blob: 3e14ba25f678bf8869e005a34dc9742e117e90df [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/exec.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * #!-checking implemented by tytso.
9 */
10/*
11 * Demand-loading implemented 01.12.91 - no need to read anything but
12 * the header into memory. The inode of the executable is put into
13 * "current->executable", and page faults do the actual loading. Clean.
14 *
15 * Once more I can proudly say that linux stood up to being changed: it
16 * was less than 2 hours work to get demand-loading completely implemented.
17 *
18 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
19 * current->executable is only used by the procfs. This allows a dispatch
20 * table to check for several different types of binary formats. We keep
21 * trying until we recognize the file or we run out of supported binary
Aleksa Sarai613cc2b2016-12-21 16:26:24 +110022 * formats.
Linus Torvalds1da177e2005-04-16 15:20:36 -070023 */
24
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/slab.h>
26#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -040027#include <linux/fdtable.h>
Hugh Dickinsba92a432008-07-25 01:45:43 -070028#include <linux/mm.h>
Davidlohr Bueso615d6e82014-04-07 15:37:25 -070029#include <linux/vmacache.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/stat.h>
31#include <linux/fcntl.h>
Hugh Dickinsba92a432008-07-25 01:45:43 -070032#include <linux/swap.h>
Neil Horman74aadce2007-10-16 23:26:35 -070033#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/init.h>
Ingo Molnar6e84f312017-02-08 18:51:29 +010035#include <linux/sched/mm.h>
Ingo Molnarf7ccbae2017-02-08 18:51:30 +010036#include <linux/sched/coredump.h>
Ingo Molnar3f07c012017-02-08 18:51:30 +010037#include <linux/sched/signal.h>
Ingo Molnar6a3827d2017-02-08 18:51:31 +010038#include <linux/sched/numa_balancing.h>
Ingo Molnar29930022017-02-08 18:51:36 +010039#include <linux/sched/task.h>
Hugh Dickinsca5b1722008-07-28 15:46:18 -070040#include <linux/pagemap.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020041#include <linux/perf_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/highmem.h>
43#include <linux/spinlock.h>
44#include <linux/key.h>
45#include <linux/personality.h>
46#include <linux/binfmts.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/utsname.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080048#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/module.h>
50#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/mount.h>
52#include <linux/security.h>
53#include <linux/syscalls.h>
Jay Lan8f0ab512006-09-30 23:28:59 -070054#include <linux/tsacct_kern.h>
Matt Helsley9f460802005-11-07 00:59:16 -080055#include <linux/cn_proc.h>
Al Viro473ae302006-04-26 14:04:08 -040056#include <linux/audit.h>
Roland McGrath6341c392008-07-25 19:45:44 -070057#include <linux/tracehook.h>
Johannes Berg5f4123b2008-07-09 10:28:40 +020058#include <linux/kmod.h>
Eric Paris6110e3a2008-12-17 13:53:20 -050059#include <linux/fsnotify.h>
Al Viro5ad4e532009-03-29 19:50:06 -040060#include <linux/fs_struct.h>
Neil Horman61be2282009-09-23 15:56:58 -070061#include <linux/pipe_fs_i.h>
Ying Han3d5992d2010-10-26 14:21:23 -070062#include <linux/oom.h>
Oleg Nesterov0e028462011-03-06 18:02:54 +010063#include <linux/compat.h>
Mimi Zoharb44a7df2015-12-28 16:02:29 -050064#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080066#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067#include <asm/mmu_context.h>
Ollie Wildb6a2fea2007-07-19 01:48:16 -070068#include <asm/tlb.h>
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080069
70#include <trace/events/task.h>
David Howellsa6f76f22008-11-14 10:39:24 +110071#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
David Smith4ff16c22012-02-07 10:11:05 -060073#include <trace/events/sched.h>
74
Alan Coxd6e71142005-06-23 00:09:43 -070075int suid_dumpable = 0;
76
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070077static LIST_HEAD(formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070078static DEFINE_RWLOCK(binfmt_lock);
79
Al Viro8fc3dc52012-03-17 03:05:16 -040080void __register_binfmt(struct linux_binfmt * fmt, int insert)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081{
Al Viro8fc3dc52012-03-17 03:05:16 -040082 BUG_ON(!fmt);
Oleg Nesterov92eaa562013-09-11 14:24:42 -070083 if (WARN_ON(!fmt->load_binary))
84 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 write_lock(&binfmt_lock);
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070086 insert ? list_add(&fmt->lh, &formats) :
87 list_add_tail(&fmt->lh, &formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 write_unlock(&binfmt_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089}
90
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070091EXPORT_SYMBOL(__register_binfmt);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
Alexey Dobriyanf6b450d2007-10-16 23:26:04 -070093void unregister_binfmt(struct linux_binfmt * fmt)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094{
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 write_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070096 list_del(&fmt->lh);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 write_unlock(&binfmt_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098}
99
100EXPORT_SYMBOL(unregister_binfmt);
101
102static inline void put_binfmt(struct linux_binfmt * fmt)
103{
104 module_put(fmt->module);
105}
106
Eric W. Biederman90f85722015-06-29 14:42:03 -0500107bool path_noexec(const struct path *path)
108{
109 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
110 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
111}
112
Josh Triplett69369a72014-04-03 14:48:27 -0700113#ifdef CONFIG_USELIB
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114/*
115 * Note that a shared library must be both readable and executable due to
116 * security reasons.
117 *
118 * Also note that we take the address to load from from the file itself.
119 */
Heiko Carstens1e7bfb22009-01-14 14:14:29 +0100120SYSCALL_DEFINE1(uselib, const char __user *, library)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121{
Al Viro72c2d532013-09-22 16:27:52 -0400122 struct linux_binfmt *fmt;
Al Viro964bd182008-07-26 03:33:14 -0400123 struct file *file;
Jeff Layton91a27b22012-10-10 15:25:28 -0400124 struct filename *tmp = getname(library);
Al Viro964bd182008-07-26 03:33:14 -0400125 int error = PTR_ERR(tmp);
Al Viro47c805d2011-02-23 17:44:09 -0500126 static const struct open_flags uselib_flags = {
127 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
Al Viro62fb4a12015-12-26 22:33:24 -0500128 .acc_mode = MAY_READ | MAY_EXEC,
Al Virof9652e12013-06-11 08:23:01 +0400129 .intent = LOOKUP_OPEN,
130 .lookup_flags = LOOKUP_FOLLOW,
Al Viro47c805d2011-02-23 17:44:09 -0500131 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
Al Viro6e8341a2009-04-06 11:16:22 -0400133 if (IS_ERR(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 goto out;
135
Al Virof9652e12013-06-11 08:23:01 +0400136 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
Al Viro6e8341a2009-04-06 11:16:22 -0400137 putname(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 error = PTR_ERR(file);
139 if (IS_ERR(file))
140 goto out;
141
Al Viro6e8341a2009-04-06 11:16:22 -0400142 error = -EINVAL;
Al Viro496ad9a2013-01-23 17:07:38 -0500143 if (!S_ISREG(file_inode(file)->i_mode))
Al Viro6e8341a2009-04-06 11:16:22 -0400144 goto exit;
145
146 error = -EACCES;
Eric W. Biederman90f85722015-06-29 14:42:03 -0500147 if (path_noexec(&file->f_path))
Al Viro6e8341a2009-04-06 11:16:22 -0400148 goto exit;
149
Eric Paris2a12a9d2009-12-17 21:24:21 -0500150 fsnotify_open(file);
Eric Paris6110e3a2008-12-17 13:53:20 -0500151
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 error = -ENOEXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
Al Viro72c2d532013-09-22 16:27:52 -0400154 read_lock(&binfmt_lock);
155 list_for_each_entry(fmt, &formats, lh) {
156 if (!fmt->load_shlib)
157 continue;
158 if (!try_module_get(fmt->module))
159 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 read_unlock(&binfmt_lock);
Al Viro72c2d532013-09-22 16:27:52 -0400161 error = fmt->load_shlib(file);
162 read_lock(&binfmt_lock);
163 put_binfmt(fmt);
164 if (error != -ENOEXEC)
165 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 }
Al Viro72c2d532013-09-22 16:27:52 -0400167 read_unlock(&binfmt_lock);
Al Viro6e8341a2009-04-06 11:16:22 -0400168exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 fput(file);
170out:
171 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172}
Josh Triplett69369a72014-04-03 14:48:27 -0700173#endif /* #ifdef CONFIG_USELIB */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700175#ifdef CONFIG_MMU
Oleg Nesterovae6b5852011-03-06 18:03:11 +0100176/*
177 * The nascent bprm->mm is not visible until exec_mmap() but it can
178 * use a lot of memory, account these pages in current->mm temporary
179 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
180 * change the counter back via acct_arg_size(0).
181 */
Oleg Nesterov0e028462011-03-06 18:02:54 +0100182static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100183{
184 struct mm_struct *mm = current->mm;
185 long diff = (long)(pages - bprm->vma_pages);
186
187 if (!mm || !diff)
188 return;
189
190 bprm->vma_pages = pages;
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100191 add_mm_counter(mm, MM_ANONPAGES, diff);
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100192}
193
Oleg Nesterov0e028462011-03-06 18:02:54 +0100194static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700195 int write)
196{
197 struct page *page;
198 int ret;
Lorenzo Stoakes9beae1e2016-10-13 01:20:17 +0100199 unsigned int gup_flags = FOLL_FORCE;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700200
201#ifdef CONFIG_STACK_GROWSUP
202 if (write) {
Michal Hockod05f3162011-05-24 17:11:44 -0700203 ret = expand_downwards(bprm->vma, pos);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700204 if (ret < 0)
205 return NULL;
206 }
207#endif
Lorenzo Stoakes9beae1e2016-10-13 01:20:17 +0100208
209 if (write)
210 gup_flags |= FOLL_WRITE;
211
Dave Hansen1e987792016-02-12 13:01:54 -0800212 /*
213 * We are doing an exec(). 'current' is the process
214 * doing the exec and bprm->mm is the new process's mm.
215 */
Lorenzo Stoakes9beae1e2016-10-13 01:20:17 +0100216 ret = get_user_pages_remote(current, bprm->mm, pos, 1, gup_flags,
Lorenzo Stoakes5b56d492016-12-14 15:06:52 -0800217 &page, NULL, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700218 if (ret <= 0)
219 return NULL;
220
221 if (write) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700222 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
Kees Cookda029c12017-07-07 11:57:29 -0700223 unsigned long ptr_size, limit;
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800224
Kees Cook98da7d02017-06-23 15:08:57 -0700225 /*
226 * Since the stack will hold pointers to the strings, we
227 * must account for them as well.
228 *
229 * The size calculation is the entire vma while each arg page is
230 * built, so each time we get here it's calculating how far it
231 * is currently (rather than each call being just the newly
232 * added size from the arg page). As a result, we need to
233 * always add the entire size of the pointers, so that on the
234 * last call to get_arg_page() we'll actually have the entire
235 * correct size.
236 */
237 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
238 if (ptr_size > ULONG_MAX - size)
239 goto fail;
240 size += ptr_size;
241
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100242 acct_arg_size(bprm, size / PAGE_SIZE);
243
Linus Torvaldsa64e7152008-03-03 10:12:14 -0800244 /*
245 * We've historically supported up to 32 pages (ARG_MAX)
246 * of argument strings even with small stacks
247 */
248 if (size <= ARG_MAX)
249 return page;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700250
251 /*
Kees Cookda029c12017-07-07 11:57:29 -0700252 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
253 * (whichever is smaller) for the argv+env strings.
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700254 * This ensures that:
255 * - the remaining binfmt code will not run out of stack space,
256 * - the program will have a reasonable amount of stack left
257 * to work from.
258 */
Kees Cookda029c12017-07-07 11:57:29 -0700259 limit = _STK_LIM / 4 * 3;
260 limit = min(limit, rlimit(RLIMIT_STACK) / 4);
261 if (size > limit)
Kees Cook98da7d02017-06-23 15:08:57 -0700262 goto fail;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700263 }
264
265 return page;
Kees Cook98da7d02017-06-23 15:08:57 -0700266
267fail:
268 put_page(page);
269 return NULL;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700270}
271
272static void put_arg_page(struct page *page)
273{
274 put_page(page);
275}
276
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700277static void free_arg_pages(struct linux_binprm *bprm)
278{
279}
280
281static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
282 struct page *page)
283{
284 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
285}
286
287static int __bprm_mm_init(struct linux_binprm *bprm)
288{
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800289 int err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700290 struct vm_area_struct *vma = NULL;
291 struct mm_struct *mm = bprm->mm;
292
293 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
294 if (!vma)
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800295 return -ENOMEM;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700296
Michal Hockof268dfe2016-05-23 16:26:02 -0700297 if (down_write_killable(&mm->mmap_sem)) {
298 err = -EINTR;
299 goto err_free;
300 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700301 vma->vm_mm = mm;
302
303 /*
304 * Place the stack at the largest stack address the architecture
305 * supports. Later, we'll move this to an appropriate place. We don't
306 * use STACK_TOP because that can depend on attributes which aren't
307 * configured yet.
308 */
Michal Hockoaacb3d12011-07-26 16:08:40 -0700309 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700310 vma->vm_end = STACK_TOP_MAX;
311 vma->vm_start = vma->vm_end - PAGE_SIZE;
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -0700312 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
Coly Li3ed75eb2007-10-18 23:39:15 -0700313 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Rik van Riel5beb4932010-03-05 13:42:07 -0800314 INIT_LIST_HEAD(&vma->anon_vma_chain);
Tavis Ormandy462e635e2010-12-09 15:29:42 +0100315
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700316 err = insert_vm_struct(mm, vma);
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800317 if (err)
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700318 goto err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700319
320 mm->stack_vm = mm->total_vm = 1;
Dave Hansenfe3d1972014-11-14 07:18:29 -0800321 arch_bprm_mm_init(mm, vma);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700322 up_write(&mm->mmap_sem);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700323 bprm->p = vma->vm_end - sizeof(void *);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700324 return 0;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700325err:
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800326 up_write(&mm->mmap_sem);
Michal Hockof268dfe2016-05-23 16:26:02 -0700327err_free:
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800328 bprm->vma = NULL;
329 kmem_cache_free(vm_area_cachep, vma);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700330 return err;
331}
332
333static bool valid_arg_len(struct linux_binprm *bprm, long len)
334{
335 return len <= MAX_ARG_STRLEN;
336}
337
338#else
339
Oleg Nesterov0e028462011-03-06 18:02:54 +0100340static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100341{
342}
343
Oleg Nesterov0e028462011-03-06 18:02:54 +0100344static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700345 int write)
346{
347 struct page *page;
348
349 page = bprm->page[pos / PAGE_SIZE];
350 if (!page && write) {
351 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
352 if (!page)
353 return NULL;
354 bprm->page[pos / PAGE_SIZE] = page;
355 }
356
357 return page;
358}
359
360static void put_arg_page(struct page *page)
361{
362}
363
364static void free_arg_page(struct linux_binprm *bprm, int i)
365{
366 if (bprm->page[i]) {
367 __free_page(bprm->page[i]);
368 bprm->page[i] = NULL;
369 }
370}
371
372static void free_arg_pages(struct linux_binprm *bprm)
373{
374 int i;
375
376 for (i = 0; i < MAX_ARG_PAGES; i++)
377 free_arg_page(bprm, i);
378}
379
380static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
381 struct page *page)
382{
383}
384
385static int __bprm_mm_init(struct linux_binprm *bprm)
386{
387 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
388 return 0;
389}
390
391static bool valid_arg_len(struct linux_binprm *bprm, long len)
392{
393 return len <= bprm->p;
394}
395
396#endif /* CONFIG_MMU */
397
398/*
399 * Create a new mm_struct and populate it with a temporary stack
400 * vm_area_struct. We don't have enough context at this point to set the stack
401 * flags, permissions, and offset, so we use temporary values. We'll update
402 * them later in setup_arg_pages().
403 */
Yuanhan Liu9cc64ce2013-02-20 13:16:01 +1100404static int bprm_mm_init(struct linux_binprm *bprm)
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700405{
406 int err;
407 struct mm_struct *mm = NULL;
408
409 bprm->mm = mm = mm_alloc();
410 err = -ENOMEM;
411 if (!mm)
412 goto err;
413
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700414 err = __bprm_mm_init(bprm);
415 if (err)
416 goto err;
417
418 return 0;
419
420err:
421 if (mm) {
422 bprm->mm = NULL;
423 mmdrop(mm);
424 }
425
426 return err;
427}
428
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100429struct user_arg_ptr {
Oleg Nesterov0e028462011-03-06 18:02:54 +0100430#ifdef CONFIG_COMPAT
431 bool is_compat;
432#endif
433 union {
434 const char __user *const __user *native;
435#ifdef CONFIG_COMPAT
Al Viro38b983b2012-09-30 13:38:55 -0400436 const compat_uptr_t __user *compat;
Oleg Nesterov0e028462011-03-06 18:02:54 +0100437#endif
438 } ptr;
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100439};
440
441static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100442{
Oleg Nesterov0e028462011-03-06 18:02:54 +0100443 const char __user *native;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100444
Oleg Nesterov0e028462011-03-06 18:02:54 +0100445#ifdef CONFIG_COMPAT
446 if (unlikely(argv.is_compat)) {
447 compat_uptr_t compat;
448
449 if (get_user(compat, argv.ptr.compat + nr))
450 return ERR_PTR(-EFAULT);
451
452 return compat_ptr(compat);
453 }
454#endif
455
456 if (get_user(native, argv.ptr.native + nr))
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100457 return ERR_PTR(-EFAULT);
458
Oleg Nesterov0e028462011-03-06 18:02:54 +0100459 return native;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100460}
461
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462/*
463 * count() counts the number of strings in array ARGV.
464 */
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100465static int count(struct user_arg_ptr argv, int max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466{
467 int i = 0;
468
Oleg Nesterov0e028462011-03-06 18:02:54 +0100469 if (argv.ptr.native != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 for (;;) {
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100471 const char __user *p = get_user_arg_ptr(argv, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 if (!p)
474 break;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100475
476 if (IS_ERR(p))
477 return -EFAULT;
478
Xi Wang6d92d4f2013-01-11 14:31:48 -0800479 if (i >= max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 return -E2BIG;
Xi Wang6d92d4f2013-01-11 14:31:48 -0800481 ++i;
Roland McGrath9aea5a62010-09-07 19:37:06 -0700482
483 if (fatal_signal_pending(current))
484 return -ERESTARTNOHAND;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 cond_resched();
486 }
487 }
488 return i;
489}
490
491/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700492 * 'copy_strings()' copies argument/environment strings from the old
493 * processes's memory to the new process's stack. The call to get_user_pages()
494 * ensures the destination page is created and not swapped out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 */
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100496static int copy_strings(int argc, struct user_arg_ptr argv,
Adrian Bunk75c96f82005-05-05 16:16:09 -0700497 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498{
499 struct page *kmapped_page = NULL;
500 char *kaddr = NULL;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700501 unsigned long kpos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 int ret;
503
504 while (argc-- > 0) {
David Howellsd7627462010-08-17 23:52:56 +0100505 const char __user *str;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 int len;
507 unsigned long pos;
508
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100509 ret = -EFAULT;
510 str = get_user_arg_ptr(argv, argc);
511 if (IS_ERR(str))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100514 len = strnlen_user(str, MAX_ARG_STRLEN);
515 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 goto out;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100517
518 ret = -E2BIG;
519 if (!valid_arg_len(bprm, len))
520 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700522 /* We're going to work our way backwords. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 pos = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700524 str += len;
525 bprm->p -= len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
527 while (len > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 int offset, bytes_to_copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
Roland McGrath9aea5a62010-09-07 19:37:06 -0700530 if (fatal_signal_pending(current)) {
531 ret = -ERESTARTNOHAND;
532 goto out;
533 }
Roland McGrath7993bc12010-09-07 19:36:28 -0700534 cond_resched();
535
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 offset = pos % PAGE_SIZE;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700537 if (offset == 0)
538 offset = PAGE_SIZE;
539
540 bytes_to_copy = offset;
541 if (bytes_to_copy > len)
542 bytes_to_copy = len;
543
544 offset -= bytes_to_copy;
545 pos -= bytes_to_copy;
546 str -= bytes_to_copy;
547 len -= bytes_to_copy;
548
549 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
550 struct page *page;
551
552 page = get_arg_page(bprm, pos, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 if (!page) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700554 ret = -E2BIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 goto out;
556 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700558 if (kmapped_page) {
559 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700561 put_arg_page(kmapped_page);
562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 kmapped_page = page;
564 kaddr = kmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700565 kpos = pos & PAGE_MASK;
566 flush_arg_page(bprm, kpos, kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700568 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 ret = -EFAULT;
570 goto out;
571 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 }
573 }
574 ret = 0;
575out:
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700576 if (kmapped_page) {
577 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700579 put_arg_page(kmapped_page);
580 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 return ret;
582}
583
584/*
585 * Like copy_strings, but get argv and its values from kernel memory.
586 */
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100587int copy_strings_kernel(int argc, const char *const *__argv,
David Howellsd7627462010-08-17 23:52:56 +0100588 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589{
590 int r;
591 mm_segment_t oldfs = get_fs();
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100592 struct user_arg_ptr argv = {
Oleg Nesterov0e028462011-03-06 18:02:54 +0100593 .ptr.native = (const char __user *const __user *)__argv,
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100594 };
595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 set_fs(KERNEL_DS);
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100597 r = copy_strings(argc, argv, bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 set_fs(oldfs);
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100599
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 return r;
601}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602EXPORT_SYMBOL(copy_strings_kernel);
603
604#ifdef CONFIG_MMU
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700605
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700607 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
608 * the binfmt code determines where the new stack should reside, we shift it to
609 * its final location. The process proceeds as follows:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 *
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700611 * 1) Use shift to calculate the new vma endpoints.
612 * 2) Extend vma to cover both the old and new ranges. This ensures the
613 * arguments passed to subsequent functions are consistent.
614 * 3) Move vma's page tables to the new range.
615 * 4) Free up any cleared pgd range.
616 * 5) Shrink the vma to cover only the new range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700618static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
620 struct mm_struct *mm = vma->vm_mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700621 unsigned long old_start = vma->vm_start;
622 unsigned long old_end = vma->vm_end;
623 unsigned long length = old_end - old_start;
624 unsigned long new_start = old_start - shift;
625 unsigned long new_end = old_end - shift;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700626 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700628 BUG_ON(new_start > new_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700630 /*
631 * ensure there are no vmas between where we want to go
632 * and where we are
633 */
634 if (vma != find_vma(mm, new_start))
635 return -EFAULT;
636
637 /*
638 * cover the whole range: [new_start, old_end)
639 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800640 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
641 return -ENOMEM;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700642
643 /*
644 * move the page tables downwards, on failure we rely on
645 * process cleanup to remove whatever mess we made.
646 */
647 if (length != move_page_tables(vma, old_start,
Michel Lespinasse38a76012012-10-08 16:31:50 -0700648 vma, new_start, length, false))
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700649 return -ENOMEM;
650
651 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -0700652 tlb_gather_mmu(&tlb, mm, old_start, old_end);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700653 if (new_end > old_start) {
654 /*
655 * when the old and new regions overlap clear from new_end.
656 */
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700657 free_pgd_range(&tlb, new_end, old_end, new_end,
Hugh Dickins6ee86302013-04-29 15:07:44 -0700658 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700659 } else {
660 /*
661 * otherwise, clean from old_start; this is done to not touch
662 * the address space in [new_end, old_start) some architectures
663 * have constraints on va-space that make this illegal (IA64) -
664 * for the others its just a little faster.
665 */
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700666 free_pgd_range(&tlb, old_start, old_end, new_end,
Hugh Dickins6ee86302013-04-29 15:07:44 -0700667 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 }
Linus Torvalds2b047252013-08-15 11:42:25 -0700669 tlb_finish_mmu(&tlb, old_start, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700671 /*
Rik van Riel5beb4932010-03-05 13:42:07 -0800672 * Shrink the vma to just the new range. Always succeeds.
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700673 */
674 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
675
676 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677}
678
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700679/*
680 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
681 * the stack is optionally relocated, and some extra space is added.
682 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683int setup_arg_pages(struct linux_binprm *bprm,
684 unsigned long stack_top,
685 int executable_stack)
686{
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700687 unsigned long ret;
688 unsigned long stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 struct mm_struct *mm = current->mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700690 struct vm_area_struct *vma = bprm->vma;
691 struct vm_area_struct *prev = NULL;
692 unsigned long vm_flags;
693 unsigned long stack_base;
Michael Neuling803bf5e2010-02-10 13:56:42 -0800694 unsigned long stack_size;
695 unsigned long stack_expand;
696 unsigned long rlim_stack;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
698#ifdef CONFIG_STACK_GROWSUP
James Hogand71f2902014-05-13 23:58:24 +0100699 /* Limit stack size */
Jiri Slabyd554ed892010-03-05 13:42:42 -0800700 stack_base = rlimit_max(RLIMIT_STACK);
James Hogand71f2902014-05-13 23:58:24 +0100701 if (stack_base > STACK_SIZE_MAX)
702 stack_base = STACK_SIZE_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
Helge Dellerd045c772015-05-11 22:01:27 +0200704 /* Add space for stack randomization. */
705 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
706
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700707 /* Make sure we didn't let the argument array grow too large. */
708 if (vma->vm_end - vma->vm_start > stack_base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 return -ENOMEM;
710
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700711 stack_base = PAGE_ALIGN(stack_top - stack_base);
712
713 stack_shift = vma->vm_start - stack_base;
714 mm->arg_start = bprm->p - stack_shift;
715 bprm->p = vma->vm_end - stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716#else
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700717 stack_top = arch_align_stack(stack_top);
718 stack_top = PAGE_ALIGN(stack_top);
Roland McGrath1b528182010-09-07 19:35:49 -0700719
720 if (unlikely(stack_top < mmap_min_addr) ||
721 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
722 return -ENOMEM;
723
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700724 stack_shift = vma->vm_end - stack_top;
725
726 bprm->p -= stack_shift;
727 mm->arg_start = bprm->p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728#endif
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700729
730 if (bprm->loader)
731 bprm->loader -= stack_shift;
732 bprm->exec -= stack_shift;
733
Michal Hockof268dfe2016-05-23 16:26:02 -0700734 if (down_write_killable(&mm->mmap_sem))
735 return -EINTR;
736
Hugh Dickins96a8e132008-07-10 21:19:20 +0100737 vm_flags = VM_STACK_FLAGS;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700738
739 /*
740 * Adjust stack execute permissions; explicitly enable for
741 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
742 * (arch default) otherwise.
743 */
744 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
745 vm_flags |= VM_EXEC;
746 else if (executable_stack == EXSTACK_DISABLE_X)
747 vm_flags &= ~VM_EXEC;
748 vm_flags |= mm->def_flags;
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700749 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700750
751 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
752 vm_flags);
753 if (ret)
754 goto out_unlock;
755 BUG_ON(prev != vma);
756
757 /* Move stack pages down in memory. */
758 if (stack_shift) {
759 ret = shift_arg_pages(vma, stack_shift);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800760 if (ret)
761 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
763
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700764 /* mprotect_fixup is overkill to remove the temporary stack flags */
765 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
766
Michael Neuling5ef097d2010-03-05 13:42:57 -0800767 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
Michael Neuling803bf5e2010-02-10 13:56:42 -0800768 stack_size = vma->vm_end - vma->vm_start;
769 /*
770 * Align this down to a page boundary as expand_stack
771 * will align it up.
772 */
773 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700774#ifdef CONFIG_STACK_GROWSUP
Michael Neuling803bf5e2010-02-10 13:56:42 -0800775 if (stack_size + stack_expand > rlim_stack)
776 stack_base = vma->vm_start + rlim_stack;
777 else
778 stack_base = vma->vm_end + stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700779#else
Michael Neuling803bf5e2010-02-10 13:56:42 -0800780 if (stack_size + stack_expand > rlim_stack)
781 stack_base = vma->vm_end - rlim_stack;
782 else
783 stack_base = vma->vm_start - stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700784#endif
Eric B Munson3af9e852010-05-18 15:30:49 +0100785 current->mm->start_stack = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700786 ret = expand_stack(vma, stack_base);
787 if (ret)
788 ret = -EFAULT;
789
790out_unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 up_write(&mm->mmap_sem);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800792 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794EXPORT_SYMBOL(setup_arg_pages);
795
Nicolas Pitre7e7ec6a2016-07-24 11:30:18 -0400796#else
797
798/*
799 * Transfer the program arguments and environment from the holding pages
800 * onto the stack. The provided stack pointer is adjusted accordingly.
801 */
802int transfer_args_to_stack(struct linux_binprm *bprm,
803 unsigned long *sp_location)
804{
805 unsigned long index, stop, sp;
806 int ret = 0;
807
808 stop = bprm->p >> PAGE_SHIFT;
809 sp = *sp_location;
810
811 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
812 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
813 char *src = kmap(bprm->page[index]) + offset;
814 sp -= PAGE_SIZE - offset;
815 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
816 ret = -EFAULT;
817 kunmap(bprm->page[index]);
818 if (ret)
819 goto out;
820 }
821
822 *sp_location = sp;
823
824out:
825 return ret;
826}
827EXPORT_SYMBOL(transfer_args_to_stack);
828
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829#endif /* CONFIG_MMU */
830
David Drysdale51f39a12014-12-12 16:57:29 -0800831static struct file *do_open_execat(int fd, struct filename *name, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 struct file *file;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200834 int err;
David Drysdale51f39a12014-12-12 16:57:29 -0800835 struct open_flags open_exec_flags = {
Al Viro47c805d2011-02-23 17:44:09 -0500836 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
Al Viro62fb4a12015-12-26 22:33:24 -0500837 .acc_mode = MAY_EXEC,
Al Virof9652e12013-06-11 08:23:01 +0400838 .intent = LOOKUP_OPEN,
839 .lookup_flags = LOOKUP_FOLLOW,
Al Viro47c805d2011-02-23 17:44:09 -0500840 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
David Drysdale51f39a12014-12-12 16:57:29 -0800842 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
843 return ERR_PTR(-EINVAL);
844 if (flags & AT_SYMLINK_NOFOLLOW)
845 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
846 if (flags & AT_EMPTY_PATH)
847 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
848
849 file = do_filp_open(fd, name, &open_exec_flags);
Al Viro6e8341a2009-04-06 11:16:22 -0400850 if (IS_ERR(file))
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200851 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200853 err = -EACCES;
Al Viro496ad9a2013-01-23 17:07:38 -0500854 if (!S_ISREG(file_inode(file)->i_mode))
Al Viro6e8341a2009-04-06 11:16:22 -0400855 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200856
Eric W. Biederman90f85722015-06-29 14:42:03 -0500857 if (path_noexec(&file->f_path))
Al Viro6e8341a2009-04-06 11:16:22 -0400858 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200859
860 err = deny_write_access(file);
Al Viro6e8341a2009-04-06 11:16:22 -0400861 if (err)
862 goto exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
David Drysdale51f39a12014-12-12 16:57:29 -0800864 if (name->name[0] != '\0')
865 fsnotify_open(file);
866
Al Viro6e8341a2009-04-06 11:16:22 -0400867out:
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200868 return file;
869
Al Viro6e8341a2009-04-06 11:16:22 -0400870exit:
871 fput(file);
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200872 return ERR_PTR(err);
873}
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800874
875struct file *open_exec(const char *name)
876{
Paul Moore51689102015-01-22 00:00:03 -0500877 struct filename *filename = getname_kernel(name);
878 struct file *f = ERR_CAST(filename);
879
880 if (!IS_ERR(filename)) {
881 f = do_open_execat(AT_FDCWD, filename, 0);
882 putname(filename);
883 }
884 return f;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800885}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886EXPORT_SYMBOL(open_exec);
887
Mimi Zoharb44a7df2015-12-28 16:02:29 -0500888int kernel_read_file(struct file *file, void **buf, loff_t *size,
Mimi Zoharbc8ca5b2016-01-24 10:07:32 -0500889 loff_t max_size, enum kernel_read_file_id id)
Mimi Zoharb44a7df2015-12-28 16:02:29 -0500890{
891 loff_t i_size, pos;
892 ssize_t bytes = 0;
893 int ret;
894
895 if (!S_ISREG(file_inode(file)->i_mode) || max_size < 0)
896 return -EINVAL;
897
Mimi Zohar39eeb4f2016-01-30 22:23:26 -0500898 ret = security_kernel_read_file(file, id);
899 if (ret)
900 return ret;
901
Dmitry Kasatkin39d637a2014-10-26 12:42:07 +0200902 ret = deny_write_access(file);
903 if (ret)
904 return ret;
905
Mimi Zoharb44a7df2015-12-28 16:02:29 -0500906 i_size = i_size_read(file_inode(file));
Dmitry Kasatkin39d637a2014-10-26 12:42:07 +0200907 if (max_size > 0 && i_size > max_size) {
908 ret = -EFBIG;
909 goto out;
910 }
911 if (i_size <= 0) {
912 ret = -EINVAL;
913 goto out;
914 }
Mimi Zoharb44a7df2015-12-28 16:02:29 -0500915
Stephen Boyda098ecd2016-08-02 14:04:28 -0700916 if (id != READING_FIRMWARE_PREALLOC_BUFFER)
917 *buf = vmalloc(i_size);
Dmitry Kasatkin39d637a2014-10-26 12:42:07 +0200918 if (!*buf) {
919 ret = -ENOMEM;
920 goto out;
921 }
Mimi Zoharb44a7df2015-12-28 16:02:29 -0500922
923 pos = 0;
924 while (pos < i_size) {
Christoph Hellwigbdd1d2d2017-09-01 17:39:13 +0200925 bytes = kernel_read(file, *buf + pos, i_size - pos, &pos);
Mimi Zoharb44a7df2015-12-28 16:02:29 -0500926 if (bytes < 0) {
927 ret = bytes;
928 goto out;
929 }
930
931 if (bytes == 0)
932 break;
Mimi Zoharb44a7df2015-12-28 16:02:29 -0500933 }
934
935 if (pos != i_size) {
936 ret = -EIO;
Dmitry Kasatkin39d637a2014-10-26 12:42:07 +0200937 goto out_free;
Mimi Zoharb44a7df2015-12-28 16:02:29 -0500938 }
939
Mimi Zoharbc8ca5b2016-01-24 10:07:32 -0500940 ret = security_kernel_post_read_file(file, *buf, i_size, id);
Mimi Zoharb44a7df2015-12-28 16:02:29 -0500941 if (!ret)
942 *size = pos;
943
Dmitry Kasatkin39d637a2014-10-26 12:42:07 +0200944out_free:
Mimi Zoharb44a7df2015-12-28 16:02:29 -0500945 if (ret < 0) {
Stephen Boyda098ecd2016-08-02 14:04:28 -0700946 if (id != READING_FIRMWARE_PREALLOC_BUFFER) {
947 vfree(*buf);
948 *buf = NULL;
949 }
Mimi Zoharb44a7df2015-12-28 16:02:29 -0500950 }
Dmitry Kasatkin39d637a2014-10-26 12:42:07 +0200951
952out:
953 allow_write_access(file);
Mimi Zoharb44a7df2015-12-28 16:02:29 -0500954 return ret;
955}
956EXPORT_SYMBOL_GPL(kernel_read_file);
957
Mimi Zohar711aab12017-09-12 22:45:33 -0400958int kernel_read_file_from_path(const char *path, void **buf, loff_t *size,
Mimi Zohar09596b92015-11-19 12:39:22 -0500959 loff_t max_size, enum kernel_read_file_id id)
960{
961 struct file *file;
962 int ret;
963
964 if (!path || !*path)
965 return -EINVAL;
966
967 file = filp_open(path, O_RDONLY, 0);
968 if (IS_ERR(file))
969 return PTR_ERR(file);
970
971 ret = kernel_read_file(file, buf, size, max_size, id);
972 fput(file);
973 return ret;
974}
975EXPORT_SYMBOL_GPL(kernel_read_file_from_path);
976
Mimi Zoharb844f0e2016-02-01 08:36:21 -0500977int kernel_read_file_from_fd(int fd, void **buf, loff_t *size, loff_t max_size,
978 enum kernel_read_file_id id)
979{
980 struct fd f = fdget(fd);
981 int ret = -EBADF;
982
983 if (!f.file)
984 goto out;
985
986 ret = kernel_read_file(f.file, buf, size, max_size, id);
987out:
988 fdput(f);
989 return ret;
990}
991EXPORT_SYMBOL_GPL(kernel_read_file_from_fd);
992
Al Viro3dc20cb2013-04-13 20:31:37 -0400993ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
994{
Al Viroec695572014-02-04 19:08:21 -0500995 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
Al Viro3dc20cb2013-04-13 20:31:37 -0400996 if (res > 0)
997 flush_icache_range(addr, addr + len);
998 return res;
999}
1000EXPORT_SYMBOL(read_code);
1001
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002static int exec_mmap(struct mm_struct *mm)
1003{
1004 struct task_struct *tsk;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001005 struct mm_struct *old_mm, *active_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006
1007 /* Notify parent that we're no longer interested in the old VM */
1008 tsk = current;
1009 old_mm = current->mm;
1010 mm_release(tsk, old_mm);
1011
1012 if (old_mm) {
Konstantin Khlebnikov4fe7efd2012-06-20 12:53:01 -07001013 sync_mm_rss(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 /*
1015 * Make sure that if there is a core dump in progress
1016 * for the old mm, we get out and die instead of going
1017 * through with the exec. We must hold mmap_sem around
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001018 * checking core_state and changing tsk->mm.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 */
1020 down_read(&old_mm->mmap_sem);
Oleg Nesterov999d9fc2008-07-25 01:47:41 -07001021 if (unlikely(old_mm->core_state)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 up_read(&old_mm->mmap_sem);
1023 return -EINTR;
1024 }
1025 }
1026 task_lock(tsk);
1027 active_mm = tsk->active_mm;
1028 tsk->mm = mm;
1029 tsk->active_mm = mm;
1030 activate_mm(active_mm, mm);
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001031 tsk->mm->vmacache_seqnum = 0;
1032 vmacache_flush(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 task_unlock(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 if (old_mm) {
1035 up_read(&old_mm->mmap_sem);
Eric Sesterhenn7dddb122006-04-01 01:13:38 +02001036 BUG_ON(active_mm != old_mm);
Oleg Nesterov701085b2012-03-19 17:04:01 +01001037 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
Balbir Singh31a78f22008-09-28 23:09:31 +01001038 mm_update_next_owner(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 mmput(old_mm);
1040 return 0;
1041 }
1042 mmdrop(active_mm);
1043 return 0;
1044}
1045
1046/*
1047 * This function makes sure the current process has its own signal table,
1048 * so that flush_signal_handlers can later reset the handlers without
1049 * disturbing other processes. (Other processes might share the signal
1050 * table via the CLONE_SIGHAND option to clone().)
1051 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001052static int de_thread(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053{
1054 struct signal_struct *sig = tsk->signal;
Oleg Nesterovb2c903b2007-10-16 23:27:22 -07001055 struct sighand_struct *oldsighand = tsk->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 spinlock_t *lock = &oldsighand->siglock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001058 if (thread_group_empty(tsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 goto no_thread_group;
1060
1061 /*
1062 * Kill all other threads in the thread group.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 spin_lock_irq(lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001065 if (signal_group_exit(sig)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 /*
1067 * Another group action in progress, just
1068 * return so that the signal is processed.
1069 */
1070 spin_unlock_irq(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071 return -EAGAIN;
1072 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073
Oleg Nesterovd3441932010-05-26 14:43:11 -07001074 sig->group_exit_task = tsk;
1075 sig->notify_count = zap_other_threads(tsk);
1076 if (!thread_group_leader(tsk))
1077 sig->notify_count--;
1078
1079 while (sig->notify_count) {
Oleg Nesterovd5bbd432012-10-08 19:13:01 +02001080 __set_current_state(TASK_KILLABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 spin_unlock_irq(lock);
1082 schedule();
Oleg Nesterovd5bbd432012-10-08 19:13:01 +02001083 if (unlikely(__fatal_signal_pending(tsk)))
1084 goto killed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 spin_lock_irq(lock);
1086 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 spin_unlock_irq(lock);
1088
1089 /*
1090 * At this point all other threads have exited, all we have to
1091 * do is to wait for the thread group leader to become inactive,
1092 * and to assume its PID:
1093 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001094 if (!thread_group_leader(tsk)) {
Oleg Nesterov81879262008-12-01 14:18:16 -08001095 struct task_struct *leader = tsk->group_leader;
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001096
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001097 for (;;) {
Ingo Molnar780de9d2017-02-02 11:50:56 +01001098 cgroup_threadgroup_change_begin(tsk);
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001099 write_lock_irq(&tasklist_lock);
Kirill Tkhaidfcce792015-04-16 12:48:01 -07001100 /*
1101 * Do this under tasklist_lock to ensure that
1102 * exit_notify() can't miss ->group_exit_task
1103 */
1104 sig->notify_count = -1;
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001105 if (likely(leader->exit_state))
1106 break;
Oleg Nesterovd5bbd432012-10-08 19:13:01 +02001107 __set_current_state(TASK_KILLABLE);
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001108 write_unlock_irq(&tasklist_lock);
Ingo Molnar780de9d2017-02-02 11:50:56 +01001109 cgroup_threadgroup_change_end(tsk);
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001110 schedule();
Oleg Nesterovd5bbd432012-10-08 19:13:01 +02001111 if (unlikely(__fatal_signal_pending(tsk)))
1112 goto killed;
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001113 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114
Roland McGrathf5e90282006-04-10 22:54:16 -07001115 /*
1116 * The only record we have of the real-time age of a
1117 * process, regardless of execs it's done, is start_time.
1118 * All the past CPU time is accumulated in signal_struct
1119 * from sister threads now dead. But in this non-leader
1120 * exec, nothing survives from the original leader thread,
1121 * whose birth marks the true age of this process now.
1122 * When we take on its identity by switching to its PID, we
1123 * also take its birthdate (always earlier than our own).
1124 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001125 tsk->start_time = leader->start_time;
Oleg Nesterov266b7a02013-07-03 15:08:35 -07001126 tsk->real_start_time = leader->real_start_time;
Roland McGrathf5e90282006-04-10 22:54:16 -07001127
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -07001128 BUG_ON(!same_thread_group(leader, tsk));
1129 BUG_ON(has_group_leader_pid(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 /*
1131 * An exec() starts a new thread group with the
1132 * TGID of the previous thread group. Rehash the
1133 * two threads with a switched PID, and release
1134 * the former thread group leader:
1135 */
Eric W. Biedermand73d6522006-03-28 16:11:03 -08001136
1137 /* Become a process group leader with the old leader's pid.
Eric W. Biedermanc18258c2006-09-27 01:51:06 -07001138 * The old leader becomes a thread of the this thread group.
1139 * Note: The old leader also uses this pid until release_task
Eric W. Biedermand73d6522006-03-28 16:11:03 -08001140 * is called. Odd but simple and correct.
1141 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001142 tsk->pid = leader->pid;
Oleg Nesterov3f418542013-07-03 15:08:25 -07001143 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001144 transfer_pid(leader, tsk, PIDTYPE_PGID);
1145 transfer_pid(leader, tsk, PIDTYPE_SID);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -08001146
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001147 list_replace_rcu(&leader->tasks, &tsk->tasks);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -08001148 list_replace_init(&leader->sibling, &tsk->sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001150 tsk->group_leader = tsk;
1151 leader->group_leader = tsk;
Eric W. Biedermande12a782006-04-10 17:16:49 -06001152
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001153 tsk->exit_signal = SIGCHLD;
Oleg Nesterov087806b2011-06-22 23:10:26 +02001154 leader->exit_signal = -1;
Oleg Nesterov962b5642005-11-23 13:37:43 -08001155
1156 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1157 leader->exit_state = EXIT_DEAD;
Oleg Nesteroveac1b5e2011-07-21 20:00:43 +02001158
1159 /*
1160 * We are going to release_task()->ptrace_unlink() silently,
1161 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1162 * the tracer wont't block again waiting for this thread.
1163 */
1164 if (unlikely(leader->ptrace))
1165 __wake_up_parent(leader, leader->parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 write_unlock_irq(&tasklist_lock);
Ingo Molnar780de9d2017-02-02 11:50:56 +01001167 cgroup_threadgroup_change_end(tsk);
Oleg Nesterov81879262008-12-01 14:18:16 -08001168
1169 release_task(leader);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001170 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001172 sig->group_exit_task = NULL;
1173 sig->notify_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
1175no_thread_group:
Oleg Nesterove6368252012-03-19 17:03:22 +01001176 /* we have changed execution domain */
1177 tsk->exit_signal = SIGCHLD;
1178
Nicolas Pitrebaa73d92016-11-11 00:10:10 -05001179#ifdef CONFIG_POSIX_TIMERS
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 exit_itimers(sig);
Oleg Nesterovcbaffba2008-05-26 20:55:42 +04001181 flush_itimer_signals();
Nicolas Pitrebaa73d92016-11-11 00:10:10 -05001182#endif
Oleg Nesterov329f7db2005-11-07 21:12:43 +03001183
Oleg Nesterovb2c903b2007-10-16 23:27:22 -07001184 if (atomic_read(&oldsighand->count) != 1) {
1185 struct sighand_struct *newsighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 /*
Oleg Nesterovb2c903b2007-10-16 23:27:22 -07001187 * This ->sighand is shared with the CLONE_SIGHAND
1188 * but not CLONE_THREAD task, switch to the new one.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 */
Oleg Nesterovb2c903b2007-10-16 23:27:22 -07001190 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1191 if (!newsighand)
1192 return -ENOMEM;
1193
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 atomic_set(&newsighand->count, 1);
1195 memcpy(newsighand->action, oldsighand->action,
1196 sizeof(newsighand->action));
1197
1198 write_lock_irq(&tasklist_lock);
1199 spin_lock(&oldsighand->siglock);
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001200 rcu_assign_pointer(tsk->sighand, newsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 spin_unlock(&oldsighand->siglock);
1202 write_unlock_irq(&tasklist_lock);
1203
Davide Libenzifba2afa2007-05-10 22:23:13 -07001204 __cleanup_sighand(oldsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 }
1206
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001207 BUG_ON(!thread_group_leader(tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 return 0;
Oleg Nesterovd5bbd432012-10-08 19:13:01 +02001209
1210killed:
1211 /* protects against exit_notify() and __exit_signal() */
1212 read_lock(&tasklist_lock);
1213 sig->group_exit_task = NULL;
1214 sig->notify_count = 0;
1215 read_unlock(&tasklist_lock);
1216 return -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217}
Oleg Nesterov0840a902007-10-16 23:27:22 -07001218
Andrew Morton59714d62008-02-04 22:27:21 -08001219char *get_task_comm(char *buf, struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220{
1221 /* buf must be at least sizeof(tsk->comm) in size */
1222 task_lock(tsk);
1223 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1224 task_unlock(tsk);
Andrew Morton59714d62008-02-04 22:27:21 -08001225 return buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226}
J Freyensee7d74f492011-05-06 16:56:47 -07001227EXPORT_SYMBOL_GPL(get_task_comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
Al Viro6a6d27d2012-08-21 09:56:33 -04001229/*
1230 * These functions flushes out all traces of the currently running executable
1231 * so that a new one can be started
1232 */
1233
Adrian Hunter82b89772014-05-28 11:45:04 +03001234void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235{
1236 task_lock(tsk);
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -08001237 trace_task_rename(tsk, buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1239 task_unlock(tsk);
Adrian Hunter82b89772014-05-28 11:45:04 +03001240 perf_event_comm(tsk, exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241}
1242
Kees Cooka9208e42017-07-18 15:25:30 -07001243/*
1244 * Calling this is the point of no return. None of the failures will be
1245 * seen by userspace since either the process is already taking a fatal
1246 * signal (via de_thread() or coredump), or will have SEGV raised
1247 * (after exec_mmap()) by search_binary_handlers (see below).
1248 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249int flush_old_exec(struct linux_binprm * bprm)
1250{
Linus Torvalds221af7f2010-01-28 22:14:42 -08001251 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
1253 /*
1254 * Make sure we have a private signal table and that
1255 * we are unassociated from the previous thread group.
1256 */
1257 retval = de_thread(current);
1258 if (retval)
1259 goto out;
1260
Davidlohr Bueso6e399cd2015-04-16 12:47:59 -07001261 /*
1262 * Must be called _before_ exec_mmap() as bprm->mm is
1263 * not visibile until then. This also enables the update
1264 * to be lockless.
1265 */
Matt Helsley925d1c42008-04-29 01:01:36 -07001266 set_mm_exe_file(bprm->mm, bprm->file);
Davidlohr Bueso6e399cd2015-04-16 12:47:59 -07001267
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 * Release all of the old mmap stuff
1270 */
Oleg Nesterov3c77f842010-11-30 20:55:34 +01001271 acct_arg_size(bprm, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 retval = exec_mmap(bprm->mm);
1273 if (retval)
Al Virofd8328b2008-04-22 05:11:59 -04001274 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
Kees Cooka9208e42017-07-18 15:25:30 -07001276 /*
1277 * After clearing bprm->mm (to mark that current is using the
1278 * prepared mm now), we have nothing left of the original
1279 * process. If anything from here on returns an error, the check
1280 * in search_binary_handler() will SEGV current.
1281 */
1282 bprm->mm = NULL;
Linus Torvalds7ab02af2010-02-02 12:37:44 -08001283
Mathias Krausedac853a2011-06-09 20:05:18 +02001284 set_fs(USER_DS);
Zhang Yib88fae62014-01-23 15:55:57 -08001285 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1286 PF_NOFREEZE | PF_NO_SETAFFINITY);
Linus Torvalds7ab02af2010-02-02 12:37:44 -08001287 flush_thread();
1288 current->personality &= ~bprm->per_clear;
1289
Aleksa Sarai613cc2b2016-12-21 16:26:24 +11001290 /*
1291 * We have to apply CLOEXEC before we change whether the process is
1292 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1293 * trying to access the should-be-closed file descriptors of a process
1294 * undergoing exec(2).
1295 */
1296 do_close_on_exec(current->files);
Linus Torvalds221af7f2010-01-28 22:14:42 -08001297 return 0;
1298
1299out:
1300 return retval;
1301}
1302EXPORT_SYMBOL(flush_old_exec);
1303
Al Viro1b5d7832011-06-19 12:49:47 -04001304void would_dump(struct linux_binprm *bprm, struct file *file)
1305{
Eric W. Biedermanf84df2a2016-11-16 22:06:51 -06001306 struct inode *inode = file_inode(file);
1307 if (inode_permission(inode, MAY_READ) < 0) {
1308 struct user_namespace *old, *user_ns;
Al Viro1b5d7832011-06-19 12:49:47 -04001309 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
Eric W. Biedermanf84df2a2016-11-16 22:06:51 -06001310
1311 /* Ensure mm->user_ns contains the executable */
1312 user_ns = old = bprm->mm->user_ns;
1313 while ((user_ns != &init_user_ns) &&
1314 !privileged_wrt_inode_uidgid(user_ns, inode))
1315 user_ns = user_ns->parent;
1316
1317 if (old != user_ns) {
1318 bprm->mm->user_ns = get_user_ns(user_ns);
1319 put_user_ns(old);
1320 }
1321 }
Al Viro1b5d7832011-06-19 12:49:47 -04001322}
1323EXPORT_SYMBOL(would_dump);
1324
Linus Torvalds221af7f2010-01-28 22:14:42 -08001325void setup_new_exec(struct linux_binprm * bprm)
1326{
Kees Cook46d98eb2017-07-18 15:25:27 -07001327 /*
1328 * Once here, prepare_binrpm() will not be called any more, so
1329 * the final state of setuid/setgid/fscaps can be merged into the
1330 * secureexec flag.
1331 */
1332 bprm->secureexec |= bprm->cap_elevated;
1333
Kees Cook64701de2017-07-18 15:25:35 -07001334 if (bprm->secureexec) {
Kees Cookfe8993b2017-07-18 15:25:36 -07001335 /* Make sure parent cannot signal privileged process. */
1336 current->pdeath_signal = 0;
1337
Kees Cook64701de2017-07-18 15:25:35 -07001338 /*
1339 * For secureexec, reset the stack limit to sane default to
1340 * avoid bad behavior from the prior rlimits. This has to
1341 * happen before arch_pick_mmap_layout(), which examines
1342 * RLIMIT_STACK, but after the point of no return to avoid
1343 * needing to clean up the change on failure.
1344 */
1345 if (current->signal->rlim[RLIMIT_STACK].rlim_cur > _STK_LIM)
1346 current->signal->rlim[RLIMIT_STACK].rlim_cur = _STK_LIM;
1347 }
1348
Linus Torvalds221af7f2010-01-28 22:14:42 -08001349 arch_pick_mmap_layout(current->mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 current->sas_ss_sp = current->sas_ss_size = 0;
1352
Kees Cook473d8962017-07-18 15:25:34 -07001353 /* Figure out dumpability. */
1354 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1355 bprm->secureexec)
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001356 set_dumpable(current->mm, suid_dumpable);
Kees Cook473d8962017-07-18 15:25:34 -07001357 else
1358 set_dumpable(current->mm, SUID_DUMP_USER);
Alan Coxd6e71142005-06-23 00:09:43 -07001359
Kyle Hueye9ea1e72017-03-20 01:16:26 -07001360 arch_setup_new_exec();
Peter Zijlstrae041e322014-05-21 17:32:19 +02001361 perf_event_exec();
Adrian Hunter82b89772014-05-28 11:45:04 +03001362 __set_task_comm(current, kbasename(bprm->filename), true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
Benjamin Herrenschmidt0551fbd2006-02-28 16:59:19 -08001364 /* Set the new mm task size. We have to do that late because it may
1365 * depend on TIF_32BIT which is only updated in flush_thread() on
1366 * some architectures like powerpc
1367 */
1368 current->mm->task_size = TASK_SIZE;
1369
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 /* An exec changes our domain. We are no longer part of the thread
1371 group */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 current->self_exec_id++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 flush_signal_handlers(current, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374}
Linus Torvalds221af7f2010-01-28 22:14:42 -08001375EXPORT_SYMBOL(setup_new_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
David Howellsa6f76f22008-11-14 10:39:24 +11001377/*
Oleg Nesterova2a84742009-09-05 11:17:13 -07001378 * Prepare credentials and lock ->cred_guard_mutex.
1379 * install_exec_creds() commits the new creds and drops the lock.
1380 * Or, if exec fails before, free_bprm() should release ->cred and
1381 * and unlock.
1382 */
1383int prepare_bprm_creds(struct linux_binprm *bprm)
1384{
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001385 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
Oleg Nesterova2a84742009-09-05 11:17:13 -07001386 return -ERESTARTNOINTR;
1387
1388 bprm->cred = prepare_exec_creds();
1389 if (likely(bprm->cred))
1390 return 0;
1391
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001392 mutex_unlock(&current->signal->cred_guard_mutex);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001393 return -ENOMEM;
1394}
1395
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -08001396static void free_bprm(struct linux_binprm *bprm)
Oleg Nesterova2a84742009-09-05 11:17:13 -07001397{
1398 free_arg_pages(bprm);
1399 if (bprm->cred) {
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001400 mutex_unlock(&current->signal->cred_guard_mutex);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001401 abort_creds(bprm->cred);
1402 }
Oleg Nesterov63e46b92014-01-23 15:55:51 -08001403 if (bprm->file) {
1404 allow_write_access(bprm->file);
1405 fput(bprm->file);
1406 }
Kees Cookb66c5982012-12-20 15:05:16 -08001407 /* If a binfmt changed the interp, free it. */
1408 if (bprm->interp != bprm->filename)
1409 kfree(bprm->interp);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001410 kfree(bprm);
1411}
1412
Oleg Nesterovc2315c12017-10-03 16:15:42 -07001413int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
Kees Cookb66c5982012-12-20 15:05:16 -08001414{
1415 /* If a binfmt changed the interp, free it first. */
1416 if (bprm->interp != bprm->filename)
1417 kfree(bprm->interp);
1418 bprm->interp = kstrdup(interp, GFP_KERNEL);
1419 if (!bprm->interp)
1420 return -ENOMEM;
1421 return 0;
1422}
1423EXPORT_SYMBOL(bprm_change_interp);
1424
Oleg Nesterova2a84742009-09-05 11:17:13 -07001425/*
David Howellsa6f76f22008-11-14 10:39:24 +11001426 * install the new credentials for this executable
1427 */
1428void install_exec_creds(struct linux_binprm *bprm)
1429{
1430 security_bprm_committing_creds(bprm);
1431
1432 commit_creds(bprm->cred);
1433 bprm->cred = NULL;
Stephane Eranian2976b102013-06-20 11:36:28 +02001434
1435 /*
1436 * Disable monitoring for regular users
1437 * when executing setuid binaries. Must
1438 * wait until new credentials are committed
1439 * by commit_creds() above
1440 */
1441 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1442 perf_event_exit_task(current);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001443 /*
1444 * cred_guard_mutex must be held at least to this point to prevent
David Howellsa6f76f22008-11-14 10:39:24 +11001445 * ptrace_attach() from altering our determination of the task's
Oleg Nesterova2a84742009-09-05 11:17:13 -07001446 * credentials; any time after this it may be unlocked.
1447 */
David Howellsa6f76f22008-11-14 10:39:24 +11001448 security_bprm_committed_creds(bprm);
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001449 mutex_unlock(&current->signal->cred_guard_mutex);
David Howellsa6f76f22008-11-14 10:39:24 +11001450}
1451EXPORT_SYMBOL(install_exec_creds);
1452
1453/*
1454 * determine how safe it is to execute the proposed program
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001455 * - the caller must hold ->cred_guard_mutex to protect against
Kees Cookc2e1f2e2014-06-05 00:23:17 -07001456 * PTRACE_ATTACH or seccomp thread-sync
David Howellsa6f76f22008-11-14 10:39:24 +11001457 */
Oleg Nesterov9e00cdb2014-01-23 15:55:50 -08001458static void check_unsafe_exec(struct linux_binprm *bprm)
David Howellsa6f76f22008-11-14 10:39:24 +11001459{
David Howells0bf2f3a2009-02-06 11:45:46 +00001460 struct task_struct *p = current, *t;
Al Virof1191b52009-03-30 07:35:18 -04001461 unsigned n_fs;
David Howellsa6f76f22008-11-14 10:39:24 +11001462
Eric W. Biederman9227dd22017-01-23 17:26:31 +13001463 if (p->ptrace)
1464 bprm->unsafe |= LSM_UNSAFE_PTRACE;
David Howellsa6f76f22008-11-14 10:39:24 +11001465
Andy Lutomirski259e5e62012-04-12 16:47:50 -05001466 /*
1467 * This isn't strictly necessary, but it makes it harder for LSMs to
1468 * mess up.
1469 */
Kees Cook1d4457f2014-05-21 15:23:46 -07001470 if (task_no_new_privs(current))
Andy Lutomirski259e5e62012-04-12 16:47:50 -05001471 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1472
Oleg Nesterov83f62a22014-01-23 15:55:49 -08001473 t = p;
David Howells0bf2f3a2009-02-06 11:45:46 +00001474 n_fs = 1;
Nick Piggin2a4419b2010-08-18 04:37:33 +10001475 spin_lock(&p->fs->lock);
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001476 rcu_read_lock();
Oleg Nesterov83f62a22014-01-23 15:55:49 -08001477 while_each_thread(p, t) {
David Howells0bf2f3a2009-02-06 11:45:46 +00001478 if (t->fs == p->fs)
1479 n_fs++;
David Howells0bf2f3a2009-02-06 11:45:46 +00001480 }
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001481 rcu_read_unlock();
David Howells0bf2f3a2009-02-06 11:45:46 +00001482
Oleg Nesterov9e00cdb2014-01-23 15:55:50 -08001483 if (p->fs->users > n_fs)
David Howellsa6f76f22008-11-14 10:39:24 +11001484 bprm->unsafe |= LSM_UNSAFE_SHARE;
Oleg Nesterov9e00cdb2014-01-23 15:55:50 -08001485 else
1486 p->fs->in_exec = 1;
Nick Piggin2a4419b2010-08-18 04:37:33 +10001487 spin_unlock(&p->fs->lock);
David Howellsa6f76f22008-11-14 10:39:24 +11001488}
1489
Jann Horn8b01fc82015-04-19 02:48:39 +02001490static void bprm_fill_uid(struct linux_binprm *bprm)
1491{
1492 struct inode *inode;
1493 unsigned int mode;
1494 kuid_t uid;
1495 kgid_t gid;
1496
Kees Cookcb6fd682016-05-17 12:14:39 -07001497 /*
1498 * Since this can be called multiple times (via prepare_binprm),
1499 * we must clear any previous work done when setting set[ug]id
1500 * bits from any earlier bprm->file uses (for example when run
1501 * first for a setuid script then again for its interpreter).
1502 */
Jann Horn8b01fc82015-04-19 02:48:39 +02001503 bprm->cred->euid = current_euid();
1504 bprm->cred->egid = current_egid();
1505
Andy Lutomirski380cf5b2016-06-23 16:41:05 -05001506 if (!mnt_may_suid(bprm->file->f_path.mnt))
Jann Horn8b01fc82015-04-19 02:48:39 +02001507 return;
1508
1509 if (task_no_new_privs(current))
1510 return;
1511
Vivek Goyalfea6d2a62017-02-13 15:45:26 -05001512 inode = bprm->file->f_path.dentry->d_inode;
Jann Horn8b01fc82015-04-19 02:48:39 +02001513 mode = READ_ONCE(inode->i_mode);
1514 if (!(mode & (S_ISUID|S_ISGID)))
1515 return;
1516
1517 /* Be careful if suid/sgid is set */
Al Viro59551022016-01-22 15:40:57 -05001518 inode_lock(inode);
Jann Horn8b01fc82015-04-19 02:48:39 +02001519
1520 /* reload atomically mode/uid/gid now that lock held */
1521 mode = inode->i_mode;
1522 uid = inode->i_uid;
1523 gid = inode->i_gid;
Al Viro59551022016-01-22 15:40:57 -05001524 inode_unlock(inode);
Jann Horn8b01fc82015-04-19 02:48:39 +02001525
1526 /* We ignore suid/sgid if there are no mappings for them in the ns */
1527 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1528 !kgid_has_mapping(bprm->cred->user_ns, gid))
1529 return;
1530
1531 if (mode & S_ISUID) {
1532 bprm->per_clear |= PER_CLEAR_ON_SETID;
1533 bprm->cred->euid = uid;
1534 }
1535
1536 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1537 bprm->per_clear |= PER_CLEAR_ON_SETID;
1538 bprm->cred->egid = gid;
1539 }
1540}
1541
Oleg Nesterov9e00cdb2014-01-23 15:55:50 -08001542/*
1543 * Fill the binprm structure from the inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
David Howellsa6f76f22008-11-14 10:39:24 +11001545 *
1546 * This may be called multiple times for binary chains (scripts for example).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 */
1548int prepare_binprm(struct linux_binprm *bprm)
1549{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 int retval;
Christoph Hellwigbdd1d2d2017-09-01 17:39:13 +02001551 loff_t pos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
Jann Horn8b01fc82015-04-19 02:48:39 +02001553 bprm_fill_uid(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
1555 /* fill in binprm security blob */
David Howellsa6f76f22008-11-14 10:39:24 +11001556 retval = security_bprm_set_creds(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557 if (retval)
1558 return retval;
Kees Cookddb4a142017-07-18 15:25:23 -07001559 bprm->called_set_creds = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560
David Howellsa6f76f22008-11-14 10:39:24 +11001561 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
Christoph Hellwigbdd1d2d2017-09-01 17:39:13 +02001562 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563}
1564
1565EXPORT_SYMBOL(prepare_binprm);
1566
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001567/*
1568 * Arguments are '\0' separated strings found at the location bprm->p
1569 * points to; chop off the first by relocating brpm->p to right after
1570 * the first '\0' encountered.
1571 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001572int remove_arg_zero(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573{
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001574 int ret = 0;
1575 unsigned long offset;
1576 char *kaddr;
1577 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001579 if (!bprm->argc)
1580 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001582 do {
1583 offset = bprm->p & ~PAGE_MASK;
1584 page = get_arg_page(bprm, bprm->p, 0);
1585 if (!page) {
1586 ret = -EFAULT;
1587 goto out;
1588 }
Cong Wange8e3c3d2011-11-25 23:14:27 +08001589 kaddr = kmap_atomic(page);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001590
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001591 for (; offset < PAGE_SIZE && kaddr[offset];
1592 offset++, bprm->p++)
1593 ;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001594
Cong Wange8e3c3d2011-11-25 23:14:27 +08001595 kunmap_atomic(kaddr);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001596 put_arg_page(page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001597 } while (offset == PAGE_SIZE);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001598
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001599 bprm->p++;
1600 bprm->argc--;
1601 ret = 0;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001602
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001603out:
1604 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606EXPORT_SYMBOL(remove_arg_zero);
1607
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001608#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609/*
1610 * cycle the list of binary formats handler, until one recognizes the image
1611 */
Al Viro3c456bf2012-10-20 21:53:31 -04001612int search_binary_handler(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613{
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001614 bool need_retry = IS_ENABLED(CONFIG_MODULES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 struct linux_binfmt *fmt;
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001616 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617
Kees Cookd7402692012-12-17 16:03:20 -08001618 /* This allows 4 levels of binfmt rewrites before failing hard. */
Oleg Nesterov131b2f92013-09-11 14:24:39 -07001619 if (bprm->recursion_depth > 5)
Kees Cookd7402692012-12-17 16:03:20 -08001620 return -ELOOP;
1621
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622 retval = security_bprm_check(bprm);
1623 if (retval)
1624 return retval;
1625
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 retval = -ENOENT;
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001627 retry:
1628 read_lock(&binfmt_lock);
1629 list_for_each_entry(fmt, &formats, lh) {
1630 if (!try_module_get(fmt->module))
1631 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 read_unlock(&binfmt_lock);
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001633 bprm->recursion_depth++;
1634 retval = fmt->load_binary(bprm);
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001635 read_lock(&binfmt_lock);
1636 put_binfmt(fmt);
Al Viro19d860a2014-05-04 20:11:36 -04001637 bprm->recursion_depth--;
1638 if (retval < 0 && !bprm->mm) {
1639 /* we got to flush_old_exec() and failed after it */
1640 read_unlock(&binfmt_lock);
1641 force_sigsegv(SIGSEGV, current);
1642 return retval;
1643 }
1644 if (retval != -ENOEXEC || !bprm->file) {
1645 read_unlock(&binfmt_lock);
1646 return retval;
1647 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 }
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001649 read_unlock(&binfmt_lock);
1650
Al Viro19d860a2014-05-04 20:11:36 -04001651 if (need_retry) {
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001652 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1653 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1654 return retval;
Oleg Nesterov4e0621a2013-09-11 14:24:45 -07001655 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1656 return retval;
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001657 need_retry = false;
1658 goto retry;
1659 }
1660
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661 return retval;
1662}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663EXPORT_SYMBOL(search_binary_handler);
1664
Oleg Nesterov5d1baf32013-09-11 14:24:38 -07001665static int exec_binprm(struct linux_binprm *bprm)
1666{
1667 pid_t old_pid, old_vpid;
1668 int ret;
1669
1670 /* Need to fetch pid before load_binary changes it */
1671 old_pid = current->pid;
1672 rcu_read_lock();
1673 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1674 rcu_read_unlock();
1675
1676 ret = search_binary_handler(bprm);
1677 if (ret >= 0) {
Linus Torvalds3eaded862013-11-21 19:18:14 -08001678 audit_bprm(bprm);
Oleg Nesterov5d1baf32013-09-11 14:24:38 -07001679 trace_sched_process_exec(current, old_pid, bprm);
1680 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
Oleg Nesterov9beb2662013-09-11 14:24:40 -07001681 proc_exec_connector(current);
Oleg Nesterov5d1baf32013-09-11 14:24:38 -07001682 }
1683
1684 return ret;
1685}
1686
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687/*
1688 * sys_execve() executes a new program.
1689 */
David Drysdale51f39a12014-12-12 16:57:29 -08001690static int do_execveat_common(int fd, struct filename *filename,
1691 struct user_arg_ptr argv,
1692 struct user_arg_ptr envp,
1693 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694{
David Drysdale51f39a12014-12-12 16:57:29 -08001695 char *pathbuf = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 struct linux_binprm *bprm;
1697 struct file *file;
Al Viro3b125382008-04-22 05:31:30 -04001698 struct files_struct *displaced;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 int retval;
Vasiliy Kulikov72fa5992011-08-08 19:02:04 +04001700
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -08001701 if (IS_ERR(filename))
1702 return PTR_ERR(filename);
1703
Vasiliy Kulikov72fa5992011-08-08 19:02:04 +04001704 /*
1705 * We move the actual failure in case of RLIMIT_NPROC excess from
1706 * set*uid() to execve() because too many poorly written programs
1707 * don't check setuid() return code. Here we additionally recheck
1708 * whether NPROC limit is still exceeded.
1709 */
1710 if ((current->flags & PF_NPROC_EXCEEDED) &&
Oleg Nesterovbd9d43f2013-07-03 15:08:34 -07001711 atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
Vasiliy Kulikov72fa5992011-08-08 19:02:04 +04001712 retval = -EAGAIN;
1713 goto out_ret;
1714 }
1715
1716 /* We're below the limit (still or again), so we don't want to make
1717 * further execve() calls fail. */
1718 current->flags &= ~PF_NPROC_EXCEEDED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719
Al Viro3b125382008-04-22 05:31:30 -04001720 retval = unshare_files(&displaced);
Al Virofd8328b2008-04-22 05:11:59 -04001721 if (retval)
1722 goto out_ret;
1723
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 retval = -ENOMEM;
Oliver Neukum11b0b5a2006-03-25 03:08:13 -08001725 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 if (!bprm)
Al Virofd8328b2008-04-22 05:11:59 -04001727 goto out_files;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
Oleg Nesterova2a84742009-09-05 11:17:13 -07001729 retval = prepare_bprm_creds(bprm);
1730 if (retval)
David Howellsa6f76f22008-11-14 10:39:24 +11001731 goto out_free;
Al Viro498052b2009-03-30 07:20:30 -04001732
Oleg Nesterov9e00cdb2014-01-23 15:55:50 -08001733 check_unsafe_exec(bprm);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001734 current->in_execve = 1;
David Howellsa6f76f22008-11-14 10:39:24 +11001735
David Drysdale51f39a12014-12-12 16:57:29 -08001736 file = do_open_execat(fd, filename, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 retval = PTR_ERR(file);
1738 if (IS_ERR(file))
Al Viro498052b2009-03-30 07:20:30 -04001739 goto out_unmark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740
1741 sched_exec();
1742
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 bprm->file = file;
David Drysdale51f39a12014-12-12 16:57:29 -08001744 if (fd == AT_FDCWD || filename->name[0] == '/') {
1745 bprm->filename = filename->name;
1746 } else {
1747 if (filename->name[0] == '\0')
Michal Hocko0ee931c2017-09-13 16:28:29 -07001748 pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
David Drysdale51f39a12014-12-12 16:57:29 -08001749 else
Michal Hocko0ee931c2017-09-13 16:28:29 -07001750 pathbuf = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
David Drysdale51f39a12014-12-12 16:57:29 -08001751 fd, filename->name);
1752 if (!pathbuf) {
1753 retval = -ENOMEM;
1754 goto out_unmark;
1755 }
1756 /*
1757 * Record that a name derived from an O_CLOEXEC fd will be
1758 * inaccessible after exec. Relies on having exclusive access to
1759 * current->files (due to unshare_files above).
1760 */
1761 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1762 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1763 bprm->filename = pathbuf;
1764 }
1765 bprm->interp = bprm->filename;
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001766
1767 retval = bprm_mm_init(bprm);
1768 if (retval)
Oleg Nesterov63e46b92014-01-23 15:55:51 -08001769 goto out_unmark;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001771 bprm->argc = count(argv, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 if ((retval = bprm->argc) < 0)
David Howellsa6f76f22008-11-14 10:39:24 +11001773 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001775 bprm->envc = count(envp, MAX_ARG_STRINGS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 if ((retval = bprm->envc) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 goto out;
1778
1779 retval = prepare_binprm(bprm);
1780 if (retval < 0)
1781 goto out;
1782
1783 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1784 if (retval < 0)
1785 goto out;
1786
1787 bprm->exec = bprm->p;
1788 retval = copy_strings(bprm->envc, envp, bprm);
1789 if (retval < 0)
1790 goto out;
1791
1792 retval = copy_strings(bprm->argc, argv, bprm);
1793 if (retval < 0)
1794 goto out;
1795
Eric W. Biedermanf84df2a2016-11-16 22:06:51 -06001796 would_dump(bprm, bprm->file);
1797
Oleg Nesterov5d1baf32013-09-11 14:24:38 -07001798 retval = exec_binprm(bprm);
David Howellsa6f76f22008-11-14 10:39:24 +11001799 if (retval < 0)
1800 goto out;
1801
1802 /* execve succeeded */
Al Viro498052b2009-03-30 07:20:30 -04001803 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001804 current->in_execve = 0;
Mathieu Desnoyersa961e402017-10-19 13:30:15 -04001805 membarrier_execve(current);
David Howellsa6f76f22008-11-14 10:39:24 +11001806 acct_update_integrals(current);
Rik van Riel82727012013-10-07 11:29:28 +01001807 task_numa_free(current);
David Howellsa6f76f22008-11-14 10:39:24 +11001808 free_bprm(bprm);
David Drysdale51f39a12014-12-12 16:57:29 -08001809 kfree(pathbuf);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -08001810 putname(filename);
David Howellsa6f76f22008-11-14 10:39:24 +11001811 if (displaced)
1812 put_files_struct(displaced);
1813 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814
1815out:
Oleg Nesterov3c77f842010-11-30 20:55:34 +01001816 if (bprm->mm) {
1817 acct_arg_size(bprm, 0);
1818 mmput(bprm->mm);
1819 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820
Al Viro498052b2009-03-30 07:20:30 -04001821out_unmark:
Oleg Nesterov9e00cdb2014-01-23 15:55:50 -08001822 current->fs->in_exec = 0;
Kentaro Takedaf9ce1f12009-02-05 17:18:11 +09001823 current->in_execve = 0;
David Howellsa6f76f22008-11-14 10:39:24 +11001824
1825out_free:
Al Viro08a6fac2008-05-10 16:38:25 -04001826 free_bprm(bprm);
David Drysdale51f39a12014-12-12 16:57:29 -08001827 kfree(pathbuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828
Al Virofd8328b2008-04-22 05:11:59 -04001829out_files:
Al Viro3b125382008-04-22 05:31:30 -04001830 if (displaced)
1831 reset_files_struct(displaced);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832out_ret:
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -08001833 putname(filename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834 return retval;
1835}
1836
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -08001837int do_execve(struct filename *filename,
Oleg Nesterovba2d01622011-03-06 18:02:37 +01001838 const char __user *const __user *__argv,
Al Viroda3d4c52012-10-20 21:49:33 -04001839 const char __user *const __user *__envp)
Oleg Nesterovba2d01622011-03-06 18:02:37 +01001840{
Oleg Nesterov0e028462011-03-06 18:02:54 +01001841 struct user_arg_ptr argv = { .ptr.native = __argv };
1842 struct user_arg_ptr envp = { .ptr.native = __envp };
David Drysdale51f39a12014-12-12 16:57:29 -08001843 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1844}
1845
1846int do_execveat(int fd, struct filename *filename,
1847 const char __user *const __user *__argv,
1848 const char __user *const __user *__envp,
1849 int flags)
1850{
1851 struct user_arg_ptr argv = { .ptr.native = __argv };
1852 struct user_arg_ptr envp = { .ptr.native = __envp };
1853
1854 return do_execveat_common(fd, filename, argv, envp, flags);
Oleg Nesterovba2d01622011-03-06 18:02:37 +01001855}
1856
Oleg Nesterov0e028462011-03-06 18:02:54 +01001857#ifdef CONFIG_COMPAT
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -08001858static int compat_do_execve(struct filename *filename,
Al Viro38b983b2012-09-30 13:38:55 -04001859 const compat_uptr_t __user *__argv,
Al Virod03d26e2012-10-20 21:46:25 -04001860 const compat_uptr_t __user *__envp)
Oleg Nesterov0e028462011-03-06 18:02:54 +01001861{
1862 struct user_arg_ptr argv = {
1863 .is_compat = true,
1864 .ptr.compat = __argv,
1865 };
1866 struct user_arg_ptr envp = {
1867 .is_compat = true,
1868 .ptr.compat = __envp,
1869 };
David Drysdale51f39a12014-12-12 16:57:29 -08001870 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1871}
1872
1873static int compat_do_execveat(int fd, struct filename *filename,
1874 const compat_uptr_t __user *__argv,
1875 const compat_uptr_t __user *__envp,
1876 int flags)
1877{
1878 struct user_arg_ptr argv = {
1879 .is_compat = true,
1880 .ptr.compat = __argv,
1881 };
1882 struct user_arg_ptr envp = {
1883 .is_compat = true,
1884 .ptr.compat = __envp,
1885 };
1886 return do_execveat_common(fd, filename, argv, envp, flags);
Oleg Nesterov0e028462011-03-06 18:02:54 +01001887}
1888#endif
1889
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001890void set_binfmt(struct linux_binfmt *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891{
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001892 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07001894 if (mm->binfmt)
1895 module_put(mm->binfmt->module);
1896
1897 mm->binfmt = new;
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07001898 if (new)
1899 __module_get(new->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901EXPORT_SYMBOL(set_binfmt);
1902
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001903/*
Oleg Nesterov7288e112014-01-23 15:55:32 -08001904 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001905 */
1906void set_dumpable(struct mm_struct *mm, int value)
1907{
Oleg Nesterovabacd2f2014-01-23 15:55:31 -08001908 unsigned long old, new;
1909
Oleg Nesterov7288e112014-01-23 15:55:32 -08001910 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1911 return;
1912
Oleg Nesterovabacd2f2014-01-23 15:55:31 -08001913 do {
1914 old = ACCESS_ONCE(mm->flags);
Oleg Nesterov7288e112014-01-23 15:55:32 -08001915 new = (old & ~MMF_DUMPABLE_MASK) | value;
Oleg Nesterovabacd2f2014-01-23 15:55:31 -08001916 } while (cmpxchg(&mm->flags, old, new) != old);
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001917}
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07001918
Al Viro38b983b2012-09-30 13:38:55 -04001919SYSCALL_DEFINE3(execve,
1920 const char __user *, filename,
1921 const char __user *const __user *, argv,
1922 const char __user *const __user *, envp)
1923{
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -08001924 return do_execve(getname(filename), argv, envp);
Al Viro38b983b2012-09-30 13:38:55 -04001925}
David Drysdale51f39a12014-12-12 16:57:29 -08001926
1927SYSCALL_DEFINE5(execveat,
1928 int, fd, const char __user *, filename,
1929 const char __user *const __user *, argv,
1930 const char __user *const __user *, envp,
1931 int, flags)
1932{
1933 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1934
1935 return do_execveat(fd,
1936 getname_flags(filename, lookup_flags, NULL),
1937 argv, envp, flags);
1938}
1939
Al Viro38b983b2012-09-30 13:38:55 -04001940#ifdef CONFIG_COMPAT
Heiko Carstens625b1d72014-03-04 10:53:50 +01001941COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1942 const compat_uptr_t __user *, argv,
1943 const compat_uptr_t __user *, envp)
Al Viro38b983b2012-09-30 13:38:55 -04001944{
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -08001945 return compat_do_execve(getname(filename), argv, envp);
Al Viro38b983b2012-09-30 13:38:55 -04001946}
David Drysdale51f39a12014-12-12 16:57:29 -08001947
1948COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1949 const char __user *, filename,
1950 const compat_uptr_t __user *, argv,
1951 const compat_uptr_t __user *, envp,
1952 int, flags)
1953{
1954 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1955
1956 return compat_do_execveat(fd,
1957 getname_flags(filename, lookup_flags, NULL),
1958 argv, envp, flags);
1959}
Al Viro38b983b2012-09-30 13:38:55 -04001960#endif