xref: /linux/arch/um/os-Linux/start_up.c (revision f8343685643f2901fe11aa9d0358cafbeaf7b4c3)
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 				  "\nDisabling SYSEMU support.\n");
148 		non_fatal("check_ptrace : child exited with exitcode %d, while "
149 			  "expecting %d; status 0x%x\n", exit_with,
150 			  exitcode, status);
151 		if (mustexit)
152 			exit(1);
153 		ret = -1;
154 	}
155 
156 	if(munmap(stack, PAGE_SIZE) < 0)
157 		fatal_perror("check_ptrace : munmap failed");
158 	return ret;
159 }
160 
161 /* Changed only during early boot */
162 int ptrace_faultinfo = 1;
163 int ptrace_ldt = 1;
164 int proc_mm = 1;
165 int skas_needs_stub = 0;
166 
167 static int __init skas0_cmd_param(char *str, int* add)
168 {
169 	ptrace_faultinfo = proc_mm = 0;
170 	return 0;
171 }
172 
173 /* The two __uml_setup would conflict, without this stupid alias. */
174 
175 static int __init mode_skas0_cmd_param(char *str, int* add)
176 	__attribute__((alias("skas0_cmd_param")));
177 
178 __uml_setup("skas0", skas0_cmd_param,
179 		"skas0\n"
180 		"    Disables SKAS3 usage, so that SKAS0 is used, unless \n"
181 	        "    you specify mode=tt.\n\n");
182 
183 __uml_setup("mode=skas0", mode_skas0_cmd_param,
184 		"mode=skas0\n"
185 		"    Disables SKAS3 usage, so that SKAS0 is used, unless you \n"
186 		"    specify mode=tt. Note that this was recently added - on \n"
187 		"    older kernels you must use simply \"skas0\".\n\n");
188 
189 /* Changed only during early boot */
190 static int force_sysemu_disabled = 0;
191 
192 static int __init nosysemu_cmd_param(char *str, int* add)
193 {
194 	force_sysemu_disabled = 1;
195 	return 0;
196 }
197 
198 __uml_setup("nosysemu", nosysemu_cmd_param,
199 "nosysemu\n"
200 "    Turns off syscall emulation patch for ptrace (SYSEMU) on.\n"
201 "    SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
202 "    behaviour of ptrace() and helps reducing host context switch rate.\n"
203 "    To make it working, you need a kernel patch for your host, too.\n"
204 "    See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
205 "    information.\n\n");
206 
207 static void __init check_sysemu(void)
208 {
209 	void *stack;
210 	unsigned long regs[MAX_REG_NR];
211 	int pid, n, status, count=0;
212 
213 	non_fatal("Checking syscall emulation patch for ptrace...");
214 	sysemu_supported = 0;
215 	pid = start_ptraced_child(&stack);
216 
217 	if(ptrace(PTRACE_SYSEMU, pid, 0, 0) < 0)
218 		goto fail;
219 
220 	CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
221 	if (n < 0)
222 		fatal_perror("check_sysemu : wait failed");
223 	if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGTRAP))
224 		fatal("check_sysemu : expected SIGTRAP, got status = %d",
225 		      status);
226 
227 	if(ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
228 		fatal_perror("check_sysemu : PTRACE_GETREGS failed");
229 	if(PT_SYSCALL_NR(regs) != __NR_getpid){
230 		non_fatal("check_sysemu got system call number %d, "
231 			  "expected %d...", PT_SYSCALL_NR(regs), __NR_getpid);
232 		goto fail;
233 	}
234 
235 	n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET, os_getpid());
236 	if(n < 0){
237 		non_fatal("check_sysemu : failed to modify system call "
238 			  "return");
239 		goto fail;
240 	}
241 
242 	if (stop_ptraced_child(pid, stack, 0, 0) < 0)
243 		goto fail_stopped;
244 
245 	sysemu_supported = 1;
246 	non_fatal("OK\n");
247 	set_using_sysemu(!force_sysemu_disabled);
248 
249 	non_fatal("Checking advanced syscall emulation patch for ptrace...");
250 	pid = start_ptraced_child(&stack);
251 
252 	if((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
253 		   (void *) PTRACE_O_TRACESYSGOOD) < 0))
254 		fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
255 
256 	while(1){
257 		count++;
258 		if(ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
259 			goto fail;
260 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
261 		if(n < 0)
262 			fatal_perror("check_ptrace : wait failed");
263 
264 		if(WIFSTOPPED(status) && (WSTOPSIG(status) == (SIGTRAP|0x80))){
265 			if (!count)
266 				fatal("check_ptrace : SYSEMU_SINGLESTEP "
267 				      "doesn't singlestep");
268 			n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
269 				   os_getpid());
270 			if(n < 0)
271 				fatal_perror("check_sysemu : failed to modify "
272 					     "system call return");
273 			break;
274 		}
275 		else if(WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
276 			count++;
277 		else
278 			fatal("check_ptrace : expected SIGTRAP or "
279 			      "(SIGTRAP | 0x80), got status = %d", status);
280 	}
281 	if (stop_ptraced_child(pid, stack, 0, 0) < 0)
282 		goto fail_stopped;
283 
284 	sysemu_supported = 2;
285 	non_fatal("OK\n");
286 
287 	if ( !force_sysemu_disabled )
288 		set_using_sysemu(sysemu_supported);
289 	return;
290 
291 fail:
292 	stop_ptraced_child(pid, stack, 1, 0);
293 fail_stopped:
294 	non_fatal("missing\n");
295 }
296 
297 static void __init check_ptrace(void)
298 {
299 	void *stack;
300 	int pid, syscall, n, status;
301 
302 	non_fatal("Checking that ptrace can change system call numbers...");
303 	pid = start_ptraced_child(&stack);
304 
305 	if((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
306 		   (void *) PTRACE_O_TRACESYSGOOD) < 0))
307 		fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
308 
309 	while(1){
310 		if(ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
311 			fatal_perror("check_ptrace : ptrace failed");
312 
313 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
314 		if(n < 0)
315 			fatal_perror("check_ptrace : wait failed");
316 
317 		if(!WIFSTOPPED(status) ||
318 		   (WSTOPSIG(status) != (SIGTRAP | 0x80)))
319 			fatal("check_ptrace : expected (SIGTRAP|0x80), "
320 			       "got status = %d", status);
321 
322 		syscall = ptrace(PTRACE_PEEKUSR, pid, PT_SYSCALL_NR_OFFSET,
323 				 0);
324 		if(syscall == __NR_getpid){
325 			n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
326 				   __NR_getppid);
327 			if(n < 0)
328 				fatal_perror("check_ptrace : failed to modify "
329 					     "system call");
330 			break;
331 		}
332 	}
333 	stop_ptraced_child(pid, stack, 0, 1);
334 	non_fatal("OK\n");
335 	check_sysemu();
336 }
337 
338 extern void check_tmpexec(void);
339 
340 static void __init check_coredump_limit(void)
341 {
342 	struct rlimit lim;
343 	int err = getrlimit(RLIMIT_CORE, &lim);
344 
345 	if(err){
346 		perror("Getting core dump limit");
347 		return;
348 	}
349 
350 	printf("Core dump limits :\n\tsoft - ");
351 	if(lim.rlim_cur == RLIM_INFINITY)
352 		printf("NONE\n");
353 	else printf("%lu\n", lim.rlim_cur);
354 
355 	printf("\thard - ");
356 	if(lim.rlim_max == RLIM_INFINITY)
357 		printf("NONE\n");
358 	else printf("%lu\n", lim.rlim_max);
359 }
360 
361 void __init os_early_checks(void)
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 
374 static int __init noprocmm_cmd_param(char *str, int* add)
375 {
376 	proc_mm = 0;
377 	return 0;
378 }
379 
380 __uml_setup("noprocmm", noprocmm_cmd_param,
381 "noprocmm\n"
382 "    Turns off usage of /proc/mm, even if host supports it.\n"
383 "    To support /proc/mm, the host needs to be patched using\n"
384 "    the current skas3 patch.\n\n");
385 
386 static int __init noptracefaultinfo_cmd_param(char *str, int* add)
387 {
388 	ptrace_faultinfo = 0;
389 	return 0;
390 }
391 
392 __uml_setup("noptracefaultinfo", noptracefaultinfo_cmd_param,
393 "noptracefaultinfo\n"
394 "    Turns off usage of PTRACE_FAULTINFO, even if host supports\n"
395 "    it. To support PTRACE_FAULTINFO, the host needs to be patched\n"
396 "    using the current skas3 patch.\n\n");
397 
398 static int __init noptraceldt_cmd_param(char *str, int* add)
399 {
400 	ptrace_ldt = 0;
401 	return 0;
402 }
403 
404 __uml_setup("noptraceldt", noptraceldt_cmd_param,
405 "noptraceldt\n"
406 "    Turns off usage of PTRACE_LDT, even if host supports it.\n"
407 "    To support PTRACE_LDT, the host needs to be patched using\n"
408 "    the current skas3 patch.\n\n");
409 
410 #ifdef UML_CONFIG_MODE_SKAS
411 static inline void check_skas3_ptrace_faultinfo(void)
412 {
413 	struct ptrace_faultinfo fi;
414 	void *stack;
415 	int pid, n;
416 
417 	non_fatal("  - PTRACE_FAULTINFO...");
418 	pid = start_ptraced_child(&stack);
419 
420 	n = ptrace(PTRACE_FAULTINFO, pid, 0, &fi);
421 	if (n < 0) {
422 		ptrace_faultinfo = 0;
423 		if(errno == EIO)
424 			non_fatal("not found\n");
425 		else
426 			perror("not found");
427 	}
428 	else {
429 		if (!ptrace_faultinfo)
430 			non_fatal("found but disabled on command line\n");
431 		else
432 			non_fatal("found\n");
433 	}
434 
435 	init_registers(pid);
436 	stop_ptraced_child(pid, stack, 1, 1);
437 }
438 
439 static inline void check_skas3_ptrace_ldt(void)
440 {
441 #ifdef PTRACE_LDT
442 	void *stack;
443 	int pid, n;
444 	unsigned char ldtbuf[40];
445 	struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
446 		.func = 2, /* read default ldt */
447 		.ptr = ldtbuf,
448 		.bytecount = sizeof(ldtbuf)};
449 
450 	non_fatal("  - PTRACE_LDT...");
451 	pid = start_ptraced_child(&stack);
452 
453 	n = ptrace(PTRACE_LDT, pid, 0, (unsigned long) &ldt_op);
454 	if (n < 0) {
455 		if(errno == EIO)
456 			non_fatal("not found\n");
457 		else {
458 			perror("not found");
459 		}
460 		ptrace_ldt = 0;
461 	}
462 	else {
463 		if(ptrace_ldt)
464 			non_fatal("found\n");
465 		else
466 			non_fatal("found, but use is disabled\n");
467 	}
468 
469 	stop_ptraced_child(pid, stack, 1, 1);
470 #else
471 	/* PTRACE_LDT might be disabled via cmdline option.
472 	 * We want to override this, else we might use the stub
473 	 * without real need
474 	 */
475 	ptrace_ldt = 1;
476 #endif
477 }
478 
479 static inline void check_skas3_proc_mm(void)
480 {
481 	non_fatal("  - /proc/mm...");
482 	if (access("/proc/mm", W_OK) < 0) {
483 		proc_mm = 0;
484 		perror("not found");
485 	}
486 	else {
487 		if (!proc_mm)
488 			non_fatal("found but disabled on command line\n");
489 		else
490 			non_fatal("found\n");
491 	}
492 }
493 
494 int can_do_skas(void)
495 {
496 	non_fatal("Checking for the skas3 patch in the host:\n");
497 
498 	check_skas3_proc_mm();
499 	check_skas3_ptrace_faultinfo();
500 	check_skas3_ptrace_ldt();
501 
502 	if(!proc_mm || !ptrace_faultinfo || !ptrace_ldt)
503 		skas_needs_stub = 1;
504 
505 	return 1;
506 }
507 #else
508 int can_do_skas(void)
509 {
510 	return 0;
511 }
512 #endif
513 
514 int __init parse_iomem(char *str, int *add)
515 {
516 	struct iomem_region *new;
517 	struct stat64 buf;
518 	char *file, *driver;
519 	int fd, size;
520 
521 	driver = str;
522 	file = strchr(str,',');
523 	if(file == NULL){
524 		printf("parse_iomem : failed to parse iomem\n");
525 		goto out;
526 	}
527 	*file = '\0';
528 	file++;
529 	fd = open(file, O_RDWR, 0);
530 	if(fd < 0){
531 		os_print_error(fd, "parse_iomem - Couldn't open io file");
532 		goto out;
533 	}
534 
535 	if(fstat64(fd, &buf) < 0){
536 		perror("parse_iomem - cannot stat_fd file");
537 		goto out_close;
538 	}
539 
540 	new = malloc(sizeof(*new));
541 	if(new == NULL){
542 		perror("Couldn't allocate iomem_region struct");
543 		goto out_close;
544 	}
545 
546 	size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
547 
548 	*new = ((struct iomem_region) { .next		= iomem_regions,
549 					.driver		= driver,
550 					.fd		= fd,
551 					.size		= size,
552 					.phys		= 0,
553 					.virt		= 0 });
554 	iomem_regions = new;
555 	iomem_size += new->size + UM_KERN_PAGE_SIZE;
556 
557 	return 0;
558  out_close:
559 	close(fd);
560  out:
561 	return 1;
562 }
563