blob: ec5ef10ce2dbbd70cf271a2f3ebc714f88cfc193 [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * linux/fs/exec.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 */
7
8/*
9 * #!-checking implemented by tytso.
10 */
11/*
12 * Demand-loading implemented 01.12.91 - no need to read anything but
13 * the header into memory. The inode of the executable is put into
14 * "current->executable", and page faults do the actual loading. Clean.
15 *
16 * Once more I can proudly say that linux stood up to being changed: it
17 * was less than 2 hours work to get demand-loading completely implemented.
18 *
19 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
20 * current->executable is only used by the procfs. This allows a dispatch
21 * table to check for several different types of binary formats. We keep
22 * trying until we recognize the file or we run out of supported binary
Aleksa Sarai613cc2b2016-12-21 16:26:24 +110023 * formats.
Linus Torvalds1da177e2005-04-16 15:20:36 -070024 */
25
Scott Brandenb89999d02020-10-02 10:38:15 -070026#include <linux/kernel_read_file.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/slab.h>
28#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -040029#include <linux/fdtable.h>
Hugh Dickinsba92a432008-07-25 01:45:43 -070030#include <linux/mm.h>
Davidlohr Bueso615d6e82014-04-07 15:37:25 -070031#include <linux/vmacache.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/stat.h>
33#include <linux/fcntl.h>
Hugh Dickinsba92a432008-07-25 01:45:43 -070034#include <linux/swap.h>
Neil Horman74aadce2007-10-16 23:26:35 -070035#include <linux/string.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/init.h>
Ingo Molnar6e84f312017-02-08 18:51:29 +010037#include <linux/sched/mm.h>
Ingo Molnarf7ccbae2017-02-08 18:51:30 +010038#include <linux/sched/coredump.h>
Ingo Molnar3f07c012017-02-08 18:51:30 +010039#include <linux/sched/signal.h>
Ingo Molnar6a3827d2017-02-08 18:51:31 +010040#include <linux/sched/numa_balancing.h>
Ingo Molnar29930022017-02-08 18:51:36 +010041#include <linux/sched/task.h>
Hugh Dickinsca5b1722008-07-28 15:46:18 -070042#include <linux/pagemap.h>
Ingo Molnarcdd6c482009-09-21 12:02:48 +020043#include <linux/perf_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <linux/highmem.h>
45#include <linux/spinlock.h>
46#include <linux/key.h>
47#include <linux/personality.h>
48#include <linux/binfmts.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/utsname.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080050#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/module.h>
52#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include <linux/mount.h>
54#include <linux/security.h>
55#include <linux/syscalls.h>
Jay Lan8f0ab512006-09-30 23:28:59 -070056#include <linux/tsacct_kern.h>
Matt Helsley9f460802005-11-07 00:59:16 -080057#include <linux/cn_proc.h>
Al Viro473ae302006-04-26 14:04:08 -040058#include <linux/audit.h>
Roland McGrath6341c392008-07-25 19:45:44 -070059#include <linux/tracehook.h>
Johannes Berg5f4123b2008-07-09 10:28:40 +020060#include <linux/kmod.h>
Eric Paris6110e3a2008-12-17 13:53:20 -050061#include <linux/fsnotify.h>
Al Viro5ad4e532009-03-29 19:50:06 -040062#include <linux/fs_struct.h>
Ying Han3d5992d2010-10-26 14:21:23 -070063#include <linux/oom.h>
Oleg Nesterov0e028462011-03-06 18:02:54 +010064#include <linux/compat.h>
Mimi Zoharb44a7df2015-12-28 16:02:29 -050065#include <linux/vmalloc.h>
Jens Axboe0f212202020-09-13 13:09:39 -060066#include <linux/io_uring.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080068#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070069#include <asm/mmu_context.h>
Ollie Wildb6a2fea2007-07-19 01:48:16 -070070#include <asm/tlb.h>
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080071
72#include <trace/events/task.h>
David Howellsa6f76f22008-11-14 10:39:24 +110073#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
David Smith4ff16c22012-02-07 10:11:05 -060075#include <trace/events/sched.h>
ted.lincadbca12021-05-26 22:00:14 +080076#include <trace/hooks/sched.h>
David Smith4ff16c22012-02-07 10:11:05 -060077
Liujie Xie016d3f72021-06-03 17:41:28 +080078EXPORT_TRACEPOINT_SYMBOL_GPL(task_rename);
79
Eric W. Biederman56305aa2020-05-29 22:00:54 -050080static int bprm_creds_from_file(struct linux_binprm *bprm);
81
Alan Coxd6e71142005-06-23 00:09:43 -070082int suid_dumpable = 0;
83
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -070084static LIST_HEAD(formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085static DEFINE_RWLOCK(binfmt_lock);
86
Al Viro8fc3dc52012-03-17 03:05:16 -040087void __register_binfmt(struct linux_binfmt * fmt, int insert)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
Al Viro8fc3dc52012-03-17 03:05:16 -040089 BUG_ON(!fmt);
Oleg Nesterov92eaa562013-09-11 14:24:42 -070090 if (WARN_ON(!fmt->load_binary))
91 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 write_lock(&binfmt_lock);
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070093 insert ? list_add(&fmt->lh, &formats) :
94 list_add_tail(&fmt->lh, &formats);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 write_unlock(&binfmt_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096}
97
Ivan Kokshaysky74641f52009-04-30 15:08:49 -070098EXPORT_SYMBOL(__register_binfmt);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
Alexey Dobriyanf6b450d2007-10-16 23:26:04 -0700100void unregister_binfmt(struct linux_binfmt * fmt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 write_lock(&binfmt_lock);
Alexey Dobriyane4dc1b12007-10-16 23:26:03 -0700103 list_del(&fmt->lh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 write_unlock(&binfmt_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105}
106
107EXPORT_SYMBOL(unregister_binfmt);
108
109static inline void put_binfmt(struct linux_binfmt * fmt)
110{
111 module_put(fmt->module);
112}
113
Eric W. Biederman90f85722015-06-29 14:42:03 -0500114bool path_noexec(const struct path *path)
115{
116 return (path->mnt->mnt_flags & MNT_NOEXEC) ||
117 (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
118}
119
Josh Triplett69369a72014-04-03 14:48:27 -0700120#ifdef CONFIG_USELIB
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121/*
122 * Note that a shared library must be both readable and executable due to
123 * security reasons.
124 *
125 * Also note that we take the address to load from from the file itself.
126 */
Heiko Carstens1e7bfb22009-01-14 14:14:29 +0100127SYSCALL_DEFINE1(uselib, const char __user *, library)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128{
Al Viro72c2d532013-09-22 16:27:52 -0400129 struct linux_binfmt *fmt;
Al Viro964bd182008-07-26 03:33:14 -0400130 struct file *file;
Jeff Layton91a27b22012-10-10 15:25:28 -0400131 struct filename *tmp = getname(library);
Al Viro964bd182008-07-26 03:33:14 -0400132 int error = PTR_ERR(tmp);
Al Viro47c805d2011-02-23 17:44:09 -0500133 static const struct open_flags uselib_flags = {
134 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
Al Viro62fb4a12015-12-26 22:33:24 -0500135 .acc_mode = MAY_READ | MAY_EXEC,
Al Virof9652e12013-06-11 08:23:01 +0400136 .intent = LOOKUP_OPEN,
137 .lookup_flags = LOOKUP_FOLLOW,
Al Viro47c805d2011-02-23 17:44:09 -0500138 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
Al Viro6e8341a2009-04-06 11:16:22 -0400140 if (IS_ERR(tmp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 goto out;
142
Al Virof9652e12013-06-11 08:23:01 +0400143 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
Al Viro6e8341a2009-04-06 11:16:22 -0400144 putname(tmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 error = PTR_ERR(file);
146 if (IS_ERR(file))
147 goto out;
148
Kees Cook633fb6a2020-08-11 18:36:26 -0700149 /*
150 * may_open() has already checked for this, so it should be
151 * impossible to trip now. But we need to be extra cautious
152 * and check again at the very end too.
153 */
Kees Cookdb19c912020-08-11 18:36:23 -0700154 error = -EACCES;
Kees Cook0fd338b2020-08-11 18:36:30 -0700155 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
156 path_noexec(&file->f_path)))
Al Viro6e8341a2009-04-06 11:16:22 -0400157 goto exit;
158
Eric Paris2a12a9d2009-12-17 21:24:21 -0500159 fsnotify_open(file);
Eric Paris6110e3a2008-12-17 13:53:20 -0500160
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 error = -ENOEXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Al Viro72c2d532013-09-22 16:27:52 -0400163 read_lock(&binfmt_lock);
164 list_for_each_entry(fmt, &formats, lh) {
165 if (!fmt->load_shlib)
166 continue;
167 if (!try_module_get(fmt->module))
168 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 read_unlock(&binfmt_lock);
Al Viro72c2d532013-09-22 16:27:52 -0400170 error = fmt->load_shlib(file);
171 read_lock(&binfmt_lock);
172 put_binfmt(fmt);
173 if (error != -ENOEXEC)
174 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 }
Al Viro72c2d532013-09-22 16:27:52 -0400176 read_unlock(&binfmt_lock);
Al Viro6e8341a2009-04-06 11:16:22 -0400177exit:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 fput(file);
179out:
180 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181}
Josh Triplett69369a72014-04-03 14:48:27 -0700182#endif /* #ifdef CONFIG_USELIB */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700184#ifdef CONFIG_MMU
Oleg Nesterovae6b5852011-03-06 18:03:11 +0100185/*
186 * The nascent bprm->mm is not visible until exec_mmap() but it can
187 * use a lot of memory, account these pages in current->mm temporary
188 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
189 * change the counter back via acct_arg_size(0).
190 */
Oleg Nesterov0e028462011-03-06 18:02:54 +0100191static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100192{
193 struct mm_struct *mm = current->mm;
194 long diff = (long)(pages - bprm->vma_pages);
195
196 if (!mm || !diff)
197 return;
198
199 bprm->vma_pages = pages;
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100200 add_mm_counter(mm, MM_ANONPAGES, diff);
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100201}
202
Oleg Nesterov0e028462011-03-06 18:02:54 +0100203static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700204 int write)
205{
206 struct page *page;
207 int ret;
Lorenzo Stoakes9beae1e2016-10-13 01:20:17 +0100208 unsigned int gup_flags = FOLL_FORCE;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700209
210#ifdef CONFIG_STACK_GROWSUP
211 if (write) {
Michal Hockod05f3162011-05-24 17:11:44 -0700212 ret = expand_downwards(bprm->vma, pos);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700213 if (ret < 0)
214 return NULL;
215 }
216#endif
Lorenzo Stoakes9beae1e2016-10-13 01:20:17 +0100217
218 if (write)
219 gup_flags |= FOLL_WRITE;
220
Dave Hansen1e987792016-02-12 13:01:54 -0800221 /*
222 * We are doing an exec(). 'current' is the process
223 * doing the exec and bprm->mm is the new process's mm.
224 */
Peter Xu64019a22020-08-11 18:39:01 -0700225 ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
Lorenzo Stoakes5b56d492016-12-14 15:06:52 -0800226 &page, NULL, NULL);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700227 if (ret <= 0)
228 return NULL;
229
Oleg Nesterov655c16a2019-01-03 15:28:11 -0800230 if (write)
231 acct_arg_size(bprm, vma_pages(bprm->vma));
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700232
233 return page;
234}
235
236static void put_arg_page(struct page *page)
237{
Minchan Kimd3b077b2021-03-18 13:30:17 -0700238 put_user_page(page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700239}
240
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700241static void free_arg_pages(struct linux_binprm *bprm)
242{
243}
244
245static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
246 struct page *page)
247{
248 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
249}
250
251static int __bprm_mm_init(struct linux_binprm *bprm)
252{
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800253 int err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700254 struct vm_area_struct *vma = NULL;
255 struct mm_struct *mm = bprm->mm;
256
Linus Torvalds490fc052018-07-21 15:24:03 -0700257 bprm->vma = vma = vm_area_alloc(mm);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700258 if (!vma)
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800259 return -ENOMEM;
Kirill A. Shutemovbfd40ea2018-07-26 16:37:35 -0700260 vma_set_anonymous(vma);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700261
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700262 if (mmap_write_lock_killable(mm)) {
Michal Hockof268dfe2016-05-23 16:26:02 -0700263 err = -EINTR;
264 goto err_free;
265 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700266
267 /*
268 * Place the stack at the largest stack address the architecture
269 * supports. Later, we'll move this to an appropriate place. We don't
270 * use STACK_TOP because that can depend on attributes which aren't
271 * configured yet.
272 */
Michal Hockoaacb3d12011-07-26 16:08:40 -0700273 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700274 vma->vm_end = STACK_TOP_MAX;
275 vma->vm_start = vma->vm_end - PAGE_SIZE;
Cyrill Gorcunovd9104d12013-09-11 14:22:24 -0700276 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
Coly Li3ed75eb2007-10-18 23:39:15 -0700277 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
Tavis Ormandy462e635e2010-12-09 15:29:42 +0100278
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700279 err = insert_vm_struct(mm, vma);
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800280 if (err)
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700281 goto err;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700282
283 mm->stack_vm = mm->total_vm = 1;
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700284 mmap_write_unlock(mm);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700285 bprm->p = vma->vm_end - sizeof(void *);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700286 return 0;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700287err:
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700288 mmap_write_unlock(mm);
Michal Hockof268dfe2016-05-23 16:26:02 -0700289err_free:
Luiz Fernando N. Capitulinoeaccbfa2009-01-06 14:40:44 -0800290 bprm->vma = NULL;
Linus Torvalds3928d4f2018-07-21 13:48:51 -0700291 vm_area_free(vma);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700292 return err;
293}
294
295static bool valid_arg_len(struct linux_binprm *bprm, long len)
296{
297 return len <= MAX_ARG_STRLEN;
298}
299
300#else
301
Oleg Nesterov0e028462011-03-06 18:02:54 +0100302static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
Oleg Nesterov3c77f842010-11-30 20:55:34 +0100303{
304}
305
Oleg Nesterov0e028462011-03-06 18:02:54 +0100306static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700307 int write)
308{
309 struct page *page;
310
311 page = bprm->page[pos / PAGE_SIZE];
312 if (!page && write) {
313 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
314 if (!page)
315 return NULL;
316 bprm->page[pos / PAGE_SIZE] = page;
317 }
318
319 return page;
320}
321
322static void put_arg_page(struct page *page)
323{
324}
325
326static void free_arg_page(struct linux_binprm *bprm, int i)
327{
328 if (bprm->page[i]) {
329 __free_page(bprm->page[i]);
330 bprm->page[i] = NULL;
331 }
332}
333
334static void free_arg_pages(struct linux_binprm *bprm)
335{
336 int i;
337
338 for (i = 0; i < MAX_ARG_PAGES; i++)
339 free_arg_page(bprm, i);
340}
341
342static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
343 struct page *page)
344{
345}
346
347static int __bprm_mm_init(struct linux_binprm *bprm)
348{
349 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
350 return 0;
351}
352
353static bool valid_arg_len(struct linux_binprm *bprm, long len)
354{
355 return len <= bprm->p;
356}
357
358#endif /* CONFIG_MMU */
359
360/*
361 * Create a new mm_struct and populate it with a temporary stack
362 * vm_area_struct. We don't have enough context at this point to set the stack
363 * flags, permissions, and offset, so we use temporary values. We'll update
364 * them later in setup_arg_pages().
365 */
Yuanhan Liu9cc64ce2013-02-20 13:16:01 +1100366static int bprm_mm_init(struct linux_binprm *bprm)
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700367{
368 int err;
369 struct mm_struct *mm = NULL;
370
371 bprm->mm = mm = mm_alloc();
372 err = -ENOMEM;
373 if (!mm)
374 goto err;
375
Kees Cookc31dbb142018-04-10 16:35:01 -0700376 /* Save current stack limit for all calculations made during exec. */
377 task_lock(current->group_leader);
378 bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
379 task_unlock(current->group_leader);
380
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700381 err = __bprm_mm_init(bprm);
382 if (err)
383 goto err;
384
385 return 0;
386
387err:
388 if (mm) {
389 bprm->mm = NULL;
390 mmdrop(mm);
391 }
392
393 return err;
394}
395
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100396struct user_arg_ptr {
Oleg Nesterov0e028462011-03-06 18:02:54 +0100397#ifdef CONFIG_COMPAT
398 bool is_compat;
399#endif
400 union {
401 const char __user *const __user *native;
402#ifdef CONFIG_COMPAT
Al Viro38b983b2012-09-30 13:38:55 -0400403 const compat_uptr_t __user *compat;
Oleg Nesterov0e028462011-03-06 18:02:54 +0100404#endif
405 } ptr;
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100406};
407
408static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100409{
Oleg Nesterov0e028462011-03-06 18:02:54 +0100410 const char __user *native;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100411
Oleg Nesterov0e028462011-03-06 18:02:54 +0100412#ifdef CONFIG_COMPAT
413 if (unlikely(argv.is_compat)) {
414 compat_uptr_t compat;
415
416 if (get_user(compat, argv.ptr.compat + nr))
417 return ERR_PTR(-EFAULT);
418
419 return compat_ptr(compat);
420 }
421#endif
422
423 if (get_user(native, argv.ptr.native + nr))
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100424 return ERR_PTR(-EFAULT);
425
Oleg Nesterov0e028462011-03-06 18:02:54 +0100426 return native;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100427}
428
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429/*
430 * count() counts the number of strings in array ARGV.
431 */
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100432static int count(struct user_arg_ptr argv, int max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433{
434 int i = 0;
435
Oleg Nesterov0e028462011-03-06 18:02:54 +0100436 if (argv.ptr.native != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 for (;;) {
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100438 const char __user *p = get_user_arg_ptr(argv, i);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 if (!p)
441 break;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100442
443 if (IS_ERR(p))
444 return -EFAULT;
445
Xi Wang6d92d4f2013-01-11 14:31:48 -0800446 if (i >= max)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 return -E2BIG;
Xi Wang6d92d4f2013-01-11 14:31:48 -0800448 ++i;
Roland McGrath9aea5a62010-09-07 19:37:06 -0700449
450 if (fatal_signal_pending(current))
451 return -ERESTARTNOHAND;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 cond_resched();
453 }
454 }
455 return i;
456}
457
Eric W. Biedermanbe619f72020-07-13 12:06:48 -0500458static int count_strings_kernel(const char *const *argv)
459{
460 int i;
461
462 if (!argv)
463 return 0;
464
465 for (i = 0; argv[i]; ++i) {
466 if (i >= MAX_ARG_STRINGS)
467 return -E2BIG;
468 if (fatal_signal_pending(current))
469 return -ERESTARTNOHAND;
470 cond_resched();
471 }
472 return i;
473}
474
Eric W. Biedermand8b9cd52020-07-12 08:23:54 -0500475static int bprm_stack_limits(struct linux_binprm *bprm)
Oleg Nesterov655c16a2019-01-03 15:28:11 -0800476{
477 unsigned long limit, ptr_size;
478
Oleg Nesterov655c16a2019-01-03 15:28:11 -0800479 /*
480 * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
481 * (whichever is smaller) for the argv+env strings.
482 * This ensures that:
483 * - the remaining binfmt code will not run out of stack space,
484 * - the program will have a reasonable amount of stack left
485 * to work from.
486 */
487 limit = _STK_LIM / 4 * 3;
488 limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
489 /*
490 * We've historically supported up to 32 pages (ARG_MAX)
491 * of argument strings even with small stacks
492 */
493 limit = max_t(unsigned long, limit, ARG_MAX);
494 /*
495 * We must account for the size of all the argv and envp pointers to
496 * the argv and envp strings, since they will also take up space in
497 * the stack. They aren't stored until much later when we can't
498 * signal to the parent that the child has run out of stack space.
499 * Instead, calculate it here so it's possible to fail gracefully.
500 */
501 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
502 if (limit <= ptr_size)
503 return -E2BIG;
504 limit -= ptr_size;
505
506 bprm->argmin = bprm->p - limit;
507 return 0;
508}
509
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700511 * 'copy_strings()' copies argument/environment strings from the old
512 * processes's memory to the new process's stack. The call to get_user_pages()
513 * ensures the destination page is created and not swapped out.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 */
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100515static int copy_strings(int argc, struct user_arg_ptr argv,
Adrian Bunk75c96f82005-05-05 16:16:09 -0700516 struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517{
518 struct page *kmapped_page = NULL;
519 char *kaddr = NULL;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700520 unsigned long kpos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 int ret;
522
523 while (argc-- > 0) {
David Howellsd7627462010-08-17 23:52:56 +0100524 const char __user *str;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 int len;
526 unsigned long pos;
527
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100528 ret = -EFAULT;
529 str = get_user_arg_ptr(argv, argc);
530 if (IS_ERR(str))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100533 len = strnlen_user(str, MAX_ARG_STRLEN);
534 if (!len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 goto out;
Oleg Nesterov1d1dbf82011-03-06 18:02:21 +0100536
537 ret = -E2BIG;
538 if (!valid_arg_len(bprm, len))
539 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700541 /* We're going to work our way backwords. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 pos = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700543 str += len;
544 bprm->p -= len;
Oleg Nesterov655c16a2019-01-03 15:28:11 -0800545#ifdef CONFIG_MMU
546 if (bprm->p < bprm->argmin)
547 goto out;
548#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
550 while (len > 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 int offset, bytes_to_copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
Roland McGrath9aea5a62010-09-07 19:37:06 -0700553 if (fatal_signal_pending(current)) {
554 ret = -ERESTARTNOHAND;
555 goto out;
556 }
Roland McGrath7993bc12010-09-07 19:36:28 -0700557 cond_resched();
558
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 offset = pos % PAGE_SIZE;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700560 if (offset == 0)
561 offset = PAGE_SIZE;
562
563 bytes_to_copy = offset;
564 if (bytes_to_copy > len)
565 bytes_to_copy = len;
566
567 offset -= bytes_to_copy;
568 pos -= bytes_to_copy;
569 str -= bytes_to_copy;
570 len -= bytes_to_copy;
571
572 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
573 struct page *page;
574
575 page = get_arg_page(bprm, pos, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 if (!page) {
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700577 ret = -E2BIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 goto out;
579 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700581 if (kmapped_page) {
582 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700584 put_arg_page(kmapped_page);
585 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 kmapped_page = page;
587 kaddr = kmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700588 kpos = pos & PAGE_MASK;
589 flush_arg_page(bprm, kpos, kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 }
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700591 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 ret = -EFAULT;
593 goto out;
594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 }
596 }
597 ret = 0;
598out:
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700599 if (kmapped_page) {
600 flush_kernel_dcache_page(kmapped_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 kunmap(kmapped_page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700602 put_arg_page(kmapped_page);
603 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 return ret;
605}
606
607/*
Christoph Hellwig986db2d2020-06-04 16:51:14 -0700608 * Copy and argument/environment string from the kernel to the processes stack.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 */
Christoph Hellwig986db2d2020-06-04 16:51:14 -0700610int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611{
Christoph Hellwig762a3af2020-06-04 16:51:18 -0700612 int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
613 unsigned long pos = bprm->p;
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100614
Christoph Hellwig762a3af2020-06-04 16:51:18 -0700615 if (len == 0)
616 return -EFAULT;
617 if (!valid_arg_len(bprm, len))
618 return -E2BIG;
Oleg Nesterovba2d01622011-03-06 18:02:37 +0100619
Christoph Hellwig762a3af2020-06-04 16:51:18 -0700620 /* We're going to work our way backwards. */
621 arg += len;
622 bprm->p -= len;
623 if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
624 return -E2BIG;
625
626 while (len > 0) {
627 unsigned int bytes_to_copy = min_t(unsigned int, len,
628 min_not_zero(offset_in_page(pos), PAGE_SIZE));
629 struct page *page;
630 char *kaddr;
631
632 pos -= bytes_to_copy;
633 arg -= bytes_to_copy;
634 len -= bytes_to_copy;
635
636 page = get_arg_page(bprm, pos, 1);
637 if (!page)
638 return -E2BIG;
639 kaddr = kmap_atomic(page);
640 flush_arg_page(bprm, pos & PAGE_MASK, page);
641 memcpy(kaddr + offset_in_page(pos), arg, bytes_to_copy);
642 flush_kernel_dcache_page(page);
643 kunmap_atomic(kaddr);
644 put_arg_page(page);
645 }
646
647 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648}
Christoph Hellwig986db2d2020-06-04 16:51:14 -0700649EXPORT_SYMBOL(copy_string_kernel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Eric W. Biedermanbe619f72020-07-13 12:06:48 -0500651static int copy_strings_kernel(int argc, const char *const *argv,
652 struct linux_binprm *bprm)
653{
654 while (argc-- > 0) {
655 int ret = copy_string_kernel(argv[argc], bprm);
656 if (ret < 0)
657 return ret;
658 if (fatal_signal_pending(current))
659 return -ERESTARTNOHAND;
660 cond_resched();
661 }
662 return 0;
663}
664
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665#ifdef CONFIG_MMU
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700666
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667/*
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700668 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
669 * the binfmt code determines where the new stack should reside, we shift it to
670 * its final location. The process proceeds as follows:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 *
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700672 * 1) Use shift to calculate the new vma endpoints.
673 * 2) Extend vma to cover both the old and new ranges. This ensures the
674 * arguments passed to subsequent functions are consistent.
675 * 3) Move vma's page tables to the new range.
676 * 4) Free up any cleared pgd range.
677 * 5) Shrink the vma to cover only the new range.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700679static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680{
681 struct mm_struct *mm = vma->vm_mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700682 unsigned long old_start = vma->vm_start;
683 unsigned long old_end = vma->vm_end;
684 unsigned long length = old_end - old_start;
685 unsigned long new_start = old_start - shift;
686 unsigned long new_end = old_end - shift;
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700687 struct mmu_gather tlb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700689 BUG_ON(new_start > new_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700691 /*
692 * ensure there are no vmas between where we want to go
693 * and where we are
694 */
695 if (vma != find_vma(mm, new_start))
696 return -EFAULT;
697
698 /*
699 * cover the whole range: [new_start, old_end)
700 */
Rik van Riel5beb4932010-03-05 13:42:07 -0800701 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
702 return -ENOMEM;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700703
704 /*
705 * move the page tables downwards, on failure we rely on
706 * process cleanup to remove whatever mess we made.
707 */
708 if (length != move_page_tables(vma, old_start,
Michel Lespinasse38a76012012-10-08 16:31:50 -0700709 vma, new_start, length, false))
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700710 return -ENOMEM;
711
712 lru_add_drain();
Linus Torvalds2b047252013-08-15 11:42:25 -0700713 tlb_gather_mmu(&tlb, mm, old_start, old_end);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700714 if (new_end > old_start) {
715 /*
716 * when the old and new regions overlap clear from new_end.
717 */
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700718 free_pgd_range(&tlb, new_end, old_end, new_end,
Hugh Dickins6ee86302013-04-29 15:07:44 -0700719 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700720 } else {
721 /*
722 * otherwise, clean from old_start; this is done to not touch
723 * the address space in [new_end, old_start) some architectures
724 * have constraints on va-space that make this illegal (IA64) -
725 * for the others its just a little faster.
726 */
Peter Zijlstrad16dfc52011-05-24 17:11:45 -0700727 free_pgd_range(&tlb, old_start, old_end, new_end,
Hugh Dickins6ee86302013-04-29 15:07:44 -0700728 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 }
Linus Torvalds2b047252013-08-15 11:42:25 -0700730 tlb_finish_mmu(&tlb, old_start, old_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700732 /*
Rik van Riel5beb4932010-03-05 13:42:07 -0800733 * Shrink the vma to just the new range. Always succeeds.
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700734 */
735 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
736
737 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738}
739
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700740/*
741 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
742 * the stack is optionally relocated, and some extra space is added.
743 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744int setup_arg_pages(struct linux_binprm *bprm,
745 unsigned long stack_top,
746 int executable_stack)
747{
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700748 unsigned long ret;
749 unsigned long stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 struct mm_struct *mm = current->mm;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700751 struct vm_area_struct *vma = bprm->vma;
752 struct vm_area_struct *prev = NULL;
753 unsigned long vm_flags;
754 unsigned long stack_base;
Michael Neuling803bf5e2010-02-10 13:56:42 -0800755 unsigned long stack_size;
756 unsigned long stack_expand;
757 unsigned long rlim_stack;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
759#ifdef CONFIG_STACK_GROWSUP
James Hogand71f2902014-05-13 23:58:24 +0100760 /* Limit stack size */
Kees Cookc31dbb142018-04-10 16:35:01 -0700761 stack_base = bprm->rlim_stack.rlim_max;
James Hogand71f2902014-05-13 23:58:24 +0100762 if (stack_base > STACK_SIZE_MAX)
763 stack_base = STACK_SIZE_MAX;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Helge Dellerd045c772015-05-11 22:01:27 +0200765 /* Add space for stack randomization. */
766 stack_base += (STACK_RND_MASK << PAGE_SHIFT);
767
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700768 /* Make sure we didn't let the argument array grow too large. */
769 if (vma->vm_end - vma->vm_start > stack_base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 return -ENOMEM;
771
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700772 stack_base = PAGE_ALIGN(stack_top - stack_base);
773
774 stack_shift = vma->vm_start - stack_base;
775 mm->arg_start = bprm->p - stack_shift;
776 bprm->p = vma->vm_end - stack_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777#else
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700778 stack_top = arch_align_stack(stack_top);
779 stack_top = PAGE_ALIGN(stack_top);
Roland McGrath1b528182010-09-07 19:35:49 -0700780
781 if (unlikely(stack_top < mmap_min_addr) ||
782 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
783 return -ENOMEM;
784
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700785 stack_shift = vma->vm_end - stack_top;
786
787 bprm->p -= stack_shift;
788 mm->arg_start = bprm->p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789#endif
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700790
791 if (bprm->loader)
792 bprm->loader -= stack_shift;
793 bprm->exec -= stack_shift;
794
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700795 if (mmap_write_lock_killable(mm))
Michal Hockof268dfe2016-05-23 16:26:02 -0700796 return -EINTR;
797
Hugh Dickins96a8e132008-07-10 21:19:20 +0100798 vm_flags = VM_STACK_FLAGS;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700799
800 /*
801 * Adjust stack execute permissions; explicitly enable for
802 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
803 * (arch default) otherwise.
804 */
805 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
806 vm_flags |= VM_EXEC;
807 else if (executable_stack == EXSTACK_DISABLE_X)
808 vm_flags &= ~VM_EXEC;
809 vm_flags |= mm->def_flags;
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700810 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700811
812 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
813 vm_flags);
814 if (ret)
815 goto out_unlock;
816 BUG_ON(prev != vma);
817
Alexey Dobriyan47a2ebb2020-01-30 22:17:29 -0800818 if (unlikely(vm_flags & VM_EXEC)) {
819 pr_warn_once("process '%pD4' started with executable stack\n",
820 bprm->file);
821 }
822
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700823 /* Move stack pages down in memory. */
824 if (stack_shift) {
825 ret = shift_arg_pages(vma, stack_shift);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800826 if (ret)
827 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 }
829
Mel Gormana8bef8ff2010-05-24 14:32:24 -0700830 /* mprotect_fixup is overkill to remove the temporary stack flags */
831 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
832
Michael Neuling5ef097d2010-03-05 13:42:57 -0800833 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
Michael Neuling803bf5e2010-02-10 13:56:42 -0800834 stack_size = vma->vm_end - vma->vm_start;
835 /*
836 * Align this down to a page boundary as expand_stack
837 * will align it up.
838 */
Kees Cookc31dbb142018-04-10 16:35:01 -0700839 rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700840#ifdef CONFIG_STACK_GROWSUP
Michael Neuling803bf5e2010-02-10 13:56:42 -0800841 if (stack_size + stack_expand > rlim_stack)
842 stack_base = vma->vm_start + rlim_stack;
843 else
844 stack_base = vma->vm_end + stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700845#else
Michael Neuling803bf5e2010-02-10 13:56:42 -0800846 if (stack_size + stack_expand > rlim_stack)
847 stack_base = vma->vm_end - rlim_stack;
848 else
849 stack_base = vma->vm_start - stack_expand;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700850#endif
Eric B Munson3af9e852010-05-18 15:30:49 +0100851 current->mm->start_stack = bprm->p;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700852 ret = expand_stack(vma, stack_base);
853 if (ret)
854 ret = -EFAULT;
855
856out_unlock:
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700857 mmap_write_unlock(mm);
Anton Blanchardfc63cf22009-11-11 14:26:48 -0800858 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860EXPORT_SYMBOL(setup_arg_pages);
861
Nicolas Pitre7e7ec6a2016-07-24 11:30:18 -0400862#else
863
864/*
865 * Transfer the program arguments and environment from the holding pages
866 * onto the stack. The provided stack pointer is adjusted accordingly.
867 */
868int transfer_args_to_stack(struct linux_binprm *bprm,
869 unsigned long *sp_location)
870{
871 unsigned long index, stop, sp;
872 int ret = 0;
873
874 stop = bprm->p >> PAGE_SHIFT;
875 sp = *sp_location;
876
877 for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
878 unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
879 char *src = kmap(bprm->page[index]) + offset;
880 sp -= PAGE_SIZE - offset;
881 if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
882 ret = -EFAULT;
883 kunmap(bprm->page[index]);
884 if (ret)
885 goto out;
886 }
887
888 *sp_location = sp;
889
890out:
891 return ret;
892}
893EXPORT_SYMBOL(transfer_args_to_stack);
894
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895#endif /* CONFIG_MMU */
896
David Drysdale51f39a12014-12-12 16:57:29 -0800897static struct file *do_open_execat(int fd, struct filename *name, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 struct file *file;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200900 int err;
David Drysdale51f39a12014-12-12 16:57:29 -0800901 struct open_flags open_exec_flags = {
Al Viro47c805d2011-02-23 17:44:09 -0500902 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
Al Viro62fb4a12015-12-26 22:33:24 -0500903 .acc_mode = MAY_EXEC,
Al Virof9652e12013-06-11 08:23:01 +0400904 .intent = LOOKUP_OPEN,
905 .lookup_flags = LOOKUP_FOLLOW,
Al Viro47c805d2011-02-23 17:44:09 -0500906 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
David Drysdale51f39a12014-12-12 16:57:29 -0800908 if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
909 return ERR_PTR(-EINVAL);
910 if (flags & AT_SYMLINK_NOFOLLOW)
911 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
912 if (flags & AT_EMPTY_PATH)
913 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
914
915 file = do_filp_open(fd, name, &open_exec_flags);
Al Viro6e8341a2009-04-06 11:16:22 -0400916 if (IS_ERR(file))
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200917 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918
Kees Cook633fb6a2020-08-11 18:36:26 -0700919 /*
920 * may_open() has already checked for this, so it should be
921 * impossible to trip now. But we need to be extra cautious
922 * and check again at the very end too.
923 */
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200924 err = -EACCES;
Kees Cook0fd338b2020-08-11 18:36:30 -0700925 if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
926 path_noexec(&file->f_path)))
Al Viro6e8341a2009-04-06 11:16:22 -0400927 goto exit;
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200928
929 err = deny_write_access(file);
Al Viro6e8341a2009-04-06 11:16:22 -0400930 if (err)
931 goto exit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
David Drysdale51f39a12014-12-12 16:57:29 -0800933 if (name->name[0] != '\0')
934 fsnotify_open(file);
935
Al Viro6e8341a2009-04-06 11:16:22 -0400936out:
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200937 return file;
938
Al Viro6e8341a2009-04-06 11:16:22 -0400939exit:
940 fput(file);
Christoph Hellwige56b6a52008-05-19 07:53:34 +0200941 return ERR_PTR(err);
942}
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800943
944struct file *open_exec(const char *name)
945{
Paul Moore51689102015-01-22 00:00:03 -0500946 struct filename *filename = getname_kernel(name);
947 struct file *f = ERR_CAST(filename);
948
949 if (!IS_ERR(filename)) {
950 f = do_open_execat(AT_FDCWD, filename, 0);
951 putname(filename);
952 }
953 return f;
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -0800954}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955EXPORT_SYMBOL(open_exec);
956
Christoph Hellwig48304f72020-06-07 21:42:40 -0700957#if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
958 defined(CONFIG_BINFMT_ELF_FDPIC)
Al Viro3dc20cb2013-04-13 20:31:37 -0400959ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
960{
Al Viroec695572014-02-04 19:08:21 -0500961 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
Al Viro3dc20cb2013-04-13 20:31:37 -0400962 if (res > 0)
Christoph Hellwigbce2b682020-06-07 21:42:43 -0700963 flush_icache_user_range(addr, addr + len);
Al Viro3dc20cb2013-04-13 20:31:37 -0400964 return res;
965}
966EXPORT_SYMBOL(read_code);
Christoph Hellwig48304f72020-06-07 21:42:40 -0700967#endif
Al Viro3dc20cb2013-04-13 20:31:37 -0400968
Eric W. Biedermaneea96732020-03-25 10:03:36 -0500969/*
970 * Maps the mm_struct mm into the current task struct.
Eric W. Biedermanab7709b2020-12-03 14:12:00 -0600971 * On success, this function returns with exec_update_lock
972 * held for writing.
Eric W. Biedermaneea96732020-03-25 10:03:36 -0500973 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974static int exec_mmap(struct mm_struct *mm)
975{
976 struct task_struct *tsk;
Davidlohr Bueso615d6e82014-04-07 15:37:25 -0700977 struct mm_struct *old_mm, *active_mm;
Eric W. Biedermaneea96732020-03-25 10:03:36 -0500978 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979
980 /* Notify parent that we're no longer interested in the old VM */
981 tsk = current;
982 old_mm = current->mm;
Thomas Gleixner4610ba72019-11-06 22:55:38 +0100983 exec_mm_release(tsk, old_mm);
Eric W. Biedermana28bf132020-03-30 16:33:39 -0500984 if (old_mm)
985 sync_mm_rss(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986
Eric W. Biedermanab7709b2020-12-03 14:12:00 -0600987 ret = down_write_killable(&tsk->signal->exec_update_lock);
Eric W. Biedermaneea96732020-03-25 10:03:36 -0500988 if (ret)
989 return ret;
990
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 if (old_mm) {
992 /*
993 * Make sure that if there is a core dump in progress
994 * for the old mm, we get out and die instead of going
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -0700995 * through with the exec. We must hold mmap_lock around
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700996 * checking core_state and changing tsk->mm.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 */
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700998 mmap_read_lock(old_mm);
Oleg Nesterov999d9fc2008-07-25 01:47:41 -0700999 if (unlikely(old_mm->core_state)) {
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07001000 mmap_read_unlock(old_mm);
Eric W. Biedermanab7709b2020-12-03 14:12:00 -06001001 up_write(&tsk->signal->exec_update_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 return -EINTR;
1003 }
1004 }
Eric W. Biedermaneea96732020-03-25 10:03:36 -05001005
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 task_lock(tsk);
Mathieu Desnoyers227a4aa2019-09-19 13:37:02 -04001007 membarrier_exec_mmap(mm);
Nicholas Piggind53c3df2020-09-14 14:52:16 +10001008
1009 local_irq_disable();
1010 active_mm = tsk->active_mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 tsk->active_mm = mm;
Nicholas Piggind53c3df2020-09-14 14:52:16 +10001012 tsk->mm = mm;
1013 /*
1014 * This prevents preemption while active_mm is being loaded and
1015 * it and mm are being updated, which could cause problems for
1016 * lazy tlb mm refcounting when these are updated by context
1017 * switches. Not all architectures can handle irqs off over
1018 * activate_mm yet.
1019 */
1020 if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1021 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 activate_mm(active_mm, mm);
Nicholas Piggind53c3df2020-09-14 14:52:16 +10001023 if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1024 local_irq_enable();
Davidlohr Bueso615d6e82014-04-07 15:37:25 -07001025 tsk->mm->vmacache_seqnum = 0;
1026 vmacache_flush(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 task_unlock(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 if (old_mm) {
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07001029 mmap_read_unlock(old_mm);
Eric Sesterhenn7dddb122006-04-01 01:13:38 +02001030 BUG_ON(active_mm != old_mm);
Oleg Nesterov701085b22012-03-19 17:04:01 +01001031 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
Balbir Singh31a78f22008-09-28 23:09:31 +01001032 mm_update_next_owner(old_mm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 mmput(old_mm);
1034 return 0;
1035 }
1036 mmdrop(active_mm);
1037 return 0;
1038}
1039
Arjan van de Ven858119e2006-01-14 13:20:43 -08001040static int de_thread(struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041{
1042 struct signal_struct *sig = tsk->signal;
Oleg Nesterovb2c903b2007-10-16 23:27:22 -07001043 struct sighand_struct *oldsighand = tsk->sighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 spinlock_t *lock = &oldsighand->siglock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001046 if (thread_group_empty(tsk))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 goto no_thread_group;
1048
1049 /*
1050 * Kill all other threads in the thread group.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 spin_lock_irq(lock);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001053 if (signal_group_exit(sig)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 /*
1055 * Another group action in progress, just
1056 * return so that the signal is processed.
1057 */
1058 spin_unlock_irq(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 return -EAGAIN;
1060 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061
Oleg Nesterovd3441932010-05-26 14:43:11 -07001062 sig->group_exit_task = tsk;
1063 sig->notify_count = zap_other_threads(tsk);
1064 if (!thread_group_leader(tsk))
1065 sig->notify_count--;
1066
1067 while (sig->notify_count) {
Oleg Nesterovd5bbd432012-10-08 19:13:01 +02001068 __set_current_state(TASK_KILLABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 spin_unlock_irq(lock);
Rafael J. Wysockia72173e2018-12-03 13:04:18 +01001070 schedule();
Davidlohr Bueso08d405c2019-01-03 15:28:58 -08001071 if (__fatal_signal_pending(tsk))
Oleg Nesterovd5bbd432012-10-08 19:13:01 +02001072 goto killed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 spin_lock_irq(lock);
1074 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 spin_unlock_irq(lock);
1076
1077 /*
1078 * At this point all other threads have exited, all we have to
1079 * do is to wait for the thread group leader to become inactive,
1080 * and to assume its PID:
1081 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001082 if (!thread_group_leader(tsk)) {
Oleg Nesterov81879262008-12-01 14:18:16 -08001083 struct task_struct *leader = tsk->group_leader;
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001084
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001085 for (;;) {
Ingo Molnar780de9d2017-02-02 11:50:56 +01001086 cgroup_threadgroup_change_begin(tsk);
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001087 write_lock_irq(&tasklist_lock);
Kirill Tkhaidfcce792015-04-16 12:48:01 -07001088 /*
1089 * Do this under tasklist_lock to ensure that
1090 * exit_notify() can't miss ->group_exit_task
1091 */
1092 sig->notify_count = -1;
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001093 if (likely(leader->exit_state))
1094 break;
Oleg Nesterovd5bbd432012-10-08 19:13:01 +02001095 __set_current_state(TASK_KILLABLE);
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001096 write_unlock_irq(&tasklist_lock);
Ingo Molnar780de9d2017-02-02 11:50:56 +01001097 cgroup_threadgroup_change_end(tsk);
Rafael J. Wysockia72173e2018-12-03 13:04:18 +01001098 schedule();
Davidlohr Bueso08d405c2019-01-03 15:28:58 -08001099 if (__fatal_signal_pending(tsk))
Oleg Nesterovd5bbd432012-10-08 19:13:01 +02001100 goto killed;
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001101 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
Roland McGrathf5e90282006-04-10 22:54:16 -07001103 /*
1104 * The only record we have of the real-time age of a
1105 * process, regardless of execs it's done, is start_time.
1106 * All the past CPU time is accumulated in signal_struct
1107 * from sister threads now dead. But in this non-leader
1108 * exec, nothing survives from the original leader thread,
1109 * whose birth marks the true age of this process now.
1110 * When we take on its identity by switching to its PID, we
1111 * also take its birthdate (always earlier than our own).
1112 */
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001113 tsk->start_time = leader->start_time;
Peter Zijlstracf25e242019-11-07 11:07:58 +01001114 tsk->start_boottime = leader->start_boottime;
Roland McGrathf5e90282006-04-10 22:54:16 -07001115
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -07001116 BUG_ON(!same_thread_group(leader, tsk));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 /*
1118 * An exec() starts a new thread group with the
1119 * TGID of the previous thread group. Rehash the
1120 * two threads with a switched PID, and release
1121 * the former thread group leader:
1122 */
Eric W. Biedermand73d6522006-03-28 16:11:03 -08001123
1124 /* Become a process group leader with the old leader's pid.
Eric W. Biedermanc18258c2006-09-27 01:51:06 -07001125 * The old leader becomes a thread of the this thread group.
Eric W. Biedermand73d6522006-03-28 16:11:03 -08001126 */
Eric W. Biederman6b03d132020-04-19 06:35:02 -05001127 exchange_tids(tsk, leader);
Eric W. Biederman6883f812017-06-04 04:32:13 -05001128 transfer_pid(leader, tsk, PIDTYPE_TGID);
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001129 transfer_pid(leader, tsk, PIDTYPE_PGID);
1130 transfer_pid(leader, tsk, PIDTYPE_SID);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -08001131
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001132 list_replace_rcu(&leader->tasks, &tsk->tasks);
Oleg Nesterov9cd80bb2009-12-17 15:27:15 -08001133 list_replace_init(&leader->sibling, &tsk->sibling);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001135 tsk->group_leader = tsk;
1136 leader->group_leader = tsk;
Eric W. Biedermande12a782006-04-10 17:16:49 -06001137
Eric W. Biedermanaafe6c22006-09-27 01:51:13 -07001138 tsk->exit_signal = SIGCHLD;
Oleg Nesterov087806b2011-06-22 23:10:26 +02001139 leader->exit_signal = -1;
Oleg Nesterov962b5642005-11-23 13:37:43 -08001140
1141 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1142 leader->exit_state = EXIT_DEAD;
Oleg Nesteroveac1b5e2011-07-21 20:00:43 +02001143
1144 /*
1145 * We are going to release_task()->ptrace_unlink() silently,
1146 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1147 * the tracer wont't block again waiting for this thread.
1148 */
1149 if (unlikely(leader->ptrace))
1150 __wake_up_parent(leader, leader->parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 write_unlock_irq(&tasklist_lock);
Ingo Molnar780de9d2017-02-02 11:50:56 +01001152 cgroup_threadgroup_change_end(tsk);
Oleg Nesterov81879262008-12-01 14:18:16 -08001153
1154 release_task(leader);
Oleg Nesteroved5d2ca2008-02-04 22:27:24 -08001155 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156
Oleg Nesterov6db840f2007-10-16 23:27:23 -07001157 sig->group_exit_task = NULL;
1158 sig->notify_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159
1160no_thread_group:
Oleg Nesterove6368252012-03-19 17:03:22 +01001161 /* we have changed execution domain */
1162 tsk->exit_signal = SIGCHLD;
1163
Eric W. Biederman02169152020-03-25 10:00:21 -05001164 BUG_ON(!thread_group_leader(tsk));
1165 return 0;
1166
1167killed:
1168 /* protects against exit_notify() and __exit_signal() */
1169 read_lock(&tasklist_lock);
1170 sig->group_exit_task = NULL;
1171 sig->notify_count = 0;
1172 read_unlock(&tasklist_lock);
1173 return -EAGAIN;
1174}
1175
1176
Eric W. Biederman7a60ef42020-03-08 12:04:44 -05001177/*
1178 * This function makes sure the current process has its own signal table,
1179 * so that flush_signal_handlers can later reset the handlers without
1180 * disturbing other processes. (Other processes might share the signal
1181 * table via the CLONE_SIGHAND option to clone().)
1182 */
Eric W. Biederman02169152020-03-25 10:00:21 -05001183static int unshare_sighand(struct task_struct *me)
1184{
1185 struct sighand_struct *oldsighand = me->sighand;
Oleg Nesterov329f7db2005-11-07 21:12:43 +03001186
Elena Reshetovad036bda2019-01-18 14:27:26 +02001187 if (refcount_read(&oldsighand->count) != 1) {
Oleg Nesterovb2c903b2007-10-16 23:27:22 -07001188 struct sighand_struct *newsighand;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 /*
Oleg Nesterovb2c903b2007-10-16 23:27:22 -07001190 * This ->sighand is shared with the CLONE_SIGHAND
1191 * but not CLONE_THREAD task, switch to the new one.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 */
Oleg Nesterovb2c903b2007-10-16 23:27:22 -07001193 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1194 if (!newsighand)
1195 return -ENOMEM;
1196
Elena Reshetovad036bda2019-01-18 14:27:26 +02001197 refcount_set(&newsighand->count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 memcpy(newsighand->action, oldsighand->action,
1199 sizeof(newsighand->action));
1200
1201 write_lock_irq(&tasklist_lock);
1202 spin_lock(&oldsighand->siglock);
Eric W. Biederman02169152020-03-25 10:00:21 -05001203 rcu_assign_pointer(me->sighand, newsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 spin_unlock(&oldsighand->siglock);
1205 write_unlock_irq(&tasklist_lock);
1206
Davide Libenzifba2afa2007-05-10 22:23:13 -07001207 __cleanup_sighand(oldsighand);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 return 0;
1210}
Oleg Nesterov0840a902007-10-16 23:27:22 -07001211
Arnd Bergmann3756f642017-12-14 15:32:41 -08001212char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 task_lock(tsk);
Arnd Bergmann3756f642017-12-14 15:32:41 -08001215 strncpy(buf, tsk->comm, buf_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 task_unlock(tsk);
Andrew Morton59714d62008-02-04 22:27:21 -08001217 return buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218}
Arnd Bergmann3756f642017-12-14 15:32:41 -08001219EXPORT_SYMBOL_GPL(__get_task_comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220
Al Viro6a6d27d2012-08-21 09:56:33 -04001221/*
1222 * These functions flushes out all traces of the currently running executable
1223 * so that a new one can be started
1224 */
1225
Adrian Hunter82b89772014-05-28 11:45:04 +03001226void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227{
1228 task_lock(tsk);
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -08001229 trace_task_rename(tsk, buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1231 task_unlock(tsk);
Adrian Hunter82b89772014-05-28 11:45:04 +03001232 perf_event_comm(tsk, exec);
ted.lincadbca12021-05-26 22:00:14 +08001233 trace_android_vh_set_task_comm(tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234}
1235
Kees Cooka9208e42017-07-18 15:25:30 -07001236/*
1237 * Calling this is the point of no return. None of the failures will be
1238 * seen by userspace since either the process is already taking a fatal
1239 * signal (via de_thread() or coredump), or will have SEGV raised
Eric W. Biederman13c432b2020-03-19 17:16:12 -05001240 * (after exec_mmap()) by search_binary_handler (see below).
Kees Cooka9208e42017-07-18 15:25:30 -07001241 */
Eric W. Biederman23887772020-05-03 07:54:10 -05001242int begin_new_exec(struct linux_binprm * bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243{
Eric W. Biederman2ca7be72020-03-25 10:00:07 -05001244 struct task_struct *me = current;
Linus Torvalds221af7f2010-01-28 22:14:42 -08001245 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
Eric W. Biederman56305aa2020-05-29 22:00:54 -05001247 /* Once we are committed compute the creds */
1248 retval = bprm_creds_from_file(bprm);
1249 if (retval)
1250 return retval;
1251
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 /*
Eric W. Biederman6834e0b2020-04-04 12:01:37 -05001253 * Ensure all future errors are fatal.
1254 */
1255 bprm->point_of_no_return = true;
1256
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 /*
Eric W. Biederman02169152020-03-25 10:00:21 -05001258 * Make this the only thread in the thread group.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 */
Eric W. Biederman2ca7be72020-03-25 10:00:07 -05001260 retval = de_thread(me);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 if (retval)
1262 goto out;
1263
Davidlohr Bueso6e399cd2015-04-16 12:47:59 -07001264 /*
1265 * Must be called _before_ exec_mmap() as bprm->mm is
1266 * not visibile until then. This also enables the update
1267 * to be lockless.
1268 */
Matt Helsley925d1c42008-04-29 01:01:36 -07001269 set_mm_exe_file(bprm->mm, bprm->file);
Davidlohr Bueso6e399cd2015-04-16 12:47:59 -07001270
Eric W. Biedermanb8a61c92020-05-14 15:17:40 -05001271 /* If the binary is not readable then enforce mm->dumpable=0 */
Eric W. Biedermanf87d1c92020-05-16 16:29:20 -05001272 would_dump(bprm, bprm->file);
Eric W. Biedermanb8a61c92020-05-14 15:17:40 -05001273 if (bprm->have_execfd)
1274 would_dump(bprm, bprm->executable);
Eric W. Biedermanf87d1c92020-05-16 16:29:20 -05001275
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 * Release all of the old mmap stuff
1278 */
Oleg Nesterov3c77f842010-11-30 20:55:34 +01001279 acct_arg_size(bprm, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 retval = exec_mmap(bprm->mm);
1281 if (retval)
Al Virofd8328b2008-04-22 05:11:59 -04001282 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
Kees Cooka9208e42017-07-18 15:25:30 -07001284 bprm->mm = NULL;
Linus Torvalds7ab02af2010-02-02 12:37:44 -08001285
Eric W. Biedermanccf0fa62020-03-25 10:03:21 -05001286#ifdef CONFIG_POSIX_TIMERS
1287 exit_itimers(me->signal);
1288 flush_itimer_signals();
1289#endif
1290
1291 /*
1292 * Make the signal table private.
1293 */
1294 retval = unshare_sighand(me);
1295 if (retval)
Eric W. Biederman89826cc2020-04-02 18:04:54 -05001296 goto out_unlock;
Eric W. Biedermanccf0fa62020-03-25 10:03:21 -05001297
Christoph Hellwigfe814172020-08-11 18:33:50 -07001298 /*
1299 * Ensure that the uaccess routines can actually operate on userspace
1300 * pointers:
1301 */
1302 force_uaccess_begin();
1303
Eric W. Biederman2ca7be72020-03-25 10:00:07 -05001304 me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
Zhang Yib88fae62014-01-23 15:55:57 -08001305 PF_NOFREEZE | PF_NO_SETAFFINITY);
Linus Torvalds7ab02af2010-02-02 12:37:44 -08001306 flush_thread();
Eric W. Biederman2ca7be72020-03-25 10:00:07 -05001307 me->personality &= ~bprm->per_clear;
Linus Torvalds7ab02af2010-02-02 12:37:44 -08001308
Aleksa Sarai613cc2b2016-12-21 16:26:24 +11001309 /*
1310 * We have to apply CLOEXEC before we change whether the process is
1311 * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1312 * trying to access the should-be-closed file descriptors of a process
1313 * undergoing exec(2).
1314 */
Eric W. Biederman2ca7be72020-03-25 10:00:07 -05001315 do_close_on_exec(me->files);
Eric W. Biedermandf9e4d22020-05-03 07:15:28 -05001316
Eric W. Biedermandf9e4d22020-05-03 07:15:28 -05001317 if (bprm->secureexec) {
1318 /* Make sure parent cannot signal privileged process. */
1319 me->pdeath_signal = 0;
1320
1321 /*
1322 * For secureexec, reset the stack limit to sane default to
1323 * avoid bad behavior from the prior rlimits. This has to
1324 * happen before arch_pick_mmap_layout(), which examines
1325 * RLIMIT_STACK, but after the point of no return to avoid
1326 * needing to clean up the change on failure.
1327 */
1328 if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1329 bprm->rlim_stack.rlim_cur = _STK_LIM;
1330 }
1331
1332 me->sas_ss_sp = me->sas_ss_size = 0;
1333
1334 /*
1335 * Figure out dumpability. Note that this checking only of current
1336 * is wrong, but userspace depends on it. This should be testing
1337 * bprm->secureexec instead.
1338 */
1339 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1340 !(uid_eq(current_euid(), current_uid()) &&
1341 gid_eq(current_egid(), current_gid())))
1342 set_dumpable(current->mm, suid_dumpable);
1343 else
1344 set_dumpable(current->mm, SUID_DUMP_USER);
1345
1346 perf_event_exec();
1347 __set_task_comm(me, kbasename(bprm->filename), true);
1348
1349 /* An exec changes our domain. We are no longer part of the thread
1350 group */
1351 WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1352 flush_signal_handlers(me, 0);
1353
1354 /*
1355 * install the new credentials for this executable
1356 */
1357 security_bprm_committing_creds(bprm);
1358
1359 commit_creds(bprm->cred);
1360 bprm->cred = NULL;
1361
1362 /*
1363 * Disable monitoring for regular users
1364 * when executing setuid binaries. Must
1365 * wait until new credentials are committed
1366 * by commit_creds() above
1367 */
1368 if (get_dumpable(me->mm) != SUID_DUMP_USER)
1369 perf_event_exit_task(me);
1370 /*
1371 * cred_guard_mutex must be held at least to this point to prevent
1372 * ptrace_attach() from altering our determination of the task's
1373 * credentials; any time after this it may be unlocked.
1374 */
1375 security_bprm_committed_creds(bprm);
Eric W. Biedermanb8a61c92020-05-14 15:17:40 -05001376
1377 /* Pass the opened binary to the interpreter. */
1378 if (bprm->have_execfd) {
1379 retval = get_unused_fd_flags(0);
1380 if (retval < 0)
1381 goto out_unlock;
1382 fd_install(retval, bprm->executable);
1383 bprm->executable = NULL;
1384 bprm->execfd = retval;
1385 }
Linus Torvalds221af7f2010-01-28 22:14:42 -08001386 return 0;
1387
Eric W. Biederman89826cc2020-04-02 18:04:54 -05001388out_unlock:
Eric W. Biedermanab7709b2020-12-03 14:12:00 -06001389 up_write(&me->signal->exec_update_lock);
Linus Torvalds221af7f2010-01-28 22:14:42 -08001390out:
1391 return retval;
1392}
Eric W. Biederman23887772020-05-03 07:54:10 -05001393EXPORT_SYMBOL(begin_new_exec);
Linus Torvalds221af7f2010-01-28 22:14:42 -08001394
Al Viro1b5d7832011-06-19 12:49:47 -04001395void would_dump(struct linux_binprm *bprm, struct file *file)
1396{
Eric W. Biedermanf84df2a2016-11-16 22:06:51 -06001397 struct inode *inode = file_inode(file);
1398 if (inode_permission(inode, MAY_READ) < 0) {
1399 struct user_namespace *old, *user_ns;
Al Viro1b5d7832011-06-19 12:49:47 -04001400 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
Eric W. Biedermanf84df2a2016-11-16 22:06:51 -06001401
1402 /* Ensure mm->user_ns contains the executable */
1403 user_ns = old = bprm->mm->user_ns;
1404 while ((user_ns != &init_user_ns) &&
1405 !privileged_wrt_inode_uidgid(user_ns, inode))
1406 user_ns = user_ns->parent;
1407
1408 if (old != user_ns) {
1409 bprm->mm->user_ns = get_user_ns(user_ns);
1410 put_user_ns(old);
1411 }
1412 }
Al Viro1b5d7832011-06-19 12:49:47 -04001413}
1414EXPORT_SYMBOL(would_dump);
1415
Linus Torvalds221af7f2010-01-28 22:14:42 -08001416void setup_new_exec(struct linux_binprm * bprm)
1417{
Eric W. Biedermandf9e4d22020-05-03 07:15:28 -05001418 /* Setup things that can depend upon the personality */
Eric W. Biederman7d503fe2020-04-02 18:35:14 -05001419 struct task_struct *me = current;
Kees Cook46d98eb2017-07-18 15:25:27 -07001420
Eric W. Biederman7d503fe2020-04-02 18:35:14 -05001421 arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
Alan Coxd6e71142005-06-23 00:09:43 -07001422
Kyle Hueye9ea1e72017-03-20 01:16:26 -07001423 arch_setup_new_exec();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424
Benjamin Herrenschmidt0551fbd2006-02-28 16:59:19 -08001425 /* Set the new mm task size. We have to do that late because it may
1426 * depend on TIF_32BIT which is only updated in flush_thread() on
1427 * some architectures like powerpc
1428 */
Eric W. Biederman7d503fe2020-04-02 18:35:14 -05001429 me->mm->task_size = TASK_SIZE;
Eric W. Biedermanab7709b2020-12-03 14:12:00 -06001430 up_write(&me->signal->exec_update_lock);
Eric W. Biederman7d503fe2020-04-02 18:35:14 -05001431 mutex_unlock(&me->signal->cred_guard_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432}
Linus Torvalds221af7f2010-01-28 22:14:42 -08001433EXPORT_SYMBOL(setup_new_exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434
Kees Cookb8383832018-04-10 16:34:57 -07001435/* Runs immediately before start_thread() takes over. */
1436void finalize_exec(struct linux_binprm *bprm)
1437{
Kees Cookc31dbb142018-04-10 16:35:01 -07001438 /* Store any stack rlimit changes before starting thread. */
1439 task_lock(current->group_leader);
1440 current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1441 task_unlock(current->group_leader);
Kees Cookb8383832018-04-10 16:34:57 -07001442}
1443EXPORT_SYMBOL(finalize_exec);
1444
David Howellsa6f76f22008-11-14 10:39:24 +11001445/*
Oleg Nesterova2a84742009-09-05 11:17:13 -07001446 * Prepare credentials and lock ->cred_guard_mutex.
Eric W. Biederman96ecee29b2020-05-03 06:48:17 -05001447 * setup_new_exec() commits the new creds and drops the lock.
Oleg Nesterova2a84742009-09-05 11:17:13 -07001448 * Or, if exec fails before, free_bprm() should release ->cred and
1449 * and unlock.
1450 */
Chanho Min4addd262018-12-10 16:49:54 +09001451static int prepare_bprm_creds(struct linux_binprm *bprm)
Oleg Nesterova2a84742009-09-05 11:17:13 -07001452{
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001453 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
Oleg Nesterova2a84742009-09-05 11:17:13 -07001454 return -ERESTARTNOINTR;
1455
1456 bprm->cred = prepare_exec_creds();
1457 if (likely(bprm->cred))
1458 return 0;
1459
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001460 mutex_unlock(&current->signal->cred_guard_mutex);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001461 return -ENOMEM;
1462}
1463
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -08001464static void free_bprm(struct linux_binprm *bprm)
Oleg Nesterova2a84742009-09-05 11:17:13 -07001465{
Eric W. Biedermanf18ac552020-07-10 15:54:54 -05001466 if (bprm->mm) {
1467 acct_arg_size(bprm, 0);
1468 mmput(bprm->mm);
1469 }
Oleg Nesterova2a84742009-09-05 11:17:13 -07001470 free_arg_pages(bprm);
1471 if (bprm->cred) {
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001472 mutex_unlock(&current->signal->cred_guard_mutex);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001473 abort_creds(bprm->cred);
1474 }
Oleg Nesterov63e46b92014-01-23 15:55:51 -08001475 if (bprm->file) {
1476 allow_write_access(bprm->file);
1477 fput(bprm->file);
1478 }
Eric W. Biedermanb8a61c92020-05-14 15:17:40 -05001479 if (bprm->executable)
1480 fput(bprm->executable);
Kees Cookb66c5982012-12-20 15:05:16 -08001481 /* If a binfmt changed the interp, free it. */
1482 if (bprm->interp != bprm->filename)
1483 kfree(bprm->interp);
Eric W. Biederman60d9ad12020-07-11 08:16:15 -05001484 kfree(bprm->fdpath);
Oleg Nesterova2a84742009-09-05 11:17:13 -07001485 kfree(bprm);
1486}
1487
Eric W. Biederman60d9ad12020-07-11 08:16:15 -05001488static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
Eric W. Biederman0a8f36e2020-07-10 15:39:45 -05001489{
1490 struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
Eric W. Biederman60d9ad12020-07-11 08:16:15 -05001491 int retval = -ENOMEM;
Eric W. Biederman0a8f36e2020-07-10 15:39:45 -05001492 if (!bprm)
Eric W. Biederman60d9ad12020-07-11 08:16:15 -05001493 goto out;
1494
1495 if (fd == AT_FDCWD || filename->name[0] == '/') {
1496 bprm->filename = filename->name;
1497 } else {
1498 if (filename->name[0] == '\0')
1499 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1500 else
1501 bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1502 fd, filename->name);
1503 if (!bprm->fdpath)
1504 goto out_free;
1505
1506 bprm->filename = bprm->fdpath;
1507 }
1508 bprm->interp = bprm->filename;
Eric W. Biedermanf18ac552020-07-10 15:54:54 -05001509
1510 retval = bprm_mm_init(bprm);
1511 if (retval)
1512 goto out_free;
Eric W. Biederman0a8f36e2020-07-10 15:39:45 -05001513 return bprm;
Eric W. Biederman60d9ad12020-07-11 08:16:15 -05001514
1515out_free:
1516 free_bprm(bprm);
1517out:
1518 return ERR_PTR(retval);
Eric W. Biederman0a8f36e2020-07-10 15:39:45 -05001519}
1520
Oleg Nesterovc2315c12017-10-03 16:15:42 -07001521int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
Kees Cookb66c5982012-12-20 15:05:16 -08001522{
1523 /* If a binfmt changed the interp, free it first. */
1524 if (bprm->interp != bprm->filename)
1525 kfree(bprm->interp);
1526 bprm->interp = kstrdup(interp, GFP_KERNEL);
1527 if (!bprm->interp)
1528 return -ENOMEM;
1529 return 0;
1530}
1531EXPORT_SYMBOL(bprm_change_interp);
1532
Oleg Nesterova2a84742009-09-05 11:17:13 -07001533/*
David Howellsa6f76f22008-11-14 10:39:24 +11001534 * determine how safe it is to execute the proposed program
KOSAKI Motohiro9b1bf122010-10-27 15:34:08 -07001535 * - the caller must hold ->cred_guard_mutex to protect against
Kees Cookc2e1f2e2014-06-05 00:23:17 -07001536 * PTRACE_ATTACH or seccomp thread-sync
David Howellsa6f76f22008-11-14 10:39:24 +11001537 */
Oleg Nesterov9e00cdb2014-01-23 15:55:50 -08001538static void check_unsafe_exec(struct linux_binprm *bprm)
David Howellsa6f76f22008-11-14 10:39:24 +11001539{
David Howells0bf2f3a2009-02-06 11:45:46 +00001540 struct task_struct *p = current, *t;
Al Virof1191b52009-03-30 07:35:18 -04001541 unsigned n_fs;
David Howellsa6f76f22008-11-14 10:39:24 +11001542
Eric W. Biederman9227dd22017-01-23 17:26:31 +13001543 if (p->ptrace)
1544 bprm->unsafe |= LSM_UNSAFE_PTRACE;
David Howellsa6f76f22008-11-14 10:39:24 +11001545
Andy Lutomirski259e5e62012-04-12 16:47:50 -05001546 /*
1547 * This isn't strictly necessary, but it makes it harder for LSMs to
1548 * mess up.
1549 */
Kees Cook1d4457f2014-05-21 15:23:46 -07001550 if (task_no_new_privs(current))
Andy Lutomirski259e5e62012-04-12 16:47:50 -05001551 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1552
Oleg Nesterov83f62a22014-01-23 15:55:49 -08001553 t = p;
David Howells0bf2f3a2009-02-06 11:45:46 +00001554 n_fs = 1;
Nick Piggin2a4419b2010-08-18 04:37:33 +10001555 spin_lock(&p->fs->lock);
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001556 rcu_read_lock();
Oleg Nesterov83f62a22014-01-23 15:55:49 -08001557 while_each_thread(p, t) {
David Howells0bf2f3a2009-02-06 11:45:46 +00001558 if (t->fs == p->fs)
1559 n_fs++;
David Howells0bf2f3a2009-02-06 11:45:46 +00001560 }
Oleg Nesterov437f7fd2009-04-24 01:02:45 +02001561 rcu_read_unlock();
David Howells0bf2f3a2009-02-06 11:45:46 +00001562
Oleg Nesterov9e00cdb2014-01-23 15:55:50 -08001563 if (p->fs->users > n_fs)
David Howellsa6f76f22008-11-14 10:39:24 +11001564 bprm->unsafe |= LSM_UNSAFE_SHARE;
Oleg Nesterov9e00cdb2014-01-23 15:55:50 -08001565 else
1566 p->fs->in_exec = 1;
Nick Piggin2a4419b2010-08-18 04:37:33 +10001567 spin_unlock(&p->fs->lock);
David Howellsa6f76f22008-11-14 10:39:24 +11001568}
1569
Eric W. Biederman56305aa2020-05-29 22:00:54 -05001570static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
Jann Horn8b01fc82015-04-19 02:48:39 +02001571{
Eric W. Biederman56305aa2020-05-29 22:00:54 -05001572 /* Handle suid and sgid on files */
Jann Horn8b01fc82015-04-19 02:48:39 +02001573 struct inode *inode;
1574 unsigned int mode;
1575 kuid_t uid;
1576 kgid_t gid;
1577
Eric W. Biederman56305aa2020-05-29 22:00:54 -05001578 if (!mnt_may_suid(file->f_path.mnt))
Jann Horn8b01fc82015-04-19 02:48:39 +02001579 return;
1580
1581 if (task_no_new_privs(current))
1582 return;
1583
Eric W. Biederman56305aa2020-05-29 22:00:54 -05001584 inode = file->f_path.dentry->d_inode;
Jann Horn8b01fc82015-04-19 02:48:39 +02001585 mode = READ_ONCE(inode->i_mode);
1586 if (!(mode & (S_ISUID|S_ISGID)))
1587 return;
1588
1589 /* Be careful if suid/sgid is set */
Al Viro59551022016-01-22 15:40:57 -05001590 inode_lock(inode);
Jann Horn8b01fc82015-04-19 02:48:39 +02001591
1592 /* reload atomically mode/uid/gid now that lock held */
1593 mode = inode->i_mode;
1594 uid = inode->i_uid;
1595 gid = inode->i_gid;
Al Viro59551022016-01-22 15:40:57 -05001596 inode_unlock(inode);
Jann Horn8b01fc82015-04-19 02:48:39 +02001597
1598 /* We ignore suid/sgid if there are no mappings for them in the ns */
1599 if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1600 !kgid_has_mapping(bprm->cred->user_ns, gid))
1601 return;
1602
1603 if (mode & S_ISUID) {
1604 bprm->per_clear |= PER_CLEAR_ON_SETID;
1605 bprm->cred->euid = uid;
1606 }
1607
1608 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1609 bprm->per_clear |= PER_CLEAR_ON_SETID;
1610 bprm->cred->egid = gid;
1611 }
1612}
1613
Oleg Nesterov9e00cdb2014-01-23 15:55:50 -08001614/*
Eric W. Biederman56305aa2020-05-29 22:00:54 -05001615 * Compute brpm->cred based upon the final binary.
1616 */
1617static int bprm_creds_from_file(struct linux_binprm *bprm)
1618{
1619 /* Compute creds based on which file? */
1620 struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1621
1622 bprm_fill_uid(bprm, file);
1623 return security_bprm_creds_from_file(bprm, file);
1624}
1625
1626/*
Oleg Nesterov9e00cdb2014-01-23 15:55:50 -08001627 * Fill the binprm structure from the inode.
Eric W. Biederman56305aa2020-05-29 22:00:54 -05001628 * Read the first BINPRM_BUF_SIZE bytes
David Howellsa6f76f22008-11-14 10:39:24 +11001629 *
1630 * This may be called multiple times for binary chains (scripts for example).
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 */
Eric W. Biederman8b72ca92020-05-13 22:25:20 -05001632static int prepare_binprm(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633{
Christoph Hellwigbdd1d2d2017-09-01 17:39:13 +02001634 loff_t pos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635
David Howellsa6f76f22008-11-14 10:39:24 +11001636 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
Christoph Hellwigbdd1d2d2017-09-01 17:39:13 +02001637 return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638}
1639
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001640/*
1641 * Arguments are '\0' separated strings found at the location bprm->p
1642 * points to; chop off the first by relocating brpm->p to right after
1643 * the first '\0' encountered.
1644 */
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001645int remove_arg_zero(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646{
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001647 int ret = 0;
1648 unsigned long offset;
1649 char *kaddr;
1650 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001652 if (!bprm->argc)
1653 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001655 do {
1656 offset = bprm->p & ~PAGE_MASK;
1657 page = get_arg_page(bprm, bprm->p, 0);
1658 if (!page) {
1659 ret = -EFAULT;
1660 goto out;
1661 }
Cong Wange8e3c3d2011-11-25 23:14:27 +08001662 kaddr = kmap_atomic(page);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001663
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001664 for (; offset < PAGE_SIZE && kaddr[offset];
1665 offset++, bprm->p++)
1666 ;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001667
Cong Wange8e3c3d2011-11-25 23:14:27 +08001668 kunmap_atomic(kaddr);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001669 put_arg_page(page);
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001670 } while (offset == PAGE_SIZE);
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001671
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001672 bprm->p++;
1673 bprm->argc--;
1674 ret = 0;
Nick Piggin4fc75ff2007-05-08 00:25:16 -07001675
Ollie Wildb6a2fea2007-07-19 01:48:16 -07001676out:
1677 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679EXPORT_SYMBOL(remove_arg_zero);
1680
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001681#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682/*
1683 * cycle the list of binary formats handler, until one recognizes the image
1684 */
Eric W. Biedermanbc2bf332020-05-18 18:43:20 -05001685static int search_binary_handler(struct linux_binprm *bprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686{
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001687 bool need_retry = IS_ENABLED(CONFIG_MODULES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 struct linux_binfmt *fmt;
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001689 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690
Eric W. Biederman8b72ca92020-05-13 22:25:20 -05001691 retval = prepare_binprm(bprm);
1692 if (retval < 0)
1693 return retval;
Kees Cookd7402692012-12-17 16:03:20 -08001694
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 retval = security_bprm_check(bprm);
1696 if (retval)
1697 return retval;
1698
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 retval = -ENOENT;
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001700 retry:
1701 read_lock(&binfmt_lock);
1702 list_for_each_entry(fmt, &formats, lh) {
1703 if (!try_module_get(fmt->module))
1704 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 read_unlock(&binfmt_lock);
Alexey Dobriyand53ddd02019-05-14 15:44:37 -07001706
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001707 retval = fmt->load_binary(bprm);
Alexey Dobriyand53ddd02019-05-14 15:44:37 -07001708
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001709 read_lock(&binfmt_lock);
1710 put_binfmt(fmt);
Eric W. Biedermanbc2bf332020-05-18 18:43:20 -05001711 if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
Al Viro19d860a2014-05-04 20:11:36 -04001712 read_unlock(&binfmt_lock);
1713 return retval;
1714 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 }
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001716 read_unlock(&binfmt_lock);
1717
Al Viro19d860a2014-05-04 20:11:36 -04001718 if (need_retry) {
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001719 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1720 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1721 return retval;
Oleg Nesterov4e0621a2013-09-11 14:24:45 -07001722 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1723 return retval;
Oleg Nesterovcb7b6b12013-09-11 14:24:44 -07001724 need_retry = false;
1725 goto retry;
1726 }
1727
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 return retval;
1729}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730
Oleg Nesterov5d1baf32013-09-11 14:24:38 -07001731static int exec_binprm(struct linux_binprm *bprm)
1732{
1733 pid_t old_pid, old_vpid;
Eric W. Biedermanbc2bf332020-05-18 18:43:20 -05001734 int ret, depth;
Oleg Nesterov5d1baf32013-09-11 14:24:38 -07001735
1736 /* Need to fetch pid before load_binary changes it */
1737 old_pid = current->pid;
1738 rcu_read_lock();
1739 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1740 rcu_read_unlock();
1741
Eric W. Biedermanbc2bf332020-05-18 18:43:20 -05001742 /* This allows 4 levels of binfmt rewrites before failing hard. */
1743 for (depth = 0;; depth++) {
1744 struct file *exec;
1745 if (depth > 5)
1746 return -ELOOP;
1747
1748 ret = search_binary_handler(bprm);
1749 if (ret < 0)
1750 return ret;
1751 if (!bprm->interpreter)
1752 break;
1753
1754 exec = bprm->file;
1755 bprm->file = bprm->interpreter;
1756 bprm->interpreter = NULL;
1757
1758 allow_write_access(exec);
1759 if (unlikely(bprm->have_execfd)) {
1760 if (bprm->executable) {
1761 fput(exec);
1762 return -ENOEXEC;
1763 }
1764 bprm->executable = exec;
1765 } else
1766 fput(exec);
Oleg Nesterov5d1baf32013-09-11 14:24:38 -07001767 }
1768
Eric W. Biedermanbc2bf332020-05-18 18:43:20 -05001769 audit_bprm(bprm);
1770 trace_sched_process_exec(current, old_pid, bprm);
1771 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1772 proc_exec_connector(current);
1773 return 0;
Oleg Nesterov5d1baf32013-09-11 14:24:38 -07001774}
1775
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776/*
1777 * sys_execve() executes a new program.
1778 */
Eric W. Biederman0c9cdff2020-07-12 07:17:50 -05001779static int bprm_execve(struct linux_binprm *bprm,
1780 int fd, struct filename *filename, int flags)
1781{
1782 struct file *file;
1783 struct files_struct *displaced;
1784 int retval;
1785
Jens Axboe0f212202020-09-13 13:09:39 -06001786 /*
1787 * Cancel any io_uring activity across execve
1788 */
1789 io_uring_task_cancel();
1790
Eric W. Biederman0c9cdff2020-07-12 07:17:50 -05001791 retval = unshare_files(&displaced);
1792 if (retval)
1793 return retval;
1794
1795 retval = prepare_bprm_creds(bprm);
1796 if (retval)
1797 goto out_files;
1798
1799 check_unsafe_exec(bprm);
1800 current->in_execve = 1;
1801
1802 file = do_open_execat(fd, filename, flags);
1803 retval = PTR_ERR(file);
1804 if (IS_ERR(file))
1805 goto out_unmark;
1806
1807 sched_exec();
1808
1809 bprm->file = file;
1810 /*
1811 * Record that a name derived from an O_CLOEXEC fd will be
1812 * inaccessible after exec. Relies on having exclusive access to
1813 * current->files (due to unshare_files above).
1814 */
1815 if (bprm->fdpath &&
1816 close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1817 bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1818
1819 /* Set the unchanging part of bprm->cred */
1820 retval = security_bprm_creds_for_exec(bprm);
1821 if (retval)
1822 goto out;
1823
1824 retval = exec_binprm(bprm);
1825 if (retval < 0)
1826 goto out;
1827
1828 /* execve succeeded */
1829 current->fs->in_exec = 0;
1830 current->in_execve = 0;
1831 rseq_execve(current);
1832 acct_update_integrals(current);
1833 task_numa_free(current, false);
1834 if (displaced)
1835 put_files_struct(displaced);
1836 return retval;
1837
1838out:
1839 /*
1840 * If past the point of no return ensure the the code never
1841 * returns to the userspace process. Use an existing fatal
1842 * signal if present otherwise terminate the process with
1843 * SIGSEGV.
1844 */
1845 if (bprm->point_of_no_return && !fatal_signal_pending(current))
1846 force_sigsegv(SIGSEGV);
1847
1848out_unmark:
1849 current->fs->in_exec = 0;
1850 current->in_execve = 0;
1851
1852out_files:
1853 if (displaced)
1854 reset_files_struct(displaced);
1855
1856 return retval;
1857}
1858
Eric W. Biederman25cf3362020-06-25 13:56:40 -05001859static int do_execveat_common(int fd, struct filename *filename,
1860 struct user_arg_ptr argv,
1861 struct user_arg_ptr envp,
1862 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863{
1864 struct linux_binprm *bprm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 int retval;
Vasiliy Kulikov72fa5992011-08-08 19:02:04 +04001866
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -08001867 if (IS_ERR(filename))
1868 return PTR_ERR(filename);
1869
Vasiliy Kulikov72fa5992011-08-08 19:02:04 +04001870 /*
1871 * We move the actual failure in case of RLIMIT_NPROC excess from
1872 * set*uid() to execve() because too many poorly written programs
1873 * don't check setuid() return code. Here we additionally recheck
1874 * whether NPROC limit is still exceeded.
1875 */
1876 if ((current->flags & PF_NPROC_EXCEEDED) &&
Oleg Nesterovbd9d43f2013-07-03 15:08:34 -07001877 atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
Vasiliy Kulikov72fa5992011-08-08 19:02:04 +04001878 retval = -EAGAIN;
1879 goto out_ret;
1880 }
1881
1882 /* We're below the limit (still or again), so we don't want to make
1883 * further execve() calls fail. */
1884 current->flags &= ~PF_NPROC_EXCEEDED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885
Eric W. Biederman60d9ad12020-07-11 08:16:15 -05001886 bprm = alloc_bprm(fd, filename);
Eric W. Biederman0a8f36e2020-07-10 15:39:45 -05001887 if (IS_ERR(bprm)) {
1888 retval = PTR_ERR(bprm);
1889 goto out_ret;
1890 }
1891
Eric W. Biedermand8b9cd52020-07-12 08:23:54 -05001892 retval = count(argv, MAX_ARG_STRINGS);
1893 if (retval < 0)
1894 goto out_free;
1895 bprm->argc = retval;
1896
1897 retval = count(envp, MAX_ARG_STRINGS);
1898 if (retval < 0)
1899 goto out_free;
1900 bprm->envc = retval;
1901
1902 retval = bprm_stack_limits(bprm);
Oleg Nesterov655c16a2019-01-03 15:28:11 -08001903 if (retval < 0)
Eric W. Biederman0c9cdff2020-07-12 07:17:50 -05001904 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905
Christoph Hellwig986db2d2020-06-04 16:51:14 -07001906 retval = copy_string_kernel(bprm->filename, bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 if (retval < 0)
Eric W. Biederman0c9cdff2020-07-12 07:17:50 -05001908 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 bprm->exec = bprm->p;
Eric W. Biederman0c9cdff2020-07-12 07:17:50 -05001910
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 retval = copy_strings(bprm->envc, envp, bprm);
1912 if (retval < 0)
Eric W. Biederman0c9cdff2020-07-12 07:17:50 -05001913 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
1915 retval = copy_strings(bprm->argc, argv, bprm);
1916 if (retval < 0)
Eric W. Biederman0c9cdff2020-07-12 07:17:50 -05001917 goto out_free;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918
Eric W. Biederman0c9cdff2020-07-12 07:17:50 -05001919 retval = bprm_execve(bprm, fd, filename, flags);
David Howellsa6f76f22008-11-14 10:39:24 +11001920out_free:
Al Viro08a6fac2008-05-10 16:38:25 -04001921 free_bprm(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922
1923out_ret:
Eric W. Biederman25cf3362020-06-25 13:56:40 -05001924 putname(filename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 return retval;
1926}
1927
Eric W. Biedermanbe619f72020-07-13 12:06:48 -05001928int kernel_execve(const char *kernel_filename,
1929 const char *const *argv, const char *const *envp)
1930{
1931 struct filename *filename;
1932 struct linux_binprm *bprm;
1933 int fd = AT_FDCWD;
1934 int retval;
1935
1936 filename = getname_kernel(kernel_filename);
1937 if (IS_ERR(filename))
1938 return PTR_ERR(filename);
1939
1940 bprm = alloc_bprm(fd, filename);
1941 if (IS_ERR(bprm)) {
1942 retval = PTR_ERR(bprm);
1943 goto out_ret;
1944 }
1945
1946 retval = count_strings_kernel(argv);
1947 if (retval < 0)
1948 goto out_free;
1949 bprm->argc = retval;
1950
1951 retval = count_strings_kernel(envp);
1952 if (retval < 0)
1953 goto out_free;
1954 bprm->envc = retval;
1955
1956 retval = bprm_stack_limits(bprm);
1957 if (retval < 0)
1958 goto out_free;
1959
1960 retval = copy_string_kernel(bprm->filename, bprm);
1961 if (retval < 0)
1962 goto out_free;
1963 bprm->exec = bprm->p;
1964
1965 retval = copy_strings_kernel(bprm->envc, envp, bprm);
1966 if (retval < 0)
1967 goto out_free;
1968
1969 retval = copy_strings_kernel(bprm->argc, argv, bprm);
1970 if (retval < 0)
1971 goto out_free;
1972
1973 retval = bprm_execve(bprm, fd, filename, 0);
1974out_free:
1975 free_bprm(bprm);
1976out_ret:
1977 putname(filename);
1978 return retval;
1979}
1980
1981static int do_execve(struct filename *filename,
Oleg Nesterovba2d01622011-03-06 18:02:37 +01001982 const char __user *const __user *__argv,
Al Viroda3d4c52012-10-20 21:49:33 -04001983 const char __user *const __user *__envp)
Oleg Nesterovba2d01622011-03-06 18:02:37 +01001984{
Oleg Nesterov0e028462011-03-06 18:02:54 +01001985 struct user_arg_ptr argv = { .ptr.native = __argv };
1986 struct user_arg_ptr envp = { .ptr.native = __envp };
David Drysdale51f39a12014-12-12 16:57:29 -08001987 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1988}
1989
Eric W. Biedermanbe619f72020-07-13 12:06:48 -05001990static int do_execveat(int fd, struct filename *filename,
David Drysdale51f39a12014-12-12 16:57:29 -08001991 const char __user *const __user *__argv,
1992 const char __user *const __user *__envp,
1993 int flags)
1994{
1995 struct user_arg_ptr argv = { .ptr.native = __argv };
1996 struct user_arg_ptr envp = { .ptr.native = __envp };
1997
1998 return do_execveat_common(fd, filename, argv, envp, flags);
Oleg Nesterovba2d01622011-03-06 18:02:37 +01001999}
2000
Oleg Nesterov0e028462011-03-06 18:02:54 +01002001#ifdef CONFIG_COMPAT
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -08002002static int compat_do_execve(struct filename *filename,
Al Viro38b983b2012-09-30 13:38:55 -04002003 const compat_uptr_t __user *__argv,
Al Virod03d26e2012-10-20 21:46:25 -04002004 const compat_uptr_t __user *__envp)
Oleg Nesterov0e028462011-03-06 18:02:54 +01002005{
2006 struct user_arg_ptr argv = {
2007 .is_compat = true,
2008 .ptr.compat = __argv,
2009 };
2010 struct user_arg_ptr envp = {
2011 .is_compat = true,
2012 .ptr.compat = __envp,
2013 };
David Drysdale51f39a12014-12-12 16:57:29 -08002014 return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2015}
2016
2017static int compat_do_execveat(int fd, struct filename *filename,
2018 const compat_uptr_t __user *__argv,
2019 const compat_uptr_t __user *__envp,
2020 int flags)
2021{
2022 struct user_arg_ptr argv = {
2023 .is_compat = true,
2024 .ptr.compat = __argv,
2025 };
2026 struct user_arg_ptr envp = {
2027 .is_compat = true,
2028 .ptr.compat = __envp,
2029 };
2030 return do_execveat_common(fd, filename, argv, envp, flags);
Oleg Nesterov0e028462011-03-06 18:02:54 +01002031}
2032#endif
2033
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07002034void set_binfmt(struct linux_binfmt *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035{
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07002036 struct mm_struct *mm = current->mm;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
Hiroshi Shimamoto801460d2009-09-23 15:57:41 -07002038 if (mm->binfmt)
2039 module_put(mm->binfmt->module);
2040
2041 mm->binfmt = new;
Oleg Nesterov964ee7d2009-09-23 15:56:59 -07002042 if (new)
2043 __module_get(new->module);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045EXPORT_SYMBOL(set_binfmt);
2046
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07002047/*
Oleg Nesterov7288e112014-01-23 15:55:32 -08002048 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07002049 */
2050void set_dumpable(struct mm_struct *mm, int value)
2051{
Oleg Nesterov7288e112014-01-23 15:55:32 -08002052 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2053 return;
2054
Vineet Gupta26e15222019-03-07 16:29:23 -08002055 set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07002056}
Kawai, Hidehiro6c5d5232007-07-19 01:48:27 -07002057
Al Viro38b983b2012-09-30 13:38:55 -04002058SYSCALL_DEFINE3(execve,
2059 const char __user *, filename,
2060 const char __user *const __user *, argv,
2061 const char __user *const __user *, envp)
2062{
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -08002063 return do_execve(getname(filename), argv, envp);
Al Viro38b983b2012-09-30 13:38:55 -04002064}
David Drysdale51f39a12014-12-12 16:57:29 -08002065
2066SYSCALL_DEFINE5(execveat,
2067 int, fd, const char __user *, filename,
2068 const char __user *const __user *, argv,
2069 const char __user *const __user *, envp,
2070 int, flags)
2071{
2072 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2073
2074 return do_execveat(fd,
2075 getname_flags(filename, lookup_flags, NULL),
2076 argv, envp, flags);
2077}
2078
Al Viro38b983b2012-09-30 13:38:55 -04002079#ifdef CONFIG_COMPAT
Heiko Carstens625b1d72014-03-04 10:53:50 +01002080COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2081 const compat_uptr_t __user *, argv,
2082 const compat_uptr_t __user *, envp)
Al Viro38b983b2012-09-30 13:38:55 -04002083{
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -08002084 return compat_do_execve(getname(filename), argv, envp);
Al Viro38b983b2012-09-30 13:38:55 -04002085}
David Drysdale51f39a12014-12-12 16:57:29 -08002086
2087COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2088 const char __user *, filename,
2089 const compat_uptr_t __user *, argv,
2090 const compat_uptr_t __user *, envp,
2091 int, flags)
2092{
2093 int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
2094
2095 return compat_do_execveat(fd,
2096 getname_flags(filename, lookup_flags, NULL),
2097 argv, envp, flags);
2098}
Al Viro38b983b2012-09-30 13:38:55 -04002099#endif