shim_exec.c 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496
  1. /* -*- mode:c; c-file-style:"k&r"; c-basic-offset: 4; tab-width:4; indent-tabs-mode:nil; mode:auto-fill; fill-column:78; -*- */
  2. /* vim: set ts=4 sw=4 et tw=78 fo=cqt wm=0: */
  3. /* Copyright (C) 2014 Stony Brook University
  4. This file is part of Graphene Library OS.
  5. Graphene Library OS is free software: you can redistribute it and/or
  6. modify it under the terms of the GNU Lesser General Public License
  7. as published by the Free Software Foundation, either version 3 of the
  8. License, or (at your option) any later version.
  9. Graphene Library OS is distributed in the hope that it will be useful,
  10. but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. GNU Lesser General Public License for more details.
  13. You should have received a copy of the GNU Lesser General Public License
  14. along with this program. If not, see <http://www.gnu.org/licenses/>. */
  15. /*
  16. * shim_epoll.c
  17. *
  18. * Implementation of system call "execve".
  19. */
  20. #include <shim_internal.h>
  21. #include <shim_table.h>
  22. #include <shim_thread.h>
  23. #include <shim_fs.h>
  24. #include <shim_ipc.h>
  25. #include <shim_profile.h>
  26. #include <pal.h>
  27. #include <pal_error.h>
  28. #include <errno.h>
  29. #include <linux/futex.h>
  30. #include <sys/syscall.h>
  31. #include <sys/mman.h>
  32. #include <asm/prctl.h>
  33. static int close_on_exec (struct shim_fd_handle * fd_hdl,
  34. struct shim_handle_map * map, void * arg)
  35. {
  36. if (fd_hdl->flags & FD_CLOEXEC) {
  37. struct shim_handle * hdl = __detach_fd_handle(fd_hdl, NULL, map);
  38. close_handle(hdl);
  39. }
  40. return 0;
  41. }
  42. static int close_cloexec_handle (struct shim_handle_map * map)
  43. {
  44. return walk_handle_map(&close_on_exec, map, NULL);
  45. }
  46. DEFINE_PROFILE_CATEGORY(exec_rtld, exec);
  47. DEFINE_PROFILE_INTERVAL(alloc_new_stack_for_exec, exec_rtld);
  48. DEFINE_PROFILE_INTERVAL(arrange_arguments_for_exec, exec_rtld);
  49. DEFINE_PROFILE_INTERVAL(unmap_executable_for_exec, exec_rtld);
  50. DEFINE_PROFILE_INTERVAL(unmap_loaded_binaries_for_exec, exec_rtld);
  51. DEFINE_PROFILE_INTERVAL(unmap_all_vmas_for_exec, exec_rtld);
  52. DEFINE_PROFILE_INTERVAL(load_new_executable_for_exec, exec_rtld);
  53. static void * old_stack_top, * old_stack, * old_stack_red;
  54. static const char ** new_argp;
  55. static int new_argc;
  56. static int * new_argcp;
  57. static elf_auxv_t * new_auxp;
  58. int init_brk_from_executable (struct shim_handle * exec);
  59. int shim_do_execve_rtld (struct shim_handle * hdl, const char ** argv,
  60. const char ** envp)
  61. {
  62. BEGIN_PROFILE_INTERVAL();
  63. struct shim_thread * cur_thread = get_cur_thread();
  64. int ret;
  65. if ((ret = close_cloexec_handle(cur_thread->handle_map)) < 0)
  66. return ret;
  67. SAVE_PROFILE_INTERVAL(close_CLOEXEC_files_for_exec);
  68. void * tcb = malloc(sizeof(__libc_tcb_t));
  69. if (!tcb)
  70. return -ENOMEM;
  71. populate_tls(tcb, false);
  72. __disable_preempt(&((__libc_tcb_t *) tcb)->shim_tcb); // Temporarily disable preemption
  73. // during execve().
  74. debug("set tcb to %p\n", tcb);
  75. put_handle(cur_thread->exec);
  76. get_handle(hdl);
  77. cur_thread->exec = hdl;
  78. old_stack_top = cur_thread->stack_top;
  79. old_stack = cur_thread->stack;
  80. old_stack_red = cur_thread->stack_red;
  81. cur_thread->stack_top = NULL;
  82. cur_thread->stack = NULL;
  83. cur_thread->stack_red = NULL;
  84. initial_envp = NULL;
  85. new_argc = 0;
  86. for (const char ** a = argv ; *a ; a++, new_argc++);
  87. new_argcp = &new_argc;
  88. if ((ret = init_stack(argv, envp, &new_argcp, &new_argp, &new_auxp)) < 0)
  89. return ret;
  90. SAVE_PROFILE_INTERVAL(alloc_new_stack_for_exec);
  91. SWITCH_STACK(new_argp);
  92. cur_thread = get_cur_thread();
  93. UPDATE_PROFILE_INTERVAL();
  94. DkVirtualMemoryFree(old_stack, old_stack_top - old_stack);
  95. DkVirtualMemoryFree(old_stack_red, old_stack - old_stack_red);
  96. if (bkeep_munmap(old_stack, old_stack_top - old_stack, 0) < 0 ||
  97. bkeep_munmap(old_stack_red, old_stack - old_stack_red, 0) < 0)
  98. BUG();
  99. remove_loaded_libraries();
  100. clean_link_map_list();
  101. SAVE_PROFILE_INTERVAL(unmap_loaded_binaries_for_exec);
  102. reset_brk();
  103. size_t count = DEFAULT_VMA_COUNT;
  104. struct shim_vma_val * vmas = malloc(sizeof(struct shim_vma_val) * count);
  105. if (!vmas)
  106. return -ENOMEM;
  107. retry_dump_vmas:
  108. ret = dump_all_vmas(vmas, count);
  109. if (ret == -EOVERFLOW) {
  110. struct shim_vma_val * new_vmas
  111. = malloc(sizeof(struct shim_vma_val) * count * 2);
  112. if (!new_vmas) {
  113. free(vmas);
  114. return -ENOMEM;
  115. }
  116. free(vmas);
  117. vmas = new_vmas;
  118. count *= 2;
  119. goto retry_dump_vmas;
  120. }
  121. if (ret < 0) {
  122. free(vmas);
  123. return ret;
  124. }
  125. count = ret;
  126. for (struct shim_vma_val * vma = vmas ; vma < vmas + count ; vma++) {
  127. /* Don't free the current stack */
  128. if (vma->addr == cur_thread->stack)
  129. continue;
  130. /* Free all the mapped VMAs */
  131. if (!(vma->flags & VMA_UNMAPPED))
  132. DkVirtualMemoryFree(vma->addr, vma->length);
  133. /* Remove the VMAs */
  134. bkeep_munmap(vma->addr, vma->length, vma->flags);
  135. }
  136. free_vma_val_array(vmas, count);
  137. SAVE_PROFILE_INTERVAL(unmap_all_vmas_for_exec);
  138. if ((ret = load_elf_object(cur_thread->exec, NULL, 0)) < 0)
  139. shim_terminate(ret);
  140. init_brk_from_executable(cur_thread->exec);
  141. load_elf_interp(cur_thread->exec);
  142. SAVE_PROFILE_INTERVAL(load_new_executable_for_exec);
  143. cur_thread->robust_list = NULL;
  144. #ifdef PROFILE
  145. if (ENTER_TIME)
  146. SAVE_PROFILE_INTERVAL_SINCE(syscall_execve, ENTER_TIME);
  147. #endif
  148. debug("execve: start execution\n");
  149. execute_elf_object(cur_thread->exec, new_argcp, new_argp, new_auxp);
  150. return 0;
  151. }
  152. #include <shim_checkpoint.h>
  153. DEFINE_PROFILE_CATEGORY(exec, );
  154. DEFINE_PROFILE_INTERVAL(search_and_check_file_for_exec, exec);
  155. DEFINE_PROFILE_INTERVAL(open_file_for_exec, exec);
  156. DEFINE_PROFILE_INTERVAL(close_CLOEXEC_files_for_exec, exec);
  157. static int migrate_execve (struct shim_cp_store * cpstore,
  158. struct shim_thread * thread,
  159. struct shim_process * process, va_list ap)
  160. {
  161. struct shim_handle_map * handle_map;
  162. const char ** envp = va_arg(ap, const char **);
  163. int ret;
  164. BEGIN_PROFILE_INTERVAL();
  165. if ((ret = dup_handle_map(&handle_map, thread->handle_map)) < 0)
  166. return ret;
  167. set_handle_map(thread, handle_map);
  168. if ((ret = close_cloexec_handle(handle_map)) < 0)
  169. return ret;
  170. SAVE_PROFILE_INTERVAL(close_CLOEXEC_files_for_exec);
  171. /* Now we start to migrate bookkeeping for exec.
  172. The data we need to migrate are:
  173. 1. cur_threadrent thread
  174. 2. cur_threadrent filesystem
  175. 3. handle mapping
  176. 4. each handle */
  177. BEGIN_MIGRATION_DEF(execve,
  178. struct shim_thread * thread,
  179. struct shim_process * proc,
  180. const char ** envp)
  181. {
  182. DEFINE_MIGRATE(process, proc, sizeof(struct shim_process));
  183. DEFINE_MIGRATE(all_mounts, NULL, 0);
  184. DEFINE_MIGRATE(running_thread, thread, sizeof(struct shim_thread));
  185. DEFINE_MIGRATE(handle_map, thread->handle_map,
  186. sizeof (struct shim_handle_map));
  187. DEFINE_MIGRATE(migratable, NULL, 0);
  188. DEFINE_MIGRATE(environ, envp, 0);
  189. }
  190. END_MIGRATION_DEF(execve)
  191. return START_MIGRATE(cpstore, execve, thread, process, envp);
  192. }
  193. int shim_do_execve (const char * file, const char ** argv,
  194. const char ** envp)
  195. {
  196. struct shim_thread * cur_thread = get_cur_thread();
  197. struct shim_dentry * dent = NULL;
  198. int ret = 0, argc = 0;
  199. if (test_user_string(file))
  200. return -EFAULT;
  201. for (const char** a = argv; /* no condition*/; a++, argc++) {
  202. if (test_user_memory(a, sizeof(*a), false))
  203. return -EFAULT;
  204. if (*a == NULL)
  205. break;
  206. if (test_user_string(*a))
  207. return -EFAULT;
  208. }
  209. if (!envp)
  210. envp = initial_envp;
  211. for (const char** e = envp; /* no condition*/; e++) {
  212. if (test_user_memory(e, sizeof(*e), false))
  213. return -EFAULT;
  214. if (*e == NULL)
  215. break;
  216. if (test_user_string(*e))
  217. return -EFAULT;
  218. }
  219. BEGIN_PROFILE_INTERVAL();
  220. DEFINE_LIST(sharg);
  221. struct sharg {
  222. LIST_TYPE(sharg) list;
  223. int len;
  224. char arg[0];
  225. };
  226. DEFINE_LISTP(sharg);
  227. LISTP_TYPE(sharg) shargs;
  228. INIT_LISTP(&shargs);
  229. reopen:
  230. /* XXX: Not sure what to do here yet */
  231. assert(cur_thread);
  232. if ((ret = path_lookupat(NULL, file, LOOKUP_OPEN, &dent, NULL)) < 0)
  233. return ret;
  234. struct shim_mount * fs = dent->fs;
  235. get_dentry(dent);
  236. if (!fs->d_ops->open) {
  237. ret = -EACCES;
  238. err:
  239. put_dentry(dent);
  240. return ret;
  241. }
  242. if (fs->d_ops->mode) {
  243. __kernel_mode_t mode;
  244. if ((ret = fs->d_ops->mode(dent, &mode, 1)) < 0)
  245. goto err;
  246. }
  247. SAVE_PROFILE_INTERVAL(search_and_check_file_for_exec);
  248. struct shim_handle * exec = NULL;
  249. if (!(exec = get_new_handle())) {
  250. ret = -ENOMEM;
  251. goto err;
  252. }
  253. set_handle_fs(exec, fs);
  254. exec->flags = O_RDONLY;
  255. exec->acc_mode = MAY_READ;
  256. ret = fs->d_ops->open(exec, dent, O_RDONLY);
  257. if (qstrempty(&exec->uri)) {
  258. put_handle(exec);
  259. return -EACCES;
  260. }
  261. int pathlen;
  262. char *path = dentry_get_path(dent, true, &pathlen);
  263. qstrsetstr(&exec->path, path, pathlen);
  264. if ((ret = check_elf_object(exec)) < 0 && ret != -EINVAL) {
  265. put_handle(exec);
  266. return ret;
  267. }
  268. if (ret == -EINVAL) { /* it's a shebang */
  269. LISTP_TYPE(sharg) new_shargs = LISTP_INIT;
  270. struct sharg * next = NULL;
  271. bool ended = false, started = false;
  272. char buf[80];
  273. do {
  274. ret = do_handle_read(exec, buf, 80);
  275. if (ret <= 0)
  276. break;
  277. char * s = buf, * c = buf, * e = buf + ret;
  278. if (!started) {
  279. if (ret < 2 || buf[0] != '#' || buf[1] != '!')
  280. break;
  281. s += 2;
  282. c += 2;
  283. started = true;
  284. }
  285. for (; c < e ; c++) {
  286. if (*c == ' ' || *c == '\n' || c == e - 1) {
  287. int l = (*c == ' ' || * c == '\n') ? c - s : e - s;
  288. if (next) {
  289. struct sharg * sh =
  290. __alloca(sizeof(struct sharg) + next->len + l + 1);
  291. sh->len = next->len + l;
  292. memcpy(sh->arg, next->arg, next->len);
  293. memcpy(sh->arg + next->len, s, l);
  294. sh->arg[next->len + l] = 0;
  295. next = sh;
  296. } else {
  297. next = __alloca(sizeof(struct sharg) + l + 1);
  298. next->len = l;
  299. memcpy(next->arg, s, l);
  300. next->arg[l] = 0;
  301. }
  302. if (*c == ' ' || *c == '\n') {
  303. INIT_LIST_HEAD(next, list);
  304. LISTP_ADD_TAIL(next, &new_shargs, list);
  305. next = NULL;
  306. s = c + 1;
  307. if (*c == '\n') {
  308. ended = true;
  309. break;
  310. }
  311. }
  312. }
  313. }
  314. } while (!ended);
  315. if (started) {
  316. if (next) {
  317. INIT_LIST_HEAD(next, list);
  318. LISTP_ADD_TAIL(next, &new_shargs, list);
  319. }
  320. struct sharg * first =
  321. LISTP_FIRST_ENTRY(&new_shargs, struct sharg, list);
  322. assert(first);
  323. debug("detected as script: run by %s\n", first->arg);
  324. file = first->arg;
  325. LISTP_SPLICE(&new_shargs, &shargs, list, sharg);
  326. put_handle(exec);
  327. goto reopen;
  328. }
  329. }
  330. SAVE_PROFILE_INTERVAL(open_file_for_exec);
  331. #if EXECVE_RTLD == 1
  332. if (!strcmp_static(PAL_CB(host_type), "Linux-SGX")) {
  333. int is_last = check_last_thread(cur_thread) == 0;
  334. if (is_last) {
  335. debug("execve() in the same process\n");
  336. return shim_do_execve_rtld(exec, argv, envp);
  337. }
  338. debug("execve() in a new process\n");
  339. }
  340. #endif
  341. INC_PROFILE_OCCURENCE(syscall_use_ipc);
  342. if (!LISTP_EMPTY(&shargs)) {
  343. struct sharg * sh;
  344. int shargc = 0, cnt = 0;
  345. LISTP_FOR_EACH_ENTRY(sh, &shargs, list)
  346. shargc++;
  347. const char ** new_argv =
  348. __alloca(sizeof(const char *) * (argc + shargc + 1));
  349. LISTP_FOR_EACH_ENTRY(sh, &shargs, list)
  350. new_argv[cnt++] = sh->arg;
  351. for (cnt = 0 ; cnt < argc ; cnt++)
  352. new_argv[shargc + cnt] = argv[cnt];
  353. new_argv[shargc + argc] = NULL;
  354. argv = new_argv;
  355. }
  356. lock(&cur_thread->lock);
  357. put_handle(cur_thread->exec);
  358. cur_thread->exec = exec;
  359. void * stack = cur_thread->stack;
  360. void * stack_top = cur_thread->stack_top;
  361. __libc_tcb_t * tcb = cur_thread->tcb;
  362. bool user_tcb = cur_thread->user_tcb;
  363. void * frameptr = cur_thread->frameptr;
  364. cur_thread->stack = NULL;
  365. cur_thread->stack_top = NULL;
  366. cur_thread->frameptr = NULL;
  367. cur_thread->tcb = NULL;
  368. cur_thread->user_tcb = false;
  369. cur_thread->in_vm = false;
  370. unlock(&cur_thread->lock);
  371. ret = do_migrate_process(&migrate_execve, exec, argv, cur_thread, envp);
  372. lock(&cur_thread->lock);
  373. cur_thread->stack = stack;
  374. cur_thread->stack_top = stack_top;
  375. cur_thread->frameptr = frameptr;
  376. cur_thread->tcb = tcb;
  377. cur_thread->user_tcb = user_tcb;
  378. if (ret < 0) {
  379. cur_thread->in_vm = true;
  380. unlock(&cur_thread->lock);
  381. return ret;
  382. }
  383. struct shim_handle_map * handle_map = cur_thread->handle_map;
  384. cur_thread->handle_map = NULL;
  385. unlock(&cur_thread->lock);
  386. if (handle_map)
  387. put_handle_map(handle_map);
  388. if (cur_thread->dummy)
  389. switch_dummy_thread(cur_thread);
  390. try_process_exit(0, 0);
  391. return 0;
  392. }