Merge branch 'srp' of master.kernel.org:/pub/scm/linux/kernel/git/roland/infiniband
[linux-2.6] / arch / um / os-Linux / start_up.c
1 /*
2  * Copyright (C) 2000, 2001, 2002 Jeff Dike (jdike@karaya.com)
3  * Licensed under the GPL
4  */
5
6 #include <stdio.h>
7 #include <stddef.h>
8 #include <stdarg.h>
9 #include <stdlib.h>
10 #include <string.h>
11 #include <unistd.h>
12 #include <signal.h>
13 #include <sched.h>
14 #include <fcntl.h>
15 #include <errno.h>
16 #include <setjmp.h>
17 #include <sys/time.h>
18 #include <sys/wait.h>
19 #include <sys/mman.h>
20 #include <asm/unistd.h>
21 #include <asm/page.h>
22 #include <sys/types.h>
23 #include "user_util.h"
24 #include "kern_util.h"
25 #include "user.h"
26 #include "signal_kern.h"
27 #include "signal_user.h"
28 #include "sysdep/ptrace.h"
29 #include "sysdep/sigcontext.h"
30 #include "irq_user.h"
31 #include "ptrace_user.h"
32 #include "mem_user.h"
33 #include "time_user.h"
34 #include "init.h"
35 #include "os.h"
36 #include "uml-config.h"
37 #include "choose-mode.h"
38 #include "mode.h"
39 #include "tempfile.h"
40 #include "kern_constants.h"
41
42 #ifdef UML_CONFIG_MODE_SKAS
43 #include "skas.h"
44 #include "skas_ptrace.h"
45 #include "registers.h"
46 #endif
47
48 static int ptrace_child(void *arg)
49 {
50         int ret;
51         int pid = os_getpid(), ppid = getppid();
52         int sc_result;
53
54         if(ptrace(PTRACE_TRACEME, 0, 0, 0) < 0){
55                 perror("ptrace");
56                 os_kill_process(pid, 0);
57         }
58         os_stop_process(pid);
59
60         /*This syscall will be intercepted by the parent. Don't call more than
61          * once, please.*/
62         sc_result = os_getpid();
63
64         if (sc_result == pid)
65                 ret = 1; /*Nothing modified by the parent, we are running
66                            normally.*/
67         else if (sc_result == ppid)
68                 ret = 0; /*Expected in check_ptrace and check_sysemu when they
69                            succeed in modifying the stack frame*/
70         else
71                 ret = 2; /*Serious trouble! This could be caused by a bug in
72                            host 2.6 SKAS3/2.6 patch before release -V6, together
73                            with a bug in the UML code itself.*/
74         _exit(ret);
75 }
76
77 static int start_ptraced_child(void **stack_out)
78 {
79         void *stack;
80         unsigned long sp;
81         int pid, n, status;
82
83         stack = mmap(NULL, PAGE_SIZE, PROT_READ | PROT_WRITE | PROT_EXEC,
84                      MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
85         if(stack == MAP_FAILED)
86                 panic("check_ptrace : mmap failed, errno = %d", errno);
87         sp = (unsigned long) stack + PAGE_SIZE - sizeof(void *);
88         pid = clone(ptrace_child, (void *) sp, SIGCHLD, NULL);
89         if(pid < 0)
90                 panic("start_ptraced_child : clone failed, errno = %d", errno);
91         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
92         if(n < 0)
93                 panic("check_ptrace : clone failed, errno = %d", errno);
94         if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGSTOP))
95                 panic("check_ptrace : expected SIGSTOP, got status = %d",
96                       status);
97
98         *stack_out = stack;
99         return(pid);
100 }
101
102 /* When testing for SYSEMU support, if it is one of the broken versions, we
103  * must just avoid using sysemu, not panic, but only if SYSEMU features are
104  * broken.
105  * So only for SYSEMU features we test mustpanic, while normal host features
106  * must work anyway!
107  */
108 static int stop_ptraced_child(int pid, void *stack, int exitcode,
109                               int mustpanic)
110 {
111         int status, n, ret = 0;
112
113         if(ptrace(PTRACE_CONT, pid, 0, 0) < 0)
114                 panic("check_ptrace : ptrace failed, errno = %d", errno);
115         CATCH_EINTR(n = waitpid(pid, &status, 0));
116         if(!WIFEXITED(status) || (WEXITSTATUS(status) != exitcode)) {
117                 int exit_with = WEXITSTATUS(status);
118                 if (exit_with == 2)
119                         printk("check_ptrace : child exited with status 2. "
120                                "Serious trouble happening! Try updating your "
121                                "host skas patch!\nDisabling SYSEMU support.");
122                 printk("check_ptrace : child exited with exitcode %d, while "
123                       "expecting %d; status 0x%x", exit_with,
124                       exitcode, status);
125                 if (mustpanic)
126                         panic("\n");
127                 else
128                         printk("\n");
129                 ret = -1;
130         }
131
132         if(munmap(stack, PAGE_SIZE) < 0)
133                 panic("check_ptrace : munmap failed, errno = %d", errno);
134         return ret;
135 }
136
137 int ptrace_faultinfo = 1;
138 int proc_mm = 1;
139
140 static int __init skas0_cmd_param(char *str, int* add)
141 {
142         ptrace_faultinfo = proc_mm = 0;
143         return 0;
144 }
145
146 /* The two __uml_setup would conflict, without this stupid alias. */
147
148 static int __init mode_skas0_cmd_param(char *str, int* add)
149         __attribute__((alias("skas0_cmd_param")));
150
151 __uml_setup("skas0", skas0_cmd_param,
152                 "skas0\n"
153                 "    Disables SKAS3 usage, so that SKAS0 is used, unless \n"
154                 "    you specify mode=tt.\n\n");
155
156 __uml_setup("mode=skas0", mode_skas0_cmd_param,
157                 "mode=skas0\n"
158                 "    Disables SKAS3 usage, so that SKAS0 is used, unless you \n"
159                 "    specify mode=tt. Note that this was recently added - on \n"
160                 "    older kernels you must use simply \"skas0\".\n\n");
161
162 static int force_sysemu_disabled = 0;
163
164 static int __init nosysemu_cmd_param(char *str, int* add)
165 {
166         force_sysemu_disabled = 1;
167         return 0;
168 }
169
170 __uml_setup("nosysemu", nosysemu_cmd_param,
171 "nosysemu\n"
172 "    Turns off syscall emulation patch for ptrace (SYSEMU) on.\n"
173 "    SYSEMU is a performance-patch introduced by Laurent Vivier. It changes\n"
174 "    behaviour of ptrace() and helps reducing host context switch rate.\n"
175 "    To make it working, you need a kernel patch for your host, too.\n"
176 "    See http://perso.wanadoo.fr/laurent.vivier/UML/ for further \n"
177 "    information.\n\n");
178
179 static void __init check_sysemu(void)
180 {
181         void *stack;
182         int pid, n, status, count=0;
183
184         printk("Checking syscall emulation patch for ptrace...");
185         sysemu_supported = 0;
186         pid = start_ptraced_child(&stack);
187
188         if(ptrace(PTRACE_SYSEMU, pid, 0, 0) < 0)
189                 goto fail;
190
191         CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
192         if (n < 0)
193                 panic("check_sysemu : wait failed, errno = %d", errno);
194         if(!WIFSTOPPED(status) || (WSTOPSIG(status) != SIGTRAP))
195                 panic("check_sysemu : expected SIGTRAP, "
196                       "got status = %d", status);
197
198         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
199                    os_getpid());
200         if(n < 0)
201                 panic("check_sysemu : failed to modify system "
202                       "call return, errno = %d", errno);
203
204         if (stop_ptraced_child(pid, stack, 0, 0) < 0)
205                 goto fail_stopped;
206
207         sysemu_supported = 1;
208         printk("OK\n");
209         set_using_sysemu(!force_sysemu_disabled);
210
211         printk("Checking advanced syscall emulation patch for ptrace...");
212         pid = start_ptraced_child(&stack);
213
214         if(ptrace(PTRACE_OLDSETOPTIONS, pid, 0,
215                   (void *) PTRACE_O_TRACESYSGOOD) < 0)
216                 panic("check_ptrace: PTRACE_OLDSETOPTIONS failed, errno = %d",
217                       errno);
218
219         while(1){
220                 count++;
221                 if(ptrace(PTRACE_SYSEMU_SINGLESTEP, pid, 0, 0) < 0)
222                         goto fail;
223                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
224                 if(n < 0)
225                         panic("check_ptrace : wait failed, errno = %d", errno);
226                 if(WIFSTOPPED(status) && (WSTOPSIG(status) == (SIGTRAP|0x80))){
227                         if (!count)
228                                 panic("check_ptrace : SYSEMU_SINGLESTEP "
229                                       "doesn't singlestep");
230                         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_RET_OFFSET,
231                                    os_getpid());
232                         if(n < 0)
233                                 panic("check_sysemu : failed to modify system "
234                                       "call return, errno = %d", errno);
235                         break;
236                 }
237                 else if(WIFSTOPPED(status) && (WSTOPSIG(status) == SIGTRAP))
238                         count++;
239                 else
240                         panic("check_ptrace : expected SIGTRAP or "
241                               "(SIGTRAP|0x80), got status = %d", status);
242         }
243         if (stop_ptraced_child(pid, stack, 0, 0) < 0)
244                 goto fail_stopped;
245
246         sysemu_supported = 2;
247         printk("OK\n");
248
249         if ( !force_sysemu_disabled )
250                 set_using_sysemu(sysemu_supported);
251         return;
252
253 fail:
254         stop_ptraced_child(pid, stack, 1, 0);
255 fail_stopped:
256         printk("missing\n");
257 }
258
259 static void __init check_ptrace(void)
260 {
261         void *stack;
262         int pid, syscall, n, status;
263
264         printk("Checking that ptrace can change system call numbers...");
265         pid = start_ptraced_child(&stack);
266
267         if(ptrace(PTRACE_OLDSETOPTIONS, pid, 0, (void *)PTRACE_O_TRACESYSGOOD) < 0)
268                 panic("check_ptrace: PTRACE_OLDSETOPTIONS failed, errno = %d", errno);
269
270         while(1){
271                 if(ptrace(PTRACE_SYSCALL, pid, 0, 0) < 0)
272                         panic("check_ptrace : ptrace failed, errno = %d",
273                               errno);
274                 CATCH_EINTR(n = waitpid(pid, &status, WUNTRACED));
275                 if(n < 0)
276                         panic("check_ptrace : wait failed, errno = %d", errno);
277                 if(!WIFSTOPPED(status) || (WSTOPSIG(status) != (SIGTRAP|0x80)))
278                         panic("check_ptrace : expected (SIGTRAP|0x80), "
279                               "got status = %d", status);
280
281                 syscall = ptrace(PTRACE_PEEKUSR, pid, PT_SYSCALL_NR_OFFSET,
282                                  0);
283                 if(syscall == __NR_getpid){
284                         n = ptrace(PTRACE_POKEUSR, pid, PT_SYSCALL_NR_OFFSET,
285                                    __NR_getppid);
286                         if(n < 0)
287                                 panic("check_ptrace : failed to modify system "
288                                       "call, errno = %d", errno);
289                         break;
290                 }
291         }
292         stop_ptraced_child(pid, stack, 0, 1);
293         printk("OK\n");
294         check_sysemu();
295 }
296
297 extern int create_tmp_file(unsigned long len);
298
299 static void check_tmpexec(void)
300 {
301         void *addr;
302         int err, fd = create_tmp_file(UM_KERN_PAGE_SIZE);
303
304         addr = mmap(NULL, UM_KERN_PAGE_SIZE,
305                     PROT_READ | PROT_WRITE | PROT_EXEC, MAP_PRIVATE, fd, 0);
306         printf("Checking PROT_EXEC mmap in /tmp...");
307         fflush(stdout);
308         if(addr == MAP_FAILED){
309                 err = errno;
310                 perror("failed");
311                 if(err == EPERM)
312                         printf("/tmp must be not mounted noexec\n");
313                 exit(1);
314         }
315         printf("OK\n");
316         munmap(addr, UM_KERN_PAGE_SIZE);
317
318         close(fd);
319 }
320
321 void os_early_checks(void)
322 {
323         check_ptrace();
324
325         /* Need to check this early because mmapping happens before the
326          * kernel is running.
327          */
328         check_tmpexec();
329 }
330
331 static int __init noprocmm_cmd_param(char *str, int* add)
332 {
333         proc_mm = 0;
334         return 0;
335 }
336
337 __uml_setup("noprocmm", noprocmm_cmd_param,
338 "noprocmm\n"
339 "    Turns off usage of /proc/mm, even if host supports it.\n"
340 "    To support /proc/mm, the host needs to be patched using\n"
341 "    the current skas3 patch.\n\n");
342
343 static int __init noptracefaultinfo_cmd_param(char *str, int* add)
344 {
345         ptrace_faultinfo = 0;
346         return 0;
347 }
348
349 __uml_setup("noptracefaultinfo", noptracefaultinfo_cmd_param,
350 "noptracefaultinfo\n"
351 "    Turns off usage of PTRACE_FAULTINFO, even if host supports\n"
352 "    it. To support PTRACE_FAULTINFO, the host needs to be patched\n"
353 "    using the current skas3 patch.\n\n");
354
355 #ifdef UML_CONFIG_MODE_SKAS
356 static inline void check_skas3_ptrace_support(void)
357 {
358         struct ptrace_faultinfo fi;
359         void *stack;
360         int pid, n;
361
362         printf("Checking for the skas3 patch in the host...");
363         pid = start_ptraced_child(&stack);
364
365         n = ptrace(PTRACE_FAULTINFO, pid, 0, &fi);
366         if (n < 0) {
367                 ptrace_faultinfo = 0;
368                 if(errno == EIO)
369                         printf("not found\n");
370                 else
371                         perror("not found");
372         }
373         else {
374                 if (!ptrace_faultinfo)
375                         printf("found but disabled on command line\n");
376                 else
377                         printf("found\n");
378         }
379
380         init_registers(pid);
381         stop_ptraced_child(pid, stack, 1, 1);
382 }
383
384 int can_do_skas(void)
385 {
386         printf("Checking for /proc/mm...");
387         if (os_access("/proc/mm", OS_ACC_W_OK) < 0) {
388                 proc_mm = 0;
389                 printf("not found\n");
390         }
391         else {
392                 if (!proc_mm)
393                         printf("found but disabled on command line\n");
394                 else
395                         printf("found\n");
396         }
397
398         check_skas3_ptrace_support();
399         return 1;
400 }
401 #else
402 int can_do_skas(void)
403 {
404         return(0);
405 }
406 #endif
407
408 int have_devanon = 0;
409
410 void check_devanon(void)
411 {
412         int fd;
413
414         printk("Checking for /dev/anon on the host...");
415         fd = open("/dev/anon", O_RDWR);
416         if(fd < 0){
417                 printk("Not available (open failed with errno %d)\n", errno);
418                 return;
419         }
420
421         printk("OK\n");
422         have_devanon = 1;
423 }
424
425 int __init parse_iomem(char *str, int *add)
426 {
427         struct iomem_region *new;
428         struct uml_stat buf;
429         char *file, *driver;
430         int fd, err, size;
431
432         driver = str;
433         file = strchr(str,',');
434         if(file == NULL){
435                 printf("parse_iomem : failed to parse iomem\n");
436                 goto out;
437         }
438         *file = '\0';
439         file++;
440         fd = os_open_file(file, of_rdwr(OPENFLAGS()), 0);
441         if(fd < 0){
442                 os_print_error(fd, "parse_iomem - Couldn't open io file");
443                 goto out;
444         }
445
446         err = os_stat_fd(fd, &buf);
447         if(err < 0){
448                 os_print_error(err, "parse_iomem - cannot stat_fd file");
449                 goto out_close;
450         }
451
452         new = malloc(sizeof(*new));
453         if(new == NULL){
454                 perror("Couldn't allocate iomem_region struct");
455                 goto out_close;
456         }
457
458         size = (buf.ust_size + UM_KERN_PAGE_SIZE) & ~(UM_KERN_PAGE_SIZE - 1);
459
460         *new = ((struct iomem_region) { .next           = iomem_regions,
461                                         .driver         = driver,
462                                         .fd             = fd,
463                                         .size           = size,
464                                         .phys           = 0,
465                                         .virt           = 0 });
466         iomem_regions = new;
467         iomem_size += new->size + UM_KERN_PAGE_SIZE;
468
469         return(0);
470  out_close:
471         os_close_file(fd);
472  out:
473         return(1);
474 }
475