Linux Audio

Check our new training course

Loading...
v5.14.15
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
  4 * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
 
  5 */
  6
  7#include <stdlib.h>
  8#include <unistd.h>
  9#include <sched.h>
 10#include <errno.h>
 11#include <string.h>
 12#include <sys/mman.h>
 13#include <sys/wait.h>
 14#include <asm/unistd.h>
 15#include <as-layout.h>
 16#include <init.h>
 17#include <kern_util.h>
 18#include <mem.h>
 19#include <os.h>
 
 20#include <ptrace_user.h>
 21#include <registers.h>
 22#include <skas.h>
 
 23#include <sysdep/stub.h>
 24#include <linux/threads.h>
 25
 26int is_skas_winch(int pid, int fd, void *data)
 27{
 28	return pid == getpgrp();
 29}
 30
 31static const char *ptrace_reg_name(int idx)
 32{
 33#define R(n) case HOST_##n: return #n
 34
 35	switch (idx) {
 36#ifdef __x86_64__
 37	R(BX);
 38	R(CX);
 39	R(DI);
 40	R(SI);
 41	R(DX);
 42	R(BP);
 43	R(AX);
 44	R(R8);
 45	R(R9);
 46	R(R10);
 47	R(R11);
 48	R(R12);
 49	R(R13);
 50	R(R14);
 51	R(R15);
 52	R(ORIG_AX);
 53	R(CS);
 54	R(SS);
 55	R(EFLAGS);
 56#elif defined(__i386__)
 57	R(IP);
 58	R(SP);
 59	R(EFLAGS);
 60	R(AX);
 61	R(BX);
 62	R(CX);
 63	R(DX);
 64	R(SI);
 65	R(DI);
 66	R(BP);
 67	R(CS);
 68	R(SS);
 69	R(DS);
 70	R(FS);
 71	R(ES);
 72	R(GS);
 73	R(ORIG_AX);
 74#endif
 75	}
 76	return "";
 77}
 78
 79static int ptrace_dump_regs(int pid)
 80{
 81	unsigned long regs[MAX_REG_NR];
 82	int i;
 83
 84	if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
 85		return -errno;
 86
 87	printk(UM_KERN_ERR "Stub registers -\n");
 88	for (i = 0; i < ARRAY_SIZE(regs); i++) {
 89		const char *regname = ptrace_reg_name(i);
 90
 91		printk(UM_KERN_ERR "\t%s\t(%2d): %lx\n", regname, i, regs[i]);
 92	}
 93
 94	return 0;
 95}
 96
 97/*
 98 * Signals that are OK to receive in the stub - we'll just continue it.
 99 * SIGWINCH will happen when UML is inside a detached screen.
100 */
101#define STUB_SIG_MASK ((1 << SIGALRM) | (1 << SIGWINCH))
102
103/* Signals that the stub will finish with - anything else is an error */
104#define STUB_DONE_MASK (1 << SIGTRAP)
105
106void wait_stub_done(int pid)
107{
108	int n, status, err;
109
110	while (1) {
111		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
112		if ((n < 0) || !WIFSTOPPED(status))
113			goto bad_wait;
114
115		if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
116			break;
117
118		err = ptrace(PTRACE_CONT, pid, 0, 0);
119		if (err) {
120			printk(UM_KERN_ERR "wait_stub_done : continue failed, "
121			       "errno = %d\n", errno);
122			fatal_sigsegv();
123		}
124	}
125
126	if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
127		return;
 
 
128
129bad_wait:
130	err = ptrace_dump_regs(pid);
131	if (err)
132		printk(UM_KERN_ERR "Failed to get registers from stub, "
133		       "errno = %d\n", -err);
134	printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, "
135	       "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno,
136	       status);
137	fatal_sigsegv();
 
 
 
138}
139
140extern unsigned long current_stub_stack(void);
141
142static void get_skas_faultinfo(int pid, struct faultinfo *fi, unsigned long *aux_fp_regs)
143{
144	int err;
145
146	err = get_fp_registers(pid, aux_fp_regs);
147	if (err < 0) {
148		printk(UM_KERN_ERR "save_fp_registers returned %d\n",
149		       err);
150		fatal_sigsegv();
151	}
152	err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
153	if (err) {
154		printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
155		       "errno = %d\n", pid, errno);
156		fatal_sigsegv();
 
 
157	}
158	wait_stub_done(pid);
 
159
160	/*
161	 * faultinfo is prepared by the stub_segv_handler at start of
162	 * the stub stack page. We just have to copy it.
163	 */
164	memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
 
 
 
 
 
 
 
 
165
166	err = put_fp_registers(pid, aux_fp_regs);
167	if (err < 0) {
168		printk(UM_KERN_ERR "put_fp_registers returned %d\n",
169		       err);
170		fatal_sigsegv();
 
 
 
 
 
 
 
171	}
172}
173
174static void handle_segv(int pid, struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
175{
176	get_skas_faultinfo(pid, &regs->faultinfo, aux_fp_regs);
177	segv(regs->faultinfo, 0, 1, NULL);
178}
179
180/*
181 * To use the same value of using_sysemu as the caller, ask it that value
182 * (in local_using_sysemu
183 */
184static void handle_trap(int pid, struct uml_pt_regs *regs,
185			int local_using_sysemu)
186{
187	int err, status;
188
189	if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
190		fatal_sigsegv();
191
 
 
 
192	if (!local_using_sysemu)
193	{
194		err = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
195			     __NR_getpid);
196		if (err < 0) {
197			printk(UM_KERN_ERR "handle_trap - nullifying syscall "
198			       "failed, errno = %d\n", errno);
199			fatal_sigsegv();
200		}
201
202		err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
203		if (err < 0) {
204			printk(UM_KERN_ERR "handle_trap - continuing to end of "
205			       "syscall failed, errno = %d\n", errno);
206			fatal_sigsegv();
207		}
208
209		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
210		if ((err < 0) || !WIFSTOPPED(status) ||
211		    (WSTOPSIG(status) != SIGTRAP + 0x80)) {
212			err = ptrace_dump_regs(pid);
213			if (err)
214				printk(UM_KERN_ERR "Failed to get registers "
215				       "from process, errno = %d\n", -err);
216			printk(UM_KERN_ERR "handle_trap - failed to wait at "
217			       "end of syscall, errno = %d, status = %d\n",
218			       errno, status);
219			fatal_sigsegv();
220		}
221	}
222
223	handle_syscall(regs);
224}
225
226extern char __syscall_stub_start[];
227
228/**
229 * userspace_tramp() - userspace trampoline
230 * @stack:	pointer to the new userspace stack page, can be NULL, if? FIXME:
231 *
232 * The userspace trampoline is used to setup a new userspace process in start_userspace() after it was clone()'ed.
233 * This function will run on a temporary stack page.
234 * It ptrace()'es itself, then
235 * Two pages are mapped into the userspace address space:
236 * - STUB_CODE (with EXEC), which contains the skas stub code
237 * - STUB_DATA (with R/W), which contains a data page that is used to transfer certain data between the UML userspace process and the UML kernel.
238 * Also for the userspace process a SIGSEGV handler is installed to catch pagefaults in the userspace process.
239 * And last the process stops itself to give control to the UML kernel for this userspace process.
240 *
241 * Return: Always zero, otherwise the current userspace process is ended with non null exit() call
242 */
243static int userspace_tramp(void *stack)
244{
245	void *addr;
246	int fd;
247	unsigned long long offset;
248
249	ptrace(PTRACE_TRACEME, 0, 0, 0);
250
251	signal(SIGTERM, SIG_DFL);
252	signal(SIGWINCH, SIG_IGN);
253
254	fd = phys_mapping(to_phys(__syscall_stub_start), &offset);
255	addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
256		      PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
257	if (addr == MAP_FAILED) {
258		printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
259		       "errno = %d\n", STUB_CODE, errno);
260		exit(1);
261	}
262
263	if (stack != NULL) {
264		fd = phys_mapping(to_phys(stack), &offset);
265		addr = mmap((void *) STUB_DATA,
266			    UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
267			    MAP_FIXED | MAP_SHARED, fd, offset);
 
 
 
 
 
268		if (addr == MAP_FAILED) {
269			printk(UM_KERN_ERR "mapping segfault stack "
270			       "at 0x%lx failed, errno = %d\n",
271			       STUB_DATA, errno);
272			exit(1);
273		}
 
 
 
 
 
 
 
 
 
 
 
 
 
274	}
275	if (stack != NULL) {
276		struct sigaction sa;
277
278		unsigned long v = STUB_CODE +
279				  (unsigned long) stub_segv_handler -
280				  (unsigned long) __syscall_stub_start;
281
282		set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
283		sigemptyset(&sa.sa_mask);
284		sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
285		sa.sa_sigaction = (void *) v;
286		sa.sa_restorer = NULL;
287		if (sigaction(SIGSEGV, &sa, NULL) < 0) {
288			printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
289			       "handler failed - errno = %d\n", errno);
290			exit(1);
291		}
292	}
293
294	kill(os_getpid(), SIGSTOP);
295	return 0;
296}
297
 
 
 
298int userspace_pid[NR_CPUS];
299int kill_userspace_mm[NR_CPUS];
300
301/**
302 * start_userspace() - prepare a new userspace process
303 * @stub_stack:	pointer to the stub stack. Can be NULL, if? FIXME:
304 *
305 * Setups a new temporary stack page that is used while userspace_tramp() runs
306 * Clones the kernel process into a new userspace process, with FDs only.
307 *
308 * Return: When positive: the process id of the new userspace process,
309 *         when negative: an error number.
310 * FIXME: can PIDs become negative?!
311 */
312int start_userspace(unsigned long stub_stack)
313{
314	void *stack;
315	unsigned long sp;
316	int pid, status, n, flags, err;
317
318	/* setup a temporary stack page */
319	stack = mmap(NULL, UM_KERN_PAGE_SIZE,
320		     PROT_READ | PROT_WRITE | PROT_EXEC,
321		     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
322	if (stack == MAP_FAILED) {
323		err = -errno;
324		printk(UM_KERN_ERR "start_userspace : mmap failed, "
325		       "errno = %d\n", errno);
326		return err;
327	}
328
329	/* set stack pointer to the end of the stack page, so it can grow downwards */
330	sp = (unsigned long)stack + UM_KERN_PAGE_SIZE;
331
332	flags = CLONE_FILES | SIGCHLD;
 
 
 
 
333
334	/* clone into new userspace process */
335	pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
336	if (pid < 0) {
337		err = -errno;
338		printk(UM_KERN_ERR "start_userspace : clone failed, "
339		       "errno = %d\n", errno);
340		return err;
341	}
342
343	do {
344		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
345		if (n < 0) {
346			err = -errno;
347			printk(UM_KERN_ERR "start_userspace : wait failed, "
348			       "errno = %d\n", errno);
349			goto out_kill;
350		}
351	} while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
352
353	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
354		err = -EINVAL;
355		printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
356		       "status = %d\n", status);
357		goto out_kill;
358	}
359
360	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
361		   (void *) PTRACE_O_TRACESYSGOOD) < 0) {
362		err = -errno;
363		printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
364		       "failed, errno = %d\n", errno);
365		goto out_kill;
366	}
367
368	if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
369		err = -errno;
370		printk(UM_KERN_ERR "start_userspace : munmap failed, "
371		       "errno = %d\n", errno);
372		goto out_kill;
373	}
374
375	return pid;
376
377 out_kill:
378	os_kill_ptraced_process(pid, 1);
379	return err;
380}
381
382void userspace(struct uml_pt_regs *regs, unsigned long *aux_fp_regs)
383{
 
 
384	int err, status, op, pid = userspace_pid[0];
385	/* To prevent races if using_sysemu changes under us.*/
386	int local_using_sysemu;
387	siginfo_t si;
388
389	/* Handle any immediate reschedules or signals */
390	interrupt_end();
391
392	while (1) {
393		if (kill_userspace_mm[0])
394			fatal_sigsegv();
 
 
395
 
396		/*
397		 * This can legitimately fail if the process loads a
398		 * bogus value into a segment register.  It will
399		 * segfault and PTRACE_GETREGS will read that value
400		 * out of the process.  However, PTRACE_SETREGS will
401		 * fail.  In this case, there is nothing to do but
402		 * just kill the process.
403		 */
404		if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp)) {
405			printk(UM_KERN_ERR "userspace - ptrace set regs "
406			       "failed, errno = %d\n", errno);
407			fatal_sigsegv();
408		}
409
410		if (put_fp_registers(pid, regs->fp)) {
411			printk(UM_KERN_ERR "userspace - ptrace set fp regs "
412			       "failed, errno = %d\n", errno);
413			fatal_sigsegv();
414		}
415
416		/* Now we set local_using_sysemu to be used for one loop */
417		local_using_sysemu = get_using_sysemu();
418
419		op = SELECT_PTRACE_OPERATION(local_using_sysemu,
420					     singlestepping(NULL));
421
422		if (ptrace(op, pid, 0, 0)) {
423			printk(UM_KERN_ERR "userspace - ptrace continue "
424			       "failed, op = %d, errno = %d\n", op, errno);
425			fatal_sigsegv();
426		}
427
428		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
429		if (err < 0) {
430			printk(UM_KERN_ERR "userspace - wait failed, "
431			       "errno = %d\n", errno);
432			fatal_sigsegv();
433		}
434
435		regs->is_user = 1;
436		if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
437			printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
438			       "errno = %d\n", errno);
439			fatal_sigsegv();
440		}
441
442		if (get_fp_registers(pid, regs->fp)) {
443			printk(UM_KERN_ERR "userspace -  get_fp_registers failed, "
444			       "errno = %d\n", errno);
445			fatal_sigsegv();
446		}
447
448		UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
449
450		if (WIFSTOPPED(status)) {
451			int sig = WSTOPSIG(status);
452
453			/* These signal handlers need the si argument.
454			 * The SIGIO and SIGALARM handlers which constitute the
455			 * majority of invocations, do not use it.
456			 */
457			switch (sig) {
458			case SIGSEGV:
459			case SIGTRAP:
460			case SIGILL:
461			case SIGBUS:
462			case SIGFPE:
463			case SIGWINCH:
464				ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
465				break;
466			}
467
468			switch (sig) {
469			case SIGSEGV:
470				if (PTRACE_FULL_FAULTINFO) {
 
471					get_skas_faultinfo(pid,
472							   &regs->faultinfo, aux_fp_regs);
473					(*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
474							     regs);
475				}
476				else handle_segv(pid, regs, aux_fp_regs);
477				break;
478			case SIGTRAP + 0x80:
479			        handle_trap(pid, regs, local_using_sysemu);
480				break;
481			case SIGTRAP:
482				relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
483				break;
484			case SIGALRM:
 
 
 
 
 
 
 
 
 
 
 
485				break;
486			case SIGIO:
487			case SIGILL:
488			case SIGBUS:
489			case SIGFPE:
490			case SIGWINCH:
491				block_signals_trace();
492				(*sig_info[sig])(sig, (struct siginfo *)&si, regs);
493				unblock_signals_trace();
494				break;
495			default:
496				printk(UM_KERN_ERR "userspace - child stopped "
497				       "with signal %d\n", sig);
498				fatal_sigsegv();
499			}
500			pid = userspace_pid[0];
501			interrupt_end();
502
503			/* Avoid -ERESTARTSYS handling in host */
504			if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
505				PT_SYSCALL_NR(regs->gp) = -1;
506		}
507	}
508}
509
510static unsigned long thread_regs[MAX_REG_NR];
511static unsigned long thread_fp_regs[FP_SIZE];
512
513static int __init init_thread_regs(void)
514{
515	get_safe_registers(thread_regs, thread_fp_regs);
516	/* Set parent's instruction pointer to start of clone-stub */
517	thread_regs[REGS_IP_INDEX] = STUB_CODE +
518				(unsigned long) stub_clone_handler -
519				(unsigned long) __syscall_stub_start;
520	thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
521		sizeof(void *);
522#ifdef __SIGNAL_FRAMESIZE
523	thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
524#endif
525	return 0;
526}
527
528__initcall(init_thread_regs);
529
530int copy_context_skas0(unsigned long new_stack, int pid)
531{
 
532	int err;
533	unsigned long current_stack = current_stub_stack();
534	struct stub_data *data = (struct stub_data *) current_stack;
535	struct stub_data *child_data = (struct stub_data *) new_stack;
536	unsigned long long new_offset;
537	int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
538
539	/*
540	 * prepare offset and fd of child's stack as argument for parent's
541	 * and child's mmap2 calls
542	 */
543	*data = ((struct stub_data) {
544		.offset	= MMAP_OFFSET(new_offset),
545		.fd     = new_fd,
546		.parent_err = -ESRCH,
547		.child_err = 0,
548	});
549
550	*child_data = ((struct stub_data) {
551		.child_err = -ESRCH,
552	});
553
554	err = ptrace_setregs(pid, thread_regs);
555	if (err < 0) {
556		err = -errno;
557		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
558		       "failed, pid = %d, errno = %d\n", pid, -err);
559		return err;
560	}
561
562	err = put_fp_registers(pid, thread_fp_regs);
563	if (err < 0) {
564		printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
565		       "failed, pid = %d, err = %d\n", pid, err);
566		return err;
567	}
568
 
 
 
569	/*
570	 * Wait, until parent has finished its work: read child's pid from
571	 * parent's stack, and check, if bad result.
572	 */
573	err = ptrace(PTRACE_CONT, pid, 0, 0);
574	if (err) {
575		err = -errno;
576		printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
577		       "errno = %d\n", pid, errno);
578		return err;
579	}
580
581	wait_stub_done(pid);
582
583	pid = data->parent_err;
584	if (pid < 0) {
585		printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
586		       "error %d\n", -pid);
587		return pid;
588	}
589
590	/*
591	 * Wait, until child has finished too: read child's result from
592	 * child's stack and check it.
593	 */
594	wait_stub_done(pid);
595	if (child_data->child_err != STUB_DATA) {
596		printk(UM_KERN_ERR "copy_context_skas0 - stub-child %d reports "
597		       "error %ld\n", pid, data->child_err);
598		err = data->child_err;
599		goto out_kill;
600	}
601
602	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
603		   (void *)PTRACE_O_TRACESYSGOOD) < 0) {
604		err = -errno;
605		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
606		       "failed, errno = %d\n", errno);
607		goto out_kill;
608	}
609
610	return pid;
611
612 out_kill:
613	os_kill_ptraced_process(pid, 1);
614	return err;
615}
616
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
617void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
618{
619	(*buf)[0].JB_IP = (unsigned long) handler;
620	(*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
621		sizeof(void *);
622}
623
624#define INIT_JMP_NEW_THREAD 0
625#define INIT_JMP_CALLBACK 1
626#define INIT_JMP_HALT 2
627#define INIT_JMP_REBOOT 3
628
629void switch_threads(jmp_buf *me, jmp_buf *you)
630{
631	if (UML_SETJMP(me) == 0)
632		UML_LONGJMP(you, 1);
633}
634
635static jmp_buf initial_jmpbuf;
636
637/* XXX Make these percpu */
638static void (*cb_proc)(void *arg);
639static void *cb_arg;
640static jmp_buf *cb_back;
641
642int start_idle_thread(void *stack, jmp_buf *switch_buf)
643{
644	int n;
645
646	set_handler(SIGWINCH);
647
648	/*
649	 * Can't use UML_SETJMP or UML_LONGJMP here because they save
650	 * and restore signals, with the possible side-effect of
651	 * trying to handle any signals which came when they were
652	 * blocked, which can't be done on this stack.
653	 * Signals must be blocked when jumping back here and restored
654	 * after returning to the jumper.
655	 */
656	n = setjmp(initial_jmpbuf);
657	switch (n) {
658	case INIT_JMP_NEW_THREAD:
659		(*switch_buf)[0].JB_IP = (unsigned long) uml_finishsetup;
660		(*switch_buf)[0].JB_SP = (unsigned long) stack +
661			UM_THREAD_SIZE - sizeof(void *);
662		break;
663	case INIT_JMP_CALLBACK:
664		(*cb_proc)(cb_arg);
665		longjmp(*cb_back, 1);
666		break;
667	case INIT_JMP_HALT:
668		kmalloc_ok = 0;
669		return 0;
670	case INIT_JMP_REBOOT:
671		kmalloc_ok = 0;
672		return 1;
673	default:
674		printk(UM_KERN_ERR "Bad sigsetjmp return in "
675		       "start_idle_thread - %d\n", n);
676		fatal_sigsegv();
677	}
678	longjmp(*switch_buf, 1);
679
680	/* unreachable */
681	printk(UM_KERN_ERR "impossible long jump!");
682	fatal_sigsegv();
683	return 0;
684}
685
686void initial_thread_cb_skas(void (*proc)(void *), void *arg)
687{
688	jmp_buf here;
689
690	cb_proc = proc;
691	cb_arg = arg;
692	cb_back = &here;
693
694	block_signals_trace();
695	if (UML_SETJMP(&here) == 0)
696		UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
697	unblock_signals_trace();
698
699	cb_proc = NULL;
700	cb_arg = NULL;
701	cb_back = NULL;
702}
703
704void halt_skas(void)
705{
706	block_signals_trace();
707	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
708}
709
710void reboot_skas(void)
711{
712	block_signals_trace();
713	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
714}
715
716void __switch_mm(struct mm_id *mm_idp)
717{
718	userspace_pid[0] = mm_idp->u.pid;
719	kill_userspace_mm[0] = mm_idp->kill;
 
 
 
 
 
 
 
 
 
 
 
720}
v3.15
 
  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
 26int is_skas_winch(int pid, int fd, void *data)
 27{
 28	return pid == getpgrp();
 29}
 30
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 31static 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
 55void wait_stub_done(int pid)
 56{
 57	int n, status, err, bad_stop = 0;
 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	else
 78		bad_stop = 1;
 79
 80bad_wait:
 81	err = ptrace_dump_regs(pid);
 82	if (err)
 83		printk(UM_KERN_ERR "Failed to get registers from stub, "
 84		       "errno = %d\n", -err);
 85	printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, "
 86	       "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno,
 87	       status);
 88	if (bad_stop)
 89		kill(pid, SIGKILL);
 90	else
 91		fatal_sigsegv();
 92}
 93
 94extern unsigned long current_stub_stack(void);
 95
 96static void get_skas_faultinfo(int pid, struct faultinfo *fi)
 97{
 98	int err;
 99
100	if (ptrace_faultinfo) {
101		err = ptrace(PTRACE_FAULTINFO, pid, 0, fi);
102		if (err) {
103			printk(UM_KERN_ERR "get_skas_faultinfo - "
104			       "PTRACE_FAULTINFO failed, errno = %d\n", errno);
105			fatal_sigsegv();
106		}
107
108		/* Special handling for i386, which has different structs */
109		if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo))
110			memset((char *)fi + sizeof(struct ptrace_faultinfo), 0,
111			       sizeof(struct faultinfo) -
112			       sizeof(struct ptrace_faultinfo));
113	}
114	else {
115		unsigned long fpregs[FP_SIZE];
116
117		err = get_fp_registers(pid, fpregs);
118		if (err < 0) {
119			printk(UM_KERN_ERR "save_fp_registers returned %d\n",
120			       err);
121			fatal_sigsegv();
122		}
123		err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
124		if (err) {
125			printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
126			       "errno = %d\n", pid, errno);
127			fatal_sigsegv();
128		}
129		wait_stub_done(pid);
130
131		/*
132		 * faultinfo is prepared by the stub-segv-handler at start of
133		 * the stub stack page. We just have to copy it.
134		 */
135		memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
136
137		err = put_fp_registers(pid, fpregs);
138		if (err < 0) {
139			printk(UM_KERN_ERR "put_fp_registers returned %d\n",
140			       err);
141			fatal_sigsegv();
142		}
143	}
144}
145
146static void handle_segv(int pid, struct uml_pt_regs * regs)
147{
148	get_skas_faultinfo(pid, &regs->faultinfo);
149	segv(regs->faultinfo, 0, 1, NULL);
150}
151
152/*
153 * To use the same value of using_sysemu as the caller, ask it that value
154 * (in local_using_sysemu
155 */
156static void handle_trap(int pid, struct uml_pt_regs *regs,
157			int local_using_sysemu)
158{
159	int err, status;
160
161	if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
162		fatal_sigsegv();
163
164	/* Mark this as a syscall */
165	UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->gp);
166
167	if (!local_using_sysemu)
168	{
169		err = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
170			     __NR_getpid);
171		if (err < 0) {
172			printk(UM_KERN_ERR "handle_trap - nullifying syscall "
173			       "failed, errno = %d\n", errno);
174			fatal_sigsegv();
175		}
176
177		err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
178		if (err < 0) {
179			printk(UM_KERN_ERR "handle_trap - continuing to end of "
180			       "syscall failed, errno = %d\n", errno);
181			fatal_sigsegv();
182		}
183
184		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
185		if ((err < 0) || !WIFSTOPPED(status) ||
186		    (WSTOPSIG(status) != SIGTRAP + 0x80)) {
187			err = ptrace_dump_regs(pid);
188			if (err)
189				printk(UM_KERN_ERR "Failed to get registers "
190				       "from process, errno = %d\n", -err);
191			printk(UM_KERN_ERR "handle_trap - failed to wait at "
192			       "end of syscall, errno = %d, status = %d\n",
193			       errno, status);
194			fatal_sigsegv();
195		}
196	}
197
198	handle_syscall(regs);
199}
200
201extern int __syscall_stub_start;
202
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
203static int userspace_tramp(void *stack)
204{
205	void *addr;
206	int err;
 
207
208	ptrace(PTRACE_TRACEME, 0, 0, 0);
209
210	signal(SIGTERM, SIG_DFL);
211	signal(SIGWINCH, SIG_IGN);
212	err = set_interval();
213	if (err) {
214		printk(UM_KERN_ERR "userspace_tramp - setting timer failed, "
215		       "errno = %d\n", err);
 
 
 
216		exit(1);
217	}
218
219	if (!proc_mm) {
220		/*
221		 * This has a pte, but it can't be mapped in with the usual
222		 * tlb_flush mechanism because this is part of that mechanism
223		 */
224		int fd;
225		unsigned long long offset;
226		fd = phys_mapping(to_phys(&__syscall_stub_start), &offset);
227		addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
228			      PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
229		if (addr == MAP_FAILED) {
230			printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
231			       "errno = %d\n", STUB_CODE, errno);
 
232			exit(1);
233		}
234
235		if (stack != NULL) {
236			fd = phys_mapping(to_phys(stack), &offset);
237			addr = mmap((void *) STUB_DATA,
238				    UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
239				    MAP_FIXED | MAP_SHARED, fd, offset);
240			if (addr == MAP_FAILED) {
241				printk(UM_KERN_ERR "mapping segfault stack "
242				       "at 0x%lx failed, errno = %d\n",
243				       STUB_DATA, errno);
244				exit(1);
245			}
246		}
247	}
248	if (!ptrace_faultinfo && (stack != NULL)) {
249		struct sigaction sa;
250
251		unsigned long v = STUB_CODE +
252				  (unsigned long) stub_segv_handler -
253				  (unsigned long) &__syscall_stub_start;
254
255		set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
256		sigemptyset(&sa.sa_mask);
257		sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
258		sa.sa_sigaction = (void *) v;
259		sa.sa_restorer = NULL;
260		if (sigaction(SIGSEGV, &sa, NULL) < 0) {
261			printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
262			       "handler failed - errno = %d\n", errno);
263			exit(1);
264		}
265	}
266
267	kill(os_getpid(), SIGSTOP);
268	return 0;
269}
270
271/* Each element set once, and only accessed by a single processor anyway */
272#undef NR_CPUS
273#define NR_CPUS 1
274int userspace_pid[NR_CPUS];
 
275
 
 
 
 
 
 
 
 
 
 
 
276int start_userspace(unsigned long stub_stack)
277{
278	void *stack;
279	unsigned long sp;
280	int pid, status, n, flags, err;
281
 
282	stack = mmap(NULL, UM_KERN_PAGE_SIZE,
283		     PROT_READ | PROT_WRITE | PROT_EXEC,
284		     MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
285	if (stack == MAP_FAILED) {
286		err = -errno;
287		printk(UM_KERN_ERR "start_userspace : mmap failed, "
288		       "errno = %d\n", errno);
289		return err;
290	}
291
292	sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
 
293
294	flags = CLONE_FILES;
295	if (proc_mm)
296		flags |= CLONE_VM;
297	else
298		flags |= SIGCHLD;
299
 
300	pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
301	if (pid < 0) {
302		err = -errno;
303		printk(UM_KERN_ERR "start_userspace : clone failed, "
304		       "errno = %d\n", errno);
305		return err;
306	}
307
308	do {
309		CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
310		if (n < 0) {
311			err = -errno;
312			printk(UM_KERN_ERR "start_userspace : wait failed, "
313			       "errno = %d\n", errno);
314			goto out_kill;
315		}
316	} while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
317
318	if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
319		err = -EINVAL;
320		printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
321		       "status = %d\n", status);
322		goto out_kill;
323	}
324
325	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
326		   (void *) PTRACE_O_TRACESYSGOOD) < 0) {
327		err = -errno;
328		printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
329		       "failed, errno = %d\n", errno);
330		goto out_kill;
331	}
332
333	if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
334		err = -errno;
335		printk(UM_KERN_ERR "start_userspace : munmap failed, "
336		       "errno = %d\n", errno);
337		goto out_kill;
338	}
339
340	return pid;
341
342 out_kill:
343	os_kill_ptraced_process(pid, 1);
344	return err;
345}
346
347void userspace(struct uml_pt_regs *regs)
348{
349	struct itimerval timer;
350	unsigned long long nsecs, now;
351	int err, status, op, pid = userspace_pid[0];
352	/* To prevent races if using_sysemu changes under us.*/
353	int local_using_sysemu;
354	siginfo_t si;
355
356	/* Handle any immediate reschedules or signals */
357	interrupt_end();
358
359	if (getitimer(ITIMER_VIRTUAL, &timer))
360		printk(UM_KERN_ERR "Failed to get itimer, errno = %d\n", errno);
361	nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC +
362		timer.it_value.tv_usec * UM_NSEC_PER_USEC;
363	nsecs += os_nsecs();
364
365	while (1) {
366		/*
367		 * This can legitimately fail if the process loads a
368		 * bogus value into a segment register.  It will
369		 * segfault and PTRACE_GETREGS will read that value
370		 * out of the process.  However, PTRACE_SETREGS will
371		 * fail.  In this case, there is nothing to do but
372		 * just kill the process.
373		 */
374		if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp))
 
 
375			fatal_sigsegv();
 
376
377		if (put_fp_registers(pid, regs->fp))
 
 
378			fatal_sigsegv();
 
379
380		/* Now we set local_using_sysemu to be used for one loop */
381		local_using_sysemu = get_using_sysemu();
382
383		op = SELECT_PTRACE_OPERATION(local_using_sysemu,
384					     singlestepping(NULL));
385
386		if (ptrace(op, pid, 0, 0)) {
387			printk(UM_KERN_ERR "userspace - ptrace continue "
388			       "failed, op = %d, errno = %d\n", op, errno);
389			fatal_sigsegv();
390		}
391
392		CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
393		if (err < 0) {
394			printk(UM_KERN_ERR "userspace - wait failed, "
395			       "errno = %d\n", errno);
396			fatal_sigsegv();
397		}
398
399		regs->is_user = 1;
400		if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
401			printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
402			       "errno = %d\n", errno);
403			fatal_sigsegv();
404		}
405
406		if (get_fp_registers(pid, regs->fp)) {
407			printk(UM_KERN_ERR "userspace -  get_fp_registers failed, "
408			       "errno = %d\n", errno);
409			fatal_sigsegv();
410		}
411
412		UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
413
414		if (WIFSTOPPED(status)) {
415			int sig = WSTOPSIG(status);
416
417			ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
 
 
 
 
 
 
 
 
 
 
 
 
 
418
419			switch (sig) {
420			case SIGSEGV:
421				if (PTRACE_FULL_FAULTINFO ||
422				    !ptrace_faultinfo) {
423					get_skas_faultinfo(pid,
424							   &regs->faultinfo);
425					(*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
426							     regs);
427				}
428				else handle_segv(pid, regs);
429				break;
430			case SIGTRAP + 0x80:
431			        handle_trap(pid, regs, local_using_sysemu);
432				break;
433			case SIGTRAP:
434				relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
435				break;
436			case SIGVTALRM:
437				now = os_nsecs();
438				if (now < nsecs)
439					break;
440				block_signals();
441				(*sig_info[sig])(sig, (struct siginfo *)&si, regs);
442				unblock_signals();
443				nsecs = timer.it_value.tv_sec *
444					UM_NSEC_PER_SEC +
445					timer.it_value.tv_usec *
446					UM_NSEC_PER_USEC;
447				nsecs += os_nsecs();
448				break;
449			case SIGIO:
450			case SIGILL:
451			case SIGBUS:
452			case SIGFPE:
453			case SIGWINCH:
454				block_signals();
455				(*sig_info[sig])(sig, (struct siginfo *)&si, regs);
456				unblock_signals();
457				break;
458			default:
459				printk(UM_KERN_ERR "userspace - child stopped "
460				       "with signal %d\n", sig);
461				fatal_sigsegv();
462			}
463			pid = userspace_pid[0];
464			interrupt_end();
465
466			/* Avoid -ERESTARTSYS handling in host */
467			if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
468				PT_SYSCALL_NR(regs->gp) = -1;
469		}
470	}
471}
472
473static unsigned long thread_regs[MAX_REG_NR];
474static unsigned long thread_fp_regs[FP_SIZE];
475
476static int __init init_thread_regs(void)
477{
478	get_safe_registers(thread_regs, thread_fp_regs);
479	/* Set parent's instruction pointer to start of clone-stub */
480	thread_regs[REGS_IP_INDEX] = STUB_CODE +
481				(unsigned long) stub_clone_handler -
482				(unsigned long) &__syscall_stub_start;
483	thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
484		sizeof(void *);
485#ifdef __SIGNAL_FRAMESIZE
486	thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
487#endif
488	return 0;
489}
490
491__initcall(init_thread_regs);
492
493int copy_context_skas0(unsigned long new_stack, int pid)
494{
495	struct timeval tv = { .tv_sec = 0, .tv_usec = UM_USEC_PER_SEC / UM_HZ };
496	int err;
497	unsigned long current_stack = current_stub_stack();
498	struct stub_data *data = (struct stub_data *) current_stack;
499	struct stub_data *child_data = (struct stub_data *) new_stack;
500	unsigned long long new_offset;
501	int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
502
503	/*
504	 * prepare offset and fd of child's stack as argument for parent's
505	 * and child's mmap2 calls
506	 */
507	*data = ((struct stub_data) { .offset	= MMAP_OFFSET(new_offset),
508				      .fd	= new_fd,
509				      .timer    = ((struct itimerval)
510					           { .it_value = tv,
511						     .it_interval = tv }) });
 
 
 
 
 
512
513	err = ptrace_setregs(pid, thread_regs);
514	if (err < 0) {
515		err = -errno;
516		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
517		       "failed, pid = %d, errno = %d\n", pid, -err);
518		return err;
519	}
520
521	err = put_fp_registers(pid, thread_fp_regs);
522	if (err < 0) {
523		printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
524		       "failed, pid = %d, err = %d\n", pid, err);
525		return err;
526	}
527
528	/* set a well known return code for detection of child write failure */
529	child_data->err = 12345678;
530
531	/*
532	 * Wait, until parent has finished its work: read child's pid from
533	 * parent's stack, and check, if bad result.
534	 */
535	err = ptrace(PTRACE_CONT, pid, 0, 0);
536	if (err) {
537		err = -errno;
538		printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
539		       "errno = %d\n", pid, errno);
540		return err;
541	}
542
543	wait_stub_done(pid);
544
545	pid = data->err;
546	if (pid < 0) {
547		printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
548		       "error %d\n", -pid);
549		return pid;
550	}
551
552	/*
553	 * Wait, until child has finished too: read child's result from
554	 * child's stack and check it.
555	 */
556	wait_stub_done(pid);
557	if (child_data->err != STUB_DATA) {
558		printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
559		       "error %ld\n", child_data->err);
560		err = child_data->err;
561		goto out_kill;
562	}
563
564	if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
565		   (void *)PTRACE_O_TRACESYSGOOD) < 0) {
566		err = -errno;
567		printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
568		       "failed, errno = %d\n", errno);
569		goto out_kill;
570	}
571
572	return pid;
573
574 out_kill:
575	os_kill_ptraced_process(pid, 1);
576	return err;
577}
578
579/*
580 * This is used only, if stub pages are needed, while proc_mm is
581 * available. Opening /proc/mm creates a new mm_context, which lacks
582 * the stub-pages. Thus, we map them using /proc/mm-fd
583 */
584int map_stub_pages(int fd, unsigned long code, unsigned long data,
585		   unsigned long stack)
586{
587	struct proc_mm_op mmop;
588	int n;
589	unsigned long long code_offset;
590	int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start),
591				   &code_offset);
592
593	mmop = ((struct proc_mm_op) { .op        = MM_MMAP,
594				      .u         =
595				      { .mmap    =
596					{ .addr    = code,
597					  .len     = UM_KERN_PAGE_SIZE,
598					  .prot    = PROT_EXEC,
599					  .flags   = MAP_FIXED | MAP_PRIVATE,
600					  .fd      = code_fd,
601					  .offset  = code_offset
602	} } });
603	CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
604	if (n != sizeof(mmop)) {
605		n = errno;
606		printk(UM_KERN_ERR "mmap args - addr = 0x%lx, fd = %d, "
607		       "offset = %llx\n", code, code_fd,
608		       (unsigned long long) code_offset);
609		printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for code "
610		       "failed, err = %d\n", n);
611		return -n;
612	}
613
614	if (stack) {
615		unsigned long long map_offset;
616		int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
617		mmop = ((struct proc_mm_op)
618				{ .op        = MM_MMAP,
619				  .u         =
620				  { .mmap    =
621				    { .addr    = data,
622				      .len     = UM_KERN_PAGE_SIZE,
623				      .prot    = PROT_READ | PROT_WRITE,
624				      .flags   = MAP_FIXED | MAP_SHARED,
625				      .fd      = map_fd,
626				      .offset  = map_offset
627		} } });
628		CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
629		if (n != sizeof(mmop)) {
630			n = errno;
631			printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for "
632			       "data failed, err = %d\n", n);
633			return -n;
634		}
635	}
636
637	return 0;
638}
639
640void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
641{
642	(*buf)[0].JB_IP = (unsigned long) handler;
643	(*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
644		sizeof(void *);
645}
646
647#define INIT_JMP_NEW_THREAD 0
648#define INIT_JMP_CALLBACK 1
649#define INIT_JMP_HALT 2
650#define INIT_JMP_REBOOT 3
651
652void switch_threads(jmp_buf *me, jmp_buf *you)
653{
654	if (UML_SETJMP(me) == 0)
655		UML_LONGJMP(you, 1);
656}
657
658static jmp_buf initial_jmpbuf;
659
660/* XXX Make these percpu */
661static void (*cb_proc)(void *arg);
662static void *cb_arg;
663static jmp_buf *cb_back;
664
665int start_idle_thread(void *stack, jmp_buf *switch_buf)
666{
667	int n;
668
669	set_handler(SIGWINCH);
670
671	/*
672	 * Can't use UML_SETJMP or UML_LONGJMP here because they save
673	 * and restore signals, with the possible side-effect of
674	 * trying to handle any signals which came when they were
675	 * blocked, which can't be done on this stack.
676	 * Signals must be blocked when jumping back here and restored
677	 * after returning to the jumper.
678	 */
679	n = setjmp(initial_jmpbuf);
680	switch (n) {
681	case INIT_JMP_NEW_THREAD:
682		(*switch_buf)[0].JB_IP = (unsigned long) new_thread_handler;
683		(*switch_buf)[0].JB_SP = (unsigned long) stack +
684			UM_THREAD_SIZE - sizeof(void *);
685		break;
686	case INIT_JMP_CALLBACK:
687		(*cb_proc)(cb_arg);
688		longjmp(*cb_back, 1);
689		break;
690	case INIT_JMP_HALT:
691		kmalloc_ok = 0;
692		return 0;
693	case INIT_JMP_REBOOT:
694		kmalloc_ok = 0;
695		return 1;
696	default:
697		printk(UM_KERN_ERR "Bad sigsetjmp return in "
698		       "start_idle_thread - %d\n", n);
699		fatal_sigsegv();
700	}
701	longjmp(*switch_buf, 1);
 
 
 
 
 
702}
703
704void initial_thread_cb_skas(void (*proc)(void *), void *arg)
705{
706	jmp_buf here;
707
708	cb_proc = proc;
709	cb_arg = arg;
710	cb_back = &here;
711
712	block_signals();
713	if (UML_SETJMP(&here) == 0)
714		UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
715	unblock_signals();
716
717	cb_proc = NULL;
718	cb_arg = NULL;
719	cb_back = NULL;
720}
721
722void halt_skas(void)
723{
724	block_signals();
725	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
726}
727
728void reboot_skas(void)
729{
730	block_signals();
731	UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
732}
733
734void __switch_mm(struct mm_id *mm_idp)
735{
736	int err;
737
738	/* FIXME: need cpu pid in __switch_mm */
739	if (proc_mm) {
740		err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
741			     mm_idp->u.mm_fd);
742		if (err) {
743			printk(UM_KERN_ERR "__switch_mm - PTRACE_SWITCH_MM "
744			       "failed, errno = %d\n", errno);
745			fatal_sigsegv();
746		}
747	}
748	else userspace_pid[0] = mm_idp->u.pid;
749}