xref: /openbmc/linux/arch/um/os-Linux/start_up.c (revision 8e367065)
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 <setjmp.h>
18 #include <sys/time.h>
19 #include <sys/wait.h>
20 #include <sys/mman.h>
21 #include <asm/unistd.h>
22 #include <asm/page.h>
23 #include <sys/types.h>
24 #include "user_util.h"
25 #include "kern_util.h"
26 #include "user.h"
27 #include "signal_kern.h"
28 #include "sysdep/ptrace.h"
29 #include "sysdep/sigcontext.h"
30 #include "irq_user.h"
31 #include "ptrace_user.h"
32 #include "mem_user.h"
33 #include "init.h"
34 #include "os.h"
35 #include "uml-config.h"
36 #include "choose-mode.h"
37 #include "mode.h"
38 #include "tempfile.h"
39 #include "kern_constants.h"
40 
41 #ifdef UML_CONFIG_MODE_SKAS
42 #include "skas.h"
43 #include "skas_ptrace.h"
44 #include "registers.h"
45 #endif
46 
47 static int ptrace_child(void *arg)
48 {
49 	int ret;
50 	int pid = os_getpid(), ppid = getppid();
51 	int sc_result;
52 
53 	change_sig(SIGWINCH, 0);
54 	if(ptrace(PTRACE_TRACEME, 0, 0, 0) < 0){
55 		perror("ptrace");
56 		os_kill_process(pid, 0);
57 	}
58 	os_stop_process(pid);
59 
60 	/*This syscall will be intercepted by the parent. Don't call more than
61 	 * once, please.*/
62 	sc_result = os_getpid();
63 
64 	if (sc_result == pid)
65 		ret = 1; /*Nothing modified by the parent, we are running
66 			   normally.*/
67 	else if (sc_result == ppid)
68 		ret = 0; /*Expected in check_ptrace and check_sysemu when they
69 			   succeed in modifying the stack frame*/
70 	else
71 		ret = 2; /*Serious trouble! This could be caused by a bug in
72 			   host 2.6 SKAS3/2.6 patch before release -V6, together
73 			   with a bug in the UML code itself.*/
74 	_exit(ret);
75 }
76 
77 static int start_ptraced_child(void **stack_out)
78 {
79 	void *stack;
80 	unsigned long sp;
81 	int pid, n, status;
82 
83 	stack = mmap(NULL, PAGE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
84 		     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
85 	if(stack == MAP_FAILED)
86 		panic("check_ptrace : mmap failed, errno = %d", errno);
87 	sp = (unsigned long) stack + PAGE_SIZE - sizeof(void *);
88 	pid = clone(ptrace_child, (void *) sp, SIGCHLD, NULL);
89 	if(pid < 0)
90 		panic("start_ptraced_child : clone failed, errno = %d", errno);
91 	CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
92 	if(n < 0)
93 		panic("check_ptrace : clone failed, errno = %d", errno);
94 	if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
95 		panic("check_ptrace : expected SIGSTOP, got status = %d",
96 		      status);
97 
98 	*stack_out = stack;
99 	return(pid);
100 }
101 
102 /* When testing for SYSEMU support, if it is one of the broken versions, we
103  * must just avoid using sysemu, not panic, but only if SYSEMU features are
104  * broken.
105  * So only for SYSEMU features we test mustpanic, while normal host features
106  * must work anyway!
107  */
108 static int stop_ptraced_child(int pid, void *stack, int exitcode,
109 			      int mustpanic)
110 {
111 	int status, n, ret = 0;
112 
113 	if(ptrace(PTRACE_CONT, pid, 0, 0) < 0)
114 		panic("check_ptrace : ptrace failed, errno = %d", errno);
115 	CATCH_EINTR(n = waitpid(pid, &status, 0));
116 	if(!WIFEXITED(status) || (WEXITSTATUS(status) != exitcode)) {
117 		int exit_with = WEXITSTATUS(status);
118 		if (exit_with == 2)
119 			printf("check_ptrace : child exited with status 2. "
120 			       "Serious trouble happening! Try updating your "
121 			       "host skas patch!\nDisabling SYSEMU support.");
122 		printf("check_ptrace : child exited with exitcode %d, while "
123 		      "expecting %d; status 0x%x", exit_with,
124 		      exitcode, status);
125 		if (mustpanic)
126 			panic("\n");
127 		else
128 			printf("\n");
129 		ret = -1;
130 	}
131 
132 	if(munmap(stack, PAGE_SIZE) < 0)
133 		panic("check_ptrace : munmap failed, errno = %d", errno);
134 	return ret;
135 }
136 
137 int ptrace_faultinfo = 1;
138 int ptrace_ldt = 1;
139 int proc_mm = 1;
140 int skas_needs_stub = 0;
141 
142 static int __init skas0_cmd_param(char *str, int* add)
143 {
144 	ptrace_faultinfo = proc_mm = 0;
145 	return 0;
146 }
147 
148 /* The two __uml_setup would conflict, without this stupid alias. */
149 
150 static int __init mode_skas0_cmd_param(char *str, int* add)
151 	__attribute__((alias("skas0_cmd_param")));
152 
153 __uml_setup("skas0", skas0_cmd_param,
154 		"skas0\n"
155 		"    Disables SKAS3 usage, so that SKAS0 is used, unless \n"
156 	        "    you specify mode=tt.\n\n");
157 
158 __uml_setup("mode=skas0", mode_skas0_cmd_param,
159 		"mode=skas0\n"
160 		"    Disables SKAS3 usage, so that SKAS0 is used, unless you \n"
161 		"    specify mode=tt. Note that this was recently added - on \n"
162 		"    older kernels you must use simply \"skas0\".\n\n");
163 
164 static int force_sysemu_disabled = 0;
165 
166 static int __init nosysemu_cmd_param(char *str, int* add)
167 {
168 	force_sysemu_disabled = 1;
169 	return 0;
170 }
171 
172 __uml_setup("nosysemu", nosysemu_cmd_param,
173 "nosysemu\n"
174 "    Turns off syscall emulation patch for ptrace (SYSEMU) on.\n"
175 "    SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
176 "    behaviour of ptrace() and helps reducing host context switch rate.\n"
177 "    To make it working, you need a kernel patch for your host, too.\n"
178 "    See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
179 "    information.\n\n");
180 
181 static void __init check_sysemu(void)
182 {
183 	void *stack;
184  	int pid, n, status, count=0;
185 
186 	printf("Checking syscall emulation patch for ptrace...");
187 	sysemu_supported = 0;
188 	pid = start_ptraced_child(&stack);
189 
190 	if(ptrace(PTRACE_SYSEMU, pid, 0, 0) < 0)
191 		goto fail;
192 
193 	CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
194 	if (n < 0)
195 		panic("check_sysemu : wait failed, errno = %d", errno);
196 	if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGTRAP))
197 		panic("check_sysemu : expected SIGTRAP, "
198 		      "got status = %d", status);
199 
200 	n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
201 		   os_getpid());
202 	if(n < 0)
203 		panic("check_sysemu : failed to modify system "
204 		      "call return, errno = %d", errno);
205 
206 	if (stop_ptraced_child(pid, stack, 0, 0) < 0)
207 		goto fail_stopped;
208 
209 	sysemu_supported = 1;
210 	printf("OK\n");
211 	set_using_sysemu(!force_sysemu_disabled);
212 
213 	printf("Checking advanced syscall emulation patch for ptrace...");
214 	pid = start_ptraced_child(&stack);
215 
216 	if(ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
217 		  (void *) PTRACE_O_TRACESYSGOOD) < 0)
218 		panic("check_ptrace: PTRACE_OLDSETOPTIONS failed, errno = %d",
219 		      errno);
220 
221 	while(1){
222 		count++;
223 		if(ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
224 			goto fail;
225 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
226 		if(n < 0)
227 			panic("check_ptrace : wait failed, errno = %d", errno);
228 		if(WIFSTOPPED(status) && (WSTOPSIG(status) == (SIGTRAP|0x80))){
229 			if (!count)
230 				panic("check_ptrace : SYSEMU_SINGLESTEP "
231 				      "doesn't singlestep");
232 			n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
233 				   os_getpid());
234 			if(n < 0)
235 				panic("check_sysemu : failed to modify system "
236 				      "call return, errno = %d", errno);
237 			break;
238 		}
239 		else if(WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
240 			count++;
241 		else
242 			panic("check_ptrace : expected SIGTRAP or "
243 			      "(SIGTRAP|0x80), got status = %d", status);
244 	}
245 	if (stop_ptraced_child(pid, stack, 0, 0) < 0)
246 		goto fail_stopped;
247 
248 	sysemu_supported = 2;
249 	printf("OK\n");
250 
251 	if ( !force_sysemu_disabled )
252 		set_using_sysemu(sysemu_supported);
253 	return;
254 
255 fail:
256 	stop_ptraced_child(pid, stack, 1, 0);
257 fail_stopped:
258 	printf("missing\n");
259 }
260 
261 static void __init check_ptrace(void)
262 {
263 	void *stack;
264 	int pid, syscall, n, status;
265 
266 	printf("Checking that ptrace can change system call numbers...");
267 	pid = start_ptraced_child(&stack);
268 
269 	if(ptrace(PTRACE_OLDSETOPTIONS, pid, 0, (void *)PTRACE_O_TRACESYSGOOD) < 0)
270 		panic("check_ptrace: PTRACE_OLDSETOPTIONS failed, errno = %d", errno);
271 
272 	while(1){
273 		if(ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
274 			panic("check_ptrace : ptrace failed, errno = %d",
275 			      errno);
276 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
277 		if(n < 0)
278 			panic("check_ptrace : wait failed, errno = %d", errno);
279 		if(!WIFSTOPPED(status) || (WSTOPSIG(status) != (SIGTRAP|0x80)))
280 			panic("check_ptrace : expected (SIGTRAP|0x80), "
281 			      "got status = %d", status);
282 
283 		syscall = ptrace(PTRACE_PEEKUSR, pid, PT_SYSCALL_NR_OFFSET,
284 				 0);
285 		if(syscall == __NR_getpid){
286 			n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
287 				   __NR_getppid);
288 			if(n < 0)
289 				panic("check_ptrace : failed to modify system "
290 				      "call, errno = %d", errno);
291 			break;
292 		}
293 	}
294 	stop_ptraced_child(pid, stack, 0, 1);
295 	printf("OK\n");
296 	check_sysemu();
297 }
298 
299 extern int create_tmp_file(unsigned long long len);
300 
301 static void check_tmpexec(void)
302 {
303 	void *addr;
304 	int err, fd = create_tmp_file(UM_KERN_PAGE_SIZE);
305 
306 	addr = mmap(NULL, UM_KERN_PAGE_SIZE,
307 		    PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE, fd, 0);
308 	printf("Checking PROT_EXEC mmap in /tmp...");
309 	fflush(stdout);
310 	if(addr == MAP_FAILED){
311 		err = errno;
312 		perror("failed");
313 		if(err == EPERM)
314 			printf("/tmp must be not mounted noexec\n");
315 		exit(1);
316 	}
317 	printf("OK\n");
318 	munmap(addr, UM_KERN_PAGE_SIZE);
319 
320 	close(fd);
321 }
322 
323 void os_early_checks(void)
324 {
325 	check_ptrace();
326 
327 	/* Need to check this early because mmapping happens before the
328 	 * kernel is running.
329 	 */
330 	check_tmpexec();
331 }
332 
333 static int __init noprocmm_cmd_param(char *str, int* add)
334 {
335 	proc_mm = 0;
336 	return 0;
337 }
338 
339 __uml_setup("noprocmm", noprocmm_cmd_param,
340 "noprocmm\n"
341 "    Turns off usage of /proc/mm, even if host supports it.\n"
342 "    To support /proc/mm, the host needs to be patched using\n"
343 "    the current skas3 patch.\n\n");
344 
345 static int __init noptracefaultinfo_cmd_param(char *str, int* add)
346 {
347 	ptrace_faultinfo = 0;
348 	return 0;
349 }
350 
351 __uml_setup("noptracefaultinfo", noptracefaultinfo_cmd_param,
352 "noptracefaultinfo\n"
353 "    Turns off usage of PTRACE_FAULTINFO, even if host supports\n"
354 "    it. To support PTRACE_FAULTINFO, the host needs to be patched\n"
355 "    using the current skas3 patch.\n\n");
356 
357 static int __init noptraceldt_cmd_param(char *str, int* add)
358 {
359 	ptrace_ldt = 0;
360 	return 0;
361 }
362 
363 __uml_setup("noptraceldt", noptraceldt_cmd_param,
364 "noptraceldt\n"
365 "    Turns off usage of PTRACE_LDT, even if host supports it.\n"
366 "    To support PTRACE_LDT, the host needs to be patched using\n"
367 "    the current skas3 patch.\n\n");
368 
369 #ifdef UML_CONFIG_MODE_SKAS
370 static inline void check_skas3_ptrace_faultinfo(void)
371 {
372 	struct ptrace_faultinfo fi;
373 	void *stack;
374 	int pid, n;
375 
376 	printf("  - PTRACE_FAULTINFO...");
377 	pid = start_ptraced_child(&stack);
378 
379 	n = ptrace(PTRACE_FAULTINFO, pid, 0, &fi);
380 	if (n < 0) {
381 		ptrace_faultinfo = 0;
382 		if(errno == EIO)
383 			printf("not found\n");
384 		else
385 			perror("not found");
386 	}
387 	else {
388 		if (!ptrace_faultinfo)
389 			printf("found but disabled on command line\n");
390 		else
391 			printf("found\n");
392 	}
393 
394 	init_registers(pid);
395 	stop_ptraced_child(pid, stack, 1, 1);
396 }
397 
398 static inline void check_skas3_ptrace_ldt(void)
399 {
400 #ifdef PTRACE_LDT
401 	void *stack;
402 	int pid, n;
403 	unsigned char ldtbuf[40];
404 	struct ptrace_ldt ldt_op = (struct ptrace_ldt) {
405 		.func = 2, /* read default ldt */
406 		.ptr = ldtbuf,
407 		.bytecount = sizeof(ldtbuf)};
408 
409 	printf("  - PTRACE_LDT...");
410 	pid = start_ptraced_child(&stack);
411 
412 	n = ptrace(PTRACE_LDT, pid, 0, (unsigned long) &ldt_op);
413 	if (n < 0) {
414 		if(errno == EIO)
415 			printf("not found\n");
416 		else {
417 			perror("not found");
418 		}
419 		ptrace_ldt = 0;
420 	}
421 	else {
422 		if(ptrace_ldt)
423 			printf("found\n");
424 		else
425 			printf("found, but use is disabled\n");
426 	}
427 
428 	stop_ptraced_child(pid, stack, 1, 1);
429 #else
430 	/* PTRACE_LDT might be disabled via cmdline option.
431 	 * We want to override this, else we might use the stub
432 	 * without real need
433 	 */
434 	ptrace_ldt = 1;
435 #endif
436 }
437 
438 static inline void check_skas3_proc_mm(void)
439 {
440 	printf("  - /proc/mm...");
441 	if (os_access("/proc/mm", OS_ACC_W_OK) < 0) {
442  		proc_mm = 0;
443 		printf("not found\n");
444 	}
445 	else {
446 		if (!proc_mm)
447 			printf("found but disabled on command line\n");
448 		else
449 			printf("found\n");
450 	}
451 }
452 
453 int can_do_skas(void)
454 {
455 	printf("Checking for the skas3 patch in the host:\n");
456 
457 	check_skas3_proc_mm();
458 	check_skas3_ptrace_faultinfo();
459 	check_skas3_ptrace_ldt();
460 
461 	if(!proc_mm || !ptrace_faultinfo || !ptrace_ldt)
462 		skas_needs_stub = 1;
463 
464 	return 1;
465 }
466 #else
467 int can_do_skas(void)
468 {
469 	return(0);
470 }
471 #endif
472 
473 int have_devanon = 0;
474 
475 /* Runs on boot kernel stack - already safe to use printk. */
476 
477 void check_devanon(void)
478 {
479 	int fd;
480 
481 	printk("Checking for /dev/anon on the host...");
482 	fd = open("/dev/anon", O_RDWR);
483 	if(fd < 0){
484 		printk("Not available (open failed with errno %d)\n", errno);
485 		return;
486 	}
487 
488 	printk("OK\n");
489 	have_devanon = 1;
490 }
491 
492 int __init parse_iomem(char *str, int *add)
493 {
494 	struct iomem_region *new;
495 	struct uml_stat buf;
496 	char *file, *driver;
497 	int fd, err, size;
498 
499 	driver = str;
500 	file = strchr(str,',');
501 	if(file == NULL){
502 		printf("parse_iomem : failed to parse iomem\n");
503 		goto out;
504 	}
505 	*file = '\0';
506 	file++;
507 	fd = os_open_file(file, of_rdwr(OPENFLAGS()), 0);
508 	if(fd < 0){
509 		os_print_error(fd, "parse_iomem - Couldn't open io file");
510 		goto out;
511 	}
512 
513 	err = os_stat_fd(fd, &buf);
514 	if(err < 0){
515 		os_print_error(err, "parse_iomem - cannot stat_fd file");
516 		goto out_close;
517 	}
518 
519 	new = malloc(sizeof(*new));
520 	if(new == NULL){
521 		perror("Couldn't allocate iomem_region struct");
522 		goto out_close;
523 	}
524 
525 	size = (buf.ust_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
526 
527 	*new = ((struct iomem_region) { .next		= iomem_regions,
528 					.driver		= driver,
529 					.fd		= fd,
530 					.size		= size,
531 					.phys		= 0,
532 					.virt		= 0 });
533 	iomem_regions = new;
534 	iomem_size += new->size + UM_KERN_PAGE_SIZE;
535 
536 	return(0);
537  out_close:
538 	os_close_file(fd);
539  out:
540 	return(1);
541 }
542 
543 
544 /* Changed during early boot */
545 int pty_output_sigio = 0;
546 int pty_close_sigio = 0;
547 
548 /* Used as a flag during SIGIO testing early in boot */
549 static volatile int got_sigio = 0;
550 
551 static void __init handler(int sig)
552 {
553 	got_sigio = 1;
554 }
555 
556 struct openpty_arg {
557 	int master;
558 	int slave;
559 	int err;
560 };
561 
562 static void openpty_cb(void *arg)
563 {
564 	struct openpty_arg *info = arg;
565 
566 	info->err = 0;
567 	if(openpty(&info->master, &info->slave, NULL, NULL, NULL))
568 		info->err = -errno;
569 }
570 
571 static void __init check_one_sigio(void (*proc)(int, int))
572 {
573 	struct sigaction old, new;
574 	struct openpty_arg pty = { .master = -1, .slave = -1 };
575 	int master, slave, err;
576 
577 	initial_thread_cb(openpty_cb, &pty);
578 	if(pty.err){
579 		printk("openpty failed, errno = %d\n", -pty.err);
580 		return;
581 	}
582 
583 	master = pty.master;
584 	slave = pty.slave;
585 
586 	if((master == -1) || (slave == -1)){
587 		printk("openpty failed to allocate a pty\n");
588 		return;
589 	}
590 
591 	/* Not now, but complain so we now where we failed. */
592 	err = raw(master);
593 	if (err < 0)
594 		panic("check_sigio : __raw failed, errno = %d\n", -err);
595 
596 	err = os_sigio_async(master, slave);
597 	if(err < 0)
598 		panic("tty_fds : sigio_async failed, err = %d\n", -err);
599 
600 	if(sigaction(SIGIO, NULL, &old) < 0)
601 		panic("check_sigio : sigaction 1 failed, errno = %d\n", errno);
602 	new = old;
603 	new.sa_handler = handler;
604 	if(sigaction(SIGIO, &new, NULL) < 0)
605 		panic("check_sigio : sigaction 2 failed, errno = %d\n", errno);
606 
607 	got_sigio = 0;
608 	(*proc)(master, slave);
609 
610 	close(master);
611 	close(slave);
612 
613 	if(sigaction(SIGIO, &old, NULL) < 0)
614 		panic("check_sigio : sigaction 3 failed, errno = %d\n", errno);
615 }
616 
617 static void tty_output(int master, int slave)
618 {
619 	int n;
620 	char buf[512];
621 
622 	printk("Checking that host ptys support output SIGIO...");
623 
624 	memset(buf, 0, sizeof(buf));
625 
626 	while(os_write_file(master, buf, sizeof(buf)) > 0) ;
627 	if(errno != EAGAIN)
628 		panic("check_sigio : write failed, errno = %d\n", errno);
629 	while(((n = os_read_file(slave, buf, sizeof(buf))) > 0) && !got_sigio) ;
630 
631 	if(got_sigio){
632 		printk("Yes\n");
633 		pty_output_sigio = 1;
634 	}
635 	else if(n == -EAGAIN) printk("No, enabling workaround\n");
636 	else panic("check_sigio : read failed, err = %d\n", n);
637 }
638 
639 static void tty_close(int master, int slave)
640 {
641 	printk("Checking that host ptys support SIGIO on close...");
642 
643 	close(slave);
644 	if(got_sigio){
645 		printk("Yes\n");
646 		pty_close_sigio = 1;
647 	}
648 	else printk("No, enabling workaround\n");
649 }
650 
651 void __init check_sigio(void)
652 {
653 	if((os_access("/dev/ptmx", OS_ACC_R_OK) < 0) &&
654 	   (os_access("/dev/ptyp0", OS_ACC_R_OK) < 0)){
655 		printk("No pseudo-terminals available - skipping pty SIGIO "
656 		       "check\n");
657 		return;
658 	}
659 	check_one_sigio(tty_output);
660 	check_one_sigio(tty_close);
661 }
662 
663 void os_check_bugs(void)
664 {
665 	check_ptrace();
666 	check_sigio();
667 	check_devanon();
668 }
669 
670