xref: /linux/arch/um/os-Linux/main.c (revision f0c76bc82d4544bde874497e830b2c9fa04bafe1)
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>
1352c653b3SJeff Dike #include <sys/resource.h>
1468b9883cSBenjamin Berg #include <sys/personality.h>
1537185b33SAl Viro #include <as-layout.h>
1637185b33SAl Viro #include <init.h>
1737185b33SAl Viro #include <kern_util.h>
1837185b33SAl Viro #include <os.h>
1937185b33SAl Viro #include <um_malloc.h>
20847d3abcSTiwei Bie #include "internal.h"
2152c653b3SJeff Dike 
2252c653b3SJeff Dike #define PGD_BOUND (4 * 1024 * 1024)
2352c653b3SJeff Dike #define STACKSIZE (8 * 1024 * 1024)
2452c653b3SJeff Dike #define THREAD_NAME_LEN (256)
2552c653b3SJeff Dike 
26b743ac54SRichard Weinberger long elf_aux_hwcap;
27b743ac54SRichard Weinberger 
28*f0c76bc8STiwei Bie static void __init set_stklim(void)
2952c653b3SJeff Dike {
3052c653b3SJeff Dike 	struct rlimit lim;
3152c653b3SJeff Dike 
3252c653b3SJeff Dike 	if (getrlimit(RLIMIT_STACK, &lim) < 0) {
3352c653b3SJeff Dike 		perror("getrlimit");
3452c653b3SJeff Dike 		exit(1);
3552c653b3SJeff Dike 	}
3652c653b3SJeff Dike 	if ((lim.rlim_cur == RLIM_INFINITY) || (lim.rlim_cur > STACKSIZE)) {
3752c653b3SJeff Dike 		lim.rlim_cur = STACKSIZE;
3852c653b3SJeff Dike 		if (setrlimit(RLIMIT_STACK, &lim) < 0) {
3952c653b3SJeff Dike 			perror("setrlimit");
4052c653b3SJeff Dike 			exit(1);
4152c653b3SJeff Dike 		}
4252c653b3SJeff Dike 	}
4352c653b3SJeff Dike }
4452c653b3SJeff Dike 
4552c653b3SJeff Dike static void last_ditch_exit(int sig)
4652c653b3SJeff Dike {
4752c653b3SJeff Dike 	uml_cleanup();
4852c653b3SJeff Dike 	exit(1);
4952c653b3SJeff Dike }
5052c653b3SJeff Dike 
514b84c69bSJeff Dike static void install_fatal_handler(int sig)
524b84c69bSJeff Dike {
534b84c69bSJeff Dike 	struct sigaction action;
544b84c69bSJeff Dike 
554b84c69bSJeff Dike 	/* All signals are enabled in this handler ... */
564b84c69bSJeff Dike 	sigemptyset(&action.sa_mask);
574b84c69bSJeff Dike 
58ba180fd4SJeff Dike 	/*
59ba180fd4SJeff Dike 	 * ... including the signal being handled, plus we want the
604b84c69bSJeff Dike 	 * handler reset to the default behavior, so that if an exit
614b84c69bSJeff Dike 	 * handler is hanging for some reason, the UML will just die
624b84c69bSJeff Dike 	 * after this signal is sent a second time.
634b84c69bSJeff Dike 	 */
644b84c69bSJeff Dike 	action.sa_flags = SA_RESETHAND | SA_NODEFER;
654b84c69bSJeff Dike 	action.sa_restorer = NULL;
664b84c69bSJeff Dike 	action.sa_handler = last_ditch_exit;
674b84c69bSJeff Dike 	if (sigaction(sig, &action, NULL) < 0) {
680936d4f3SMasami Hiramatsu 		os_warn("failed to install handler for signal %d "
690936d4f3SMasami Hiramatsu 			"- errno = %d\n", sig, errno);
704b84c69bSJeff Dike 		exit(1);
714b84c69bSJeff Dike 	}
724b84c69bSJeff Dike }
734b84c69bSJeff Dike 
740ce451acSRichard Weinberger #define UML_LIB_PATH	":" OS_LIB_PATH "/uml"
75cb98cdcdSMattia Dongili 
76cb98cdcdSMattia Dongili static void setup_env_path(void)
77cb98cdcdSMattia Dongili {
78cb98cdcdSMattia Dongili 	char *new_path = NULL;
79cb98cdcdSMattia Dongili 	char *old_path = NULL;
80cb98cdcdSMattia Dongili 	int path_len = 0;
81cb98cdcdSMattia Dongili 
82cb98cdcdSMattia Dongili 	old_path = getenv("PATH");
83ba180fd4SJeff Dike 	/*
84ba180fd4SJeff Dike 	 * if no PATH variable is set or it has an empty value
85cb98cdcdSMattia Dongili 	 * just use the default + /usr/lib/uml
86cb98cdcdSMattia Dongili 	 */
87cb98cdcdSMattia Dongili 	if (!old_path || (path_len = strlen(old_path)) == 0) {
88c9a3072dSWANG Cong 		if (putenv("PATH=:/bin:/usr/bin/" UML_LIB_PATH))
89c9a3072dSWANG Cong 			perror("couldn't putenv");
90cb98cdcdSMattia Dongili 		return;
91cb98cdcdSMattia Dongili 	}
92cb98cdcdSMattia Dongili 
93cb98cdcdSMattia Dongili 	/* append /usr/lib/uml to the existing path */
94cb98cdcdSMattia Dongili 	path_len += strlen("PATH=" UML_LIB_PATH) + 1;
95cb98cdcdSMattia Dongili 	new_path = malloc(path_len);
96cb98cdcdSMattia Dongili 	if (!new_path) {
97c9a3072dSWANG Cong 		perror("couldn't malloc to set a new PATH");
98cb98cdcdSMattia Dongili 		return;
99cb98cdcdSMattia Dongili 	}
100cb98cdcdSMattia Dongili 	snprintf(new_path, path_len, "PATH=%s" UML_LIB_PATH, old_path);
101c9a3072dSWANG Cong 	if (putenv(new_path)) {
102c9a3072dSWANG Cong 		perror("couldn't putenv to set a new PATH");
103c9a3072dSWANG Cong 		free(new_path);
104c9a3072dSWANG Cong 	}
105cb98cdcdSMattia Dongili }
106cb98cdcdSMattia Dongili 
10736e45463SJeff Dike int __init main(int argc, char **argv, char **envp)
10852c653b3SJeff Dike {
10952c653b3SJeff Dike 	char **new_argv;
11052c653b3SJeff Dike 	int ret, i, err;
11152c653b3SJeff Dike 
11268b9883cSBenjamin Berg 	/* Disable randomization and re-exec if it was changed successfully */
11368b9883cSBenjamin Berg 	ret = personality(PER_LINUX | ADDR_NO_RANDOMIZE);
11468b9883cSBenjamin Berg 	if (ret >= 0 && (ret & (PER_LINUX | ADDR_NO_RANDOMIZE)) !=
115031acdcfSJohannes Berg 			 (PER_LINUX | ADDR_NO_RANDOMIZE)) {
116d61ac4a7SJohannes Berg 		char buf[4096] = {};
117031acdcfSJohannes Berg 		ssize_t ret;
118031acdcfSJohannes Berg 
119031acdcfSJohannes Berg 		ret = readlink("/proc/self/exe", buf, sizeof(buf));
120031acdcfSJohannes Berg 		if (ret < 0 || ret >= sizeof(buf)) {
121031acdcfSJohannes Berg 			perror("readlink failure");
122031acdcfSJohannes Berg 			exit(1);
123031acdcfSJohannes Berg 		}
124031acdcfSJohannes Berg 		execve(buf, argv, envp);
125031acdcfSJohannes Berg 	}
12668b9883cSBenjamin Berg 
12752c653b3SJeff Dike 	set_stklim();
12852c653b3SJeff Dike 
129cb98cdcdSMattia Dongili 	setup_env_path();
130cb98cdcdSMattia Dongili 
13125012721SRichard Weinberger 	setsid();
13225012721SRichard Weinberger 
13352c653b3SJeff Dike 	new_argv = malloc((argc + 1) * sizeof(char *));
13452c653b3SJeff Dike 	if (new_argv == NULL) {
13552c653b3SJeff Dike 		perror("Mallocing argv");
13652c653b3SJeff Dike 		exit(1);
13752c653b3SJeff Dike 	}
13852c653b3SJeff Dike 	for (i = 0; i < argc; i++) {
13952c653b3SJeff Dike 		new_argv[i] = strdup(argv[i]);
14052c653b3SJeff Dike 		if (new_argv[i] == NULL) {
14152c653b3SJeff Dike 			perror("Mallocing an arg");
14252c653b3SJeff Dike 			exit(1);
14352c653b3SJeff Dike 		}
14452c653b3SJeff Dike 	}
14552c653b3SJeff Dike 	new_argv[argc] = NULL;
14652c653b3SJeff Dike 
147ba180fd4SJeff Dike 	/*
148ba180fd4SJeff Dike 	 * Allow these signals to bring down a UML if all other
1494b84c69bSJeff Dike 	 * methods of control fail.
1504b84c69bSJeff Dike 	 */
1514b84c69bSJeff Dike 	install_fatal_handler(SIGINT);
1524b84c69bSJeff Dike 	install_fatal_handler(SIGTERM);
15352c653b3SJeff Dike 
154b743ac54SRichard Weinberger #ifdef CONFIG_ARCH_REUSE_HOST_VSYSCALL_AREA
15552c653b3SJeff Dike 	scan_elf_aux(envp);
156b743ac54SRichard Weinberger #endif
15752c653b3SJeff Dike 
1589fcb663bSAnton Ivanov 	change_sig(SIGPIPE, 0);
15968b9883cSBenjamin Berg 	ret = linux_main(argc, argv, envp);
16052c653b3SJeff Dike 
161ba180fd4SJeff Dike 	/*
162ba180fd4SJeff Dike 	 * Disable SIGPROF - I have no idea why libc doesn't do this or turn
16352c653b3SJeff Dike 	 * off the profiling time, but UML dies with a SIGPROF just before
16452c653b3SJeff Dike 	 * exiting when profiling is active.
16552c653b3SJeff Dike 	 */
16652c653b3SJeff Dike 	change_sig(SIGPROF, 0);
16752c653b3SJeff Dike 
168ba180fd4SJeff Dike 	/*
169ba180fd4SJeff Dike 	 * This signal stuff used to be in the reboot case.  However,
1702eb5f31bSAnton Ivanov 	 * sometimes a timer signal can come in when we're halting (reproducably
17152c653b3SJeff Dike 	 * when writing out gcov information, presumably because that takes
17252c653b3SJeff Dike 	 * some time) and cause a segfault.
17352c653b3SJeff Dike 	 */
17452c653b3SJeff Dike 
1752eb5f31bSAnton Ivanov 	/* stop timers and set timer signal to be ignored */
1762eb5f31bSAnton Ivanov 	os_timer_disable();
17752c653b3SJeff Dike 
17852c653b3SJeff Dike 	/* disable SIGIO for the fds and set SIGIO to be ignored */
17952c653b3SJeff Dike 	err = deactivate_all_fds();
18052c653b3SJeff Dike 	if (err)
1810936d4f3SMasami Hiramatsu 		os_warn("deactivate_all_fds failed, errno = %d\n", -err);
18252c653b3SJeff Dike 
183ba180fd4SJeff Dike 	/*
184ba180fd4SJeff Dike 	 * Let any pending signals fire now.  This ensures
18552c653b3SJeff Dike 	 * that they won't be delivered after the exec, when
18652c653b3SJeff Dike 	 * they are definitely not expected.
18752c653b3SJeff Dike 	 */
1885c1f33e2SJohannes Berg 	unblock_signals();
18952c653b3SJeff Dike 
190d3878bb8SMasami Hiramatsu 	os_info("\n");
19152c653b3SJeff Dike 	/* Reboot */
19252c653b3SJeff Dike 	if (ret) {
19352c653b3SJeff Dike 		execvp(new_argv[0], new_argv);
19452c653b3SJeff Dike 		perror("Failed to exec kernel");
19552c653b3SJeff Dike 		ret = 1;
19652c653b3SJeff Dike 	}
197a5ed1ffaSJeff Dike 	return uml_exitcode;
19852c653b3SJeff Dike }
19952c653b3SJeff Dike 
20052c653b3SJeff Dike extern void *__real_malloc(int);
201855f6e18STiwei Bie extern void __real_free(void *);
20252c653b3SJeff Dike 
20349ff7d87STiwei Bie /* workaround for -Wmissing-prototypes warnings */
20449ff7d87STiwei Bie void *__wrap_malloc(int size);
20549ff7d87STiwei Bie void *__wrap_calloc(int n, int size);
20649ff7d87STiwei Bie void __wrap_free(void *ptr);
20749ff7d87STiwei Bie 
20852c653b3SJeff Dike void *__wrap_malloc(int size)
20952c653b3SJeff Dike {
21052c653b3SJeff Dike 	void *ret;
21152c653b3SJeff Dike 
2126aa802ceSJeff Dike 	if (!kmalloc_ok)
213a5ed1ffaSJeff Dike 		return __real_malloc(size);
214c539ab73SJeff Dike 	else if (size <= UM_KERN_PAGE_SIZE)
215c539ab73SJeff Dike 		/* finding contiguous pages can be hard*/
21643f5b308SJeff Dike 		ret = uml_kmalloc(size, UM_GFP_KERNEL);
217e4c4bf99SJeff Dike 	else ret = vmalloc(size);
21852c653b3SJeff Dike 
219ba180fd4SJeff Dike 	/*
220ba180fd4SJeff Dike 	 * glibc people insist that if malloc fails, errno should be
22152c653b3SJeff Dike 	 * set by malloc as well. So we do.
22252c653b3SJeff Dike 	 */
22352c653b3SJeff Dike 	if (ret == NULL)
22452c653b3SJeff Dike 		errno = ENOMEM;
22552c653b3SJeff Dike 
226a5ed1ffaSJeff Dike 	return ret;
22752c653b3SJeff Dike }
22852c653b3SJeff Dike 
22952c653b3SJeff Dike void *__wrap_calloc(int n, int size)
23052c653b3SJeff Dike {
23152c653b3SJeff Dike 	void *ptr = __wrap_malloc(n * size);
23252c653b3SJeff Dike 
233a5ed1ffaSJeff Dike 	if (ptr == NULL)
234a5ed1ffaSJeff Dike 		return NULL;
23552c653b3SJeff Dike 	memset(ptr, 0, n * size);
236a5ed1ffaSJeff Dike 	return ptr;
23752c653b3SJeff Dike }
23852c653b3SJeff Dike 
23952c653b3SJeff Dike void __wrap_free(void *ptr)
24052c653b3SJeff Dike {
24152c653b3SJeff Dike 	unsigned long addr = (unsigned long) ptr;
24252c653b3SJeff Dike 
243ba180fd4SJeff Dike 	/*
244ba180fd4SJeff Dike 	 * We need to know how the allocation happened, so it can be correctly
24552c653b3SJeff Dike 	 * freed.  This is done by seeing what region of memory the pointer is
24652c653b3SJeff Dike 	 * in -
24752c653b3SJeff Dike 	 * 	physical memory - kmalloc/kfree
24852c653b3SJeff Dike 	 *	kernel virtual memory - vmalloc/vfree
24952c653b3SJeff Dike 	 * 	anywhere else - malloc/free
25052c653b3SJeff Dike 	 * If kmalloc is not yet possible, then either high_physmem and/or
25152c653b3SJeff Dike 	 * end_vm are still 0 (as at startup), in which case we call free, or
25252c653b3SJeff Dike 	 * we have set them, but anyway addr has not been allocated from those
25352c653b3SJeff Dike 	 * areas. So, in both cases __real_free is called.
25452c653b3SJeff Dike 	 *
25552c653b3SJeff Dike 	 * CAN_KMALLOC is checked because it would be bad to free a buffer
25652c653b3SJeff Dike 	 * with kmalloc/vmalloc after they have been turned off during
25752c653b3SJeff Dike 	 * shutdown.
25852c653b3SJeff Dike 	 * XXX: However, we sometimes shutdown CAN_KMALLOC temporarily, so
25952c653b3SJeff Dike 	 * there is a possibility for memory leaks.
26052c653b3SJeff Dike 	 */
26152c653b3SJeff Dike 
26252c653b3SJeff Dike 	if ((addr >= uml_physmem) && (addr < high_physmem)) {
2636aa802ceSJeff Dike 		if (kmalloc_ok)
26452c653b3SJeff Dike 			kfree(ptr);
26552c653b3SJeff Dike 	}
26652c653b3SJeff Dike 	else if ((addr >= start_vm) && (addr < end_vm)) {
2676aa802ceSJeff Dike 		if (kmalloc_ok)
26852c653b3SJeff Dike 			vfree(ptr);
26952c653b3SJeff Dike 	}
27052c653b3SJeff Dike 	else __real_free(ptr);
27152c653b3SJeff Dike }
272