]> git.karo-electronics.de Git - karo-tx-linux.git/blob - arch/um/os-Linux/skas/process.c
Merge tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
[karo-tx-linux.git] / arch / um / os-Linux / skas / process.c
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
26 int is_skas_winch(int pid, int fd, void *data)
27 {
28         return pid == getpgrp();
29 }
30
31 static 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
55 void 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
78 bad_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
89 extern unsigned long current_stub_stack(void);
90
91 static void get_skas_faultinfo(int pid, struct faultinfo *fi)
92 {
93         int err;
94         unsigned long fpregs[FP_SIZE];
95
96         err = get_fp_registers(pid, fpregs);
97         if (err < 0) {
98                 printk(UM_KERN_ERR "save_fp_registers returned %d\n",
99                        err);
100                 fatal_sigsegv();
101         }
102         err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV);
103         if (err) {
104                 printk(UM_KERN_ERR "Failed to continue stub, pid = %d, "
105                        "errno = %d\n", pid, errno);
106                 fatal_sigsegv();
107         }
108         wait_stub_done(pid);
109
110         /*
111          * faultinfo is prepared by the stub_segv_handler at start of
112          * the stub stack page. We just have to copy it.
113          */
114         memcpy(fi, (void *)current_stub_stack(), sizeof(*fi));
115
116         err = put_fp_registers(pid, fpregs);
117         if (err < 0) {
118                 printk(UM_KERN_ERR "put_fp_registers returned %d\n",
119                        err);
120                 fatal_sigsegv();
121         }
122 }
123
124 static void handle_segv(int pid, struct uml_pt_regs * regs)
125 {
126         get_skas_faultinfo(pid, &regs->faultinfo);
127         segv(regs->faultinfo, 0, 1, NULL);
128 }
129
130 /*
131  * To use the same value of using_sysemu as the caller, ask it that value
132  * (in local_using_sysemu
133  */
134 static void handle_trap(int pid, struct uml_pt_regs *regs,
135                         int local_using_sysemu)
136 {
137         int err, status;
138
139         if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END))
140                 fatal_sigsegv();
141
142         if (!local_using_sysemu)
143         {
144                 err = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
145                              __NR_getpid);
146                 if (err < 0) {
147                         printk(UM_KERN_ERR "handle_trap - nullifying syscall "
148                                "failed, errno = %d\n", errno);
149                         fatal_sigsegv();
150                 }
151
152                 err = ptrace(PTRACE_SYSCALL, pid, 0, 0);
153                 if (err < 0) {
154                         printk(UM_KERN_ERR "handle_trap - continuing to end of "
155                                "syscall failed, errno = %d\n", errno);
156                         fatal_sigsegv();
157                 }
158
159                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
160                 if ((err < 0) || !WIFSTOPPED(status) ||
161                     (WSTOPSIG(status) != SIGTRAP + 0x80)) {
162                         err = ptrace_dump_regs(pid);
163                         if (err)
164                                 printk(UM_KERN_ERR "Failed to get registers "
165                                        "from process, errno = %d\n", -err);
166                         printk(UM_KERN_ERR "handle_trap - failed to wait at "
167                                "end of syscall, errno = %d, status = %d\n",
168                                errno, status);
169                         fatal_sigsegv();
170                 }
171         }
172
173         handle_syscall(regs);
174 }
175
176 extern char __syscall_stub_start[];
177
178 /**
179  * userspace_tramp() - userspace trampoline
180  * @stack:      pointer to the new userspace stack page, can be NULL, if? FIXME:
181  *
182  * The userspace trampoline is used to setup a new userspace process in start_userspace() after it was clone()'ed.
183  * This function will run on a temporary stack page.
184  * It ptrace()'es itself, then
185  * Two pages are mapped into the userspace address space:
186  * - STUB_CODE (with EXEC), which contains the skas stub code
187  * - 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.
188  * Also for the userspace process a SIGSEGV handler is installed to catch pagefaults in the userspace process.
189  * And last the process stops itself to give control to the UML kernel for this userspace process.
190  *
191  * Return: Always zero, otherwise the current userspace process is ended with non null exit() call
192  */
193 static int userspace_tramp(void *stack)
194 {
195         void *addr;
196         int fd;
197         unsigned long long offset;
198
199         ptrace(PTRACE_TRACEME, 0, 0, 0);
200
201         signal(SIGTERM, SIG_DFL);
202         signal(SIGWINCH, SIG_IGN);
203
204         /*
205          * This has a pte, but it can't be mapped in with the usual
206          * tlb_flush mechanism because this is part of that mechanism
207          */
208         fd = phys_mapping(to_phys(__syscall_stub_start), &offset);
209         addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE,
210                       PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset);
211         if (addr == MAP_FAILED) {
212                 printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, "
213                        "errno = %d\n", STUB_CODE, errno);
214                 exit(1);
215         }
216
217         if (stack != NULL) {
218                 fd = phys_mapping(to_phys(stack), &offset);
219                 addr = mmap((void *) STUB_DATA,
220                             UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE,
221                             MAP_FIXED | MAP_SHARED, fd, offset);
222                 if (addr == MAP_FAILED) {
223                         printk(UM_KERN_ERR "mapping segfault stack "
224                                "at 0x%lx failed, errno = %d\n",
225                                STUB_DATA, errno);
226                         exit(1);
227                 }
228         }
229         if (stack != NULL) {
230                 struct sigaction sa;
231
232                 unsigned long v = STUB_CODE +
233                                   (unsigned long) stub_segv_handler -
234                                   (unsigned long) __syscall_stub_start;
235
236                 set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE);
237                 sigemptyset(&sa.sa_mask);
238                 sa.sa_flags = SA_ONSTACK | SA_NODEFER | SA_SIGINFO;
239                 sa.sa_sigaction = (void *) v;
240                 sa.sa_restorer = NULL;
241                 if (sigaction(SIGSEGV, &sa, NULL) < 0) {
242                         printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV "
243                                "handler failed - errno = %d\n", errno);
244                         exit(1);
245                 }
246         }
247
248         kill(os_getpid(), SIGSTOP);
249         return 0;
250 }
251
252 int userspace_pid[NR_CPUS];
253
254 /**
255  * start_userspace() - prepare a new userspace process
256  * @stub_stack: pointer to the stub stack. Can be NULL, if? FIXME:
257  *
258  * Setups a new temporary stack page that is used while userspace_tramp() runs
259  * Clones the kernel process into a new userspace process, with FDs only.
260  *
261  * Return: When positive: the process id of the new userspace process,
262  *         when negative: an error number.
263  * FIXME: can PIDs become negative?!
264  */
265 int start_userspace(unsigned long stub_stack)
266 {
267         void *stack;
268         unsigned long sp;
269         int pid, status, n, flags, err;
270
271         /* setup a temporary stack page */
272         stack = mmap(NULL, UM_KERN_PAGE_SIZE,
273                      PROT_READ | PROT_WRITE | PROT_EXEC,
274                      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
275         if (stack == MAP_FAILED) {
276                 err = -errno;
277                 printk(UM_KERN_ERR "start_userspace : mmap failed, "
278                        "errno = %d\n", errno);
279                 return err;
280         }
281
282         /* set stack pointer to the end of the stack page, so it can grow downwards */
283         sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *);
284
285         flags = CLONE_FILES | SIGCHLD;
286
287         /* clone into new userspace process */
288         pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack);
289         if (pid < 0) {
290                 err = -errno;
291                 printk(UM_KERN_ERR "start_userspace : clone failed, "
292                        "errno = %d\n", errno);
293                 return err;
294         }
295
296         do {
297                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL));
298                 if (n < 0) {
299                         err = -errno;
300                         printk(UM_KERN_ERR "start_userspace : wait failed, "
301                                "errno = %d\n", errno);
302                         goto out_kill;
303                 }
304         } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGALRM));
305
306         if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) {
307                 err = -EINVAL;
308                 printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got "
309                        "status = %d\n", status);
310                 goto out_kill;
311         }
312
313         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
314                    (void *) PTRACE_O_TRACESYSGOOD) < 0) {
315                 err = -errno;
316                 printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS "
317                        "failed, errno = %d\n", errno);
318                 goto out_kill;
319         }
320
321         if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) {
322                 err = -errno;
323                 printk(UM_KERN_ERR "start_userspace : munmap failed, "
324                        "errno = %d\n", errno);
325                 goto out_kill;
326         }
327
328         return pid;
329
330  out_kill:
331         os_kill_ptraced_process(pid, 1);
332         return err;
333 }
334
335 void userspace(struct uml_pt_regs *regs)
336 {
337         int err, status, op, pid = userspace_pid[0];
338         /* To prevent races if using_sysemu changes under us.*/
339         int local_using_sysemu;
340         siginfo_t si;
341
342         /* Handle any immediate reschedules or signals */
343         interrupt_end();
344
345         while (1) {
346
347                 /*
348                  * This can legitimately fail if the process loads a
349                  * bogus value into a segment register.  It will
350                  * segfault and PTRACE_GETREGS will read that value
351                  * out of the process.  However, PTRACE_SETREGS will
352                  * fail.  In this case, there is nothing to do but
353                  * just kill the process.
354                  */
355                 if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp)) {
356                         printk(UM_KERN_ERR "userspace - ptrace set regs "
357                                "failed, errno = %d\n", errno);
358                         fatal_sigsegv();
359                 }
360
361                 if (put_fp_registers(pid, regs->fp)) {
362                         printk(UM_KERN_ERR "userspace - ptrace set fp regs "
363                                "failed, errno = %d\n", errno);
364                         fatal_sigsegv();
365                 }
366
367                 /* Now we set local_using_sysemu to be used for one loop */
368                 local_using_sysemu = get_using_sysemu();
369
370                 op = SELECT_PTRACE_OPERATION(local_using_sysemu,
371                                              singlestepping(NULL));
372
373                 if (ptrace(op, pid, 0, 0)) {
374                         printk(UM_KERN_ERR "userspace - ptrace continue "
375                                "failed, op = %d, errno = %d\n", op, errno);
376                         fatal_sigsegv();
377                 }
378
379                 CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL));
380                 if (err < 0) {
381                         printk(UM_KERN_ERR "userspace - wait failed, "
382                                "errno = %d\n", errno);
383                         fatal_sigsegv();
384                 }
385
386                 regs->is_user = 1;
387                 if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) {
388                         printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, "
389                                "errno = %d\n", errno);
390                         fatal_sigsegv();
391                 }
392
393                 if (get_fp_registers(pid, regs->fp)) {
394                         printk(UM_KERN_ERR "userspace -  get_fp_registers failed, "
395                                "errno = %d\n", errno);
396                         fatal_sigsegv();
397                 }
398
399                 UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */
400
401                 if (WIFSTOPPED(status)) {
402                         int sig = WSTOPSIG(status);
403
404                         ptrace(PTRACE_GETSIGINFO, pid, 0, (struct siginfo *)&si);
405
406                         switch (sig) {
407                         case SIGSEGV:
408                                 if (PTRACE_FULL_FAULTINFO) {
409                                         get_skas_faultinfo(pid,
410                                                            &regs->faultinfo);
411                                         (*sig_info[SIGSEGV])(SIGSEGV, (struct siginfo *)&si,
412                                                              regs);
413                                 }
414                                 else handle_segv(pid, regs);
415                                 break;
416                         case SIGTRAP + 0x80:
417                                 handle_trap(pid, regs, local_using_sysemu);
418                                 break;
419                         case SIGTRAP:
420                                 relay_signal(SIGTRAP, (struct siginfo *)&si, regs);
421                                 break;
422                         case SIGALRM:
423                                 break;
424                         case SIGIO:
425                         case SIGILL:
426                         case SIGBUS:
427                         case SIGFPE:
428                         case SIGWINCH:
429                                 block_signals();
430                                 (*sig_info[sig])(sig, (struct siginfo *)&si, regs);
431                                 unblock_signals();
432                                 break;
433                         default:
434                                 printk(UM_KERN_ERR "userspace - child stopped "
435                                        "with signal %d\n", sig);
436                                 fatal_sigsegv();
437                         }
438                         pid = userspace_pid[0];
439                         interrupt_end();
440
441                         /* Avoid -ERESTARTSYS handling in host */
442                         if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET)
443                                 PT_SYSCALL_NR(regs->gp) = -1;
444                 }
445         }
446 }
447
448 static unsigned long thread_regs[MAX_REG_NR];
449 static unsigned long thread_fp_regs[FP_SIZE];
450
451 static int __init init_thread_regs(void)
452 {
453         get_safe_registers(thread_regs, thread_fp_regs);
454         /* Set parent's instruction pointer to start of clone-stub */
455         thread_regs[REGS_IP_INDEX] = STUB_CODE +
456                                 (unsigned long) stub_clone_handler -
457                                 (unsigned long) __syscall_stub_start;
458         thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE -
459                 sizeof(void *);
460 #ifdef __SIGNAL_FRAMESIZE
461         thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE;
462 #endif
463         return 0;
464 }
465
466 __initcall(init_thread_regs);
467
468 int copy_context_skas0(unsigned long new_stack, int pid)
469 {
470         int err;
471         unsigned long current_stack = current_stub_stack();
472         struct stub_data *data = (struct stub_data *) current_stack;
473         struct stub_data *child_data = (struct stub_data *) new_stack;
474         unsigned long long new_offset;
475         int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset);
476
477         /*
478          * prepare offset and fd of child's stack as argument for parent's
479          * and child's mmap2 calls
480          */
481         *data = ((struct stub_data) {
482                         .offset = MMAP_OFFSET(new_offset),
483                         .fd     = new_fd
484         });
485
486         err = ptrace_setregs(pid, thread_regs);
487         if (err < 0) {
488                 err = -errno;
489                 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS "
490                        "failed, pid = %d, errno = %d\n", pid, -err);
491                 return err;
492         }
493
494         err = put_fp_registers(pid, thread_fp_regs);
495         if (err < 0) {
496                 printk(UM_KERN_ERR "copy_context_skas0 : put_fp_registers "
497                        "failed, pid = %d, err = %d\n", pid, err);
498                 return err;
499         }
500
501         /* set a well known return code for detection of child write failure */
502         child_data->err = 12345678;
503
504         /*
505          * Wait, until parent has finished its work: read child's pid from
506          * parent's stack, and check, if bad result.
507          */
508         err = ptrace(PTRACE_CONT, pid, 0, 0);
509         if (err) {
510                 err = -errno;
511                 printk(UM_KERN_ERR "Failed to continue new process, pid = %d, "
512                        "errno = %d\n", pid, errno);
513                 return err;
514         }
515
516         wait_stub_done(pid);
517
518         pid = data->err;
519         if (pid < 0) {
520                 printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports "
521                        "error %d\n", -pid);
522                 return pid;
523         }
524
525         /*
526          * Wait, until child has finished too: read child's result from
527          * child's stack and check it.
528          */
529         wait_stub_done(pid);
530         if (child_data->err != STUB_DATA) {
531                 printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports "
532                        "error %ld\n", child_data->err);
533                 err = child_data->err;
534                 goto out_kill;
535         }
536
537         if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL,
538                    (void *)PTRACE_O_TRACESYSGOOD) < 0) {
539                 err = -errno;
540                 printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS "
541                        "failed, errno = %d\n", errno);
542                 goto out_kill;
543         }
544
545         return pid;
546
547  out_kill:
548         os_kill_ptraced_process(pid, 1);
549         return err;
550 }
551
552 void new_thread(void *stack, jmp_buf *buf, void (*handler)(void))
553 {
554         (*buf)[0].JB_IP = (unsigned long) handler;
555         (*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE -
556                 sizeof(void *);
557 }
558
559 #define INIT_JMP_NEW_THREAD 0
560 #define INIT_JMP_CALLBACK 1
561 #define INIT_JMP_HALT 2
562 #define INIT_JMP_REBOOT 3
563
564 void switch_threads(jmp_buf *me, jmp_buf *you)
565 {
566         if (UML_SETJMP(me) == 0)
567                 UML_LONGJMP(you, 1);
568 }
569
570 static jmp_buf initial_jmpbuf;
571
572 /* XXX Make these percpu */
573 static void (*cb_proc)(void *arg);
574 static void *cb_arg;
575 static jmp_buf *cb_back;
576
577 int start_idle_thread(void *stack, jmp_buf *switch_buf)
578 {
579         int n;
580
581         set_handler(SIGWINCH);
582
583         /*
584          * Can't use UML_SETJMP or UML_LONGJMP here because they save
585          * and restore signals, with the possible side-effect of
586          * trying to handle any signals which came when they were
587          * blocked, which can't be done on this stack.
588          * Signals must be blocked when jumping back here and restored
589          * after returning to the jumper.
590          */
591         n = setjmp(initial_jmpbuf);
592         switch (n) {
593         case INIT_JMP_NEW_THREAD:
594                 (*switch_buf)[0].JB_IP = (unsigned long) uml_finishsetup;
595                 (*switch_buf)[0].JB_SP = (unsigned long) stack +
596                         UM_THREAD_SIZE - sizeof(void *);
597                 break;
598         case INIT_JMP_CALLBACK:
599                 (*cb_proc)(cb_arg);
600                 longjmp(*cb_back, 1);
601                 break;
602         case INIT_JMP_HALT:
603                 kmalloc_ok = 0;
604                 return 0;
605         case INIT_JMP_REBOOT:
606                 kmalloc_ok = 0;
607                 return 1;
608         default:
609                 printk(UM_KERN_ERR "Bad sigsetjmp return in "
610                        "start_idle_thread - %d\n", n);
611                 fatal_sigsegv();
612         }
613         longjmp(*switch_buf, 1);
614 }
615
616 void initial_thread_cb_skas(void (*proc)(void *), void *arg)
617 {
618         jmp_buf here;
619
620         cb_proc = proc;
621         cb_arg = arg;
622         cb_back = &here;
623
624         block_signals();
625         if (UML_SETJMP(&here) == 0)
626                 UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK);
627         unblock_signals();
628
629         cb_proc = NULL;
630         cb_arg = NULL;
631         cb_back = NULL;
632 }
633
634 void halt_skas(void)
635 {
636         block_signals();
637         UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT);
638 }
639
640 void reboot_skas(void)
641 {
642         block_signals();
643         UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT);
644 }
645
646 void __switch_mm(struct mm_id *mm_idp)
647 {
648         userspace_pid[0] = mm_idp->u.pid;
649 }