xref: /openbmc/linux/arch/um/os-Linux/start_up.c (revision 64c70b1c)
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, UM_KERN_PAGE_SIZE,
111 		     PROT_READ | PROT_WRITE | PROT_EXEC,
112 		     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
113 	if(stack == MAP_FAILED)
114 		fatal_perror("check_ptrace : mmap failed");
115 	sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
116 	pid = clone(ptrace_child, (void *) sp, SIGCHLD, NULL);
117 	if(pid < 0)
118 		fatal_perror("start_ptraced_child : clone failed");
119 	CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
120 	if(n < 0)
121 		fatal_perror("check_ptrace : clone failed");
122 	if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
123 		fatal("check_ptrace : expected SIGSTOP, got status = %d",
124 		      status);
125 
126 	*stack_out = stack;
127 	return pid;
128 }
129 
130 /* When testing for SYSEMU support, if it is one of the broken versions, we
131  * must just avoid using sysemu, not panic, but only if SYSEMU features are
132  * broken.
133  * So only for SYSEMU features we test mustpanic, while normal host features
134  * must work anyway!
135  */
136 static int stop_ptraced_child(int pid, void *stack, int exitcode,
137 			      int mustexit)
138 {
139 	int status, n, ret = 0;
140 
141 	if(ptrace(PTRACE_CONT, pid, 0, 0) < 0)
142 		fatal_perror("stop_ptraced_child : ptrace failed");
143 	CATCH_EINTR(n = waitpid(pid, &status, 0));
144 	if(!WIFEXITED(status) || (WEXITSTATUS(status) != exitcode)) {
145 		int exit_with = WEXITSTATUS(status);
146 		if (exit_with == 2)
147 			non_fatal("check_ptrace : child exited with status 2. "
148 				  "\nDisabling SYSEMU support.\n");
149 		non_fatal("check_ptrace : child exited with exitcode %d, while "
150 			  "expecting %d; status 0x%x\n", exit_with,
151 			  exitcode, status);
152 		if (mustexit)
153 			exit(1);
154 		ret = -1;
155 	}
156 
157 	if(munmap(stack, UM_KERN_PAGE_SIZE) < 0)
158 		fatal_perror("check_ptrace : munmap failed");
159 	return ret;
160 }
161 
162 /* Changed only during early boot */
163 int ptrace_faultinfo = 1;
164 int ptrace_ldt = 1;
165 int proc_mm = 1;
166 int skas_needs_stub = 0;
167 
168 static int __init skas0_cmd_param(char *str, int* add)
169 {
170 	ptrace_faultinfo = proc_mm = 0;
171 	return 0;
172 }
173 
174 /* The two __uml_setup would conflict, without this stupid alias. */
175 
176 static int __init mode_skas0_cmd_param(char *str, int* add)
177 	__attribute__((alias("skas0_cmd_param")));
178 
179 __uml_setup("skas0", skas0_cmd_param,
180 		"skas0\n"
181 		"    Disables SKAS3 usage, so that SKAS0 is used, unless \n"
182 	        "    you specify mode=tt.\n\n");
183 
184 __uml_setup("mode=skas0", mode_skas0_cmd_param,
185 		"mode=skas0\n"
186 		"    Disables SKAS3 usage, so that SKAS0 is used, unless you \n"
187 		"    specify mode=tt. Note that this was recently added - on \n"
188 		"    older kernels you must use simply \"skas0\".\n\n");
189 
190 /* Changed only during early boot */
191 static int force_sysemu_disabled = 0;
192 
193 static int __init nosysemu_cmd_param(char *str, int* add)
194 {
195 	force_sysemu_disabled = 1;
196 	return 0;
197 }
198 
199 __uml_setup("nosysemu", nosysemu_cmd_param,
200 "nosysemu\n"
201 "    Turns off syscall emulation patch for ptrace (SYSEMU) on.\n"
202 "    SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
203 "    behaviour of ptrace() and helps reducing host context switch rate.\n"
204 "    To make it working, you need a kernel patch for your host, too.\n"
205 "    See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
206 "    information.\n\n");
207 
208 static void __init check_sysemu(void)
209 {
210 	void *stack;
211 	unsigned long regs[MAX_REG_NR];
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 	if(ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
229 		fatal_perror("check_sysemu : PTRACE_GETREGS failed");
230 	if(PT_SYSCALL_NR(regs) != __NR_getpid){
231 		non_fatal("check_sysemu got system call number %d, "
232 			  "expected %d...", PT_SYSCALL_NR(regs), __NR_getpid);
233 		goto fail;
234 	}
235 
236 	n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET, os_getpid());
237 	if(n < 0){
238 		non_fatal("check_sysemu : failed to modify system call "
239 			  "return");
240 		goto fail;
241 	}
242 
243 	if (stop_ptraced_child(pid, stack, 0, 0) < 0)
244 		goto fail_stopped;
245 
246 	sysemu_supported = 1;
247 	non_fatal("OK\n");
248 	set_using_sysemu(!force_sysemu_disabled);
249 
250 	non_fatal("Checking advanced syscall emulation patch for ptrace...");
251 	pid = start_ptraced_child(&stack);
252 
253 	if((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
254 		   (void *) PTRACE_O_TRACESYSGOOD) < 0))
255 		fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
256 
257 	while(1){
258 		count++;
259 		if(ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
260 			goto fail;
261 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
262 		if(n < 0)
263 			fatal_perror("check_ptrace : wait failed");
264 
265 		if(WIFSTOPPED(status) && (WSTOPSIG(status) == (SIGTRAP|0x80))){
266 			if (!count)
267 				fatal("check_ptrace : SYSEMU_SINGLESTEP "
268 				      "doesn't singlestep");
269 			n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
270 				   os_getpid());
271 			if(n < 0)
272 				fatal_perror("check_sysemu : failed to modify "
273 					     "system call return");
274 			break;
275 		}
276 		else if(WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
277 			count++;
278 		else
279 			fatal("check_ptrace : expected SIGTRAP or "
280 			      "(SIGTRAP | 0x80), got status = %d", status);
281 	}
282 	if (stop_ptraced_child(pid, stack, 0, 0) < 0)
283 		goto fail_stopped;
284 
285 	sysemu_supported = 2;
286 	non_fatal("OK\n");
287 
288 	if ( !force_sysemu_disabled )
289 		set_using_sysemu(sysemu_supported);
290 	return;
291 
292 fail:
293 	stop_ptraced_child(pid, stack, 1, 0);
294 fail_stopped:
295 	non_fatal("missing\n");
296 }
297 
298 static void __init check_ptrace(void)
299 {
300 	void *stack;
301 	int pid, syscall, n, status;
302 
303 	non_fatal("Checking that ptrace can change system call numbers...");
304 	pid = start_ptraced_child(&stack);
305 
306 	if((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
307 		   (void *) PTRACE_O_TRACESYSGOOD) < 0))
308 		fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
309 
310 	while(1){
311 		if(ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
312 			fatal_perror("check_ptrace : ptrace failed");
313 
314 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
315 		if(n < 0)
316 			fatal_perror("check_ptrace : wait failed");
317 
318 		if(!WIFSTOPPED(status) ||
319 		   (WSTOPSIG(status) != (SIGTRAP | 0x80)))
320 			fatal("check_ptrace : expected (SIGTRAP|0x80), "
321 			       "got status = %d", status);
322 
323 		syscall = ptrace(PTRACE_PEEKUSR, pid, PT_SYSCALL_NR_OFFSET,
324 				 0);
325 		if(syscall == __NR_getpid){
326 			n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
327 				   __NR_getppid);
328 			if(n < 0)
329 				fatal_perror("check_ptrace : failed to modify "
330 					     "system call");
331 			break;
332 		}
333 	}
334 	stop_ptraced_child(pid, stack, 0, 1);
335 	non_fatal("OK\n");
336 	check_sysemu();
337 }
338 
339 extern void check_tmpexec(void);
340 
341 static void __init check_coredump_limit(void)
342 {
343 	struct rlimit lim;
344 	int err = getrlimit(RLIMIT_CORE, &lim);
345 
346 	if(err){
347 		perror("Getting core dump limit");
348 		return;
349 	}
350 
351 	printf("Core dump limits :\n\tsoft - ");
352 	if(lim.rlim_cur == RLIM_INFINITY)
353 		printf("NONE\n");
354 	else printf("%lu\n", lim.rlim_cur);
355 
356 	printf("\thard - ");
357 	if(lim.rlim_max == RLIM_INFINITY)
358 		printf("NONE\n");
359 	else printf("%lu\n", lim.rlim_max);
360 }
361 
362 void __init os_early_checks(void)
363 {
364 	/* Print out the core dump limits early */
365 	check_coredump_limit();
366 
367 	check_ptrace();
368 
369 	/* Need to check this early because mmapping happens before the
370 	 * kernel is running.
371 	 */
372 	check_tmpexec();
373 }
374 
375 static int __init noprocmm_cmd_param(char *str, int* add)
376 {
377 	proc_mm = 0;
378 	return 0;
379 }
380 
381 __uml_setup("noprocmm", noprocmm_cmd_param,
382 "noprocmm\n"
383 "    Turns off usage of /proc/mm, even if host supports it.\n"
384 "    To support /proc/mm, the host needs to be patched using\n"
385 "    the current skas3 patch.\n\n");
386 
387 static int __init noptracefaultinfo_cmd_param(char *str, int* add)
388 {
389 	ptrace_faultinfo = 0;
390 	return 0;
391 }
392 
393 __uml_setup("noptracefaultinfo", noptracefaultinfo_cmd_param,
394 "noptracefaultinfo\n"
395 "    Turns off usage of PTRACE_FAULTINFO, even if host supports\n"
396 "    it. To support PTRACE_FAULTINFO, the host needs to be patched\n"
397 "    using the current skas3 patch.\n\n");
398 
399 static int __init noptraceldt_cmd_param(char *str, int* add)
400 {
401 	ptrace_ldt = 0;
402 	return 0;
403 }
404 
405 __uml_setup("noptraceldt", noptraceldt_cmd_param,
406 "noptraceldt\n"
407 "    Turns off usage of PTRACE_LDT, even if host supports it.\n"
408 "    To support PTRACE_LDT, the host needs to be patched using\n"
409 "    the current skas3 patch.\n\n");
410 
411 #ifdef UML_CONFIG_MODE_SKAS
412 static inline void check_skas3_ptrace_faultinfo(void)
413 {
414 	struct ptrace_faultinfo fi;
415 	void *stack;
416 	int pid, n;
417 
418 	non_fatal("  - PTRACE_FAULTINFO...");
419 	pid = start_ptraced_child(&stack);
420 
421 	n = ptrace(PTRACE_FAULTINFO, pid, 0, &fi);
422 	if (n < 0) {
423 		ptrace_faultinfo = 0;
424 		if(errno == EIO)
425 			non_fatal("not found\n");
426 		else
427 			perror("not found");
428 	}
429 	else {
430 		if (!ptrace_faultinfo)
431 			non_fatal("found but disabled on command line\n");
432 		else
433 			non_fatal("found\n");
434 	}
435 
436 	init_registers(pid);
437 	stop_ptraced_child(pid, stack, 1, 1);
438 }
439 
440 static inline void check_skas3_ptrace_ldt(void)
441 {
442 #ifdef PTRACE_LDT
443 	void *stack;
444 	int pid, n;
445 	unsigned char ldtbuf[40];
446 	struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
447 		.func = 2, /* read default ldt */
448 		.ptr = ldtbuf,
449 		.bytecount = sizeof(ldtbuf)};
450 
451 	non_fatal("  - PTRACE_LDT...");
452 	pid = start_ptraced_child(&stack);
453 
454 	n = ptrace(PTRACE_LDT, pid, 0, (unsigned long) &ldt_op);
455 	if (n < 0) {
456 		if(errno == EIO)
457 			non_fatal("not found\n");
458 		else {
459 			perror("not found");
460 		}
461 		ptrace_ldt = 0;
462 	}
463 	else {
464 		if(ptrace_ldt)
465 			non_fatal("found\n");
466 		else
467 			non_fatal("found, but use is disabled\n");
468 	}
469 
470 	stop_ptraced_child(pid, stack, 1, 1);
471 #else
472 	/* PTRACE_LDT might be disabled via cmdline option.
473 	 * We want to override this, else we might use the stub
474 	 * without real need
475 	 */
476 	ptrace_ldt = 1;
477 #endif
478 }
479 
480 static inline void check_skas3_proc_mm(void)
481 {
482 	non_fatal("  - /proc/mm...");
483 	if (access("/proc/mm", W_OK) < 0) {
484 		proc_mm = 0;
485 		perror("not found");
486 	}
487 	else {
488 		if (!proc_mm)
489 			non_fatal("found but disabled on command line\n");
490 		else
491 			non_fatal("found\n");
492 	}
493 }
494 
495 int can_do_skas(void)
496 {
497 	non_fatal("Checking for the skas3 patch in the host:\n");
498 
499 	check_skas3_proc_mm();
500 	check_skas3_ptrace_faultinfo();
501 	check_skas3_ptrace_ldt();
502 
503 	if(!proc_mm || !ptrace_faultinfo || !ptrace_ldt)
504 		skas_needs_stub = 1;
505 
506 	return 1;
507 }
508 #else
509 int can_do_skas(void)
510 {
511 	return 0;
512 }
513 #endif
514 
515 int __init parse_iomem(char *str, int *add)
516 {
517 	struct iomem_region *new;
518 	struct stat64 buf;
519 	char *file, *driver;
520 	int fd, size;
521 
522 	driver = str;
523 	file = strchr(str,',');
524 	if(file == NULL){
525 		printf("parse_iomem : failed to parse iomem\n");
526 		goto out;
527 	}
528 	*file = '\0';
529 	file++;
530 	fd = open(file, O_RDWR, 0);
531 	if(fd < 0){
532 		os_print_error(fd, "parse_iomem - Couldn't open io file");
533 		goto out;
534 	}
535 
536 	if(fstat64(fd, &buf) < 0){
537 		perror("parse_iomem - cannot stat_fd file");
538 		goto out_close;
539 	}
540 
541 	new = malloc(sizeof(*new));
542 	if(new == NULL){
543 		perror("Couldn't allocate iomem_region struct");
544 		goto out_close;
545 	}
546 
547 	size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
548 
549 	*new = ((struct iomem_region) { .next		= iomem_regions,
550 					.driver		= driver,
551 					.fd		= fd,
552 					.size		= size,
553 					.phys		= 0,
554 					.virt		= 0 });
555 	iomem_regions = new;
556 	iomem_size += new->size + UM_KERN_PAGE_SIZE;
557 
558 	return 0;
559  out_close:
560 	close(fd);
561  out:
562 	return 1;
563 }
564