *fmt, ...)
86 {
87 	va_list list;
88 
89 	va_start(list, fmt);
90 	vfprintf(stderr, fmt, list);
91 	va_end(list);
92 }
93 
94 static int start_ptraced_child(void)
95 {
96 	int pid, n, status;
97 
98 	pid = fork();
99 	if (pid == 0)
100 		ptrace_child();
101 	else if (pid < 0)
102 		fatal_perror("start_ptraced_child : fork failed");
103 
104 	CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
105 	if (n < 0)
106 		fatal_perror("check_ptrace : waitpid failed");
107 	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
108 		fatal("check_ptrace : expected SIGSTOP, got status = %d",
109 		      status);
110 
111 	return pid;
112 }
113 
114 /* When testing for SYSEMU support, if it is one of the broken versions, we
115  * must just avoid using sysemu, not panic, but only if SYSEMU features are
116  * broken.
117  * So only for SYSEMU features we test mustpanic, while normal host features
118  * must work anyway!
119  */
120 static int stop_ptraced_child(int pid, int exitcode, int mustexit)
121 {
122 	int status, n, ret = 0;
123 
124 	if (ptrace(PTRACE_CONT, pid, 0, 0) < 0) {
125 		perror("stop_ptraced_child : ptrace failed");
126 		return -1;
127 	}
128 	CATCH_EINTR(n = waitpid(pid, &status, 0));
129 	if (!WIFEXITED(status) || (WEXITSTATUS(status) != exitcode)) {
130 		int exit_with = WEXITSTATUS(status);
131 		if (exit_with == 2)
132 			non_fatal("check_ptrace : child exited with status 2. "
133 				  "\nDisabling SYSEMU support.\n");
134 		non_fatal("check_ptrace : child exited with exitcode %d, while "
135 			  "expecting %d; status 0x%x\n", exit_with,
136 			  exitcode, status);
137 		if (mustexit)
138 			exit(1);
139 		ret = -1;
140 	}
141 
142 	return ret;
143 }
144 
145 /* Changed only during early boot */
146 int ptrace_faultinfo;
147 static int disable_ptrace_faultinfo;
148 
149 int ptrace_ldt;
150 static int disable_ptrace_ldt;
151 
152 int proc_mm;
153 static int disable_proc_mm;
154 
155 int have_switch_mm;
156 static int disable_switch_mm;
157 
158 int skas_needs_stub;
159 
160 static int __init skas0_cmd_param(char *str, int* add)
161 {
162 	disable_ptrace_faultinfo = 1;
163 	disable_ptrace_ldt = 1;
164 	disable_proc_mm = 1;
165 	disable_switch_mm = 1;
166 
167 	return 0;
168 }
169 
170 /* The two __uml_setup would conflict, without this stupid alias. */
171 
172 static int __init mode_skas0_cmd_param(char *str, int* add)
173 	__attribute__((alias("skas0_cmd_param")));
174 
175 __uml_setup("skas0", skas0_cmd_param,
176 "skas0\n"
177 "    Disables SKAS3 and SKAS4 usage, so that SKAS0 is used\n\n");
178 
179 __uml_setup("mode=skas0", mode_skas0_cmd_param,
180 "mode=skas0\n"
181 "    Disables SKAS3 and SKAS4 usage, so that SKAS0 is used.\n\n");
182 
183 /* Changed only during early boot */
184 static int force_sysemu_disabled = 0;
185 
186 static int __init nosysemu_cmd_param(char *str, int* add)
187 {
188 	force_sysemu_disabled = 1;
189 	return 0;
190 }
191 
192 __uml_setup("nosysemu", nosysemu_cmd_param,
193 "nosysemu\n"
194 "    Turns off syscall emulation patch for ptrace (SYSEMU) on.\n"
195 "    SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
196 "    behaviour of ptrace() and helps reducing host context switch rate.\n"
197 "    To make it working, you need a kernel patch for your host, too.\n"
198 "    See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
199 "    information.\n\n");
200 
201 static void __init check_sysemu(void)
202 {
203 	unsigned long regs[MAX_REG_NR];
204 	int pid, n, status, count=0;
205 
206 	non_fatal("Checking syscall emulation patch for ptrace...");
207 	sysemu_supported = 0;
208 	pid = start_ptraced_child();
209 
210 	if (ptrace(PTRACE_SYSEMU, pid, 0, 0) < 0)
211 		goto fail;
212 
213 	CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
214 	if (n < 0)
215 		fatal_perror("check_sysemu : wait failed");
216 	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGTRAP))
217 		fatal("check_sysemu : expected SIGTRAP, got status = %d\n",
218 		      status);
219 
220 	if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
221 		fatal_perror("check_sysemu : PTRACE_GETREGS failed");
222 	if (PT_SYSCALL_NR(regs) != __NR_getpid) {
223 		non_fatal("check_sysemu got system call number %d, "
224 			  "expected %d...", PT_SYSCALL_NR(regs), __NR_getpid);
225 		goto fail;
226 	}
227 
228 	n = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_RET_OFFSET, os_getpid());
229 	if (n < 0) {
230 		non_fatal("check_sysemu : failed to modify system call "
231 			  "return");
232 		goto fail;
233 	}
234 
235 	if (stop_ptraced_child(pid, 0, 0) < 0)
236 		goto fail_stopped;
237 
238 	sysemu_supported = 1;
239 	non_fatal("OK\n");
240 	set_using_sysemu(!force_sysemu_disabled);
241 
242 	non_fatal("Checking advanced syscall emulation patch for ptrace...");
243 	pid = start_ptraced_child();
244 
245 	if ((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
246 		   (void *) PTRACE_O_TRACESYSGOOD) < 0))
247 		fatal_perror("check_sysemu: PTRACE_OLDSETOPTIONS failed");
248 
249 	while (1) {
250 		count++;
251 		if (ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
252 			goto fail;
253 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
254 		if (n < 0)
255 			fatal_perror("check_sysemu: wait failed");
256 
257 		if (WIFSTOPPED(status) &&
258 		    (WSTOPSIG(status) == (SIGTRAP|0x80))) {
259 			if (!count) {
260 				non_fatal("check_sysemu: SYSEMU_SINGLESTEP "
261 					  "doesn't singlestep");
262 				goto fail;
263 			}
264 			n = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_RET_OFFSET,
265 				   os_getpid());
266 			if (n < 0)
267 				fatal_perror("check_sysemu : failed to modify "
268 					     "system call return");
269 			break;
270 		}
271 		else if (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
272 			count++;
273 		else {
274 			non_fatal("check_sysemu: expected SIGTRAP or "
275 				  "(SIGTRAP | 0x80), got status = %d\n",
276 				  status);
277 			goto fail;
278 		}
279 	}
280 	if (stop_ptraced_child(pid, 0, 0) < 0)
281 		goto fail_stopped;
282 
283 	sysemu_supported = 2;
284 	non_fatal("OK\n");
285 
286 	if (!force_sysemu_disabled)
287 		set_using_sysemu(sysemu_supported);
288 	return;
289 
290 fail:
291 	stop_ptraced_child(pid, 1, 0);
292 fail_stopped:
293 	non_fatal("missing\n");
294 }
295 
296 static void __init check_ptrace(void)
297 {
298 	int pid, syscall, n, status;
299 
300 	non_fatal("Checking that ptrace can change system call numbers...");
301 	pid = start_ptraced_child();
302 
303 	if ((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
304 		   (void *) PTRACE_O_TRACESYSGOOD) < 0))
305 		fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
306 
307 	while (1) {
308 		if (ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
309 			fatal_perror("check_ptrace : ptrace failed");
310 
311 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
312 		if (n < 0)
313 			fatal_perror("check_ptrace : wait failed");
314 
315 		if (!WIFSTOPPED(status) ||
316 		   (WSTOPSIG(status) != (SIGTRAP | 0x80)))
317 			fatal("check_ptrace : expected (SIGTRAP|0x80), "
318 			       "got status = %d", status);
319 
320 		syscall = ptrace(PTRACE_PEEKUSER, pid, PT_SYSCALL_NR_OFFSET,
321 				 0);
322 		if (syscall == __NR_getpid) {
323 			n = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
324 				   __NR_getppid);
325 			if (n < 0)
326 				fatal_perror("check_ptrace : failed to modify "
327 					     "system call");
328 			break;
329 		}
330 	}
331 	stop_ptraced_child(pid, 0, 1);
332 	non_fatal("OK\n");
333 	check_sysemu();
334 }
335 
336 extern void check_tmpexec(void);
337 
338 static void __init check_coredump_limit(void)
339 {
340 	struct rlimit lim;
341 	int err = getrlimit(RLIMIT_CORE, &lim);
342 
343 	if (err) {
344 		perror("Getting core dump limit");
345 		return;
346 	}
347 
348 	printf("Core dump limits :\n\tsoft - ");
349 	if (lim.rlim_cur == RLIM_INFINITY)
350 		printf("NONE\n");
351 	else printf("%lu\n", lim.rlim_cur);
352 
353 	printf("\thard - ");
354 	if (lim.rlim_max == RLIM_INFINITY)
355 		printf("NONE\n");
356 	else printf("%lu\n", lim.rlim_max);
357 }
358 
359 void __init os_early_checks(void)
360 {
361 	int pid;
362 
363 	/* Print out the core dump limits early */
364 	check_coredump_limit();
365 
366 	check_ptrace();
367 
368 	/* Need to check this early because mmapping happens before the
369 	 * kernel is running.
370 	 */
371 	check_tmpexec();
372 
373 	pid = start_ptraced_child();
374 	if (init_registers(pid))
375 		fatal("Failed to initialize default registers");
376 	stop_ptraced_child(pid, 1, 1);
377 }
378 
379 static int __init noprocmm_cmd_param(char *str, int* add)
380 {
381 	disable_proc_mm = 1;
382 	return 0;
383 }
384 
385 __uml_setup("noprocmm", noprocmm_cmd_param,
386 "noprocmm\n"
387 "    Turns off usage of /proc/mm, even if host supports it.\n"
388 "    To support /proc/mm, the host needs to be patched using\n"
389 "    the current skas3 patch.\n\n");
390 
391 static int __init noptracefaultinfo_cmd_param(char *str, int* add)
392 {
393 	disable_ptrace_faultinfo = 1;
394 	return 0;
395 }
396 
397 __uml_setup("noptracefaultinfo", noptracefaultinfo_cmd_param,
398 "noptracefaultinfo\n"
399 "    Turns off usage of PTRACE_FAULTINFO, even if host supports\n"
400 "    it. To support PTRACE_FAULTINFO, the host needs to be patched\n"
401 "    using the current skas3 patch.\n\n");
402 
403 static int __init noptraceldt_cmd_param(char *str, int* add)
404 {
405 	disable_ptrace_ldt = 1;
406 	return 0;
407 }
408 
409 __uml_setup("noptraceldt", noptraceldt_cmd_param,
410 "noptraceldt\n"
411 "    Turns off usage of PTRACE_LDT, even if host supports it.\n"
412 "    To support PTRACE_LDT, the host needs to be patched using\n"
413 "    the current skas3 patch.\n\n");
414 
415 static inline void check_skas3_ptrace_faultinfo(void)
416 {
417 	struct ptrace_faultinfo fi;
418 	int pid, n;
419 
420 	non_fatal("  - PTRACE_FAULTINFO...");
421 	pid = start_ptraced_child();
422 
423 	n = ptrace(PTRACE_FAULTINFO, pid, 0, &fi);
424 	if (n < 0) {
425 		if (errno == EIO)
426 			non_fatal("not found\n");
427 		else
428 			perror("not found");
429 	} else if (disable_ptrace_faultinfo)
430 		non_fatal("found but disabled on command line\n");
431 	else {
432 		ptrace_faultinfo = 1;
433 		non_fatal("found\n");
434 	}
435 
436 	stop_ptraced_child(pid, 1, 1);
437 }
438 
439 static inline void check_skas3_ptrace_ldt(void)
440 {
441 #ifdef PTRACE_LDT
442 	int pid, n;
443 	unsigned char ldtbuf[40];
444 	struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
445 		.func = 2, /* read default ldt */
446 		.ptr = ldtbuf,
447 		.bytecount = sizeof(ldtbuf)};
448 
449 	non_fatal("  - PTRACE_LDT...");
450 	pid = start_ptraced_child();
451 
452 	n = ptrace(PTRACE_LDT, pid, 0, (unsigned long) &ldt_op);
453 	if (n < 0) {
454 		if (errno == EIO)
455 			non_fatal("not found\n");
456 		else
457 			perror("not found");
458 	} else if (disable_ptrace_ldt)
459 		non_fatal("found, but use is disabled\n");
460 	else {
461 		ptrace_ldt = 1;
462 		non_fatal("found\n");
463 	}
464 
465 	stop_ptraced_child(pid, 1, 1);
466 #endif
467 }
468 
469 static inline void check_skas3_proc_mm(void)
470 {
471 	non_fatal("  - /proc/mm...");
472 	if (access("/proc/mm", W_OK) < 0)
473 		perror("not found");
474 	else if (disable_proc_mm)
475 		non_fatal("found but disabled on command line\n");
476 	else {
477 		proc_mm = 1;
478 		non_fatal("found\n");
479 	}
480 }
481 
482 void can_do_skas(void)
483 {
484 	non_fatal("Checking for the skas3 patch in the host:\n");
485 
486 	check_skas3_proc_mm();
487 	check_skas3_ptrace_faultinfo();
488 	check_skas3_ptrace_ldt();
489 
490 	if (!proc_mm || !ptrace_faultinfo || !ptrace_ldt)
491 		skas_needs_stub = 1;
492 }
493 
494 int __init parse_iomem(char *str, int *add)
495 {
496 	struct iomem_region *new;
497 	struct stat64 buf;
498 	char *file, *driver;
499 	int fd, size;
500 
501 	driver = str;
502 	file = strchr(str,',');
503 	if (file == NULL) {
504 		fprintf(stderr, "parse_iomem : failed to parse iomem\n");
505 		goto out;
506 	}
507 	*file = '\0';
508 	file++;
509 	fd = open(file, O_RDWR, 0);
510 	if (fd < 0) {
511 		perror("parse_iomem - Couldn't open io file");
512 		goto out;
513 	}
514 
515 	if (fstat64(fd, &buf) < 0) {
516 		perror("parse_iomem - cannot stat_fd file");
517 		goto out_close;
518 	}
519 
520 	new = malloc(sizeof(*new));
521 	if (new == NULL) {
522 		perror("Couldn't allocate iomem_region struct");
523 		goto out_close;
524 	}
525 
526 	size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
527 
528 	*new = ((struct iomem_region) { .next		= iomem_regions,
529 					.driver		= driver,
530 					.fd		= fd,
531 					.size		= size,
532 					.phys		= 0,
533 					.virt		= 0 });
534 	iomem_regions = new;
535 	iomem_size += new->size + UM_KERN_PAGE_SIZE;
536 
537 	return 0;
538  out_close:
539 	close(fd);
540  out:
541 	return 1;
542 }
543