~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/exec.c

Version: ~ [ linux-5.10-rc1 ] ~ [ linux-5.9.1 ] ~ [ linux-5.8.16 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.72 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.152 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.202 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.240 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.240 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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/stat.h>
 30 #include <linux/fcntl.h>
 31 #include <linux/swap.h>
 32 #include <linux/string.h>
 33 #include <linux/init.h>
 34 #include <linux/pagemap.h>
 35 #include <linux/perf_event.h>
 36 #include <linux/highmem.h>
 37 #include <linux/spinlock.h>
 38 #include <linux/key.h>
 39 #include <linux/personality.h>
 40 #include <linux/binfmts.h>
 41 #include <linux/utsname.h>
 42 #include <linux/pid_namespace.h>
 43 #include <linux/module.h>
 44 #include <linux/namei.h>
 45 #include <linux/mount.h>
 46 #include <linux/security.h>
 47 #include <linux/syscalls.h>
 48 #include <linux/tsacct_kern.h>
 49 #include <linux/cn_proc.h>
 50 #include <linux/audit.h>
 51 #include <linux/tracehook.h>
 52 #include <linux/kmod.h>
 53 #include <linux/fsnotify.h>
 54 #include <linux/fs_struct.h>
 55 #include <linux/pipe_fs_i.h>
 56 #include <linux/oom.h>
 57 #include <linux/compat.h>
 58 
 59 #include <asm/uaccess.h>
 60 #include <asm/mmu_context.h>
 61 #include <asm/tlb.h>
 62 
 63 #include <trace/events/task.h>
 64 #include "internal.h"
 65 #include "coredump.h"
 66 
 67 #include <trace/events/sched.h>
 68 
 69 int suid_dumpable = 0;
 70 
 71 static LIST_HEAD(formats);
 72 static DEFINE_RWLOCK(binfmt_lock);
 73 
 74 void __register_binfmt(struct linux_binfmt * fmt, int insert)
 75 {
 76         BUG_ON(!fmt);
 77         write_lock(&binfmt_lock);
 78         insert ? list_add(&fmt->lh, &formats) :
 79                  list_add_tail(&fmt->lh, &formats);
 80         write_unlock(&binfmt_lock);
 81 }
 82 
 83 EXPORT_SYMBOL(__register_binfmt);
 84 
 85 void unregister_binfmt(struct linux_binfmt * fmt)
 86 {
 87         write_lock(&binfmt_lock);
 88         list_del(&fmt->lh);
 89         write_unlock(&binfmt_lock);
 90 }
 91 
 92 EXPORT_SYMBOL(unregister_binfmt);
 93 
 94 static inline void put_binfmt(struct linux_binfmt * fmt)
 95 {
 96         module_put(fmt->module);
 97 }
 98 
 99 /*
100  * Note that a shared library must be both readable and executable due to
101  * security reasons.
102  *
103  * Also note that we take the address to load from from the file itself.
104  */
105 SYSCALL_DEFINE1(uselib, const char __user *, library)
106 {
107         struct file *file;
108         struct filename *tmp = getname(library);
109         int error = PTR_ERR(tmp);
110         static const struct open_flags uselib_flags = {
111                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
112                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
113                 .intent = LOOKUP_OPEN
114         };
115 
116         if (IS_ERR(tmp))
117                 goto out;
118 
119         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
120         putname(tmp);
121         error = PTR_ERR(file);
122         if (IS_ERR(file))
123                 goto out;
124 
125         error = -EINVAL;
126         if (!S_ISREG(file_inode(file)->i_mode))
127                 goto exit;
128 
129         error = -EACCES;
130         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
131                 goto exit;
132 
133         fsnotify_open(file);
134 
135         error = -ENOEXEC;
136         if(file->f_op) {
137                 struct linux_binfmt * fmt;
138 
139                 read_lock(&binfmt_lock);
140                 list_for_each_entry(fmt, &formats, lh) {
141                         if (!fmt->load_shlib)
142                                 continue;
143                         if (!try_module_get(fmt->module))
144                                 continue;
145                         read_unlock(&binfmt_lock);
146                         error = fmt->load_shlib(file);
147                         read_lock(&binfmt_lock);
148                         put_binfmt(fmt);
149                         if (error != -ENOEXEC)
150                                 break;
151                 }
152                 read_unlock(&binfmt_lock);
153         }
154 exit:
155         fput(file);
156 out:
157         return error;
158 }
159 
160 #ifdef CONFIG_MMU
161 /*
162  * The nascent bprm->mm is not visible until exec_mmap() but it can
163  * use a lot of memory, account these pages in current->mm temporary
164  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
165  * change the counter back via acct_arg_size(0).
166  */
167 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
168 {
169         struct mm_struct *mm = current->mm;
170         long diff = (long)(pages - bprm->vma_pages);
171 
172         if (!mm || !diff)
173                 return;
174 
175         bprm->vma_pages = pages;
176         add_mm_counter(mm, MM_ANONPAGES, diff);
177 }
178 
179 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
180                 int write)
181 {
182         struct page *page;
183         int ret;
184 
185 #ifdef CONFIG_STACK_GROWSUP
186         if (write) {
187                 ret = expand_downwards(bprm->vma, pos);
188                 if (ret < 0)
189                         return NULL;
190         }
191 #endif
192         ret = get_user_pages(current, bprm->mm, pos,
193                         1, write, 1, &page, NULL);
194         if (ret <= 0)
195                 return NULL;
196 
197         if (write) {
198                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
199                 unsigned long ptr_size;
200                 struct rlimit *rlim;
201 
202                 /*
203                  * Since the stack will hold pointers to the strings, we
204                  * must account for them as well.
205                  *
206                  * The size calculation is the entire vma while each arg page is
207                  * built, so each time we get here it's calculating how far it
208                  * is currently (rather than each call being just the newly
209                  * added size from the arg page).  As a result, we need to
210                  * always add the entire size of the pointers, so that on the
211                  * last call to get_arg_page() we'll actually have the entire
212                  * correct size.
213                  */
214                 ptr_size = (bprm->argc + bprm->envc) * sizeof(void *);
215                 if (ptr_size > ULONG_MAX - size)
216                         goto fail;
217                 size += ptr_size;
218 
219                 acct_arg_size(bprm, size / PAGE_SIZE);
220 
221                 /*
222                  * We've historically supported up to 32 pages (ARG_MAX)
223                  * of argument strings even with small stacks
224                  */
225                 if (size <= ARG_MAX)
226                         return page;
227 
228                 /*
229                  * Limit to 1/4-th the stack size for the argv+env strings.
230                  * This ensures that:
231                  *  - the remaining binfmt code will not run out of stack space,
232                  *  - the program will have a reasonable amount of stack left
233                  *    to work from.
234                  */
235                 rlim = current->signal->rlim;
236                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4)
237                         goto fail;
238         }
239 
240         return page;
241 
242 fail:
243         put_page(page);
244         return NULL;
245 }
246 
247 static void put_arg_page(struct page *page)
248 {
249         put_page(page);
250 }
251 
252 static void free_arg_page(struct linux_binprm *bprm, int i)
253 {
254 }
255 
256 static void free_arg_pages(struct linux_binprm *bprm)
257 {
258 }
259 
260 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
261                 struct page *page)
262 {
263         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
264 }
265 
266 static int __bprm_mm_init(struct linux_binprm *bprm)
267 {
268         int err;
269         struct vm_area_struct *vma = NULL;
270         struct mm_struct *mm = bprm->mm;
271 
272         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
273         if (!vma)
274                 return -ENOMEM;
275 
276         down_write(&mm->mmap_sem);
277         vma->vm_mm = mm;
278 
279         /*
280          * Place the stack at the largest stack address the architecture
281          * supports. Later, we'll move this to an appropriate place. We don't
282          * use STACK_TOP because that can depend on attributes which aren't
283          * configured yet.
284          */
285         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
286         vma->vm_end = STACK_TOP_MAX;
287         vma->vm_start = vma->vm_end - PAGE_SIZE;
288         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
289         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
290         INIT_LIST_HEAD(&vma->anon_vma_chain);
291 
292         err = insert_vm_struct(mm, vma);
293         if (err)
294                 goto err;
295 
296         mm->stack_vm = mm->total_vm = 1;
297         up_write(&mm->mmap_sem);
298         bprm->p = vma->vm_end - sizeof(void *);
299         return 0;
300 err:
301         up_write(&mm->mmap_sem);
302         bprm->vma = NULL;
303         kmem_cache_free(vm_area_cachep, vma);
304         return err;
305 }
306 
307 static bool valid_arg_len(struct linux_binprm *bprm, long len)
308 {
309         return len <= MAX_ARG_STRLEN;
310 }
311 
312 #else
313 
314 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
315 {
316 }
317 
318 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
319                 int write)
320 {
321         struct page *page;
322 
323         page = bprm->page[pos / PAGE_SIZE];
324         if (!page && write) {
325                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
326                 if (!page)
327                         return NULL;
328                 bprm->page[pos / PAGE_SIZE] = page;
329         }
330 
331         return page;
332 }
333 
334 static void put_arg_page(struct page *page)
335 {
336 }
337 
338 static void free_arg_page(struct linux_binprm *bprm, int i)
339 {
340         if (bprm->page[i]) {
341                 __free_page(bprm->page[i]);
342                 bprm->page[i] = NULL;
343         }
344 }
345 
346 static void free_arg_pages(struct linux_binprm *bprm)
347 {
348         int i;
349 
350         for (i = 0; i < MAX_ARG_PAGES; i++)
351                 free_arg_page(bprm, i);
352 }
353 
354 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
355                 struct page *page)
356 {
357 }
358 
359 static int __bprm_mm_init(struct linux_binprm *bprm)
360 {
361         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
362         return 0;
363 }
364 
365 static bool valid_arg_len(struct linux_binprm *bprm, long len)
366 {
367         return len <= bprm->p;
368 }
369 
370 #endif /* CONFIG_MMU */
371 
372 /*
373  * Create a new mm_struct and populate it with a temporary stack
374  * vm_area_struct.  We don't have enough context at this point to set the stack
375  * flags, permissions, and offset, so we use temporary values.  We'll update
376  * them later in setup_arg_pages().
377  */
378 static int bprm_mm_init(struct linux_binprm *bprm)
379 {
380         int err;
381         struct mm_struct *mm = NULL;
382 
383         bprm->mm = mm = mm_alloc();
384         err = -ENOMEM;
385         if (!mm)
386                 goto err;
387 
388         err = init_new_context(current, mm);
389         if (err)
390                 goto err;
391 
392         err = __bprm_mm_init(bprm);
393         if (err)
394                 goto err;
395 
396         return 0;
397 
398 err:
399         if (mm) {
400                 bprm->mm = NULL;
401                 mmdrop(mm);
402         }
403 
404         return err;
405 }
406 
407 struct user_arg_ptr {
408 #ifdef CONFIG_COMPAT
409         bool is_compat;
410 #endif
411         union {
412                 const char __user *const __user *native;
413 #ifdef CONFIG_COMPAT
414                 const compat_uptr_t __user *compat;
415 #endif
416         } ptr;
417 };
418 
419 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
420 {
421         const char __user *native;
422 
423 #ifdef CONFIG_COMPAT
424         if (unlikely(argv.is_compat)) {
425                 compat_uptr_t compat;
426 
427                 if (get_user(compat, argv.ptr.compat + nr))
428                         return ERR_PTR(-EFAULT);
429 
430                 return compat_ptr(compat);
431         }
432 #endif
433 
434         if (get_user(native, argv.ptr.native + nr))
435                 return ERR_PTR(-EFAULT);
436 
437         return native;
438 }
439 
440 /*
441  * count() counts the number of strings in array ARGV.
442  */
443 static int count(struct user_arg_ptr argv, int max)
444 {
445         int i = 0;
446 
447         if (argv.ptr.native != NULL) {
448                 for (;;) {
449                         const char __user *p = get_user_arg_ptr(argv, i);
450 
451                         if (!p)
452                                 break;
453 
454                         if (IS_ERR(p))
455                                 return -EFAULT;
456 
457                         if (i >= max)
458                                 return -E2BIG;
459                         ++i;
460 
461                         if (fatal_signal_pending(current))
462                                 return -ERESTARTNOHAND;
463                         cond_resched();
464                 }
465         }
466         return i;
467 }
468 
469 /*
470  * 'copy_strings()' copies argument/environment strings from the old
471  * processes's memory to the new process's stack.  The call to get_user_pages()
472  * ensures the destination page is created and not swapped out.
473  */
474 static int copy_strings(int argc, struct user_arg_ptr argv,
475                         struct linux_binprm *bprm)
476 {
477         struct page *kmapped_page = NULL;
478         char *kaddr = NULL;
479         unsigned long kpos = 0;
480         int ret;
481 
482         while (argc-- > 0) {
483                 const char __user *str;
484                 int len;
485                 unsigned long pos;
486 
487                 ret = -EFAULT;
488                 str = get_user_arg_ptr(argv, argc);
489                 if (IS_ERR(str))
490                         goto out;
491 
492                 len = strnlen_user(str, MAX_ARG_STRLEN);
493                 if (!len)
494                         goto out;
495 
496                 ret = -E2BIG;
497                 if (!valid_arg_len(bprm, len))
498                         goto out;
499 
500                 /* We're going to work our way backwords. */
501                 pos = bprm->p;
502                 str += len;
503                 bprm->p -= len;
504 
505                 while (len > 0) {
506                         int offset, bytes_to_copy;
507 
508                         if (fatal_signal_pending(current)) {
509                                 ret = -ERESTARTNOHAND;
510                                 goto out;
511                         }
512                         cond_resched();
513 
514                         offset = pos % PAGE_SIZE;
515                         if (offset == 0)
516                                 offset = PAGE_SIZE;
517 
518                         bytes_to_copy = offset;
519                         if (bytes_to_copy > len)
520                                 bytes_to_copy = len;
521 
522                         offset -= bytes_to_copy;
523                         pos -= bytes_to_copy;
524                         str -= bytes_to_copy;
525                         len -= bytes_to_copy;
526 
527                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
528                                 struct page *page;
529 
530                                 page = get_arg_page(bprm, pos, 1);
531                                 if (!page) {
532                                         ret = -E2BIG;
533                                         goto out;
534                                 }
535 
536                                 if (kmapped_page) {
537                                         flush_kernel_dcache_page(kmapped_page);
538                                         kunmap(kmapped_page);
539                                         put_arg_page(kmapped_page);
540                                 }
541                                 kmapped_page = page;
542                                 kaddr = kmap(kmapped_page);
543                                 kpos = pos & PAGE_MASK;
544                                 flush_arg_page(bprm, kpos, kmapped_page);
545                         }
546                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
547                                 ret = -EFAULT;
548                                 goto out;
549                         }
550                 }
551         }
552         ret = 0;
553 out:
554         if (kmapped_page) {
555                 flush_kernel_dcache_page(kmapped_page);
556                 kunmap(kmapped_page);
557                 put_arg_page(kmapped_page);
558         }
559         return ret;
560 }
561 
562 /*
563  * Like copy_strings, but get argv and its values from kernel memory.
564  */
565 int copy_strings_kernel(int argc, const char *const *__argv,
566                         struct linux_binprm *bprm)
567 {
568         int r;
569         mm_segment_t oldfs = get_fs();
570         struct user_arg_ptr argv = {
571                 .ptr.native = (const char __user *const  __user *)__argv,
572         };
573 
574         set_fs(KERNEL_DS);
575         r = copy_strings(argc, argv, bprm);
576         set_fs(oldfs);
577 
578         return r;
579 }
580 EXPORT_SYMBOL(copy_strings_kernel);
581 
582 #ifdef CONFIG_MMU
583 
584 /*
585  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
586  * the binfmt code determines where the new stack should reside, we shift it to
587  * its final location.  The process proceeds as follows:
588  *
589  * 1) Use shift to calculate the new vma endpoints.
590  * 2) Extend vma to cover both the old and new ranges.  This ensures the
591  *    arguments passed to subsequent functions are consistent.
592  * 3) Move vma's page tables to the new range.
593  * 4) Free up any cleared pgd range.
594  * 5) Shrink the vma to cover only the new range.
595  */
596 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
597 {
598         struct mm_struct *mm = vma->vm_mm;
599         unsigned long old_start = vma->vm_start;
600         unsigned long old_end = vma->vm_end;
601         unsigned long length = old_end - old_start;
602         unsigned long new_start = old_start - shift;
603         unsigned long new_end = old_end - shift;
604         struct mmu_gather tlb;
605 
606         BUG_ON(new_start > new_end);
607 
608         /*
609          * ensure there are no vmas between where we want to go
610          * and where we are
611          */
612         if (vma != find_vma(mm, new_start))
613                 return -EFAULT;
614 
615         /*
616          * cover the whole range: [new_start, old_end)
617          */
618         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
619                 return -ENOMEM;
620 
621         /*
622          * move the page tables downwards, on failure we rely on
623          * process cleanup to remove whatever mess we made.
624          */
625         if (length != move_page_tables(vma, old_start,
626                                        vma, new_start, length, false))
627                 return -ENOMEM;
628 
629         lru_add_drain();
630         tlb_gather_mmu(&tlb, mm, old_start, old_end);
631         if (new_end > old_start) {
632                 /*
633                  * when the old and new regions overlap clear from new_end.
634                  */
635                 free_pgd_range(&tlb, new_end, old_end, new_end,
636                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
637         } else {
638                 /*
639                  * otherwise, clean from old_start; this is done to not touch
640                  * the address space in [new_end, old_start) some architectures
641                  * have constraints on va-space that make this illegal (IA64) -
642                  * for the others its just a little faster.
643                  */
644                 free_pgd_range(&tlb, old_start, old_end, new_end,
645                         vma->vm_next ? vma->vm_next->vm_start : USER_PGTABLES_CEILING);
646         }
647         tlb_finish_mmu(&tlb, old_start, old_end);
648 
649         /*
650          * Shrink the vma to just the new range.  Always succeeds.
651          */
652         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
653 
654         return 0;
655 }
656 
657 /*
658  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
659  * the stack is optionally relocated, and some extra space is added.
660  */
661 int setup_arg_pages(struct linux_binprm *bprm,
662                     unsigned long stack_top,
663                     int executable_stack)
664 {
665         unsigned long ret;
666         unsigned long stack_shift;
667         struct mm_struct *mm = current->mm;
668         struct vm_area_struct *vma = bprm->vma;
669         struct vm_area_struct *prev = NULL;
670         unsigned long vm_flags;
671         unsigned long stack_base;
672         unsigned long stack_size;
673         unsigned long stack_expand;
674         unsigned long rlim_stack;
675 
676 #ifdef CONFIG_STACK_GROWSUP
677         /* Limit stack size */
678         stack_base = rlimit_max(RLIMIT_STACK);
679         if (stack_base > STACK_SIZE_MAX)
680                 stack_base = STACK_SIZE_MAX;
681 
682         /* Make sure we didn't let the argument array grow too large. */
683         if (vma->vm_end - vma->vm_start > stack_base)
684                 return -ENOMEM;
685 
686         stack_base = PAGE_ALIGN(stack_top - stack_base);
687 
688         stack_shift = vma->vm_start - stack_base;
689         mm->arg_start = bprm->p - stack_shift;
690         bprm->p = vma->vm_end - stack_shift;
691 #else
692         stack_top = arch_align_stack(stack_top);
693         stack_top = PAGE_ALIGN(stack_top);
694 
695         if (unlikely(stack_top < mmap_min_addr) ||
696             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
697                 return -ENOMEM;
698 
699         stack_shift = vma->vm_end - stack_top;
700 
701         bprm->p -= stack_shift;
702         mm->arg_start = bprm->p;
703 #endif
704 
705         if (bprm->loader)
706                 bprm->loader -= stack_shift;
707         bprm->exec -= stack_shift;
708 
709         down_write(&mm->mmap_sem);
710         vm_flags = VM_STACK_FLAGS;
711 
712         /*
713          * Adjust stack execute permissions; explicitly enable for
714          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
715          * (arch default) otherwise.
716          */
717         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
718                 vm_flags |= VM_EXEC;
719         else if (executable_stack == EXSTACK_DISABLE_X)
720                 vm_flags &= ~VM_EXEC;
721         vm_flags |= mm->def_flags;
722         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
723 
724         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
725                         vm_flags);
726         if (ret)
727                 goto out_unlock;
728         BUG_ON(prev != vma);
729 
730         /* Move stack pages down in memory. */
731         if (stack_shift) {
732                 ret = shift_arg_pages(vma, stack_shift);
733                 if (ret)
734                         goto out_unlock;
735         }
736 
737         /* mprotect_fixup is overkill to remove the temporary stack flags */
738         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
739 
740         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
741         stack_size = vma->vm_end - vma->vm_start;
742         /*
743          * Align this down to a page boundary as expand_stack
744          * will align it up.
745          */
746         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
747 #ifdef CONFIG_STACK_GROWSUP
748         if (stack_size + stack_expand > rlim_stack)
749                 stack_base = vma->vm_start + rlim_stack;
750         else
751                 stack_base = vma->vm_end + stack_expand;
752 #else
753         if (stack_size + stack_expand > rlim_stack)
754                 stack_base = vma->vm_end - rlim_stack;
755         else
756                 stack_base = vma->vm_start - stack_expand;
757 #endif
758         current->mm->start_stack = bprm->p;
759         ret = expand_stack(vma, stack_base);
760         if (ret)
761                 ret = -EFAULT;
762 
763 out_unlock:
764         up_write(&mm->mmap_sem);
765         return ret;
766 }
767 EXPORT_SYMBOL(setup_arg_pages);
768 
769 #endif /* CONFIG_MMU */
770 
771 struct file *open_exec(const char *name)
772 {
773         struct file *file;
774         int err;
775         struct filename tmp = { .name = name };
776         static const struct open_flags open_exec_flags = {
777                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
778                 .acc_mode = MAY_EXEC | MAY_OPEN,
779                 .intent = LOOKUP_OPEN
780         };
781 
782         file = do_filp_open(AT_FDCWD, &tmp, &open_exec_flags, LOOKUP_FOLLOW);
783         if (IS_ERR(file))
784                 goto out;
785 
786         err = -EACCES;
787         if (!S_ISREG(file_inode(file)->i_mode))
788                 goto exit;
789 
790         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
791                 goto exit;
792 
793         fsnotify_open(file);
794 
795         err = deny_write_access(file);
796         if (err)
797                 goto exit;
798 
799 out:
800         return file;
801 
802 exit:
803         fput(file);
804         return ERR_PTR(err);
805 }
806 EXPORT_SYMBOL(open_exec);
807 
808 int kernel_read(struct file *file, loff_t offset,
809                 char *addr, unsigned long count)
810 {
811         mm_segment_t old_fs;
812         loff_t pos = offset;
813         int result;
814 
815         old_fs = get_fs();
816         set_fs(get_ds());
817         /* The cast to a user pointer is valid due to the set_fs() */
818         result = vfs_read(file, (void __user *)addr, count, &pos);
819         set_fs(old_fs);
820         return result;
821 }
822 
823 EXPORT_SYMBOL(kernel_read);
824 
825 ssize_t read_code(struct file *file, unsigned long addr, loff_t pos, size_t len)
826 {
827         ssize_t res = file->f_op->read(file, (void __user *)addr, len, &pos);
828         if (res > 0)
829                 flush_icache_range(addr, addr + len);
830         return res;
831 }
832 EXPORT_SYMBOL(read_code);
833 
834 static int exec_mmap(struct mm_struct *mm)
835 {
836         struct task_struct *tsk;
837         struct mm_struct * old_mm, *active_mm;
838 
839         /* Notify parent that we're no longer interested in the old VM */
840         tsk = current;
841         old_mm = current->mm;
842         mm_release(tsk, old_mm);
843 
844         if (old_mm) {
845                 sync_mm_rss(old_mm);
846                 /*
847                  * Make sure that if there is a core dump in progress
848                  * for the old mm, we get out and die instead of going
849                  * through with the exec.  We must hold mmap_sem around
850                  * checking core_state and changing tsk->mm.
851                  */
852                 down_read(&old_mm->mmap_sem);
853                 if (unlikely(old_mm->core_state)) {
854                         up_read(&old_mm->mmap_sem);
855                         return -EINTR;
856                 }
857         }
858         task_lock(tsk);
859         active_mm = tsk->active_mm;
860         tsk->mm = mm;
861         tsk->active_mm = mm;
862         activate_mm(active_mm, mm);
863         task_unlock(tsk);
864         arch_pick_mmap_layout(mm);
865         if (old_mm) {
866                 up_read(&old_mm->mmap_sem);
867                 BUG_ON(active_mm != old_mm);
868                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
869                 mm_update_next_owner(old_mm);
870                 mmput(old_mm);
871                 return 0;
872         }
873         mmdrop(active_mm);
874         return 0;
875 }
876 
877 /*
878  * This function makes sure the current process has its own signal table,
879  * so that flush_signal_handlers can later reset the handlers without
880  * disturbing other processes.  (Other processes might share the signal
881  * table via the CLONE_SIGHAND option to clone().)
882  */
883 static int de_thread(struct task_struct *tsk)
884 {
885         struct signal_struct *sig = tsk->signal;
886         struct sighand_struct *oldsighand = tsk->sighand;
887         spinlock_t *lock = &oldsighand->siglock;
888 
889         if (thread_group_empty(tsk))
890                 goto no_thread_group;
891 
892         /*
893          * Kill all other threads in the thread group.
894          */
895         spin_lock_irq(lock);
896         if (signal_group_exit(sig)) {
897                 /*
898                  * Another group action in progress, just
899                  * return so that the signal is processed.
900                  */
901                 spin_unlock_irq(lock);
902                 return -EAGAIN;
903         }
904 
905         sig->group_exit_task = tsk;
906         sig->notify_count = zap_other_threads(tsk);
907         if (!thread_group_leader(tsk))
908                 sig->notify_count--;
909 
910         while (sig->notify_count) {
911                 __set_current_state(TASK_KILLABLE);
912                 spin_unlock_irq(lock);
913                 schedule();
914                 if (unlikely(__fatal_signal_pending(tsk)))
915                         goto killed;
916                 spin_lock_irq(lock);
917         }
918         spin_unlock_irq(lock);
919 
920         /*
921          * At this point all other threads have exited, all we have to
922          * do is to wait for the thread group leader to become inactive,
923          * and to assume its PID:
924          */
925         if (!thread_group_leader(tsk)) {
926                 struct task_struct *leader = tsk->group_leader;
927 
928                 sig->notify_count = -1; /* for exit_notify() */
929                 for (;;) {
930                         threadgroup_change_begin(tsk);
931                         write_lock_irq(&tasklist_lock);
932                         if (likely(leader->exit_state))
933                                 break;
934                         __set_current_state(TASK_KILLABLE);
935                         write_unlock_irq(&tasklist_lock);
936                         threadgroup_change_end(tsk);
937                         schedule();
938                         if (unlikely(__fatal_signal_pending(tsk)))
939                                 goto killed;
940                 }
941 
942                 /*
943                  * The only record we have of the real-time age of a
944                  * process, regardless of execs it's done, is start_time.
945                  * All the past CPU time is accumulated in signal_struct
946                  * from sister threads now dead.  But in this non-leader
947                  * exec, nothing survives from the original leader thread,
948                  * whose birth marks the true age of this process now.
949                  * When we take on its identity by switching to its PID, we
950                  * also take its birthdate (always earlier than our own).
951                  */
952                 tsk->start_time = leader->start_time;
953 
954                 BUG_ON(!same_thread_group(leader, tsk));
955                 BUG_ON(has_group_leader_pid(tsk));
956                 /*
957                  * An exec() starts a new thread group with the
958                  * TGID of the previous thread group. Rehash the
959                  * two threads with a switched PID, and release
960                  * the former thread group leader:
961                  */
962 
963                 /* Become a process group leader with the old leader's pid.
964                  * The old leader becomes a thread of the this thread group.
965                  * Note: The old leader also uses this pid until release_task
966                  *       is called.  Odd but simple and correct.
967                  */
968                 detach_pid(tsk, PIDTYPE_PID);
969                 tsk->pid = leader->pid;
970                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
971                 transfer_pid(leader, tsk, PIDTYPE_PGID);
972                 transfer_pid(leader, tsk, PIDTYPE_SID);
973 
974                 list_replace_rcu(&leader->tasks, &tsk->tasks);
975                 list_replace_init(&leader->sibling, &tsk->sibling);
976 
977                 tsk->group_leader = tsk;
978                 leader->group_leader = tsk;
979 
980                 tsk->exit_signal = SIGCHLD;
981                 leader->exit_signal = -1;
982 
983                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
984                 leader->exit_state = EXIT_DEAD;
985 
986                 /*
987                  * We are going to release_task()->ptrace_unlink() silently,
988                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
989                  * the tracer wont't block again waiting for this thread.
990                  */
991                 if (unlikely(leader->ptrace))
992                         __wake_up_parent(leader, leader->parent);
993                 write_unlock_irq(&tasklist_lock);
994                 threadgroup_change_end(tsk);
995 
996                 release_task(leader);
997         }
998 
999         sig->group_exit_task = NULL;
1000         sig->notify_count = 0;
1001 
1002 no_thread_group:
1003         /* we have changed execution domain */
1004         tsk->exit_signal = SIGCHLD;
1005 
1006         exit_itimers(sig);
1007         flush_itimer_signals();
1008 
1009         if (atomic_read(&oldsighand->count) != 1) {
1010                 struct sighand_struct *newsighand;
1011                 /*
1012                  * This ->sighand is shared with the CLONE_SIGHAND
1013                  * but not CLONE_THREAD task, switch to the new one.
1014                  */
1015                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
1016                 if (!newsighand)
1017                         return -ENOMEM;
1018 
1019                 atomic_set(&newsighand->count, 1);
1020                 memcpy(newsighand->action, oldsighand->action,
1021                        sizeof(newsighand->action));
1022 
1023                 write_lock_irq(&tasklist_lock);
1024                 spin_lock(&oldsighand->siglock);
1025                 rcu_assign_pointer(tsk->sighand, newsighand);
1026                 spin_unlock(&oldsighand->siglock);
1027                 write_unlock_irq(&tasklist_lock);
1028 
1029                 __cleanup_sighand(oldsighand);
1030         }
1031 
1032         BUG_ON(!thread_group_leader(tsk));
1033         return 0;
1034 
1035 killed:
1036         /* protects against exit_notify() and __exit_signal() */
1037         read_lock(&tasklist_lock);
1038         sig->group_exit_task = NULL;
1039         sig->notify_count = 0;
1040         read_unlock(&tasklist_lock);
1041         return -EAGAIN;
1042 }
1043 
1044 char *get_task_comm(char *buf, struct task_struct *tsk)
1045 {
1046         /* buf must be at least sizeof(tsk->comm) in size */
1047         task_lock(tsk);
1048         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1049         task_unlock(tsk);
1050         return buf;
1051 }
1052 EXPORT_SYMBOL_GPL(get_task_comm);
1053 
1054 /*
1055  * These functions flushes out all traces of the currently running executable
1056  * so that a new one can be started
1057  */
1058 
1059 void set_task_comm(struct task_struct *tsk, char *buf)
1060 {
1061         task_lock(tsk);
1062         trace_task_rename(tsk, buf);
1063         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1064         task_unlock(tsk);
1065         perf_event_comm(tsk);
1066 }
1067 
1068 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1069 {
1070         int i, ch;
1071 
1072         /* Copies the binary name from after last slash */
1073         for (i = 0; (ch = *(fn++)) != '\0';) {
1074                 if (ch == '/')
1075                         i = 0; /* overwrite what we wrote */
1076                 else
1077                         if (i < len - 1)
1078                                 tcomm[i++] = ch;
1079         }
1080         tcomm[i] = '\0';
1081 }
1082 
1083 int flush_old_exec(struct linux_binprm * bprm)
1084 {
1085         int retval;
1086 
1087         /*
1088          * Make sure we have a private signal table and that
1089          * we are unassociated from the previous thread group.
1090          */
1091         retval = de_thread(current);
1092         if (retval)
1093                 goto out;
1094 
1095         set_mm_exe_file(bprm->mm, bprm->file);
1096 
1097         filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1098         /*
1099          * Release all of the old mmap stuff
1100          */
1101         acct_arg_size(bprm, 0);
1102         retval = exec_mmap(bprm->mm);
1103         if (retval)
1104                 goto out;
1105 
1106         bprm->mm = NULL;                /* We're using it now */
1107 
1108         set_fs(USER_DS);
1109         current->flags &=
1110                 ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD | PF_NOFREEZE);
1111         flush_thread();
1112         current->personality &= ~bprm->per_clear;
1113 
1114         /*
1115          * We have to apply CLOEXEC before we change whether the process is
1116          * dumpable (in setup_new_exec) to avoid a race with a process in userspace
1117          * trying to access the should-be-closed file descriptors of a process
1118          * undergoing exec(2).
1119          */
1120         do_close_on_exec(current->files);
1121         return 0;
1122 
1123 out:
1124         return retval;
1125 }
1126 EXPORT_SYMBOL(flush_old_exec);
1127 
1128 void would_dump(struct linux_binprm *bprm, struct file *file)
1129 {
1130         if (inode_permission(file_inode(file), MAY_READ) < 0)
1131                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1132 }
1133 EXPORT_SYMBOL(would_dump);
1134 
1135 void setup_new_exec(struct linux_binprm * bprm)
1136 {
1137         arch_pick_mmap_layout(current->mm);
1138 
1139         /* This is the point of no return */
1140         current->sas_ss_sp = current->sas_ss_size = 0;
1141 
1142         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1143                 set_dumpable(current->mm, SUID_DUMP_USER);
1144         else
1145                 set_dumpable(current->mm, suid_dumpable);
1146 
1147         set_task_comm(current, bprm->tcomm);
1148 
1149         /* Set the new mm task size. We have to do that late because it may
1150          * depend on TIF_32BIT which is only updated in flush_thread() on
1151          * some architectures like powerpc
1152          */
1153         current->mm->task_size = TASK_SIZE;
1154 
1155         /* install the new credentials */
1156         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1157             !gid_eq(bprm->cred->gid, current_egid())) {
1158                 current->pdeath_signal = 0;
1159         } else {
1160                 would_dump(bprm, bprm->file);
1161                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1162                         set_dumpable(current->mm, suid_dumpable);
1163         }
1164 
1165         /* An exec changes our domain. We are no longer part of the thread
1166            group */
1167 
1168         current->self_exec_id++;
1169                         
1170         flush_signal_handlers(current, 0);
1171 }
1172 EXPORT_SYMBOL(setup_new_exec);
1173 
1174 /*
1175  * Prepare credentials and lock ->cred_guard_mutex.
1176  * install_exec_creds() commits the new creds and drops the lock.
1177  * Or, if exec fails before, free_bprm() should release ->cred and
1178  * and unlock.
1179  */
1180 int prepare_bprm_creds(struct linux_binprm *bprm)
1181 {
1182         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1183                 return -ERESTARTNOINTR;
1184 
1185         bprm->cred = prepare_exec_creds();
1186         if (likely(bprm->cred))
1187                 return 0;
1188 
1189         mutex_unlock(&current->signal->cred_guard_mutex);
1190         return -ENOMEM;
1191 }
1192 
1193 void free_bprm(struct linux_binprm *bprm)
1194 {
1195         free_arg_pages(bprm);
1196         if (bprm->cred) {
1197                 mutex_unlock(&current->signal->cred_guard_mutex);
1198                 abort_creds(bprm->cred);
1199         }
1200         /* If a binfmt changed the interp, free it. */
1201         if (bprm->interp != bprm->filename)
1202                 kfree(bprm->interp);
1203         kfree(bprm);
1204 }
1205 
1206 int bprm_change_interp(char *interp, struct linux_binprm *bprm)
1207 {
1208         /* If a binfmt changed the interp, free it first. */
1209         if (bprm->interp != bprm->filename)
1210                 kfree(bprm->interp);
1211         bprm->interp = kstrdup(interp, GFP_KERNEL);
1212         if (!bprm->interp)
1213                 return -ENOMEM;
1214         return 0;
1215 }
1216 EXPORT_SYMBOL(bprm_change_interp);
1217 
1218 /*
1219  * install the new credentials for this executable
1220  */
1221 void install_exec_creds(struct linux_binprm *bprm)
1222 {
1223         security_bprm_committing_creds(bprm);
1224 
1225         commit_creds(bprm->cred);
1226         bprm->cred = NULL;
1227 
1228         /*
1229          * Disable monitoring for regular users
1230          * when executing setuid binaries. Must
1231          * wait until new credentials are committed
1232          * by commit_creds() above
1233          */
1234         if (get_dumpable(current->mm) != SUID_DUMP_USER)
1235                 perf_event_exit_task(current);
1236         /*
1237          * cred_guard_mutex must be held at least to this point to prevent
1238          * ptrace_attach() from altering our determination of the task's
1239          * credentials; any time after this it may be unlocked.
1240          */
1241         security_bprm_committed_creds(bprm);
1242         mutex_unlock(&current->signal->cred_guard_mutex);
1243 }
1244 EXPORT_SYMBOL(install_exec_creds);
1245 
1246 /*
1247  * determine how safe it is to execute the proposed program
1248  * - the caller must hold ->cred_guard_mutex to protect against
1249  *   PTRACE_ATTACH
1250  */
1251 static int check_unsafe_exec(struct linux_binprm *bprm)
1252 {
1253         struct task_struct *p = current, *t;
1254         unsigned n_fs;
1255         int res = 0;
1256 
1257         if (p->ptrace) {
1258                 if (p->ptrace & PT_PTRACE_CAP)
1259                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1260                 else
1261                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1262         }
1263 
1264         /*
1265          * This isn't strictly necessary, but it makes it harder for LSMs to
1266          * mess up.
1267          */
1268         if (current->no_new_privs)
1269                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1270 
1271         n_fs = 1;
1272         spin_lock(&p->fs->lock);
1273         rcu_read_lock();
1274         for (t = next_thread(p); t != p; t = next_thread(t)) {
1275                 if (t->fs == p->fs)
1276                         n_fs++;
1277         }
1278         rcu_read_unlock();
1279 
1280         if (p->fs->users > n_fs) {
1281                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1282         } else {
1283                 res = -EAGAIN;
1284                 if (!p->fs->in_exec) {
1285                         p->fs->in_exec = 1;
1286                         res = 1;
1287                 }
1288         }
1289         spin_unlock(&p->fs->lock);
1290 
1291         return res;
1292 }
1293 
1294 static void bprm_fill_uid(struct linux_binprm *bprm)
1295 {
1296         struct inode *inode;
1297         unsigned int mode;
1298         kuid_t uid;
1299         kgid_t gid;
1300 
1301         /* clear any previous set[ug]id data from a previous binary */
1302         bprm->cred->euid = current_euid();
1303         bprm->cred->egid = current_egid();
1304 
1305         if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
1306                 return;
1307 
1308         if (current->no_new_privs)
1309                 return;
1310 
1311         inode = file_inode(bprm->file);
1312         mode = ACCESS_ONCE(inode->i_mode);
1313         if (!(mode & (S_ISUID|S_ISGID)))
1314                 return;
1315 
1316         /* Be careful if suid/sgid is set */
1317         mutex_lock(&inode->i_mutex);
1318 
1319         /* reload atomically mode/uid/gid now that lock held */
1320         mode = inode->i_mode;
1321         uid = inode->i_uid;
1322         gid = inode->i_gid;
1323         mutex_unlock(&inode->i_mutex);
1324 
1325         /* We ignore suid/sgid if there are no mappings for them in the ns */
1326         if (!kuid_has_mapping(bprm->cred->user_ns, uid) ||
1327                  !kgid_has_mapping(bprm->cred->user_ns, gid))
1328                 return;
1329 
1330         if (mode & S_ISUID) {
1331                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1332                 bprm->cred->euid = uid;
1333         }
1334 
1335         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1336                 bprm->per_clear |= PER_CLEAR_ON_SETID;
1337                 bprm->cred->egid = gid;
1338         }
1339 }
1340 
1341 /* 
1342  * Fill the binprm structure from the inode. 
1343  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1344  *
1345  * This may be called multiple times for binary chains (scripts for example).
1346  */
1347 int prepare_binprm(struct linux_binprm *bprm)
1348 {
1349         int retval;
1350 
1351         if (bprm->file->f_op == NULL)
1352                 return -EACCES;
1353 
1354         bprm_fill_uid(bprm);
1355 
1356         /* fill in binprm security blob */
1357         retval = security_bprm_set_creds(bprm);
1358         if (retval)
1359                 return retval;
1360         bprm->cred_prepared = 1;
1361 
1362         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1363         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1364 }
1365 
1366 EXPORT_SYMBOL(prepare_binprm);
1367 
1368 /*
1369  * Arguments are '\0' separated strings found at the location bprm->p
1370  * points to; chop off the first by relocating brpm->p to right after
1371  * the first '\0' encountered.
1372  */
1373 int remove_arg_zero(struct linux_binprm *bprm)
1374 {
1375         int ret = 0;
1376         unsigned long offset;
1377         char *kaddr;
1378         struct page *page;
1379 
1380         if (!bprm->argc)
1381                 return 0;
1382 
1383         do {
1384                 offset = bprm->p & ~PAGE_MASK;
1385                 page = get_arg_page(bprm, bprm->p, 0);
1386                 if (!page) {
1387                         ret = -EFAULT;
1388                         goto out;
1389                 }
1390                 kaddr = kmap_atomic(page);
1391 
1392                 for (; offset < PAGE_SIZE && kaddr[offset];
1393                                 offset++, bprm->p++)
1394                         ;
1395 
1396                 kunmap_atomic(kaddr);
1397                 put_arg_page(page);
1398 
1399                 if (offset == PAGE_SIZE)
1400                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1401         } while (offset == PAGE_SIZE);
1402 
1403         bprm->p++;
1404         bprm->argc--;
1405         ret = 0;
1406 
1407 out:
1408         return ret;
1409 }
1410 EXPORT_SYMBOL(remove_arg_zero);
1411 
1412 /*
1413  * cycle the list of binary formats handler, until one recognizes the image
1414  */
1415 int search_binary_handler(struct linux_binprm *bprm)
1416 {
1417         unsigned int depth = bprm->recursion_depth;
1418         int try,retval;
1419         struct linux_binfmt *fmt;
1420         pid_t old_pid, old_vpid;
1421 
1422         /* This allows 4 levels of binfmt rewrites before failing hard. */
1423         if (depth > 5)
1424                 return -ELOOP;
1425 
1426         retval = security_bprm_check(bprm);
1427         if (retval)
1428                 return retval;
1429 
1430         retval = audit_bprm(bprm);
1431         if (retval)
1432                 return retval;
1433 
1434         /* Need to fetch pid before load_binary changes it */
1435         old_pid = current->pid;
1436         rcu_read_lock();
1437         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1438         rcu_read_unlock();
1439 
1440         retval = -ENOENT;
1441         for (try=0; try<2; try++) {
1442                 read_lock(&binfmt_lock);
1443                 list_for_each_entry(fmt, &formats, lh) {
1444                         int (*fn)(struct linux_binprm *) = fmt->load_binary;
1445                         if (!fn)
1446                                 continue;
1447                         if (!try_module_get(fmt->module))
1448                                 continue;
1449                         read_unlock(&binfmt_lock);
1450                         bprm->recursion_depth = depth + 1;
1451                         retval = fn(bprm);
1452                         bprm->recursion_depth = depth;
1453                         if (retval >= 0) {
1454                                 if (depth == 0) {
1455                                         trace_sched_process_exec(current, old_pid, bprm);
1456                                         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1457                                 }
1458                                 put_binfmt(fmt);
1459                                 allow_write_access(bprm->file);
1460                                 if (bprm->file)
1461                                         fput(bprm->file);
1462                                 bprm->file = NULL;
1463                                 current->did_exec = 1;
1464                                 proc_exec_connector(current);
1465                                 return retval;
1466                         }
1467                         read_lock(&binfmt_lock);
1468                         put_binfmt(fmt);
1469                         if (retval != -ENOEXEC || bprm->mm == NULL)
1470                                 break;
1471                         if (!bprm->file) {
1472                                 read_unlock(&binfmt_lock);
1473                                 return retval;
1474                         }
1475                 }
1476                 read_unlock(&binfmt_lock);
1477 #ifdef CONFIG_MODULES
1478                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1479                         break;
1480                 } else {
1481 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1482                         if (printable(bprm->buf[0]) &&
1483                             printable(bprm->buf[1]) &&
1484                             printable(bprm->buf[2]) &&
1485                             printable(bprm->buf[3]))
1486                                 break; /* -ENOEXEC */
1487                         if (try)
1488                                 break; /* -ENOEXEC */
1489                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1490                 }
1491 #else
1492                 break;
1493 #endif
1494         }
1495         return retval;
1496 }
1497 
1498 EXPORT_SYMBOL(search_binary_handler);
1499 
1500 /*
1501  * sys_execve() executes a new program.
1502  */
1503 static int do_execve_common(const char *filename,
1504                                 struct user_arg_ptr argv,
1505                                 struct user_arg_ptr envp)
1506 {
1507         struct linux_binprm *bprm;
1508         struct file *file;
1509         struct files_struct *displaced;
1510         bool clear_in_exec;
1511         int retval;
1512         const struct cred *cred = current_cred();
1513 
1514         /*
1515          * We move the actual failure in case of RLIMIT_NPROC excess from
1516          * set*uid() to execve() because too many poorly written programs
1517          * don't check setuid() return code.  Here we additionally recheck
1518          * whether NPROC limit is still exceeded.
1519          */
1520         if ((current->flags & PF_NPROC_EXCEEDED) &&
1521             atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1522                 retval = -EAGAIN;
1523                 goto out_ret;
1524         }
1525 
1526         /* We're below the limit (still or again), so we don't want to make
1527          * further execve() calls fail. */
1528         current->flags &= ~PF_NPROC_EXCEEDED;
1529 
1530         retval = unshare_files(&displaced);
1531         if (retval)
1532                 goto out_ret;
1533 
1534         retval = -ENOMEM;
1535         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1536         if (!bprm)
1537                 goto out_files;
1538 
1539         retval = prepare_bprm_creds(bprm);
1540         if (retval)
1541                 goto out_free;
1542 
1543         retval = check_unsafe_exec(bprm);
1544         if (retval < 0)
1545                 goto out_free;
1546         clear_in_exec = retval;
1547         current->in_execve = 1;
1548 
1549         file = open_exec(filename);
1550         retval = PTR_ERR(file);
1551         if (IS_ERR(file))
1552                 goto out_unmark;
1553 
1554         sched_exec();
1555 
1556         bprm->file = file;
1557         bprm->filename = filename;
1558         bprm->interp = filename;
1559 
1560         retval = bprm_mm_init(bprm);
1561         if (retval)
1562                 goto out_file;
1563 
1564         bprm->argc = count(argv, MAX_ARG_STRINGS);
1565         if ((retval = bprm->argc) < 0)
1566                 goto out;
1567 
1568         bprm->envc = count(envp, MAX_ARG_STRINGS);
1569         if ((retval = bprm->envc) < 0)
1570                 goto out;
1571 
1572         retval = prepare_binprm(bprm);
1573         if (retval < 0)
1574                 goto out;
1575 
1576         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1577         if (retval < 0)
1578                 goto out;
1579 
1580         bprm->exec = bprm->p;
1581         retval = copy_strings(bprm->envc, envp, bprm);
1582         if (retval < 0)
1583                 goto out;
1584 
1585         retval = copy_strings(bprm->argc, argv, bprm);
1586         if (retval < 0)
1587                 goto out;
1588 
1589         retval = ccs_search_binary_handler(bprm);
1590         if (retval < 0)
1591                 goto out;
1592 
1593         /* execve succeeded */
1594         current->fs->in_exec = 0;
1595         current->in_execve = 0;
1596         acct_update_integrals(current);
1597         free_bprm(bprm);
1598         if (displaced)
1599                 put_files_struct(displaced);
1600         return retval;
1601 
1602 out:
1603         if (bprm->mm) {
1604                 acct_arg_size(bprm, 0);
1605                 mmput(bprm->mm);
1606         }
1607 
1608 out_file:
1609         if (bprm->file) {
1610                 allow_write_access(bprm->file);
1611                 fput(bprm->file);
1612         }
1613 
1614 out_unmark:
1615         if (clear_in_exec)
1616                 current->fs->in_exec = 0;
1617         current->in_execve = 0;
1618 
1619 out_free:
1620         free_bprm(bprm);
1621 
1622 out_files:
1623         if (displaced)
1624                 reset_files_struct(displaced);
1625 out_ret:
1626         return retval;
1627 }
1628 
1629 int do_execve(const char *filename,
1630         const char __user *const __user *__argv,
1631         const char __user *const __user *__envp)
1632 {
1633         struct user_arg_ptr argv = { .ptr.native = __argv };
1634         struct user_arg_ptr envp = { .ptr.native = __envp };
1635         return do_execve_common(filename, argv, envp);
1636 }
1637 
1638 #ifdef CONFIG_COMPAT
1639 static int compat_do_execve(const char *filename,
1640         const compat_uptr_t __user *__argv,
1641         const compat_uptr_t __user *__envp)
1642 {
1643         struct user_arg_ptr argv = {
1644                 .is_compat = true,
1645                 .ptr.compat = __argv,
1646         };
1647         struct user_arg_ptr envp = {
1648                 .is_compat = true,
1649                 .ptr.compat = __envp,
1650         };
1651         return do_execve_common(filename, argv, envp);
1652 }
1653 #endif
1654 
1655 void set_binfmt(struct linux_binfmt *new)
1656 {
1657         struct mm_struct *mm = current->mm;
1658 
1659         if (mm->binfmt)
1660                 module_put(mm->binfmt->module);
1661 
1662         mm->binfmt = new;
1663         if (new)
1664                 __module_get(new->module);
1665 }
1666 
1667 EXPORT_SYMBOL(set_binfmt);
1668 
1669 /*
1670  * set_dumpable converts traditional three-value dumpable to two flags and
1671  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1672  * these bits are not changed atomically.  So get_dumpable can observe the
1673  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1674  * return either old dumpable or new one by paying attention to the order of
1675  * modifying the bits.
1676  *
1677  * dumpable |   mm->flags (binary)
1678  * old  new | initial interim  final
1679  * ---------+-----------------------
1680  *  0    1  |   00      01      01
1681  *  0    2  |   00      10(*)   11
1682  *  1    0  |   01      00      00
1683  *  1    2  |   01      11      11
1684  *  2    0  |   11      10(*)   00
1685  *  2    1  |   11      11      01
1686  *
1687  * (*) get_dumpable regards interim value of 10 as 11.
1688  */
1689 void set_dumpable(struct mm_struct *mm, int value)
1690 {
1691         switch (value) {
1692         case SUID_DUMP_DISABLE:
1693                 clear_bit(MMF_DUMPABLE, &mm->flags);
1694                 smp_wmb();
1695                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1696                 break;
1697         case SUID_DUMP_USER:
1698                 set_bit(MMF_DUMPABLE, &mm->flags);
1699                 smp_wmb();
1700                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1701                 break;
1702         case SUID_DUMP_ROOT:
1703                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1704                 smp_wmb();
1705                 set_bit(MMF_DUMPABLE, &mm->flags);
1706                 break;
1707         }
1708 }
1709 
1710 int __get_dumpable(unsigned long mm_flags)
1711 {
1712         int ret;
1713 
1714         ret = mm_flags & MMF_DUMPABLE_MASK;
1715         return (ret > SUID_DUMP_USER) ? SUID_DUMP_ROOT : ret;
1716 }
1717 
1718 /*
1719  * This returns the actual value of the suid_dumpable flag. For things
1720  * that are using this for checking for privilege transitions, it must
1721  * test against SUID_DUMP_USER rather than treating it as a boolean
1722  * value.
1723  */
1724 int get_dumpable(struct mm_struct *mm)
1725 {
1726         return __get_dumpable(mm->flags);
1727 }
1728 
1729 SYSCALL_DEFINE3(execve,
1730                 const char __user *, filename,
1731                 const char __user *const __user *, argv,
1732                 const char __user *const __user *, envp)
1733 {
1734         struct filename *path = getname(filename);
1735         int error = PTR_ERR(path);
1736         if (!IS_ERR(path)) {
1737                 error = do_execve(path->name, argv, envp);
1738                 putname(path);
1739         }
1740         return error;
1741 }
1742 #ifdef CONFIG_COMPAT
1743 asmlinkage long compat_sys_execve(const char __user * filename,
1744         const compat_uptr_t __user * argv,
1745         const compat_uptr_t __user * envp)
1746 {
1747         struct filename *path = getname(filename);
1748         int error = PTR_ERR(path);
1749         if (!IS_ERR(path)) {
1750                 error = compat_do_execve(path->name, argv, envp);
1751                 putname(path);
1752         }
1753         return error;
1754 }
1755 #endif
1756 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp