Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
4 */
5
6#include <stdio.h>
7#include <stdlib.h>
8#include <stdarg.h>
9#include <unistd.h>
10#include <errno.h>
11#include <fcntl.h>
12#include <sched.h>
13#include <signal.h>
14#include <string.h>
15#include <sys/mman.h>
16#include <sys/stat.h>
17#include <sys/wait.h>
18#include <sys/time.h>
19#include <sys/resource.h>
20#include <asm/ldt.h>
21#include <asm/unistd.h>
22#include <init.h>
23#include <os.h>
24#include <kern_util.h>
25#include <mem_user.h>
26#include <ptrace_user.h>
27#include <registers.h>
28#include <skas.h>
29#include "internal.h"
30
31static void ptrace_child(void)
32{
33 int ret;
34 /* Calling os_getpid because some libcs cached getpid incorrectly */
35 int pid = os_getpid(), ppid = getppid();
36 int sc_result;
37
38 if (change_sig(SIGWINCH, 0) < 0 ||
39 ptrace(PTRACE_TRACEME, 0, 0, 0) < 0) {
40 perror("ptrace");
41 kill(pid, SIGKILL);
42 }
43 kill(pid, SIGSTOP);
44
45 /*
46 * This syscall will be intercepted by the parent. Don't call more than
47 * once, please.
48 */
49 sc_result = os_getpid();
50
51 if (sc_result == pid)
52 /* Nothing modified by the parent, we are running normally. */
53 ret = 1;
54 else if (sc_result == ppid)
55 /*
56 * Expected in check_ptrace and check_sysemu when they succeed
57 * in modifying the stack frame
58 */
59 ret = 0;
60 else
61 /* Serious trouble! This could be caused by a bug in host 2.6
62 * SKAS3/2.6 patch before release -V6, together with a bug in
63 * the UML code itself.
64 */
65 ret = 2;
66
67 exit(ret);
68}
69
70static void fatal_perror(const char *str)
71{
72 perror(str);
73 exit(1);
74}
75
76static void fatal(char *fmt, ...)
77{
78 va_list list;
79
80 va_start(list, fmt);
81 vfprintf(stderr, fmt, list);
82 va_end(list);
83
84 exit(1);
85}
86
87static void non_fatal(char *fmt, ...)
88{
89 va_list list;
90
91 va_start(list, fmt);
92 vfprintf(stderr, fmt, list);
93 va_end(list);
94}
95
96static int start_ptraced_child(void)
97{
98 int pid, n, status;
99
100 fflush(stdout);
101
102 pid = fork();
103 if (pid == 0)
104 ptrace_child();
105 else if (pid < 0)
106 fatal_perror("start_ptraced_child : fork failed");
107
108 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
109 if (n < 0)
110 fatal_perror("check_ptrace : waitpid failed");
111 if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
112 fatal("check_ptrace : expected SIGSTOP, got status = %d",
113 status);
114
115 return pid;
116}
117
118static void stop_ptraced_child(int pid, int exitcode)
119{
120 int status, n;
121
122 if (ptrace(PTRACE_CONT, pid, 0, 0) < 0)
123 fatal_perror("stop_ptraced_child : ptrace failed");
124
125 CATCH_EINTR(n = waitpid(pid, &status, 0));
126 if (!WIFEXITED(status) || (WEXITSTATUS(status) != exitcode)) {
127 int exit_with = WEXITSTATUS(status);
128 fatal("stop_ptraced_child : child exited with exitcode %d, "
129 "while expecting %d; status 0x%x\n", exit_with,
130 exitcode, status);
131 }
132}
133
134static void __init check_sysemu(void)
135{
136 int pid, n, status, count=0;
137
138 os_info("Checking syscall emulation for ptrace...");
139 pid = start_ptraced_child();
140
141 if ((ptrace(PTRACE_SETOPTIONS, pid, 0,
142 (void *) PTRACE_O_TRACESYSGOOD) < 0))
143 fatal_perror("check_sysemu: PTRACE_SETOPTIONS failed");
144
145 while (1) {
146 count++;
147 if (ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
148 goto fail;
149 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
150 if (n < 0)
151 fatal_perror("check_sysemu: wait failed");
152
153 if (WIFSTOPPED(status) &&
154 (WSTOPSIG(status) == (SIGTRAP|0x80))) {
155 if (!count) {
156 non_fatal("check_sysemu: SYSEMU_SINGLESTEP "
157 "doesn't singlestep");
158 goto fail;
159 }
160 n = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_RET_OFFSET,
161 os_getpid());
162 if (n < 0)
163 fatal_perror("check_sysemu : failed to modify "
164 "system call return");
165 break;
166 }
167 else if (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
168 count++;
169 else {
170 non_fatal("check_sysemu: expected SIGTRAP or "
171 "(SIGTRAP | 0x80), got status = %d\n",
172 status);
173 goto fail;
174 }
175 }
176 stop_ptraced_child(pid, 0);
177
178 os_info("OK\n");
179
180 return;
181
182fail:
183 stop_ptraced_child(pid, 1);
184 fatal("missing\n");
185}
186
187static void __init check_ptrace(void)
188{
189 int pid, syscall, n, status;
190
191 os_info("Checking that ptrace can change system call numbers...");
192 pid = start_ptraced_child();
193
194 if ((ptrace(PTRACE_SETOPTIONS, pid, 0,
195 (void *) PTRACE_O_TRACESYSGOOD) < 0))
196 fatal_perror("check_ptrace: PTRACE_SETOPTIONS failed");
197
198 while (1) {
199 if (ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
200 fatal_perror("check_ptrace : ptrace failed");
201
202 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
203 if (n < 0)
204 fatal_perror("check_ptrace : wait failed");
205
206 if (!WIFSTOPPED(status) ||
207 (WSTOPSIG(status) != (SIGTRAP | 0x80)))
208 fatal("check_ptrace : expected (SIGTRAP|0x80), "
209 "got status = %d", status);
210
211 syscall = ptrace(PTRACE_PEEKUSER, pid, PT_SYSCALL_NR_OFFSET,
212 0);
213 if (syscall == __NR_getpid) {
214 n = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
215 __NR_getppid);
216 if (n < 0)
217 fatal_perror("check_ptrace : failed to modify "
218 "system call");
219 break;
220 }
221 }
222 stop_ptraced_child(pid, 0);
223 os_info("OK\n");
224 check_sysemu();
225}
226
227static void __init check_coredump_limit(void)
228{
229 struct rlimit lim;
230 int err = getrlimit(RLIMIT_CORE, &lim);
231
232 if (err) {
233 perror("Getting core dump limit");
234 return;
235 }
236
237 os_info("Core dump limits :\n\tsoft - ");
238 if (lim.rlim_cur == RLIM_INFINITY)
239 os_info("NONE\n");
240 else
241 os_info("%llu\n", (unsigned long long)lim.rlim_cur);
242
243 os_info("\thard - ");
244 if (lim.rlim_max == RLIM_INFINITY)
245 os_info("NONE\n");
246 else
247 os_info("%llu\n", (unsigned long long)lim.rlim_max);
248}
249
250void __init get_host_cpu_features(
251 void (*flags_helper_func)(char *line),
252 void (*cache_helper_func)(char *line))
253{
254 FILE *cpuinfo;
255 char *line = NULL;
256 size_t len = 0;
257 int done_parsing = 0;
258
259 cpuinfo = fopen("/proc/cpuinfo", "r");
260 if (cpuinfo == NULL) {
261 os_info("Failed to get host CPU features\n");
262 } else {
263 while ((getline(&line, &len, cpuinfo)) != -1) {
264 if (strstr(line, "flags")) {
265 flags_helper_func(line);
266 done_parsing++;
267 }
268 if (strstr(line, "cache_alignment")) {
269 cache_helper_func(line);
270 done_parsing++;
271 }
272 free(line);
273 line = NULL;
274 if (done_parsing > 1)
275 break;
276 }
277 fclose(cpuinfo);
278 }
279}
280
281
282void __init os_early_checks(void)
283{
284 int pid;
285
286 /* Print out the core dump limits early */
287 check_coredump_limit();
288
289 check_ptrace();
290
291 /* Need to check this early because mmapping happens before the
292 * kernel is running.
293 */
294 check_tmpexec();
295
296 pid = start_ptraced_child();
297 if (init_pid_registers(pid))
298 fatal("Failed to initialize default registers");
299 stop_ptraced_child(pid, 1);
300}
301
302int __init parse_iomem(char *str, int *add)
303{
304 struct iomem_region *new;
305 struct stat64 buf;
306 char *file, *driver;
307 int fd, size;
308
309 driver = str;
310 file = strchr(str,',');
311 if (file == NULL) {
312 os_warn("parse_iomem : failed to parse iomem\n");
313 goto out;
314 }
315 *file = '\0';
316 file++;
317 fd = open(file, O_RDWR, 0);
318 if (fd < 0) {
319 perror("parse_iomem - Couldn't open io file");
320 goto out;
321 }
322
323 if (fstat64(fd, &buf) < 0) {
324 perror("parse_iomem - cannot stat_fd file");
325 goto out_close;
326 }
327
328 new = malloc(sizeof(*new));
329 if (new == NULL) {
330 perror("Couldn't allocate iomem_region struct");
331 goto out_close;
332 }
333
334 size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
335
336 *new = ((struct iomem_region) { .next = iomem_regions,
337 .driver = driver,
338 .fd = fd,
339 .size = size,
340 .phys = 0,
341 .virt = 0 });
342 iomem_regions = new;
343 iomem_size += new->size + UM_KERN_PAGE_SIZE;
344
345 return 0;
346 out_close:
347 close(fd);
348 out:
349 return 1;
350}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
4 */
5
6#include <stdio.h>
7#include <stdlib.h>
8#include <stdarg.h>
9#include <unistd.h>
10#include <errno.h>
11#include <fcntl.h>
12#include <sched.h>
13#include <signal.h>
14#include <string.h>
15#include <sys/mman.h>
16#include <sys/stat.h>
17#include <sys/wait.h>
18#include <sys/time.h>
19#include <sys/resource.h>
20#include <asm/unistd.h>
21#include <init.h>
22#include <os.h>
23#include <mem_user.h>
24#include <ptrace_user.h>
25#include <registers.h>
26#include <skas.h>
27
28static void ptrace_child(void)
29{
30 int ret;
31 /* Calling os_getpid because some libcs cached getpid incorrectly */
32 int pid = os_getpid(), ppid = getppid();
33 int sc_result;
34
35 if (change_sig(SIGWINCH, 0) < 0 ||
36 ptrace(PTRACE_TRACEME, 0, 0, 0) < 0) {
37 perror("ptrace");
38 kill(pid, SIGKILL);
39 }
40 kill(pid, SIGSTOP);
41
42 /*
43 * This syscall will be intercepted by the parent. Don't call more than
44 * once, please.
45 */
46 sc_result = os_getpid();
47
48 if (sc_result == pid)
49 /* Nothing modified by the parent, we are running normally. */
50 ret = 1;
51 else if (sc_result == ppid)
52 /*
53 * Expected in check_ptrace and check_sysemu when they succeed
54 * in modifying the stack frame
55 */
56 ret = 0;
57 else
58 /* Serious trouble! This could be caused by a bug in host 2.6
59 * SKAS3/2.6 patch before release -V6, together with a bug in
60 * the UML code itself.
61 */
62 ret = 2;
63
64 exit(ret);
65}
66
67static void fatal_perror(const char *str)
68{
69 perror(str);
70 exit(1);
71}
72
73static void fatal(char *fmt, ...)
74{
75 va_list list;
76
77 va_start(list, fmt);
78 vfprintf(stderr, fmt, list);
79 va_end(list);
80
81 exit(1);
82}
83
84static void non_fatal(char *fmt, ...)
85{
86 va_list list;
87
88 va_start(list, fmt);
89 vfprintf(stderr, fmt, list);
90 va_end(list);
91}
92
93static int start_ptraced_child(void)
94{
95 int pid, n, status;
96
97 fflush(stdout);
98
99 pid = fork();
100 if (pid == 0)
101 ptrace_child();
102 else if (pid < 0)
103 fatal_perror("start_ptraced_child : fork failed");
104
105 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
106 if (n < 0)
107 fatal_perror("check_ptrace : waitpid failed");
108 if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
109 fatal("check_ptrace : expected SIGSTOP, got status = %d",
110 status);
111
112 return pid;
113}
114
115/* When testing for SYSEMU support, if it is one of the broken versions, we
116 * must just avoid using sysemu, not panic, but only if SYSEMU features are
117 * broken.
118 * So only for SYSEMU features we test mustpanic, while normal host features
119 * must work anyway!
120 */
121static int stop_ptraced_child(int pid, int exitcode, int mustexit)
122{
123 int status, n, ret = 0;
124
125 if (ptrace(PTRACE_CONT, pid, 0, 0) < 0) {
126 perror("stop_ptraced_child : ptrace failed");
127 return -1;
128 }
129 CATCH_EINTR(n = waitpid(pid, &status, 0));
130 if (!WIFEXITED(status) || (WEXITSTATUS(status) != exitcode)) {
131 int exit_with = WEXITSTATUS(status);
132 if (exit_with == 2)
133 non_fatal("check_ptrace : child exited with status 2. "
134 "\nDisabling SYSEMU support.\n");
135 non_fatal("check_ptrace : child exited with exitcode %d, while "
136 "expecting %d; status 0x%x\n", exit_with,
137 exitcode, status);
138 if (mustexit)
139 exit(1);
140 ret = -1;
141 }
142
143 return ret;
144}
145
146/* Changed only during early boot */
147static int force_sysemu_disabled = 0;
148
149static int __init nosysemu_cmd_param(char *str, int* add)
150{
151 force_sysemu_disabled = 1;
152 return 0;
153}
154
155__uml_setup("nosysemu", nosysemu_cmd_param,
156"nosysemu\n"
157" Turns off syscall emulation patch for ptrace (SYSEMU).\n"
158" SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
159" behaviour of ptrace() and helps reduce host context switch rates.\n"
160" To make it work, you need a kernel patch for your host, too.\n"
161" See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
162" information.\n\n");
163
164static void __init check_sysemu(void)
165{
166 unsigned long regs[MAX_REG_NR];
167 int pid, n, status, count=0;
168
169 os_info("Checking syscall emulation patch for ptrace...");
170 sysemu_supported = 0;
171 pid = start_ptraced_child();
172
173 if (ptrace(PTRACE_SYSEMU, pid, 0, 0) < 0)
174 goto fail;
175
176 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
177 if (n < 0)
178 fatal_perror("check_sysemu : wait failed");
179 if (!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGTRAP))
180 fatal("check_sysemu : expected SIGTRAP, got status = %d\n",
181 status);
182
183 if (ptrace(PTRACE_GETREGS, pid, 0, regs) < 0)
184 fatal_perror("check_sysemu : PTRACE_GETREGS failed");
185 if (PT_SYSCALL_NR(regs) != __NR_getpid) {
186 non_fatal("check_sysemu got system call number %d, "
187 "expected %d...", PT_SYSCALL_NR(regs), __NR_getpid);
188 goto fail;
189 }
190
191 n = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_RET_OFFSET, os_getpid());
192 if (n < 0) {
193 non_fatal("check_sysemu : failed to modify system call "
194 "return");
195 goto fail;
196 }
197
198 if (stop_ptraced_child(pid, 0, 0) < 0)
199 goto fail_stopped;
200
201 sysemu_supported = 1;
202 os_info("OK\n");
203 set_using_sysemu(!force_sysemu_disabled);
204
205 os_info("Checking advanced syscall emulation patch for ptrace...");
206 pid = start_ptraced_child();
207
208 if ((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
209 (void *) PTRACE_O_TRACESYSGOOD) < 0))
210 fatal_perror("check_sysemu: PTRACE_OLDSETOPTIONS failed");
211
212 while (1) {
213 count++;
214 if (ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
215 goto fail;
216 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
217 if (n < 0)
218 fatal_perror("check_sysemu: wait failed");
219
220 if (WIFSTOPPED(status) &&
221 (WSTOPSIG(status) == (SIGTRAP|0x80))) {
222 if (!count) {
223 non_fatal("check_sysemu: SYSEMU_SINGLESTEP "
224 "doesn't singlestep");
225 goto fail;
226 }
227 n = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_RET_OFFSET,
228 os_getpid());
229 if (n < 0)
230 fatal_perror("check_sysemu : failed to modify "
231 "system call return");
232 break;
233 }
234 else if (WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
235 count++;
236 else {
237 non_fatal("check_sysemu: expected SIGTRAP or "
238 "(SIGTRAP | 0x80), got status = %d\n",
239 status);
240 goto fail;
241 }
242 }
243 if (stop_ptraced_child(pid, 0, 0) < 0)
244 goto fail_stopped;
245
246 sysemu_supported = 2;
247 os_info("OK\n");
248
249 if (!force_sysemu_disabled)
250 set_using_sysemu(sysemu_supported);
251 return;
252
253fail:
254 stop_ptraced_child(pid, 1, 0);
255fail_stopped:
256 non_fatal("missing\n");
257}
258
259static void __init check_ptrace(void)
260{
261 int pid, syscall, n, status;
262
263 os_info("Checking that ptrace can change system call numbers...");
264 pid = start_ptraced_child();
265
266 if ((ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
267 (void *) PTRACE_O_TRACESYSGOOD) < 0))
268 fatal_perror("check_ptrace: PTRACE_OLDSETOPTIONS failed");
269
270 while (1) {
271 if (ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
272 fatal_perror("check_ptrace : ptrace failed");
273
274 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
275 if (n < 0)
276 fatal_perror("check_ptrace : wait failed");
277
278 if (!WIFSTOPPED(status) ||
279 (WSTOPSIG(status) != (SIGTRAP | 0x80)))
280 fatal("check_ptrace : expected (SIGTRAP|0x80), "
281 "got status = %d", status);
282
283 syscall = ptrace(PTRACE_PEEKUSER, pid, PT_SYSCALL_NR_OFFSET,
284 0);
285 if (syscall == __NR_getpid) {
286 n = ptrace(PTRACE_POKEUSER, pid, PT_SYSCALL_NR_OFFSET,
287 __NR_getppid);
288 if (n < 0)
289 fatal_perror("check_ptrace : failed to modify "
290 "system call");
291 break;
292 }
293 }
294 stop_ptraced_child(pid, 0, 1);
295 os_info("OK\n");
296 check_sysemu();
297}
298
299extern void check_tmpexec(void);
300
301static void __init check_coredump_limit(void)
302{
303 struct rlimit lim;
304 int err = getrlimit(RLIMIT_CORE, &lim);
305
306 if (err) {
307 perror("Getting core dump limit");
308 return;
309 }
310
311 os_info("Core dump limits :\n\tsoft - ");
312 if (lim.rlim_cur == RLIM_INFINITY)
313 os_info("NONE\n");
314 else
315 os_info("%llu\n", (unsigned long long)lim.rlim_cur);
316
317 os_info("\thard - ");
318 if (lim.rlim_max == RLIM_INFINITY)
319 os_info("NONE\n");
320 else
321 os_info("%llu\n", (unsigned long long)lim.rlim_max);
322}
323
324void __init get_host_cpu_features(
325 void (*flags_helper_func)(char *line),
326 void (*cache_helper_func)(char *line))
327{
328 FILE *cpuinfo;
329 char *line = NULL;
330 size_t len = 0;
331 int done_parsing = 0;
332
333 cpuinfo = fopen("/proc/cpuinfo", "r");
334 if (cpuinfo == NULL) {
335 os_info("Failed to get host CPU features\n");
336 } else {
337 while ((getline(&line, &len, cpuinfo)) != -1) {
338 if (strstr(line, "flags")) {
339 flags_helper_func(line);
340 done_parsing++;
341 }
342 if (strstr(line, "cache_alignment")) {
343 cache_helper_func(line);
344 done_parsing++;
345 }
346 free(line);
347 line = NULL;
348 if (done_parsing > 1)
349 break;
350 }
351 fclose(cpuinfo);
352 }
353}
354
355
356void __init os_early_checks(void)
357{
358 int pid;
359
360 /* Print out the core dump limits early */
361 check_coredump_limit();
362
363 check_ptrace();
364
365 /* Need to check this early because mmapping happens before the
366 * kernel is running.
367 */
368 check_tmpexec();
369
370 pid = start_ptraced_child();
371 if (init_pid_registers(pid))
372 fatal("Failed to initialize default registers");
373 stop_ptraced_child(pid, 1, 1);
374}
375
376int __init parse_iomem(char *str, int *add)
377{
378 struct iomem_region *new;
379 struct stat64 buf;
380 char *file, *driver;
381 int fd, size;
382
383 driver = str;
384 file = strchr(str,',');
385 if (file == NULL) {
386 os_warn("parse_iomem : failed to parse iomem\n");
387 goto out;
388 }
389 *file = '\0';
390 file++;
391 fd = open(file, O_RDWR, 0);
392 if (fd < 0) {
393 perror("parse_iomem - Couldn't open io file");
394 goto out;
395 }
396
397 if (fstat64(fd, &buf) < 0) {
398 perror("parse_iomem - cannot stat_fd file");
399 goto out_close;
400 }
401
402 new = malloc(sizeof(*new));
403 if (new == NULL) {
404 perror("Couldn't allocate iomem_region struct");
405 goto out_close;
406 }
407
408 size = (buf.st_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
409
410 *new = ((struct iomem_region) { .next = iomem_regions,
411 .driver = driver,
412 .fd = fd,
413 .size = size,
414 .phys = 0,
415 .virt = 0 });
416 iomem_regions = new;
417 iomem_size += new->size + UM_KERN_PAGE_SIZE;
418
419 return 0;
420 out_close:
421 close(fd);
422 out:
423 return 1;
424}