2 * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
3 * Licensed under the GPL
13 #include <asm/unistd.h>
14 #include "as-layout.h"
16 #include "kern_util.h"
20 #include "ptrace_user.h"
21 #include "registers.h"
23 #include "skas_ptrace.h"
24 #include "sysdep/stub.h"
26 int is_skas_winch(int pid, int fd, void *data)
28 return pid == getpgrp();
31 static int ptrace_dump_regs(int pid)
33 unsigned long regs[MAX_REG_NR];
36 if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
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]);
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.
50 #define STUB_SIG_MASK ((1 << SIGVTALRM) | (1 << SIGWINCH))
52 /* Signals that the stub will finish with - anything else is an error */
53 #define STUB_DONE_MASK (1 << SIGTRAP)
55 void wait_stub_done(int pid)
60 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
61 if ((n < 0) || !WIFSTOPPED(status))
64 if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0)
67 err = ptrace(PTRACE_CONT, pid, 0, 0);
69 printk(UM_KERN_ERR "wait_stub_done : continue failed, "
70 "errno = %d\n", errno);
75 if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0)
79 err = ptrace_dump_regs(pid);
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,
89 extern unsigned long current_stub_stack(void);
91 static void get_skas_faultinfo(int pid, struct faultinfo *fi)
95 if (ptrace_faultinfo) {
96 err = ptrace(PTRACE_FAULTINFO, pid, 0, fi);
98 printk(UM_KERN_ERR "get_skas_faultinfo - "
99 "PTRACE_FAULTINFO failed, errno = %d\n", errno);
103 /* Special handling for i386, which has different structs */
104 if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo))
105 memset((char *)fi + sizeof(struct ptrace_faultinfo), 0,
106 sizeof(struct faultinfo) -
107 sizeof(struct ptrace_faultinfo));
110 unsigned long fpregs[FP_SIZE];
112 err = get_fp_registers(pid, fpregs);
114 printk(UM_KERN_ERR "save_fp_registers returned %d\n",
118 err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
120 printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
121 "errno = %d\n", pid, errno);
127 * faultinfo is prepared by the stub-segv-handler at start of
128 * the stub stack page. We just have to copy it.
130 memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
132 err = put_fp_registers(pid, fpregs);
134 printk(UM_KERN_ERR "put_fp_registers returned %d\n",
141 static void handle_segv(int pid, struct uml_pt_regs * regs)
143 get_skas_faultinfo(pid, ®s->faultinfo);
144 segv(regs->faultinfo, 0, 1, NULL);
148 * To use the same value of using_sysemu as the caller, ask it that value
149 * (in local_using_sysemu
151 static void handle_trap(int pid, struct uml_pt_regs *regs,
152 int local_using_sysemu)
156 if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
159 /* Mark this as a syscall */
160 UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->gp);
162 if (!local_using_sysemu)
164 err = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
167 printk(UM_KERN_ERR "handle_trap - nullifying syscall "
168 "failed, errno = %d\n", errno);
172 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
174 printk(UM_KERN_ERR "handle_trap - continuing to end of "
175 "syscall failed, errno = %d\n", errno);
179 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
180 if ((err < 0) || !WIFSTOPPED(status) ||
181 (WSTOPSIG(status) != SIGTRAP + 0x80)) {
182 err = ptrace_dump_regs(pid);
184 printk(UM_KERN_ERR "Failed to get registers "
185 "from process, errno = %d\n", -err);
186 printk(UM_KERN_ERR "handle_trap - failed to wait at "
187 "end of syscall, errno = %d, status = %d\n",
193 handle_syscall(regs);
196 extern int __syscall_stub_start;
198 static int userspace_tramp(void *stack)
203 ptrace(PTRACE_TRACEME, 0, 0, 0);
205 signal(SIGTERM, SIG_DFL);
206 signal(SIGWINCH, SIG_IGN);
207 err = set_interval();
209 printk(UM_KERN_ERR "userspace_tramp - setting timer failed, "
210 "errno = %d\n", err);
216 * This has a pte, but it can't be mapped in with the usual
217 * tlb_flush mechanism because this is part of that mechanism
220 unsigned long long offset;
221 fd = phys_mapping(to_phys(&__syscall_stub_start), &offset);
222 addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
223 PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
224 if (addr == MAP_FAILED) {
225 printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
226 "errno = %d\n", STUB_CODE, errno);
231 fd = phys_mapping(to_phys(stack), &offset);
232 addr = mmap((void *) STUB_DATA,
233 UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
234 MAP_FIXED | MAP_SHARED, fd, offset);
235 if (addr == MAP_FAILED) {
236 printk(UM_KERN_ERR "mapping segfault stack "
237 "at 0x%lx failed, errno = %d\n",
243 if (!ptrace_faultinfo && (stack != NULL)) {
246 unsigned long v = STUB_CODE +
247 (unsigned long) stub_segv_handler -
248 (unsigned long) &__syscall_stub_start;
250 set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
251 sigemptyset(&sa.sa_mask);
252 sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
253 sa.sa_sigaction = (void *) v;
254 sa.sa_restorer = NULL;
255 if (sigaction(SIGSEGV, &sa, NULL) < 0) {
256 printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
257 "handler failed - errno = %d\n", errno);
262 kill(os_getpid(), SIGSTOP);
266 /* Each element set once, and only accessed by a single processor anyway */
269 int userspace_pid[NR_CPUS];
271 int start_userspace(unsigned long stub_stack)
275 int pid, status, n, flags, err;
277 stack = mmap(NULL, UM_KERN_PAGE_SIZE,
278 PROT_READ | PROT_WRITE | PROT_EXEC,
279 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
280 if (stack == MAP_FAILED) {
282 printk(UM_KERN_ERR "start_userspace : mmap failed, "
283 "errno = %d\n", errno);
287 sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
295 pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
298 printk(UM_KERN_ERR "start_userspace : clone failed, "
299 "errno = %d\n", errno);
304 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
307 printk(UM_KERN_ERR "start_userspace : wait failed, "
308 "errno = %d\n", errno);
311 } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM));
313 if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
315 printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
316 "status = %d\n", status);
320 if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
321 (void *) PTRACE_O_TRACESYSGOOD) < 0) {
323 printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
324 "failed, errno = %d\n", errno);
328 if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
330 printk(UM_KERN_ERR "start_userspace : munmap failed, "
331 "errno = %d\n", errno);
338 os_kill_ptraced_process(pid, 1);
342 void userspace(struct uml_pt_regs *regs)
344 struct itimerval timer;
345 unsigned long long nsecs, now;
346 int err, status, op, pid = userspace_pid[0];
347 /* To prevent races if using_sysemu changes under us.*/
348 int local_using_sysemu;
350 if (getitimer(ITIMER_VIRTUAL, &timer))
351 printk(UM_KERN_ERR "Failed to get itimer, errno = %d\n", errno);
352 nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC +
353 timer.it_value.tv_usec * UM_NSEC_PER_USEC;
358 * This can legitimately fail if the process loads a
359 * bogus value into a segment register. It will
360 * segfault and PTRACE_GETREGS will read that value
361 * out of the process. However, PTRACE_SETREGS will
362 * fail. In this case, there is nothing to do but
363 * just kill the process.
365 if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp))
368 if (put_fp_registers(pid, regs->fp))
371 /* Now we set local_using_sysemu to be used for one loop */
372 local_using_sysemu = get_using_sysemu();
374 op = SELECT_PTRACE_OPERATION(local_using_sysemu,
375 singlestepping(NULL));
377 if (ptrace(op, pid, 0, 0)) {
378 printk(UM_KERN_ERR "userspace - ptrace continue "
379 "failed, op = %d, errno = %d\n", op, errno);
383 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
385 printk(UM_KERN_ERR "userspace - wait failed, "
386 "errno = %d\n", errno);
391 if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
392 printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
393 "errno = %d\n", errno);
397 if (get_fp_registers(pid, regs->fp)) {
398 printk(UM_KERN_ERR "userspace - get_fp_registers failed, "
399 "errno = %d\n", errno);
403 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
405 if (WIFSTOPPED(status)) {
406 int sig = WSTOPSIG(status);
409 if (PTRACE_FULL_FAULTINFO ||
411 get_skas_faultinfo(pid,
413 (*sig_info[SIGSEGV])(SIGSEGV, regs);
415 else handle_segv(pid, regs);
418 handle_trap(pid, regs, local_using_sysemu);
421 relay_signal(SIGTRAP, regs);
428 (*sig_info[sig])(sig, regs);
430 nsecs = timer.it_value.tv_sec *
432 timer.it_value.tv_usec *
442 (*sig_info[sig])(sig, regs);
446 printk(UM_KERN_ERR "userspace - child stopped "
447 "with signal %d\n", sig);
450 pid = userspace_pid[0];
453 /* Avoid -ERESTARTSYS handling in host */
454 if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
455 PT_SYSCALL_NR(regs->gp) = -1;
460 static unsigned long thread_regs[MAX_REG_NR];
461 static unsigned long thread_fp_regs[FP_SIZE];
463 static int __init init_thread_regs(void)
465 get_safe_registers(thread_regs, thread_fp_regs);
466 /* Set parent's instruction pointer to start of clone-stub */
467 thread_regs[REGS_IP_INDEX] = STUB_CODE +
468 (unsigned long) stub_clone_handler -
469 (unsigned long) &__syscall_stub_start;
470 thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
472 #ifdef __SIGNAL_FRAMESIZE
473 thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
478 __initcall(init_thread_regs);
480 int copy_context_skas0(unsigned long new_stack, int pid)
482 struct timeval tv = { .tv_sec = 0, .tv_usec = UM_USEC_PER_SEC / UM_HZ };
484 unsigned long current_stack = current_stub_stack();
485 struct stub_data *data = (struct stub_data *) current_stack;
486 struct stub_data *child_data = (struct stub_data *) new_stack;
487 unsigned long long new_offset;
488 int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
491 * prepare offset and fd of child's stack as argument for parent's
492 * and child's mmap2 calls
494 *data = ((struct stub_data) { .offset = MMAP_OFFSET(new_offset),
496 .timer = ((struct itimerval)
498 .it_interval = tv }) });
500 err = ptrace_setregs(pid, thread_regs);
503 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
504 "failed, pid = %d, errno = %d\n", pid, -err);
508 err = put_fp_registers(pid, thread_fp_regs);
510 printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
511 "failed, pid = %d, err = %d\n", pid, err);
515 /* set a well known return code for detection of child write failure */
516 child_data->err = 12345678;
519 * Wait, until parent has finished its work: read child's pid from
520 * parent's stack, and check, if bad result.
522 err = ptrace(PTRACE_CONT, pid, 0, 0);
525 printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
526 "errno = %d\n", pid, errno);
534 printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
540 * Wait, until child has finished too: read child's result from
541 * child's stack and check it.
544 if (child_data->err != STUB_DATA) {
545 printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
546 "error %ld\n", child_data->err);
547 err = child_data->err;
551 if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
552 (void *)PTRACE_O_TRACESYSGOOD) < 0) {
554 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
555 "failed, errno = %d\n", errno);
562 os_kill_ptraced_process(pid, 1);
567 * This is used only, if stub pages are needed, while proc_mm is
568 * available. Opening /proc/mm creates a new mm_context, which lacks
569 * the stub-pages. Thus, we map them using /proc/mm-fd
571 int map_stub_pages(int fd, unsigned long code, unsigned long data,
574 struct proc_mm_op mmop;
576 unsigned long long code_offset;
577 int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start),
580 mmop = ((struct proc_mm_op) { .op = MM_MMAP,
584 .len = UM_KERN_PAGE_SIZE,
586 .flags = MAP_FIXED | MAP_PRIVATE,
588 .offset = code_offset
590 CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
591 if (n != sizeof(mmop)) {
593 printk(UM_KERN_ERR "mmap args - addr = 0x%lx, fd = %d, "
594 "offset = %llx\n", code, code_fd,
595 (unsigned long long) code_offset);
596 printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for code "
597 "failed, err = %d\n", n);
602 unsigned long long map_offset;
603 int map_fd = phys_mapping(to_phys((void *)stack), &map_offset);
604 mmop = ((struct proc_mm_op)
609 .len = UM_KERN_PAGE_SIZE,
610 .prot = PROT_READ | PROT_WRITE,
611 .flags = MAP_FIXED | MAP_SHARED,
615 CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop)));
616 if (n != sizeof(mmop)) {
618 printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for "
619 "data failed, err = %d\n", n);
627 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
629 (*buf)[0].JB_IP = (unsigned long) handler;
630 (*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
634 #define INIT_JMP_NEW_THREAD 0
635 #define INIT_JMP_CALLBACK 1
636 #define INIT_JMP_HALT 2
637 #define INIT_JMP_REBOOT 3
639 void switch_threads(jmp_buf *me, jmp_buf *you)
641 if (UML_SETJMP(me) == 0)
645 static jmp_buf initial_jmpbuf;
647 /* XXX Make these percpu */
648 static void (*cb_proc)(void *arg);
650 static jmp_buf *cb_back;
652 int start_idle_thread(void *stack, jmp_buf *switch_buf)
656 set_handler(SIGWINCH);
659 * Can't use UML_SETJMP or UML_LONGJMP here because they save
660 * and restore signals, with the possible side-effect of
661 * trying to handle any signals which came when they were
662 * blocked, which can't be done on this stack.
663 * Signals must be blocked when jumping back here and restored
664 * after returning to the jumper.
666 n = setjmp(initial_jmpbuf);
668 case INIT_JMP_NEW_THREAD:
669 (*switch_buf)[0].JB_IP = (unsigned long) new_thread_handler;
670 (*switch_buf)[0].JB_SP = (unsigned long) stack +
671 UM_THREAD_SIZE - sizeof(void *);
673 case INIT_JMP_CALLBACK:
675 longjmp(*cb_back, 1);
680 case INIT_JMP_REBOOT:
684 printk(UM_KERN_ERR "Bad sigsetjmp return in "
685 "start_idle_thread - %d\n", n);
688 longjmp(*switch_buf, 1);
691 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
700 if (UML_SETJMP(&here) == 0)
701 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
712 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
715 void reboot_skas(void)
718 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
721 void __switch_mm(struct mm_id *mm_idp)
725 /* FIXME: need cpu pid in __switch_mm */
727 err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0,
730 printk(UM_KERN_ERR "__switch_mm - PTRACE_SWITCH_MM "
731 "failed, errno = %d\n", errno);
735 else userspace_pid[0] = mm_idp->u.pid;