152c653b3SJeff Dike /* 252c653b3SJeff Dike * Copyright (C) 2000, 2001 Jeff Dike (jdike@karaya.com) 352c653b3SJeff Dike * Licensed under the GPL 452c653b3SJeff Dike */ 552c653b3SJeff Dike 652c653b3SJeff Dike #include <unistd.h> 752c653b3SJeff Dike #include <stdio.h> 852c653b3SJeff Dike #include <stdlib.h> 952c653b3SJeff Dike #include <string.h> 1052c653b3SJeff Dike #include <signal.h> 1152c653b3SJeff Dike #include <errno.h> 1252c653b3SJeff Dike #include <sys/resource.h> 1352c653b3SJeff Dike #include <sys/mman.h> 1452c653b3SJeff Dike #include <sys/user.h> 1552c653b3SJeff Dike #include <asm/page.h> 1652c653b3SJeff Dike #include "user_util.h" 1752c653b3SJeff Dike #include "kern_util.h" 1852c653b3SJeff Dike #include "mem_user.h" 1952c653b3SJeff Dike #include "irq_user.h" 2052c653b3SJeff Dike #include "user.h" 2152c653b3SJeff Dike #include "init.h" 2252c653b3SJeff Dike #include "mode.h" 2352c653b3SJeff Dike #include "choose-mode.h" 2452c653b3SJeff Dike #include "uml-config.h" 2552c653b3SJeff Dike #include "os.h" 26c13e5690SPaolo 'Blaisorblade' Giarrusso #include "um_malloc.h" 2752c653b3SJeff Dike 2811100b1dSJeff Dike /* Set in main, unchanged thereafter */ 2952c653b3SJeff Dike char *linux_prog; 3052c653b3SJeff Dike 3152c653b3SJeff Dike #define PGD_BOUND (4 * 1024 * 1024) 3252c653b3SJeff Dike #define STACKSIZE (8 * 1024 * 1024) 3352c653b3SJeff Dike #define THREAD_NAME_LEN (256) 3452c653b3SJeff Dike 3552c653b3SJeff Dike static void set_stklim(void) 3652c653b3SJeff Dike { 3752c653b3SJeff Dike struct rlimit lim; 3852c653b3SJeff Dike 3952c653b3SJeff Dike if(getrlimit(RLIMIT_STACK, &lim) < 0){ 4052c653b3SJeff Dike perror("getrlimit"); 4152c653b3SJeff Dike exit(1); 4252c653b3SJeff Dike } 4352c653b3SJeff Dike if((lim.rlim_cur == RLIM_INFINITY) || (lim.rlim_cur > STACKSIZE)){ 4452c653b3SJeff Dike lim.rlim_cur = STACKSIZE; 4552c653b3SJeff Dike if(setrlimit(RLIMIT_STACK, &lim) < 0){ 4652c653b3SJeff Dike perror("setrlimit"); 4752c653b3SJeff Dike exit(1); 4852c653b3SJeff Dike } 4952c653b3SJeff Dike } 5052c653b3SJeff Dike } 5152c653b3SJeff Dike 5252c653b3SJeff Dike static __init void do_uml_initcalls(void) 5352c653b3SJeff Dike { 5452c653b3SJeff Dike initcall_t *call; 5552c653b3SJeff Dike 5652c653b3SJeff Dike call = &__uml_initcall_start; 57f2183125SJeff Dike while (call < &__uml_initcall_end){ 5852c653b3SJeff Dike (*call)(); 5952c653b3SJeff Dike call++; 6052c653b3SJeff Dike } 6152c653b3SJeff Dike } 6252c653b3SJeff Dike 6352c653b3SJeff Dike static void last_ditch_exit(int sig) 6452c653b3SJeff Dike { 6552c653b3SJeff Dike uml_cleanup(); 6652c653b3SJeff Dike exit(1); 6752c653b3SJeff Dike } 6852c653b3SJeff Dike 694b84c69bSJeff Dike static void install_fatal_handler(int sig) 704b84c69bSJeff Dike { 714b84c69bSJeff Dike struct sigaction action; 724b84c69bSJeff Dike 734b84c69bSJeff Dike /* All signals are enabled in this handler ... */ 744b84c69bSJeff Dike sigemptyset(&action.sa_mask); 754b84c69bSJeff Dike 764b84c69bSJeff Dike /* ... including the signal being handled, plus we want the 774b84c69bSJeff Dike * handler reset to the default behavior, so that if an exit 784b84c69bSJeff Dike * handler is hanging for some reason, the UML will just die 794b84c69bSJeff Dike * after this signal is sent a second time. 804b84c69bSJeff Dike */ 814b84c69bSJeff Dike action.sa_flags = SA_RESETHAND | SA_NODEFER; 824b84c69bSJeff Dike action.sa_restorer = NULL; 834b84c69bSJeff Dike action.sa_handler = last_ditch_exit; 844b84c69bSJeff Dike if(sigaction(sig, &action, NULL) < 0){ 854b84c69bSJeff Dike printf("failed to install handler for signal %d - errno = %d\n", 864b84c69bSJeff Dike errno); 874b84c69bSJeff Dike exit(1); 884b84c69bSJeff Dike } 894b84c69bSJeff Dike } 904b84c69bSJeff Dike 91cb98cdcdSMattia Dongili #define UML_LIB_PATH ":/usr/lib/uml" 92cb98cdcdSMattia Dongili 93cb98cdcdSMattia Dongili static void setup_env_path(void) 94cb98cdcdSMattia Dongili { 95cb98cdcdSMattia Dongili char *new_path = NULL; 96cb98cdcdSMattia Dongili char *old_path = NULL; 97cb98cdcdSMattia Dongili int path_len = 0; 98cb98cdcdSMattia Dongili 99cb98cdcdSMattia Dongili old_path = getenv("PATH"); 100cb98cdcdSMattia Dongili /* if no PATH variable is set or it has an empty value 101cb98cdcdSMattia Dongili * just use the default + /usr/lib/uml 102cb98cdcdSMattia Dongili */ 103cb98cdcdSMattia Dongili if (!old_path || (path_len = strlen(old_path)) == 0) { 104cb98cdcdSMattia Dongili putenv("PATH=:/bin:/usr/bin/" UML_LIB_PATH); 105cb98cdcdSMattia Dongili return; 106cb98cdcdSMattia Dongili } 107cb98cdcdSMattia Dongili 108cb98cdcdSMattia Dongili /* append /usr/lib/uml to the existing path */ 109cb98cdcdSMattia Dongili path_len += strlen("PATH=" UML_LIB_PATH) + 1; 110cb98cdcdSMattia Dongili new_path = malloc(path_len); 111cb98cdcdSMattia Dongili if (!new_path) { 112cb98cdcdSMattia Dongili perror("coudn't malloc to set a new PATH"); 113cb98cdcdSMattia Dongili return; 114cb98cdcdSMattia Dongili } 115cb98cdcdSMattia Dongili snprintf(new_path, path_len, "PATH=%s" UML_LIB_PATH, old_path); 116cb98cdcdSMattia Dongili putenv(new_path); 117cb98cdcdSMattia Dongili } 118cb98cdcdSMattia Dongili 11952c653b3SJeff Dike extern int uml_exitcode; 12052c653b3SJeff Dike 12152c653b3SJeff Dike extern void scan_elf_aux( char **envp); 12252c653b3SJeff Dike 12352c653b3SJeff Dike int main(int argc, char **argv, char **envp) 12452c653b3SJeff Dike { 12552c653b3SJeff Dike char **new_argv; 12652c653b3SJeff Dike int ret, i, err; 12752c653b3SJeff Dike 12852c653b3SJeff Dike #ifdef UML_CONFIG_CMDLINE_ON_HOST 12952c653b3SJeff Dike /* Allocate memory for thread command lines */ 13052c653b3SJeff Dike if(argc < 2 || strlen(argv[1]) < THREAD_NAME_LEN - 1){ 13152c653b3SJeff Dike 13252c653b3SJeff Dike char padding[THREAD_NAME_LEN] = { 13352c653b3SJeff Dike [ 0 ... THREAD_NAME_LEN - 2] = ' ', '\0' 13452c653b3SJeff Dike }; 13552c653b3SJeff Dike 13652c653b3SJeff Dike new_argv = malloc((argc + 2) * sizeof(char*)); 13752c653b3SJeff Dike if(!new_argv) { 13852c653b3SJeff Dike perror("Allocating extended argv"); 13952c653b3SJeff Dike exit(1); 14052c653b3SJeff Dike } 14152c653b3SJeff Dike 14252c653b3SJeff Dike new_argv[0] = argv[0]; 14352c653b3SJeff Dike new_argv[1] = padding; 14452c653b3SJeff Dike 14552c653b3SJeff Dike for(i = 2; i <= argc; i++) 14652c653b3SJeff Dike new_argv[i] = argv[i - 1]; 14752c653b3SJeff Dike new_argv[argc + 1] = NULL; 14852c653b3SJeff Dike 14952c653b3SJeff Dike execvp(new_argv[0], new_argv); 15052c653b3SJeff Dike perror("execing with extended args"); 15152c653b3SJeff Dike exit(1); 15252c653b3SJeff Dike } 15352c653b3SJeff Dike #endif 15452c653b3SJeff Dike 15552c653b3SJeff Dike linux_prog = argv[0]; 15652c653b3SJeff Dike 15752c653b3SJeff Dike set_stklim(); 15852c653b3SJeff Dike 159cb98cdcdSMattia Dongili setup_env_path(); 160cb98cdcdSMattia Dongili 16152c653b3SJeff Dike new_argv = malloc((argc + 1) * sizeof(char *)); 16252c653b3SJeff Dike if(new_argv == NULL){ 16352c653b3SJeff Dike perror("Mallocing argv"); 16452c653b3SJeff Dike exit(1); 16552c653b3SJeff Dike } 16652c653b3SJeff Dike for(i=0;i<argc;i++){ 16752c653b3SJeff Dike new_argv[i] = strdup(argv[i]); 16852c653b3SJeff Dike if(new_argv[i] == NULL){ 16952c653b3SJeff Dike perror("Mallocing an arg"); 17052c653b3SJeff Dike exit(1); 17152c653b3SJeff Dike } 17252c653b3SJeff Dike } 17352c653b3SJeff Dike new_argv[argc] = NULL; 17452c653b3SJeff Dike 1754b84c69bSJeff Dike /* Allow these signals to bring down a UML if all other 1764b84c69bSJeff Dike * methods of control fail. 1774b84c69bSJeff Dike */ 1784b84c69bSJeff Dike install_fatal_handler(SIGINT); 1794b84c69bSJeff Dike install_fatal_handler(SIGTERM); 1804b84c69bSJeff Dike install_fatal_handler(SIGHUP); 18152c653b3SJeff Dike 18252c653b3SJeff Dike scan_elf_aux( envp); 18352c653b3SJeff Dike 18452c653b3SJeff Dike do_uml_initcalls(); 18552c653b3SJeff Dike ret = linux_main(argc, argv); 18652c653b3SJeff Dike 18752c653b3SJeff Dike /* Disable SIGPROF - I have no idea why libc doesn't do this or turn 18852c653b3SJeff Dike * off the profiling time, but UML dies with a SIGPROF just before 18952c653b3SJeff Dike * exiting when profiling is active. 19052c653b3SJeff Dike */ 19152c653b3SJeff Dike change_sig(SIGPROF, 0); 19252c653b3SJeff Dike 19352c653b3SJeff Dike /* This signal stuff used to be in the reboot case. However, 19452c653b3SJeff Dike * sometimes a SIGVTALRM can come in when we're halting (reproducably 19552c653b3SJeff Dike * when writing out gcov information, presumably because that takes 19652c653b3SJeff Dike * some time) and cause a segfault. 19752c653b3SJeff Dike */ 19852c653b3SJeff Dike 19952c653b3SJeff Dike /* stop timers and set SIG*ALRM to be ignored */ 20052c653b3SJeff Dike disable_timer(); 20152c653b3SJeff Dike 20252c653b3SJeff Dike /* disable SIGIO for the fds and set SIGIO to be ignored */ 20352c653b3SJeff Dike err = deactivate_all_fds(); 20452c653b3SJeff Dike if(err) 20552c653b3SJeff Dike printf("deactivate_all_fds failed, errno = %d\n", -err); 20652c653b3SJeff Dike 20752c653b3SJeff Dike /* Let any pending signals fire now. This ensures 20852c653b3SJeff Dike * that they won't be delivered after the exec, when 20952c653b3SJeff Dike * they are definitely not expected. 21052c653b3SJeff Dike */ 21152c653b3SJeff Dike unblock_signals(); 21252c653b3SJeff Dike 21352c653b3SJeff Dike /* Reboot */ 21452c653b3SJeff Dike if(ret){ 21552c653b3SJeff Dike printf("\n"); 21652c653b3SJeff Dike execvp(new_argv[0], new_argv); 21752c653b3SJeff Dike perror("Failed to exec kernel"); 21852c653b3SJeff Dike ret = 1; 21952c653b3SJeff Dike } 22052c653b3SJeff Dike printf("\n"); 22152c653b3SJeff Dike return(uml_exitcode); 22252c653b3SJeff Dike } 22352c653b3SJeff Dike 22452c653b3SJeff Dike #define CAN_KMALLOC() \ 22552c653b3SJeff Dike (kmalloc_ok && CHOOSE_MODE((os_getpid() != tracing_pid), 1)) 22652c653b3SJeff Dike 22752c653b3SJeff Dike extern void *__real_malloc(int); 22852c653b3SJeff Dike 22952c653b3SJeff Dike void *__wrap_malloc(int size) 23052c653b3SJeff Dike { 23152c653b3SJeff Dike void *ret; 23252c653b3SJeff Dike 23352c653b3SJeff Dike if(!CAN_KMALLOC()) 23452c653b3SJeff Dike return(__real_malloc(size)); 23552c653b3SJeff Dike else if(size <= PAGE_SIZE) /* finding contiguos pages can be hard*/ 23652c653b3SJeff Dike ret = um_kmalloc(size); 23752c653b3SJeff Dike else ret = um_vmalloc(size); 23852c653b3SJeff Dike 23952c653b3SJeff Dike /* glibc people insist that if malloc fails, errno should be 24052c653b3SJeff Dike * set by malloc as well. So we do. 24152c653b3SJeff Dike */ 24252c653b3SJeff Dike if(ret == NULL) 24352c653b3SJeff Dike errno = ENOMEM; 24452c653b3SJeff Dike 24552c653b3SJeff Dike return(ret); 24652c653b3SJeff Dike } 24752c653b3SJeff Dike 24852c653b3SJeff Dike void *__wrap_calloc(int n, int size) 24952c653b3SJeff Dike { 25052c653b3SJeff Dike void *ptr = __wrap_malloc(n * size); 25152c653b3SJeff Dike 25252c653b3SJeff Dike if(ptr == NULL) return(NULL); 25352c653b3SJeff Dike memset(ptr, 0, n * size); 25452c653b3SJeff Dike return(ptr); 25552c653b3SJeff Dike } 25652c653b3SJeff Dike 25752c653b3SJeff Dike extern void __real_free(void *); 25852c653b3SJeff Dike 25952c653b3SJeff Dike extern unsigned long high_physmem; 26052c653b3SJeff Dike 26152c653b3SJeff Dike void __wrap_free(void *ptr) 26252c653b3SJeff Dike { 26352c653b3SJeff Dike unsigned long addr = (unsigned long) ptr; 26452c653b3SJeff Dike 26552c653b3SJeff Dike /* We need to know how the allocation happened, so it can be correctly 26652c653b3SJeff Dike * freed. This is done by seeing what region of memory the pointer is 26752c653b3SJeff Dike * in - 26852c653b3SJeff Dike * physical memory - kmalloc/kfree 26952c653b3SJeff Dike * kernel virtual memory - vmalloc/vfree 27052c653b3SJeff Dike * anywhere else - malloc/free 27152c653b3SJeff Dike * If kmalloc is not yet possible, then either high_physmem and/or 27252c653b3SJeff Dike * end_vm are still 0 (as at startup), in which case we call free, or 27352c653b3SJeff Dike * we have set them, but anyway addr has not been allocated from those 27452c653b3SJeff Dike * areas. So, in both cases __real_free is called. 27552c653b3SJeff Dike * 27652c653b3SJeff Dike * CAN_KMALLOC is checked because it would be bad to free a buffer 27752c653b3SJeff Dike * with kmalloc/vmalloc after they have been turned off during 27852c653b3SJeff Dike * shutdown. 27952c653b3SJeff Dike * XXX: However, we sometimes shutdown CAN_KMALLOC temporarily, so 28052c653b3SJeff Dike * there is a possibility for memory leaks. 28152c653b3SJeff Dike */ 28252c653b3SJeff Dike 28352c653b3SJeff Dike if((addr >= uml_physmem) && (addr < high_physmem)){ 28452c653b3SJeff Dike if(CAN_KMALLOC()) 28552c653b3SJeff Dike kfree(ptr); 28652c653b3SJeff Dike } 28752c653b3SJeff Dike else if((addr >= start_vm) && (addr < end_vm)){ 28852c653b3SJeff Dike if(CAN_KMALLOC()) 28952c653b3SJeff Dike vfree(ptr); 29052c653b3SJeff Dike } 29152c653b3SJeff Dike else __real_free(ptr); 29252c653b3SJeff Dike } 293