Loading...
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}
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, ®s->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 ®s->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}