Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
2 : /*
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
23 : * formats.
24 : */
25 :
26 : #include <linux/kernel_read_file.h>
27 : #include <linux/slab.h>
28 : #include <linux/file.h>
29 : #include <linux/fdtable.h>
30 : #include <linux/mm.h>
31 : #include <linux/vmacache.h>
32 : #include <linux/stat.h>
33 : #include <linux/fcntl.h>
34 : #include <linux/swap.h>
35 : #include <linux/string.h>
36 : #include <linux/init.h>
37 : #include <linux/sched/mm.h>
38 : #include <linux/sched/coredump.h>
39 : #include <linux/sched/signal.h>
40 : #include <linux/sched/numa_balancing.h>
41 : #include <linux/sched/task.h>
42 : #include <linux/pagemap.h>
43 : #include <linux/perf_event.h>
44 : #include <linux/highmem.h>
45 : #include <linux/spinlock.h>
46 : #include <linux/key.h>
47 : #include <linux/personality.h>
48 : #include <linux/binfmts.h>
49 : #include <linux/utsname.h>
50 : #include <linux/pid_namespace.h>
51 : #include <linux/module.h>
52 : #include <linux/namei.h>
53 : #include <linux/mount.h>
54 : #include <linux/security.h>
55 : #include <linux/syscalls.h>
56 : #include <linux/tsacct_kern.h>
57 : #include <linux/cn_proc.h>
58 : #include <linux/audit.h>
59 : #include <linux/kmod.h>
60 : #include <linux/fsnotify.h>
61 : #include <linux/fs_struct.h>
62 : #include <linux/oom.h>
63 : #include <linux/compat.h>
64 : #include <linux/vmalloc.h>
65 : #include <linux/io_uring.h>
66 : #include <linux/syscall_user_dispatch.h>
67 : #include <linux/coredump.h>
68 :
69 : #include <linux/uaccess.h>
70 : #include <asm/mmu_context.h>
71 : #include <asm/tlb.h>
72 :
73 : #include <trace/events/task.h>
74 : #include "internal.h"
75 :
76 : #include <trace/events/sched.h>
77 :
78 : static int bprm_creds_from_file(struct linux_binprm *bprm);
79 :
80 : int suid_dumpable = 0;
81 :
82 : static LIST_HEAD(formats);
83 : static DEFINE_RWLOCK(binfmt_lock);
84 :
85 2 : void __register_binfmt(struct linux_binfmt * fmt, int insert)
86 : {
87 2 : write_lock(&binfmt_lock);
88 2 : insert ? list_add(&fmt->lh, &formats) :
89 2 : list_add_tail(&fmt->lh, &formats);
90 2 : write_unlock(&binfmt_lock);
91 2 : }
92 :
93 : EXPORT_SYMBOL(__register_binfmt);
94 :
95 0 : void unregister_binfmt(struct linux_binfmt * fmt)
96 : {
97 0 : write_lock(&binfmt_lock);
98 0 : list_del(&fmt->lh);
99 0 : write_unlock(&binfmt_lock);
100 0 : }
101 :
102 : EXPORT_SYMBOL(unregister_binfmt);
103 :
104 : static inline void put_binfmt(struct linux_binfmt * fmt)
105 : {
106 0 : module_put(fmt->module);
107 : }
108 :
109 0 : bool path_noexec(const struct path *path)
110 : {
111 0 : return (path->mnt->mnt_flags & MNT_NOEXEC) ||
112 0 : (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
113 : }
114 :
115 : #ifdef CONFIG_USELIB
116 : /*
117 : * Note that a shared library must be both readable and executable due to
118 : * security reasons.
119 : *
120 : * Also note that we take the address to load from the file itself.
121 : */
122 : SYSCALL_DEFINE1(uselib, const char __user *, library)
123 : {
124 : struct linux_binfmt *fmt;
125 : struct file *file;
126 : struct filename *tmp = getname(library);
127 : int error = PTR_ERR(tmp);
128 : static const struct open_flags uselib_flags = {
129 : .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
130 : .acc_mode = MAY_READ | MAY_EXEC,
131 : .intent = LOOKUP_OPEN,
132 : .lookup_flags = LOOKUP_FOLLOW,
133 : };
134 :
135 : if (IS_ERR(tmp))
136 : goto out;
137 :
138 : file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
139 : putname(tmp);
140 : error = PTR_ERR(file);
141 : if (IS_ERR(file))
142 : goto out;
143 :
144 : /*
145 : * may_open() has already checked for this, so it should be
146 : * impossible to trip now. But we need to be extra cautious
147 : * and check again at the very end too.
148 : */
149 : error = -EACCES;
150 : if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
151 : path_noexec(&file->f_path)))
152 : goto exit;
153 :
154 : fsnotify_open(file);
155 :
156 : error = -ENOEXEC;
157 :
158 : read_lock(&binfmt_lock);
159 : list_for_each_entry(fmt, &formats, lh) {
160 : if (!fmt->load_shlib)
161 : continue;
162 : if (!try_module_get(fmt->module))
163 : continue;
164 : read_unlock(&binfmt_lock);
165 : error = fmt->load_shlib(file);
166 : read_lock(&binfmt_lock);
167 : put_binfmt(fmt);
168 : if (error != -ENOEXEC)
169 : break;
170 : }
171 : read_unlock(&binfmt_lock);
172 : exit:
173 : fput(file);
174 : out:
175 : return error;
176 : }
177 : #endif /* #ifdef CONFIG_USELIB */
178 :
179 : #ifdef CONFIG_MMU
180 : /*
181 : * The nascent bprm->mm is not visible until exec_mmap() but it can
182 : * use a lot of memory, account these pages in current->mm temporary
183 : * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
184 : * change the counter back via acct_arg_size(0).
185 : */
186 0 : static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
187 : {
188 0 : struct mm_struct *mm = current->mm;
189 0 : long diff = (long)(pages - bprm->vma_pages);
190 :
191 0 : if (!mm || !diff)
192 : return;
193 :
194 0 : bprm->vma_pages = pages;
195 : add_mm_counter(mm, MM_ANONPAGES, diff);
196 : }
197 :
198 0 : static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
199 : int write)
200 : {
201 : struct page *page;
202 : int ret;
203 0 : unsigned int gup_flags = FOLL_FORCE;
204 :
205 : #ifdef CONFIG_STACK_GROWSUP
206 : if (write) {
207 : ret = expand_downwards(bprm->vma, pos);
208 : if (ret < 0)
209 : return NULL;
210 : }
211 : #endif
212 :
213 0 : if (write)
214 0 : gup_flags |= FOLL_WRITE;
215 :
216 : /*
217 : * We are doing an exec(). 'current' is the process
218 : * doing the exec and bprm->mm is the new process's mm.
219 : */
220 0 : mmap_read_lock(bprm->mm);
221 0 : ret = get_user_pages_remote(bprm->mm, pos, 1, gup_flags,
222 : &page, NULL, NULL);
223 0 : mmap_read_unlock(bprm->mm);
224 0 : if (ret <= 0)
225 : return NULL;
226 :
227 0 : if (write)
228 0 : acct_arg_size(bprm, vma_pages(bprm->vma));
229 :
230 0 : return page;
231 : }
232 :
233 : static void put_arg_page(struct page *page)
234 : {
235 0 : put_page(page);
236 : }
237 :
238 : static void free_arg_pages(struct linux_binprm *bprm)
239 : {
240 : }
241 :
242 : static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
243 : struct page *page)
244 : {
245 0 : flush_cache_page(bprm->vma, pos, page_to_pfn(page));
246 : }
247 :
248 0 : static int __bprm_mm_init(struct linux_binprm *bprm)
249 : {
250 : int err;
251 0 : struct vm_area_struct *vma = NULL;
252 0 : struct mm_struct *mm = bprm->mm;
253 :
254 0 : bprm->vma = vma = vm_area_alloc(mm);
255 0 : if (!vma)
256 : return -ENOMEM;
257 0 : vma_set_anonymous(vma);
258 :
259 0 : if (mmap_write_lock_killable(mm)) {
260 : err = -EINTR;
261 : goto err_free;
262 : }
263 :
264 : /*
265 : * Place the stack at the largest stack address the architecture
266 : * supports. Later, we'll move this to an appropriate place. We don't
267 : * use STACK_TOP because that can depend on attributes which aren't
268 : * configured yet.
269 : */
270 : BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
271 0 : vma->vm_end = STACK_TOP_MAX;
272 0 : vma->vm_start = vma->vm_end - PAGE_SIZE;
273 0 : vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
274 0 : vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
275 :
276 0 : err = insert_vm_struct(mm, vma);
277 0 : if (err)
278 : goto err;
279 :
280 0 : mm->stack_vm = mm->total_vm = 1;
281 0 : mmap_write_unlock(mm);
282 0 : bprm->p = vma->vm_end - sizeof(void *);
283 0 : return 0;
284 : err:
285 : mmap_write_unlock(mm);
286 : err_free:
287 0 : bprm->vma = NULL;
288 0 : vm_area_free(vma);
289 0 : return err;
290 : }
291 :
292 : static bool valid_arg_len(struct linux_binprm *bprm, long len)
293 : {
294 0 : return len <= MAX_ARG_STRLEN;
295 : }
296 :
297 : #else
298 :
299 : static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
300 : {
301 : }
302 :
303 : static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
304 : int write)
305 : {
306 : struct page *page;
307 :
308 : page = bprm->page[pos / PAGE_SIZE];
309 : if (!page && write) {
310 : page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
311 : if (!page)
312 : return NULL;
313 : bprm->page[pos / PAGE_SIZE] = page;
314 : }
315 :
316 : return page;
317 : }
318 :
319 : static void put_arg_page(struct page *page)
320 : {
321 : }
322 :
323 : static void free_arg_page(struct linux_binprm *bprm, int i)
324 : {
325 : if (bprm->page[i]) {
326 : __free_page(bprm->page[i]);
327 : bprm->page[i] = NULL;
328 : }
329 : }
330 :
331 : static void free_arg_pages(struct linux_binprm *bprm)
332 : {
333 : int i;
334 :
335 : for (i = 0; i < MAX_ARG_PAGES; i++)
336 : free_arg_page(bprm, i);
337 : }
338 :
339 : static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
340 : struct page *page)
341 : {
342 : }
343 :
344 : static int __bprm_mm_init(struct linux_binprm *bprm)
345 : {
346 : bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
347 : return 0;
348 : }
349 :
350 : static bool valid_arg_len(struct linux_binprm *bprm, long len)
351 : {
352 : return len <= bprm->p;
353 : }
354 :
355 : #endif /* CONFIG_MMU */
356 :
357 : /*
358 : * Create a new mm_struct and populate it with a temporary stack
359 : * vm_area_struct. We don't have enough context at this point to set the stack
360 : * flags, permissions, and offset, so we use temporary values. We'll update
361 : * them later in setup_arg_pages().
362 : */
363 0 : static int bprm_mm_init(struct linux_binprm *bprm)
364 : {
365 : int err;
366 0 : struct mm_struct *mm = NULL;
367 :
368 0 : bprm->mm = mm = mm_alloc();
369 0 : err = -ENOMEM;
370 0 : if (!mm)
371 : goto err;
372 :
373 : /* Save current stack limit for all calculations made during exec. */
374 0 : task_lock(current->group_leader);
375 0 : bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
376 0 : task_unlock(current->group_leader);
377 :
378 0 : err = __bprm_mm_init(bprm);
379 0 : if (err)
380 : goto err;
381 :
382 : return 0;
383 :
384 : err:
385 0 : if (mm) {
386 0 : bprm->mm = NULL;
387 : mmdrop(mm);
388 : }
389 :
390 : return err;
391 : }
392 :
393 : struct user_arg_ptr {
394 : #ifdef CONFIG_COMPAT
395 : bool is_compat;
396 : #endif
397 : union {
398 : const char __user *const __user *native;
399 : #ifdef CONFIG_COMPAT
400 : const compat_uptr_t __user *compat;
401 : #endif
402 : } ptr;
403 : };
404 :
405 0 : static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
406 : {
407 : const char __user *native;
408 :
409 : #ifdef CONFIG_COMPAT
410 : if (unlikely(argv.is_compat)) {
411 : compat_uptr_t compat;
412 :
413 : if (get_user(compat, argv.ptr.compat + nr))
414 : return ERR_PTR(-EFAULT);
415 :
416 : return compat_ptr(compat);
417 : }
418 : #endif
419 :
420 0 : if (get_user(native, argv.ptr.native + nr))
421 : return ERR_PTR(-EFAULT);
422 :
423 0 : return native;
424 : }
425 :
426 : /*
427 : * count() counts the number of strings in array ARGV.
428 : */
429 0 : static int count(struct user_arg_ptr argv, int max)
430 : {
431 0 : int i = 0;
432 :
433 0 : if (argv.ptr.native != NULL) {
434 0 : for (;;) {
435 0 : const char __user *p = get_user_arg_ptr(argv, i);
436 :
437 0 : if (!p)
438 : break;
439 :
440 0 : if (IS_ERR(p))
441 : return -EFAULT;
442 :
443 0 : if (i >= max)
444 : return -E2BIG;
445 0 : ++i;
446 :
447 0 : if (fatal_signal_pending(current))
448 : return -ERESTARTNOHAND;
449 0 : cond_resched();
450 : }
451 : }
452 : return i;
453 : }
454 :
455 0 : static int count_strings_kernel(const char *const *argv)
456 : {
457 : int i;
458 :
459 0 : if (!argv)
460 : return 0;
461 :
462 0 : for (i = 0; argv[i]; ++i) {
463 0 : if (i >= MAX_ARG_STRINGS)
464 : return -E2BIG;
465 0 : if (fatal_signal_pending(current))
466 : return -ERESTARTNOHAND;
467 0 : cond_resched();
468 : }
469 : return i;
470 : }
471 :
472 : static int bprm_stack_limits(struct linux_binprm *bprm)
473 : {
474 : unsigned long limit, ptr_size;
475 :
476 : /*
477 : * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
478 : * (whichever is smaller) for the argv+env strings.
479 : * This ensures that:
480 : * - the remaining binfmt code will not run out of stack space,
481 : * - the program will have a reasonable amount of stack left
482 : * to work from.
483 : */
484 0 : limit = _STK_LIM / 4 * 3;
485 0 : limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
486 : /*
487 : * We've historically supported up to 32 pages (ARG_MAX)
488 : * of argument strings even with small stacks
489 : */
490 0 : limit = max_t(unsigned long, limit, ARG_MAX);
491 : /*
492 : * We must account for the size of all the argv and envp pointers to
493 : * the argv and envp strings, since they will also take up space in
494 : * the stack. They aren't stored until much later when we can't
495 : * signal to the parent that the child has run out of stack space.
496 : * Instead, calculate it here so it's possible to fail gracefully.
497 : *
498 : * In the case of argc = 0, make sure there is space for adding a
499 : * empty string (which will bump argc to 1), to ensure confused
500 : * userspace programs don't start processing from argv[1], thinking
501 : * argc can never be 0, to keep them from walking envp by accident.
502 : * See do_execveat_common().
503 : */
504 0 : ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
505 0 : if (limit <= ptr_size)
506 : return -E2BIG;
507 0 : limit -= ptr_size;
508 :
509 0 : bprm->argmin = bprm->p - limit;
510 : return 0;
511 : }
512 :
513 : /*
514 : * 'copy_strings()' copies argument/environment strings from the old
515 : * processes's memory to the new process's stack. The call to get_user_pages()
516 : * ensures the destination page is created and not swapped out.
517 : */
518 0 : static int copy_strings(int argc, struct user_arg_ptr argv,
519 : struct linux_binprm *bprm)
520 : {
521 0 : struct page *kmapped_page = NULL;
522 0 : char *kaddr = NULL;
523 0 : unsigned long kpos = 0;
524 : int ret;
525 :
526 0 : while (argc-- > 0) {
527 : const char __user *str;
528 : int len;
529 : unsigned long pos;
530 :
531 0 : ret = -EFAULT;
532 0 : str = get_user_arg_ptr(argv, argc);
533 0 : if (IS_ERR(str))
534 : goto out;
535 :
536 0 : len = strnlen_user(str, MAX_ARG_STRLEN);
537 0 : if (!len)
538 : goto out;
539 :
540 0 : ret = -E2BIG;
541 0 : if (!valid_arg_len(bprm, len))
542 : goto out;
543 :
544 : /* We're going to work our way backwards. */
545 0 : pos = bprm->p;
546 0 : str += len;
547 0 : bprm->p -= len;
548 : #ifdef CONFIG_MMU
549 0 : if (bprm->p < bprm->argmin)
550 : goto out;
551 : #endif
552 :
553 0 : while (len > 0) {
554 : int offset, bytes_to_copy;
555 :
556 0 : if (fatal_signal_pending(current)) {
557 : ret = -ERESTARTNOHAND;
558 : goto out;
559 : }
560 0 : cond_resched();
561 :
562 0 : offset = pos % PAGE_SIZE;
563 0 : if (offset == 0)
564 0 : offset = PAGE_SIZE;
565 :
566 0 : bytes_to_copy = offset;
567 0 : if (bytes_to_copy > len)
568 0 : bytes_to_copy = len;
569 :
570 0 : offset -= bytes_to_copy;
571 0 : pos -= bytes_to_copy;
572 0 : str -= bytes_to_copy;
573 0 : len -= bytes_to_copy;
574 :
575 0 : if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
576 : struct page *page;
577 :
578 0 : page = get_arg_page(bprm, pos, 1);
579 0 : if (!page) {
580 : ret = -E2BIG;
581 : goto out;
582 : }
583 :
584 0 : if (kmapped_page) {
585 0 : flush_dcache_page(kmapped_page);
586 0 : kunmap(kmapped_page);
587 : put_arg_page(kmapped_page);
588 : }
589 0 : kmapped_page = page;
590 0 : kaddr = kmap(kmapped_page);
591 0 : kpos = pos & PAGE_MASK;
592 0 : flush_arg_page(bprm, kpos, kmapped_page);
593 : }
594 0 : if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
595 : ret = -EFAULT;
596 : goto out;
597 : }
598 : }
599 : }
600 : ret = 0;
601 : out:
602 0 : if (kmapped_page) {
603 0 : flush_dcache_page(kmapped_page);
604 0 : kunmap(kmapped_page);
605 : put_arg_page(kmapped_page);
606 : }
607 0 : return ret;
608 : }
609 :
610 : /*
611 : * Copy and argument/environment string from the kernel to the processes stack.
612 : */
613 0 : int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
614 : {
615 0 : int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
616 0 : unsigned long pos = bprm->p;
617 :
618 0 : if (len == 0)
619 : return -EFAULT;
620 0 : if (!valid_arg_len(bprm, len))
621 : return -E2BIG;
622 :
623 : /* We're going to work our way backwards. */
624 0 : arg += len;
625 0 : bprm->p -= len;
626 0 : if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
627 : return -E2BIG;
628 :
629 0 : while (len > 0) {
630 0 : unsigned int bytes_to_copy = min_t(unsigned int, len,
631 : min_not_zero(offset_in_page(pos), PAGE_SIZE));
632 : struct page *page;
633 : char *kaddr;
634 :
635 0 : pos -= bytes_to_copy;
636 0 : arg -= bytes_to_copy;
637 0 : len -= bytes_to_copy;
638 :
639 0 : page = get_arg_page(bprm, pos, 1);
640 0 : if (!page)
641 : return -E2BIG;
642 0 : kaddr = kmap_atomic(page);
643 0 : flush_arg_page(bprm, pos & PAGE_MASK, page);
644 0 : memcpy(kaddr + offset_in_page(pos), arg, bytes_to_copy);
645 0 : flush_dcache_page(page);
646 0 : kunmap_atomic(kaddr);
647 : put_arg_page(page);
648 : }
649 :
650 : return 0;
651 : }
652 : EXPORT_SYMBOL(copy_string_kernel);
653 :
654 0 : static int copy_strings_kernel(int argc, const char *const *argv,
655 : struct linux_binprm *bprm)
656 : {
657 0 : while (argc-- > 0) {
658 0 : int ret = copy_string_kernel(argv[argc], bprm);
659 0 : if (ret < 0)
660 : return ret;
661 0 : if (fatal_signal_pending(current))
662 : return -ERESTARTNOHAND;
663 0 : cond_resched();
664 : }
665 : return 0;
666 : }
667 :
668 : #ifdef CONFIG_MMU
669 :
670 : /*
671 : * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
672 : * the binfmt code determines where the new stack should reside, we shift it to
673 : * its final location. The process proceeds as follows:
674 : *
675 : * 1) Use shift to calculate the new vma endpoints.
676 : * 2) Extend vma to cover both the old and new ranges. This ensures the
677 : * arguments passed to subsequent functions are consistent.
678 : * 3) Move vma's page tables to the new range.
679 : * 4) Free up any cleared pgd range.
680 : * 5) Shrink the vma to cover only the new range.
681 : */
682 0 : static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
683 : {
684 0 : struct mm_struct *mm = vma->vm_mm;
685 0 : unsigned long old_start = vma->vm_start;
686 0 : unsigned long old_end = vma->vm_end;
687 0 : unsigned long length = old_end - old_start;
688 0 : unsigned long new_start = old_start - shift;
689 0 : unsigned long new_end = old_end - shift;
690 : struct mmu_gather tlb;
691 :
692 0 : BUG_ON(new_start > new_end);
693 :
694 : /*
695 : * ensure there are no vmas between where we want to go
696 : * and where we are
697 : */
698 0 : if (vma != find_vma(mm, new_start))
699 : return -EFAULT;
700 :
701 : /*
702 : * cover the whole range: [new_start, old_end)
703 : */
704 0 : if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
705 : return -ENOMEM;
706 :
707 : /*
708 : * move the page tables downwards, on failure we rely on
709 : * process cleanup to remove whatever mess we made.
710 : */
711 0 : if (length != move_page_tables(vma, old_start,
712 : vma, new_start, length, false))
713 : return -ENOMEM;
714 :
715 0 : lru_add_drain();
716 0 : tlb_gather_mmu(&tlb, mm);
717 0 : if (new_end > old_start) {
718 : /*
719 : * when the old and new regions overlap clear from new_end.
720 : */
721 0 : free_pgd_range(&tlb, new_end, old_end, new_end,
722 0 : vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
723 : } else {
724 : /*
725 : * otherwise, clean from old_start; this is done to not touch
726 : * the address space in [new_end, old_start) some architectures
727 : * have constraints on va-space that make this illegal (IA64) -
728 : * for the others its just a little faster.
729 : */
730 0 : free_pgd_range(&tlb, old_start, old_end, new_end,
731 0 : vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
732 : }
733 0 : tlb_finish_mmu(&tlb);
734 :
735 : /*
736 : * Shrink the vma to just the new range. Always succeeds.
737 : */
738 0 : vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
739 :
740 0 : return 0;
741 : }
742 :
743 : /*
744 : * Finalizes the stack vm_area_struct. The flags and permissions are updated,
745 : * the stack is optionally relocated, and some extra space is added.
746 : */
747 0 : int setup_arg_pages(struct linux_binprm *bprm,
748 : unsigned long stack_top,
749 : int executable_stack)
750 : {
751 : unsigned long ret;
752 : unsigned long stack_shift;
753 0 : struct mm_struct *mm = current->mm;
754 0 : struct vm_area_struct *vma = bprm->vma;
755 0 : struct vm_area_struct *prev = NULL;
756 : unsigned long vm_flags;
757 : unsigned long stack_base;
758 : unsigned long stack_size;
759 : unsigned long stack_expand;
760 : unsigned long rlim_stack;
761 :
762 : #ifdef CONFIG_STACK_GROWSUP
763 : /* Limit stack size */
764 : stack_base = bprm->rlim_stack.rlim_max;
765 :
766 : stack_base = calc_max_stack_size(stack_base);
767 :
768 : /* Add space for stack randomization. */
769 : stack_base += (STACK_RND_MASK << PAGE_SHIFT);
770 :
771 : /* Make sure we didn't let the argument array grow too large. */
772 : if (vma->vm_end - vma->vm_start > stack_base)
773 : return -ENOMEM;
774 :
775 : stack_base = PAGE_ALIGN(stack_top - stack_base);
776 :
777 : stack_shift = vma->vm_start - stack_base;
778 : mm->arg_start = bprm->p - stack_shift;
779 : bprm->p = vma->vm_end - stack_shift;
780 : #else
781 0 : stack_top = arch_align_stack(stack_top);
782 0 : stack_top = PAGE_ALIGN(stack_top);
783 :
784 0 : if (unlikely(stack_top < mmap_min_addr) ||
785 0 : unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
786 : return -ENOMEM;
787 :
788 0 : stack_shift = vma->vm_end - stack_top;
789 :
790 0 : bprm->p -= stack_shift;
791 0 : mm->arg_start = bprm->p;
792 : #endif
793 :
794 0 : if (bprm->loader)
795 0 : bprm->loader -= stack_shift;
796 0 : bprm->exec -= stack_shift;
797 :
798 0 : if (mmap_write_lock_killable(mm))
799 : return -EINTR;
800 :
801 0 : vm_flags = VM_STACK_FLAGS;
802 :
803 : /*
804 : * Adjust stack execute permissions; explicitly enable for
805 : * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
806 : * (arch default) otherwise.
807 : */
808 0 : if (unlikely(executable_stack == EXSTACK_ENABLE_X))
809 : vm_flags |= VM_EXEC;
810 0 : else if (executable_stack == EXSTACK_DISABLE_X)
811 0 : vm_flags &= ~VM_EXEC;
812 0 : vm_flags |= mm->def_flags;
813 0 : vm_flags |= VM_STACK_INCOMPLETE_SETUP;
814 :
815 0 : ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
816 : vm_flags);
817 0 : if (ret)
818 : goto out_unlock;
819 0 : BUG_ON(prev != vma);
820 :
821 0 : if (unlikely(vm_flags & VM_EXEC)) {
822 0 : pr_warn_once("process '%pD4' started with executable stack\n",
823 : bprm->file);
824 : }
825 :
826 : /* Move stack pages down in memory. */
827 0 : if (stack_shift) {
828 0 : ret = shift_arg_pages(vma, stack_shift);
829 0 : if (ret)
830 : goto out_unlock;
831 : }
832 :
833 : /* mprotect_fixup is overkill to remove the temporary stack flags */
834 0 : vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
835 :
836 0 : stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
837 0 : stack_size = vma->vm_end - vma->vm_start;
838 : /*
839 : * Align this down to a page boundary as expand_stack
840 : * will align it up.
841 : */
842 0 : rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
843 : #ifdef CONFIG_STACK_GROWSUP
844 : if (stack_size + stack_expand > rlim_stack)
845 : stack_base = vma->vm_start + rlim_stack;
846 : else
847 : stack_base = vma->vm_end + stack_expand;
848 : #else
849 0 : if (stack_size + stack_expand > rlim_stack)
850 0 : stack_base = vma->vm_end - rlim_stack;
851 : else
852 0 : stack_base = vma->vm_start - stack_expand;
853 : #endif
854 0 : current->mm->start_stack = bprm->p;
855 0 : ret = expand_stack(vma, stack_base);
856 0 : if (ret)
857 0 : ret = -EFAULT;
858 :
859 : out_unlock:
860 0 : mmap_write_unlock(mm);
861 0 : return ret;
862 : }
863 : EXPORT_SYMBOL(setup_arg_pages);
864 :
865 : #else
866 :
867 : /*
868 : * Transfer the program arguments and environment from the holding pages
869 : * onto the stack. The provided stack pointer is adjusted accordingly.
870 : */
871 : int transfer_args_to_stack(struct linux_binprm *bprm,
872 : unsigned long *sp_location)
873 : {
874 : unsigned long index, stop, sp;
875 : int ret = 0;
876 :
877 : stop = bprm->p >> PAGE_SHIFT;
878 : sp = *sp_location;
879 :
880 : for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
881 : unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
882 : char *src = kmap(bprm->page[index]) + offset;
883 : sp -= PAGE_SIZE - offset;
884 : if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
885 : ret = -EFAULT;
886 : kunmap(bprm->page[index]);
887 : if (ret)
888 : goto out;
889 : }
890 :
891 : *sp_location = sp;
892 :
893 : out:
894 : return ret;
895 : }
896 : EXPORT_SYMBOL(transfer_args_to_stack);
897 :
898 : #endif /* CONFIG_MMU */
899 :
900 0 : static struct file *do_open_execat(int fd, struct filename *name, int flags)
901 : {
902 : struct file *file;
903 : int err;
904 0 : struct open_flags open_exec_flags = {
905 : .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
906 : .acc_mode = MAY_EXEC,
907 : .intent = LOOKUP_OPEN,
908 : .lookup_flags = LOOKUP_FOLLOW,
909 : };
910 :
911 0 : if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
912 : return ERR_PTR(-EINVAL);
913 0 : if (flags & AT_SYMLINK_NOFOLLOW)
914 0 : open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
915 0 : if (flags & AT_EMPTY_PATH)
916 0 : open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
917 :
918 0 : file = do_filp_open(fd, name, &open_exec_flags);
919 0 : if (IS_ERR(file))
920 : goto out;
921 :
922 : /*
923 : * may_open() has already checked for this, so it should be
924 : * impossible to trip now. But we need to be extra cautious
925 : * and check again at the very end too.
926 : */
927 0 : err = -EACCES;
928 0 : if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
929 : path_noexec(&file->f_path)))
930 : goto exit;
931 :
932 0 : err = deny_write_access(file);
933 0 : if (err)
934 : goto exit;
935 :
936 0 : if (name->name[0] != '\0')
937 0 : fsnotify_open(file);
938 :
939 : out:
940 : return file;
941 :
942 : exit:
943 0 : fput(file);
944 0 : return ERR_PTR(err);
945 : }
946 :
947 0 : struct file *open_exec(const char *name)
948 : {
949 0 : struct filename *filename = getname_kernel(name);
950 0 : struct file *f = ERR_CAST(filename);
951 :
952 0 : if (!IS_ERR(filename)) {
953 0 : f = do_open_execat(AT_FDCWD, filename, 0);
954 0 : putname(filename);
955 : }
956 0 : return f;
957 : }
958 : EXPORT_SYMBOL(open_exec);
959 :
960 : #if defined(CONFIG_HAVE_AOUT) || defined(CONFIG_BINFMT_FLAT) || \
961 : defined(CONFIG_BINFMT_ELF_FDPIC)
962 : ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
963 : {
964 : ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
965 : if (res > 0)
966 : flush_icache_user_range(addr, addr + len);
967 : return res;
968 : }
969 : EXPORT_SYMBOL(read_code);
970 : #endif
971 :
972 : /*
973 : * Maps the mm_struct mm into the current task struct.
974 : * On success, this function returns with exec_update_lock
975 : * held for writing.
976 : */
977 0 : static int exec_mmap(struct mm_struct *mm)
978 : {
979 : struct task_struct *tsk;
980 : struct mm_struct *old_mm, *active_mm;
981 : int ret;
982 :
983 : /* Notify parent that we're no longer interested in the old VM */
984 0 : tsk = current;
985 0 : old_mm = current->mm;
986 0 : exec_mm_release(tsk, old_mm);
987 : if (old_mm)
988 : sync_mm_rss(old_mm);
989 :
990 0 : ret = down_write_killable(&tsk->signal->exec_update_lock);
991 0 : if (ret)
992 : return ret;
993 :
994 0 : if (old_mm) {
995 : /*
996 : * If there is a pending fatal signal perhaps a signal
997 : * whose default action is to create a coredump get
998 : * out and die instead of going through with the exec.
999 : */
1000 0 : ret = mmap_read_lock_killable(old_mm);
1001 0 : if (ret) {
1002 0 : up_write(&tsk->signal->exec_update_lock);
1003 0 : return ret;
1004 : }
1005 : }
1006 :
1007 0 : task_lock(tsk);
1008 0 : membarrier_exec_mmap(mm);
1009 :
1010 : local_irq_disable();
1011 0 : active_mm = tsk->active_mm;
1012 0 : tsk->active_mm = mm;
1013 0 : tsk->mm = mm;
1014 : /*
1015 : * This prevents preemption while active_mm is being loaded and
1016 : * it and mm are being updated, which could cause problems for
1017 : * lazy tlb mm refcounting when these are updated by context
1018 : * switches. Not all architectures can handle irqs off over
1019 : * activate_mm yet.
1020 : */
1021 : if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1022 : local_irq_enable();
1023 0 : activate_mm(active_mm, mm);
1024 : if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1025 : local_irq_enable();
1026 0 : tsk->mm->vmacache_seqnum = 0;
1027 0 : vmacache_flush(tsk);
1028 0 : task_unlock(tsk);
1029 0 : if (old_mm) {
1030 0 : mmap_read_unlock(old_mm);
1031 0 : BUG_ON(active_mm != old_mm);
1032 0 : setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1033 0 : mm_update_next_owner(old_mm);
1034 0 : mmput(old_mm);
1035 0 : return 0;
1036 : }
1037 : mmdrop(active_mm);
1038 : return 0;
1039 : }
1040 :
1041 0 : static int de_thread(struct task_struct *tsk)
1042 : {
1043 0 : struct signal_struct *sig = tsk->signal;
1044 0 : struct sighand_struct *oldsighand = tsk->sighand;
1045 0 : spinlock_t *lock = &oldsighand->siglock;
1046 :
1047 0 : if (thread_group_empty(tsk))
1048 : goto no_thread_group;
1049 :
1050 : /*
1051 : * Kill all other threads in the thread group.
1052 : */
1053 0 : spin_lock_irq(lock);
1054 0 : if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1055 : /*
1056 : * Another group action in progress, just
1057 : * return so that the signal is processed.
1058 : */
1059 0 : spin_unlock_irq(lock);
1060 0 : return -EAGAIN;
1061 : }
1062 :
1063 0 : sig->group_exec_task = tsk;
1064 0 : sig->notify_count = zap_other_threads(tsk);
1065 0 : if (!thread_group_leader(tsk))
1066 0 : sig->notify_count--;
1067 :
1068 0 : while (sig->notify_count) {
1069 0 : __set_current_state(TASK_KILLABLE);
1070 0 : spin_unlock_irq(lock);
1071 0 : schedule();
1072 0 : if (__fatal_signal_pending(tsk))
1073 : goto killed;
1074 : spin_lock_irq(lock);
1075 : }
1076 0 : spin_unlock_irq(lock);
1077 :
1078 : /*
1079 : * At this point all other threads have exited, all we have to
1080 : * do is to wait for the thread group leader to become inactive,
1081 : * and to assume its PID:
1082 : */
1083 0 : if (!thread_group_leader(tsk)) {
1084 0 : struct task_struct *leader = tsk->group_leader;
1085 :
1086 : for (;;) {
1087 0 : cgroup_threadgroup_change_begin(tsk);
1088 0 : write_lock_irq(&tasklist_lock);
1089 : /*
1090 : * Do this under tasklist_lock to ensure that
1091 : * exit_notify() can't miss ->group_exec_task
1092 : */
1093 0 : sig->notify_count = -1;
1094 0 : if (likely(leader->exit_state))
1095 : break;
1096 0 : __set_current_state(TASK_KILLABLE);
1097 0 : write_unlock_irq(&tasklist_lock);
1098 0 : cgroup_threadgroup_change_end(tsk);
1099 0 : schedule();
1100 0 : if (__fatal_signal_pending(tsk))
1101 : goto killed;
1102 : }
1103 :
1104 : /*
1105 : * The only record we have of the real-time age of a
1106 : * process, regardless of execs it's done, is start_time.
1107 : * All the past CPU time is accumulated in signal_struct
1108 : * from sister threads now dead. But in this non-leader
1109 : * exec, nothing survives from the original leader thread,
1110 : * whose birth marks the true age of this process now.
1111 : * When we take on its identity by switching to its PID, we
1112 : * also take its birthdate (always earlier than our own).
1113 : */
1114 0 : tsk->start_time = leader->start_time;
1115 0 : tsk->start_boottime = leader->start_boottime;
1116 :
1117 0 : BUG_ON(!same_thread_group(leader, tsk));
1118 : /*
1119 : * An exec() starts a new thread group with the
1120 : * TGID of the previous thread group. Rehash the
1121 : * two threads with a switched PID, and release
1122 : * the former thread group leader:
1123 : */
1124 :
1125 : /* Become a process group leader with the old leader's pid.
1126 : * The old leader becomes a thread of the this thread group.
1127 : */
1128 0 : exchange_tids(tsk, leader);
1129 0 : transfer_pid(leader, tsk, PIDTYPE_TGID);
1130 0 : transfer_pid(leader, tsk, PIDTYPE_PGID);
1131 0 : transfer_pid(leader, tsk, PIDTYPE_SID);
1132 :
1133 0 : list_replace_rcu(&leader->tasks, &tsk->tasks);
1134 0 : list_replace_init(&leader->sibling, &tsk->sibling);
1135 :
1136 0 : tsk->group_leader = tsk;
1137 0 : leader->group_leader = tsk;
1138 :
1139 0 : tsk->exit_signal = SIGCHLD;
1140 0 : leader->exit_signal = -1;
1141 :
1142 0 : BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1143 0 : leader->exit_state = EXIT_DEAD;
1144 :
1145 : /*
1146 : * We are going to release_task()->ptrace_unlink() silently,
1147 : * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1148 : * the tracer wont't block again waiting for this thread.
1149 : */
1150 0 : if (unlikely(leader->ptrace))
1151 0 : __wake_up_parent(leader, leader->parent);
1152 0 : write_unlock_irq(&tasklist_lock);
1153 0 : cgroup_threadgroup_change_end(tsk);
1154 :
1155 0 : release_task(leader);
1156 : }
1157 :
1158 0 : sig->group_exec_task = NULL;
1159 0 : sig->notify_count = 0;
1160 :
1161 : no_thread_group:
1162 : /* we have changed execution domain */
1163 0 : tsk->exit_signal = SIGCHLD;
1164 :
1165 0 : BUG_ON(!thread_group_leader(tsk));
1166 : return 0;
1167 :
1168 : killed:
1169 : /* protects against exit_notify() and __exit_signal() */
1170 0 : read_lock(&tasklist_lock);
1171 0 : sig->group_exec_task = NULL;
1172 0 : sig->notify_count = 0;
1173 0 : read_unlock(&tasklist_lock);
1174 0 : return -EAGAIN;
1175 : }
1176 :
1177 :
1178 : /*
1179 : * This function makes sure the current process has its own signal table,
1180 : * so that flush_signal_handlers can later reset the handlers without
1181 : * disturbing other processes. (Other processes might share the signal
1182 : * table via the CLONE_SIGHAND option to clone().)
1183 : */
1184 0 : static int unshare_sighand(struct task_struct *me)
1185 : {
1186 0 : struct sighand_struct *oldsighand = me->sighand;
1187 :
1188 0 : if (refcount_read(&oldsighand->count) != 1) {
1189 : struct sighand_struct *newsighand;
1190 : /*
1191 : * This ->sighand is shared with the CLONE_SIGHAND
1192 : * but not CLONE_THREAD task, switch to the new one.
1193 : */
1194 0 : newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1195 0 : if (!newsighand)
1196 : return -ENOMEM;
1197 :
1198 0 : refcount_set(&newsighand->count, 1);
1199 0 : memcpy(newsighand->action, oldsighand->action,
1200 : sizeof(newsighand->action));
1201 :
1202 0 : write_lock_irq(&tasklist_lock);
1203 0 : spin_lock(&oldsighand->siglock);
1204 0 : rcu_assign_pointer(me->sighand, newsighand);
1205 0 : spin_unlock(&oldsighand->siglock);
1206 0 : write_unlock_irq(&tasklist_lock);
1207 :
1208 0 : __cleanup_sighand(oldsighand);
1209 : }
1210 : return 0;
1211 : }
1212 :
1213 0 : char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1214 : {
1215 0 : task_lock(tsk);
1216 : /* Always NUL terminated and zero-padded */
1217 0 : strscpy_pad(buf, tsk->comm, buf_size);
1218 0 : task_unlock(tsk);
1219 0 : return buf;
1220 : }
1221 : EXPORT_SYMBOL_GPL(__get_task_comm);
1222 :
1223 : /*
1224 : * These functions flushes out all traces of the currently running executable
1225 : * so that a new one can be started
1226 : */
1227 :
1228 106 : void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1229 : {
1230 106 : task_lock(tsk);
1231 106 : trace_task_rename(tsk, buf);
1232 106 : strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1233 106 : task_unlock(tsk);
1234 106 : perf_event_comm(tsk, exec);
1235 106 : }
1236 :
1237 : /*
1238 : * Calling this is the point of no return. None of the failures will be
1239 : * seen by userspace since either the process is already taking a fatal
1240 : * signal (via de_thread() or coredump), or will have SEGV raised
1241 : * (after exec_mmap()) by search_binary_handler (see below).
1242 : */
1243 0 : int begin_new_exec(struct linux_binprm * bprm)
1244 : {
1245 0 : struct task_struct *me = current;
1246 : int retval;
1247 :
1248 : /* Once we are committed compute the creds */
1249 0 : retval = bprm_creds_from_file(bprm);
1250 0 : if (retval)
1251 : return retval;
1252 :
1253 : /*
1254 : * Ensure all future errors are fatal.
1255 : */
1256 0 : bprm->point_of_no_return = true;
1257 :
1258 : /*
1259 : * Make this the only thread in the thread group.
1260 : */
1261 0 : retval = de_thread(me);
1262 0 : if (retval)
1263 : goto out;
1264 :
1265 : /*
1266 : * Cancel any io_uring activity across execve
1267 : */
1268 0 : io_uring_task_cancel();
1269 :
1270 : /* Ensure the files table is not shared. */
1271 0 : retval = unshare_files();
1272 0 : if (retval)
1273 : goto out;
1274 :
1275 : /*
1276 : * Must be called _before_ exec_mmap() as bprm->mm is
1277 : * not visible until then. This also enables the update
1278 : * to be lockless.
1279 : */
1280 0 : retval = set_mm_exe_file(bprm->mm, bprm->file);
1281 0 : if (retval)
1282 : goto out;
1283 :
1284 : /* If the binary is not readable then enforce mm->dumpable=0 */
1285 0 : would_dump(bprm, bprm->file);
1286 0 : if (bprm->have_execfd)
1287 0 : would_dump(bprm, bprm->executable);
1288 :
1289 : /*
1290 : * Release all of the old mmap stuff
1291 : */
1292 0 : acct_arg_size(bprm, 0);
1293 0 : retval = exec_mmap(bprm->mm);
1294 0 : if (retval)
1295 : goto out;
1296 :
1297 0 : bprm->mm = NULL;
1298 :
1299 : #ifdef CONFIG_POSIX_TIMERS
1300 0 : exit_itimers(me->signal);
1301 0 : flush_itimer_signals();
1302 : #endif
1303 :
1304 : /*
1305 : * Make the signal table private.
1306 : */
1307 0 : retval = unshare_sighand(me);
1308 0 : if (retval)
1309 : goto out_unlock;
1310 :
1311 0 : if (me->flags & PF_KTHREAD)
1312 0 : free_kthread_struct(me);
1313 0 : me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1314 : PF_NOFREEZE | PF_NO_SETAFFINITY);
1315 0 : flush_thread();
1316 0 : me->personality &= ~bprm->per_clear;
1317 :
1318 0 : clear_syscall_work_syscall_user_dispatch(me);
1319 :
1320 : /*
1321 : * We have to apply CLOEXEC before we change whether the process is
1322 : * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1323 : * trying to access the should-be-closed file descriptors of a process
1324 : * undergoing exec(2).
1325 : */
1326 0 : do_close_on_exec(me->files);
1327 :
1328 0 : if (bprm->secureexec) {
1329 : /* Make sure parent cannot signal privileged process. */
1330 0 : me->pdeath_signal = 0;
1331 :
1332 : /*
1333 : * For secureexec, reset the stack limit to sane default to
1334 : * avoid bad behavior from the prior rlimits. This has to
1335 : * happen before arch_pick_mmap_layout(), which examines
1336 : * RLIMIT_STACK, but after the point of no return to avoid
1337 : * needing to clean up the change on failure.
1338 : */
1339 0 : if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1340 0 : bprm->rlim_stack.rlim_cur = _STK_LIM;
1341 : }
1342 :
1343 0 : me->sas_ss_sp = me->sas_ss_size = 0;
1344 :
1345 : /*
1346 : * Figure out dumpability. Note that this checking only of current
1347 : * is wrong, but userspace depends on it. This should be testing
1348 : * bprm->secureexec instead.
1349 : */
1350 0 : if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1351 0 : !(uid_eq(current_euid(), current_uid()) &&
1352 0 : gid_eq(current_egid(), current_gid())))
1353 0 : set_dumpable(current->mm, suid_dumpable);
1354 : else
1355 0 : set_dumpable(current->mm, SUID_DUMP_USER);
1356 :
1357 : perf_event_exec();
1358 0 : __set_task_comm(me, kbasename(bprm->filename), true);
1359 :
1360 : /* An exec changes our domain. We are no longer part of the thread
1361 : group */
1362 0 : WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1363 0 : flush_signal_handlers(me, 0);
1364 :
1365 0 : retval = set_cred_ucounts(bprm->cred);
1366 0 : if (retval < 0)
1367 : goto out_unlock;
1368 :
1369 : /*
1370 : * install the new credentials for this executable
1371 : */
1372 0 : security_bprm_committing_creds(bprm);
1373 :
1374 0 : commit_creds(bprm->cred);
1375 0 : bprm->cred = NULL;
1376 :
1377 : /*
1378 : * Disable monitoring for regular users
1379 : * when executing setuid binaries. Must
1380 : * wait until new credentials are committed
1381 : * by commit_creds() above
1382 : */
1383 0 : if (get_dumpable(me->mm) != SUID_DUMP_USER)
1384 : perf_event_exit_task(me);
1385 : /*
1386 : * cred_guard_mutex must be held at least to this point to prevent
1387 : * ptrace_attach() from altering our determination of the task's
1388 : * credentials; any time after this it may be unlocked.
1389 : */
1390 0 : security_bprm_committed_creds(bprm);
1391 :
1392 : /* Pass the opened binary to the interpreter. */
1393 0 : if (bprm->have_execfd) {
1394 0 : retval = get_unused_fd_flags(0);
1395 0 : if (retval < 0)
1396 : goto out_unlock;
1397 0 : fd_install(retval, bprm->executable);
1398 0 : bprm->executable = NULL;
1399 0 : bprm->execfd = retval;
1400 : }
1401 : return 0;
1402 :
1403 : out_unlock:
1404 0 : up_write(&me->signal->exec_update_lock);
1405 : out:
1406 : return retval;
1407 : }
1408 : EXPORT_SYMBOL(begin_new_exec);
1409 :
1410 0 : void would_dump(struct linux_binprm *bprm, struct file *file)
1411 : {
1412 0 : struct inode *inode = file_inode(file);
1413 0 : struct user_namespace *mnt_userns = file_mnt_user_ns(file);
1414 0 : if (inode_permission(mnt_userns, inode, MAY_READ) < 0) {
1415 : struct user_namespace *old, *user_ns;
1416 0 : bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1417 :
1418 : /* Ensure mm->user_ns contains the executable */
1419 0 : user_ns = old = bprm->mm->user_ns;
1420 0 : while ((user_ns != &init_user_ns) &&
1421 0 : !privileged_wrt_inode_uidgid(user_ns, mnt_userns, inode))
1422 0 : user_ns = user_ns->parent;
1423 :
1424 0 : if (old != user_ns) {
1425 0 : bprm->mm->user_ns = get_user_ns(user_ns);
1426 0 : put_user_ns(old);
1427 : }
1428 : }
1429 0 : }
1430 : EXPORT_SYMBOL(would_dump);
1431 :
1432 0 : void setup_new_exec(struct linux_binprm * bprm)
1433 : {
1434 : /* Setup things that can depend upon the personality */
1435 0 : struct task_struct *me = current;
1436 :
1437 0 : arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1438 :
1439 : arch_setup_new_exec();
1440 :
1441 : /* Set the new mm task size. We have to do that late because it may
1442 : * depend on TIF_32BIT which is only updated in flush_thread() on
1443 : * some architectures like powerpc
1444 : */
1445 0 : me->mm->task_size = TASK_SIZE;
1446 0 : up_write(&me->signal->exec_update_lock);
1447 0 : mutex_unlock(&me->signal->cred_guard_mutex);
1448 0 : }
1449 : EXPORT_SYMBOL(setup_new_exec);
1450 :
1451 : /* Runs immediately before start_thread() takes over. */
1452 0 : void finalize_exec(struct linux_binprm *bprm)
1453 : {
1454 : /* Store any stack rlimit changes before starting thread. */
1455 0 : task_lock(current->group_leader);
1456 0 : current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1457 0 : task_unlock(current->group_leader);
1458 0 : }
1459 : EXPORT_SYMBOL(finalize_exec);
1460 :
1461 : /*
1462 : * Prepare credentials and lock ->cred_guard_mutex.
1463 : * setup_new_exec() commits the new creds and drops the lock.
1464 : * Or, if exec fails before, free_bprm() should release ->cred
1465 : * and unlock.
1466 : */
1467 0 : static int prepare_bprm_creds(struct linux_binprm *bprm)
1468 : {
1469 0 : if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1470 : return -ERESTARTNOINTR;
1471 :
1472 0 : bprm->cred = prepare_exec_creds();
1473 0 : if (likely(bprm->cred))
1474 : return 0;
1475 :
1476 0 : mutex_unlock(¤t->signal->cred_guard_mutex);
1477 : return -ENOMEM;
1478 : }
1479 :
1480 0 : static void free_bprm(struct linux_binprm *bprm)
1481 : {
1482 0 : if (bprm->mm) {
1483 0 : acct_arg_size(bprm, 0);
1484 0 : mmput(bprm->mm);
1485 : }
1486 0 : free_arg_pages(bprm);
1487 0 : if (bprm->cred) {
1488 0 : mutex_unlock(¤t->signal->cred_guard_mutex);
1489 0 : abort_creds(bprm->cred);
1490 : }
1491 0 : if (bprm->file) {
1492 0 : allow_write_access(bprm->file);
1493 0 : fput(bprm->file);
1494 : }
1495 0 : if (bprm->executable)
1496 0 : fput(bprm->executable);
1497 : /* If a binfmt changed the interp, free it. */
1498 0 : if (bprm->interp != bprm->filename)
1499 0 : kfree(bprm->interp);
1500 0 : kfree(bprm->fdpath);
1501 0 : kfree(bprm);
1502 0 : }
1503 :
1504 0 : static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1505 : {
1506 0 : struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1507 0 : int retval = -ENOMEM;
1508 0 : if (!bprm)
1509 : goto out;
1510 :
1511 0 : if (fd == AT_FDCWD || filename->name[0] == '/') {
1512 0 : bprm->filename = filename->name;
1513 : } else {
1514 0 : if (filename->name[0] == '\0')
1515 0 : bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1516 : else
1517 0 : bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1518 : fd, filename->name);
1519 0 : if (!bprm->fdpath)
1520 : goto out_free;
1521 :
1522 0 : bprm->filename = bprm->fdpath;
1523 : }
1524 0 : bprm->interp = bprm->filename;
1525 :
1526 0 : retval = bprm_mm_init(bprm);
1527 0 : if (retval)
1528 : goto out_free;
1529 : return bprm;
1530 :
1531 : out_free:
1532 0 : free_bprm(bprm);
1533 : out:
1534 0 : return ERR_PTR(retval);
1535 : }
1536 :
1537 0 : int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1538 : {
1539 : /* If a binfmt changed the interp, free it first. */
1540 0 : if (bprm->interp != bprm->filename)
1541 0 : kfree(bprm->interp);
1542 0 : bprm->interp = kstrdup(interp, GFP_KERNEL);
1543 0 : if (!bprm->interp)
1544 : return -ENOMEM;
1545 0 : return 0;
1546 : }
1547 : EXPORT_SYMBOL(bprm_change_interp);
1548 :
1549 : /*
1550 : * determine how safe it is to execute the proposed program
1551 : * - the caller must hold ->cred_guard_mutex to protect against
1552 : * PTRACE_ATTACH or seccomp thread-sync
1553 : */
1554 0 : static void check_unsafe_exec(struct linux_binprm *bprm)
1555 : {
1556 0 : struct task_struct *p = current, *t;
1557 : unsigned n_fs;
1558 :
1559 0 : if (p->ptrace)
1560 0 : bprm->unsafe |= LSM_UNSAFE_PTRACE;
1561 :
1562 : /*
1563 : * This isn't strictly necessary, but it makes it harder for LSMs to
1564 : * mess up.
1565 : */
1566 0 : if (task_no_new_privs(current))
1567 0 : bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1568 :
1569 0 : t = p;
1570 0 : n_fs = 1;
1571 0 : spin_lock(&p->fs->lock);
1572 : rcu_read_lock();
1573 0 : while_each_thread(p, t) {
1574 0 : if (t->fs == p->fs)
1575 0 : n_fs++;
1576 : }
1577 : rcu_read_unlock();
1578 :
1579 0 : if (p->fs->users > n_fs)
1580 0 : bprm->unsafe |= LSM_UNSAFE_SHARE;
1581 : else
1582 0 : p->fs->in_exec = 1;
1583 0 : spin_unlock(&p->fs->lock);
1584 0 : }
1585 :
1586 0 : static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1587 : {
1588 : /* Handle suid and sgid on files */
1589 : struct user_namespace *mnt_userns;
1590 : struct inode *inode;
1591 : unsigned int mode;
1592 : kuid_t uid;
1593 : kgid_t gid;
1594 :
1595 0 : if (!mnt_may_suid(file->f_path.mnt))
1596 : return;
1597 :
1598 0 : if (task_no_new_privs(current))
1599 : return;
1600 :
1601 0 : inode = file->f_path.dentry->d_inode;
1602 0 : mode = READ_ONCE(inode->i_mode);
1603 0 : if (!(mode & (S_ISUID|S_ISGID)))
1604 : return;
1605 :
1606 0 : mnt_userns = file_mnt_user_ns(file);
1607 :
1608 : /* Be careful if suid/sgid is set */
1609 0 : inode_lock(inode);
1610 :
1611 : /* reload atomically mode/uid/gid now that lock held */
1612 0 : mode = inode->i_mode;
1613 0 : uid = i_uid_into_mnt(mnt_userns, inode);
1614 0 : gid = i_gid_into_mnt(mnt_userns, inode);
1615 0 : inode_unlock(inode);
1616 :
1617 : /* We ignore suid/sgid if there are no mappings for them in the ns */
1618 0 : if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1619 0 : !kgid_has_mapping(bprm->cred->user_ns, gid))
1620 : return;
1621 :
1622 0 : if (mode & S_ISUID) {
1623 0 : bprm->per_clear |= PER_CLEAR_ON_SETID;
1624 0 : bprm->cred->euid = uid;
1625 : }
1626 :
1627 0 : if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1628 0 : bprm->per_clear |= PER_CLEAR_ON_SETID;
1629 0 : bprm->cred->egid = gid;
1630 : }
1631 : }
1632 :
1633 : /*
1634 : * Compute brpm->cred based upon the final binary.
1635 : */
1636 0 : static int bprm_creds_from_file(struct linux_binprm *bprm)
1637 : {
1638 : /* Compute creds based on which file? */
1639 0 : struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1640 :
1641 0 : bprm_fill_uid(bprm, file);
1642 0 : return security_bprm_creds_from_file(bprm, file);
1643 : }
1644 :
1645 : /*
1646 : * Fill the binprm structure from the inode.
1647 : * Read the first BINPRM_BUF_SIZE bytes
1648 : *
1649 : * This may be called multiple times for binary chains (scripts for example).
1650 : */
1651 0 : static int prepare_binprm(struct linux_binprm *bprm)
1652 : {
1653 0 : loff_t pos = 0;
1654 :
1655 0 : memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1656 0 : return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1657 : }
1658 :
1659 : /*
1660 : * Arguments are '\0' separated strings found at the location bprm->p
1661 : * points to; chop off the first by relocating brpm->p to right after
1662 : * the first '\0' encountered.
1663 : */
1664 0 : int remove_arg_zero(struct linux_binprm *bprm)
1665 : {
1666 0 : int ret = 0;
1667 : unsigned long offset;
1668 : char *kaddr;
1669 : struct page *page;
1670 :
1671 0 : if (!bprm->argc)
1672 : return 0;
1673 :
1674 : do {
1675 0 : offset = bprm->p & ~PAGE_MASK;
1676 0 : page = get_arg_page(bprm, bprm->p, 0);
1677 0 : if (!page) {
1678 : ret = -EFAULT;
1679 : goto out;
1680 : }
1681 0 : kaddr = kmap_atomic(page);
1682 :
1683 0 : for (; offset < PAGE_SIZE && kaddr[offset];
1684 0 : offset++, bprm->p++)
1685 : ;
1686 :
1687 0 : kunmap_atomic(kaddr);
1688 0 : put_arg_page(page);
1689 0 : } while (offset == PAGE_SIZE);
1690 :
1691 0 : bprm->p++;
1692 0 : bprm->argc--;
1693 0 : ret = 0;
1694 :
1695 : out:
1696 : return ret;
1697 : }
1698 : EXPORT_SYMBOL(remove_arg_zero);
1699 :
1700 : #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1701 : /*
1702 : * cycle the list of binary formats handler, until one recognizes the image
1703 : */
1704 0 : static int search_binary_handler(struct linux_binprm *bprm)
1705 : {
1706 0 : bool need_retry = IS_ENABLED(CONFIG_MODULES);
1707 : struct linux_binfmt *fmt;
1708 : int retval;
1709 :
1710 0 : retval = prepare_binprm(bprm);
1711 0 : if (retval < 0)
1712 : return retval;
1713 :
1714 0 : retval = security_bprm_check(bprm);
1715 : if (retval)
1716 : return retval;
1717 :
1718 0 : retval = -ENOENT;
1719 : retry:
1720 0 : read_lock(&binfmt_lock);
1721 0 : list_for_each_entry(fmt, &formats, lh) {
1722 0 : if (!try_module_get(fmt->module))
1723 : continue;
1724 0 : read_unlock(&binfmt_lock);
1725 :
1726 0 : retval = fmt->load_binary(bprm);
1727 :
1728 0 : read_lock(&binfmt_lock);
1729 0 : put_binfmt(fmt);
1730 0 : if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1731 0 : read_unlock(&binfmt_lock);
1732 0 : return retval;
1733 : }
1734 : }
1735 0 : read_unlock(&binfmt_lock);
1736 :
1737 : if (need_retry) {
1738 : if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1739 : printable(bprm->buf[2]) && printable(bprm->buf[3]))
1740 : return retval;
1741 : if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1742 : return retval;
1743 : need_retry = false;
1744 : goto retry;
1745 : }
1746 :
1747 0 : return retval;
1748 : }
1749 :
1750 0 : static int exec_binprm(struct linux_binprm *bprm)
1751 : {
1752 : pid_t old_pid, old_vpid;
1753 : int ret, depth;
1754 :
1755 : /* Need to fetch pid before load_binary changes it */
1756 0 : old_pid = current->pid;
1757 : rcu_read_lock();
1758 0 : old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1759 : rcu_read_unlock();
1760 :
1761 : /* This allows 4 levels of binfmt rewrites before failing hard. */
1762 0 : for (depth = 0;; depth++) {
1763 : struct file *exec;
1764 0 : if (depth > 5)
1765 : return -ELOOP;
1766 :
1767 0 : ret = search_binary_handler(bprm);
1768 0 : if (ret < 0)
1769 : return ret;
1770 0 : if (!bprm->interpreter)
1771 : break;
1772 :
1773 0 : exec = bprm->file;
1774 0 : bprm->file = bprm->interpreter;
1775 0 : bprm->interpreter = NULL;
1776 :
1777 0 : allow_write_access(exec);
1778 0 : if (unlikely(bprm->have_execfd)) {
1779 0 : if (bprm->executable) {
1780 0 : fput(exec);
1781 0 : return -ENOEXEC;
1782 : }
1783 0 : bprm->executable = exec;
1784 : } else
1785 0 : fput(exec);
1786 : }
1787 :
1788 0 : audit_bprm(bprm);
1789 0 : trace_sched_process_exec(current, old_pid, bprm);
1790 0 : ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1791 0 : proc_exec_connector(current);
1792 0 : return 0;
1793 : }
1794 :
1795 : /*
1796 : * sys_execve() executes a new program.
1797 : */
1798 0 : static int bprm_execve(struct linux_binprm *bprm,
1799 : int fd, struct filename *filename, int flags)
1800 : {
1801 : struct file *file;
1802 : int retval;
1803 :
1804 0 : retval = prepare_bprm_creds(bprm);
1805 0 : if (retval)
1806 : return retval;
1807 :
1808 0 : check_unsafe_exec(bprm);
1809 0 : current->in_execve = 1;
1810 :
1811 0 : file = do_open_execat(fd, filename, flags);
1812 0 : retval = PTR_ERR(file);
1813 0 : if (IS_ERR(file))
1814 : goto out_unmark;
1815 :
1816 : sched_exec();
1817 :
1818 0 : bprm->file = file;
1819 : /*
1820 : * Record that a name derived from an O_CLOEXEC fd will be
1821 : * inaccessible after exec. This allows the code in exec to
1822 : * choose to fail when the executable is not mmaped into the
1823 : * interpreter and an open file descriptor is not passed to
1824 : * the interpreter. This makes for a better user experience
1825 : * than having the interpreter start and then immediately fail
1826 : * when it finds the executable is inaccessible.
1827 : */
1828 0 : if (bprm->fdpath && get_close_on_exec(fd))
1829 0 : bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1830 :
1831 : /* Set the unchanging part of bprm->cred */
1832 0 : retval = security_bprm_creds_for_exec(bprm);
1833 : if (retval)
1834 : goto out;
1835 :
1836 0 : retval = exec_binprm(bprm);
1837 0 : if (retval < 0)
1838 : goto out;
1839 :
1840 : /* execve succeeded */
1841 0 : current->fs->in_exec = 0;
1842 0 : current->in_execve = 0;
1843 0 : rseq_execve(current);
1844 0 : acct_update_integrals(current);
1845 0 : task_numa_free(current, false);
1846 0 : return retval;
1847 :
1848 : out:
1849 : /*
1850 : * If past the point of no return ensure the code never
1851 : * returns to the userspace process. Use an existing fatal
1852 : * signal if present otherwise terminate the process with
1853 : * SIGSEGV.
1854 : */
1855 0 : if (bprm->point_of_no_return && !fatal_signal_pending(current))
1856 0 : force_fatal_sig(SIGSEGV);
1857 :
1858 : out_unmark:
1859 0 : current->fs->in_exec = 0;
1860 0 : current->in_execve = 0;
1861 :
1862 0 : return retval;
1863 : }
1864 :
1865 0 : static int do_execveat_common(int fd, struct filename *filename,
1866 : struct user_arg_ptr argv,
1867 : struct user_arg_ptr envp,
1868 : int flags)
1869 : {
1870 : struct linux_binprm *bprm;
1871 : int retval;
1872 :
1873 0 : if (IS_ERR(filename))
1874 0 : return PTR_ERR(filename);
1875 :
1876 : /*
1877 : * We move the actual failure in case of RLIMIT_NPROC excess from
1878 : * set*uid() to execve() because too many poorly written programs
1879 : * don't check setuid() return code. Here we additionally recheck
1880 : * whether NPROC limit is still exceeded.
1881 : */
1882 0 : if ((current->flags & PF_NPROC_EXCEEDED) &&
1883 0 : is_ucounts_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1884 : retval = -EAGAIN;
1885 : goto out_ret;
1886 : }
1887 :
1888 : /* We're below the limit (still or again), so we don't want to make
1889 : * further execve() calls fail. */
1890 0 : current->flags &= ~PF_NPROC_EXCEEDED;
1891 :
1892 0 : bprm = alloc_bprm(fd, filename);
1893 0 : if (IS_ERR(bprm)) {
1894 0 : retval = PTR_ERR(bprm);
1895 0 : goto out_ret;
1896 : }
1897 :
1898 0 : retval = count(argv, MAX_ARG_STRINGS);
1899 0 : if (retval == 0)
1900 0 : pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1901 : current->comm, bprm->filename);
1902 0 : if (retval < 0)
1903 : goto out_free;
1904 0 : bprm->argc = retval;
1905 :
1906 0 : retval = count(envp, MAX_ARG_STRINGS);
1907 0 : if (retval < 0)
1908 : goto out_free;
1909 0 : bprm->envc = retval;
1910 :
1911 0 : retval = bprm_stack_limits(bprm);
1912 0 : if (retval < 0)
1913 : goto out_free;
1914 :
1915 0 : retval = copy_string_kernel(bprm->filename, bprm);
1916 0 : if (retval < 0)
1917 : goto out_free;
1918 0 : bprm->exec = bprm->p;
1919 :
1920 0 : retval = copy_strings(bprm->envc, envp, bprm);
1921 0 : if (retval < 0)
1922 : goto out_free;
1923 :
1924 0 : retval = copy_strings(bprm->argc, argv, bprm);
1925 0 : if (retval < 0)
1926 : goto out_free;
1927 :
1928 : /*
1929 : * When argv is empty, add an empty string ("") as argv[0] to
1930 : * ensure confused userspace programs that start processing
1931 : * from argv[1] won't end up walking envp. See also
1932 : * bprm_stack_limits().
1933 : */
1934 0 : if (bprm->argc == 0) {
1935 0 : retval = copy_string_kernel("", bprm);
1936 0 : if (retval < 0)
1937 : goto out_free;
1938 0 : bprm->argc = 1;
1939 : }
1940 :
1941 0 : retval = bprm_execve(bprm, fd, filename, flags);
1942 : out_free:
1943 0 : free_bprm(bprm);
1944 :
1945 : out_ret:
1946 0 : putname(filename);
1947 0 : return retval;
1948 : }
1949 :
1950 0 : int kernel_execve(const char *kernel_filename,
1951 : const char *const *argv, const char *const *envp)
1952 : {
1953 : struct filename *filename;
1954 : struct linux_binprm *bprm;
1955 0 : int fd = AT_FDCWD;
1956 : int retval;
1957 :
1958 0 : filename = getname_kernel(kernel_filename);
1959 0 : if (IS_ERR(filename))
1960 0 : return PTR_ERR(filename);
1961 :
1962 0 : bprm = alloc_bprm(fd, filename);
1963 0 : if (IS_ERR(bprm)) {
1964 0 : retval = PTR_ERR(bprm);
1965 0 : goto out_ret;
1966 : }
1967 :
1968 0 : retval = count_strings_kernel(argv);
1969 0 : if (WARN_ON_ONCE(retval == 0))
1970 0 : retval = -EINVAL;
1971 0 : if (retval < 0)
1972 : goto out_free;
1973 0 : bprm->argc = retval;
1974 :
1975 0 : retval = count_strings_kernel(envp);
1976 0 : if (retval < 0)
1977 : goto out_free;
1978 0 : bprm->envc = retval;
1979 :
1980 0 : retval = bprm_stack_limits(bprm);
1981 0 : if (retval < 0)
1982 : goto out_free;
1983 :
1984 0 : retval = copy_string_kernel(bprm->filename, bprm);
1985 0 : if (retval < 0)
1986 : goto out_free;
1987 0 : bprm->exec = bprm->p;
1988 :
1989 0 : retval = copy_strings_kernel(bprm->envc, envp, bprm);
1990 0 : if (retval < 0)
1991 : goto out_free;
1992 :
1993 0 : retval = copy_strings_kernel(bprm->argc, argv, bprm);
1994 0 : if (retval < 0)
1995 : goto out_free;
1996 :
1997 0 : retval = bprm_execve(bprm, fd, filename, 0);
1998 : out_free:
1999 0 : free_bprm(bprm);
2000 : out_ret:
2001 0 : putname(filename);
2002 0 : return retval;
2003 : }
2004 :
2005 : static int do_execve(struct filename *filename,
2006 : const char __user *const __user *__argv,
2007 : const char __user *const __user *__envp)
2008 : {
2009 0 : struct user_arg_ptr argv = { .ptr.native = __argv };
2010 0 : struct user_arg_ptr envp = { .ptr.native = __envp };
2011 0 : return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2012 : }
2013 :
2014 : static int do_execveat(int fd, struct filename *filename,
2015 : const char __user *const __user *__argv,
2016 : const char __user *const __user *__envp,
2017 : int flags)
2018 : {
2019 0 : struct user_arg_ptr argv = { .ptr.native = __argv };
2020 0 : struct user_arg_ptr envp = { .ptr.native = __envp };
2021 :
2022 0 : return do_execveat_common(fd, filename, argv, envp, flags);
2023 : }
2024 :
2025 : #ifdef CONFIG_COMPAT
2026 : static int compat_do_execve(struct filename *filename,
2027 : const compat_uptr_t __user *__argv,
2028 : const compat_uptr_t __user *__envp)
2029 : {
2030 : struct user_arg_ptr argv = {
2031 : .is_compat = true,
2032 : .ptr.compat = __argv,
2033 : };
2034 : struct user_arg_ptr envp = {
2035 : .is_compat = true,
2036 : .ptr.compat = __envp,
2037 : };
2038 : return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2039 : }
2040 :
2041 : static int compat_do_execveat(int fd, struct filename *filename,
2042 : const compat_uptr_t __user *__argv,
2043 : const compat_uptr_t __user *__envp,
2044 : int flags)
2045 : {
2046 : struct user_arg_ptr argv = {
2047 : .is_compat = true,
2048 : .ptr.compat = __argv,
2049 : };
2050 : struct user_arg_ptr envp = {
2051 : .is_compat = true,
2052 : .ptr.compat = __envp,
2053 : };
2054 : return do_execveat_common(fd, filename, argv, envp, flags);
2055 : }
2056 : #endif
2057 :
2058 0 : void set_binfmt(struct linux_binfmt *new)
2059 : {
2060 0 : struct mm_struct *mm = current->mm;
2061 :
2062 0 : if (mm->binfmt)
2063 : module_put(mm->binfmt->module);
2064 :
2065 0 : mm->binfmt = new;
2066 : if (new)
2067 : __module_get(new->module);
2068 0 : }
2069 : EXPORT_SYMBOL(set_binfmt);
2070 :
2071 : /*
2072 : * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2073 : */
2074 0 : void set_dumpable(struct mm_struct *mm, int value)
2075 : {
2076 0 : if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2077 : return;
2078 :
2079 0 : set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2080 : }
2081 :
2082 0 : SYSCALL_DEFINE3(execve,
2083 : const char __user *, filename,
2084 : const char __user *const __user *, argv,
2085 : const char __user *const __user *, envp)
2086 : {
2087 0 : return do_execve(getname(filename), argv, envp);
2088 : }
2089 :
2090 0 : SYSCALL_DEFINE5(execveat,
2091 : int, fd, const char __user *, filename,
2092 : const char __user *const __user *, argv,
2093 : const char __user *const __user *, envp,
2094 : int, flags)
2095 : {
2096 0 : return do_execveat(fd,
2097 : getname_uflags(filename, flags),
2098 : argv, envp, flags);
2099 : }
2100 :
2101 : #ifdef CONFIG_COMPAT
2102 : COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2103 : const compat_uptr_t __user *, argv,
2104 : const compat_uptr_t __user *, envp)
2105 : {
2106 : return compat_do_execve(getname(filename), argv, envp);
2107 : }
2108 :
2109 : COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2110 : const char __user *, filename,
2111 : const compat_uptr_t __user *, argv,
2112 : const compat_uptr_t __user *, envp,
2113 : int, flags)
2114 : {
2115 : return compat_do_execveat(fd,
2116 : getname_uflags(filename, flags),
2117 : argv, envp, flags);
2118 : }
2119 : #endif
2120 :
2121 : #ifdef CONFIG_SYSCTL
2122 :
2123 0 : static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2124 : void *buffer, size_t *lenp, loff_t *ppos)
2125 : {
2126 0 : int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2127 :
2128 0 : if (!error)
2129 0 : validate_coredump_safety();
2130 0 : return error;
2131 : }
2132 :
2133 : static struct ctl_table fs_exec_sysctls[] = {
2134 : {
2135 : .procname = "suid_dumpable",
2136 : .data = &suid_dumpable,
2137 : .maxlen = sizeof(int),
2138 : .mode = 0644,
2139 : .proc_handler = proc_dointvec_minmax_coredump,
2140 : .extra1 = SYSCTL_ZERO,
2141 : .extra2 = SYSCTL_TWO,
2142 : },
2143 : { }
2144 : };
2145 :
2146 1 : static int __init init_fs_exec_sysctls(void)
2147 : {
2148 1 : register_sysctl_init("fs", fs_exec_sysctls);
2149 1 : return 0;
2150 : }
2151 :
2152 : fs_initcall(init_fs_exec_sysctls);
2153 : #endif /* CONFIG_SYSCTL */
|
