]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/exec.c
Merge branch 'timers-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[mirror_ubuntu-bionic-kernel.git] / fs / exec.c
CommitLineData
1da177e4
LT
1/*
2 * linux/fs/exec.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * #!-checking implemented by tytso.
9 */
10/*
11 * Demand-loading implemented 01.12.91 - no need to read anything but
12 * the header into memory. The inode of the executable is put into
13 * "current->executable", and page faults do the actual loading. Clean.
14 *
15 * Once more I can proudly say that linux stood up to being changed: it
16 * was less than 2 hours work to get demand-loading completely implemented.
17 *
18 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
19 * current->executable is only used by the procfs. This allows a dispatch
20 * table to check for several different types of binary formats. We keep
21 * trying until we recognize the file or we run out of supported binary
22 * formats.
23 */
24
1da177e4
LT
25#include <linux/slab.h>
26#include <linux/file.h>
9f3acc31 27#include <linux/fdtable.h>
ba92a43d 28#include <linux/mm.h>
615d6e87 29#include <linux/vmacache.h>
1da177e4
LT
30#include <linux/stat.h>
31#include <linux/fcntl.h>
ba92a43d 32#include <linux/swap.h>
74aadce9 33#include <linux/string.h>
1da177e4 34#include <linux/init.h>
ca5b172b 35#include <linux/pagemap.h>
cdd6c482 36#include <linux/perf_event.h>
1da177e4
LT
37#include <linux/highmem.h>
38#include <linux/spinlock.h>
39#include <linux/key.h>
40#include <linux/personality.h>
41#include <linux/binfmts.h>
1da177e4 42#include <linux/utsname.h>
84d73786 43#include <linux/pid_namespace.h>
1da177e4
LT
44#include <linux/module.h>
45#include <linux/namei.h>
1da177e4
LT
46#include <linux/mount.h>
47#include <linux/security.h>
48#include <linux/syscalls.h>
8f0ab514 49#include <linux/tsacct_kern.h>
9f46080c 50#include <linux/cn_proc.h>
473ae30b 51#include <linux/audit.h>
6341c393 52#include <linux/tracehook.h>
5f4123be 53#include <linux/kmod.h>
6110e3ab 54#include <linux/fsnotify.h>
5ad4e53b 55#include <linux/fs_struct.h>
61be228a 56#include <linux/pipe_fs_i.h>
3d5992d2 57#include <linux/oom.h>
0e028465 58#include <linux/compat.h>
1da177e4
LT
59
60#include <asm/uaccess.h>
61#include <asm/mmu_context.h>
b6a2fea3 62#include <asm/tlb.h>
43d2b113
KH
63
64#include <trace/events/task.h>
a6f76f23 65#include "internal.h"
1da177e4 66
4ff16c25
DS
67#include <trace/events/sched.h>
68
d6e71144
AC
69int suid_dumpable = 0;
70
e4dc1b14 71static LIST_HEAD(formats);
1da177e4
LT
72static DEFINE_RWLOCK(binfmt_lock);
73
8fc3dc5a 74void __register_binfmt(struct linux_binfmt * fmt, int insert)
1da177e4 75{
8fc3dc5a 76 BUG_ON(!fmt);
92eaa565
ON
77 if (WARN_ON(!fmt->load_binary))
78 return;
1da177e4 79 write_lock(&binfmt_lock);
74641f58
IK
80 insert ? list_add(&fmt->lh, &formats) :
81 list_add_tail(&fmt->lh, &formats);
1da177e4 82 write_unlock(&binfmt_lock);
1da177e4
LT
83}
84
74641f58 85EXPORT_SYMBOL(__register_binfmt);
1da177e4 86
f6b450d4 87void unregister_binfmt(struct linux_binfmt * fmt)
1da177e4 88{
1da177e4 89 write_lock(&binfmt_lock);
e4dc1b14 90 list_del(&fmt->lh);
1da177e4 91 write_unlock(&binfmt_lock);
1da177e4
LT
92}
93
94EXPORT_SYMBOL(unregister_binfmt);
95
96static inline void put_binfmt(struct linux_binfmt * fmt)
97{
98 module_put(fmt->module);
99}
100
69369a70 101#ifdef CONFIG_USELIB
1da177e4
LT
102/*
103 * Note that a shared library must be both readable and executable due to
104 * security reasons.
105 *
106 * Also note that we take the address to load from from the file itself.
107 */
1e7bfb21 108SYSCALL_DEFINE1(uselib, const char __user *, library)
1da177e4 109{
72c2d531 110 struct linux_binfmt *fmt;
964bd183 111 struct file *file;
91a27b2a 112 struct filename *tmp = getname(library);
964bd183 113 int error = PTR_ERR(tmp);
47c805dc
AV
114 static const struct open_flags uselib_flags = {
115 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
116 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
f9652e10
AV
117 .intent = LOOKUP_OPEN,
118 .lookup_flags = LOOKUP_FOLLOW,
47c805dc 119 };
964bd183 120
6e8341a1
AV
121 if (IS_ERR(tmp))
122 goto out;
123
f9652e10 124 file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
6e8341a1
AV
125 putname(tmp);
126 error = PTR_ERR(file);
127 if (IS_ERR(file))
1da177e4
LT
128 goto out;
129
130 error = -EINVAL;
496ad9aa 131 if (!S_ISREG(file_inode(file)->i_mode))
1da177e4
LT
132 goto exit;
133
30524472 134 error = -EACCES;
6e8341a1 135 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1da177e4
LT
136 goto exit;
137
2a12a9d7 138 fsnotify_open(file);
6110e3ab 139
1da177e4 140 error = -ENOEXEC;
1da177e4 141
72c2d531
AV
142 read_lock(&binfmt_lock);
143 list_for_each_entry(fmt, &formats, lh) {
144 if (!fmt->load_shlib)
145 continue;
146 if (!try_module_get(fmt->module))
147 continue;
1da177e4 148 read_unlock(&binfmt_lock);
72c2d531
AV
149 error = fmt->load_shlib(file);
150 read_lock(&binfmt_lock);
151 put_binfmt(fmt);
152 if (error != -ENOEXEC)
153 break;
1da177e4 154 }
72c2d531 155 read_unlock(&binfmt_lock);
6e8341a1 156exit:
1da177e4
LT
157 fput(file);
158out:
159 return error;
1da177e4 160}
69369a70 161#endif /* #ifdef CONFIG_USELIB */
1da177e4 162
b6a2fea3 163#ifdef CONFIG_MMU
ae6b585e
ON
164/*
165 * The nascent bprm->mm is not visible until exec_mmap() but it can
166 * use a lot of memory, account these pages in current->mm temporary
167 * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
168 * change the counter back via acct_arg_size(0).
169 */
0e028465 170static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
3c77f845
ON
171{
172 struct mm_struct *mm = current->mm;
173 long diff = (long)(pages - bprm->vma_pages);
174
175 if (!mm || !diff)
176 return;
177
178 bprm->vma_pages = pages;
3c77f845 179 add_mm_counter(mm, MM_ANONPAGES, diff);
3c77f845
ON
180}
181
0e028465 182static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
b6a2fea3
OW
183 int write)
184{
185 struct page *page;
186 int ret;
187
188#ifdef CONFIG_STACK_GROWSUP
189 if (write) {
d05f3169 190 ret = expand_downwards(bprm->vma, pos);
b6a2fea3
OW
191 if (ret < 0)
192 return NULL;
193 }
194#endif
195 ret = get_user_pages(current, bprm->mm, pos,
196 1, write, 1, &page, NULL);
197 if (ret <= 0)
198 return NULL;
199
200 if (write) {
b6a2fea3 201 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
a64e715f
LT
202 struct rlimit *rlim;
203
3c77f845
ON
204 acct_arg_size(bprm, size / PAGE_SIZE);
205
a64e715f
LT
206 /*
207 * We've historically supported up to 32 pages (ARG_MAX)
208 * of argument strings even with small stacks
209 */
210 if (size <= ARG_MAX)
211 return page;
b6a2fea3
OW
212
213 /*
214 * Limit to 1/4-th the stack size for the argv+env strings.
215 * This ensures that:
216 * - the remaining binfmt code will not run out of stack space,
217 * - the program will have a reasonable amount of stack left
218 * to work from.
219 */
a64e715f 220 rlim = current->signal->rlim;
d554ed89 221 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
b6a2fea3
OW
222 put_page(page);
223 return NULL;
224 }
225 }
226
227 return page;
228}
229
230static void put_arg_page(struct page *page)
231{
232 put_page(page);
233}
234
235static void free_arg_page(struct linux_binprm *bprm, int i)
236{
237}
238
239static void free_arg_pages(struct linux_binprm *bprm)
240{
241}
242
243static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
244 struct page *page)
245{
246 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
247}
248
249static int __bprm_mm_init(struct linux_binprm *bprm)
250{
eaccbfa5 251 int err;
b6a2fea3
OW
252 struct vm_area_struct *vma = NULL;
253 struct mm_struct *mm = bprm->mm;
254
255 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
256 if (!vma)
eaccbfa5 257 return -ENOMEM;
b6a2fea3
OW
258
259 down_write(&mm->mmap_sem);
260 vma->vm_mm = mm;
261
262 /*
263 * Place the stack at the largest stack address the architecture
264 * supports. Later, we'll move this to an appropriate place. We don't
265 * use STACK_TOP because that can depend on attributes which aren't
266 * configured yet.
267 */
aacb3d17 268 BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
b6a2fea3
OW
269 vma->vm_end = STACK_TOP_MAX;
270 vma->vm_start = vma->vm_end - PAGE_SIZE;
d9104d1c 271 vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
3ed75eb8 272 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
5beb4930 273 INIT_LIST_HEAD(&vma->anon_vma_chain);
462e635e 274
b6a2fea3 275 err = insert_vm_struct(mm, vma);
eaccbfa5 276 if (err)
b6a2fea3 277 goto err;
b6a2fea3
OW
278
279 mm->stack_vm = mm->total_vm = 1;
280 up_write(&mm->mmap_sem);
b6a2fea3 281 bprm->p = vma->vm_end - sizeof(void *);
b6a2fea3 282 return 0;
b6a2fea3 283err:
eaccbfa5
LFC
284 up_write(&mm->mmap_sem);
285 bprm->vma = NULL;
286 kmem_cache_free(vm_area_cachep, vma);
b6a2fea3
OW
287 return err;
288}
289
290static bool valid_arg_len(struct linux_binprm *bprm, long len)
291{
292 return len <= MAX_ARG_STRLEN;
293}
294
295#else
296
0e028465 297static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
3c77f845
ON
298{
299}
300
0e028465 301static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
b6a2fea3
OW
302 int write)
303{
304 struct page *page;
305
306 page = bprm->page[pos / PAGE_SIZE];
307 if (!page && write) {
308 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
309 if (!page)
310 return NULL;
311 bprm->page[pos / PAGE_SIZE] = page;
312 }
313
314 return page;
315}
316
317static void put_arg_page(struct page *page)
318{
319}
320
321static void free_arg_page(struct linux_binprm *bprm, int i)
322{
323 if (bprm->page[i]) {
324 __free_page(bprm->page[i]);
325 bprm->page[i] = NULL;
326 }
327}
328
329static void free_arg_pages(struct linux_binprm *bprm)
330{
331 int i;
332
333 for (i = 0; i < MAX_ARG_PAGES; i++)
334 free_arg_page(bprm, i);
335}
336
337static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
338 struct page *page)
339{
340}
341
342static int __bprm_mm_init(struct linux_binprm *bprm)
343{
344 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
345 return 0;
346}
347
348static bool valid_arg_len(struct linux_binprm *bprm, long len)
349{
350 return len <= bprm->p;
351}
352
353#endif /* CONFIG_MMU */
354
355/*
356 * Create a new mm_struct and populate it with a temporary stack
357 * vm_area_struct. We don't have enough context at this point to set the stack
358 * flags, permissions, and offset, so we use temporary values. We'll update
359 * them later in setup_arg_pages().
360 */
9cc64cea 361static int bprm_mm_init(struct linux_binprm *bprm)
b6a2fea3
OW
362{
363 int err;
364 struct mm_struct *mm = NULL;
365
366 bprm->mm = mm = mm_alloc();
367 err = -ENOMEM;
368 if (!mm)
369 goto err;
370
371 err = init_new_context(current, mm);
372 if (err)
373 goto err;
374
375 err = __bprm_mm_init(bprm);
376 if (err)
377 goto err;
378
379 return 0;
380
381err:
382 if (mm) {
383 bprm->mm = NULL;
384 mmdrop(mm);
385 }
386
387 return err;
388}
389
ba2d0162 390struct user_arg_ptr {
0e028465
ON
391#ifdef CONFIG_COMPAT
392 bool is_compat;
393#endif
394 union {
395 const char __user *const __user *native;
396#ifdef CONFIG_COMPAT
38b983b3 397 const compat_uptr_t __user *compat;
0e028465
ON
398#endif
399 } ptr;
ba2d0162
ON
400};
401
402static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
1d1dbf81 403{
0e028465
ON
404 const char __user *native;
405
406#ifdef CONFIG_COMPAT
407 if (unlikely(argv.is_compat)) {
408 compat_uptr_t compat;
409
410 if (get_user(compat, argv.ptr.compat + nr))
411 return ERR_PTR(-EFAULT);
1d1dbf81 412
0e028465
ON
413 return compat_ptr(compat);
414 }
415#endif
416
417 if (get_user(native, argv.ptr.native + nr))
1d1dbf81
ON
418 return ERR_PTR(-EFAULT);
419
0e028465 420 return native;
1d1dbf81
ON
421}
422
1da177e4
LT
423/*
424 * count() counts the number of strings in array ARGV.
425 */
ba2d0162 426static int count(struct user_arg_ptr argv, int max)
1da177e4
LT
427{
428 int i = 0;
429
0e028465 430 if (argv.ptr.native != NULL) {
1da177e4 431 for (;;) {
1d1dbf81 432 const char __user *p = get_user_arg_ptr(argv, i);
1da177e4 433
1da177e4
LT
434 if (!p)
435 break;
1d1dbf81
ON
436
437 if (IS_ERR(p))
438 return -EFAULT;
439
6d92d4f6 440 if (i >= max)
1da177e4 441 return -E2BIG;
6d92d4f6 442 ++i;
9aea5a65
RM
443
444 if (fatal_signal_pending(current))
445 return -ERESTARTNOHAND;
1da177e4
LT
446 cond_resched();
447 }
448 }
449 return i;
450}
451
452/*
b6a2fea3
OW
453 * 'copy_strings()' copies argument/environment strings from the old
454 * processes's memory to the new process's stack. The call to get_user_pages()
455 * ensures the destination page is created and not swapped out.
1da177e4 456 */
ba2d0162 457static int copy_strings(int argc, struct user_arg_ptr argv,
75c96f85 458 struct linux_binprm *bprm)
1da177e4
LT
459{
460 struct page *kmapped_page = NULL;
461 char *kaddr = NULL;
b6a2fea3 462 unsigned long kpos = 0;
1da177e4
LT
463 int ret;
464
465 while (argc-- > 0) {
d7627467 466 const char __user *str;
1da177e4
LT
467 int len;
468 unsigned long pos;
469
1d1dbf81
ON
470 ret = -EFAULT;
471 str = get_user_arg_ptr(argv, argc);
472 if (IS_ERR(str))
1da177e4 473 goto out;
1da177e4 474
1d1dbf81
ON
475 len = strnlen_user(str, MAX_ARG_STRLEN);
476 if (!len)
477 goto out;
478
479 ret = -E2BIG;
480 if (!valid_arg_len(bprm, len))
1da177e4 481 goto out;
1da177e4 482
b6a2fea3 483 /* We're going to work our way backwords. */
1da177e4 484 pos = bprm->p;
b6a2fea3
OW
485 str += len;
486 bprm->p -= len;
1da177e4
LT
487
488 while (len > 0) {
1da177e4 489 int offset, bytes_to_copy;
1da177e4 490
9aea5a65
RM
491 if (fatal_signal_pending(current)) {
492 ret = -ERESTARTNOHAND;
493 goto out;
494 }
7993bc1f
RM
495 cond_resched();
496
1da177e4 497 offset = pos % PAGE_SIZE;
b6a2fea3
OW
498 if (offset == 0)
499 offset = PAGE_SIZE;
500
501 bytes_to_copy = offset;
502 if (bytes_to_copy > len)
503 bytes_to_copy = len;
504
505 offset -= bytes_to_copy;
506 pos -= bytes_to_copy;
507 str -= bytes_to_copy;
508 len -= bytes_to_copy;
509
510 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
511 struct page *page;
512
513 page = get_arg_page(bprm, pos, 1);
1da177e4 514 if (!page) {
b6a2fea3 515 ret = -E2BIG;
1da177e4
LT
516 goto out;
517 }
1da177e4 518
b6a2fea3
OW
519 if (kmapped_page) {
520 flush_kernel_dcache_page(kmapped_page);
1da177e4 521 kunmap(kmapped_page);
b6a2fea3
OW
522 put_arg_page(kmapped_page);
523 }
1da177e4
LT
524 kmapped_page = page;
525 kaddr = kmap(kmapped_page);
b6a2fea3
OW
526 kpos = pos & PAGE_MASK;
527 flush_arg_page(bprm, kpos, kmapped_page);
1da177e4 528 }
b6a2fea3 529 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
1da177e4
LT
530 ret = -EFAULT;
531 goto out;
532 }
1da177e4
LT
533 }
534 }
535 ret = 0;
536out:
b6a2fea3
OW
537 if (kmapped_page) {
538 flush_kernel_dcache_page(kmapped_page);
1da177e4 539 kunmap(kmapped_page);
b6a2fea3
OW
540 put_arg_page(kmapped_page);
541 }
1da177e4
LT
542 return ret;
543}
544
545/*
546 * Like copy_strings, but get argv and its values from kernel memory.
547 */
ba2d0162 548int copy_strings_kernel(int argc, const char *const *__argv,
d7627467 549 struct linux_binprm *bprm)
1da177e4
LT
550{
551 int r;
552 mm_segment_t oldfs = get_fs();
ba2d0162 553 struct user_arg_ptr argv = {
0e028465 554 .ptr.native = (const char __user *const __user *)__argv,
ba2d0162
ON
555 };
556
1da177e4 557 set_fs(KERNEL_DS);
ba2d0162 558 r = copy_strings(argc, argv, bprm);
1da177e4 559 set_fs(oldfs);
ba2d0162 560
1da177e4
LT
561 return r;
562}
1da177e4
LT
563EXPORT_SYMBOL(copy_strings_kernel);
564
565#ifdef CONFIG_MMU
b6a2fea3 566
1da177e4 567/*
b6a2fea3
OW
568 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
569 * the binfmt code determines where the new stack should reside, we shift it to
570 * its final location. The process proceeds as follows:
1da177e4 571 *
b6a2fea3
OW
572 * 1) Use shift to calculate the new vma endpoints.
573 * 2) Extend vma to cover both the old and new ranges. This ensures the
574 * arguments passed to subsequent functions are consistent.
575 * 3) Move vma's page tables to the new range.
576 * 4) Free up any cleared pgd range.
577 * 5) Shrink the vma to cover only the new range.
1da177e4 578 */
b6a2fea3 579static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
1da177e4
LT
580{
581 struct mm_struct *mm = vma->vm_mm;
b6a2fea3
OW
582 unsigned long old_start = vma->vm_start;
583 unsigned long old_end = vma->vm_end;
584 unsigned long length = old_end - old_start;
585 unsigned long new_start = old_start - shift;
586 unsigned long new_end = old_end - shift;
d16dfc55 587 struct mmu_gather tlb;
1da177e4 588
b6a2fea3 589 BUG_ON(new_start > new_end);
1da177e4 590
b6a2fea3
OW
591 /*
592 * ensure there are no vmas between where we want to go
593 * and where we are
594 */
595 if (vma != find_vma(mm, new_start))
596 return -EFAULT;
597
598 /*
599 * cover the whole range: [new_start, old_end)
600 */
5beb4930
RR
601 if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
602 return -ENOMEM;
b6a2fea3
OW
603
604 /*
605 * move the page tables downwards, on failure we rely on
606 * process cleanup to remove whatever mess we made.
607 */
608 if (length != move_page_tables(vma, old_start,
38a76013 609 vma, new_start, length, false))
b6a2fea3
OW
610 return -ENOMEM;
611
612 lru_add_drain();
2b047252 613 tlb_gather_mmu(&tlb, mm, old_start, old_end);
b6a2fea3
OW
614 if (new_end > old_start) {
615 /*
616 * when the old and new regions overlap clear from new_end.
617 */
d16dfc55 618 free_pgd_range(&tlb, new_end, old_end, new_end,
6ee8630e 619 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
b6a2fea3
OW
620 } else {
621 /*
622 * otherwise, clean from old_start; this is done to not touch
623 * the address space in [new_end, old_start) some architectures
624 * have constraints on va-space that make this illegal (IA64) -
625 * for the others its just a little faster.
626 */
d16dfc55 627 free_pgd_range(&tlb, old_start, old_end, new_end,
6ee8630e 628 vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
1da177e4 629 }
2b047252 630 tlb_finish_mmu(&tlb, old_start, old_end);
b6a2fea3
OW
631
632 /*
5beb4930 633 * Shrink the vma to just the new range. Always succeeds.
b6a2fea3
OW
634 */
635 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
636
637 return 0;
1da177e4
LT
638}
639
b6a2fea3
OW
640/*
641 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
642 * the stack is optionally relocated, and some extra space is added.
643 */
1da177e4
LT
644int setup_arg_pages(struct linux_binprm *bprm,
645 unsigned long stack_top,
646 int executable_stack)
647{
b6a2fea3
OW
648 unsigned long ret;
649 unsigned long stack_shift;
1da177e4 650 struct mm_struct *mm = current->mm;
b6a2fea3
OW
651 struct vm_area_struct *vma = bprm->vma;
652 struct vm_area_struct *prev = NULL;
653 unsigned long vm_flags;
654 unsigned long stack_base;
803bf5ec
MN
655 unsigned long stack_size;
656 unsigned long stack_expand;
657 unsigned long rlim_stack;
1da177e4
LT
658
659#ifdef CONFIG_STACK_GROWSUP
d71f290b 660 /* Limit stack size */
d554ed89 661 stack_base = rlimit_max(RLIMIT_STACK);
d71f290b
JH
662 if (stack_base > STACK_SIZE_MAX)
663 stack_base = STACK_SIZE_MAX;
1da177e4 664
b6a2fea3
OW
665 /* Make sure we didn't let the argument array grow too large. */
666 if (vma->vm_end - vma->vm_start > stack_base)
667 return -ENOMEM;
1da177e4 668
b6a2fea3 669 stack_base = PAGE_ALIGN(stack_top - stack_base);
1da177e4 670
b6a2fea3
OW
671 stack_shift = vma->vm_start - stack_base;
672 mm->arg_start = bprm->p - stack_shift;
673 bprm->p = vma->vm_end - stack_shift;
1da177e4 674#else
b6a2fea3
OW
675 stack_top = arch_align_stack(stack_top);
676 stack_top = PAGE_ALIGN(stack_top);
1b528181
RM
677
678 if (unlikely(stack_top < mmap_min_addr) ||
679 unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
680 return -ENOMEM;
681
b6a2fea3
OW
682 stack_shift = vma->vm_end - stack_top;
683
684 bprm->p -= stack_shift;
1da177e4 685 mm->arg_start = bprm->p;
1da177e4
LT
686#endif
687
1da177e4 688 if (bprm->loader)
b6a2fea3
OW
689 bprm->loader -= stack_shift;
690 bprm->exec -= stack_shift;
1da177e4 691
1da177e4 692 down_write(&mm->mmap_sem);
96a8e13e 693 vm_flags = VM_STACK_FLAGS;
b6a2fea3
OW
694
695 /*
696 * Adjust stack execute permissions; explicitly enable for
697 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
698 * (arch default) otherwise.
699 */
700 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
701 vm_flags |= VM_EXEC;
702 else if (executable_stack == EXSTACK_DISABLE_X)
703 vm_flags &= ~VM_EXEC;
704 vm_flags |= mm->def_flags;
a8bef8ff 705 vm_flags |= VM_STACK_INCOMPLETE_SETUP;
b6a2fea3
OW
706
707 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
708 vm_flags);
709 if (ret)
710 goto out_unlock;
711 BUG_ON(prev != vma);
712
713 /* Move stack pages down in memory. */
714 if (stack_shift) {
715 ret = shift_arg_pages(vma, stack_shift);
fc63cf23
AB
716 if (ret)
717 goto out_unlock;
1da177e4
LT
718 }
719
a8bef8ff
MG
720 /* mprotect_fixup is overkill to remove the temporary stack flags */
721 vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
722
5ef097dd 723 stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
803bf5ec
MN
724 stack_size = vma->vm_end - vma->vm_start;
725 /*
726 * Align this down to a page boundary as expand_stack
727 * will align it up.
728 */
729 rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
b6a2fea3 730#ifdef CONFIG_STACK_GROWSUP
803bf5ec
MN
731 if (stack_size + stack_expand > rlim_stack)
732 stack_base = vma->vm_start + rlim_stack;
733 else
734 stack_base = vma->vm_end + stack_expand;
b6a2fea3 735#else
803bf5ec
MN
736 if (stack_size + stack_expand > rlim_stack)
737 stack_base = vma->vm_end - rlim_stack;
738 else
739 stack_base = vma->vm_start - stack_expand;
b6a2fea3 740#endif
3af9e859 741 current->mm->start_stack = bprm->p;
b6a2fea3
OW
742 ret = expand_stack(vma, stack_base);
743 if (ret)
744 ret = -EFAULT;
745
746out_unlock:
1da177e4 747 up_write(&mm->mmap_sem);
fc63cf23 748 return ret;
1da177e4 749}
1da177e4
LT
750EXPORT_SYMBOL(setup_arg_pages);
751
1da177e4
LT
752#endif /* CONFIG_MMU */
753
c4ad8f98 754static struct file *do_open_exec(struct filename *name)
1da177e4 755{
1da177e4 756 struct file *file;
e56b6a5d 757 int err;
47c805dc
AV
758 static const struct open_flags open_exec_flags = {
759 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
760 .acc_mode = MAY_EXEC | MAY_OPEN,
f9652e10
AV
761 .intent = LOOKUP_OPEN,
762 .lookup_flags = LOOKUP_FOLLOW,
47c805dc 763 };
1da177e4 764
c4ad8f98 765 file = do_filp_open(AT_FDCWD, name, &open_exec_flags);
6e8341a1 766 if (IS_ERR(file))
e56b6a5d
CH
767 goto out;
768
769 err = -EACCES;
496ad9aa 770 if (!S_ISREG(file_inode(file)->i_mode))
6e8341a1 771 goto exit;
e56b6a5d 772
6e8341a1
AV
773 if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
774 goto exit;
e56b6a5d 775
2a12a9d7 776 fsnotify_open(file);
6110e3ab 777
e56b6a5d 778 err = deny_write_access(file);
6e8341a1
AV
779 if (err)
780 goto exit;
1da177e4 781
6e8341a1 782out:
e56b6a5d
CH
783 return file;
784
6e8341a1
AV
785exit:
786 fput(file);
e56b6a5d
CH
787 return ERR_PTR(err);
788}
c4ad8f98
LT
789
790struct file *open_exec(const char *name)
791{
792 struct filename tmp = { .name = name };
793 return do_open_exec(&tmp);
794}
1da177e4
LT
795EXPORT_SYMBOL(open_exec);
796
6777d773
MZ
797int kernel_read(struct file *file, loff_t offset,
798 char *addr, unsigned long count)
1da177e4
LT
799{
800 mm_segment_t old_fs;
801 loff_t pos = offset;
802 int result;
803
804 old_fs = get_fs();
805 set_fs(get_ds());
806 /* The cast to a user pointer is valid due to the set_fs() */
807 result = vfs_read(file, (void __user *)addr, count, &pos);
808 set_fs(old_fs);
809 return result;
810}
811
812EXPORT_SYMBOL(kernel_read);
813
3dc20cb2
AV
814ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
815{
ec695579 816 ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
3dc20cb2
AV
817 if (res > 0)
818 flush_icache_range(addr, addr + len);
819 return res;
820}
821EXPORT_SYMBOL(read_code);
822
1da177e4
LT
823static int exec_mmap(struct mm_struct *mm)
824{
825 struct task_struct *tsk;
615d6e87 826 struct mm_struct *old_mm, *active_mm;
1da177e4
LT
827
828 /* Notify parent that we're no longer interested in the old VM */
829 tsk = current;
830 old_mm = current->mm;
831 mm_release(tsk, old_mm);
832
833 if (old_mm) {
4fe7efdb 834 sync_mm_rss(old_mm);
1da177e4
LT
835 /*
836 * Make sure that if there is a core dump in progress
837 * for the old mm, we get out and die instead of going
838 * through with the exec. We must hold mmap_sem around
999d9fc1 839 * checking core_state and changing tsk->mm.
1da177e4
LT
840 */
841 down_read(&old_mm->mmap_sem);
999d9fc1 842 if (unlikely(old_mm->core_state)) {
1da177e4
LT
843 up_read(&old_mm->mmap_sem);
844 return -EINTR;
845 }
846 }
847 task_lock(tsk);
848 active_mm = tsk->active_mm;
849 tsk->mm = mm;
850 tsk->active_mm = mm;
851 activate_mm(active_mm, mm);
615d6e87
DB
852 tsk->mm->vmacache_seqnum = 0;
853 vmacache_flush(tsk);
1da177e4 854 task_unlock(tsk);
1da177e4
LT
855 if (old_mm) {
856 up_read(&old_mm->mmap_sem);
7dddb12c 857 BUG_ON(active_mm != old_mm);
701085b2 858 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
31a78f23 859 mm_update_next_owner(old_mm);
1da177e4
LT
860 mmput(old_mm);
861 return 0;
862 }
863 mmdrop(active_mm);
864 return 0;
865}
866
867/*
868 * This function makes sure the current process has its own signal table,
869 * so that flush_signal_handlers can later reset the handlers without
870 * disturbing other processes. (Other processes might share the signal
871 * table via the CLONE_SIGHAND option to clone().)
872 */
858119e1 873static int de_thread(struct task_struct *tsk)
1da177e4
LT
874{
875 struct signal_struct *sig = tsk->signal;
b2c903b8 876 struct sighand_struct *oldsighand = tsk->sighand;
1da177e4 877 spinlock_t *lock = &oldsighand->siglock;
1da177e4 878
aafe6c2a 879 if (thread_group_empty(tsk))
1da177e4
LT
880 goto no_thread_group;
881
882 /*
883 * Kill all other threads in the thread group.
1da177e4 884 */
1da177e4 885 spin_lock_irq(lock);
ed5d2cac 886 if (signal_group_exit(sig)) {
1da177e4
LT
887 /*
888 * Another group action in progress, just
889 * return so that the signal is processed.
890 */
891 spin_unlock_irq(lock);
1da177e4
LT
892 return -EAGAIN;
893 }
d344193a 894
ed5d2cac 895 sig->group_exit_task = tsk;
d344193a
ON
896 sig->notify_count = zap_other_threads(tsk);
897 if (!thread_group_leader(tsk))
898 sig->notify_count--;
1da177e4 899
d344193a 900 while (sig->notify_count) {
d5bbd43d 901 __set_current_state(TASK_KILLABLE);
1da177e4
LT
902 spin_unlock_irq(lock);
903 schedule();
d5bbd43d
ON
904 if (unlikely(__fatal_signal_pending(tsk)))
905 goto killed;
1da177e4
LT
906 spin_lock_irq(lock);
907 }
1da177e4
LT
908 spin_unlock_irq(lock);
909
910 /*
911 * At this point all other threads have exited, all we have to
912 * do is to wait for the thread group leader to become inactive,
913 * and to assume its PID:
914 */
aafe6c2a 915 if (!thread_group_leader(tsk)) {
8187926b 916 struct task_struct *leader = tsk->group_leader;
6db840fa 917
2800d8d1 918 sig->notify_count = -1; /* for exit_notify() */
6db840fa 919 for (;;) {
e56fb287 920 threadgroup_change_begin(tsk);
6db840fa
ON
921 write_lock_irq(&tasklist_lock);
922 if (likely(leader->exit_state))
923 break;
d5bbd43d 924 __set_current_state(TASK_KILLABLE);
6db840fa 925 write_unlock_irq(&tasklist_lock);
e56fb287 926 threadgroup_change_end(tsk);
6db840fa 927 schedule();
d5bbd43d
ON
928 if (unlikely(__fatal_signal_pending(tsk)))
929 goto killed;
6db840fa 930 }
1da177e4 931
f5e90281
RM
932 /*
933 * The only record we have of the real-time age of a
934 * process, regardless of execs it's done, is start_time.
935 * All the past CPU time is accumulated in signal_struct
936 * from sister threads now dead. But in this non-leader
937 * exec, nothing survives from the original leader thread,
938 * whose birth marks the true age of this process now.
939 * When we take on its identity by switching to its PID, we
940 * also take its birthdate (always earlier than our own).
941 */
aafe6c2a 942 tsk->start_time = leader->start_time;
266b7a02 943 tsk->real_start_time = leader->real_start_time;
f5e90281 944
bac0abd6
PE
945 BUG_ON(!same_thread_group(leader, tsk));
946 BUG_ON(has_group_leader_pid(tsk));
1da177e4
LT
947 /*
948 * An exec() starts a new thread group with the
949 * TGID of the previous thread group. Rehash the
950 * two threads with a switched PID, and release
951 * the former thread group leader:
952 */
d73d6529
EB
953
954 /* Become a process group leader with the old leader's pid.
c18258c6
EB
955 * The old leader becomes a thread of the this thread group.
956 * Note: The old leader also uses this pid until release_task
d73d6529
EB
957 * is called. Odd but simple and correct.
958 */
aafe6c2a 959 tsk->pid = leader->pid;
3f418548 960 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
aafe6c2a
EB
961 transfer_pid(leader, tsk, PIDTYPE_PGID);
962 transfer_pid(leader, tsk, PIDTYPE_SID);
9cd80bbb 963
aafe6c2a 964 list_replace_rcu(&leader->tasks, &tsk->tasks);
9cd80bbb 965 list_replace_init(&leader->sibling, &tsk->sibling);
1da177e4 966
aafe6c2a
EB
967 tsk->group_leader = tsk;
968 leader->group_leader = tsk;
de12a787 969
aafe6c2a 970 tsk->exit_signal = SIGCHLD;
087806b1 971 leader->exit_signal = -1;
962b564c
ON
972
973 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
974 leader->exit_state = EXIT_DEAD;
eac1b5e5
ON
975
976 /*
977 * We are going to release_task()->ptrace_unlink() silently,
978 * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
979 * the tracer wont't block again waiting for this thread.
980 */
981 if (unlikely(leader->ptrace))
982 __wake_up_parent(leader, leader->parent);
1da177e4 983 write_unlock_irq(&tasklist_lock);
e56fb287 984 threadgroup_change_end(tsk);
8187926b
ON
985
986 release_task(leader);
ed5d2cac 987 }
1da177e4 988
6db840fa
ON
989 sig->group_exit_task = NULL;
990 sig->notify_count = 0;
1da177e4
LT
991
992no_thread_group:
e6368253
ON
993 /* we have changed execution domain */
994 tsk->exit_signal = SIGCHLD;
995
1da177e4 996 exit_itimers(sig);
cbaffba1 997 flush_itimer_signals();
329f7dba 998
b2c903b8
ON
999 if (atomic_read(&oldsighand->count) != 1) {
1000 struct sighand_struct *newsighand;
1da177e4 1001 /*
b2c903b8
ON
1002 * This ->sighand is shared with the CLONE_SIGHAND
1003 * but not CLONE_THREAD task, switch to the new one.
1da177e4 1004 */
b2c903b8
ON
1005 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1006 if (!newsighand)
1007 return -ENOMEM;
1008
1da177e4
LT
1009 atomic_set(&newsighand->count, 1);
1010 memcpy(newsighand->action, oldsighand->action,
1011 sizeof(newsighand->action));
1012
1013 write_lock_irq(&tasklist_lock);
1014 spin_lock(&oldsighand->siglock);
aafe6c2a 1015 rcu_assign_pointer(tsk->sighand, newsighand);
1da177e4
LT
1016 spin_unlock(&oldsighand->siglock);
1017 write_unlock_irq(&tasklist_lock);
1018
fba2afaa 1019 __cleanup_sighand(oldsighand);
1da177e4
LT
1020 }
1021
aafe6c2a 1022 BUG_ON(!thread_group_leader(tsk));
1da177e4 1023 return 0;
d5bbd43d
ON
1024
1025killed:
1026 /* protects against exit_notify() and __exit_signal() */
1027 read_lock(&tasklist_lock);
1028 sig->group_exit_task = NULL;
1029 sig->notify_count = 0;
1030 read_unlock(&tasklist_lock);
1031 return -EAGAIN;
1da177e4 1032}
0840a90d 1033
59714d65 1034char *get_task_comm(char *buf, struct task_struct *tsk)
1da177e4
LT
1035{
1036 /* buf must be at least sizeof(tsk->comm) in size */
1037 task_lock(tsk);
1038 strncpy(buf, tsk->comm, sizeof(tsk->comm));
1039 task_unlock(tsk);
59714d65 1040 return buf;
1da177e4 1041}
7d74f492 1042EXPORT_SYMBOL_GPL(get_task_comm);
1da177e4 1043
6a6d27de
AV
1044/*
1045 * These functions flushes out all traces of the currently running executable
1046 * so that a new one can be started
1047 */
1048
82b89778 1049void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1da177e4
LT
1050{
1051 task_lock(tsk);
43d2b113 1052 trace_task_rename(tsk, buf);
1da177e4
LT
1053 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1054 task_unlock(tsk);
82b89778 1055 perf_event_comm(tsk, exec);
1da177e4
LT
1056}
1057
1058int flush_old_exec(struct linux_binprm * bprm)
1059{
221af7f8 1060 int retval;
1da177e4
LT
1061
1062 /*
1063 * Make sure we have a private signal table and that
1064 * we are unassociated from the previous thread group.
1065 */
1066 retval = de_thread(current);
1067 if (retval)
1068 goto out;
1069
925d1c40 1070 set_mm_exe_file(bprm->mm, bprm->file);
1da177e4
LT
1071 /*
1072 * Release all of the old mmap stuff
1073 */
3c77f845 1074 acct_arg_size(bprm, 0);
1da177e4
LT
1075 retval = exec_mmap(bprm->mm);
1076 if (retval)
fd8328be 1077 goto out;
1da177e4
LT
1078
1079 bprm->mm = NULL; /* We're using it now */
7ab02af4 1080
dac853ae 1081 set_fs(USER_DS);
b88fae64
ZY
1082 current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1083 PF_NOFREEZE | PF_NO_SETAFFINITY);
7ab02af4
LT
1084 flush_thread();
1085 current->personality &= ~bprm->per_clear;
1086
221af7f8
LT
1087 return 0;
1088
1089out:
1090 return retval;
1091}
1092EXPORT_SYMBOL(flush_old_exec);
1093
1b5d783c
AV
1094void would_dump(struct linux_binprm *bprm, struct file *file)
1095{
496ad9aa 1096 if (inode_permission(file_inode(file), MAY_READ) < 0)
1b5d783c
AV
1097 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1098}
1099EXPORT_SYMBOL(would_dump);
1100
221af7f8
LT
1101void setup_new_exec(struct linux_binprm * bprm)
1102{
221af7f8 1103 arch_pick_mmap_layout(current->mm);
1da177e4
LT
1104
1105 /* This is the point of no return */
1da177e4
LT
1106 current->sas_ss_sp = current->sas_ss_size = 0;
1107
8e96e3b7 1108 if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
e579d2c2 1109 set_dumpable(current->mm, SUID_DUMP_USER);
d6e71144 1110 else
6c5d5238 1111 set_dumpable(current->mm, suid_dumpable);
d6e71144 1112
e041e328 1113 perf_event_exec();
82b89778 1114 __set_task_comm(current, kbasename(bprm->filename), true);
1da177e4 1115
0551fbd2
BH
1116 /* Set the new mm task size. We have to do that late because it may
1117 * depend on TIF_32BIT which is only updated in flush_thread() on
1118 * some architectures like powerpc
1119 */
1120 current->mm->task_size = TASK_SIZE;
1121
a6f76f23 1122 /* install the new credentials */
8e96e3b7
EB
1123 if (!uid_eq(bprm->cred->uid, current_euid()) ||
1124 !gid_eq(bprm->cred->gid, current_egid())) {
d2d56c5f 1125 current->pdeath_signal = 0;
1b5d783c
AV
1126 } else {
1127 would_dump(bprm, bprm->file);
1128 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1129 set_dumpable(current->mm, suid_dumpable);
1da177e4
LT
1130 }
1131
1132 /* An exec changes our domain. We are no longer part of the thread
1133 group */
1da177e4 1134 current->self_exec_id++;
1da177e4 1135 flush_signal_handlers(current, 0);
6a6d27de 1136 do_close_on_exec(current->files);
1da177e4 1137}
221af7f8 1138EXPORT_SYMBOL(setup_new_exec);
1da177e4 1139
a2a8474c
ON
1140/*
1141 * Prepare credentials and lock ->cred_guard_mutex.
1142 * install_exec_creds() commits the new creds and drops the lock.
1143 * Or, if exec fails before, free_bprm() should release ->cred and
1144 * and unlock.
1145 */
1146int prepare_bprm_creds(struct linux_binprm *bprm)
1147{
9b1bf12d 1148 if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
a2a8474c
ON
1149 return -ERESTARTNOINTR;
1150
1151 bprm->cred = prepare_exec_creds();
1152 if (likely(bprm->cred))
1153 return 0;
1154
9b1bf12d 1155 mutex_unlock(&current->signal->cred_guard_mutex);
a2a8474c
ON
1156 return -ENOMEM;
1157}
1158
c4ad8f98 1159static void free_bprm(struct linux_binprm *bprm)
a2a8474c
ON
1160{
1161 free_arg_pages(bprm);
1162 if (bprm->cred) {
9b1bf12d 1163 mutex_unlock(&current->signal->cred_guard_mutex);
a2a8474c
ON
1164 abort_creds(bprm->cred);
1165 }
63e46b95
ON
1166 if (bprm->file) {
1167 allow_write_access(bprm->file);
1168 fput(bprm->file);
1169 }
b66c5984
KC
1170 /* If a binfmt changed the interp, free it. */
1171 if (bprm->interp != bprm->filename)
1172 kfree(bprm->interp);
a2a8474c
ON
1173 kfree(bprm);
1174}
1175
b66c5984
KC
1176int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1177{
1178 /* If a binfmt changed the interp, free it first. */
1179 if (bprm->interp != bprm->filename)
1180 kfree(bprm->interp);
1181 bprm->interp = kstrdup(interp, GFP_KERNEL);
1182 if (!bprm->interp)
1183 return -ENOMEM;
1184 return 0;
1185}
1186EXPORT_SYMBOL(bprm_change_interp);
1187
a6f76f23
DH
1188/*
1189 * install the new credentials for this executable
1190 */
1191void install_exec_creds(struct linux_binprm *bprm)
1192{
1193 security_bprm_committing_creds(bprm);
1194
1195 commit_creds(bprm->cred);
1196 bprm->cred = NULL;
2976b10f
SE
1197
1198 /*
1199 * Disable monitoring for regular users
1200 * when executing setuid binaries. Must
1201 * wait until new credentials are committed
1202 * by commit_creds() above
1203 */
1204 if (get_dumpable(current->mm) != SUID_DUMP_USER)
1205 perf_event_exit_task(current);
a2a8474c
ON
1206 /*
1207 * cred_guard_mutex must be held at least to this point to prevent
a6f76f23 1208 * ptrace_attach() from altering our determination of the task's
a2a8474c
ON
1209 * credentials; any time after this it may be unlocked.
1210 */
a6f76f23 1211 security_bprm_committed_creds(bprm);
9b1bf12d 1212 mutex_unlock(&current->signal->cred_guard_mutex);
a6f76f23
DH
1213}
1214EXPORT_SYMBOL(install_exec_creds);
1215
1216/*
1217 * determine how safe it is to execute the proposed program
9b1bf12d 1218 * - the caller must hold ->cred_guard_mutex to protect against
a6f76f23
DH
1219 * PTRACE_ATTACH
1220 */
9e00cdb0 1221static void check_unsafe_exec(struct linux_binprm *bprm)
a6f76f23 1222{
0bf2f3ae 1223 struct task_struct *p = current, *t;
f1191b50 1224 unsigned n_fs;
a6f76f23 1225
4b9d33e6
TH
1226 if (p->ptrace) {
1227 if (p->ptrace & PT_PTRACE_CAP)
1228 bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1229 else
1230 bprm->unsafe |= LSM_UNSAFE_PTRACE;
1231 }
a6f76f23 1232
259e5e6c
AL
1233 /*
1234 * This isn't strictly necessary, but it makes it harder for LSMs to
1235 * mess up.
1236 */
1237 if (current->no_new_privs)
1238 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1239
83f62a2e 1240 t = p;
0bf2f3ae 1241 n_fs = 1;
2a4419b5 1242 spin_lock(&p->fs->lock);
437f7fdb 1243 rcu_read_lock();
83f62a2e 1244 while_each_thread(p, t) {
0bf2f3ae
DH
1245 if (t->fs == p->fs)
1246 n_fs++;
0bf2f3ae 1247 }
437f7fdb 1248 rcu_read_unlock();
0bf2f3ae 1249
9e00cdb0 1250 if (p->fs->users > n_fs)
a6f76f23 1251 bprm->unsafe |= LSM_UNSAFE_SHARE;
9e00cdb0
ON
1252 else
1253 p->fs->in_exec = 1;
2a4419b5 1254 spin_unlock(&p->fs->lock);
a6f76f23
DH
1255}
1256
9e00cdb0
ON
1257/*
1258 * Fill the binprm structure from the inode.
1da177e4 1259 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
a6f76f23
DH
1260 *
1261 * This may be called multiple times for binary chains (scripts for example).
1da177e4
LT
1262 */
1263int prepare_binprm(struct linux_binprm *bprm)
1264{
72c2d531
AV
1265 struct inode *inode = file_inode(bprm->file);
1266 umode_t mode = inode->i_mode;
1da177e4
LT
1267 int retval;
1268
1da177e4 1269
a6f76f23
DH
1270 /* clear any previous set[ug]id data from a previous binary */
1271 bprm->cred->euid = current_euid();
1272 bprm->cred->egid = current_egid();
1da177e4 1273
259e5e6c 1274 if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
3cdf5b45
EB
1275 !current->no_new_privs &&
1276 kuid_has_mapping(bprm->cred->user_ns, inode->i_uid) &&
1277 kgid_has_mapping(bprm->cred->user_ns, inode->i_gid)) {
1da177e4
LT
1278 /* Set-uid? */
1279 if (mode & S_ISUID) {
a6f76f23
DH
1280 bprm->per_clear |= PER_CLEAR_ON_SETID;
1281 bprm->cred->euid = inode->i_uid;
1da177e4
LT
1282 }
1283
1284 /* Set-gid? */
1285 /*
1286 * If setgid is set but no group execute bit then this
1287 * is a candidate for mandatory locking, not a setgid
1288 * executable.
1289 */
1290 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
a6f76f23
DH
1291 bprm->per_clear |= PER_CLEAR_ON_SETID;
1292 bprm->cred->egid = inode->i_gid;
1da177e4
LT
1293 }
1294 }
1295
1296 /* fill in binprm security blob */
a6f76f23 1297 retval = security_bprm_set_creds(bprm);
1da177e4
LT
1298 if (retval)
1299 return retval;
a6f76f23 1300 bprm->cred_prepared = 1;
1da177e4 1301
a6f76f23
DH
1302 memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1303 return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1da177e4
LT
1304}
1305
1306EXPORT_SYMBOL(prepare_binprm);
1307
4fc75ff4
NP
1308/*
1309 * Arguments are '\0' separated strings found at the location bprm->p
1310 * points to; chop off the first by relocating brpm->p to right after
1311 * the first '\0' encountered.
1312 */
b6a2fea3 1313int remove_arg_zero(struct linux_binprm *bprm)
1da177e4 1314{
b6a2fea3
OW
1315 int ret = 0;
1316 unsigned long offset;
1317 char *kaddr;
1318 struct page *page;
4fc75ff4 1319
b6a2fea3
OW
1320 if (!bprm->argc)
1321 return 0;
1da177e4 1322
b6a2fea3
OW
1323 do {
1324 offset = bprm->p & ~PAGE_MASK;
1325 page = get_arg_page(bprm, bprm->p, 0);
1326 if (!page) {
1327 ret = -EFAULT;
1328 goto out;
1329 }
e8e3c3d6 1330 kaddr = kmap_atomic(page);
4fc75ff4 1331
b6a2fea3
OW
1332 for (; offset < PAGE_SIZE && kaddr[offset];
1333 offset++, bprm->p++)
1334 ;
4fc75ff4 1335
e8e3c3d6 1336 kunmap_atomic(kaddr);
b6a2fea3 1337 put_arg_page(page);
4fc75ff4 1338
b6a2fea3
OW
1339 if (offset == PAGE_SIZE)
1340 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1341 } while (offset == PAGE_SIZE);
4fc75ff4 1342
b6a2fea3
OW
1343 bprm->p++;
1344 bprm->argc--;
1345 ret = 0;
4fc75ff4 1346
b6a2fea3
OW
1347out:
1348 return ret;
1da177e4 1349}
1da177e4
LT
1350EXPORT_SYMBOL(remove_arg_zero);
1351
cb7b6b1c 1352#define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1da177e4
LT
1353/*
1354 * cycle the list of binary formats handler, until one recognizes the image
1355 */
3c456bfc 1356int search_binary_handler(struct linux_binprm *bprm)
1da177e4 1357{
cb7b6b1c 1358 bool need_retry = IS_ENABLED(CONFIG_MODULES);
1da177e4 1359 struct linux_binfmt *fmt;
cb7b6b1c 1360 int retval;
1da177e4 1361
d7402698 1362 /* This allows 4 levels of binfmt rewrites before failing hard. */
131b2f9f 1363 if (bprm->recursion_depth > 5)
d7402698
KC
1364 return -ELOOP;
1365
1da177e4
LT
1366 retval = security_bprm_check(bprm);
1367 if (retval)
1368 return retval;
1369
1da177e4 1370 retval = -ENOENT;
cb7b6b1c
ON
1371 retry:
1372 read_lock(&binfmt_lock);
1373 list_for_each_entry(fmt, &formats, lh) {
1374 if (!try_module_get(fmt->module))
1375 continue;
1376 read_unlock(&binfmt_lock);
1377 bprm->recursion_depth++;
1378 retval = fmt->load_binary(bprm);
1379 bprm->recursion_depth--;
6b3c538f
ON
1380 if (retval >= 0 || retval != -ENOEXEC ||
1381 bprm->mm == NULL || bprm->file == NULL) {
1da177e4 1382 put_binfmt(fmt);
cb7b6b1c 1383 return retval;
1da177e4 1384 }
cb7b6b1c
ON
1385 read_lock(&binfmt_lock);
1386 put_binfmt(fmt);
1da177e4 1387 }
cb7b6b1c
ON
1388 read_unlock(&binfmt_lock);
1389
6b3c538f 1390 if (need_retry && retval == -ENOEXEC) {
cb7b6b1c
ON
1391 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1392 printable(bprm->buf[2]) && printable(bprm->buf[3]))
1393 return retval;
4e0621a0
ON
1394 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1395 return retval;
cb7b6b1c
ON
1396 need_retry = false;
1397 goto retry;
1398 }
1399
1da177e4
LT
1400 return retval;
1401}
1da177e4
LT
1402EXPORT_SYMBOL(search_binary_handler);
1403
5d1baf3b
ON
1404static int exec_binprm(struct linux_binprm *bprm)
1405{
1406 pid_t old_pid, old_vpid;
1407 int ret;
1408
1409 /* Need to fetch pid before load_binary changes it */
1410 old_pid = current->pid;
1411 rcu_read_lock();
1412 old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1413 rcu_read_unlock();
1414
1415 ret = search_binary_handler(bprm);
1416 if (ret >= 0) {
3eaded86 1417 audit_bprm(bprm);
5d1baf3b
ON
1418 trace_sched_process_exec(current, old_pid, bprm);
1419 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
9beb266f 1420 proc_exec_connector(current);
5d1baf3b
ON
1421 }
1422
1423 return ret;
1424}
1425
1da177e4
LT
1426/*
1427 * sys_execve() executes a new program.
1428 */
c4ad8f98 1429static int do_execve_common(struct filename *filename,
ba2d0162 1430 struct user_arg_ptr argv,
835ab32d 1431 struct user_arg_ptr envp)
1da177e4
LT
1432{
1433 struct linux_binprm *bprm;
1434 struct file *file;
3b125388 1435 struct files_struct *displaced;
1da177e4 1436 int retval;
72fa5997 1437
c4ad8f98
LT
1438 if (IS_ERR(filename))
1439 return PTR_ERR(filename);
1440
72fa5997
VK
1441 /*
1442 * We move the actual failure in case of RLIMIT_NPROC excess from
1443 * set*uid() to execve() because too many poorly written programs
1444 * don't check setuid() return code. Here we additionally recheck
1445 * whether NPROC limit is still exceeded.
1446 */
1447 if ((current->flags & PF_NPROC_EXCEEDED) &&
bd9d43f4 1448 atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
72fa5997
VK
1449 retval = -EAGAIN;
1450 goto out_ret;
1451 }
1452
1453 /* We're below the limit (still or again), so we don't want to make
1454 * further execve() calls fail. */
1455 current->flags &= ~PF_NPROC_EXCEEDED;
1da177e4 1456
3b125388 1457 retval = unshare_files(&displaced);
fd8328be
AV
1458 if (retval)
1459 goto out_ret;
1460
1da177e4 1461 retval = -ENOMEM;
11b0b5ab 1462 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1da177e4 1463 if (!bprm)
fd8328be 1464 goto out_files;
1da177e4 1465
a2a8474c
ON
1466 retval = prepare_bprm_creds(bprm);
1467 if (retval)
a6f76f23 1468 goto out_free;
498052bb 1469
9e00cdb0 1470 check_unsafe_exec(bprm);
a2a8474c 1471 current->in_execve = 1;
a6f76f23 1472
c4ad8f98 1473 file = do_open_exec(filename);
1da177e4
LT
1474 retval = PTR_ERR(file);
1475 if (IS_ERR(file))
498052bb 1476 goto out_unmark;
1da177e4
LT
1477
1478 sched_exec();
1479
1da177e4 1480 bprm->file = file;
c4ad8f98 1481 bprm->filename = bprm->interp = filename->name;
1da177e4 1482
b6a2fea3
OW
1483 retval = bprm_mm_init(bprm);
1484 if (retval)
63e46b95 1485 goto out_unmark;
1da177e4 1486
b6a2fea3 1487 bprm->argc = count(argv, MAX_ARG_STRINGS);
1da177e4 1488 if ((retval = bprm->argc) < 0)
a6f76f23 1489 goto out;
1da177e4 1490
b6a2fea3 1491 bprm->envc = count(envp, MAX_ARG_STRINGS);
1da177e4 1492 if ((retval = bprm->envc) < 0)
1da177e4
LT
1493 goto out;
1494
1495 retval = prepare_binprm(bprm);
1496 if (retval < 0)
1497 goto out;
1498
1499 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1500 if (retval < 0)
1501 goto out;
1502
1503 bprm->exec = bprm->p;
1504 retval = copy_strings(bprm->envc, envp, bprm);
1505 if (retval < 0)
1506 goto out;
1507
1508 retval = copy_strings(bprm->argc, argv, bprm);
1509 if (retval < 0)
1510 goto out;
1511
5d1baf3b 1512 retval = exec_binprm(bprm);
a6f76f23
DH
1513 if (retval < 0)
1514 goto out;
1da177e4 1515
a6f76f23 1516 /* execve succeeded */
498052bb 1517 current->fs->in_exec = 0;
f9ce1f1c 1518 current->in_execve = 0;
a6f76f23 1519 acct_update_integrals(current);
82727018 1520 task_numa_free(current);
a6f76f23 1521 free_bprm(bprm);
c4ad8f98 1522 putname(filename);
a6f76f23
DH
1523 if (displaced)
1524 put_files_struct(displaced);
1525 return retval;
1da177e4 1526
a6f76f23 1527out:
3c77f845
ON
1528 if (bprm->mm) {
1529 acct_arg_size(bprm, 0);
1530 mmput(bprm->mm);
1531 }
1da177e4 1532
498052bb 1533out_unmark:
9e00cdb0 1534 current->fs->in_exec = 0;
f9ce1f1c 1535 current->in_execve = 0;
a6f76f23
DH
1536
1537out_free:
08a6fac1 1538 free_bprm(bprm);
1da177e4 1539
fd8328be 1540out_files:
3b125388
AV
1541 if (displaced)
1542 reset_files_struct(displaced);
1da177e4 1543out_ret:
c4ad8f98 1544 putname(filename);
1da177e4
LT
1545 return retval;
1546}
1547
c4ad8f98 1548int do_execve(struct filename *filename,
ba2d0162 1549 const char __user *const __user *__argv,
da3d4c5f 1550 const char __user *const __user *__envp)
ba2d0162 1551{
0e028465
ON
1552 struct user_arg_ptr argv = { .ptr.native = __argv };
1553 struct user_arg_ptr envp = { .ptr.native = __envp };
835ab32d 1554 return do_execve_common(filename, argv, envp);
0e028465
ON
1555}
1556
1557#ifdef CONFIG_COMPAT
c4ad8f98 1558static int compat_do_execve(struct filename *filename,
38b983b3 1559 const compat_uptr_t __user *__argv,
d03d26e5 1560 const compat_uptr_t __user *__envp)
0e028465
ON
1561{
1562 struct user_arg_ptr argv = {
1563 .is_compat = true,
1564 .ptr.compat = __argv,
1565 };
1566 struct user_arg_ptr envp = {
1567 .is_compat = true,
1568 .ptr.compat = __envp,
1569 };
835ab32d 1570 return do_execve_common(filename, argv, envp);
ba2d0162 1571}
0e028465 1572#endif
ba2d0162 1573
964ee7df 1574void set_binfmt(struct linux_binfmt *new)
1da177e4 1575{
801460d0
HS
1576 struct mm_struct *mm = current->mm;
1577
1578 if (mm->binfmt)
1579 module_put(mm->binfmt->module);
1da177e4 1580
801460d0 1581 mm->binfmt = new;
964ee7df
ON
1582 if (new)
1583 __module_get(new->module);
1da177e4 1584}
1da177e4
LT
1585EXPORT_SYMBOL(set_binfmt);
1586
6c5d5238 1587/*
7288e118 1588 * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
6c5d5238
KH
1589 */
1590void set_dumpable(struct mm_struct *mm, int value)
1591{
abacd2fe
ON
1592 unsigned long old, new;
1593
7288e118
ON
1594 if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1595 return;
1596
abacd2fe
ON
1597 do {
1598 old = ACCESS_ONCE(mm->flags);
7288e118 1599 new = (old & ~MMF_DUMPABLE_MASK) | value;
abacd2fe 1600 } while (cmpxchg(&mm->flags, old, new) != old);
6c5d5238 1601}
6c5d5238 1602
38b983b3
AV
1603SYSCALL_DEFINE3(execve,
1604 const char __user *, filename,
1605 const char __user *const __user *, argv,
1606 const char __user *const __user *, envp)
1607{
c4ad8f98 1608 return do_execve(getname(filename), argv, envp);
38b983b3
AV
1609}
1610#ifdef CONFIG_COMPAT
625b1d7e
HC
1611COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1612 const compat_uptr_t __user *, argv,
1613 const compat_uptr_t __user *, envp)
38b983b3 1614{
c4ad8f98 1615 return compat_do_execve(getname(filename), argv, envp);
38b983b3
AV
1616}
1617#endif