• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
3  * Licensed under the GPL
4  */
5 
6 #include <stdlib.h>
7 #include <unistd.h>
8 #include <sched.h>
9 #include <errno.h>
10 #include <string.h>
11 #include <sys/mman.h>
12 #include <sys/wait.h>
13 #include <asm/unistd.h>
14 #include <as-layout.h>
15 #include <init.h>
16 #include <kern_util.h>
17 #include <mem.h>
18 #include <os.h>
19 #include <proc_mm.h>
20 #include <ptrace_user.h>
21 #include <registers.h>
22 #include <skas.h>
23 #include <skas_ptrace.h>
24 #include <sysdep/stub.h>
25 
is_skas_winch(int pid,int fd,void * data)26 int is_skas_winch(int pid, int fd, void *data)
27 {
28 	return pid == getpgrp();
29 }
30 
ptrace_dump_regs(int pid)31 static int ptrace_dump_regs(int pid)
32 {
33 	unsigned long regs[MAX_REG_NR];
34 	int i;
35 
36 	if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
37 		return -errno;
38 
39 	printk(UM_KERN_ERR "Stub registers -\n");
40 	for (i = 0; i < ARRAY_SIZE(regs); i++)
41 		printk(UM_KERN_ERR "\t%d - %lx\n", i, regs[i]);
42 
43 	return 0;
44 }
45 
46 /*
47  * Signals that are OK to receive in the stub - we'll just continue it.
48  * SIGWINCH will happen when UML is inside a detached screen.
49  */
50 #define STUB_SIG_MASK ((1 << SIGVTALRM) | (1 << SIGWINCH))
51 
52 /* Signals that the stub will finish with - anything else is an error */
53 #define STUB_DONE_MASK (1 << SIGTRAP)
54 
wait_stub_done(int pid)55 void wait_stub_done(int pid)
56 {
57 	int n, status, err;
58 
59 	while (1) {
60 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
61 		if ((n < 0) || !WIFSTOPPED(status))
62 			goto bad_wait;
63 
64 		if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
65 			break;
66 
67 		err = ptrace(PTRACE_CONT, pid, 0, 0);
68 		if (err) {
69 			printk(UM_KERN_ERR "wait_stub_done : continue failed, "
70 			       "errno = %d\n", errno);
71 			fatal_sigsegv();
72 		}
73 	}
74 
75 	if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
76 		return;
77 
78 bad_wait:
79 	err = ptrace_dump_regs(pid);
80 	if (err)
81 		printk(UM_KERN_ERR "Failed to get registers from stub, "
82 		       "errno = %d\n", -err);
83 	printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, "
84 	       "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno,
85 	       status);
86 	fatal_sigsegv();
87 }
88 
89 extern unsigned long current_stub_stack(void);
90 
get_skas_faultinfo(int pid,struct faultinfo * fi)91 static void get_skas_faultinfo(int pid, struct faultinfo *fi)
92 {
93 	int err;
94 
95 	if (ptrace_faultinfo) {
96 		err = ptrace(PTRACE_FAULTINFO, pid, 0, fi);
97 		if (err) {
98 			printk(UM_KERN_ERR "get_skas_faultinfo - "
99 			       "PTRACE_FAULTINFO failed, errno = %d\n", errno);
100 			fatal_sigsegv();
101 		}
102 
103 		/* Special handling for i386, which has different structs */
104 		if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo))
105 			memset((char *)fi + sizeof(struct ptrace_faultinfo), 0,
106 			       sizeof(struct faultinfo) -
107 			       sizeof(struct ptrace_faultinfo));
108 	}
109 	else {
110 		unsigned long fpregs[FP_SIZE];
111 
112 		err = get_fp_registers(pid, fpregs);
113 		if (err < 0) {
114 			printk(UM_KERN_ERR "save_fp_registers returned %d\n",
115 			       err);
116 			fatal_sigsegv();
117 		}
118 		err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
119 		if (err) {
120 			printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
121 			       "errno = %d\n", pid, errno);
122 			fatal_sigsegv();
123 		}
124 		wait_stub_done(pid);
125 
126 		/*
127 		 * faultinfo is prepared by the stub-segv-handler at start of
128 		 * the stub stack page. We just have to copy it.
129 		 */
130 		memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
131 
132 		err = put_fp_registers(pid, fpregs);
133 		if (err < 0) {
134 			printk(UM_KERN_ERR "put_fp_registers returned %d\n",
135 			       err);
136 			fatal_sigsegv();
137 		}
138 	}
139 }
140 
handle_segv(int pid,struct uml_pt_regs * regs)141 static void handle_segv(int pid, struct uml_pt_regs * regs)
142 {
143 	get_skas_faultinfo(pid, &regs->faultinfo);
144 	segv(regs->faultinfo, 0, 1, NULL);
145 }
146 
147 /*
148  * To use the same value of using_sysemu as the caller, ask it that value
149  * (in local_using_sysemu
150  */
handle_trap(int pid,struct uml_pt_regs * regs,int local_using_sysemu)151 static void handle_trap(int pid, struct uml_pt_regs *regs,
152 			int local_using_sysemu)
153 {
154 	int err, status;
155 
156 	if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
157 		fatal_sigsegv();
158 
159 	/* Mark this as a syscall */
160 	UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->gp);
161 
162 	if (!local_using_sysemu)
163 	{
164 		err = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
165 			     __NR_getpid);
166 		if (err < 0) {
167 			printk(UM_KERN_ERR "handle_trap - nullifying syscall "
168 			       "failed, errno = %d\n", errno);
169 			fatal_sigsegv();
170 		}
171 
172 		err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
173 		if (err < 0) {
174 			printk(UM_KERN_ERR "handle_trap - continuing to end of "
175 			       "syscall failed, errno = %d\n", errno);
176 			fatal_sigsegv();
177 		}
178 
179 		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
180 		if ((err < 0) || !WIFSTOPPED(status) ||
181 		    (WSTOPSIG(status) != SIGTRAP + 0x80)) {
182 			err = ptrace_dump_regs(pid);
183 			if (err)
184 				printk(UM_KERN_ERR "Failed to get registers "
185 				       "from process, errno = %d\n", -err);
186 			printk(UM_KERN_ERR "handle_trap - failed to wait at "
187 			       "end of syscall, errno = %d, status = %d\n",
188 			       errno, status);
189 			fatal_sigsegv();
190 		}
191 	}
192 
193 	handle_syscall(regs);
194 }
195 
196 extern int __syscall_stub_start;
197 
userspace_tramp(void * stack)198 static int userspace_tramp(void *stack)
199 {
200 	void *addr;
201 	int err;
202 
203 	ptrace(PTRACE_TRACEME, 0, 0, 0);
204 
205 	signal(SIGTERM, SIG_DFL);
206 	signal(SIGWINCH, SIG_IGN);
207 	err = set_interval();
208 	if (err) {
209 		printk(UM_KERN_ERR "userspace_tramp - setting timer failed, "
210 		       "errno = %d\n", err);
211 		exit(1);
212 	}
213 
214 	if (!proc_mm) {
215 		/*
216 		 * This has a pte, but it can't be mapped in with the usual
217 		 * tlb_flush mechanism because this is part of that mechanism
218 		 */
219 		int fd;
220 		unsigned long long offset;
221 		fd = phys_mapping(to_phys(&__syscall_stub_start), &offset);
222 		addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
223 			      PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
224 		if (addr == MAP_FAILED) {
225 			printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
226 			       "errno = %d\n", STUB_CODE, errno);
227 			exit(1);
228 		}
229 
230 		if (stack != NULL) {
231 			fd = phys_mapping(to_phys(stack), &offset);
232 			addr = mmap((void *) STUB_DATA,
233 				    UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
234 				    MAP_FIXED | MAP_SHARED, fd, offset);
235 			if (addr == MAP_FAILED) {
236 				printk(UM_KERN_ERR "mapping segfault stack "
237 				       "at 0x%lx failed, errno = %d\n",
238 				       STUB_DATA, errno);
239 				exit(1);
240 			}
241 		}
242 	}
243 	if (!ptrace_faultinfo && (stack != NULL)) {
244 		struct sigaction sa;
245 
246 		unsigned long v = STUB_CODE +
247 				  (unsigned long) stub_segv_handler -
248 				  (unsigned long) &__syscall_stub_start;
249 
250 		set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
251 		sigemptyset(&sa.sa_mask);
252 		sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
253 		sa.sa_sigaction = (void *) v;
254 		sa.sa_restorer = NULL;
255 		if (sigaction(SIGSEGV, &sa, NULL) < 0) {
256 			printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
257 			       "handler failed - errno = %d\n", errno);
258 			exit(1);
259 		}
260 	}
261 
262 	kill(os_getpid(), SIGSTOP);
263 	return 0;
264 }
265 
266 /* Each element set once, and only accessed by a single processor anyway */
267 #undef NR_CPUS
268 #define NR_CPUS 1
269 int userspace_pid[NR_CPUS];
270 
start_userspace(unsigned long stub_stack)271 int start_userspace(unsigned long stub_stack)
272 {
273 	void *stack;
274 	unsigned long sp;
275 	int pid, status, n, flags, err;
276 
277 	stack = mmap(NULL, UM_KERN_PAGE_SIZE,
278 		     PROT_READ | PROT_WRITE | PROT_EXEC,
279 		     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
280 	if (stack == MAP_FAILED) {
281 		err = -errno;
282 		printk(UM_KERN_ERR "start_userspace : mmap failed, "
283 		       "errno = %d\n", errno);
284 		return err;
285 	}
286 
287 	sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
288 
289 	flags = CLONE_FILES;
290 	if (proc_mm)
291 		flags |= CLONE_VM;
292 	else
293 		flags |= SIGCHLD;
294 
295 	pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
296 	if (pid < 0) {
297 		err = -errno;
298 		printk(UM_KERN_ERR "start_userspace : clone failed, "
299 		       "errno = %d\n", errno);
300 		return err;
301 	}
302 
303 	do {
304 		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
305 		if (n < 0) {
306 			err = -errno;
307 			printk(UM_KERN_ERR "start_userspace : wait failed, "
308 			       "errno = %d\n", errno);
309 			goto out_kill;
310 		}
311 	} while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
312 
313 	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
314 		err = -EINVAL;
315 		printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
316 		       "status = %d\n", status);
317 		goto out_kill;
318 	}
319 
320 	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
321 		   (void *) PTRACE_O_TRACESYSGOOD) < 0) {
322 		err = -errno;
323 		printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
324 		       "failed, errno = %d\n", errno);
325 		goto out_kill;
326 	}
327 
328 	if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
329 		err = -errno;
330 		printk(UM_KERN_ERR "start_userspace : munmap failed, "
331 		       "errno = %d\n", errno);
332 		goto out_kill;
333 	}
334 
335 	return pid;
336 
337  out_kill:
338 	os_kill_ptraced_process(pid, 1);
339 	return err;
340 }
341 
userspace(struct uml_pt_regs * regs)342 void userspace(struct uml_pt_regs *regs)
343 {
344 	struct itimerval timer;
345 	unsigned long long nsecs, now;
346 	int err, status, op, pid = userspace_pid[0];
347 	/* To prevent races if using_sysemu changes under us.*/
348 	int local_using_sysemu;
349 	siginfo_t si;
350 
351 	/* Handle any immediate reschedules or signals */
352 	interrupt_end();
353 
354 	if (getitimer(ITIMER_VIRTUAL, &timer))
355 		printk(UM_KERN_ERR "Failed to get itimer, errno = %d\n", errno);
356 	nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC +
357 		timer.it_value.tv_usec * UM_NSEC_PER_USEC;
358 	nsecs += os_nsecs();
359 
360 	while (1) {
361 		/*
362 		 * This can legitimately fail if the process loads a
363 		 * bogus value into a segment register.  It will
364 		 * segfault and PTRACE_GETREGS will read that value
365 		 * out of the process.  However, PTRACE_SETREGS will
366 		 * fail.  In this case, there is nothing to do but
367 		 * just kill the process.
368 		 */
369 		if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp))
370 			fatal_sigsegv();
371 
372 		if (put_fp_registers(pid, regs->fp))
373 			fatal_sigsegv();
374 
375 		/* Now we set local_using_sysemu to be used for one loop */
376 		local_using_sysemu = get_using_sysemu();
377 
378 		op = SELECT_PTRACE_OPERATION(local_using_sysemu,
379 					     singlestepping(NULL));
380 
381 		if (ptrace(op, pid, 0, 0)) {
382 			printk(UM_KERN_ERR "userspace - ptrace continue "
383 			       "failed, op = %d, errno = %d\n", op, errno);
384 			fatal_sigsegv();
385 		}
386 
387 		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
388 		if (err < 0) {
389 			printk(UM_KERN_ERR "userspace - wait failed, "
390 			       "errno = %d\n", errno);
391 			fatal_sigsegv();
392 		}
393 
394 		regs->is_user = 1;
395 		if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
396 			printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
397 			       "errno = %d\n", errno);
398 			fatal_sigsegv();
399 		}
400 
401 		if (get_fp_registers(pid, regs->fp)) {
402 			printk(UM_KERN_ERR "userspace -  get_fp_registers failed, "
403 			       "errno = %d\n", errno);
404 			fatal_sigsegv();
405 		}
406 
407 		UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
408 
409 		if (WIFSTOPPED(status)) {
410 			int sig = WSTOPSIG(status);
411 
412 			ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
413 
414 			switch (sig) {
415 			case SIGSEGV:
416 				if (PTRACE_FULL_FAULTINFO ||
417 				    !ptrace_faultinfo) {
418 					get_skas_faultinfo(pid,
419 							   &regs->faultinfo);
420 					(*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
421 							     regs);
422 				}
423 				else handle_segv(pid, regs);
424 				break;
425 			case SIGTRAP + 0x80:
426 			        handle_trap(pid, regs, local_using_sysemu);
427 				break;
428 			case SIGTRAP:
429 				relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
430 				break;
431 			case SIGVTALRM:
432 				now = os_nsecs();
433 				if (now < nsecs)
434 					break;
435 				block_signals();
436 				(*sig_info[sig])(sig, (struct siginfo *)&si, regs);
437 				unblock_signals();
438 				nsecs = timer.it_value.tv_sec *
439 					UM_NSEC_PER_SEC +
440 					timer.it_value.tv_usec *
441 					UM_NSEC_PER_USEC;
442 				nsecs += os_nsecs();
443 				break;
444 			case SIGIO:
445 			case SIGILL:
446 			case SIGBUS:
447 			case SIGFPE:
448 			case SIGWINCH:
449 				block_signals();
450 				(*sig_info[sig])(sig, (struct siginfo *)&si, regs);
451 				unblock_signals();
452 				break;
453 			default:
454 				printk(UM_KERN_ERR "userspace - child stopped "
455 				       "with signal %d\n", sig);
456 				fatal_sigsegv();
457 			}
458 			pid = userspace_pid[0];
459 			interrupt_end();
460 
461 			/* Avoid -ERESTARTSYS handling in host */
462 			if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
463 				PT_SYSCALL_NR(regs->gp) = -1;
464 		}
465 	}
466 }
467 
468 static unsigned long thread_regs[MAX_REG_NR];
469 static unsigned long thread_fp_regs[FP_SIZE];
470 
init_thread_regs(void)471 static int __init init_thread_regs(void)
472 {
473 	get_safe_registers(thread_regs, thread_fp_regs);
474 	/* Set parent's instruction pointer to start of clone-stub */
475 	thread_regs[REGS_IP_INDEX] = STUB_CODE +
476 				(unsigned long) stub_clone_handler -
477 				(unsigned long) &__syscall_stub_start;
478 	thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
479 		sizeof(void *);
480 #ifdef __SIGNAL_FRAMESIZE
481 	thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
482 #endif
483 	return 0;
484 }
485 
486 __initcall(init_thread_regs);
487 
copy_context_skas0(unsigned long new_stack,int pid)488 int copy_context_skas0(unsigned long new_stack, int pid)
489 {
490 	struct timeval tv = { .tv_sec = 0, .tv_usec = UM_USEC_PER_SEC / UM_HZ };
491 	int err;
492 	unsigned long current_stack = current_stub_stack();
493 	struct stub_data *data = (struct stub_data *) current_stack;
494 	struct stub_data *child_data = (struct stub_data *) new_stack;
495 	unsigned long long new_offset;
496 	int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
497 
498 	/*
499 	 * prepare offset and fd of child's stack as argument for parent's
500 	 * and child's mmap2 calls
501 	 */
502 	*data = ((struct stub_data) { .offset	= MMAP_OFFSET(new_offset),
503 				      .fd	= new_fd,
504 				      .timer    = ((struct itimerval)
505 					           { .it_value = tv,
506 						     .it_interval = tv }) });
507 
508 	err = ptrace_setregs(pid, thread_regs);
509 	if (err < 0) {
510 		err = -errno;
511 		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
512 		       "failed, pid = %d, errno = %d\n", pid, -err);
513 		return err;
514 	}
515 
516 	err = put_fp_registers(pid, thread_fp_regs);
517 	if (err < 0) {
518 		printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
519 		       "failed, pid = %d, err = %d\n", pid, err);
520 		return err;
521 	}
522 
523 	/* set a well known return code for detection of child write failure */
524 	child_data->err = 12345678;
525 
526 	/*
527 	 * Wait, until parent has finished its work: read child's pid from
528 	 * parent's stack, and check, if bad result.
529 	 */
530 	err = ptrace(PTRACE_CONT, pid, 0, 0);
531 	if (err) {
532 		err = -errno;
533 		printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
534 		       "errno = %d\n", pid, errno);
535 		return err;
536 	}
537 
538 	wait_stub_done(pid);
539 
540 	pid = data->err;
541 	if (pid < 0) {
542 		printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
543 		       "error %d\n", -pid);
544 		return pid;
545 	}
546 
547 	/*
548 	 * Wait, until child has finished too: read child's result from
549 	 * child's stack and check it.
550 	 */
551 	wait_stub_done(pid);
552 	if (child_data->err != STUB_DATA) {
553 		printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
554 		       "error %ld\n", child_data->err);
555 		err = child_data->err;
556 		goto out_kill;
557 	}
558 
559 	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
560 		   (void *)PTRACE_O_TRACESYSGOOD) < 0) {
561 		err = -errno;
562 		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
563 		       "failed, errno = %d\n", errno);
564 		goto out_kill;
565 	}
566 
567 	return pid;
568 
569  out_kill:
570 	os_kill_ptraced_process(pid, 1);
571 	return err;
572 }
573 
574 /*
575  * This is used only, if stub pages are needed, while proc_mm is
576  * available. Opening /proc/mm creates a new mm_context, which lacks
577  * the stub-pages. Thus, we map them using /proc/mm-fd
578  */
map_stub_pages(int fd,unsigned long code,unsigned long data,unsigned long stack)579 int map_stub_pages(int fd, unsigned long code, unsigned long data,
580 		   unsigned long stack)
581 {
582 	struct proc_mm_op mmop;
583 	int n;
584 	unsigned long long code_offset;
585 	int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start),
586 				   &code_offset);
587 
588 	mmop = ((struct proc_mm_op) { .op        = MM_MMAP,
589 				      .u         =
590 				      { .mmap    =
591 					{ .addr    = code,
592 					  .len     = UM_KERN_PAGE_SIZE,
593 					  .prot    = PROT_EXEC,
594 					  .flags   = MAP_FIXED | MAP_PRIVATE,
595 					  .fd      = code_fd,
596 					  .offset  = code_offset
597 	} } });
598 	CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
599 	if (n != sizeof(mmop)) {
600 		n = errno;
601 		printk(UM_KERN_ERR "mmap args - addr = 0x%lx, fd = %d, "
602 		       "offset = %llx\n", code, code_fd,
603 		       (unsigned long long) code_offset);
604 		printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for code "
605 		       "failed, err = %d\n", n);
606 		return -n;
607 	}
608 
609 	if (stack) {
610 		unsigned long long map_offset;
611 		int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
612 		mmop = ((struct proc_mm_op)
613 				{ .op        = MM_MMAP,
614 				  .u         =
615 				  { .mmap    =
616 				    { .addr    = data,
617 				      .len     = UM_KERN_PAGE_SIZE,
618 				      .prot    = PROT_READ | PROT_WRITE,
619 				      .flags   = MAP_FIXED | MAP_SHARED,
620 				      .fd      = map_fd,
621 				      .offset  = map_offset
622 		} } });
623 		CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
624 		if (n != sizeof(mmop)) {
625 			n = errno;
626 			printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for "
627 			       "data failed, err = %d\n", n);
628 			return -n;
629 		}
630 	}
631 
632 	return 0;
633 }
634 
new_thread(void * stack,jmp_buf * buf,void (* handler)(void))635 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
636 {
637 	(*buf)[0].JB_IP = (unsigned long) handler;
638 	(*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
639 		sizeof(void *);
640 }
641 
642 #define INIT_JMP_NEW_THREAD 0
643 #define INIT_JMP_CALLBACK 1
644 #define INIT_JMP_HALT 2
645 #define INIT_JMP_REBOOT 3
646 
switch_threads(jmp_buf * me,jmp_buf * you)647 void switch_threads(jmp_buf *me, jmp_buf *you)
648 {
649 	if (UML_SETJMP(me) == 0)
650 		UML_LONGJMP(you, 1);
651 }
652 
653 static jmp_buf initial_jmpbuf;
654 
655 /* XXX Make these percpu */
656 static void (*cb_proc)(void *arg);
657 static void *cb_arg;
658 static jmp_buf *cb_back;
659 
start_idle_thread(void * stack,jmp_buf * switch_buf)660 int start_idle_thread(void *stack, jmp_buf *switch_buf)
661 {
662 	int n;
663 
664 	set_handler(SIGWINCH);
665 
666 	/*
667 	 * Can't use UML_SETJMP or UML_LONGJMP here because they save
668 	 * and restore signals, with the possible side-effect of
669 	 * trying to handle any signals which came when they were
670 	 * blocked, which can't be done on this stack.
671 	 * Signals must be blocked when jumping back here and restored
672 	 * after returning to the jumper.
673 	 */
674 	n = setjmp(initial_jmpbuf);
675 	switch (n) {
676 	case INIT_JMP_NEW_THREAD:
677 		(*switch_buf)[0].JB_IP = (unsigned long) new_thread_handler;
678 		(*switch_buf)[0].JB_SP = (unsigned long) stack +
679 			UM_THREAD_SIZE - sizeof(void *);
680 		break;
681 	case INIT_JMP_CALLBACK:
682 		(*cb_proc)(cb_arg);
683 		longjmp(*cb_back, 1);
684 		break;
685 	case INIT_JMP_HALT:
686 		kmalloc_ok = 0;
687 		return 0;
688 	case INIT_JMP_REBOOT:
689 		kmalloc_ok = 0;
690 		return 1;
691 	default:
692 		printk(UM_KERN_ERR "Bad sigsetjmp return in "
693 		       "start_idle_thread - %d\n", n);
694 		fatal_sigsegv();
695 	}
696 	longjmp(*switch_buf, 1);
697 }
698 
initial_thread_cb_skas(void (* proc)(void *),void * arg)699 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
700 {
701 	jmp_buf here;
702 
703 	cb_proc = proc;
704 	cb_arg = arg;
705 	cb_back = &here;
706 
707 	block_signals();
708 	if (UML_SETJMP(&here) == 0)
709 		UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
710 	unblock_signals();
711 
712 	cb_proc = NULL;
713 	cb_arg = NULL;
714 	cb_back = NULL;
715 }
716 
halt_skas(void)717 void halt_skas(void)
718 {
719 	block_signals();
720 	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
721 }
722 
reboot_skas(void)723 void reboot_skas(void)
724 {
725 	block_signals();
726 	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
727 }
728 
__switch_mm(struct mm_id * mm_idp)729 void __switch_mm(struct mm_id *mm_idp)
730 {
731 	int err;
732 
733 	/* FIXME: need cpu pid in __switch_mm */
734 	if (proc_mm) {
735 		err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
736 			     mm_idp->u.mm_fd);
737 		if (err) {
738 			printk(UM_KERN_ERR "__switch_mm - PTRACE_SWITCH_MM "
739 			       "failed, errno = %d\n", errno);
740 			fatal_sigsegv();
741 		}
742 	}
743 	else userspace_pid[0] = mm_idp->u.pid;
744 }
745