Adding breaktrace & disabling timer migration
[kvmfornfv.git] / kernel / fs / exec.c
1 /*
2  *  linux/fs/exec.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  * #!-checking implemented by tytso.
9  */
10 /*
11  * Demand-loading implemented 01.12.91 - no need to read anything but
12  * the header into memory. The inode of the executable is put into
13  * "current->executable", and page faults do the actual loading. Clean.
14  *
15  * Once more I can proudly say that linux stood up to being changed: it
16  * was less than 2 hours work to get demand-loading completely implemented.
17  *
18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
19  * current->executable is only used by the procfs.  This allows a dispatch
20  * table to check for several different types  of binary formats.  We keep
21  * trying until we recognize the file or we run out of supported binary
22  * formats.
23  */
24
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
28 #include <linux/mm.h>
29 #include <linux/vmacache.h>
30 #include <linux/stat.h>
31 #include <linux/fcntl.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/perf_event.h>
37 #include <linux/highmem.h>
38 #include <linux/spinlock.h>
39 #include <linux/key.h>
40 #include <linux/personality.h>
41 #include <linux/binfmts.h>
42 #include <linux/utsname.h>
43 #include <linux/pid_namespace.h>
44 #include <linux/module.h>
45 #include <linux/namei.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52 #include <linux/tracehook.h>
53 #include <linux/kmod.h>
54 #include <linux/fsnotify.h>
55 #include <linux/fs_struct.h>
56 #include <linux/pipe_fs_i.h>
57 #include <linux/oom.h>
58 #include <linux/compat.h>
59 #include <linux/user_namespace.h>
60
61 #include <asm/uaccess.h>
62 #include <asm/mmu_context.h>
63 #include <asm/tlb.h>
64
65 #include <trace/events/task.h>
66 #include "internal.h"
67
68 #include <trace/events/sched.h>
69
70 int suid_dumpable = 0;
71
72 static LIST_HEAD(formats);
73 static DEFINE_RWLOCK(binfmt_lock);
74
75 void __register_binfmt(struct linux_binfmt * fmt, int insert)
76 {
77         BUG_ON(!fmt);
78         if (WARN_ON(!fmt->load_binary))
79                 return;
80         write_lock(&binfmt_lock);
81         insert ? list_add(&fmt->lh, &formats) :
82                  list_add_tail(&fmt->lh, &formats);
83         write_unlock(&binfmt_lock);
84 }
85
86 EXPORT_SYMBOL(__register_binfmt);
87
88 void unregister_binfmt(struct linux_binfmt * fmt)
89 {
90         write_lock(&binfmt_lock);
91         list_del(&fmt->lh);
92         write_unlock(&binfmt_lock);
93 }
94
95 EXPORT_SYMBOL(unregister_binfmt);
96
97 static inline void put_binfmt(struct linux_binfmt * fmt)
98 {
99         module_put(fmt->module);
100 }
101
102 bool path_noexec(const struct path *path)
103 {
104         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
105                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
106 }
107
108 #ifdef CONFIG_USELIB
109 /*
110  * Note that a shared library must be both readable and executable due to
111  * security reasons.
112  *
113  * Also note that we take the address to load from from the file itself.
114  */
115 SYSCALL_DEFINE1(uselib, const char __user *, library)
116 {
117         struct linux_binfmt *fmt;
118         struct file *file;
119         struct filename *tmp = getname(library);
120         int error = PTR_ERR(tmp);
121         static const struct open_flags uselib_flags = {
122                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
123                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
124                 .intent = LOOKUP_OPEN,
125                 .lookup_flags = LOOKUP_FOLLOW,
126         };
127
128         if (IS_ERR(tmp))
129                 goto out;
130
131         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags);
132         putname(tmp);
133         error = PTR_ERR(file);
134         if (IS_ERR(file))
135                 goto out;
136
137         error = -EINVAL;
138         if (!S_ISREG(file_inode(file)->i_mode))
139                 goto exit;
140
141         error = -EACCES;
142         if (path_noexec(&file->f_path))
143                 goto exit;
144
145         fsnotify_open(file);
146
147         error = -ENOEXEC;
148
149         read_lock(&binfmt_lock);
150         list_for_each_entry(fmt, &formats, lh) {
151                 if (!fmt->load_shlib)
152                         continue;
153                 if (!try_module_get(fmt->module))
154                         continue;
155                 read_unlock(&binfmt_lock);
156                 error = fmt->load_shlib(file);
157                 read_lock(&binfmt_lock);
158                 put_binfmt(fmt);
159                 if (error != -ENOEXEC)
160                         break;
161         }
162         read_unlock(&binfmt_lock);
163 exit:
164         fput(file);
165 out:
166         return error;
167 }
168 #endif /* #ifdef CONFIG_USELIB */
169
170 #ifdef CONFIG_MMU
171 /*
172  * The nascent bprm->mm is not visible until exec_mmap() but it can
173  * use a lot of memory, account these pages in current->mm temporary
174  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
175  * change the counter back via acct_arg_size(0).
176  */
177 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
178 {
179         struct mm_struct *mm = current->mm;
180         long diff = (long)(pages - bprm->vma_pages);
181
182         if (!mm || !diff)
183                 return;
184
185         bprm->vma_pages = pages;
186         add_mm_counter(mm, MM_ANONPAGES, diff);
187 }
188
189 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
190                 int write)
191 {
192         struct page *page;
193         int ret;
194
195 #ifdef CONFIG_STACK_GROWSUP
196         if (write) {
197                 ret = expand_downwards(bprm->vma, pos);
198                 if (ret < 0)
199                         return NULL;
200         }
201 #endif
202         ret = get_user_pages(current, bprm->mm, pos,
203                         1, write, 1, &page, NULL);
204         if (ret <= 0)
205                 return NULL;
206
207         if (write) {
208                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
209                 struct rlimit *rlim;
210
211                 acct_arg_size(bprm, size / PAGE_SIZE);
212
213                 /*
214                  * We've historically supported up to 32 pages (ARG_MAX)
215                  * of argument strings even with small stacks
216                  */
217                 if (size <= ARG_MAX)
218                         return page;
219
220                 /*
221                  * Limit to 1/4-th the stack size for the argv+env strings.
222                  * This ensures that:
223                  *  - the remaining binfmt code will not run out of stack space,
224                  *  - the program will have a reasonable amount of stack left
225                  *    to work from.
226                  */
227                 rlim = current->signal->rlim;
228                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
229                         put_page(page);
230                         return NULL;
231                 }
232         }
233
234         return page;
235 }
236
237 static void put_arg_page(struct page *page)
238 {
239         put_page(page);
240 }
241
242 static void free_arg_page(struct linux_binprm *bprm, int i)
243 {
244 }
245
246 static void free_arg_pages(struct linux_binprm *bprm)
247 {
248 }
249
250 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
251                 struct page *page)
252 {
253         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
254 }
255
256 static int __bprm_mm_init(struct linux_binprm *bprm)
257 {
258         int err;
259         struct vm_area_struct *vma = NULL;
260         struct mm_struct *mm = bprm->mm;
261
262         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
263         if (!vma)
264                 return -ENOMEM;
265
266         down_write(&mm->mmap_sem);
267         vma->vm_mm = mm;
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         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
276         vma->vm_end = STACK_TOP_MAX;
277         vma->vm_start = vma->vm_end - PAGE_SIZE;
278         vma->vm_flags = VM_SOFTDIRTY | VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
279         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
280         INIT_LIST_HEAD(&vma->anon_vma_chain);
281
282         err = insert_vm_struct(mm, vma);
283         if (err)
284                 goto err;
285
286         mm->stack_vm = mm->total_vm = 1;
287         arch_bprm_mm_init(mm, vma);
288         up_write(&mm->mmap_sem);
289         bprm->p = vma->vm_end - sizeof(void *);
290         return 0;
291 err:
292         up_write(&mm->mmap_sem);
293         bprm->vma = NULL;
294         kmem_cache_free(vm_area_cachep, vma);
295         return err;
296 }
297
298 static bool valid_arg_len(struct linux_binprm *bprm, long len)
299 {
300         return len <= MAX_ARG_STRLEN;
301 }
302
303 #else
304
305 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
306 {
307 }
308
309 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
310                 int write)
311 {
312         struct page *page;
313
314         page = bprm->page[pos / PAGE_SIZE];
315         if (!page && write) {
316                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
317                 if (!page)
318                         return NULL;
319                 bprm->page[pos / PAGE_SIZE] = page;
320         }
321
322         return page;
323 }
324
325 static void put_arg_page(struct page *page)
326 {
327 }
328
329 static void free_arg_page(struct linux_binprm *bprm, int i)
330 {
331         if (bprm->page[i]) {
332                 __free_page(bprm->page[i]);
333                 bprm->page[i] = NULL;
334         }
335 }
336
337 static void free_arg_pages(struct linux_binprm *bprm)
338 {
339         int i;
340
341         for (i = 0; i < MAX_ARG_PAGES; i++)
342                 free_arg_page(bprm, i);
343 }
344
345 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
346                 struct page *page)
347 {
348 }
349
350 static int __bprm_mm_init(struct linux_binprm *bprm)
351 {
352         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
353         return 0;
354 }
355
356 static bool valid_arg_len(struct linux_binprm *bprm, long len)
357 {
358         return len <= bprm->p;
359 }
360
361 #endif /* CONFIG_MMU */
362
363 /*
364  * Create a new mm_struct and populate it with a temporary stack
365  * vm_area_struct.  We don't have enough context at this point to set the stack
366  * flags, permissions, and offset, so we use temporary values.  We'll update
367  * them later in setup_arg_pages().
368  */
369 static int bprm_mm_init(struct linux_binprm *bprm)
370 {
371         int err;
372         struct mm_struct *mm = NULL;
373
374         bprm->mm = mm = mm_alloc();
375         err = -ENOMEM;
376         if (!mm)
377                 goto err;
378
379         err = __bprm_mm_init(bprm);
380         if (err)
381                 goto err;
382
383         return 0;
384
385 err:
386         if (mm) {
387                 bprm->mm = NULL;
388                 mmdrop(mm);
389         }
390
391         return err;
392 }
393
394 struct user_arg_ptr {
395 #ifdef CONFIG_COMPAT
396         bool is_compat;
397 #endif
398         union {
399                 const char __user *const __user *native;
400 #ifdef CONFIG_COMPAT
401                 const compat_uptr_t __user *compat;
402 #endif
403         } ptr;
404 };
405
406 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
407 {
408         const char __user *native;
409
410 #ifdef CONFIG_COMPAT
411         if (unlikely(argv.is_compat)) {
412                 compat_uptr_t compat;
413
414                 if (get_user(compat, argv.ptr.compat + nr))
415                         return ERR_PTR(-EFAULT);
416
417                 return compat_ptr(compat);
418         }
419 #endif
420
421         if (get_user(native, argv.ptr.native + nr))
422                 return ERR_PTR(-EFAULT);
423
424         return native;
425 }
426
427 /*
428  * count() counts the number of strings in array ARGV.
429  */
430 static int count(struct user_arg_ptr argv, int max)
431 {
432         int i = 0;
433
434         if (argv.ptr.native != NULL) {
435                 for (;;) {
436                         const char __user *p = get_user_arg_ptr(argv, i);
437
438                         if (!p)
439                                 break;
440
441                         if (IS_ERR(p))
442                                 return -EFAULT;
443
444                         if (i >= max)
445                                 return -E2BIG;
446                         ++i;
447
448                         if (fatal_signal_pending(current))
449                                 return -ERESTARTNOHAND;
450                         cond_resched();
451                 }
452         }
453         return i;
454 }
455
456 /*
457  * 'copy_strings()' copies argument/environment strings from the old
458  * processes's memory to the new process's stack.  The call to get_user_pages()
459  * ensures the destination page is created and not swapped out.
460  */
461 static int copy_strings(int argc, struct user_arg_ptr argv,
462                         struct linux_binprm *bprm)
463 {
464         struct page *kmapped_page = NULL;
465         char *kaddr = NULL;
466         unsigned long kpos = 0;
467         int ret;
468
469         while (argc-- > 0) {
470                 const char __user *str;
471                 int len;
472                 unsigned long pos;
473
474                 ret = -EFAULT;
475                 str = get_user_arg_ptr(argv, argc);
476                 if (IS_ERR(str))
477                         goto out;
478
479                 len = strnlen_user(str, MAX_ARG_STRLEN);
480                 if (!len)
481                         goto out;
482
483                 ret = -E2BIG;
484                 if (!valid_arg_len(bprm, len))
485                         goto out;
486
487                 /* We're going to work our way backwords. */
488                 pos = bprm->p;
489                 str += len;
490                 bprm->p -= len;
491
492                 while (len > 0) {
493                         int offset, bytes_to_copy;
494
495                         if (fatal_signal_pending(current)) {
496                                 ret = -ERESTARTNOHAND;
497                                 goto out;
498                         }
499                         cond_resched();
500
501                         offset = pos % PAGE_SIZE;
502                         if (offset == 0)
503                                 offset = PAGE_SIZE;
504
505                         bytes_to_copy = offset;
506                         if (bytes_to_copy > len)
507                                 bytes_to_copy = len;
508
509                         offset -= bytes_to_copy;
510                         pos -= bytes_to_copy;
511                         str -= bytes_to_copy;
512                         len -= bytes_to_copy;
513
514                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
515                                 struct page *page;
516
517                                 page = get_arg_page(bprm, pos, 1);
518                                 if (!page) {
519                                         ret = -E2BIG;
520                                         goto out;
521                                 }
522
523                                 if (kmapped_page) {
524                                         flush_kernel_dcache_page(kmapped_page);
525                                         kunmap(kmapped_page);
526                                         put_arg_page(kmapped_page);
527                                 }
528                                 kmapped_page = page;
529                                 kaddr = kmap(kmapped_page);
530                                 kpos = pos & PAGE_MASK;
531                                 flush_arg_page(bprm, kpos, kmapped_page);
532                         }
533                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
534                                 ret = -EFAULT;
535                                 goto out;
536                         }
537                 }
538         }
539         ret = 0;
540 out:
541         if (kmapped_page) {
542                 flush_kernel_dcache_page(kmapped_page);
543                 kunmap(kmapped_page);
544                 put_arg_page(kmapped_page);
545         }
546         return ret;
547 }
548
549 /*
550  * Like copy_strings, but get argv and its values from kernel memory.
551  */
552 int copy_strings_kernel(int argc, const char *const *__argv,
553                         struct linux_binprm *bprm)
554 {
555         int r;
556         mm_segment_t oldfs = get_fs();
557         struct user_arg_ptr argv = {
558                 .ptr.native = (const char __user *const  __user *)__argv,
559         };
560
561         set_fs(KERNEL_DS);
562         r = copy_strings(argc, argv, bprm);
563         set_fs(oldfs);
564
565         return r;
566 }
567 EXPORT_SYMBOL(copy_strings_kernel);
568
569 #ifdef CONFIG_MMU
570
571 /*
572  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
573  * the binfmt code determines where the new stack should reside, we shift it to
574  * its final location.  The process proceeds as follows:
575  *
576  * 1) Use shift to calculate the new vma endpoints.
577  * 2) Extend vma to cover both the old and new ranges.  This ensures the
578  *    arguments passed to subsequent functions are consistent.
579  * 3) Move vma's page tables to the new range.
580  * 4) Free up any cleared pgd range.
581  * 5) Shrink the vma to cover only the new range.
582  */
583 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
584 {
585         struct mm_struct *mm = vma->vm_mm;
586         unsigned long old_start = vma->vm_start;
587         unsigned long old_end = vma->vm_end;
588         unsigned long length = old_end - old_start;
589         unsigned long new_start = old_start - shift;
590         unsigned long new_end = old_end - shift;
591         struct mmu_gather tlb;
592
593         BUG_ON(new_start > new_end);
594
595         /*
596          * ensure there are no vmas between where we want to go
597          * and where we are
598          */
599         if (vma != find_vma(mm, new_start))
600                 return -EFAULT;
601
602         /*
603          * cover the whole range: [new_start, old_end)
604          */
605         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
606                 return -ENOMEM;
607
608         /*
609          * move the page tables downwards, on failure we rely on
610          * process cleanup to remove whatever mess we made.
611          */
612         if (length != move_page_tables(vma, old_start,
613                                        vma, new_start, length, false))
614                 return -ENOMEM;
615
616         lru_add_drain();
617         tlb_gather_mmu(&tlb, mm, old_start, old_end);
618         if (new_end > old_start) {
619                 /*
620                  * when the old and new regions overlap clear from new_end.
621                  */
622                 free_pgd_range(&tlb, new_end, old_end, new_end,
623                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
624         } else {
625                 /*
626                  * otherwise, clean from old_start; this is done to not touch
627                  * the address space in [new_end, old_start) some architectures
628                  * have constraints on va-space that make this illegal (IA64) -
629                  * for the others its just a little faster.
630                  */
631                 free_pgd_range(&tlb, old_start, old_end, new_end,
632                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
633         }
634         tlb_finish_mmu(&tlb, old_start, old_end);
635
636         /*
637          * Shrink the vma to just the new range.  Always succeeds.
638          */
639         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
640
641         return 0;
642 }
643
644 /*
645  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
646  * the stack is optionally relocated, and some extra space is added.
647  */
648 int setup_arg_pages(struct linux_binprm *bprm,
649                     unsigned long stack_top,
650                     int executable_stack)
651 {
652         unsigned long ret;
653         unsigned long stack_shift;
654         struct mm_struct *mm = current->mm;
655         struct vm_area_struct *vma = bprm->vma;
656         struct vm_area_struct *prev = NULL;
657         unsigned long vm_flags;
658         unsigned long stack_base;
659         unsigned long stack_size;
660         unsigned long stack_expand;
661         unsigned long rlim_stack;
662
663 #ifdef CONFIG_STACK_GROWSUP
664         /* Limit stack size */
665         stack_base = rlimit_max(RLIMIT_STACK);
666         if (stack_base > STACK_SIZE_MAX)
667                 stack_base = STACK_SIZE_MAX;
668
669         /* Add space for stack randomization. */
670         stack_base += (STACK_RND_MASK << PAGE_SHIFT);
671
672         /* Make sure we didn't let the argument array grow too large. */
673         if (vma->vm_end - vma->vm_start > stack_base)
674                 return -ENOMEM;
675
676         stack_base = PAGE_ALIGN(stack_top - stack_base);
677
678         stack_shift = vma->vm_start - stack_base;
679         mm->arg_start = bprm->p - stack_shift;
680         bprm->p = vma->vm_end - stack_shift;
681 #else
682         stack_top = arch_align_stack(stack_top);
683         stack_top = PAGE_ALIGN(stack_top);
684
685         if (unlikely(stack_top < mmap_min_addr) ||
686             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
687                 return -ENOMEM;
688
689         stack_shift = vma->vm_end - stack_top;
690
691         bprm->p -= stack_shift;
692         mm->arg_start = bprm->p;
693 #endif
694
695         if (bprm->loader)
696                 bprm->loader -= stack_shift;
697         bprm->exec -= stack_shift;
698
699         down_write(&mm->mmap_sem);
700         vm_flags = VM_STACK_FLAGS;
701
702         /*
703          * Adjust stack execute permissions; explicitly enable for
704          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
705          * (arch default) otherwise.
706          */
707         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
708                 vm_flags |= VM_EXEC;
709         else if (executable_stack == EXSTACK_DISABLE_X)
710                 vm_flags &= ~VM_EXEC;
711         vm_flags |= mm->def_flags;
712         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
713
714         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
715                         vm_flags);
716         if (ret)
717                 goto out_unlock;
718         BUG_ON(prev != vma);
719
720         /* Move stack pages down in memory. */
721         if (stack_shift) {
722                 ret = shift_arg_pages(vma, stack_shift);
723                 if (ret)
724                         goto out_unlock;
725         }
726
727         /* mprotect_fixup is overkill to remove the temporary stack flags */
728         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
729
730         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
731         stack_size = vma->vm_end - vma->vm_start;
732         /*
733          * Align this down to a page boundary as expand_stack
734          * will align it up.
735          */
736         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
737 #ifdef CONFIG_STACK_GROWSUP
738         if (stack_size + stack_expand > rlim_stack)
739                 stack_base = vma->vm_start + rlim_stack;
740         else
741                 stack_base = vma->vm_end + stack_expand;
742 #else
743         if (stack_size + stack_expand > rlim_stack)
744                 stack_base = vma->vm_end - rlim_stack;
745         else
746                 stack_base = vma->vm_start - stack_expand;
747 #endif
748         current->mm->start_stack = bprm->p;
749         ret = expand_stack(vma, stack_base);
750         if (ret)
751                 ret = -EFAULT;
752
753 out_unlock:
754         up_write(&mm->mmap_sem);
755         return ret;
756 }
757 EXPORT_SYMBOL(setup_arg_pages);
758
759 #endif /* CONFIG_MMU */
760
761 static struct file *do_open_execat(int fd, struct filename *name, int flags)
762 {
763         struct file *file;
764         int err;
765         struct open_flags open_exec_flags = {
766                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
767                 .acc_mode = MAY_EXEC | MAY_OPEN,
768                 .intent = LOOKUP_OPEN,
769                 .lookup_flags = LOOKUP_FOLLOW,
770         };
771
772         if ((flags & ~(AT_SYMLINK_NOFOLLOW | AT_EMPTY_PATH)) != 0)
773                 return ERR_PTR(-EINVAL);
774         if (flags & AT_SYMLINK_NOFOLLOW)
775                 open_exec_flags.lookup_flags &= ~LOOKUP_FOLLOW;
776         if (flags & AT_EMPTY_PATH)
777                 open_exec_flags.lookup_flags |= LOOKUP_EMPTY;
778
779         file = do_filp_open(fd, name, &open_exec_flags);
780         if (IS_ERR(file))
781                 goto out;
782
783         err = -EACCES;
784         if (!S_ISREG(file_inode(file)->i_mode))
785                 goto exit;
786
787         if (path_noexec(&file->f_path))
788                 goto exit;
789
790         err = deny_write_access(file);
791         if (err)
792                 goto exit;
793
794         if (name->name[0] != '\0')
795                 fsnotify_open(file);
796
797 out:
798         return file;
799
800 exit:
801         fput(file);
802         return ERR_PTR(err);
803 }
804
805 struct file *open_exec(const char *name)
806 {
807         struct filename *filename = getname_kernel(name);
808         struct file *f = ERR_CAST(filename);
809
810         if (!IS_ERR(filename)) {
811                 f = do_open_execat(AT_FDCWD, filename, 0);
812                 putname(filename);
813         }
814         return f;
815 }
816 EXPORT_SYMBOL(open_exec);
817
818 int kernel_read(struct file *file, loff_t offset,
819                 char *addr, unsigned long count)
820 {
821         mm_segment_t old_fs;
822         loff_t pos = offset;
823         int result;
824
825         old_fs = get_fs();
826         set_fs(get_ds());
827         /* The cast to a user pointer is valid due to the set_fs() */
828         result = vfs_read(file, (void __user *)addr, count, &pos);
829         set_fs(old_fs);
830         return result;
831 }
832
833 EXPORT_SYMBOL(kernel_read);
834
835 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
836 {
837         ssize_t res = vfs_read(file, (void __user *)addr, len, &pos);
838         if (res > 0)
839                 flush_icache_range(addr, addr + len);
840         return res;
841 }
842 EXPORT_SYMBOL(read_code);
843
844 static int exec_mmap(struct mm_struct *mm)
845 {
846         struct task_struct *tsk;
847         struct mm_struct *old_mm, *active_mm;
848
849         /* Notify parent that we're no longer interested in the old VM */
850         tsk = current;
851         old_mm = current->mm;
852         mm_release(tsk, old_mm);
853
854         if (old_mm) {
855                 sync_mm_rss(old_mm);
856                 /*
857                  * Make sure that if there is a core dump in progress
858                  * for the old mm, we get out and die instead of going
859                  * through with the exec.  We must hold mmap_sem around
860                  * checking core_state and changing tsk->mm.
861                  */
862                 down_read(&old_mm->mmap_sem);
863                 if (unlikely(old_mm->core_state)) {
864                         up_read(&old_mm->mmap_sem);
865                         return -EINTR;
866                 }
867         }
868         task_lock(tsk);
869         preempt_disable_rt();
870         active_mm = tsk->active_mm;
871         tsk->mm = mm;
872         tsk->active_mm = mm;
873         activate_mm(active_mm, mm);
874         tsk->mm->vmacache_seqnum = 0;
875         vmacache_flush(tsk);
876         preempt_enable_rt();
877         task_unlock(tsk);
878         if (old_mm) {
879                 up_read(&old_mm->mmap_sem);
880                 BUG_ON(active_mm != old_mm);
881                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
882                 mm_update_next_owner(old_mm);
883                 mmput(old_mm);
884                 return 0;
885         }
886         mmdrop(active_mm);
887         return 0;
888 }
889
890 /*
891  * This function makes sure the current process has its own signal table,
892  * so that flush_signal_handlers can later reset the handlers without
893  * disturbing other processes.  (Other processes might share the signal
894  * table via the CLONE_SIGHAND option to clone().)
895  */
896 static int de_thread(struct task_struct *tsk)
897 {
898         struct signal_struct *sig = tsk->signal;
899         struct sighand_struct *oldsighand = tsk->sighand;
900         spinlock_t *lock = &oldsighand->siglock;
901
902         if (thread_group_empty(tsk))
903                 goto no_thread_group;
904
905         /*
906          * Kill all other threads in the thread group.
907          */
908         spin_lock_irq(lock);
909         if (signal_group_exit(sig)) {
910                 /*
911                  * Another group action in progress, just
912                  * return so that the signal is processed.
913                  */
914                 spin_unlock_irq(lock);
915                 return -EAGAIN;
916         }
917
918         sig->group_exit_task = tsk;
919         sig->notify_count = zap_other_threads(tsk);
920         if (!thread_group_leader(tsk))
921                 sig->notify_count--;
922
923         while (sig->notify_count) {
924                 __set_current_state(TASK_KILLABLE);
925                 spin_unlock_irq(lock);
926                 schedule();
927                 if (unlikely(__fatal_signal_pending(tsk)))
928                         goto killed;
929                 spin_lock_irq(lock);
930         }
931         spin_unlock_irq(lock);
932
933         /*
934          * At this point all other threads have exited, all we have to
935          * do is to wait for the thread group leader to become inactive,
936          * and to assume its PID:
937          */
938         if (!thread_group_leader(tsk)) {
939                 struct task_struct *leader = tsk->group_leader;
940
941                 for (;;) {
942                         threadgroup_change_begin(tsk);
943                         write_lock_irq(&tasklist_lock);
944                         /*
945                          * Do this under tasklist_lock to ensure that
946                          * exit_notify() can't miss ->group_exit_task
947                          */
948                         sig->notify_count = -1;
949                         if (likely(leader->exit_state))
950                                 break;
951                         __set_current_state(TASK_KILLABLE);
952                         write_unlock_irq(&tasklist_lock);
953                         threadgroup_change_end(tsk);
954                         schedule();
955                         if (unlikely(__fatal_signal_pending(tsk)))
956                                 goto killed;
957                 }
958
959                 /*
960                  * The only record we have of the real-time age of a
961                  * process, regardless of execs it's done, is start_time.
962                  * All the past CPU time is accumulated in signal_struct
963                  * from sister threads now dead.  But in this non-leader
964                  * exec, nothing survives from the original leader thread,
965                  * whose birth marks the true age of this process now.
966                  * When we take on its identity by switching to its PID, we
967                  * also take its birthdate (always earlier than our own).
968                  */
969                 tsk->start_time = leader->start_time;
970                 tsk->real_start_time = leader->real_start_time;
971
972                 BUG_ON(!same_thread_group(leader, tsk));
973                 BUG_ON(has_group_leader_pid(tsk));
974                 /*
975                  * An exec() starts a new thread group with the
976                  * TGID of the previous thread group. Rehash the
977                  * two threads with a switched PID, and release
978                  * the former thread group leader:
979                  */
980
981                 /* Become a process group leader with the old leader's pid.
982                  * The old leader becomes a thread of the this thread group.
983                  * Note: The old leader also uses this pid until release_task
984                  *       is called.  Odd but simple and correct.
985                  */
986                 tsk->pid = leader->pid;
987                 change_pid(tsk, PIDTYPE_PID, task_pid(leader));
988                 transfer_pid(leader, tsk, PIDTYPE_PGID);
989                 transfer_pid(leader, tsk, PIDTYPE_SID);
990
991                 list_replace_rcu(&leader->tasks, &tsk->tasks);
992                 list_replace_init(&leader->sibling, &tsk->sibling);
993
994                 tsk->group_leader = tsk;
995                 leader->group_leader = tsk;
996
997                 tsk->exit_signal = SIGCHLD;
998                 leader->exit_signal = -1;
999
1000                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
1001                 leader->exit_state = EXIT_DEAD;
1002
1003                 /*
1004                  * We are going to release_task()->ptrace_unlink() silently,
1005                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
1006                  * the tracer wont't block again waiting for this thread.
1007                  */
1008                 if (unlikely(leader->ptrace))
1009                         __wake_up_parent(leader, leader->parent);
1010                 write_unlock_irq(&tasklist_lock);
1011                 threadgroup_change_end(tsk);
1012
1013                 release_task(leader);
1014         }
1015
1016         sig->group_exit_task = NULL;
1017         sig->notify_count = 0;
1018
1019 no_thread_group:
1020         /* we have changed execution domain */
1021         tsk->exit_signal = SIGCHLD;
1022
1023         exit_itimers(sig);
1024         flush_itimer_signals();
1025
1026         if (atomic_read(&oldsighand->count) != 1) {
1027                 struct sighand_struct *newsighand;
1028                 /*
1029                  * This ->sighand is shared with the CLONE_SIGHAND
1030                  * but not CLONE_THREAD task, switch to the new one.
1031                  */
1032                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1033                 if (!newsighand)
1034                         return -ENOMEM;
1035
1036                 atomic_set(&newsighand->count, 1);
1037                 memcpy(newsighand->action, oldsighand->action,
1038                        sizeof(newsighand->action));
1039
1040                 write_lock_irq(&tasklist_lock);
1041                 spin_lock(&oldsighand->siglock);
1042                 rcu_assign_pointer(tsk->sighand, newsighand);
1043                 spin_unlock(&oldsighand->siglock);
1044                 write_unlock_irq(&tasklist_lock);
1045
1046                 __cleanup_sighand(oldsighand);
1047         }
1048
1049         BUG_ON(!thread_group_leader(tsk));
1050         return 0;
1051
1052 killed:
1053         /* protects against exit_notify() and __exit_signal() */
1054         read_lock(&tasklist_lock);
1055         sig->group_exit_task = NULL;
1056         sig->notify_count = 0;
1057         read_unlock(&tasklist_lock);
1058         return -EAGAIN;
1059 }
1060
1061 char *get_task_comm(char *buf, struct task_struct *tsk)
1062 {
1063         /* buf must be at least sizeof(tsk->comm) in size */
1064         task_lock(tsk);
1065         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1066         task_unlock(tsk);
1067         return buf;
1068 }
1069 EXPORT_SYMBOL_GPL(get_task_comm);
1070
1071 /*
1072  * These functions flushes out all traces of the currently running executable
1073  * so that a new one can be started
1074  */
1075
1076 void __set_task_comm(struct task_struct *tsk, const char *buf, bool exec)
1077 {
1078         task_lock(tsk);
1079         trace_task_rename(tsk, buf);
1080         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1081         task_unlock(tsk);
1082         perf_event_comm(tsk, exec);
1083 }
1084
1085 int flush_old_exec(struct linux_binprm * bprm)
1086 {
1087         int retval;
1088
1089         /*
1090          * Make sure we have a private signal table and that
1091          * we are unassociated from the previous thread group.
1092          */
1093         retval = de_thread(current);
1094         if (retval)
1095                 goto out;
1096
1097         /*
1098          * Must be called _before_ exec_mmap() as bprm->mm is
1099          * not visibile until then. This also enables the update
1100          * to be lockless.
1101          */
1102         set_mm_exe_file(bprm->mm, bprm->file);
1103
1104         /*
1105          * Release all of the old mmap stuff
1106          */
1107         acct_arg_size(bprm, 0);
1108         retval = exec_mmap(bprm->mm);
1109         if (retval)
1110                 goto out;
1111
1112         bprm->mm = NULL;                /* We're using it now */
1113
1114         set_fs(USER_DS);
1115         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD |
1116                                         PF_NOFREEZE | PF_NO_SETAFFINITY);
1117         flush_thread();
1118         current->personality &= ~bprm->per_clear;
1119
1120         /*
1121          * We have to apply CLOEXEC before we change whether the process is
1122          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1123          * trying to access the should-be-closed file descriptors of a process
1124          * undergoing exec(2).
1125          */
1126         do_close_on_exec(current->files);
1127         return 0;
1128
1129 out:
1130         return retval;
1131 }
1132 EXPORT_SYMBOL(flush_old_exec);
1133
1134 void would_dump(struct linux_binprm *bprm, struct file *file)
1135 {
1136         struct inode *inode = file_inode(file);
1137         if (inode_permission(inode, MAY_READ) < 0) {
1138                 struct user_namespace *old, *user_ns;
1139                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1140
1141                 /* Ensure mm->user_ns contains the executable */
1142                 user_ns = old = bprm->mm->user_ns;
1143                 while ((user_ns != &init_user_ns) &&
1144                        !privileged_wrt_inode_uidgid(user_ns, inode))
1145                         user_ns = user_ns->parent;
1146
1147                 if (old != user_ns) {
1148                         bprm->mm->user_ns = get_user_ns(user_ns);
1149                         put_user_ns(old);
1150                 }
1151         }
1152 }
1153 EXPORT_SYMBOL(would_dump);
1154
1155 void setup_new_exec(struct linux_binprm * bprm)
1156 {
1157         arch_pick_mmap_layout(current->mm);
1158
1159         /* This is the point of no return */
1160         current->sas_ss_sp = current->sas_ss_size = 0;
1161
1162         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1163                 set_dumpable(current->mm, SUID_DUMP_USER);
1164         else
1165                 set_dumpable(current->mm, suid_dumpable);
1166
1167         perf_event_exec();
1168         __set_task_comm(current, kbasename(bprm->filename), true);
1169
1170         /* Set the new mm task size. We have to do that late because it may
1171          * depend on TIF_32BIT which is only updated in flush_thread() on
1172          * some architectures like powerpc
1173          */
1174         current->mm->task_size = TASK_SIZE;
1175
1176         /* install the new credentials */
1177         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1178             !gid_eq(bprm->cred->gid, current_egid())) {
1179                 current->pdeath_signal = 0;
1180         } else {
1181                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1182                         set_dumpable(current->mm, suid_dumpable);
1183         }
1184
1185         /* An exec changes our domain. We are no longer part of the thread
1186            group */
1187         current->self_exec_id++;
1188         flush_signal_handlers(current, 0);
1189 }
1190 EXPORT_SYMBOL(setup_new_exec);
1191
1192 /*
1193  * Prepare credentials and lock ->cred_guard_mutex.
1194  * install_exec_creds() commits the new creds and drops the lock.
1195  * Or, if exec fails before, free_bprm() should release ->cred and
1196  * and unlock.
1197  */
1198 int prepare_bprm_creds(struct linux_binprm *bprm)
1199 {
1200         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1201                 return -ERESTARTNOINTR;
1202
1203         bprm->cred = prepare_exec_creds();
1204         if (likely(bprm->cred))
1205                 return 0;
1206
1207         mutex_unlock(&current->signal->cred_guard_mutex);
1208         return -ENOMEM;
1209 }
1210
1211 static void free_bprm(struct linux_binprm *bprm)
1212 {
1213         free_arg_pages(bprm);
1214         if (bprm->cred) {
1215                 mutex_unlock(&current->signal->cred_guard_mutex);
1216                 abort_creds(bprm->cred);
1217         }
1218         if (bprm->file) {
1219                 allow_write_access(bprm->file);
1220                 fput(bprm->file);
1221         }
1222         /* If a binfmt changed the interp, free it. */
1223         if (bprm->interp != bprm->filename)
1224                 kfree(bprm->interp);
1225         kfree(bprm);
1226 }
1227
1228 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1229 {
1230         /* If a binfmt changed the interp, free it first. */
1231         if (bprm->interp != bprm->filename)
1232                 kfree(bprm->interp);
1233         bprm->interp = kstrdup(interp, GFP_KERNEL);
1234         if (!bprm->interp)
1235                 return -ENOMEM;
1236         return 0;
1237 }
1238 EXPORT_SYMBOL(bprm_change_interp);
1239
1240 /*
1241  * install the new credentials for this executable
1242  */
1243 void install_exec_creds(struct linux_binprm *bprm)
1244 {
1245         security_bprm_committing_creds(bprm);
1246
1247         commit_creds(bprm->cred);
1248         bprm->cred = NULL;
1249
1250         /*
1251          * Disable monitoring for regular users
1252          * when executing setuid binaries. Must
1253          * wait until new credentials are committed
1254          * by commit_creds() above
1255          */
1256         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1257                 perf_event_exit_task(current);
1258         /*
1259          * cred_guard_mutex must be held at least to this point to prevent
1260          * ptrace_attach() from altering our determination of the task's
1261          * credentials; any time after this it may be unlocked.
1262          */
1263         security_bprm_committed_creds(bprm);
1264         mutex_unlock(&current->signal->cred_guard_mutex);
1265 }
1266 EXPORT_SYMBOL(install_exec_creds);
1267
1268 /*
1269  * determine how safe it is to execute the proposed program
1270  * - the caller must hold ->cred_guard_mutex to protect against
1271  *   PTRACE_ATTACH or seccomp thread-sync
1272  */
1273 static void check_unsafe_exec(struct linux_binprm *bprm)
1274 {
1275         struct task_struct *p = current, *t;
1276         unsigned n_fs;
1277
1278         if (p->ptrace) {
1279                 if (ptracer_capable(p, current_user_ns()))
1280                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1281                 else
1282                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1283         }
1284
1285         /*
1286          * This isn't strictly necessary, but it makes it harder for LSMs to
1287          * mess up.
1288          */
1289         if (task_no_new_privs(current))
1290                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1291
1292         t = p;
1293         n_fs = 1;
1294         spin_lock(&p->fs->lock);
1295         rcu_read_lock();
1296         while_each_thread(p, t) {
1297                 if (t->fs == p->fs)
1298                         n_fs++;
1299         }
1300         rcu_read_unlock();
1301
1302         if (p->fs->users > n_fs)
1303                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1304         else
1305                 p->fs->in_exec = 1;
1306         spin_unlock(&p->fs->lock);
1307 }
1308
1309 static void bprm_fill_uid(struct linux_binprm *bprm)
1310 {
1311         struct inode *inode;
1312         unsigned int mode;
1313         kuid_t uid;
1314         kgid_t gid;
1315
1316         /* clear any previous set[ug]id data from a previous binary */
1317         bprm->cred->euid = current_euid();
1318         bprm->cred->egid = current_egid();
1319
1320         if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1321                 return;
1322
1323         if (task_no_new_privs(current))
1324                 return;
1325
1326         inode = file_inode(bprm->file);
1327         mode = READ_ONCE(inode->i_mode);
1328         if (!(mode & (S_ISUID|S_ISGID)))
1329                 return;
1330
1331         /* Be careful if suid/sgid is set */
1332         mutex_lock(&inode->i_mutex);
1333
1334         /* reload atomically mode/uid/gid now that lock held */
1335         mode = inode->i_mode;
1336         uid = inode->i_uid;
1337         gid = inode->i_gid;
1338         mutex_unlock(&inode->i_mutex);
1339
1340         /* We ignore suid/sgid if there are no mappings for them in the ns */
1341         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1342                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1343                 return;
1344
1345         if (mode & S_ISUID) {
1346                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1347                 bprm->cred->euid = uid;
1348         }
1349
1350         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1351                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1352                 bprm->cred->egid = gid;
1353         }
1354 }
1355
1356 /*
1357  * Fill the binprm structure from the inode.
1358  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1359  *
1360  * This may be called multiple times for binary chains (scripts for example).
1361  */
1362 int prepare_binprm(struct linux_binprm *bprm)
1363 {
1364         int retval;
1365
1366         bprm_fill_uid(bprm);
1367
1368         /* fill in binprm security blob */
1369         retval = security_bprm_set_creds(bprm);
1370         if (retval)
1371                 return retval;
1372         bprm->cred_prepared = 1;
1373
1374         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1375         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1376 }
1377
1378 EXPORT_SYMBOL(prepare_binprm);
1379
1380 /*
1381  * Arguments are '\0' separated strings found at the location bprm->p
1382  * points to; chop off the first by relocating brpm->p to right after
1383  * the first '\0' encountered.
1384  */
1385 int remove_arg_zero(struct linux_binprm *bprm)
1386 {
1387         int ret = 0;
1388         unsigned long offset;
1389         char *kaddr;
1390         struct page *page;
1391
1392         if (!bprm->argc)
1393                 return 0;
1394
1395         do {
1396                 offset = bprm->p & ~PAGE_MASK;
1397                 page = get_arg_page(bprm, bprm->p, 0);
1398                 if (!page) {
1399                         ret = -EFAULT;
1400                         goto out;
1401                 }
1402                 kaddr = kmap_atomic(page);
1403
1404                 for (; offset < PAGE_SIZE && kaddr[offset];
1405                                 offset++, bprm->p++)
1406                         ;
1407
1408                 kunmap_atomic(kaddr);
1409                 put_arg_page(page);
1410
1411                 if (offset == PAGE_SIZE)
1412                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1413         } while (offset == PAGE_SIZE);
1414
1415         bprm->p++;
1416         bprm->argc--;
1417         ret = 0;
1418
1419 out:
1420         return ret;
1421 }
1422 EXPORT_SYMBOL(remove_arg_zero);
1423
1424 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1425 /*
1426  * cycle the list of binary formats handler, until one recognizes the image
1427  */
1428 int search_binary_handler(struct linux_binprm *bprm)
1429 {
1430         bool need_retry = IS_ENABLED(CONFIG_MODULES);
1431         struct linux_binfmt *fmt;
1432         int retval;
1433
1434         /* This allows 4 levels of binfmt rewrites before failing hard. */
1435         if (bprm->recursion_depth > 5)
1436                 return -ELOOP;
1437
1438         retval = security_bprm_check(bprm);
1439         if (retval)
1440                 return retval;
1441
1442         retval = -ENOENT;
1443  retry:
1444         read_lock(&binfmt_lock);
1445         list_for_each_entry(fmt, &formats, lh) {
1446                 if (!try_module_get(fmt->module))
1447                         continue;
1448                 read_unlock(&binfmt_lock);
1449                 bprm->recursion_depth++;
1450                 retval = fmt->load_binary(bprm);
1451                 read_lock(&binfmt_lock);
1452                 put_binfmt(fmt);
1453                 bprm->recursion_depth--;
1454                 if (retval < 0 && !bprm->mm) {
1455                         /* we got to flush_old_exec() and failed after it */
1456                         read_unlock(&binfmt_lock);
1457                         force_sigsegv(SIGSEGV, current);
1458                         return retval;
1459                 }
1460                 if (retval != -ENOEXEC || !bprm->file) {
1461                         read_unlock(&binfmt_lock);
1462                         return retval;
1463                 }
1464         }
1465         read_unlock(&binfmt_lock);
1466
1467         if (need_retry) {
1468                 if (printable(bprm->buf[0]) && printable(bprm->buf[1]) &&
1469                     printable(bprm->buf[2]) && printable(bprm->buf[3]))
1470                         return retval;
1471                 if (request_module("binfmt-%04x", *(ushort *)(bprm->buf + 2)) < 0)
1472                         return retval;
1473                 need_retry = false;
1474                 goto retry;
1475         }
1476
1477         return retval;
1478 }
1479 EXPORT_SYMBOL(search_binary_handler);
1480
1481 static int exec_binprm(struct linux_binprm *bprm)
1482 {
1483         pid_t old_pid, old_vpid;
1484         int ret;
1485
1486         /* Need to fetch pid before load_binary changes it */
1487         old_pid = current->pid;
1488         rcu_read_lock();
1489         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1490         rcu_read_unlock();
1491
1492         ret = search_binary_handler(bprm);
1493         if (ret >= 0) {
1494                 audit_bprm(bprm);
1495                 trace_sched_process_exec(current, old_pid, bprm);
1496                 ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1497                 proc_exec_connector(current);
1498         }
1499
1500         return ret;
1501 }
1502
1503 /*
1504  * sys_execve() executes a new program.
1505  */
1506 static int do_execveat_common(int fd, struct filename *filename,
1507                               struct user_arg_ptr argv,
1508                               struct user_arg_ptr envp,
1509                               int flags)
1510 {
1511         char *pathbuf = NULL;
1512         struct linux_binprm *bprm;
1513         struct file *file;
1514         struct files_struct *displaced;
1515         int retval;
1516
1517         if (IS_ERR(filename))
1518                 return PTR_ERR(filename);
1519
1520         /*
1521          * We move the actual failure in case of RLIMIT_NPROC excess from
1522          * set*uid() to execve() because too many poorly written programs
1523          * don't check setuid() return code.  Here we additionally recheck
1524          * whether NPROC limit is still exceeded.
1525          */
1526         if ((current->flags & PF_NPROC_EXCEEDED) &&
1527             atomic_read(&current_user()->processes) > rlimit(RLIMIT_NPROC)) {
1528                 retval = -EAGAIN;
1529                 goto out_ret;
1530         }
1531
1532         /* We're below the limit (still or again), so we don't want to make
1533          * further execve() calls fail. */
1534         current->flags &= ~PF_NPROC_EXCEEDED;
1535
1536         retval = unshare_files(&displaced);
1537         if (retval)
1538                 goto out_ret;
1539
1540         retval = -ENOMEM;
1541         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1542         if (!bprm)
1543                 goto out_files;
1544
1545         retval = prepare_bprm_creds(bprm);
1546         if (retval)
1547                 goto out_free;
1548
1549         check_unsafe_exec(bprm);
1550         current->in_execve = 1;
1551
1552         file = do_open_execat(fd, filename, flags);
1553         retval = PTR_ERR(file);
1554         if (IS_ERR(file))
1555                 goto out_unmark;
1556
1557         sched_exec();
1558
1559         bprm->file = file;
1560         if (fd == AT_FDCWD || filename->name[0] == '/') {
1561                 bprm->filename = filename->name;
1562         } else {
1563                 if (filename->name[0] == '\0')
1564                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d", fd);
1565                 else
1566                         pathbuf = kasprintf(GFP_TEMPORARY, "/dev/fd/%d/%s",
1567                                             fd, filename->name);
1568                 if (!pathbuf) {
1569                         retval = -ENOMEM;
1570                         goto out_unmark;
1571                 }
1572                 /*
1573                  * Record that a name derived from an O_CLOEXEC fd will be
1574                  * inaccessible after exec. Relies on having exclusive access to
1575                  * current->files (due to unshare_files above).
1576                  */
1577                 if (close_on_exec(fd, rcu_dereference_raw(current->files->fdt)))
1578                         bprm->interp_flags |= BINPRM_FLAGS_PATH_INACCESSIBLE;
1579                 bprm->filename = pathbuf;
1580         }
1581         bprm->interp = bprm->filename;
1582
1583         retval = bprm_mm_init(bprm);
1584         if (retval)
1585                 goto out_unmark;
1586
1587         bprm->argc = count(argv, MAX_ARG_STRINGS);
1588         if ((retval = bprm->argc) < 0)
1589                 goto out;
1590
1591         bprm->envc = count(envp, MAX_ARG_STRINGS);
1592         if ((retval = bprm->envc) < 0)
1593                 goto out;
1594
1595         retval = prepare_binprm(bprm);
1596         if (retval < 0)
1597                 goto out;
1598
1599         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1600         if (retval < 0)
1601                 goto out;
1602
1603         bprm->exec = bprm->p;
1604         retval = copy_strings(bprm->envc, envp, bprm);
1605         if (retval < 0)
1606                 goto out;
1607
1608         retval = copy_strings(bprm->argc, argv, bprm);
1609         if (retval < 0)
1610                 goto out;
1611
1612         would_dump(bprm, bprm->file);
1613
1614         retval = exec_binprm(bprm);
1615         if (retval < 0)
1616                 goto out;
1617
1618         /* execve succeeded */
1619         current->fs->in_exec = 0;
1620         current->in_execve = 0;
1621         acct_update_integrals(current);
1622         task_numa_free(current);
1623         free_bprm(bprm);
1624         kfree(pathbuf);
1625         putname(filename);
1626         if (displaced)
1627                 put_files_struct(displaced);
1628         return retval;
1629
1630 out:
1631         if (bprm->mm) {
1632                 acct_arg_size(bprm, 0);
1633                 mmput(bprm->mm);
1634         }
1635
1636 out_unmark:
1637         current->fs->in_exec = 0;
1638         current->in_execve = 0;
1639
1640 out_free:
1641         free_bprm(bprm);
1642         kfree(pathbuf);
1643
1644 out_files:
1645         if (displaced)
1646                 reset_files_struct(displaced);
1647 out_ret:
1648         putname(filename);
1649         return retval;
1650 }
1651
1652 int do_execve(struct filename *filename,
1653         const char __user *const __user *__argv,
1654         const char __user *const __user *__envp)
1655 {
1656         struct user_arg_ptr argv = { .ptr.native = __argv };
1657         struct user_arg_ptr envp = { .ptr.native = __envp };
1658         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1659 }
1660
1661 int do_execveat(int fd, struct filename *filename,
1662                 const char __user *const __user *__argv,
1663                 const char __user *const __user *__envp,
1664                 int flags)
1665 {
1666         struct user_arg_ptr argv = { .ptr.native = __argv };
1667         struct user_arg_ptr envp = { .ptr.native = __envp };
1668
1669         return do_execveat_common(fd, filename, argv, envp, flags);
1670 }
1671
1672 #ifdef CONFIG_COMPAT
1673 static int compat_do_execve(struct filename *filename,
1674         const compat_uptr_t __user *__argv,
1675         const compat_uptr_t __user *__envp)
1676 {
1677         struct user_arg_ptr argv = {
1678                 .is_compat = true,
1679                 .ptr.compat = __argv,
1680         };
1681         struct user_arg_ptr envp = {
1682                 .is_compat = true,
1683                 .ptr.compat = __envp,
1684         };
1685         return do_execveat_common(AT_FDCWD, filename, argv, envp, 0);
1686 }
1687
1688 static int compat_do_execveat(int fd, struct filename *filename,
1689                               const compat_uptr_t __user *__argv,
1690                               const compat_uptr_t __user *__envp,
1691                               int flags)
1692 {
1693         struct user_arg_ptr argv = {
1694                 .is_compat = true,
1695                 .ptr.compat = __argv,
1696         };
1697         struct user_arg_ptr envp = {
1698                 .is_compat = true,
1699                 .ptr.compat = __envp,
1700         };
1701         return do_execveat_common(fd, filename, argv, envp, flags);
1702 }
1703 #endif
1704
1705 void set_binfmt(struct linux_binfmt *new)
1706 {
1707         struct mm_struct *mm = current->mm;
1708
1709         if (mm->binfmt)
1710                 module_put(mm->binfmt->module);
1711
1712         mm->binfmt = new;
1713         if (new)
1714                 __module_get(new->module);
1715 }
1716 EXPORT_SYMBOL(set_binfmt);
1717
1718 /*
1719  * set_dumpable stores three-value SUID_DUMP_* into mm->flags.
1720  */
1721 void set_dumpable(struct mm_struct *mm, int value)
1722 {
1723         unsigned long old, new;
1724
1725         if (WARN_ON((unsigned)value > SUID_DUMP_ROOT))
1726                 return;
1727
1728         do {
1729                 old = ACCESS_ONCE(mm->flags);
1730                 new = (old & ~MMF_DUMPABLE_MASK) | value;
1731         } while (cmpxchg(&mm->flags, old, new) != old);
1732 }
1733
1734 SYSCALL_DEFINE3(execve,
1735                 const char __user *, filename,
1736                 const char __user *const __user *, argv,
1737                 const char __user *const __user *, envp)
1738 {
1739         return do_execve(getname(filename), argv, envp);
1740 }
1741
1742 SYSCALL_DEFINE5(execveat,
1743                 int, fd, const char __user *, filename,
1744                 const char __user *const __user *, argv,
1745                 const char __user *const __user *, envp,
1746                 int, flags)
1747 {
1748         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1749
1750         return do_execveat(fd,
1751                            getname_flags(filename, lookup_flags, NULL),
1752                            argv, envp, flags);
1753 }
1754
1755 #ifdef CONFIG_COMPAT
1756 COMPAT_SYSCALL_DEFINE3(execve, const char __user *, filename,
1757         const compat_uptr_t __user *, argv,
1758         const compat_uptr_t __user *, envp)
1759 {
1760         return compat_do_execve(getname(filename), argv, envp);
1761 }
1762
1763 COMPAT_SYSCALL_DEFINE5(execveat, int, fd,
1764                        const char __user *, filename,
1765                        const compat_uptr_t __user *, argv,
1766                        const compat_uptr_t __user *, envp,
1767                        int,  flags)
1768 {
1769         int lookup_flags = (flags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
1770
1771         return compat_do_execveat(fd,
1772                                   getname_flags(filename, lookup_flags, NULL),
1773                                   argv, envp, flags);
1774 }
1775 #endif