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/stat.h>
32 : #include <linux/fcntl.h>
33 : #include <linux/swap.h>
34 : #include <linux/string.h>
35 : #include <linux/init.h>
36 : #include <linux/sched/mm.h>
37 : #include <linux/sched/coredump.h>
38 : #include <linux/sched/signal.h>
39 : #include <linux/sched/numa_balancing.h>
40 : #include <linux/sched/task.h>
41 : #include <linux/pagemap.h>
42 : #include <linux/perf_event.h>
43 : #include <linux/highmem.h>
44 : #include <linux/spinlock.h>
45 : #include <linux/key.h>
46 : #include <linux/personality.h>
47 : #include <linux/binfmts.h>
48 : #include <linux/utsname.h>
49 : #include <linux/pid_namespace.h>
50 : #include <linux/module.h>
51 : #include <linux/namei.h>
52 : #include <linux/mount.h>
53 : #include <linux/security.h>
54 : #include <linux/syscalls.h>
55 : #include <linux/tsacct_kern.h>
56 : #include <linux/cn_proc.h>
57 : #include <linux/audit.h>
58 : #include <linux/kmod.h>
59 : #include <linux/fsnotify.h>
60 : #include <linux/fs_struct.h>
61 : #include <linux/oom.h>
62 : #include <linux/compat.h>
63 : #include <linux/vmalloc.h>
64 : #include <linux/io_uring.h>
65 : #include <linux/syscall_user_dispatch.h>
66 : #include <linux/coredump.h>
67 : #include <linux/time_namespace.h>
68 : #include <linux/user_events.h>
69 :
70 : #include <linux/uaccess.h>
71 : #include <asm/mmu_context.h>
72 : #include <asm/tlb.h>
73 :
74 : #include <trace/events/task.h>
75 : #include "internal.h"
76 :
77 : #include <trace/events/sched.h>
78 :
79 : static int bprm_creds_from_file(struct linux_binprm *bprm);
80 :
81 : int suid_dumpable = 0;
82 :
83 : static LIST_HEAD(formats);
84 : static DEFINE_RWLOCK(binfmt_lock);
85 :
86 0 : void __register_binfmt(struct linux_binfmt * fmt, int insert)
87 : {
88 0 : write_lock(&binfmt_lock);
89 0 : insert ? list_add(&fmt->lh, &formats) :
90 0 : list_add_tail(&fmt->lh, &formats);
91 0 : write_unlock(&binfmt_lock);
92 0 : }
93 :
94 : EXPORT_SYMBOL(__register_binfmt);
95 :
96 0 : void unregister_binfmt(struct linux_binfmt * fmt)
97 : {
98 0 : write_lock(&binfmt_lock);
99 0 : list_del(&fmt->lh);
100 0 : write_unlock(&binfmt_lock);
101 0 : }
102 :
103 : EXPORT_SYMBOL(unregister_binfmt);
104 :
105 : static inline void put_binfmt(struct linux_binfmt * fmt)
106 : {
107 64908985 : module_put(fmt->module);
108 : }
109 :
110 1035028521 : bool path_noexec(const struct path *path)
111 : {
112 1035028521 : return (path->mnt->mnt_flags & MNT_NOEXEC) ||
113 1034946262 : (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
114 : }
115 :
116 : #ifdef CONFIG_USELIB
117 : /*
118 : * Note that a shared library must be both readable and executable due to
119 : * security reasons.
120 : *
121 : * Also note that we take the address to load from the file itself.
122 : */
123 : SYSCALL_DEFINE1(uselib, const char __user *, library)
124 : {
125 : struct linux_binfmt *fmt;
126 : struct file *file;
127 : struct filename *tmp = getname(library);
128 : int error = PTR_ERR(tmp);
129 : static const struct open_flags uselib_flags = {
130 : .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
131 : .acc_mode = MAY_READ | MAY_EXEC,
132 : .intent = LOOKUP_OPEN,
133 : .lookup_flags = LOOKUP_FOLLOW,
134 : };
135 :
136 : if (IS_ERR(tmp))
137 : goto out;
138 :
139 : file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
140 : putname(tmp);
141 : error = PTR_ERR(file);
142 : if (IS_ERR(file))
143 : goto out;
144 :
145 : /*
146 : * may_open() has already checked for this, so it should be
147 : * impossible to trip now. But we need to be extra cautious
148 : * and check again at the very end too.
149 : */
150 : error = -EACCES;
151 : if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
152 : path_noexec(&file->f_path)))
153 : goto exit;
154 :
155 : error = -ENOEXEC;
156 :
157 : read_lock(&binfmt_lock);
158 : list_for_each_entry(fmt, &formats, lh) {
159 : if (!fmt->load_shlib)
160 : continue;
161 : if (!try_module_get(fmt->module))
162 : continue;
163 : read_unlock(&binfmt_lock);
164 : error = fmt->load_shlib(file);
165 : read_lock(&binfmt_lock);
166 : put_binfmt(fmt);
167 : if (error != -ENOEXEC)
168 : break;
169 : }
170 : read_unlock(&binfmt_lock);
171 : exit:
172 : fput(file);
173 : out:
174 : return error;
175 : }
176 : #endif /* #ifdef CONFIG_USELIB */
177 :
178 : #ifdef CONFIG_MMU
179 : /*
180 : * The nascent bprm->mm is not visible until exec_mmap() but it can
181 : * use a lot of memory, account these pages in current->mm temporary
182 : * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
183 : * change the counter back via acct_arg_size(0).
184 : */
185 184050343 : static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
186 : {
187 184050343 : struct mm_struct *mm = current->mm;
188 184050343 : long diff = (long)(pages - bprm->vma_pages);
189 :
190 184050343 : if (!mm || !diff)
191 : return;
192 :
193 110457053 : bprm->vma_pages = pages;
194 110457053 : add_mm_counter(mm, MM_ANONPAGES, diff);
195 : }
196 :
197 148293320 : static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
198 : int write)
199 : {
200 148293320 : struct page *page;
201 148293320 : struct vm_area_struct *vma = bprm->vma;
202 148293320 : struct mm_struct *mm = bprm->mm;
203 148293320 : int ret;
204 :
205 : /*
206 : * Avoid relying on expanding the stack down in GUP (which
207 : * does not work for STACK_GROWSUP anyway), and just do it
208 : * by hand ahead of time.
209 : */
210 148293320 : if (write && pos < vma->vm_start) {
211 39263892 : mmap_write_lock(mm);
212 39265949 : ret = expand_downwards(vma, pos);
213 39260904 : if (unlikely(ret < 0)) {
214 0 : mmap_write_unlock(mm);
215 0 : return NULL;
216 : }
217 39260904 : mmap_write_downgrade(mm);
218 : } else
219 109029428 : mmap_read_lock(mm);
220 :
221 : /*
222 : * We are doing an exec(). 'current' is the process
223 : * doing the exec and 'mm' is the new process's mm.
224 : */
225 148265477 : ret = get_user_pages_remote(mm, pos, 1,
226 : write ? FOLL_WRITE : 0,
227 : &page, NULL);
228 148309071 : mmap_read_unlock(mm);
229 148281449 : if (ret <= 0)
230 : return NULL;
231 :
232 148281438 : if (write)
233 148204684 : acct_arg_size(bprm, vma_pages(vma));
234 :
235 148293235 : return page;
236 : }
237 :
238 : static void put_arg_page(struct page *page)
239 : {
240 148304649 : put_page(page);
241 148230562 : }
242 :
243 : static void free_arg_pages(struct linux_binprm *bprm)
244 : {
245 : }
246 :
247 : static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
248 : struct page *page)
249 : {
250 : flush_cache_page(bprm->vma, pos, page_to_pfn(page));
251 : }
252 :
253 35860201 : static int __bprm_mm_init(struct linux_binprm *bprm)
254 : {
255 35860201 : int err;
256 35860201 : struct vm_area_struct *vma = NULL;
257 35860201 : struct mm_struct *mm = bprm->mm;
258 :
259 35860201 : bprm->vma = vma = vm_area_alloc(mm);
260 35858234 : if (!vma)
261 : return -ENOMEM;
262 35858234 : vma_set_anonymous(vma);
263 :
264 35858234 : if (mmap_write_lock_killable(mm)) {
265 0 : err = -EINTR;
266 0 : goto err_free;
267 : }
268 :
269 : /*
270 : * Place the stack at the largest stack address the architecture
271 : * supports. Later, we'll move this to an appropriate place. We don't
272 : * use STACK_TOP because that can depend on attributes which aren't
273 : * configured yet.
274 : */
275 35857588 : BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
276 35857588 : vma->vm_end = STACK_TOP_MAX;
277 35857588 : vma->vm_start = vma->vm_end - PAGE_SIZE;
278 35857588 : vm_flags_init(vma, VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP);
279 35857588 : vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
280 :
281 35857046 : err = insert_vm_struct(mm, vma);
282 35848339 : if (err)
283 0 : goto err;
284 :
285 35848339 : mm->stack_vm = mm->total_vm = 1;
286 35848339 : mmap_write_unlock(mm);
287 35846415 : bprm->p = vma->vm_end - sizeof(void *);
288 35846415 : return 0;
289 : err:
290 0 : mmap_write_unlock(mm);
291 0 : err_free:
292 0 : bprm->vma = NULL;
293 0 : vm_area_free(vma);
294 0 : return err;
295 : }
296 :
297 : static bool valid_arg_len(struct linux_binprm *bprm, long len)
298 : {
299 7664724560 : return len <= MAX_ARG_STRLEN;
300 : }
301 :
302 : #else
303 :
304 : static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
305 : {
306 : }
307 :
308 : static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
309 : int write)
310 : {
311 : struct page *page;
312 :
313 : page = bprm->page[pos / PAGE_SIZE];
314 : if (!page && write) {
315 : page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
316 : if (!page)
317 : return NULL;
318 : bprm->page[pos / PAGE_SIZE] = page;
319 : }
320 :
321 : return page;
322 : }
323 :
324 : static void put_arg_page(struct page *page)
325 : {
326 : }
327 :
328 : static void free_arg_page(struct linux_binprm *bprm, int i)
329 : {
330 : if (bprm->page[i]) {
331 : __free_page(bprm->page[i]);
332 : bprm->page[i] = NULL;
333 : }
334 : }
335 :
336 : static void free_arg_pages(struct linux_binprm *bprm)
337 : {
338 : int i;
339 :
340 : for (i = 0; i < MAX_ARG_PAGES; i++)
341 : free_arg_page(bprm, i);
342 : }
343 :
344 : static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
345 : struct page *page)
346 : {
347 : }
348 :
349 : static int __bprm_mm_init(struct linux_binprm *bprm)
350 : {
351 : bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
352 : return 0;
353 : }
354 :
355 : static bool valid_arg_len(struct linux_binprm *bprm, long len)
356 : {
357 : return len <= bprm->p;
358 : }
359 :
360 : #endif /* CONFIG_MMU */
361 :
362 : /*
363 : * Create a new mm_struct and populate it with a temporary stack
364 : * vm_area_struct. We don't have enough context at this point to set the stack
365 : * flags, permissions, and offset, so we use temporary values. We'll update
366 : * them later in setup_arg_pages().
367 : */
368 35857116 : static int bprm_mm_init(struct linux_binprm *bprm)
369 : {
370 35857116 : int err;
371 35857116 : struct mm_struct *mm = NULL;
372 :
373 35857116 : bprm->mm = mm = mm_alloc();
374 35860287 : err = -ENOMEM;
375 35860287 : if (!mm)
376 0 : goto err;
377 :
378 : /* Save current stack limit for all calculations made during exec. */
379 35860287 : task_lock(current->group_leader);
380 35860325 : bprm->rlim_stack = current->signal->rlim[RLIMIT_STACK];
381 35860325 : task_unlock(current->group_leader);
382 :
383 35860298 : err = __bprm_mm_init(bprm);
384 35846052 : if (err)
385 0 : goto err;
386 :
387 : return 0;
388 :
389 0 : err:
390 0 : if (mm) {
391 0 : bprm->mm = NULL;
392 0 : mmdrop(mm);
393 : }
394 :
395 : return err;
396 : }
397 :
398 : struct user_arg_ptr {
399 : #ifdef CONFIG_COMPAT
400 : bool is_compat;
401 : #endif
402 : union {
403 : const char __user *const __user *native;
404 : #ifdef CONFIG_COMPAT
405 : const compat_uptr_t __user *compat;
406 : #endif
407 : } ptr;
408 : };
409 :
410 15322626141 : static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
411 : {
412 15322626141 : const char __user *native;
413 :
414 : #ifdef CONFIG_COMPAT
415 15322626141 : if (unlikely(argv.is_compat)) {
416 0 : compat_uptr_t compat;
417 :
418 0 : if (get_user(compat, argv.ptr.compat + nr))
419 : return ERR_PTR(-EFAULT);
420 :
421 0 : return compat_ptr(compat);
422 : }
423 : #endif
424 :
425 15322626141 : if (get_user(native, argv.ptr.native + nr))
426 0 : return ERR_PTR(-EFAULT);
427 :
428 : return native;
429 : }
430 :
431 : /*
432 : * count() counts the number of strings in array ARGV.
433 : */
434 71204554 : static int count(struct user_arg_ptr argv, int max)
435 : {
436 71204554 : int i = 0;
437 :
438 71204554 : if (argv.ptr.native != NULL) {
439 7629298919 : for (;;) {
440 7700493863 : const char __user *p = get_user_arg_ptr(argv, i);
441 :
442 7700711582 : if (!p)
443 : break;
444 :
445 7629487526 : if (IS_ERR(p))
446 : return -EFAULT;
447 :
448 7629487526 : if (i >= max)
449 : return -E2BIG;
450 7629487526 : ++i;
451 :
452 7629487526 : if (fatal_signal_pending(current))
453 : return -ERESTARTNOHAND;
454 7629183758 : cond_resched();
455 : }
456 : }
457 : return i;
458 : }
459 :
460 492132 : static int count_strings_kernel(const char *const *argv)
461 : {
462 492132 : int i;
463 :
464 492132 : if (!argv)
465 : return 0;
466 :
467 2214587 : for (i = 0; argv[i]; ++i) {
468 1722455 : if (i >= MAX_ARG_STRINGS)
469 : return -E2BIG;
470 1722455 : if (fatal_signal_pending(current))
471 : return -ERESTARTNOHAND;
472 1722455 : cond_resched();
473 : }
474 : return i;
475 : }
476 :
477 35855238 : static int bprm_stack_limits(struct linux_binprm *bprm)
478 : {
479 35855238 : unsigned long limit, ptr_size;
480 :
481 : /*
482 : * Limit to 1/4 of the max stack size or 3/4 of _STK_LIM
483 : * (whichever is smaller) for the argv+env strings.
484 : * This ensures that:
485 : * - the remaining binfmt code will not run out of stack space,
486 : * - the program will have a reasonable amount of stack left
487 : * to work from.
488 : */
489 35855238 : limit = _STK_LIM / 4 * 3;
490 35855238 : limit = min(limit, bprm->rlim_stack.rlim_cur / 4);
491 : /*
492 : * We've historically supported up to 32 pages (ARG_MAX)
493 : * of argument strings even with small stacks
494 : */
495 35855238 : limit = max_t(unsigned long, limit, ARG_MAX);
496 : /*
497 : * We must account for the size of all the argv and envp pointers to
498 : * the argv and envp strings, since they will also take up space in
499 : * the stack. They aren't stored until much later when we can't
500 : * signal to the parent that the child has run out of stack space.
501 : * Instead, calculate it here so it's possible to fail gracefully.
502 : *
503 : * In the case of argc = 0, make sure there is space for adding a
504 : * empty string (which will bump argc to 1), to ensure confused
505 : * userspace programs don't start processing from argv[1], thinking
506 : * argc can never be 0, to keep them from walking envp by accident.
507 : * See do_execveat_common().
508 : */
509 35855238 : ptr_size = (max(bprm->argc, 1) + bprm->envc) * sizeof(void *);
510 35855238 : if (limit <= ptr_size)
511 : return -E2BIG;
512 35855238 : limit -= ptr_size;
513 :
514 35855238 : bprm->argmin = bprm->p - limit;
515 35855238 : return 0;
516 : }
517 :
518 : /*
519 : * 'copy_strings()' copies argument/environment strings from the old
520 : * processes's memory to the new process's stack. The call to get_user_pages()
521 : * ensures the destination page is created and not swapped out.
522 : */
523 71193561 : static int copy_strings(int argc, struct user_arg_ptr argv,
524 : struct linux_binprm *bprm)
525 : {
526 71193561 : struct page *kmapped_page = NULL;
527 71193561 : char *kaddr = NULL;
528 71193561 : unsigned long kpos = 0;
529 71193561 : int ret;
530 :
531 7699592407 : while (argc-- > 0) {
532 7628365672 : const char __user *str;
533 7628365672 : int len;
534 7628365672 : unsigned long pos;
535 :
536 7628365672 : ret = -EFAULT;
537 7628365672 : str = get_user_arg_ptr(argv, argc);
538 7629027364 : if (IS_ERR(str))
539 0 : goto out;
540 :
541 7629027364 : len = strnlen_user(str, MAX_ARG_STRLEN);
542 7626975502 : if (!len)
543 0 : goto out;
544 :
545 7626975502 : ret = -E2BIG;
546 7626975502 : if (!valid_arg_len(bprm, len))
547 0 : goto out;
548 :
549 : /* We're going to work our way backwards. */
550 7626975502 : pos = bprm->p;
551 7626975502 : str += len;
552 7626975502 : bprm->p -= len;
553 : #ifdef CONFIG_MMU
554 7626975502 : if (bprm->p < bprm->argmin)
555 0 : goto out;
556 : #endif
557 :
558 15294440118 : while (len > 0) {
559 7666041272 : int offset, bytes_to_copy;
560 :
561 7666041272 : if (fatal_signal_pending(current)) {
562 80 : ret = -ERESTARTNOHAND;
563 80 : goto out;
564 : }
565 7664913341 : cond_resched();
566 :
567 7665937052 : offset = pos % PAGE_SIZE;
568 7665937052 : if (offset == 0)
569 39272962 : offset = PAGE_SIZE;
570 :
571 7665937052 : bytes_to_copy = offset;
572 7665937052 : if (bytes_to_copy > len)
573 : bytes_to_copy = len;
574 :
575 7665937052 : offset -= bytes_to_copy;
576 7665937052 : pos -= bytes_to_copy;
577 7665937052 : str -= bytes_to_copy;
578 7665937052 : len -= bytes_to_copy;
579 :
580 7665937052 : if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
581 110464820 : struct page *page;
582 :
583 110464820 : page = get_arg_page(bprm, pos, 1);
584 110474105 : if (!page) {
585 6 : ret = -E2BIG;
586 6 : goto out;
587 : }
588 :
589 110474099 : if (kmapped_page) {
590 39269210 : flush_dcache_page(kmapped_page);
591 39269210 : kunmap_local(kaddr);
592 39269210 : put_arg_page(kmapped_page);
593 : }
594 110473127 : kmapped_page = page;
595 110473127 : kaddr = kmap_local_page(kmapped_page);
596 110473127 : kpos = pos & PAGE_MASK;
597 110473127 : flush_arg_page(bprm, kpos, kmapped_page);
598 : }
599 15333409975 : if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
600 0 : ret = -EFAULT;
601 0 : goto out;
602 : }
603 : }
604 : }
605 : ret = 0;
606 71226821 : out:
607 71226821 : if (kmapped_page) {
608 71221494 : flush_dcache_page(kmapped_page);
609 71221494 : kunmap_local(kaddr);
610 71221494 : put_arg_page(kmapped_page);
611 : }
612 71224371 : return ret;
613 : }
614 :
615 : /*
616 : * Copy and argument/environment string from the kernel to the processes stack.
617 : */
618 37751129 : int copy_string_kernel(const char *arg, struct linux_binprm *bprm)
619 : {
620 37751129 : int len = strnlen(arg, MAX_ARG_STRLEN) + 1 /* terminating NUL */;
621 37749058 : unsigned long pos = bprm->p;
622 :
623 37749058 : if (len == 0)
624 : return -EFAULT;
625 37749058 : if (!valid_arg_len(bprm, len))
626 : return -E2BIG;
627 :
628 : /* We're going to work our way backwards. */
629 37749058 : arg += len;
630 37749058 : bprm->p -= len;
631 37749058 : if (IS_ENABLED(CONFIG_MMU) && bprm->p < bprm->argmin)
632 : return -E2BIG;
633 :
634 75492338 : while (len > 0) {
635 37749979 : unsigned int bytes_to_copy = min_t(unsigned int, len,
636 : min_not_zero(offset_in_page(pos), PAGE_SIZE));
637 37749979 : struct page *page;
638 :
639 37749979 : pos -= bytes_to_copy;
640 37749979 : arg -= bytes_to_copy;
641 37749979 : len -= bytes_to_copy;
642 :
643 37749979 : page = get_arg_page(bprm, pos, 1);
644 37750048 : if (!page)
645 : return -E2BIG;
646 37750043 : flush_arg_page(bprm, pos & PAGE_MASK, page);
647 37750043 : memcpy_to_page(page, offset_in_page(pos), arg, bytes_to_copy);
648 37739080 : put_arg_page(page);
649 : }
650 :
651 : return 0;
652 : }
653 : EXPORT_SYMBOL(copy_string_kernel);
654 :
655 492131 : static int copy_strings_kernel(int argc, const char *const *argv,
656 : struct linux_binprm *bprm)
657 : {
658 2214588 : while (argc-- > 0) {
659 1722458 : int ret = copy_string_kernel(argv[argc], bprm);
660 1722457 : if (ret < 0)
661 0 : return ret;
662 1722457 : if (fatal_signal_pending(current))
663 : return -ERESTARTNOHAND;
664 1722457 : cond_resched();
665 : }
666 : return 0;
667 : }
668 :
669 : #ifdef CONFIG_MMU
670 :
671 : /*
672 : * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
673 : * the binfmt code determines where the new stack should reside, we shift it to
674 : * its final location. The process proceeds as follows:
675 : *
676 : * 1) Use shift to calculate the new vma endpoints.
677 : * 2) Extend vma to cover both the old and new ranges. This ensures the
678 : * arguments passed to subsequent functions are consistent.
679 : * 3) Move vma's page tables to the new range.
680 : * 4) Free up any cleared pgd range.
681 : * 5) Shrink the vma to cover only the new range.
682 : */
683 32404023 : static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
684 : {
685 32404023 : struct mm_struct *mm = vma->vm_mm;
686 32404023 : unsigned long old_start = vma->vm_start;
687 32404023 : unsigned long old_end = vma->vm_end;
688 32404023 : unsigned long length = old_end - old_start;
689 32404023 : unsigned long new_start = old_start - shift;
690 32404023 : unsigned long new_end = old_end - shift;
691 32404023 : VMA_ITERATOR(vmi, mm, new_start);
692 32404023 : struct vm_area_struct *next;
693 32404023 : struct mmu_gather tlb;
694 :
695 32404023 : BUG_ON(new_start > new_end);
696 :
697 : /*
698 : * ensure there are no vmas between where we want to go
699 : * and where we are
700 : */
701 32404023 : if (vma != vma_next(&vmi))
702 : return -EFAULT;
703 :
704 : /*
705 : * cover the whole range: [new_start, old_end)
706 : */
707 32412524 : if (vma_expand(&vmi, vma, new_start, old_end, vma->vm_pgoff, NULL))
708 : return -ENOMEM;
709 :
710 : /*
711 : * move the page tables downwards, on failure we rely on
712 : * process cleanup to remove whatever mess we made.
713 : */
714 32401835 : if (length != move_page_tables(vma, old_start,
715 : vma, new_start, length, false))
716 : return -ENOMEM;
717 :
718 32410051 : lru_add_drain();
719 32403748 : tlb_gather_mmu(&tlb, mm);
720 32425781 : next = vma_next(&vmi);
721 32404406 : if (new_end > old_start) {
722 : /*
723 : * when the old and new regions overlap clear from new_end.
724 : */
725 13 : free_pgd_range(&tlb, new_end, old_end, new_end,
726 : next ? next->vm_start : USER_PGTABLES_CEILING);
727 : } else {
728 : /*
729 : * otherwise, clean from old_start; this is done to not touch
730 : * the address space in [new_end, old_start) some architectures
731 : * have constraints on va-space that make this illegal (IA64) -
732 : * for the others its just a little faster.
733 : */
734 32404393 : free_pgd_range(&tlb, old_start, old_end, new_end,
735 : next ? next->vm_start : USER_PGTABLES_CEILING);
736 : }
737 32414379 : tlb_finish_mmu(&tlb);
738 :
739 32417686 : vma_prev(&vmi);
740 : /* Shrink the vma to just the new range */
741 32401226 : return vma_shrink(&vmi, vma, new_start, new_end, vma->vm_pgoff);
742 : }
743 :
744 : /*
745 : * Finalizes the stack vm_area_struct. The flags and permissions are updated,
746 : * the stack is optionally relocated, and some extra space is added.
747 : */
748 32411459 : int setup_arg_pages(struct linux_binprm *bprm,
749 : unsigned long stack_top,
750 : int executable_stack)
751 : {
752 32411459 : unsigned long ret;
753 32411459 : unsigned long stack_shift;
754 32411459 : struct mm_struct *mm = current->mm;
755 32411459 : struct vm_area_struct *vma = bprm->vma;
756 32411459 : struct vm_area_struct *prev = NULL;
757 32411459 : unsigned long vm_flags;
758 32411459 : unsigned long stack_base;
759 32411459 : unsigned long stack_size;
760 32411459 : unsigned long stack_expand;
761 32411459 : unsigned long rlim_stack;
762 32411459 : struct mmu_gather tlb;
763 32411459 : struct vma_iterator vmi;
764 :
765 : #ifdef CONFIG_STACK_GROWSUP
766 : /* Limit stack size */
767 : stack_base = bprm->rlim_stack.rlim_max;
768 :
769 : stack_base = calc_max_stack_size(stack_base);
770 :
771 : /* Add space for stack randomization. */
772 : stack_base += (STACK_RND_MASK << PAGE_SHIFT);
773 :
774 : /* Make sure we didn't let the argument array grow too large. */
775 : if (vma->vm_end - vma->vm_start > stack_base)
776 : return -ENOMEM;
777 :
778 : stack_base = PAGE_ALIGN(stack_top - stack_base);
779 :
780 : stack_shift = vma->vm_start - stack_base;
781 : mm->arg_start = bprm->p - stack_shift;
782 : bprm->p = vma->vm_end - stack_shift;
783 : #else
784 32411459 : stack_top = arch_align_stack(stack_top);
785 32421875 : stack_top = PAGE_ALIGN(stack_top);
786 :
787 32421875 : if (unlikely(stack_top < mmap_min_addr) ||
788 32421875 : unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
789 : return -ENOMEM;
790 :
791 32421875 : stack_shift = vma->vm_end - stack_top;
792 :
793 32421875 : bprm->p -= stack_shift;
794 32421875 : mm->arg_start = bprm->p;
795 : #endif
796 :
797 32421875 : if (bprm->loader)
798 0 : bprm->loader -= stack_shift;
799 32421875 : bprm->exec -= stack_shift;
800 :
801 32421875 : if (mmap_write_lock_killable(mm))
802 : return -EINTR;
803 :
804 32419524 : vm_flags = VM_STACK_FLAGS;
805 :
806 : /*
807 : * Adjust stack execute permissions; explicitly enable for
808 : * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
809 : * (arch default) otherwise.
810 : */
811 32419524 : if (unlikely(executable_stack == EXSTACK_ENABLE_X))
812 : vm_flags |= VM_EXEC;
813 32419524 : else if (executable_stack == EXSTACK_DISABLE_X)
814 32419235 : vm_flags &= ~VM_EXEC;
815 32419524 : vm_flags |= mm->def_flags;
816 32419524 : vm_flags |= VM_STACK_INCOMPLETE_SETUP;
817 :
818 32419524 : vma_iter_init(&vmi, mm, vma->vm_start);
819 :
820 32410744 : tlb_gather_mmu(&tlb, mm);
821 32417281 : ret = mprotect_fixup(&vmi, &tlb, vma, &prev, vma->vm_start, vma->vm_end,
822 : vm_flags);
823 32404725 : tlb_finish_mmu(&tlb);
824 :
825 32409238 : if (ret)
826 0 : goto out_unlock;
827 32409238 : BUG_ON(prev != vma);
828 :
829 32409238 : if (unlikely(vm_flags & VM_EXEC)) {
830 0 : pr_warn_once("process '%pD4' started with executable stack\n",
831 : bprm->file);
832 : }
833 :
834 : /* Move stack pages down in memory. */
835 32409238 : if (stack_shift) {
836 32410553 : ret = shift_arg_pages(vma, stack_shift);
837 32408102 : if (ret)
838 0 : goto out_unlock;
839 : }
840 :
841 : /* mprotect_fixup is overkill to remove the temporary stack flags */
842 32406787 : vm_flags_clear(vma, VM_STACK_INCOMPLETE_SETUP);
843 :
844 32402176 : stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
845 32402176 : stack_size = vma->vm_end - vma->vm_start;
846 : /*
847 : * Align this down to a page boundary as expand_stack
848 : * will align it up.
849 : */
850 32402176 : rlim_stack = bprm->rlim_stack.rlim_cur & PAGE_MASK;
851 :
852 32402176 : stack_expand = min(rlim_stack, stack_size + stack_expand);
853 :
854 : #ifdef CONFIG_STACK_GROWSUP
855 : stack_base = vma->vm_start + stack_expand;
856 : #else
857 32402176 : stack_base = vma->vm_end - stack_expand;
858 : #endif
859 32402176 : current->mm->start_stack = bprm->p;
860 32402176 : ret = expand_stack_locked(vma, stack_base);
861 32373104 : if (ret)
862 0 : ret = -EFAULT;
863 :
864 32373104 : out_unlock:
865 32373104 : mmap_write_unlock(mm);
866 32389636 : return ret;
867 : }
868 : EXPORT_SYMBOL(setup_arg_pages);
869 :
870 : #else
871 :
872 : /*
873 : * Transfer the program arguments and environment from the holding pages
874 : * onto the stack. The provided stack pointer is adjusted accordingly.
875 : */
876 : int transfer_args_to_stack(struct linux_binprm *bprm,
877 : unsigned long *sp_location)
878 : {
879 : unsigned long index, stop, sp;
880 : int ret = 0;
881 :
882 : stop = bprm->p >> PAGE_SHIFT;
883 : sp = *sp_location;
884 :
885 : for (index = MAX_ARG_PAGES - 1; index >= stop; index--) {
886 : unsigned int offset = index == stop ? bprm->p & ~PAGE_MASK : 0;
887 : char *src = kmap_local_page(bprm->page[index]) + offset;
888 : sp -= PAGE_SIZE - offset;
889 : if (copy_to_user((void *) sp, src, PAGE_SIZE - offset) != 0)
890 : ret = -EFAULT;
891 : kunmap_local(src);
892 : if (ret)
893 : goto out;
894 : }
895 :
896 : *sp_location = sp;
897 :
898 : out:
899 : return ret;
900 : }
901 : EXPORT_SYMBOL(transfer_args_to_stack);
902 :
903 : #endif /* CONFIG_MMU */
904 :
905 68344808 : static struct file *do_open_execat(int fd, struct filename *name, int flags)
906 : {
907 68344808 : struct file *file;
908 68344808 : int err;
909 68344808 : struct open_flags open_exec_flags = {
910 : .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
911 : .acc_mode = MAY_EXEC,
912 : .intent = LOOKUP_OPEN,
913 : .lookup_flags = LOOKUP_FOLLOW,
914 : };
915 :
916 68344808 : if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
917 : return ERR_PTR(-EINVAL);
918 68344808 : if (flags & AT_SYMLINK_NOFOLLOW)
919 0 : open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
920 68344808 : if (flags & AT_EMPTY_PATH)
921 0 : open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
922 :
923 68344808 : file = do_filp_open(fd, name, &open_exec_flags);
924 68327314 : if (IS_ERR(file))
925 3430437 : goto out;
926 :
927 : /*
928 : * may_open() has already checked for this, so it should be
929 : * impossible to trip now. But we need to be extra cautious
930 : * and check again at the very end too.
931 : */
932 64896877 : err = -EACCES;
933 64896877 : if (WARN_ON_ONCE(!S_ISREG(file_inode(file)->i_mode) ||
934 : path_noexec(&file->f_path)))
935 0 : goto exit;
936 :
937 64896877 : err = deny_write_access(file);
938 64926135 : if (err)
939 0 : goto exit;
940 :
941 64926135 : out:
942 : return file;
943 :
944 0 : exit:
945 0 : fput(file);
946 0 : return ERR_PTR(err);
947 : }
948 :
949 32498159 : struct file *open_exec(const char *name)
950 : {
951 32498159 : struct filename *filename = getname_kernel(name);
952 32483442 : struct file *f = ERR_CAST(filename);
953 :
954 32483442 : if (!IS_ERR(filename)) {
955 32483442 : f = do_open_execat(AT_FDCWD, filename, 0);
956 32503370 : putname(filename);
957 : }
958 32499337 : return f;
959 : }
960 : EXPORT_SYMBOL(open_exec);
961 :
962 : #if defined(CONFIG_BINFMT_FLAT) || defined(CONFIG_BINFMT_ELF_FDPIC)
963 : ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
964 : {
965 : ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
966 : if (res > 0)
967 : flush_icache_user_range(addr, addr + len);
968 : return res;
969 : }
970 : EXPORT_SYMBOL(read_code);
971 : #endif
972 :
973 : /*
974 : * Maps the mm_struct mm into the current task struct.
975 : * On success, this function returns with exec_update_lock
976 : * held for writing.
977 : */
978 32412350 : static int exec_mmap(struct mm_struct *mm)
979 : {
980 32412350 : struct task_struct *tsk;
981 32412350 : struct mm_struct *old_mm, *active_mm;
982 32412350 : int ret;
983 :
984 : /* Notify parent that we're no longer interested in the old VM */
985 32412350 : tsk = current;
986 32412350 : old_mm = current->mm;
987 32412350 : exec_mm_release(tsk, old_mm);
988 32412968 : if (old_mm)
989 : sync_mm_rss(old_mm);
990 :
991 32412968 : ret = down_write_killable(&tsk->signal->exec_update_lock);
992 32414945 : if (ret)
993 : return ret;
994 :
995 32414945 : if (old_mm) {
996 : /*
997 : * If there is a pending fatal signal perhaps a signal
998 : * whose default action is to create a coredump get
999 : * out and die instead of going through with the exec.
1000 : */
1001 32168889 : ret = mmap_read_lock_killable(old_mm);
1002 32161528 : if (ret) {
1003 0 : up_write(&tsk->signal->exec_update_lock);
1004 0 : return ret;
1005 : }
1006 : }
1007 :
1008 32407584 : task_lock(tsk);
1009 32415012 : membarrier_exec_mmap(mm);
1010 :
1011 32403493 : local_irq_disable();
1012 32402357 : active_mm = tsk->active_mm;
1013 32402357 : tsk->active_mm = mm;
1014 32402357 : tsk->mm = mm;
1015 32402357 : mm_init_cid(mm);
1016 : /*
1017 : * This prevents preemption while active_mm is being loaded and
1018 : * it and mm are being updated, which could cause problems for
1019 : * lazy tlb mm refcounting when these are updated by context
1020 : * switches. Not all architectures can handle irqs off over
1021 : * activate_mm yet.
1022 : */
1023 32402357 : if (!IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1024 32402357 : local_irq_enable();
1025 32401868 : activate_mm(active_mm, mm);
1026 32412458 : if (IS_ENABLED(CONFIG_ARCH_WANT_IRQS_OFF_ACTIVATE_MM))
1027 32412458 : local_irq_enable();
1028 32412458 : lru_gen_add_mm(mm);
1029 32429440 : task_unlock(tsk);
1030 32429389 : lru_gen_use_mm(mm);
1031 32429389 : if (old_mm) {
1032 32183323 : mmap_read_unlock(old_mm);
1033 32180558 : BUG_ON(active_mm != old_mm);
1034 32180558 : setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
1035 32177801 : mm_update_next_owner(old_mm);
1036 32171760 : mmput(old_mm);
1037 32171760 : return 0;
1038 : }
1039 246066 : mmdrop_lazy_tlb(active_mm);
1040 246066 : return 0;
1041 : }
1042 :
1043 32409533 : static int de_thread(struct task_struct *tsk)
1044 : {
1045 32409533 : struct signal_struct *sig = tsk->signal;
1046 32409533 : struct sighand_struct *oldsighand = tsk->sighand;
1047 32409533 : spinlock_t *lock = &oldsighand->siglock;
1048 :
1049 32409533 : if (thread_group_empty(tsk))
1050 32409533 : goto no_thread_group;
1051 :
1052 : /*
1053 : * Kill all other threads in the thread group.
1054 : */
1055 0 : spin_lock_irq(lock);
1056 0 : if ((sig->flags & SIGNAL_GROUP_EXIT) || sig->group_exec_task) {
1057 : /*
1058 : * Another group action in progress, just
1059 : * return so that the signal is processed.
1060 : */
1061 0 : spin_unlock_irq(lock);
1062 0 : return -EAGAIN;
1063 : }
1064 :
1065 0 : sig->group_exec_task = tsk;
1066 0 : sig->notify_count = zap_other_threads(tsk);
1067 0 : if (!thread_group_leader(tsk))
1068 0 : sig->notify_count--;
1069 :
1070 0 : while (sig->notify_count) {
1071 0 : __set_current_state(TASK_KILLABLE);
1072 0 : spin_unlock_irq(lock);
1073 0 : schedule();
1074 0 : if (__fatal_signal_pending(tsk))
1075 0 : goto killed;
1076 0 : spin_lock_irq(lock);
1077 : }
1078 0 : spin_unlock_irq(lock);
1079 :
1080 : /*
1081 : * At this point all other threads have exited, all we have to
1082 : * do is to wait for the thread group leader to become inactive,
1083 : * and to assume its PID:
1084 : */
1085 0 : if (!thread_group_leader(tsk)) {
1086 0 : struct task_struct *leader = tsk->group_leader;
1087 :
1088 0 : for (;;) {
1089 0 : cgroup_threadgroup_change_begin(tsk);
1090 0 : write_lock_irq(&tasklist_lock);
1091 : /*
1092 : * Do this under tasklist_lock to ensure that
1093 : * exit_notify() can't miss ->group_exec_task
1094 : */
1095 0 : sig->notify_count = -1;
1096 0 : if (likely(leader->exit_state))
1097 : break;
1098 0 : __set_current_state(TASK_KILLABLE);
1099 0 : write_unlock_irq(&tasklist_lock);
1100 0 : cgroup_threadgroup_change_end(tsk);
1101 0 : schedule();
1102 0 : if (__fatal_signal_pending(tsk))
1103 0 : goto killed;
1104 : }
1105 :
1106 : /*
1107 : * The only record we have of the real-time age of a
1108 : * process, regardless of execs it's done, is start_time.
1109 : * All the past CPU time is accumulated in signal_struct
1110 : * from sister threads now dead. But in this non-leader
1111 : * exec, nothing survives from the original leader thread,
1112 : * whose birth marks the true age of this process now.
1113 : * When we take on its identity by switching to its PID, we
1114 : * also take its birthdate (always earlier than our own).
1115 : */
1116 0 : tsk->start_time = leader->start_time;
1117 0 : tsk->start_boottime = leader->start_boottime;
1118 :
1119 0 : BUG_ON(!same_thread_group(leader, tsk));
1120 : /*
1121 : * An exec() starts a new thread group with the
1122 : * TGID of the previous thread group. Rehash the
1123 : * two threads with a switched PID, and release
1124 : * the former thread group leader:
1125 : */
1126 :
1127 : /* Become a process group leader with the old leader's pid.
1128 : * The old leader becomes a thread of the this thread group.
1129 : */
1130 0 : exchange_tids(tsk, leader);
1131 0 : transfer_pid(leader, tsk, PIDTYPE_TGID);
1132 0 : transfer_pid(leader, tsk, PIDTYPE_PGID);
1133 0 : transfer_pid(leader, tsk, PIDTYPE_SID);
1134 :
1135 0 : list_replace_rcu(&leader->tasks, &tsk->tasks);
1136 0 : list_replace_init(&leader->sibling, &tsk->sibling);
1137 :
1138 0 : tsk->group_leader = tsk;
1139 0 : leader->group_leader = tsk;
1140 :
1141 0 : tsk->exit_signal = SIGCHLD;
1142 0 : leader->exit_signal = -1;
1143 :
1144 0 : BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1145 0 : leader->exit_state = EXIT_DEAD;
1146 :
1147 : /*
1148 : * We are going to release_task()->ptrace_unlink() silently,
1149 : * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1150 : * the tracer won't block again waiting for this thread.
1151 : */
1152 0 : if (unlikely(leader->ptrace))
1153 0 : __wake_up_parent(leader, leader->parent);
1154 0 : write_unlock_irq(&tasklist_lock);
1155 0 : cgroup_threadgroup_change_end(tsk);
1156 :
1157 0 : release_task(leader);
1158 : }
1159 :
1160 0 : sig->group_exec_task = NULL;
1161 0 : sig->notify_count = 0;
1162 :
1163 32409533 : no_thread_group:
1164 : /* we have changed execution domain */
1165 32409533 : tsk->exit_signal = SIGCHLD;
1166 :
1167 32409533 : BUG_ON(!thread_group_leader(tsk));
1168 : return 0;
1169 :
1170 0 : killed:
1171 : /* protects against exit_notify() and __exit_signal() */
1172 0 : read_lock(&tasklist_lock);
1173 0 : sig->group_exec_task = NULL;
1174 0 : sig->notify_count = 0;
1175 0 : read_unlock(&tasklist_lock);
1176 0 : return -EAGAIN;
1177 : }
1178 :
1179 :
1180 : /*
1181 : * This function makes sure the current process has its own signal table,
1182 : * so that flush_signal_handlers can later reset the handlers without
1183 : * disturbing other processes. (Other processes might share the signal
1184 : * table via the CLONE_SIGHAND option to clone().)
1185 : */
1186 32413539 : static int unshare_sighand(struct task_struct *me)
1187 : {
1188 32413539 : struct sighand_struct *oldsighand = me->sighand;
1189 :
1190 32413539 : if (refcount_read(&oldsighand->count) != 1) {
1191 0 : struct sighand_struct *newsighand;
1192 : /*
1193 : * This ->sighand is shared with the CLONE_SIGHAND
1194 : * but not CLONE_THREAD task, switch to the new one.
1195 : */
1196 0 : newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1197 0 : if (!newsighand)
1198 : return -ENOMEM;
1199 :
1200 0 : refcount_set(&newsighand->count, 1);
1201 :
1202 0 : write_lock_irq(&tasklist_lock);
1203 0 : spin_lock(&oldsighand->siglock);
1204 0 : memcpy(newsighand->action, oldsighand->action,
1205 : sizeof(newsighand->action));
1206 0 : rcu_assign_pointer(me->sighand, newsighand);
1207 0 : spin_unlock(&oldsighand->siglock);
1208 0 : write_unlock_irq(&tasklist_lock);
1209 :
1210 0 : __cleanup_sighand(oldsighand);
1211 : }
1212 : return 0;
1213 : }
1214 :
1215 11356808 : char *__get_task_comm(char *buf, size_t buf_size, struct task_struct *tsk)
1216 : {
1217 11356808 : task_lock(tsk);
1218 : /* Always NUL terminated and zero-padded */
1219 11356805 : strscpy_pad(buf, tsk->comm, buf_size);
1220 11356748 : task_unlock(tsk);
1221 11356807 : return buf;
1222 : }
1223 : EXPORT_SYMBOL_GPL(__get_task_comm);
1224 :
1225 : /*
1226 : * These functions flushes out all traces of the currently running executable
1227 : * so that a new one can be started
1228 : */
1229 :
1230 32535352 : void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1231 : {
1232 32535352 : task_lock(tsk);
1233 32541160 : trace_task_rename(tsk, buf);
1234 32537456 : strscpy_pad(tsk->comm, buf, sizeof(tsk->comm));
1235 32521097 : task_unlock(tsk);
1236 32521878 : perf_event_comm(tsk, exec);
1237 32521834 : }
1238 :
1239 : /*
1240 : * Calling this is the point of no return. None of the failures will be
1241 : * seen by userspace since either the process is already taking a fatal
1242 : * signal (via de_thread() or coredump), or will have SEGV raised
1243 : * (after exec_mmap()) by search_binary_handler (see below).
1244 : */
1245 32422658 : int begin_new_exec(struct linux_binprm * bprm)
1246 : {
1247 32422658 : struct task_struct *me = current;
1248 32422658 : int retval;
1249 :
1250 : /* Once we are committed compute the creds */
1251 32422658 : retval = bprm_creds_from_file(bprm);
1252 32409881 : if (retval)
1253 : return retval;
1254 :
1255 : /*
1256 : * Ensure all future errors are fatal.
1257 : */
1258 32410389 : bprm->point_of_no_return = true;
1259 :
1260 : /*
1261 : * Make this the only thread in the thread group.
1262 : */
1263 32410389 : retval = de_thread(me);
1264 32409431 : if (retval)
1265 0 : goto out;
1266 :
1267 : /*
1268 : * Cancel any io_uring activity across execve
1269 : */
1270 32409431 : io_uring_task_cancel();
1271 :
1272 : /* Ensure the files table is not shared. */
1273 32406369 : retval = unshare_files();
1274 32416642 : if (retval)
1275 0 : goto out;
1276 :
1277 : /*
1278 : * Must be called _before_ exec_mmap() as bprm->mm is
1279 : * not visible until then. This also enables the update
1280 : * to be lockless.
1281 : */
1282 32416642 : retval = set_mm_exe_file(bprm->mm, bprm->file);
1283 32423653 : if (retval)
1284 0 : goto out;
1285 :
1286 : /* If the binary is not readable then enforce mm->dumpable=0 */
1287 32423653 : would_dump(bprm, bprm->file);
1288 32416798 : if (bprm->have_execfd)
1289 0 : would_dump(bprm, bprm->executable);
1290 :
1291 : /*
1292 : * Release all of the old mmap stuff
1293 : */
1294 32416798 : acct_arg_size(bprm, 0);
1295 32417421 : retval = exec_mmap(bprm->mm);
1296 32426813 : if (retval)
1297 0 : goto out;
1298 :
1299 32426813 : bprm->mm = NULL;
1300 :
1301 32426813 : retval = exec_task_namespaces();
1302 32418394 : if (retval)
1303 0 : goto out_unlock;
1304 :
1305 : #ifdef CONFIG_POSIX_TIMERS
1306 32418394 : spin_lock_irq(&me->sighand->siglock);
1307 32426345 : posix_cpu_timers_exit(me);
1308 32422095 : spin_unlock_irq(&me->sighand->siglock);
1309 32422801 : exit_itimers(me);
1310 32417844 : flush_itimer_signals();
1311 : #endif
1312 :
1313 : /*
1314 : * Make the signal table private.
1315 : */
1316 32416532 : retval = unshare_sighand(me);
1317 32413817 : if (retval)
1318 0 : goto out_unlock;
1319 :
1320 32413817 : me->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC |
1321 : PF_NOFREEZE | PF_NO_SETAFFINITY);
1322 32413817 : flush_thread();
1323 32413728 : me->personality &= ~bprm->per_clear;
1324 :
1325 32413728 : clear_syscall_work_syscall_user_dispatch(me);
1326 :
1327 : /*
1328 : * We have to apply CLOEXEC before we change whether the process is
1329 : * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1330 : * trying to access the should-be-closed file descriptors of a process
1331 : * undergoing exec(2).
1332 : */
1333 32426224 : do_close_on_exec(me->files);
1334 :
1335 32418192 : if (bprm->secureexec) {
1336 : /* Make sure parent cannot signal privileged process. */
1337 2128 : me->pdeath_signal = 0;
1338 :
1339 : /*
1340 : * For secureexec, reset the stack limit to sane default to
1341 : * avoid bad behavior from the prior rlimits. This has to
1342 : * happen before arch_pick_mmap_layout(), which examines
1343 : * RLIMIT_STACK, but after the point of no return to avoid
1344 : * needing to clean up the change on failure.
1345 : */
1346 2128 : if (bprm->rlim_stack.rlim_cur > _STK_LIM)
1347 0 : bprm->rlim_stack.rlim_cur = _STK_LIM;
1348 : }
1349 :
1350 32418192 : me->sas_ss_sp = me->sas_ss_size = 0;
1351 :
1352 : /*
1353 : * Figure out dumpability. Note that this checking only of current
1354 : * is wrong, but userspace depends on it. This should be testing
1355 : * bprm->secureexec instead.
1356 : */
1357 32418192 : if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP ||
1358 32418089 : !(uid_eq(current_euid(), current_uid()) &&
1359 32410804 : gid_eq(current_egid(), current_gid())))
1360 8857 : set_dumpable(current->mm, suid_dumpable);
1361 : else
1362 32409335 : set_dumpable(current->mm, SUID_DUMP_USER);
1363 :
1364 32424441 : perf_event_exec();
1365 32401938 : __set_task_comm(me, kbasename(bprm->filename), true);
1366 :
1367 : /* An exec changes our domain. We are no longer part of the thread
1368 : group */
1369 32399193 : WRITE_ONCE(me->self_exec_id, me->self_exec_id + 1);
1370 32399193 : flush_signal_handlers(me, 0);
1371 :
1372 32419401 : retval = set_cred_ucounts(bprm->cred);
1373 32412934 : if (retval < 0)
1374 0 : goto out_unlock;
1375 :
1376 : /*
1377 : * install the new credentials for this executable
1378 : */
1379 32412934 : security_bprm_committing_creds(bprm);
1380 :
1381 32412934 : commit_creds(bprm->cred);
1382 32406094 : bprm->cred = NULL;
1383 :
1384 : /*
1385 : * Disable monitoring for regular users
1386 : * when executing setuid binaries. Must
1387 : * wait until new credentials are committed
1388 : * by commit_creds() above
1389 : */
1390 32406094 : if (get_dumpable(me->mm) != SUID_DUMP_USER)
1391 2166 : perf_event_exit_task(me);
1392 : /*
1393 : * cred_guard_mutex must be held at least to this point to prevent
1394 : * ptrace_attach() from altering our determination of the task's
1395 : * credentials; any time after this it may be unlocked.
1396 : */
1397 32406094 : security_bprm_committed_creds(bprm);
1398 :
1399 : /* Pass the opened binary to the interpreter. */
1400 32406094 : if (bprm->have_execfd) {
1401 0 : retval = get_unused_fd_flags(0);
1402 0 : if (retval < 0)
1403 0 : goto out_unlock;
1404 0 : fd_install(retval, bprm->executable);
1405 0 : bprm->executable = NULL;
1406 0 : bprm->execfd = retval;
1407 : }
1408 : return 0;
1409 :
1410 0 : out_unlock:
1411 0 : up_write(&me->signal->exec_update_lock);
1412 : out:
1413 : return retval;
1414 : }
1415 : EXPORT_SYMBOL(begin_new_exec);
1416 :
1417 64844673 : void would_dump(struct linux_binprm *bprm, struct file *file)
1418 : {
1419 64844673 : struct inode *inode = file_inode(file);
1420 64844673 : struct mnt_idmap *idmap = file_mnt_idmap(file);
1421 64838220 : if (inode_permission(idmap, inode, MAY_READ) < 0) {
1422 103 : struct user_namespace *old, *user_ns;
1423 103 : bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1424 :
1425 : /* Ensure mm->user_ns contains the executable */
1426 103 : user_ns = old = bprm->mm->user_ns;
1427 181 : while ((user_ns != &init_user_ns) &&
1428 52 : !privileged_wrt_inode_uidgid(user_ns, idmap, inode))
1429 26 : user_ns = user_ns->parent;
1430 :
1431 103 : if (old != user_ns) {
1432 26 : bprm->mm->user_ns = get_user_ns(user_ns);
1433 26 : put_user_ns(old);
1434 : }
1435 : }
1436 64836422 : }
1437 : EXPORT_SYMBOL(would_dump);
1438 :
1439 32405592 : void setup_new_exec(struct linux_binprm * bprm)
1440 : {
1441 : /* Setup things that can depend upon the personality */
1442 32405592 : struct task_struct *me = current;
1443 :
1444 32405592 : arch_pick_mmap_layout(me->mm, &bprm->rlim_stack);
1445 :
1446 32408337 : arch_setup_new_exec();
1447 :
1448 : /* Set the new mm task size. We have to do that late because it may
1449 : * depend on TIF_32BIT which is only updated in flush_thread() on
1450 : * some architectures like powerpc
1451 : */
1452 32408050 : me->mm->task_size = TASK_SIZE;
1453 32399720 : up_write(&me->signal->exec_update_lock);
1454 32414189 : mutex_unlock(&me->signal->cred_guard_mutex);
1455 32419187 : }
1456 : EXPORT_SYMBOL(setup_new_exec);
1457 :
1458 : /* Runs immediately before start_thread() takes over. */
1459 32416739 : void finalize_exec(struct linux_binprm *bprm)
1460 : {
1461 : /* Store any stack rlimit changes before starting thread. */
1462 32416739 : task_lock(current->group_leader);
1463 32418642 : current->signal->rlim[RLIMIT_STACK] = bprm->rlim_stack;
1464 32418642 : task_unlock(current->group_leader);
1465 32424321 : }
1466 : EXPORT_SYMBOL(finalize_exec);
1467 :
1468 : /*
1469 : * Prepare credentials and lock ->cred_guard_mutex.
1470 : * setup_new_exec() commits the new creds and drops the lock.
1471 : * Or, if exec fails before, free_bprm() should release ->cred
1472 : * and unlock.
1473 : */
1474 35850513 : static int prepare_bprm_creds(struct linux_binprm *bprm)
1475 : {
1476 35850513 : if (mutex_lock_interruptible(¤t->signal->cred_guard_mutex))
1477 : return -ERESTARTNOINTR;
1478 :
1479 35857383 : bprm->cred = prepare_exec_creds();
1480 35854731 : if (likely(bprm->cred))
1481 : return 0;
1482 :
1483 0 : mutex_unlock(¤t->signal->cred_guard_mutex);
1484 0 : return -ENOMEM;
1485 : }
1486 :
1487 35828180 : static void free_bprm(struct linux_binprm *bprm)
1488 : {
1489 35828180 : if (bprm->mm) {
1490 3430864 : acct_arg_size(bprm, 0);
1491 3430865 : mmput(bprm->mm);
1492 : }
1493 35835195 : free_arg_pages(bprm);
1494 35835195 : if (bprm->cred) {
1495 3430771 : mutex_unlock(¤t->signal->cred_guard_mutex);
1496 3430771 : abort_creds(bprm->cred);
1497 : }
1498 35833836 : if (bprm->file) {
1499 32403581 : allow_write_access(bprm->file);
1500 32414463 : fput(bprm->file);
1501 : }
1502 35855321 : if (bprm->executable)
1503 0 : fput(bprm->executable);
1504 : /* If a binfmt changed the interp, free it. */
1505 35853357 : if (bprm->interp != bprm->filename)
1506 74844 : kfree(bprm->interp);
1507 35853340 : kfree(bprm->fdpath);
1508 35853117 : kfree(bprm);
1509 35852744 : }
1510 :
1511 35853096 : static struct linux_binprm *alloc_bprm(int fd, struct filename *filename)
1512 : {
1513 35853096 : struct linux_binprm *bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1514 35857655 : int retval = -ENOMEM;
1515 35857655 : if (!bprm)
1516 0 : goto out;
1517 :
1518 35857655 : if (fd == AT_FDCWD || filename->name[0] == '/') {
1519 35857551 : bprm->filename = filename->name;
1520 : } else {
1521 104 : if (filename->name[0] == '\0')
1522 0 : bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d", fd);
1523 : else
1524 104 : bprm->fdpath = kasprintf(GFP_KERNEL, "/dev/fd/%d/%s",
1525 : fd, filename->name);
1526 104 : if (!bprm->fdpath)
1527 0 : goto out_free;
1528 :
1529 104 : bprm->filename = bprm->fdpath;
1530 : }
1531 35857655 : bprm->interp = bprm->filename;
1532 :
1533 35857655 : retval = bprm_mm_init(bprm);
1534 35844686 : if (retval)
1535 0 : goto out_free;
1536 : return bprm;
1537 :
1538 0 : out_free:
1539 0 : free_bprm(bprm);
1540 0 : out:
1541 0 : return ERR_PTR(retval);
1542 : }
1543 :
1544 74782 : int bprm_change_interp(const char *interp, struct linux_binprm *bprm)
1545 : {
1546 : /* If a binfmt changed the interp, free it first. */
1547 74782 : if (bprm->interp != bprm->filename)
1548 0 : kfree(bprm->interp);
1549 74782 : bprm->interp = kstrdup(interp, GFP_KERNEL);
1550 74828 : if (!bprm->interp)
1551 0 : return -ENOMEM;
1552 : return 0;
1553 : }
1554 : EXPORT_SYMBOL(bprm_change_interp);
1555 :
1556 : /*
1557 : * determine how safe it is to execute the proposed program
1558 : * - the caller must hold ->cred_guard_mutex to protect against
1559 : * PTRACE_ATTACH or seccomp thread-sync
1560 : */
1561 35852914 : static void check_unsafe_exec(struct linux_binprm *bprm)
1562 : {
1563 35852914 : struct task_struct *p = current, *t;
1564 35852914 : unsigned n_fs;
1565 :
1566 35852914 : if (p->ptrace)
1567 0 : bprm->unsafe |= LSM_UNSAFE_PTRACE;
1568 :
1569 : /*
1570 : * This isn't strictly necessary, but it makes it harder for LSMs to
1571 : * mess up.
1572 : */
1573 71705828 : if (task_no_new_privs(current))
1574 172 : bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1575 :
1576 : /*
1577 : * If another task is sharing our fs, we cannot safely
1578 : * suid exec because the differently privileged task
1579 : * will be able to manipulate the current directory, etc.
1580 : * It would be nice to force an unshare instead...
1581 : */
1582 35852914 : t = p;
1583 35852914 : n_fs = 1;
1584 35852914 : spin_lock(&p->fs->lock);
1585 35856935 : rcu_read_lock();
1586 35855648 : while_each_thread(p, t) {
1587 0 : if (t->fs == p->fs)
1588 0 : n_fs++;
1589 : }
1590 35855648 : rcu_read_unlock();
1591 :
1592 35849479 : if (p->fs->users > n_fs)
1593 0 : bprm->unsafe |= LSM_UNSAFE_SHARE;
1594 : else
1595 35849479 : p->fs->in_exec = 1;
1596 35849479 : spin_unlock(&p->fs->lock);
1597 35852281 : }
1598 :
1599 32415829 : static void bprm_fill_uid(struct linux_binprm *bprm, struct file *file)
1600 : {
1601 : /* Handle suid and sgid on files */
1602 32415829 : struct mnt_idmap *idmap;
1603 32415829 : struct inode *inode = file_inode(file);
1604 32415829 : unsigned int mode;
1605 32415829 : vfsuid_t vfsuid;
1606 32415829 : vfsgid_t vfsgid;
1607 :
1608 32415829 : if (!mnt_may_suid(file->f_path.mnt))
1609 : return;
1610 :
1611 64832280 : if (task_no_new_privs(current))
1612 : return;
1613 :
1614 32416138 : mode = READ_ONCE(inode->i_mode);
1615 32416138 : if (!(mode & (S_ISUID|S_ISGID)))
1616 : return;
1617 :
1618 309092 : idmap = file_mnt_idmap(file);
1619 :
1620 : /* Be careful if suid/sgid is set */
1621 309092 : inode_lock(inode);
1622 :
1623 : /* reload atomically mode/uid/gid now that lock held */
1624 309119 : mode = inode->i_mode;
1625 309119 : vfsuid = i_uid_into_vfsuid(idmap, inode);
1626 309121 : vfsgid = i_gid_into_vfsgid(idmap, inode);
1627 309106 : inode_unlock(inode);
1628 :
1629 : /* We ignore suid/sgid if there are no mappings for them in the ns */
1630 618211 : if (!vfsuid_has_mapping(bprm->cred->user_ns, vfsuid) ||
1631 309086 : !vfsgid_has_mapping(bprm->cred->user_ns, vfsgid))
1632 26 : return;
1633 :
1634 309098 : if (mode & S_ISUID) {
1635 309091 : bprm->per_clear |= PER_CLEAR_ON_SETID;
1636 309091 : bprm->cred->euid = vfsuid_into_kuid(vfsuid);
1637 : }
1638 :
1639 309098 : if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1640 78 : bprm->per_clear |= PER_CLEAR_ON_SETID;
1641 78 : bprm->cred->egid = vfsgid_into_kgid(vfsgid);
1642 : }
1643 : }
1644 :
1645 : /*
1646 : * Compute brpm->cred based upon the final binary.
1647 : */
1648 32415524 : static int bprm_creds_from_file(struct linux_binprm *bprm)
1649 : {
1650 : /* Compute creds based on which file? */
1651 32415524 : struct file *file = bprm->execfd_creds ? bprm->executable : bprm->file;
1652 :
1653 32415524 : bprm_fill_uid(bprm, file);
1654 32417706 : return security_bprm_creds_from_file(bprm, file);
1655 : }
1656 :
1657 : /*
1658 : * Fill the binprm structure from the inode.
1659 : * Read the first BINPRM_BUF_SIZE bytes
1660 : *
1661 : * This may be called multiple times for binary chains (scripts for example).
1662 : */
1663 32492755 : static int prepare_binprm(struct linux_binprm *bprm)
1664 : {
1665 32492755 : loff_t pos = 0;
1666 :
1667 32492755 : memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1668 32492755 : return kernel_read(bprm->file, bprm->buf, BINPRM_BUF_SIZE, &pos);
1669 : }
1670 :
1671 : /*
1672 : * Arguments are '\0' separated strings found at the location bprm->p
1673 : * points to; chop off the first by relocating brpm->p to right after
1674 : * the first '\0' encountered.
1675 : */
1676 74809 : int remove_arg_zero(struct linux_binprm *bprm)
1677 : {
1678 74809 : int ret = 0;
1679 74809 : unsigned long offset;
1680 74809 : char *kaddr;
1681 74809 : struct page *page;
1682 :
1683 74809 : if (!bprm->argc)
1684 : return 0;
1685 :
1686 74890 : do {
1687 74890 : offset = bprm->p & ~PAGE_MASK;
1688 74890 : page = get_arg_page(bprm, bprm->p, 0);
1689 74865 : if (!page) {
1690 0 : ret = -EFAULT;
1691 0 : goto out;
1692 : }
1693 74865 : kaddr = kmap_local_page(page);
1694 :
1695 1726625 : for (; offset < PAGE_SIZE && kaddr[offset];
1696 1651760 : offset++, bprm->p++)
1697 : ;
1698 :
1699 74865 : kunmap_local(kaddr);
1700 74865 : put_arg_page(page);
1701 74889 : } while (offset == PAGE_SIZE);
1702 :
1703 74808 : bprm->p++;
1704 74808 : bprm->argc--;
1705 74808 : ret = 0;
1706 :
1707 : out:
1708 : return ret;
1709 : }
1710 : EXPORT_SYMBOL(remove_arg_zero);
1711 :
1712 : #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1713 : /*
1714 : * cycle the list of binary formats handler, until one recognizes the image
1715 : */
1716 32499819 : static int search_binary_handler(struct linux_binprm *bprm)
1717 : {
1718 32499819 : bool need_retry = IS_ENABLED(CONFIG_MODULES);
1719 32499819 : struct linux_binfmt *fmt;
1720 32499819 : int retval;
1721 :
1722 32499819 : retval = prepare_binprm(bprm);
1723 32490151 : if (retval < 0)
1724 : return retval;
1725 :
1726 : retval = security_bprm_check(bprm);
1727 : if (retval)
1728 : return retval;
1729 :
1730 : retval = -ENOENT;
1731 32490000 : retry:
1732 32490000 : read_lock(&binfmt_lock);
1733 64910909 : list_for_each_entry(fmt, &formats, lh) {
1734 64910775 : if (!try_module_get(fmt->module))
1735 0 : continue;
1736 64897102 : read_unlock(&binfmt_lock);
1737 :
1738 64906091 : retval = fmt->load_binary(bprm);
1739 :
1740 64895504 : read_lock(&binfmt_lock);
1741 64908985 : put_binfmt(fmt);
1742 64907055 : if (bprm->point_of_no_return || (retval != -ENOEXEC)) {
1743 32493033 : read_unlock(&binfmt_lock);
1744 32493033 : return retval;
1745 : }
1746 : }
1747 134 : read_unlock(&binfmt_lock);
1748 :
1749 134 : if (need_retry) {
1750 134 : if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1751 130 : printable(bprm->buf[2]) && printable(bprm->buf[3]))
1752 130 : return retval;
1753 4 : if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1754 4 : return retval;
1755 0 : need_retry = false;
1756 0 : goto retry;
1757 : }
1758 :
1759 : return retval;
1760 : }
1761 :
1762 : /* binfmt handlers will call back into begin_new_exec() on success. */
1763 32419009 : static int exec_binprm(struct linux_binprm *bprm)
1764 : {
1765 32419009 : pid_t old_pid, old_vpid;
1766 32419009 : int ret, depth;
1767 :
1768 : /* Need to fetch pid before load_binary changes it */
1769 32419009 : old_pid = current->pid;
1770 32419009 : rcu_read_lock();
1771 32412895 : old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1772 32422714 : rcu_read_unlock();
1773 :
1774 : /* This allows 4 levels of binfmt rewrites before failing hard. */
1775 32497524 : for (depth = 0;; depth++) {
1776 32495889 : struct file *exec;
1777 32495889 : if (depth > 5)
1778 : return -ELOOP;
1779 :
1780 32495889 : ret = search_binary_handler(bprm);
1781 32485947 : if (ret < 0)
1782 695 : return ret;
1783 32485252 : if (!bprm->interpreter)
1784 : break;
1785 :
1786 74813 : exec = bprm->file;
1787 74813 : bprm->file = bprm->interpreter;
1788 74813 : bprm->interpreter = NULL;
1789 :
1790 74813 : allow_write_access(exec);
1791 74825 : if (unlikely(bprm->have_execfd)) {
1792 0 : if (bprm->executable) {
1793 0 : fput(exec);
1794 0 : return -ENOEXEC;
1795 : }
1796 0 : bprm->executable = exec;
1797 : } else
1798 74825 : fput(exec);
1799 : }
1800 :
1801 32410439 : audit_bprm(bprm);
1802 32410439 : trace_sched_process_exec(current, old_pid, bprm);
1803 32405717 : ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1804 32405599 : proc_exec_connector(current);
1805 32405599 : return 0;
1806 : }
1807 :
1808 : /*
1809 : * sys_execve() executes a new program.
1810 : */
1811 35854810 : static int bprm_execve(struct linux_binprm *bprm,
1812 : int fd, struct filename *filename, int flags)
1813 : {
1814 35854810 : struct file *file;
1815 35854810 : int retval;
1816 :
1817 35854810 : retval = prepare_bprm_creds(bprm);
1818 35853988 : if (retval)
1819 : return retval;
1820 :
1821 : /*
1822 : * Check for unsafe execution states before exec_binprm(), which
1823 : * will call back into begin_new_exec(), into bprm_creds_from_file(),
1824 : * where setuid-ness is evaluated.
1825 : */
1826 35853956 : check_unsafe_exec(bprm);
1827 35849661 : current->in_execve = 1;
1828 35849661 : sched_mm_cid_before_execve(current);
1829 :
1830 35849661 : file = do_open_execat(fd, filename, flags);
1831 35854165 : retval = PTR_ERR(file);
1832 35854165 : if (IS_ERR(file))
1833 3430154 : goto out_unmark;
1834 :
1835 32424011 : sched_exec();
1836 :
1837 32417341 : bprm->file = file;
1838 : /*
1839 : * Record that a name derived from an O_CLOEXEC fd will be
1840 : * inaccessible after exec. This allows the code in exec to
1841 : * choose to fail when the executable is not mmaped into the
1842 : * interpreter and an open file descriptor is not passed to
1843 : * the interpreter. This makes for a better user experience
1844 : * than having the interpreter start and then immediately fail
1845 : * when it finds the executable is inaccessible.
1846 : */
1847 32417341 : if (bprm->fdpath && get_close_on_exec(fd))
1848 104 : bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1849 :
1850 : /* Set the unchanging part of bprm->cred */
1851 32417341 : retval = security_bprm_creds_for_exec(bprm);
1852 32417341 : if (retval)
1853 : goto out;
1854 :
1855 32417341 : retval = exec_binprm(bprm);
1856 32404697 : if (retval < 0)
1857 695 : goto out;
1858 :
1859 32404002 : sched_mm_cid_after_execve(current);
1860 : /* execve succeeded */
1861 32404002 : current->fs->in_exec = 0;
1862 32404002 : current->in_execve = 0;
1863 32404002 : rseq_execve(current);
1864 32404002 : user_events_execve(current);
1865 32404002 : acct_update_integrals(current);
1866 32404625 : task_numa_free(current, false);
1867 32404625 : return retval;
1868 :
1869 : out:
1870 : /*
1871 : * If past the point of no return ensure the code never
1872 : * returns to the userspace process. Use an existing fatal
1873 : * signal if present otherwise terminate the process with
1874 : * SIGSEGV.
1875 : */
1876 695 : if (bprm->point_of_no_return && !fatal_signal_pending(current))
1877 0 : force_fatal_sig(SIGSEGV);
1878 :
1879 695 : out_unmark:
1880 3430849 : sched_mm_cid_after_execve(current);
1881 3430849 : current->fs->in_exec = 0;
1882 3430849 : current->in_execve = 0;
1883 :
1884 3430849 : return retval;
1885 : }
1886 :
1887 35608482 : static int do_execveat_common(int fd, struct filename *filename,
1888 : struct user_arg_ptr argv,
1889 : struct user_arg_ptr envp,
1890 : int flags)
1891 : {
1892 35608482 : struct linux_binprm *bprm;
1893 35608482 : int retval;
1894 :
1895 35608482 : if (IS_ERR(filename))
1896 0 : return PTR_ERR(filename);
1897 :
1898 : /*
1899 : * We move the actual failure in case of RLIMIT_NPROC excess from
1900 : * set*uid() to execve() because too many poorly written programs
1901 : * don't check setuid() return code. Here we additionally recheck
1902 : * whether NPROC limit is still exceeded.
1903 : */
1904 35608482 : if ((current->flags & PF_NPROC_EXCEEDED) &&
1905 0 : is_rlimit_overlimit(current_ucounts(), UCOUNT_RLIMIT_NPROC, rlimit(RLIMIT_NPROC))) {
1906 0 : retval = -EAGAIN;
1907 0 : goto out_ret;
1908 : }
1909 :
1910 : /* We're below the limit (still or again), so we don't want to make
1911 : * further execve() calls fail. */
1912 35608482 : current->flags &= ~PF_NPROC_EXCEEDED;
1913 :
1914 35608482 : bprm = alloc_bprm(fd, filename);
1915 35602718 : if (IS_ERR(bprm)) {
1916 0 : retval = PTR_ERR(bprm);
1917 0 : goto out_ret;
1918 : }
1919 :
1920 35602718 : retval = count(argv, MAX_ARG_STRINGS);
1921 35604407 : if (retval == 0)
1922 104 : pr_warn_once("process '%s' launched '%s' with NULL argv: empty string added\n",
1923 : current->comm, bprm->filename);
1924 35604407 : if (retval < 0)
1925 2 : goto out_free;
1926 35604405 : bprm->argc = retval;
1927 :
1928 35604405 : retval = count(envp, MAX_ARG_STRINGS);
1929 35613611 : if (retval < 0)
1930 4 : goto out_free;
1931 35613607 : bprm->envc = retval;
1932 :
1933 35613607 : retval = bprm_stack_limits(bprm);
1934 35606471 : if (retval < 0)
1935 0 : goto out_free;
1936 :
1937 35606471 : retval = copy_string_kernel(bprm->filename, bprm);
1938 35597271 : if (retval < 0)
1939 5 : goto out_free;
1940 35597266 : bprm->exec = bprm->p;
1941 :
1942 35597266 : retval = copy_strings(bprm->envc, envp, bprm);
1943 35607396 : if (retval < 0)
1944 84 : goto out_free;
1945 :
1946 35607312 : retval = copy_strings(bprm->argc, argv, bprm);
1947 35612994 : if (retval < 0)
1948 2 : goto out_free;
1949 :
1950 : /*
1951 : * When argv is empty, add an empty string ("") as argv[0] to
1952 : * ensure confused userspace programs that start processing
1953 : * from argv[1] won't end up walking envp. See also
1954 : * bprm_stack_limits().
1955 : */
1956 35612992 : if (bprm->argc == 0) {
1957 104 : retval = copy_string_kernel("", bprm);
1958 104 : if (retval < 0)
1959 0 : goto out_free;
1960 104 : bprm->argc = 1;
1961 : }
1962 :
1963 35612992 : retval = bprm_execve(bprm, fd, filename, flags);
1964 35601615 : out_free:
1965 35601615 : free_bprm(bprm);
1966 :
1967 35605268 : out_ret:
1968 35605268 : putname(filename);
1969 35605268 : return retval;
1970 : }
1971 :
1972 246066 : int kernel_execve(const char *kernel_filename,
1973 : const char *const *argv, const char *const *envp)
1974 : {
1975 246066 : struct filename *filename;
1976 246066 : struct linux_binprm *bprm;
1977 246066 : int fd = AT_FDCWD;
1978 246066 : int retval;
1979 :
1980 : /* It is non-sense for kernel threads to call execve */
1981 246066 : if (WARN_ON_ONCE(current->flags & PF_KTHREAD))
1982 : return -EINVAL;
1983 :
1984 246066 : filename = getname_kernel(kernel_filename);
1985 246066 : if (IS_ERR(filename))
1986 0 : return PTR_ERR(filename);
1987 :
1988 246066 : bprm = alloc_bprm(fd, filename);
1989 246066 : if (IS_ERR(bprm)) {
1990 0 : retval = PTR_ERR(bprm);
1991 0 : goto out_ret;
1992 : }
1993 :
1994 246066 : retval = count_strings_kernel(argv);
1995 246066 : if (WARN_ON_ONCE(retval == 0))
1996 : retval = -EINVAL;
1997 246066 : if (retval < 0)
1998 0 : goto out_free;
1999 246066 : bprm->argc = retval;
2000 :
2001 246066 : retval = count_strings_kernel(envp);
2002 246066 : if (retval < 0)
2003 0 : goto out_free;
2004 246066 : bprm->envc = retval;
2005 :
2006 246066 : retval = bprm_stack_limits(bprm);
2007 246066 : if (retval < 0)
2008 0 : goto out_free;
2009 :
2010 246066 : retval = copy_string_kernel(bprm->filename, bprm);
2011 246065 : if (retval < 0)
2012 0 : goto out_free;
2013 246065 : bprm->exec = bprm->p;
2014 :
2015 246065 : retval = copy_strings_kernel(bprm->envc, envp, bprm);
2016 246066 : if (retval < 0)
2017 0 : goto out_free;
2018 :
2019 246066 : retval = copy_strings_kernel(bprm->argc, argv, bprm);
2020 246065 : if (retval < 0)
2021 0 : goto out_free;
2022 :
2023 246065 : retval = bprm_execve(bprm, fd, filename, 0);
2024 246062 : out_free:
2025 246062 : free_bprm(bprm);
2026 246064 : out_ret:
2027 246064 : putname(filename);
2028 246064 : return retval;
2029 : }
2030 :
2031 35609049 : static int do_execve(struct filename *filename,
2032 : const char __user *const __user *__argv,
2033 : const char __user *const __user *__envp)
2034 : {
2035 35609049 : struct user_arg_ptr argv = { .ptr.native = __argv };
2036 35609049 : struct user_arg_ptr envp = { .ptr.native = __envp };
2037 35609049 : return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2038 : }
2039 :
2040 104 : static int do_execveat(int fd, struct filename *filename,
2041 : const char __user *const __user *__argv,
2042 : const char __user *const __user *__envp,
2043 : int flags)
2044 : {
2045 104 : struct user_arg_ptr argv = { .ptr.native = __argv };
2046 104 : struct user_arg_ptr envp = { .ptr.native = __envp };
2047 :
2048 104 : return do_execveat_common(fd, filename, argv, envp, flags);
2049 : }
2050 :
2051 : #ifdef CONFIG_COMPAT
2052 0 : static int compat_do_execve(struct filename *filename,
2053 : const compat_uptr_t __user *__argv,
2054 : const compat_uptr_t __user *__envp)
2055 : {
2056 0 : struct user_arg_ptr argv = {
2057 : .is_compat = true,
2058 : .ptr.compat = __argv,
2059 : };
2060 0 : struct user_arg_ptr envp = {
2061 : .is_compat = true,
2062 : .ptr.compat = __envp,
2063 : };
2064 0 : return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
2065 : }
2066 :
2067 0 : static int compat_do_execveat(int fd, struct filename *filename,
2068 : const compat_uptr_t __user *__argv,
2069 : const compat_uptr_t __user *__envp,
2070 : int flags)
2071 : {
2072 0 : struct user_arg_ptr argv = {
2073 : .is_compat = true,
2074 : .ptr.compat = __argv,
2075 : };
2076 0 : struct user_arg_ptr envp = {
2077 : .is_compat = true,
2078 : .ptr.compat = __envp,
2079 : };
2080 0 : return do_execveat_common(fd, filename, argv, envp, flags);
2081 : }
2082 : #endif
2083 :
2084 32429303 : void set_binfmt(struct linux_binfmt *new)
2085 : {
2086 32429303 : struct mm_struct *mm = current->mm;
2087 :
2088 32429303 : if (mm->binfmt)
2089 0 : module_put(mm->binfmt->module);
2090 :
2091 32429295 : mm->binfmt = new;
2092 32429295 : if (new)
2093 32429295 : __module_get(new->module);
2094 32429231 : }
2095 : EXPORT_SYMBOL(set_binfmt);
2096 :
2097 : /*
2098 : * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
2099 : */
2100 32494343 : void set_dumpable(struct mm_struct *mm, int value)
2101 : {
2102 32494343 : if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
2103 : return;
2104 :
2105 32494343 : set_mask_bits(&mm->flags, MMF_DUMPABLE_MASK, value);
2106 : }
2107 :
2108 71222351 : SYSCALL_DEFINE3(execve,
2109 : const char __user *, filename,
2110 : const char __user *const __user *, argv,
2111 : const char __user *const __user *, envp)
2112 : {
2113 35610761 : return do_execve(getname(filename), argv, envp);
2114 : }
2115 :
2116 208 : SYSCALL_DEFINE5(execveat,
2117 : int, fd, const char __user *, filename,
2118 : const char __user *const __user *, argv,
2119 : const char __user *const __user *, envp,
2120 : int, flags)
2121 : {
2122 104 : return do_execveat(fd,
2123 : getname_uflags(filename, flags),
2124 : argv, envp, flags);
2125 : }
2126 :
2127 : #ifdef CONFIG_COMPAT
2128 0 : COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
2129 : const compat_uptr_t __user *, argv,
2130 : const compat_uptr_t __user *, envp)
2131 : {
2132 0 : return compat_do_execve(getname(filename), argv, envp);
2133 : }
2134 :
2135 0 : COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
2136 : const char __user *, filename,
2137 : const compat_uptr_t __user *, argv,
2138 : const compat_uptr_t __user *, envp,
2139 : int, flags)
2140 : {
2141 0 : return compat_do_execveat(fd,
2142 : getname_uflags(filename, flags),
2143 : argv, envp, flags);
2144 : }
2145 : #endif
2146 :
2147 : #ifdef CONFIG_SYSCTL
2148 :
2149 0 : static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2150 : void *buffer, size_t *lenp, loff_t *ppos)
2151 : {
2152 0 : int error = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
2153 :
2154 0 : if (!error)
2155 0 : validate_coredump_safety();
2156 0 : return error;
2157 : }
2158 :
2159 : static struct ctl_table fs_exec_sysctls[] = {
2160 : {
2161 : .procname = "suid_dumpable",
2162 : .data = &suid_dumpable,
2163 : .maxlen = sizeof(int),
2164 : .mode = 0644,
2165 : .proc_handler = proc_dointvec_minmax_coredump,
2166 : .extra1 = SYSCTL_ZERO,
2167 : .extra2 = SYSCTL_TWO,
2168 : },
2169 : { }
2170 : };
2171 :
2172 0 : static int __init init_fs_exec_sysctls(void)
2173 : {
2174 0 : register_sysctl_init("fs", fs_exec_sysctls);
2175 0 : return 0;
2176 : }
2177 :
2178 : fs_initcall(init_fs_exec_sysctls);
2179 : #endif /* CONFIG_SYSCTL */
|
