Commit | Line | Data |
---|---|---|
abaf6977 | 1 | /* |
ba180fd4 | 2 | * Copyright (C) 2002- 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com) |
abaf6977 GS |
3 | * Licensed under the GPL |
4 | */ | |
5 | ||
6 | #include <stdlib.h> | |
abaf6977 | 7 | #include <unistd.h> |
abaf6977 | 8 | #include <sched.h> |
ba180fd4 JD |
9 | #include <errno.h> |
10 | #include <string.h> | |
abaf6977 | 11 | #include <sys/mman.h> |
ba180fd4 JD |
12 | #include <sys/ptrace.h> |
13 | #include <sys/wait.h> | |
14 | #include <asm/unistd.h> | |
15 | #include "as-layout.h" | |
abaf6977 | 16 | #include "chan_user.h" |
ba180fd4 | 17 | #include "kern_constants.h" |
edea1385 | 18 | #include "kern_util.h" |
abaf6977 | 19 | #include "mem.h" |
ba180fd4 | 20 | #include "os.h" |
abaf6977 | 21 | #include "process.h" |
ba180fd4 JD |
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" | |
abaf6977 GS |
29 | |
30 | int is_skas_winch(int pid, int fd, void *data) | |
31 | { | |
512b6fb1 | 32 | if (pid != getpgrp()) |
ba180fd4 | 33 | return 0; |
abaf6977 | 34 | |
42a359e3 | 35 | register_winch_irq(-1, fd, -1, data, 0); |
ba180fd4 | 36 | return 1; |
abaf6977 GS |
37 | } |
38 | ||
f30c2c98 JD |
39 | static int ptrace_dump_regs(int pid) |
40 | { | |
3e6f2ac4 JD |
41 | unsigned long regs[MAX_REG_NR]; |
42 | int i; | |
f30c2c98 | 43 | |
3e6f2ac4 JD |
44 | if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0) |
45 | return -errno; | |
ba180fd4 JD |
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]); | |
f30c2c98 | 50 | |
3e6f2ac4 | 51 | return 0; |
f30c2c98 JD |
52 | } |
53 | ||
16dd07bc JD |
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 | */ | |
3d5ede6f | 58 | #define STUB_SIG_MASK ((1 << SIGVTALRM) | (1 << SIGWINCH)) |
16dd07bc JD |
59 | |
60 | /* Signals that the stub will finish with - anything else is an error */ | |
ee3d9bd4 | 61 | #define STUB_DONE_MASK (1 << SIGTRAP) |
16dd07bc JD |
62 | |
63 | void wait_stub_done(int pid) | |
abaf6977 GS |
64 | { |
65 | int n, status, err; | |
66 | ||
ba180fd4 | 67 | while (1) { |
4dbed85a | 68 | CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL)); |
ba180fd4 | 69 | if ((n < 0) || !WIFSTOPPED(status)) |
16dd07bc JD |
70 | goto bad_wait; |
71 | ||
ba180fd4 | 72 | if (((1 << WSTOPSIG(status)) & STUB_SIG_MASK) == 0) |
16dd07bc JD |
73 | break; |
74 | ||
75 | err = ptrace(PTRACE_CONT, pid, 0, 0); | |
3e6f2ac4 JD |
76 | if (err) { |
77 | printk(UM_KERN_ERR "wait_stub_done : continue failed, " | |
78 | "errno = %d\n", errno); | |
79 | fatal_sigsegv(); | |
80 | } | |
abaf6977 | 81 | } |
16dd07bc | 82 | |
ba180fd4 | 83 | if (((1 << WSTOPSIG(status)) & STUB_DONE_MASK) != 0) |
16dd07bc JD |
84 | return; |
85 | ||
86 | bad_wait: | |
87 | err = ptrace_dump_regs(pid); | |
ba180fd4 JD |
88 | if (err) |
89 | printk(UM_KERN_ERR "Failed to get registers from stub, " | |
90 | "errno = %d\n", -err); | |
3e6f2ac4 JD |
91 | printk(UM_KERN_ERR "wait_stub_done : failed to wait for SIGTRAP, " |
92 | "pid = %d, n = %d, errno = %d, status = 0x%x\n", pid, n, errno, | |
93 | status); | |
94 | fatal_sigsegv(); | |
abaf6977 GS |
95 | } |
96 | ||
97 | extern unsigned long current_stub_stack(void); | |
98 | ||
99 | void get_skas_faultinfo(int pid, struct faultinfo * fi) | |
100 | { | |
101 | int err; | |
102 | ||
ba180fd4 | 103 | if (ptrace_faultinfo) { |
abaf6977 | 104 | err = ptrace(PTRACE_FAULTINFO, pid, 0, fi); |
3e6f2ac4 JD |
105 | if (err) { |
106 | printk(UM_KERN_ERR "get_skas_faultinfo - " | |
107 | "PTRACE_FAULTINFO failed, errno = %d\n", errno); | |
108 | fatal_sigsegv(); | |
109 | } | |
abaf6977 GS |
110 | |
111 | /* Special handling for i386, which has different structs */ | |
112 | if (sizeof(struct ptrace_faultinfo) < sizeof(struct faultinfo)) | |
113 | memset((char *)fi + sizeof(struct ptrace_faultinfo), 0, | |
114 | sizeof(struct faultinfo) - | |
115 | sizeof(struct ptrace_faultinfo)); | |
116 | } | |
117 | else { | |
2f56debd JD |
118 | unsigned long fpregs[FP_SIZE]; |
119 | ||
120 | err = get_fp_registers(pid, fpregs); | |
121 | if (err < 0) { | |
122 | printk(UM_KERN_ERR "save_fp_registers returned %d\n", | |
123 | err); | |
124 | fatal_sigsegv(); | |
125 | } | |
16dd07bc | 126 | err = ptrace(PTRACE_CONT, pid, 0, SIGSEGV); |
3e6f2ac4 JD |
127 | if (err) { |
128 | printk(UM_KERN_ERR "Failed to continue stub, pid = %d, " | |
129 | "errno = %d\n", pid, errno); | |
130 | fatal_sigsegv(); | |
131 | } | |
16dd07bc | 132 | wait_stub_done(pid); |
abaf6977 | 133 | |
ba180fd4 JD |
134 | /* |
135 | * faultinfo is prepared by the stub-segv-handler at start of | |
abaf6977 GS |
136 | * the stub stack page. We just have to copy it. |
137 | */ | |
138 | memcpy(fi, (void *)current_stub_stack(), sizeof(*fi)); | |
2f56debd JD |
139 | |
140 | err = put_fp_registers(pid, fpregs); | |
141 | if (err < 0) { | |
142 | printk(UM_KERN_ERR "put_fp_registers returned %d\n", | |
143 | err); | |
144 | fatal_sigsegv(); | |
145 | } | |
abaf6977 GS |
146 | } |
147 | } | |
148 | ||
77bf4400 | 149 | static void handle_segv(int pid, struct uml_pt_regs * regs) |
abaf6977 | 150 | { |
77bf4400 JD |
151 | get_skas_faultinfo(pid, ®s->faultinfo); |
152 | segv(regs->faultinfo, 0, 1, NULL); | |
abaf6977 GS |
153 | } |
154 | ||
ba180fd4 JD |
155 | /* |
156 | * To use the same value of using_sysemu as the caller, ask it that value | |
157 | * (in local_using_sysemu | |
158 | */ | |
159 | static void handle_trap(int pid, struct uml_pt_regs *regs, | |
160 | int local_using_sysemu) | |
abaf6977 GS |
161 | { |
162 | int err, status; | |
163 | ||
e06173bd JD |
164 | if ((UPT_IP(regs) >= STUB_START) && (UPT_IP(regs) < STUB_END)) |
165 | fatal_sigsegv(); | |
166 | ||
abaf6977 | 167 | /* Mark this as a syscall */ |
18baddda | 168 | UPT_SYSCALL_NR(regs) = PT_SYSCALL_NR(regs->gp); |
abaf6977 GS |
169 | |
170 | if (!local_using_sysemu) | |
171 | { | |
172 | err = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET, | |
173 | __NR_getpid); | |
3e6f2ac4 JD |
174 | if (err < 0) { |
175 | printk(UM_KERN_ERR "handle_trap - nullifying syscall " | |
176 | "failed, errno = %d\n", errno); | |
177 | fatal_sigsegv(); | |
178 | } | |
abaf6977 GS |
179 | |
180 | err = ptrace(PTRACE_SYSCALL, pid, 0, 0); | |
3e6f2ac4 JD |
181 | if (err < 0) { |
182 | printk(UM_KERN_ERR "handle_trap - continuing to end of " | |
183 | "syscall failed, errno = %d\n", errno); | |
184 | fatal_sigsegv(); | |
185 | } | |
abaf6977 | 186 | |
4dbed85a | 187 | CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL)); |
ba180fd4 | 188 | if ((err < 0) || !WIFSTOPPED(status) || |
3e6f2ac4 JD |
189 | (WSTOPSIG(status) != SIGTRAP + 0x80)) { |
190 | err = ptrace_dump_regs(pid); | |
191 | if (err) | |
192 | printk(UM_KERN_ERR "Failed to get registers " | |
ba180fd4 | 193 | "from process, errno = %d\n", -err); |
3e6f2ac4 JD |
194 | printk(UM_KERN_ERR "handle_trap - failed to wait at " |
195 | "end of syscall, errno = %d, status = %d\n", | |
196 | errno, status); | |
197 | fatal_sigsegv(); | |
198 | } | |
abaf6977 GS |
199 | } |
200 | ||
201 | handle_syscall(regs); | |
202 | } | |
203 | ||
204 | extern int __syscall_stub_start; | |
abaf6977 GS |
205 | |
206 | static int userspace_tramp(void *stack) | |
207 | { | |
208 | void *addr; | |
537ae946 | 209 | int err; |
abaf6977 GS |
210 | |
211 | ptrace(PTRACE_TRACEME, 0, 0, 0); | |
212 | ||
a24864a1 | 213 | signal(SIGTERM, SIG_DFL); |
ee3d9bd4 | 214 | signal(SIGWINCH, SIG_IGN); |
a2f018bf | 215 | err = set_interval(); |
3e6f2ac4 JD |
216 | if (err) { |
217 | printk(UM_KERN_ERR "userspace_tramp - setting timer failed, " | |
218 | "errno = %d\n", err); | |
219 | exit(1); | |
220 | } | |
abaf6977 | 221 | |
ba180fd4 JD |
222 | if (!proc_mm) { |
223 | /* | |
224 | * This has a pte, but it can't be mapped in with the usual | |
abaf6977 GS |
225 | * tlb_flush mechanism because this is part of that mechanism |
226 | */ | |
09ee011e | 227 | int fd; |
ba180fd4 | 228 | unsigned long long offset; |
09ee011e | 229 | fd = phys_mapping(to_phys(&__syscall_stub_start), &offset); |
54ae36f2 | 230 | addr = mmap64((void *) STUB_CODE, UM_KERN_PAGE_SIZE, |
09ee011e | 231 | PROT_EXEC, MAP_FIXED | MAP_PRIVATE, fd, offset); |
ba180fd4 | 232 | if (addr == MAP_FAILED) { |
54ae36f2 JD |
233 | printk(UM_KERN_ERR "mapping mmap stub at 0x%lx failed, " |
234 | "errno = %d\n", STUB_CODE, errno); | |
abaf6977 GS |
235 | exit(1); |
236 | } | |
237 | ||
ba180fd4 | 238 | if (stack != NULL) { |
abaf6977 | 239 | fd = phys_mapping(to_phys(stack), &offset); |
54ae36f2 | 240 | addr = mmap((void *) STUB_DATA, |
1ffb9164 | 241 | UM_KERN_PAGE_SIZE, PROT_READ | PROT_WRITE, |
abaf6977 | 242 | MAP_FIXED | MAP_SHARED, fd, offset); |
ba180fd4 JD |
243 | if (addr == MAP_FAILED) { |
244 | printk(UM_KERN_ERR "mapping segfault stack " | |
54ae36f2 JD |
245 | "at 0x%lx failed, errno = %d\n", |
246 | STUB_DATA, errno); | |
abaf6977 GS |
247 | exit(1); |
248 | } | |
249 | } | |
250 | } | |
ba180fd4 | 251 | if (!ptrace_faultinfo && (stack != NULL)) { |
4b84c69b JD |
252 | struct sigaction sa; |
253 | ||
54ae36f2 | 254 | unsigned long v = STUB_CODE + |
abaf6977 GS |
255 | (unsigned long) stub_segv_handler - |
256 | (unsigned long) &__syscall_stub_start; | |
257 | ||
54ae36f2 | 258 | set_sigstack((void *) STUB_DATA, UM_KERN_PAGE_SIZE); |
4b84c69b | 259 | sigemptyset(&sa.sa_mask); |
ee3d9bd4 | 260 | sa.sa_flags = SA_ONSTACK | SA_NODEFER; |
4b84c69b JD |
261 | sa.sa_handler = (void *) v; |
262 | sa.sa_restorer = NULL; | |
3e6f2ac4 JD |
263 | if (sigaction(SIGSEGV, &sa, NULL) < 0) { |
264 | printk(UM_KERN_ERR "userspace_tramp - setting SIGSEGV " | |
265 | "handler failed - errno = %d\n", errno); | |
266 | exit(1); | |
267 | } | |
abaf6977 GS |
268 | } |
269 | ||
512b6fb1 | 270 | kill(os_getpid(), SIGSTOP); |
ba180fd4 | 271 | return 0; |
abaf6977 GS |
272 | } |
273 | ||
274 | /* Each element set once, and only accessed by a single processor anyway */ | |
275 | #undef NR_CPUS | |
276 | #define NR_CPUS 1 | |
277 | int userspace_pid[NR_CPUS]; | |
278 | ||
279 | int start_userspace(unsigned long stub_stack) | |
280 | { | |
281 | void *stack; | |
282 | unsigned long sp; | |
3e6f2ac4 | 283 | int pid, status, n, flags, err; |
abaf6977 | 284 | |
c539ab73 JD |
285 | stack = mmap(NULL, UM_KERN_PAGE_SIZE, |
286 | PROT_READ | PROT_WRITE | PROT_EXEC, | |
abaf6977 | 287 | MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); |
3e6f2ac4 JD |
288 | if (stack == MAP_FAILED) { |
289 | err = -errno; | |
290 | printk(UM_KERN_ERR "start_userspace : mmap failed, " | |
b5498832 | 291 | "errno = %d\n", errno); |
3e6f2ac4 JD |
292 | return err; |
293 | } | |
294 | ||
c539ab73 | 295 | sp = (unsigned long) stack + UM_KERN_PAGE_SIZE - sizeof(void *); |
abaf6977 | 296 | |
4dbed85a | 297 | flags = CLONE_FILES; |
ba180fd4 JD |
298 | if (proc_mm) |
299 | flags |= CLONE_VM; | |
4dbed85a SG |
300 | else |
301 | flags |= SIGCHLD; | |
ba180fd4 | 302 | |
abaf6977 | 303 | pid = clone(userspace_tramp, (void *) sp, flags, (void *) stub_stack); |
3e6f2ac4 JD |
304 | if (pid < 0) { |
305 | err = -errno; | |
306 | printk(UM_KERN_ERR "start_userspace : clone failed, " | |
b5498832 | 307 | "errno = %d\n", errno); |
3e6f2ac4 JD |
308 | return err; |
309 | } | |
abaf6977 GS |
310 | |
311 | do { | |
4dbed85a | 312 | CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED | __WALL)); |
3e6f2ac4 JD |
313 | if (n < 0) { |
314 | err = -errno; | |
315 | printk(UM_KERN_ERR "start_userspace : wait failed, " | |
b5498832 | 316 | "errno = %d\n", errno); |
3e6f2ac4 JD |
317 | goto out_kill; |
318 | } | |
ba180fd4 | 319 | } while (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGVTALRM)); |
abaf6977 | 320 | |
3e6f2ac4 JD |
321 | if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP)) { |
322 | err = -EINVAL; | |
323 | printk(UM_KERN_ERR "start_userspace : expected SIGSTOP, got " | |
b5498832 | 324 | "status = %d\n", status); |
3e6f2ac4 JD |
325 | goto out_kill; |
326 | } | |
abaf6977 | 327 | |
ba180fd4 | 328 | if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL, |
3e6f2ac4 JD |
329 | (void *) PTRACE_O_TRACESYSGOOD) < 0) { |
330 | err = -errno; | |
331 | printk(UM_KERN_ERR "start_userspace : PTRACE_OLDSETOPTIONS " | |
332 | "failed, errno = %d\n", errno); | |
333 | goto out_kill; | |
334 | } | |
abaf6977 | 335 | |
3e6f2ac4 JD |
336 | if (munmap(stack, UM_KERN_PAGE_SIZE) < 0) { |
337 | err = -errno; | |
338 | printk(UM_KERN_ERR "start_userspace : munmap failed, " | |
339 | "errno = %d\n", errno); | |
340 | goto out_kill; | |
341 | } | |
abaf6977 | 342 | |
ba180fd4 | 343 | return pid; |
3e6f2ac4 JD |
344 | |
345 | out_kill: | |
346 | os_kill_ptraced_process(pid, 1); | |
347 | return err; | |
abaf6977 GS |
348 | } |
349 | ||
77bf4400 | 350 | void userspace(struct uml_pt_regs *regs) |
abaf6977 | 351 | { |
d2753a6d JD |
352 | struct itimerval timer; |
353 | unsigned long long nsecs, now; | |
abaf6977 | 354 | int err, status, op, pid = userspace_pid[0]; |
2ea5bc5e JD |
355 | /* To prevent races if using_sysemu changes under us.*/ |
356 | int local_using_sysemu; | |
abaf6977 | 357 | |
d2753a6d | 358 | if (getitimer(ITIMER_VIRTUAL, &timer)) |
5134d8fe | 359 | printk(UM_KERN_ERR "Failed to get itimer, errno = %d\n", errno); |
1a805219 JD |
360 | nsecs = timer.it_value.tv_sec * UM_NSEC_PER_SEC + |
361 | timer.it_value.tv_usec * UM_NSEC_PER_USEC; | |
d2753a6d JD |
362 | nsecs += os_nsecs(); |
363 | ||
ba180fd4 | 364 | while (1) { |
3e6f2ac4 JD |
365 | /* |
366 | * This can legitimately fail if the process loads a | |
367 | * bogus value into a segment register. It will | |
368 | * segfault and PTRACE_GETREGS will read that value | |
369 | * out of the process. However, PTRACE_SETREGS will | |
370 | * fail. In this case, there is nothing to do but | |
371 | * just kill the process. | |
372 | */ | |
d25f2e12 | 373 | if (ptrace(PTRACE_SETREGS, pid, 0, regs->gp)) |
3e6f2ac4 | 374 | fatal_sigsegv(); |
abaf6977 GS |
375 | |
376 | /* Now we set local_using_sysemu to be used for one loop */ | |
377 | local_using_sysemu = get_using_sysemu(); | |
378 | ||
2ea5bc5e JD |
379 | op = SELECT_PTRACE_OPERATION(local_using_sysemu, |
380 | singlestepping(NULL)); | |
abaf6977 | 381 | |
3e6f2ac4 JD |
382 | if (ptrace(op, pid, 0, 0)) { |
383 | printk(UM_KERN_ERR "userspace - ptrace continue " | |
384 | "failed, op = %d, errno = %d\n", op, errno); | |
385 | fatal_sigsegv(); | |
386 | } | |
abaf6977 | 387 | |
4dbed85a | 388 | CATCH_EINTR(err = waitpid(pid, &status, WUNTRACED | __WALL)); |
3e6f2ac4 JD |
389 | if (err < 0) { |
390 | printk(UM_KERN_ERR "userspace - wait failed, " | |
391 | "errno = %d\n", errno); | |
392 | fatal_sigsegv(); | |
393 | } | |
abaf6977 | 394 | |
77bf4400 | 395 | regs->is_user = 1; |
3e6f2ac4 JD |
396 | if (ptrace(PTRACE_GETREGS, pid, 0, regs->gp)) { |
397 | printk(UM_KERN_ERR "userspace - PTRACE_GETREGS failed, " | |
398 | "errno = %d\n", errno); | |
399 | fatal_sigsegv(); | |
400 | } | |
d25f2e12 | 401 | |
abaf6977 GS |
402 | UPT_SYSCALL_NR(regs) = -1; /* Assume: It's not a syscall */ |
403 | ||
ba180fd4 | 404 | if (WIFSTOPPED(status)) { |
16dd07bc | 405 | int sig = WSTOPSIG(status); |
5134d8fe | 406 | switch (sig) { |
abaf6977 | 407 | case SIGSEGV: |
ba180fd4 JD |
408 | if (PTRACE_FULL_FAULTINFO || |
409 | !ptrace_faultinfo) { | |
410 | get_skas_faultinfo(pid, | |
411 | ®s->faultinfo); | |
16dd07bc JD |
412 | (*sig_info[SIGSEGV])(SIGSEGV, regs); |
413 | } | |
abaf6977 GS |
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, regs); | |
421 | break; | |
abaf6977 | 422 | case SIGVTALRM: |
d2753a6d | 423 | now = os_nsecs(); |
d25f2e12 | 424 | if (now < nsecs) |
d2753a6d JD |
425 | break; |
426 | block_signals(); | |
427 | (*sig_info[sig])(sig, regs); | |
428 | unblock_signals(); | |
1a805219 JD |
429 | nsecs = timer.it_value.tv_sec * |
430 | UM_NSEC_PER_SEC + | |
431 | timer.it_value.tv_usec * | |
432 | UM_NSEC_PER_USEC; | |
d2753a6d JD |
433 | nsecs += os_nsecs(); |
434 | break; | |
435 | case SIGIO: | |
abaf6977 GS |
436 | case SIGILL: |
437 | case SIGBUS: | |
438 | case SIGFPE: | |
439 | case SIGWINCH: | |
16dd07bc JD |
440 | block_signals(); |
441 | (*sig_info[sig])(sig, regs); | |
442 | unblock_signals(); | |
abaf6977 GS |
443 | break; |
444 | default: | |
96cee304 | 445 | printk(UM_KERN_ERR "userspace - child stopped " |
ba180fd4 | 446 | "with signal %d\n", sig); |
3e6f2ac4 | 447 | fatal_sigsegv(); |
abaf6977 | 448 | } |
abaf6977 GS |
449 | pid = userspace_pid[0]; |
450 | interrupt_end(); | |
451 | ||
452 | /* Avoid -ERESTARTSYS handling in host */ | |
ba180fd4 | 453 | if (PT_SYSCALL_NR_OFFSET != PT_SYSCALL_RET_OFFSET) |
18baddda | 454 | PT_SYSCALL_NR(regs->gp) = -1; |
abaf6977 GS |
455 | } |
456 | } | |
457 | } | |
abaf6977 | 458 | |
16dd07bc | 459 | static unsigned long thread_regs[MAX_REG_NR]; |
16dd07bc JD |
460 | |
461 | static int __init init_thread_regs(void) | |
462 | { | |
42daba31 | 463 | get_safe_registers(thread_regs); |
16dd07bc | 464 | /* Set parent's instruction pointer to start of clone-stub */ |
54ae36f2 | 465 | thread_regs[REGS_IP_INDEX] = STUB_CODE + |
16dd07bc JD |
466 | (unsigned long) stub_clone_handler - |
467 | (unsigned long) &__syscall_stub_start; | |
54ae36f2 | 468 | thread_regs[REGS_SP_INDEX] = STUB_DATA + UM_KERN_PAGE_SIZE - |
16dd07bc JD |
469 | sizeof(void *); |
470 | #ifdef __SIGNAL_FRAMESIZE | |
471 | thread_regs[REGS_SP_INDEX] -= __SIGNAL_FRAMESIZE; | |
472 | #endif | |
473 | return 0; | |
474 | } | |
475 | ||
476 | __initcall(init_thread_regs); | |
477 | ||
abaf6977 GS |
478 | int copy_context_skas0(unsigned long new_stack, int pid) |
479 | { | |
1a805219 | 480 | struct timeval tv = { .tv_sec = 0, .tv_usec = UM_USEC_PER_SEC / UM_HZ }; |
abaf6977 | 481 | int err; |
abaf6977 GS |
482 | unsigned long current_stack = current_stub_stack(); |
483 | struct stub_data *data = (struct stub_data *) current_stack; | |
484 | struct stub_data *child_data = (struct stub_data *) new_stack; | |
0a7675aa | 485 | unsigned long long new_offset; |
abaf6977 GS |
486 | int new_fd = phys_mapping(to_phys((void *)new_stack), &new_offset); |
487 | ||
ba180fd4 JD |
488 | /* |
489 | * prepare offset and fd of child's stack as argument for parent's | |
abaf6977 GS |
490 | * and child's mmap2 calls |
491 | */ | |
492 | *data = ((struct stub_data) { .offset = MMAP_OFFSET(new_offset), | |
493 | .fd = new_fd, | |
494 | .timer = ((struct itimerval) | |
d2753a6d JD |
495 | { .it_value = tv, |
496 | .it_interval = tv }) }); | |
497 | ||
16dd07bc | 498 | err = ptrace_setregs(pid, thread_regs); |
3e6f2ac4 JD |
499 | if (err < 0) { |
500 | err = -errno; | |
501 | printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_SETREGS " | |
502 | "failed, pid = %d, errno = %d\n", pid, -err); | |
503 | return err; | |
504 | } | |
abaf6977 GS |
505 | |
506 | /* set a well known return code for detection of child write failure */ | |
507 | child_data->err = 12345678; | |
508 | ||
ba180fd4 JD |
509 | /* |
510 | * Wait, until parent has finished its work: read child's pid from | |
abaf6977 GS |
511 | * parent's stack, and check, if bad result. |
512 | */ | |
16dd07bc | 513 | err = ptrace(PTRACE_CONT, pid, 0, 0); |
3e6f2ac4 JD |
514 | if (err) { |
515 | err = -errno; | |
516 | printk(UM_KERN_ERR "Failed to continue new process, pid = %d, " | |
517 | "errno = %d\n", pid, errno); | |
518 | return err; | |
519 | } | |
520 | ||
16dd07bc | 521 | wait_stub_done(pid); |
abaf6977 GS |
522 | |
523 | pid = data->err; | |
3e6f2ac4 JD |
524 | if (pid < 0) { |
525 | printk(UM_KERN_ERR "copy_context_skas0 - stub-parent reports " | |
526 | "error %d\n", -pid); | |
527 | return pid; | |
528 | } | |
abaf6977 | 529 | |
ba180fd4 JD |
530 | /* |
531 | * Wait, until child has finished too: read child's result from | |
abaf6977 GS |
532 | * child's stack and check it. |
533 | */ | |
16dd07bc | 534 | wait_stub_done(pid); |
3e6f2ac4 JD |
535 | if (child_data->err != STUB_DATA) { |
536 | printk(UM_KERN_ERR "copy_context_skas0 - stub-child reports " | |
537 | "error %ld\n", child_data->err); | |
538 | err = child_data->err; | |
539 | goto out_kill; | |
540 | } | |
abaf6977 GS |
541 | |
542 | if (ptrace(PTRACE_OLDSETOPTIONS, pid, NULL, | |
3e6f2ac4 JD |
543 | (void *)PTRACE_O_TRACESYSGOOD) < 0) { |
544 | err = -errno; | |
545 | printk(UM_KERN_ERR "copy_context_skas0 : PTRACE_OLDSETOPTIONS " | |
546 | "failed, errno = %d\n", errno); | |
547 | goto out_kill; | |
548 | } | |
abaf6977 GS |
549 | |
550 | return pid; | |
3e6f2ac4 JD |
551 | |
552 | out_kill: | |
553 | os_kill_ptraced_process(pid, 1); | |
554 | return err; | |
abaf6977 GS |
555 | } |
556 | ||
557 | /* | |
558 | * This is used only, if stub pages are needed, while proc_mm is | |
ef0470c0 | 559 | * available. Opening /proc/mm creates a new mm_context, which lacks |
abaf6977 GS |
560 | * the stub-pages. Thus, we map them using /proc/mm-fd |
561 | */ | |
3e6f2ac4 JD |
562 | int map_stub_pages(int fd, unsigned long code, unsigned long data, |
563 | unsigned long stack) | |
abaf6977 GS |
564 | { |
565 | struct proc_mm_op mmop; | |
566 | int n; | |
0a7675aa | 567 | unsigned long long code_offset; |
09ee011e JD |
568 | int code_fd = phys_mapping(to_phys((void *) &__syscall_stub_start), |
569 | &code_offset); | |
abaf6977 GS |
570 | |
571 | mmop = ((struct proc_mm_op) { .op = MM_MMAP, | |
572 | .u = | |
573 | { .mmap = | |
574 | { .addr = code, | |
c539ab73 | 575 | .len = UM_KERN_PAGE_SIZE, |
abaf6977 GS |
576 | .prot = PROT_EXEC, |
577 | .flags = MAP_FIXED | MAP_PRIVATE, | |
09ee011e JD |
578 | .fd = code_fd, |
579 | .offset = code_offset | |
abaf6977 | 580 | } } }); |
a61f334f | 581 | CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop))); |
ba180fd4 | 582 | if (n != sizeof(mmop)) { |
a61f334f | 583 | n = errno; |
ba180fd4 JD |
584 | printk(UM_KERN_ERR "mmap args - addr = 0x%lx, fd = %d, " |
585 | "offset = %llx\n", code, code_fd, | |
586 | (unsigned long long) code_offset); | |
3e6f2ac4 JD |
587 | printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for code " |
588 | "failed, err = %d\n", n); | |
589 | return -n; | |
b4cf95c6 | 590 | } |
abaf6977 | 591 | |
ba180fd4 | 592 | if (stack) { |
0a7675aa | 593 | unsigned long long map_offset; |
abaf6977 GS |
594 | int map_fd = phys_mapping(to_phys((void *)stack), &map_offset); |
595 | mmop = ((struct proc_mm_op) | |
596 | { .op = MM_MMAP, | |
597 | .u = | |
598 | { .mmap = | |
599 | { .addr = data, | |
c539ab73 | 600 | .len = UM_KERN_PAGE_SIZE, |
abaf6977 GS |
601 | .prot = PROT_READ | PROT_WRITE, |
602 | .flags = MAP_FIXED | MAP_SHARED, | |
603 | .fd = map_fd, | |
604 | .offset = map_offset | |
605 | } } }); | |
a61f334f | 606 | CATCH_EINTR(n = write(fd, &mmop, sizeof(mmop))); |
3e6f2ac4 JD |
607 | if (n != sizeof(mmop)) { |
608 | n = errno; | |
609 | printk(UM_KERN_ERR "map_stub_pages : /proc/mm map for " | |
610 | "data failed, err = %d\n", n); | |
611 | return -n; | |
612 | } | |
abaf6977 | 613 | } |
3e6f2ac4 JD |
614 | |
615 | return 0; | |
abaf6977 GS |
616 | } |
617 | ||
3c917350 | 618 | void new_thread(void *stack, jmp_buf *buf, void (*handler)(void)) |
abaf6977 | 619 | { |
3c917350 | 620 | (*buf)[0].JB_IP = (unsigned long) handler; |
e1a79c40 JD |
621 | (*buf)[0].JB_SP = (unsigned long) stack + UM_THREAD_SIZE - |
622 | sizeof(void *); | |
abaf6977 GS |
623 | } |
624 | ||
e2216feb | 625 | #define INIT_JMP_NEW_THREAD 0 |
3c917350 JD |
626 | #define INIT_JMP_CALLBACK 1 |
627 | #define INIT_JMP_HALT 2 | |
628 | #define INIT_JMP_REBOOT 3 | |
abaf6977 | 629 | |
3c917350 | 630 | void switch_threads(jmp_buf *me, jmp_buf *you) |
abaf6977 | 631 | { |
ba180fd4 | 632 | if (UML_SETJMP(me) == 0) |
3c917350 | 633 | UML_LONGJMP(you, 1); |
abaf6977 GS |
634 | } |
635 | ||
ad28e029 | 636 | static jmp_buf initial_jmpbuf; |
abaf6977 GS |
637 | |
638 | /* XXX Make these percpu */ | |
639 | static void (*cb_proc)(void *arg); | |
640 | static void *cb_arg; | |
ad28e029 | 641 | static jmp_buf *cb_back; |
abaf6977 | 642 | |
3c917350 | 643 | int start_idle_thread(void *stack, jmp_buf *switch_buf) |
abaf6977 | 644 | { |
a5df0d1a | 645 | int n; |
abaf6977 GS |
646 | |
647 | set_handler(SIGWINCH, (__sighandler_t) sig_handler, | |
61b63c55 | 648 | SA_ONSTACK | SA_RESTART, SIGUSR1, SIGIO, SIGVTALRM, -1); |
abaf6977 | 649 | |
77f6af77 JD |
650 | /* |
651 | * Can't use UML_SETJMP or UML_LONGJMP here because they save | |
652 | * and restore signals, with the possible side-effect of | |
653 | * trying to handle any signals which came when they were | |
654 | * blocked, which can't be done on this stack. | |
655 | * Signals must be blocked when jumping back here and restored | |
656 | * after returning to the jumper. | |
657 | */ | |
658 | n = setjmp(initial_jmpbuf); | |
5134d8fe | 659 | switch (n) { |
abaf6977 | 660 | case INIT_JMP_NEW_THREAD: |
3c917350 JD |
661 | (*switch_buf)[0].JB_IP = (unsigned long) new_thread_handler; |
662 | (*switch_buf)[0].JB_SP = (unsigned long) stack + | |
e1a79c40 | 663 | UM_THREAD_SIZE - sizeof(void *); |
abaf6977 GS |
664 | break; |
665 | case INIT_JMP_CALLBACK: | |
666 | (*cb_proc)(cb_arg); | |
77f6af77 | 667 | longjmp(*cb_back, 1); |
abaf6977 GS |
668 | break; |
669 | case INIT_JMP_HALT: | |
670 | kmalloc_ok = 0; | |
ba180fd4 | 671 | return 0; |
abaf6977 GS |
672 | case INIT_JMP_REBOOT: |
673 | kmalloc_ok = 0; | |
ba180fd4 | 674 | return 1; |
abaf6977 | 675 | default: |
3e6f2ac4 JD |
676 | printk(UM_KERN_ERR "Bad sigsetjmp return in " |
677 | "start_idle_thread - %d\n", n); | |
678 | fatal_sigsegv(); | |
abaf6977 | 679 | } |
77f6af77 | 680 | longjmp(*switch_buf, 1); |
abaf6977 GS |
681 | } |
682 | ||
683 | void initial_thread_cb_skas(void (*proc)(void *), void *arg) | |
684 | { | |
ad28e029 | 685 | jmp_buf here; |
abaf6977 GS |
686 | |
687 | cb_proc = proc; | |
688 | cb_arg = arg; | |
689 | cb_back = &here; | |
690 | ||
691 | block_signals(); | |
ba180fd4 | 692 | if (UML_SETJMP(&here) == 0) |
ad28e029 | 693 | UML_LONGJMP(&initial_jmpbuf, INIT_JMP_CALLBACK); |
abaf6977 GS |
694 | unblock_signals(); |
695 | ||
696 | cb_proc = NULL; | |
697 | cb_arg = NULL; | |
698 | cb_back = NULL; | |
699 | } | |
700 | ||
701 | void halt_skas(void) | |
702 | { | |
703 | block_signals(); | |
ad28e029 | 704 | UML_LONGJMP(&initial_jmpbuf, INIT_JMP_HALT); |
abaf6977 GS |
705 | } |
706 | ||
707 | void reboot_skas(void) | |
708 | { | |
709 | block_signals(); | |
ad28e029 | 710 | UML_LONGJMP(&initial_jmpbuf, INIT_JMP_REBOOT); |
abaf6977 GS |
711 | } |
712 | ||
77bf4400 | 713 | void __switch_mm(struct mm_id *mm_idp) |
abaf6977 GS |
714 | { |
715 | int err; | |
716 | ||
77bf4400 | 717 | /* FIXME: need cpu pid in __switch_mm */ |
ba180fd4 | 718 | if (proc_mm) { |
abaf6977 GS |
719 | err = ptrace(PTRACE_SWITCH_MM, userspace_pid[0], 0, |
720 | mm_idp->u.mm_fd); | |
3e6f2ac4 JD |
721 | if (err) { |
722 | printk(UM_KERN_ERR "__switch_mm - PTRACE_SWITCH_MM " | |
723 | "failed, errno = %d\n", errno); | |
724 | fatal_sigsegv(); | |
725 | } | |
abaf6977 GS |
726 | } |
727 | else userspace_pid[0] = mm_idp->u.pid; | |
728 | } |