projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
uml: use ptrace directly in libc code
[deliverable/linux.git] / arch / um / os-Linux / skas / process.c
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/ptrace.h>
13 #include <sys/wait.h>
14 #include <asm/unistd.h>
15 #include "as-layout.h"
16 #include "chan_user.h"
17 #include "kern_constants.h"
18 #include "kern_util.h"
19 #include "mem.h"
20 #include "os.h"
21 #include "process.h"
22 #include "proc_mm.h"
23 #include "ptrace_user.h"
24 #include "registers.h"
25 #include "skas.h"
26 #include "skas_ptrace.h"
27 #include "user.h"
28 #include "sysdep/stub.h"
29
30 int is_skas_winch(int pid, int fd, void *data)
31 {
32 if (pid != getpgrp())
33 return 0;
34
35 register_winch_irq(-1, fd, -1, data, 0);
36 return 1;
37 }
38
39 static int ptrace_dump_regs(int pid)
40 {
41 unsigned long regs[MAX_REG_NR];
42 int i;
43
44 if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
45 return -errno;
46
47 printk(UM_KERN_ERR "Stub registers -\n");
48 for (i = 0; i < ARRAY_SIZE(regs); i++)
49 printk(UM_KERN_ERR "\t%d - %lx\n", i, regs[i]);
50
51 return 0;
52 }
53
54 /*
55 * Signals that are OK to receive in the stub - we'll just continue it.
56 * SIGWINCH will happen when UML is inside a detached screen.
57 */
58 #define STUB_SIG_MASK (1 << SIGVTALRM)
59
60 /* Signals that the stub will finish with - anything else is an error */
61 #define STUB_DONE_MASK (1 << SIGTRAP)
62
63 void wait_stub_done(int pid)
64 {
65 int n, status, err;
66
67 while (1) {
68 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
69 if ((n < 0) || !WIFSTOPPED(status))
70 goto bad_wait;
71
72 if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
73 break;
74
75 err = ptrace(PTRACE_CONT, pid, 0, 0);
76 if (err)
77 panic("wait_stub_done : continue failed, errno = %d\n",
78 errno);
79 }
80
81 if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
82 return;
83
84 bad_wait:
85 err = ptrace_dump_regs(pid);
86 if (err)
87 printk(UM_KERN_ERR "Failed to get registers from stub, "
88 "errno = %d\n", -err);
89 panic("wait_stub_done : failed to wait for SIGUSR1/SIGTRAP, pid = %d, "
90 "n = %d, errno = %d, status = 0x%x\n", pid, n, errno, status);
91 }
92
93 extern unsigned long current_stub_stack(void);
94
95 void get_skas_faultinfo(int pid, struct faultinfo * fi)
96 {
97 int err;
98
99 if (ptrace_faultinfo) {
100 err = ptrace(PTRACE_FAULTINFO, pid, 0, fi);
101 if (err)
102 panic("get_skas_faultinfo - PTRACE_FAULTINFO failed, "
103 "errno = %d\n", errno);
104
105 /* Special handling for i386, which has different structs */
106 if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo))
107 memset((char *)fi + sizeof(struct ptrace_faultinfo), 0,
108 sizeof(struct faultinfo) -
109 sizeof(struct ptrace_faultinfo));
110 }
111 else {
112 err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
113 if (err)
114 panic("Failed to continue stub, pid = %d, errno = %d\n",
115 pid, errno);
116 wait_stub_done(pid);
117
118 /*
119 * faultinfo is prepared by the stub-segv-handler at start of
120 * the stub stack page. We just have to copy it.
121 */
122 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
123 }
124 }
125
126 static void handle_segv(int pid, struct uml_pt_regs * regs)
127 {
128 get_skas_faultinfo(pid, &regs->faultinfo);
129 segv(regs->faultinfo, 0, 1, NULL);
130 }
131
132 /*
133 * To use the same value of using_sysemu as the caller, ask it that value
134 * (in local_using_sysemu
135 */
136 static void handle_trap(int pid, struct uml_pt_regs *regs,
137 int local_using_sysemu)
138 {
139 int err, status;
140
141 /* Mark this as a syscall */
142 UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->gp);
143
144 if (!local_using_sysemu)
145 {
146 err = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
147 __NR_getpid);
148 if (err < 0)
149 panic("handle_trap - nullifying syscall failed, "
150 "errno = %d\n", errno);
151
152 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
153 if (err < 0)
154 panic("handle_trap - continuing to end of syscall "
155 "failed, errno = %d\n", errno);
156
157 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
158 if ((err < 0) || !WIFSTOPPED(status) ||
159 (WSTOPSIG(status) != SIGTRAP + 0x80)) {
160 err = ptrace_dump_regs(pid);
161 if (err)
162 printk(UM_KERN_ERR "Failed to get registers "
163 "from process, errno = %d\n", -err);
164 panic("handle_trap - failed to wait at end of syscall, "
165 "errno = %d, status = %d\n", errno, status);
166 }
167 }
168
169 handle_syscall(regs);
170 }
171
172 extern int __syscall_stub_start;
173
174 static int userspace_tramp(void *stack)
175 {
176 void *addr;
177 int err;
178
179 ptrace(PTRACE_TRACEME, 0, 0, 0);
180
181 signal(SIGTERM, SIG_DFL);
182 signal(SIGWINCH, SIG_IGN);
183 err = set_interval();
184 if (err)
185 panic("userspace_tramp - setting timer failed, errno = %d\n",
186 err);
187
188 if (!proc_mm) {
189 /*
190 * This has a pte, but it can't be mapped in with the usual
191 * tlb_flush mechanism because this is part of that mechanism
192 */
193 int fd;
194 unsigned long long offset;
195 fd = phys_mapping(to_phys(&__syscall_stub_start), &offset);
196 addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
197 PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
198 if (addr == MAP_FAILED) {
199 printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
200 "errno = %d\n", STUB_CODE, errno);
201 exit(1);
202 }
203
204 if (stack != NULL) {
205 fd = phys_mapping(to_phys(stack), &offset);
206 addr = mmap((void *) STUB_DATA,
207 UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
208 MAP_FIXED | MAP_SHARED, fd, offset);
209 if (addr == MAP_FAILED) {
210 printk(UM_KERN_ERR "mapping segfault stack "
211 "at 0x%lx failed, errno = %d\n",
212 STUB_DATA, errno);
213 exit(1);
214 }
215 }
216 }
217 if (!ptrace_faultinfo && (stack != NULL)) {
218 struct sigaction sa;
219
220 unsigned long v = STUB_CODE +
221 (unsigned long) stub_segv_handler -
222 (unsigned long) &__syscall_stub_start;
223
224 set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
225 sigemptyset(&sa.sa_mask);
226 sa.sa_flags = SA_ONSTACK | SA_NODEFER;
227 sa.sa_handler = (void *) v;
228 sa.sa_restorer = NULL;
229 if (sigaction(SIGSEGV, &sa, NULL) < 0)
230 panic("userspace_tramp - setting SIGSEGV handler "
231 "failed - errno = %d\n", errno);
232 }
233
234 kill(os_getpid(), SIGSTOP);
235 return 0;
236 }
237
238 /* Each element set once, and only accessed by a single processor anyway */
239 #undef NR_CPUS
240 #define NR_CPUS 1
241 int userspace_pid[NR_CPUS];
242
243 int start_userspace(unsigned long stub_stack)
244 {
245 void *stack;
246 unsigned long sp;
247 int pid, status, n, flags;
248
249 stack = mmap(NULL, UM_KERN_PAGE_SIZE,
250 PROT_READ | PROT_WRITE | PROT_EXEC,
251 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
252 if (stack == MAP_FAILED)
253 panic("start_userspace : mmap failed, errno = %d", errno);
254 sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
255
256 flags = CLONE_FILES;
257 if (proc_mm)
258 flags |= CLONE_VM;
259 else
260 flags |= SIGCHLD;
261
262 pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
263 if (pid < 0)
264 panic("start_userspace : clone failed, errno = %d", errno);
265
266 do {
267 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
268 if (n < 0)
269 panic("start_userspace : wait failed, errno = %d",
270 errno);
271 } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
272
273 if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
274 panic("start_userspace : expected SIGSTOP, got status = %d",
275 status);
276
277 if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
278 (void *) PTRACE_O_TRACESYSGOOD) < 0)
279 panic("start_userspace : PTRACE_OLDSETOPTIONS failed, "
280 "errno = %d\n", errno);
281
282 if (munmap(stack, UM_KERN_PAGE_SIZE) < 0)
283 panic("start_userspace : munmap failed, errno = %d\n", errno);
284
285 return pid;
286 }
287
288 void userspace(struct uml_pt_regs *regs)
289 {
290 struct itimerval timer;
291 unsigned long long nsecs, now;
292 int err, status, op, pid = userspace_pid[0];
293 /* To prevent races if using_sysemu changes under us.*/
294 int local_using_sysemu;
295
296 if (getitimer(ITIMER_VIRTUAL, &timer))
297 printk("Failed to get itimer, errno = %d\n", errno);
298 nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC +
299 timer.it_value.tv_usec * UM_NSEC_PER_USEC;
300 nsecs += os_nsecs();
301
302 while (1) {
303 if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp))
304 panic("userspace - PTRACE_SETREGS failed, "
305 "errno = %d\n", errno);
306
307 /* Now we set local_using_sysemu to be used for one loop */
308 local_using_sysemu = get_using_sysemu();
309
310 op = SELECT_PTRACE_OPERATION(local_using_sysemu,
311 singlestepping(NULL));
312
313 err = ptrace(op, pid, 0, 0);
314 if (err)
315 panic("userspace - could not resume userspace process, "
316 "pid=%d, ptrace operation = %d, errno = %d\n",
317 pid, op, errno);
318
319 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
320 if (err < 0)
321 panic("userspace - waitpid failed, errno = %d\n",
322 errno);
323
324 regs->is_user = 1;
325 if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp))
326 panic("userspace - saving registers failed, "
327 "errno = %d\n", errno);
328
329 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
330
331 if (WIFSTOPPED(status)) {
332 int sig = WSTOPSIG(status);
333 switch(sig) {
334 case SIGSEGV:
335 if (PTRACE_FULL_FAULTINFO ||
336 !ptrace_faultinfo) {
337 get_skas_faultinfo(pid,
338 &regs->faultinfo);
339 (*sig_info[SIGSEGV])(SIGSEGV, regs);
340 }
341 else handle_segv(pid, regs);
342 break;
343 case SIGTRAP + 0x80:
344 handle_trap(pid, regs, local_using_sysemu);
345 break;
346 case SIGTRAP:
347 relay_signal(SIGTRAP, regs);
348 break;
349 case SIGVTALRM:
350 now = os_nsecs();
351 if (now < nsecs)
352 break;
353 block_signals();
354 (*sig_info[sig])(sig, regs);
355 unblock_signals();
356 nsecs = timer.it_value.tv_sec *
357 UM_NSEC_PER_SEC +
358 timer.it_value.tv_usec *
359 UM_NSEC_PER_USEC;
360 nsecs += os_nsecs();
361 break;
362 case SIGIO:
363 case SIGILL:
364 case SIGBUS:
365 case SIGFPE:
366 case SIGWINCH:
367 block_signals();
368 (*sig_info[sig])(sig, regs);
369 unblock_signals();
370 break;
371 default:
372 printk(UM_KERN_ERR "userspace - child stopped "
373 "with signal %d\n", sig);
374 }
375 pid = userspace_pid[0];
376 interrupt_end();
377
378 /* Avoid -ERESTARTSYS handling in host */
379 if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
380 PT_SYSCALL_NR(regs->gp) = -1;
381 }
382 }
383 }
384
385 static unsigned long thread_regs[MAX_REG_NR];
386
387 static int __init init_thread_regs(void)
388 {
389 get_safe_registers(thread_regs);
390 /* Set parent's instruction pointer to start of clone-stub */
391 thread_regs[REGS_IP_INDEX] = STUB_CODE +
392 (unsigned long) stub_clone_handler -
393 (unsigned long) &__syscall_stub_start;
394 thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
395 sizeof(void *);
396 #ifdef __SIGNAL_FRAMESIZE
397 thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
398 #endif
399 return 0;
400 }
401
402 __initcall(init_thread_regs);
403
404 int copy_context_skas0(unsigned long new_stack, int pid)
405 {
406 struct timeval tv = { .tv_sec = 0, .tv_usec = UM_USEC_PER_SEC / UM_HZ };
407 int err;
408 unsigned long current_stack = current_stub_stack();
409 struct stub_data *data = (struct stub_data *) current_stack;
410 struct stub_data *child_data = (struct stub_data *) new_stack;
411 unsigned long long new_offset;
412 int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
413
414 /*
415 * prepare offset and fd of child's stack as argument for parent's
416 * and child's mmap2 calls
417 */
418 *data = ((struct stub_data) { .offset = MMAP_OFFSET(new_offset),
419 .fd = new_fd,
420 .timer = ((struct itimerval)
421 { .it_value = tv,
422 .it_interval = tv }) });
423
424 err = ptrace_setregs(pid, thread_regs);
425 if (err < 0)
426 panic("copy_context_skas0 : PTRACE_SETREGS failed, "
427 "pid = %d, errno = %d\n", pid, -err);
428
429 /* set a well known return code for detection of child write failure */
430 child_data->err = 12345678;
431
432 /*
433 * Wait, until parent has finished its work: read child's pid from
434 * parent's stack, and check, if bad result.
435 */
436 err = ptrace(PTRACE_CONT, pid, 0, 0);
437 if (err)
438 panic("Failed to continue new process, pid = %d, "
439 "errno = %d\n", pid, errno);
440 wait_stub_done(pid);
441
442 pid = data->err;
443 if (pid < 0)
444 panic("copy_context_skas0 - stub-parent reports error %d\n",
445 -pid);
446
447 /*
448 * Wait, until child has finished too: read child's result from
449 * child's stack and check it.
450 */
451 wait_stub_done(pid);
452 if (child_data->err != STUB_DATA)
453 panic("copy_context_skas0 - stub-child reports error %ld\n",
454 child_data->err);
455
456 if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
457 (void *)PTRACE_O_TRACESYSGOOD) < 0)
458 panic("copy_context_skas0 : PTRACE_OLDSETOPTIONS failed, "
459 "errno = %d\n", errno);
460
461 return pid;
462 }
463
464 /*
465 * This is used only, if stub pages are needed, while proc_mm is
466 * available. Opening /proc/mm creates a new mm_context, which lacks
467 * the stub-pages. Thus, we map them using /proc/mm-fd
468 */
469 void map_stub_pages(int fd, unsigned long code,
470 unsigned long data, unsigned long stack)
471 {
472 struct proc_mm_op mmop;
473 int n;
474 unsigned long long code_offset;
475 int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start),
476 &code_offset);
477
478 mmop = ((struct proc_mm_op) { .op = MM_MMAP,
479 .u =
480 { .mmap =
481 { .addr = code,
482 .len = UM_KERN_PAGE_SIZE,
483 .prot = PROT_EXEC,
484 .flags = MAP_FIXED | MAP_PRIVATE,
485 .fd = code_fd,
486 .offset = code_offset
487 } } });
488 CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
489 if (n != sizeof(mmop)) {
490 n = errno;
491 printk(UM_KERN_ERR "mmap args - addr = 0x%lx, fd = %d, "
492 "offset = %llx\n", code, code_fd,
493 (unsigned long long) code_offset);
494 panic("map_stub_pages : /proc/mm map for code failed, "
495 "err = %d\n", n);
496 }
497
498 if (stack) {
499 unsigned long long map_offset;
500 int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
501 mmop = ((struct proc_mm_op)
502 { .op = MM_MMAP,
503 .u =
504 { .mmap =
505 { .addr = data,
506 .len = UM_KERN_PAGE_SIZE,
507 .prot = PROT_READ | PROT_WRITE,
508 .flags = MAP_FIXED | MAP_SHARED,
509 .fd = map_fd,
510 .offset = map_offset
511 } } });
512 CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
513 if (n != sizeof(mmop))
514 panic("map_stub_pages : /proc/mm map for data failed, "
515 "err = %d\n", errno);
516 }
517 }
518
519 void 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
531 void switch_threads(jmp_buf *me, jmp_buf *you)
532 {
533 if (UML_SETJMP(me) == 0)
534 UML_LONGJMP(you, 1);
535 }
536
537 static jmp_buf initial_jmpbuf;
538
539 /* XXX Make these percpu */
540 static void (*cb_proc)(void *arg);
541 static void *cb_arg;
542 static jmp_buf *cb_back;
543
544 int start_idle_thread(void *stack, jmp_buf *switch_buf)
545 {
546 int n;
547
548 set_handler(SIGWINCH, (__sighandler_t) sig_handler,
549 SA_ONSTACK | SA_RESTART, SIGUSR1, SIGIO, SIGVTALRM, -1);
550
551 /*
552 * Can't use UML_SETJMP or UML_LONGJMP here because they save
553 * and restore signals, with the possible side-effect of
554 * trying to handle any signals which came when they were
555 * blocked, which can't be done on this stack.
556 * Signals must be blocked when jumping back here and restored
557 * after returning to the jumper.
558 */
559 n = setjmp(initial_jmpbuf);
560 switch(n) {
561 case INIT_JMP_NEW_THREAD:
562 (*switch_buf)[0].JB_IP = (unsigned long) new_thread_handler;
563 (*switch_buf)[0].JB_SP = (unsigned long) stack +
564 UM_THREAD_SIZE - sizeof(void *);
565 break;
566 case INIT_JMP_CALLBACK:
567 (*cb_proc)(cb_arg);
568 longjmp(*cb_back, 1);
569 break;
570 case INIT_JMP_HALT:
571 kmalloc_ok = 0;
572 return 0;
573 case INIT_JMP_REBOOT:
574 kmalloc_ok = 0;
575 return 1;
576 default:
577 panic("Bad sigsetjmp return in start_idle_thread - %d\n", n);
578 }
579 longjmp(*switch_buf, 1);
580 }
581
582 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
583 {
584 jmp_buf here;
585
586 cb_proc = proc;
587 cb_arg = arg;
588 cb_back = &here;
589
590 block_signals();
591 if (UML_SETJMP(&here) == 0)
592 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
593 unblock_signals();
594
595 cb_proc = NULL;
596 cb_arg = NULL;
597 cb_back = NULL;
598 }
599
600 void halt_skas(void)
601 {
602 block_signals();
603 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
604 }
605
606 void reboot_skas(void)
607 {
608 block_signals();
609 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
610 }
611
612 void __switch_mm(struct mm_id *mm_idp)
613 {
614 int err;
615
616 /* FIXME: need cpu pid in __switch_mm */
617 if (proc_mm) {
618 err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
619 mm_idp->u.mm_fd);
620 if (err)
621 panic("__switch_mm - PTRACE_SWITCH_MM failed, "
622 "errno = %d\n", errno);
623 }
624 else userspace_pid[0] = mm_idp->u.pid;
625 }
Linux kernel - Deliverables
This page took 0.070028 seconds and 5 git commands to generate.