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