Loading...
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, ®s->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 ®s->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}
1/*
2 * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de)
3 * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
4 * Licensed under the GPL
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
25int is_skas_winch(int pid, int fd, void *data)
26{
27 return pid == getpgrp();
28}
29
30static int ptrace_dump_regs(int pid)
31{
32 unsigned long regs[MAX_REG_NR];
33 int i;
34
35 if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
36 return -errno;
37
38 printk(UM_KERN_ERR "Stub registers -\n");
39 for (i = 0; i < ARRAY_SIZE(regs); i++)
40 printk(UM_KERN_ERR "\t%d - %lx\n", i, regs[i]);
41
42 return 0;
43}
44
45/*
46 * Signals that are OK to receive in the stub - we'll just continue it.
47 * SIGWINCH will happen when UML is inside a detached screen.
48 */
49#define STUB_SIG_MASK ((1 << SIGALRM) | (1 << SIGWINCH))
50
51/* Signals that the stub will finish with - anything else is an error */
52#define STUB_DONE_MASK (1 << SIGTRAP)
53
54void wait_stub_done(int pid)
55{
56 int n, status, err;
57
58 while (1) {
59 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
60 if ((n < 0) || !WIFSTOPPED(status))
61 goto bad_wait;
62
63 if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
64 break;
65
66 err = ptrace(PTRACE_CONT, pid, 0, 0);
67 if (err) {
68 printk(UM_KERN_ERR "wait_stub_done : continue failed, "
69 "errno = %d\n", errno);
70 fatal_sigsegv();
71 }
72 }
73
74 if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
75 return;
76
77bad_wait:
78 err = ptrace_dump_regs(pid);
79 if (err)
80 printk(UM_KERN_ERR "Failed to get registers from stub, "
81 "errno = %d\n", -err);
82 printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, "
83 "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno,
84 status);
85 fatal_sigsegv();
86}
87
88extern unsigned long current_stub_stack(void);
89
90static void get_skas_faultinfo(int pid, struct faultinfo *fi)
91{
92 int err;
93 unsigned long fpregs[FP_SIZE];
94
95 err = get_fp_registers(pid, fpregs);
96 if (err < 0) {
97 printk(UM_KERN_ERR "save_fp_registers returned %d\n",
98 err);
99 fatal_sigsegv();
100 }
101 err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
102 if (err) {
103 printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
104 "errno = %d\n", pid, errno);
105 fatal_sigsegv();
106 }
107 wait_stub_done(pid);
108
109 /*
110 * faultinfo is prepared by the stub-segv-handler at start of
111 * the stub stack page. We just have to copy it.
112 */
113 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
114
115 err = put_fp_registers(pid, fpregs);
116 if (err < 0) {
117 printk(UM_KERN_ERR "put_fp_registers returned %d\n",
118 err);
119 fatal_sigsegv();
120 }
121}
122
123static void handle_segv(int pid, struct uml_pt_regs * regs)
124{
125 get_skas_faultinfo(pid, ®s->faultinfo);
126 segv(regs->faultinfo, 0, 1, NULL);
127}
128
129/*
130 * To use the same value of using_sysemu as the caller, ask it that value
131 * (in local_using_sysemu
132 */
133static void handle_trap(int pid, struct uml_pt_regs *regs,
134 int local_using_sysemu)
135{
136 int err, status;
137
138 if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
139 fatal_sigsegv();
140
141 if (!local_using_sysemu)
142 {
143 err = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
144 __NR_getpid);
145 if (err < 0) {
146 printk(UM_KERN_ERR "handle_trap - nullifying syscall "
147 "failed, errno = %d\n", errno);
148 fatal_sigsegv();
149 }
150
151 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
152 if (err < 0) {
153 printk(UM_KERN_ERR "handle_trap - continuing to end of "
154 "syscall failed, errno = %d\n", errno);
155 fatal_sigsegv();
156 }
157
158 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
159 if ((err < 0) || !WIFSTOPPED(status) ||
160 (WSTOPSIG(status) != SIGTRAP + 0x80)) {
161 err = ptrace_dump_regs(pid);
162 if (err)
163 printk(UM_KERN_ERR "Failed to get registers "
164 "from process, errno = %d\n", -err);
165 printk(UM_KERN_ERR "handle_trap - failed to wait at "
166 "end of syscall, errno = %d, status = %d\n",
167 errno, status);
168 fatal_sigsegv();
169 }
170 }
171
172 handle_syscall(regs);
173}
174
175extern char __syscall_stub_start[];
176
177static int userspace_tramp(void *stack)
178{
179 void *addr;
180 int fd;
181 unsigned long long offset;
182
183 ptrace(PTRACE_TRACEME, 0, 0, 0);
184
185 signal(SIGTERM, SIG_DFL);
186 signal(SIGWINCH, SIG_IGN);
187
188 /*
189 * This has a pte, but it can't be mapped in with the usual
190 * tlb_flush mechanism because this is part of that mechanism
191 */
192 fd = phys_mapping(to_phys(__syscall_stub_start), &offset);
193 addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
194 PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
195 if (addr == MAP_FAILED) {
196 printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
197 "errno = %d\n", STUB_CODE, errno);
198 exit(1);
199 }
200
201 if (stack != NULL) {
202 fd = phys_mapping(to_phys(stack), &offset);
203 addr = mmap((void *) STUB_DATA,
204 UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
205 MAP_FIXED | MAP_SHARED, fd, offset);
206 if (addr == MAP_FAILED) {
207 printk(UM_KERN_ERR "mapping segfault stack "
208 "at 0x%lx failed, errno = %d\n",
209 STUB_DATA, errno);
210 exit(1);
211 }
212 }
213 if (stack != NULL) {
214 struct sigaction sa;
215
216 unsigned long v = STUB_CODE +
217 (unsigned long) stub_segv_handler -
218 (unsigned long) __syscall_stub_start;
219
220 set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
221 sigemptyset(&sa.sa_mask);
222 sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
223 sa.sa_sigaction = (void *) v;
224 sa.sa_restorer = NULL;
225 if (sigaction(SIGSEGV, &sa, NULL) < 0) {
226 printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
227 "handler failed - errno = %d\n", errno);
228 exit(1);
229 }
230 }
231
232 kill(os_getpid(), SIGSTOP);
233 return 0;
234}
235
236/* Each element set once, and only accessed by a single processor anyway */
237#undef NR_CPUS
238#define NR_CPUS 1
239int userspace_pid[NR_CPUS];
240
241int start_userspace(unsigned long stub_stack)
242{
243 void *stack;
244 unsigned long sp;
245 int pid, status, n, flags, err;
246
247 stack = mmap(NULL, UM_KERN_PAGE_SIZE,
248 PROT_READ | PROT_WRITE | PROT_EXEC,
249 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
250 if (stack == MAP_FAILED) {
251 err = -errno;
252 printk(UM_KERN_ERR "start_userspace : mmap failed, "
253 "errno = %d\n", errno);
254 return err;
255 }
256
257 sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
258
259 flags = CLONE_FILES | SIGCHLD;
260
261 pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
262 if (pid < 0) {
263 err = -errno;
264 printk(UM_KERN_ERR "start_userspace : clone failed, "
265 "errno = %d\n", errno);
266 return err;
267 }
268
269 do {
270 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
271 if (n < 0) {
272 err = -errno;
273 printk(UM_KERN_ERR "start_userspace : wait failed, "
274 "errno = %d\n", errno);
275 goto out_kill;
276 }
277 } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
278
279 if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
280 err = -EINVAL;
281 printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
282 "status = %d\n", status);
283 goto out_kill;
284 }
285
286 if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
287 (void *) PTRACE_O_TRACESYSGOOD) < 0) {
288 err = -errno;
289 printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
290 "failed, errno = %d\n", errno);
291 goto out_kill;
292 }
293
294 if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
295 err = -errno;
296 printk(UM_KERN_ERR "start_userspace : munmap failed, "
297 "errno = %d\n", errno);
298 goto out_kill;
299 }
300
301 return pid;
302
303 out_kill:
304 os_kill_ptraced_process(pid, 1);
305 return err;
306}
307
308void userspace(struct uml_pt_regs *regs)
309{
310 int err, status, op, pid = userspace_pid[0];
311 /* To prevent races if using_sysemu changes under us.*/
312 int local_using_sysemu;
313 siginfo_t si;
314
315 /* Handle any immediate reschedules or signals */
316 interrupt_end();
317
318 while (1) {
319
320 /*
321 * This can legitimately fail if the process loads a
322 * bogus value into a segment register. It will
323 * segfault and PTRACE_GETREGS will read that value
324 * out of the process. However, PTRACE_SETREGS will
325 * fail. In this case, there is nothing to do but
326 * just kill the process.
327 */
328 if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp))
329 fatal_sigsegv();
330
331 if (put_fp_registers(pid, regs->fp))
332 fatal_sigsegv();
333
334 /* Now we set local_using_sysemu to be used for one loop */
335 local_using_sysemu = get_using_sysemu();
336
337 op = SELECT_PTRACE_OPERATION(local_using_sysemu,
338 singlestepping(NULL));
339
340 if (ptrace(op, pid, 0, 0)) {
341 printk(UM_KERN_ERR "userspace - ptrace continue "
342 "failed, op = %d, errno = %d\n", op, errno);
343 fatal_sigsegv();
344 }
345
346 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
347 if (err < 0) {
348 printk(UM_KERN_ERR "userspace - wait failed, "
349 "errno = %d\n", errno);
350 fatal_sigsegv();
351 }
352
353 regs->is_user = 1;
354 if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
355 printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
356 "errno = %d\n", errno);
357 fatal_sigsegv();
358 }
359
360 if (get_fp_registers(pid, regs->fp)) {
361 printk(UM_KERN_ERR "userspace - get_fp_registers failed, "
362 "errno = %d\n", errno);
363 fatal_sigsegv();
364 }
365
366 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
367
368 if (WIFSTOPPED(status)) {
369 int sig = WSTOPSIG(status);
370
371 ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
372
373 switch (sig) {
374 case SIGSEGV:
375 if (PTRACE_FULL_FAULTINFO) {
376 get_skas_faultinfo(pid,
377 ®s->faultinfo);
378 (*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
379 regs);
380 }
381 else handle_segv(pid, regs);
382 break;
383 case SIGTRAP + 0x80:
384 handle_trap(pid, regs, local_using_sysemu);
385 break;
386 case SIGTRAP:
387 relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
388 break;
389 case SIGALRM:
390 break;
391 case SIGIO:
392 case SIGILL:
393 case SIGBUS:
394 case SIGFPE:
395 case SIGWINCH:
396 block_signals();
397 (*sig_info[sig])(sig, (struct siginfo *)&si, regs);
398 unblock_signals();
399 break;
400 default:
401 printk(UM_KERN_ERR "userspace - child stopped "
402 "with signal %d\n", sig);
403 fatal_sigsegv();
404 }
405 pid = userspace_pid[0];
406 interrupt_end();
407
408 /* Avoid -ERESTARTSYS handling in host */
409 if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
410 PT_SYSCALL_NR(regs->gp) = -1;
411 }
412 }
413}
414
415static unsigned long thread_regs[MAX_REG_NR];
416static unsigned long thread_fp_regs[FP_SIZE];
417
418static int __init init_thread_regs(void)
419{
420 get_safe_registers(thread_regs, thread_fp_regs);
421 /* Set parent's instruction pointer to start of clone-stub */
422 thread_regs[REGS_IP_INDEX] = STUB_CODE +
423 (unsigned long) stub_clone_handler -
424 (unsigned long) __syscall_stub_start;
425 thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
426 sizeof(void *);
427#ifdef __SIGNAL_FRAMESIZE
428 thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
429#endif
430 return 0;
431}
432
433__initcall(init_thread_regs);
434
435int copy_context_skas0(unsigned long new_stack, int pid)
436{
437 int err;
438 unsigned long current_stack = current_stub_stack();
439 struct stub_data *data = (struct stub_data *) current_stack;
440 struct stub_data *child_data = (struct stub_data *) new_stack;
441 unsigned long long new_offset;
442 int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
443
444 /*
445 * prepare offset and fd of child's stack as argument for parent's
446 * and child's mmap2 calls
447 */
448 *data = ((struct stub_data) {
449 .offset = MMAP_OFFSET(new_offset),
450 .fd = new_fd
451 });
452
453 err = ptrace_setregs(pid, thread_regs);
454 if (err < 0) {
455 err = -errno;
456 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
457 "failed, pid = %d, errno = %d\n", pid, -err);
458 return err;
459 }
460
461 err = put_fp_registers(pid, thread_fp_regs);
462 if (err < 0) {
463 printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
464 "failed, pid = %d, err = %d\n", pid, err);
465 return err;
466 }
467
468 /* set a well known return code for detection of child write failure */
469 child_data->err = 12345678;
470
471 /*
472 * Wait, until parent has finished its work: read child's pid from
473 * parent's stack, and check, if bad result.
474 */
475 err = ptrace(PTRACE_CONT, pid, 0, 0);
476 if (err) {
477 err = -errno;
478 printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
479 "errno = %d\n", pid, errno);
480 return err;
481 }
482
483 wait_stub_done(pid);
484
485 pid = data->err;
486 if (pid < 0) {
487 printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
488 "error %d\n", -pid);
489 return pid;
490 }
491
492 /*
493 * Wait, until child has finished too: read child's result from
494 * child's stack and check it.
495 */
496 wait_stub_done(pid);
497 if (child_data->err != STUB_DATA) {
498 printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
499 "error %ld\n", child_data->err);
500 err = child_data->err;
501 goto out_kill;
502 }
503
504 if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
505 (void *)PTRACE_O_TRACESYSGOOD) < 0) {
506 err = -errno;
507 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
508 "failed, errno = %d\n", errno);
509 goto out_kill;
510 }
511
512 return pid;
513
514 out_kill:
515 os_kill_ptraced_process(pid, 1);
516 return err;
517}
518
519void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
520{
521 (*buf)[0].JB_IP = (unsigned long) handler;
522 (*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
523 sizeof(void *);
524}
525
526#define INIT_JMP_NEW_THREAD 0
527#define INIT_JMP_CALLBACK 1
528#define INIT_JMP_HALT 2
529#define INIT_JMP_REBOOT 3
530
531void switch_threads(jmp_buf *me, jmp_buf *you)
532{
533 if (UML_SETJMP(me) == 0)
534 UML_LONGJMP(you, 1);
535}
536
537static jmp_buf initial_jmpbuf;
538
539/* XXX Make these percpu */
540static void (*cb_proc)(void *arg);
541static void *cb_arg;
542static jmp_buf *cb_back;
543
544int start_idle_thread(void *stack, jmp_buf *switch_buf)
545{
546 int n;
547
548 set_handler(SIGWINCH);
549
550 /*
551 * Can't use UML_SETJMP or UML_LONGJMP here because they save
552 * and restore signals, with the possible side-effect of
553 * trying to handle any signals which came when they were
554 * blocked, which can't be done on this stack.
555 * Signals must be blocked when jumping back here and restored
556 * after returning to the jumper.
557 */
558 n = setjmp(initial_jmpbuf);
559 switch (n) {
560 case INIT_JMP_NEW_THREAD:
561 (*switch_buf)[0].JB_IP = (unsigned long) uml_finishsetup;
562 (*switch_buf)[0].JB_SP = (unsigned long) stack +
563 UM_THREAD_SIZE - sizeof(void *);
564 break;
565 case INIT_JMP_CALLBACK:
566 (*cb_proc)(cb_arg);
567 longjmp(*cb_back, 1);
568 break;
569 case INIT_JMP_HALT:
570 kmalloc_ok = 0;
571 return 0;
572 case INIT_JMP_REBOOT:
573 kmalloc_ok = 0;
574 return 1;
575 default:
576 printk(UM_KERN_ERR "Bad sigsetjmp return in "
577 "start_idle_thread - %d\n", n);
578 fatal_sigsegv();
579 }
580 longjmp(*switch_buf, 1);
581}
582
583void initial_thread_cb_skas(void (*proc)(void *), void *arg)
584{
585 jmp_buf here;
586
587 cb_proc = proc;
588 cb_arg = arg;
589 cb_back = &here;
590
591 block_signals();
592 if (UML_SETJMP(&here) == 0)
593 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
594 unblock_signals();
595
596 cb_proc = NULL;
597 cb_arg = NULL;
598 cb_back = NULL;
599}
600
601void halt_skas(void)
602{
603 block_signals();
604 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
605}
606
607void reboot_skas(void)
608{
609 block_signals();
610 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
611}
612
613void __switch_mm(struct mm_id *mm_idp)
614{
615 userspace_pid[0] = mm_idp->u.pid;
616}