197870c34SAlex Dewar // SPDX-License-Identifier: GPL-2.0 252c653b3SJeff Dike /* 32eb5f31bSAnton Ivanov * Copyright (C) 2015 Thomas Meyer (thomas@m3y3r.de) 4ba180fd4SJeff Dike * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com) 552c653b3SJeff Dike */ 652c653b3SJeff Dike 752c653b3SJeff Dike #include <stdio.h> 852c653b3SJeff Dike #include <stdlib.h> 9ba180fd4SJeff Dike #include <unistd.h> 1052c653b3SJeff Dike #include <errno.h> 11ba180fd4SJeff Dike #include <signal.h> 12ba180fd4SJeff Dike #include <string.h> 13*031acdcfSJohannes Berg #include <limits.h> 1452c653b3SJeff Dike #include <sys/resource.h> 1568b9883cSBenjamin Berg #include <sys/personality.h> 1637185b33SAl Viro #include <as-layout.h> 1737185b33SAl Viro #include <init.h> 1837185b33SAl Viro #include <kern_util.h> 1937185b33SAl Viro #include <os.h> 2037185b33SAl Viro #include <um_malloc.h> 21847d3abcSTiwei Bie #include "internal.h" 2252c653b3SJeff Dike 2352c653b3SJeff Dike #define PGD_BOUND (4 * 1024 * 1024) 2452c653b3SJeff Dike #define STACKSIZE (8 * 1024 * 1024) 2552c653b3SJeff Dike #define THREAD_NAME_LEN (256) 2652c653b3SJeff Dike 27b743ac54SRichard Weinberger long elf_aux_hwcap; 28b743ac54SRichard Weinberger 2952c653b3SJeff Dike static void set_stklim(void) 3052c653b3SJeff Dike { 3152c653b3SJeff Dike struct rlimit lim; 3252c653b3SJeff Dike 3352c653b3SJeff Dike if (getrlimit(RLIMIT_STACK, &lim) < 0) { 3452c653b3SJeff Dike perror("getrlimit"); 3552c653b3SJeff Dike exit(1); 3652c653b3SJeff Dike } 3752c653b3SJeff Dike if ((lim.rlim_cur == RLIM_INFINITY) || (lim.rlim_cur > STACKSIZE)) { 3852c653b3SJeff Dike lim.rlim_cur = STACKSIZE; 3952c653b3SJeff Dike if (setrlimit(RLIMIT_STACK, &lim) < 0) { 4052c653b3SJeff Dike perror("setrlimit"); 4152c653b3SJeff Dike exit(1); 4252c653b3SJeff Dike } 4352c653b3SJeff Dike } 4452c653b3SJeff Dike } 4552c653b3SJeff Dike 4652c653b3SJeff Dike static void last_ditch_exit(int sig) 4752c653b3SJeff Dike { 4852c653b3SJeff Dike uml_cleanup(); 4952c653b3SJeff Dike exit(1); 5052c653b3SJeff Dike } 5152c653b3SJeff Dike 524b84c69bSJeff Dike static void install_fatal_handler(int sig) 534b84c69bSJeff Dike { 544b84c69bSJeff Dike struct sigaction action; 554b84c69bSJeff Dike 564b84c69bSJeff Dike /* All signals are enabled in this handler ... */ 574b84c69bSJeff Dike sigemptyset(&action.sa_mask); 584b84c69bSJeff Dike 59ba180fd4SJeff Dike /* 60ba180fd4SJeff Dike * ... including the signal being handled, plus we want the 614b84c69bSJeff Dike * handler reset to the default behavior, so that if an exit 624b84c69bSJeff Dike * handler is hanging for some reason, the UML will just die 634b84c69bSJeff Dike * after this signal is sent a second time. 644b84c69bSJeff Dike */ 654b84c69bSJeff Dike action.sa_flags = SA_RESETHAND | SA_NODEFER; 664b84c69bSJeff Dike action.sa_restorer = NULL; 674b84c69bSJeff Dike action.sa_handler = last_ditch_exit; 684b84c69bSJeff Dike if (sigaction(sig, &action, NULL) < 0) { 690936d4f3SMasami Hiramatsu os_warn("failed to install handler for signal %d " 700936d4f3SMasami Hiramatsu "- errno = %d\n", sig, errno); 714b84c69bSJeff Dike exit(1); 724b84c69bSJeff Dike } 734b84c69bSJeff Dike } 744b84c69bSJeff Dike 750ce451acSRichard Weinberger #define UML_LIB_PATH ":" OS_LIB_PATH "/uml" 76cb98cdcdSMattia Dongili 77cb98cdcdSMattia Dongili static void setup_env_path(void) 78cb98cdcdSMattia Dongili { 79cb98cdcdSMattia Dongili char *new_path = NULL; 80cb98cdcdSMattia Dongili char *old_path = NULL; 81cb98cdcdSMattia Dongili int path_len = 0; 82cb98cdcdSMattia Dongili 83cb98cdcdSMattia Dongili old_path = getenv("PATH"); 84ba180fd4SJeff Dike /* 85ba180fd4SJeff Dike * if no PATH variable is set or it has an empty value 86cb98cdcdSMattia Dongili * just use the default + /usr/lib/uml 87cb98cdcdSMattia Dongili */ 88cb98cdcdSMattia Dongili if (!old_path || (path_len = strlen(old_path)) == 0) { 89c9a3072dSWANG Cong if (putenv("PATH=:/bin:/usr/bin/" UML_LIB_PATH)) 90c9a3072dSWANG Cong perror("couldn't putenv"); 91cb98cdcdSMattia Dongili return; 92cb98cdcdSMattia Dongili } 93cb98cdcdSMattia Dongili 94cb98cdcdSMattia Dongili /* append /usr/lib/uml to the existing path */ 95cb98cdcdSMattia Dongili path_len += strlen("PATH=" UML_LIB_PATH) + 1; 96cb98cdcdSMattia Dongili new_path = malloc(path_len); 97cb98cdcdSMattia Dongili if (!new_path) { 98c9a3072dSWANG Cong perror("couldn't malloc to set a new PATH"); 99cb98cdcdSMattia Dongili return; 100cb98cdcdSMattia Dongili } 101cb98cdcdSMattia Dongili snprintf(new_path, path_len, "PATH=%s" UML_LIB_PATH, old_path); 102c9a3072dSWANG Cong if (putenv(new_path)) { 103c9a3072dSWANG Cong perror("couldn't putenv to set a new PATH"); 104c9a3072dSWANG Cong free(new_path); 105c9a3072dSWANG Cong } 106cb98cdcdSMattia Dongili } 107cb98cdcdSMattia Dongili 10836e45463SJeff Dike int __init main(int argc, char **argv, char **envp) 10952c653b3SJeff Dike { 11052c653b3SJeff Dike char **new_argv; 11152c653b3SJeff Dike int ret, i, err; 11252c653b3SJeff Dike 11368b9883cSBenjamin Berg /* Disable randomization and re-exec if it was changed successfully */ 11468b9883cSBenjamin Berg ret = personality(PER_LINUX | ADDR_NO_RANDOMIZE); 11568b9883cSBenjamin Berg if (ret >= 0 && (ret & (PER_LINUX | ADDR_NO_RANDOMIZE)) != 116*031acdcfSJohannes Berg (PER_LINUX | ADDR_NO_RANDOMIZE)) { 117*031acdcfSJohannes Berg char buf[PATH_MAX] = {}; 118*031acdcfSJohannes Berg ssize_t ret; 119*031acdcfSJohannes Berg 120*031acdcfSJohannes Berg ret = readlink("/proc/self/exe", buf, sizeof(buf)); 121*031acdcfSJohannes Berg if (ret < 0 || ret >= sizeof(buf)) { 122*031acdcfSJohannes Berg perror("readlink failure"); 123*031acdcfSJohannes Berg exit(1); 124*031acdcfSJohannes Berg } 125*031acdcfSJohannes Berg execve(buf, argv, envp); 126*031acdcfSJohannes Berg } 12768b9883cSBenjamin Berg 12852c653b3SJeff Dike set_stklim(); 12952c653b3SJeff Dike 130cb98cdcdSMattia Dongili setup_env_path(); 131cb98cdcdSMattia Dongili 13225012721SRichard Weinberger setsid(); 13325012721SRichard Weinberger 13452c653b3SJeff Dike new_argv = malloc((argc + 1) * sizeof(char *)); 13552c653b3SJeff Dike if (new_argv == NULL) { 13652c653b3SJeff Dike perror("Mallocing argv"); 13752c653b3SJeff Dike exit(1); 13852c653b3SJeff Dike } 13952c653b3SJeff Dike for (i = 0; i < argc; i++) { 14052c653b3SJeff Dike new_argv[i] = strdup(argv[i]); 14152c653b3SJeff Dike if (new_argv[i] == NULL) { 14252c653b3SJeff Dike perror("Mallocing an arg"); 14352c653b3SJeff Dike exit(1); 14452c653b3SJeff Dike } 14552c653b3SJeff Dike } 14652c653b3SJeff Dike new_argv[argc] = NULL; 14752c653b3SJeff Dike 148ba180fd4SJeff Dike /* 149ba180fd4SJeff Dike * Allow these signals to bring down a UML if all other 1504b84c69bSJeff Dike * methods of control fail. 1514b84c69bSJeff Dike */ 1524b84c69bSJeff Dike install_fatal_handler(SIGINT); 1534b84c69bSJeff Dike install_fatal_handler(SIGTERM); 15452c653b3SJeff Dike 155b743ac54SRichard Weinberger #ifdef CONFIG_ARCH_REUSE_HOST_VSYSCALL_AREA 15652c653b3SJeff Dike scan_elf_aux(envp); 157b743ac54SRichard Weinberger #endif 15852c653b3SJeff Dike 1599fcb663bSAnton Ivanov change_sig(SIGPIPE, 0); 16068b9883cSBenjamin Berg ret = linux_main(argc, argv, envp); 16152c653b3SJeff Dike 162ba180fd4SJeff Dike /* 163ba180fd4SJeff Dike * Disable SIGPROF - I have no idea why libc doesn't do this or turn 16452c653b3SJeff Dike * off the profiling time, but UML dies with a SIGPROF just before 16552c653b3SJeff Dike * exiting when profiling is active. 16652c653b3SJeff Dike */ 16752c653b3SJeff Dike change_sig(SIGPROF, 0); 16852c653b3SJeff Dike 169ba180fd4SJeff Dike /* 170ba180fd4SJeff Dike * This signal stuff used to be in the reboot case. However, 1712eb5f31bSAnton Ivanov * sometimes a timer signal can come in when we're halting (reproducably 17252c653b3SJeff Dike * when writing out gcov information, presumably because that takes 17352c653b3SJeff Dike * some time) and cause a segfault. 17452c653b3SJeff Dike */ 17552c653b3SJeff Dike 1762eb5f31bSAnton Ivanov /* stop timers and set timer signal to be ignored */ 1772eb5f31bSAnton Ivanov os_timer_disable(); 17852c653b3SJeff Dike 17952c653b3SJeff Dike /* disable SIGIO for the fds and set SIGIO to be ignored */ 18052c653b3SJeff Dike err = deactivate_all_fds(); 18152c653b3SJeff Dike if (err) 1820936d4f3SMasami Hiramatsu os_warn("deactivate_all_fds failed, errno = %d\n", -err); 18352c653b3SJeff Dike 184ba180fd4SJeff Dike /* 185ba180fd4SJeff Dike * Let any pending signals fire now. This ensures 18652c653b3SJeff Dike * that they won't be delivered after the exec, when 18752c653b3SJeff Dike * they are definitely not expected. 18852c653b3SJeff Dike */ 1895c1f33e2SJohannes Berg unblock_signals(); 19052c653b3SJeff Dike 191d3878bb8SMasami Hiramatsu os_info("\n"); 19252c653b3SJeff Dike /* Reboot */ 19352c653b3SJeff Dike if (ret) { 19452c653b3SJeff Dike execvp(new_argv[0], new_argv); 19552c653b3SJeff Dike perror("Failed to exec kernel"); 19652c653b3SJeff Dike ret = 1; 19752c653b3SJeff Dike } 198a5ed1ffaSJeff Dike return uml_exitcode; 19952c653b3SJeff Dike } 20052c653b3SJeff Dike 20152c653b3SJeff Dike extern void *__real_malloc(int); 202855f6e18STiwei Bie extern void __real_free(void *); 20352c653b3SJeff Dike 20449ff7d87STiwei Bie /* workaround for -Wmissing-prototypes warnings */ 20549ff7d87STiwei Bie void *__wrap_malloc(int size); 20649ff7d87STiwei Bie void *__wrap_calloc(int n, int size); 20749ff7d87STiwei Bie void __wrap_free(void *ptr); 20849ff7d87STiwei Bie 20952c653b3SJeff Dike void *__wrap_malloc(int size) 21052c653b3SJeff Dike { 21152c653b3SJeff Dike void *ret; 21252c653b3SJeff Dike 2136aa802ceSJeff Dike if (!kmalloc_ok) 214a5ed1ffaSJeff Dike return __real_malloc(size); 215c539ab73SJeff Dike else if (size <= UM_KERN_PAGE_SIZE) 216c539ab73SJeff Dike /* finding contiguous pages can be hard*/ 21743f5b308SJeff Dike ret = uml_kmalloc(size, UM_GFP_KERNEL); 218e4c4bf99SJeff Dike else ret = vmalloc(size); 21952c653b3SJeff Dike 220ba180fd4SJeff Dike /* 221ba180fd4SJeff Dike * glibc people insist that if malloc fails, errno should be 22252c653b3SJeff Dike * set by malloc as well. So we do. 22352c653b3SJeff Dike */ 22452c653b3SJeff Dike if (ret == NULL) 22552c653b3SJeff Dike errno = ENOMEM; 22652c653b3SJeff Dike 227a5ed1ffaSJeff Dike return ret; 22852c653b3SJeff Dike } 22952c653b3SJeff Dike 23052c653b3SJeff Dike void *__wrap_calloc(int n, int size) 23152c653b3SJeff Dike { 23252c653b3SJeff Dike void *ptr = __wrap_malloc(n * size); 23352c653b3SJeff Dike 234a5ed1ffaSJeff Dike if (ptr == NULL) 235a5ed1ffaSJeff Dike return NULL; 23652c653b3SJeff Dike memset(ptr, 0, n * size); 237a5ed1ffaSJeff Dike return ptr; 23852c653b3SJeff Dike } 23952c653b3SJeff Dike 24052c653b3SJeff Dike void __wrap_free(void *ptr) 24152c653b3SJeff Dike { 24252c653b3SJeff Dike unsigned long addr = (unsigned long) ptr; 24352c653b3SJeff Dike 244ba180fd4SJeff Dike /* 245ba180fd4SJeff Dike * We need to know how the allocation happened, so it can be correctly 24652c653b3SJeff Dike * freed. This is done by seeing what region of memory the pointer is 24752c653b3SJeff Dike * in - 24852c653b3SJeff Dike * physical memory - kmalloc/kfree 24952c653b3SJeff Dike * kernel virtual memory - vmalloc/vfree 25052c653b3SJeff Dike * anywhere else - malloc/free 25152c653b3SJeff Dike * If kmalloc is not yet possible, then either high_physmem and/or 25252c653b3SJeff Dike * end_vm are still 0 (as at startup), in which case we call free, or 25352c653b3SJeff Dike * we have set them, but anyway addr has not been allocated from those 25452c653b3SJeff Dike * areas. So, in both cases __real_free is called. 25552c653b3SJeff Dike * 25652c653b3SJeff Dike * CAN_KMALLOC is checked because it would be bad to free a buffer 25752c653b3SJeff Dike * with kmalloc/vmalloc after they have been turned off during 25852c653b3SJeff Dike * shutdown. 25952c653b3SJeff Dike * XXX: However, we sometimes shutdown CAN_KMALLOC temporarily, so 26052c653b3SJeff Dike * there is a possibility for memory leaks. 26152c653b3SJeff Dike */ 26252c653b3SJeff Dike 26352c653b3SJeff Dike if ((addr >= uml_physmem) && (addr < high_physmem)) { 2646aa802ceSJeff Dike if (kmalloc_ok) 26552c653b3SJeff Dike kfree(ptr); 26652c653b3SJeff Dike } 26752c653b3SJeff Dike else if ((addr >= start_vm) && (addr < end_vm)) { 2686aa802ceSJeff Dike if (kmalloc_ok) 26952c653b3SJeff Dike vfree(ptr); 27052c653b3SJeff Dike } 27152c653b3SJeff Dike else __real_free(ptr); 27252c653b3SJeff Dike } 273