mm: mmu_gather rework
[linux-2.6.git] / 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/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/proc_fs.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
60 #include <asm/uaccess.h>
61 #include <asm/mmu_context.h>
62 #include <asm/tlb.h>
63 #include "internal.h"
64
65 int core_uses_pid;
66 char core_pattern[CORENAME_MAX_SIZE] = "core";
67 unsigned int core_pipe_limit;
68 int suid_dumpable = 0;
69
70 struct core_name {
71         char *corename;
72         int used, size;
73 };
74 static atomic_t call_count = ATOMIC_INIT(1);
75
76 /* The maximal length of core_pattern is also specified in sysctl.c */
77
78 static LIST_HEAD(formats);
79 static DEFINE_RWLOCK(binfmt_lock);
80
81 int __register_binfmt(struct linux_binfmt * fmt, int insert)
82 {
83         if (!fmt)
84                 return -EINVAL;
85         write_lock(&binfmt_lock);
86         insert ? list_add(&fmt->lh, &formats) :
87                  list_add_tail(&fmt->lh, &formats);
88         write_unlock(&binfmt_lock);
89         return 0;       
90 }
91
92 EXPORT_SYMBOL(__register_binfmt);
93
94 void unregister_binfmt(struct linux_binfmt * fmt)
95 {
96         write_lock(&binfmt_lock);
97         list_del(&fmt->lh);
98         write_unlock(&binfmt_lock);
99 }
100
101 EXPORT_SYMBOL(unregister_binfmt);
102
103 static inline void put_binfmt(struct linux_binfmt * fmt)
104 {
105         module_put(fmt->module);
106 }
107
108 /*
109  * Note that a shared library must be both readable and executable due to
110  * security reasons.
111  *
112  * Also note that we take the address to load from from the file itself.
113  */
114 SYSCALL_DEFINE1(uselib, const char __user *, library)
115 {
116         struct file *file;
117         char *tmp = getname(library);
118         int error = PTR_ERR(tmp);
119         static const struct open_flags uselib_flags = {
120                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
121                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
122                 .intent = LOOKUP_OPEN
123         };
124
125         if (IS_ERR(tmp))
126                 goto out;
127
128         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
129         putname(tmp);
130         error = PTR_ERR(file);
131         if (IS_ERR(file))
132                 goto out;
133
134         error = -EINVAL;
135         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
136                 goto exit;
137
138         error = -EACCES;
139         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
140                 goto exit;
141
142         fsnotify_open(file);
143
144         error = -ENOEXEC;
145         if(file->f_op) {
146                 struct linux_binfmt * fmt;
147
148                 read_lock(&binfmt_lock);
149                 list_for_each_entry(fmt, &formats, lh) {
150                         if (!fmt->load_shlib)
151                                 continue;
152                         if (!try_module_get(fmt->module))
153                                 continue;
154                         read_unlock(&binfmt_lock);
155                         error = fmt->load_shlib(file);
156                         read_lock(&binfmt_lock);
157                         put_binfmt(fmt);
158                         if (error != -ENOEXEC)
159                                 break;
160                 }
161                 read_unlock(&binfmt_lock);
162         }
163 exit:
164         fput(file);
165 out:
166         return error;
167 }
168
169 #ifdef CONFIG_MMU
170 /*
171  * The nascent bprm->mm is not visible until exec_mmap() but it can
172  * use a lot of memory, account these pages in current->mm temporary
173  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
174  * change the counter back via acct_arg_size(0).
175  */
176 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
177 {
178         struct mm_struct *mm = current->mm;
179         long diff = (long)(pages - bprm->vma_pages);
180
181         if (!mm || !diff)
182                 return;
183
184         bprm->vma_pages = pages;
185
186 #ifdef SPLIT_RSS_COUNTING
187         add_mm_counter(mm, MM_ANONPAGES, diff);
188 #else
189         spin_lock(&mm->page_table_lock);
190         add_mm_counter(mm, MM_ANONPAGES, diff);
191         spin_unlock(&mm->page_table_lock);
192 #endif
193 }
194
195 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
196                 int write)
197 {
198         struct page *page;
199         int ret;
200
201 #ifdef CONFIG_STACK_GROWSUP
202         if (write) {
203                 ret = expand_downwards(bprm->vma, pos);
204                 if (ret < 0)
205                         return NULL;
206         }
207 #endif
208         ret = get_user_pages(current, bprm->mm, pos,
209                         1, write, 1, &page, NULL);
210         if (ret <= 0)
211                 return NULL;
212
213         if (write) {
214                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
215                 struct rlimit *rlim;
216
217                 acct_arg_size(bprm, size / PAGE_SIZE);
218
219                 /*
220                  * We've historically supported up to 32 pages (ARG_MAX)
221                  * of argument strings even with small stacks
222                  */
223                 if (size <= ARG_MAX)
224                         return page;
225
226                 /*
227                  * Limit to 1/4-th the stack size for the argv+env strings.
228                  * This ensures that:
229                  *  - the remaining binfmt code will not run out of stack space,
230                  *  - the program will have a reasonable amount of stack left
231                  *    to work from.
232                  */
233                 rlim = current->signal->rlim;
234                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
235                         put_page(page);
236                         return NULL;
237                 }
238         }
239
240         return page;
241 }
242
243 static void put_arg_page(struct page *page)
244 {
245         put_page(page);
246 }
247
248 static void free_arg_page(struct linux_binprm *bprm, int i)
249 {
250 }
251
252 static void free_arg_pages(struct linux_binprm *bprm)
253 {
254 }
255
256 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
257                 struct page *page)
258 {
259         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
260 }
261
262 static int __bprm_mm_init(struct linux_binprm *bprm)
263 {
264         int err;
265         struct vm_area_struct *vma = NULL;
266         struct mm_struct *mm = bprm->mm;
267
268         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
269         if (!vma)
270                 return -ENOMEM;
271
272         down_write(&mm->mmap_sem);
273         vma->vm_mm = mm;
274
275         /*
276          * Place the stack at the largest stack address the architecture
277          * supports. Later, we'll move this to an appropriate place. We don't
278          * use STACK_TOP because that can depend on attributes which aren't
279          * configured yet.
280          */
281         BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
282         vma->vm_end = STACK_TOP_MAX;
283         vma->vm_start = vma->vm_end - PAGE_SIZE;
284         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
285         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
286         INIT_LIST_HEAD(&vma->anon_vma_chain);
287
288         err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
289         if (err)
290                 goto err;
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 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                 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
460                         if (fatal_signal_pending(current))
461                                 return -ERESTARTNOHAND;
462                         cond_resched();
463                 }
464         }
465         return i;
466 }
467
468 /*
469  * 'copy_strings()' copies argument/environment strings from the old
470  * processes's memory to the new process's stack.  The call to get_user_pages()
471  * ensures the destination page is created and not swapped out.
472  */
473 static int copy_strings(int argc, struct user_arg_ptr argv,
474                         struct linux_binprm *bprm)
475 {
476         struct page *kmapped_page = NULL;
477         char *kaddr = NULL;
478         unsigned long kpos = 0;
479         int ret;
480
481         while (argc-- > 0) {
482                 const char __user *str;
483                 int len;
484                 unsigned long pos;
485
486                 ret = -EFAULT;
487                 str = get_user_arg_ptr(argv, argc);
488                 if (IS_ERR(str))
489                         goto out;
490
491                 len = strnlen_user(str, MAX_ARG_STRLEN);
492                 if (!len)
493                         goto out;
494
495                 ret = -E2BIG;
496                 if (!valid_arg_len(bprm, len))
497                         goto out;
498
499                 /* We're going to work our way backwords. */
500                 pos = bprm->p;
501                 str += len;
502                 bprm->p -= len;
503
504                 while (len > 0) {
505                         int offset, bytes_to_copy;
506
507                         if (fatal_signal_pending(current)) {
508                                 ret = -ERESTARTNOHAND;
509                                 goto out;
510                         }
511                         cond_resched();
512
513                         offset = pos % PAGE_SIZE;
514                         if (offset == 0)
515                                 offset = PAGE_SIZE;
516
517                         bytes_to_copy = offset;
518                         if (bytes_to_copy > len)
519                                 bytes_to_copy = len;
520
521                         offset -= bytes_to_copy;
522                         pos -= bytes_to_copy;
523                         str -= bytes_to_copy;
524                         len -= bytes_to_copy;
525
526                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
527                                 struct page *page;
528
529                                 page = get_arg_page(bprm, pos, 1);
530                                 if (!page) {
531                                         ret = -E2BIG;
532                                         goto out;
533                                 }
534
535                                 if (kmapped_page) {
536                                         flush_kernel_dcache_page(kmapped_page);
537                                         kunmap(kmapped_page);
538                                         put_arg_page(kmapped_page);
539                                 }
540                                 kmapped_page = page;
541                                 kaddr = kmap(kmapped_page);
542                                 kpos = pos & PAGE_MASK;
543                                 flush_arg_page(bprm, kpos, kmapped_page);
544                         }
545                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
546                                 ret = -EFAULT;
547                                 goto out;
548                         }
549                 }
550         }
551         ret = 0;
552 out:
553         if (kmapped_page) {
554                 flush_kernel_dcache_page(kmapped_page);
555                 kunmap(kmapped_page);
556                 put_arg_page(kmapped_page);
557         }
558         return ret;
559 }
560
561 /*
562  * Like copy_strings, but get argv and its values from kernel memory.
563  */
564 int copy_strings_kernel(int argc, const char *const *__argv,
565                         struct linux_binprm *bprm)
566 {
567         int r;
568         mm_segment_t oldfs = get_fs();
569         struct user_arg_ptr argv = {
570                 .ptr.native = (const char __user *const  __user *)__argv,
571         };
572
573         set_fs(KERNEL_DS);
574         r = copy_strings(argc, argv, bprm);
575         set_fs(oldfs);
576
577         return r;
578 }
579 EXPORT_SYMBOL(copy_strings_kernel);
580
581 #ifdef CONFIG_MMU
582
583 /*
584  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
585  * the binfmt code determines where the new stack should reside, we shift it to
586  * its final location.  The process proceeds as follows:
587  *
588  * 1) Use shift to calculate the new vma endpoints.
589  * 2) Extend vma to cover both the old and new ranges.  This ensures the
590  *    arguments passed to subsequent functions are consistent.
591  * 3) Move vma's page tables to the new range.
592  * 4) Free up any cleared pgd range.
593  * 5) Shrink the vma to cover only the new range.
594  */
595 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
596 {
597         struct mm_struct *mm = vma->vm_mm;
598         unsigned long old_start = vma->vm_start;
599         unsigned long old_end = vma->vm_end;
600         unsigned long length = old_end - old_start;
601         unsigned long new_start = old_start - shift;
602         unsigned long new_end = old_end - shift;
603         struct mmu_gather tlb;
604
605         BUG_ON(new_start > new_end);
606
607         /*
608          * ensure there are no vmas between where we want to go
609          * and where we are
610          */
611         if (vma != find_vma(mm, new_start))
612                 return -EFAULT;
613
614         /*
615          * cover the whole range: [new_start, old_end)
616          */
617         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
618                 return -ENOMEM;
619
620         /*
621          * move the page tables downwards, on failure we rely on
622          * process cleanup to remove whatever mess we made.
623          */
624         if (length != move_page_tables(vma, old_start,
625                                        vma, new_start, length))
626                 return -ENOMEM;
627
628         lru_add_drain();
629         tlb_gather_mmu(&tlb, mm, 0);
630         if (new_end > old_start) {
631                 /*
632                  * when the old and new regions overlap clear from new_end.
633                  */
634                 free_pgd_range(&tlb, new_end, old_end, new_end,
635                         vma->vm_next ? vma->vm_next->vm_start : 0);
636         } else {
637                 /*
638                  * otherwise, clean from old_start; this is done to not touch
639                  * the address space in [new_end, old_start) some architectures
640                  * have constraints on va-space that make this illegal (IA64) -
641                  * for the others its just a little faster.
642                  */
643                 free_pgd_range(&tlb, old_start, old_end, new_end,
644                         vma->vm_next ? vma->vm_next->vm_start : 0);
645         }
646         tlb_finish_mmu(&tlb, new_end, old_end);
647
648         /*
649          * Shrink the vma to just the new range.  Always succeeds.
650          */
651         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
652
653         return 0;
654 }
655
656 /*
657  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
658  * the stack is optionally relocated, and some extra space is added.
659  */
660 int setup_arg_pages(struct linux_binprm *bprm,
661                     unsigned long stack_top,
662                     int executable_stack)
663 {
664         unsigned long ret;
665         unsigned long stack_shift;
666         struct mm_struct *mm = current->mm;
667         struct vm_area_struct *vma = bprm->vma;
668         struct vm_area_struct *prev = NULL;
669         unsigned long vm_flags;
670         unsigned long stack_base;
671         unsigned long stack_size;
672         unsigned long stack_expand;
673         unsigned long rlim_stack;
674
675 #ifdef CONFIG_STACK_GROWSUP
676         /* Limit stack size to 1GB */
677         stack_base = rlimit_max(RLIMIT_STACK);
678         if (stack_base > (1 << 30))
679                 stack_base = 1 << 30;
680
681         /* Make sure we didn't let the argument array grow too large. */
682         if (vma->vm_end - vma->vm_start > stack_base)
683                 return -ENOMEM;
684
685         stack_base = PAGE_ALIGN(stack_top - stack_base);
686
687         stack_shift = vma->vm_start - stack_base;
688         mm->arg_start = bprm->p - stack_shift;
689         bprm->p = vma->vm_end - stack_shift;
690 #else
691         stack_top = arch_align_stack(stack_top);
692         stack_top = PAGE_ALIGN(stack_top);
693
694         if (unlikely(stack_top < mmap_min_addr) ||
695             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
696                 return -ENOMEM;
697
698         stack_shift = vma->vm_end - stack_top;
699
700         bprm->p -= stack_shift;
701         mm->arg_start = bprm->p;
702 #endif
703
704         if (bprm->loader)
705                 bprm->loader -= stack_shift;
706         bprm->exec -= stack_shift;
707
708         down_write(&mm->mmap_sem);
709         vm_flags = VM_STACK_FLAGS;
710
711         /*
712          * Adjust stack execute permissions; explicitly enable for
713          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
714          * (arch default) otherwise.
715          */
716         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
717                 vm_flags |= VM_EXEC;
718         else if (executable_stack == EXSTACK_DISABLE_X)
719                 vm_flags &= ~VM_EXEC;
720         vm_flags |= mm->def_flags;
721         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
722
723         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
724                         vm_flags);
725         if (ret)
726                 goto out_unlock;
727         BUG_ON(prev != vma);
728
729         /* Move stack pages down in memory. */
730         if (stack_shift) {
731                 ret = shift_arg_pages(vma, stack_shift);
732                 if (ret)
733                         goto out_unlock;
734         }
735
736         /* mprotect_fixup is overkill to remove the temporary stack flags */
737         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
738
739         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
740         stack_size = vma->vm_end - vma->vm_start;
741         /*
742          * Align this down to a page boundary as expand_stack
743          * will align it up.
744          */
745         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
746 #ifdef CONFIG_STACK_GROWSUP
747         if (stack_size + stack_expand > rlim_stack)
748                 stack_base = vma->vm_start + rlim_stack;
749         else
750                 stack_base = vma->vm_end + stack_expand;
751 #else
752         if (stack_size + stack_expand > rlim_stack)
753                 stack_base = vma->vm_end - rlim_stack;
754         else
755                 stack_base = vma->vm_start - stack_expand;
756 #endif
757         current->mm->start_stack = bprm->p;
758         ret = expand_stack(vma, stack_base);
759         if (ret)
760                 ret = -EFAULT;
761
762 out_unlock:
763         up_write(&mm->mmap_sem);
764         return ret;
765 }
766 EXPORT_SYMBOL(setup_arg_pages);
767
768 #endif /* CONFIG_MMU */
769
770 struct file *open_exec(const char *name)
771 {
772         struct file *file;
773         int err;
774         static const struct open_flags open_exec_flags = {
775                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
776                 .acc_mode = MAY_EXEC | MAY_OPEN,
777                 .intent = LOOKUP_OPEN
778         };
779
780         file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
781         if (IS_ERR(file))
782                 goto out;
783
784         err = -EACCES;
785         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
786                 goto exit;
787
788         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
789                 goto exit;
790
791         fsnotify_open(file);
792
793         err = deny_write_access(file);
794         if (err)
795                 goto exit;
796
797 out:
798         return file;
799
800 exit:
801         fput(file);
802         return ERR_PTR(err);
803 }
804 EXPORT_SYMBOL(open_exec);
805
806 int kernel_read(struct file *file, loff_t offset,
807                 char *addr, unsigned long count)
808 {
809         mm_segment_t old_fs;
810         loff_t pos = offset;
811         int result;
812
813         old_fs = get_fs();
814         set_fs(get_ds());
815         /* The cast to a user pointer is valid due to the set_fs() */
816         result = vfs_read(file, (void __user *)addr, count, &pos);
817         set_fs(old_fs);
818         return result;
819 }
820
821 EXPORT_SYMBOL(kernel_read);
822
823 static int exec_mmap(struct mm_struct *mm)
824 {
825         struct task_struct *tsk;
826         struct mm_struct * old_mm, *active_mm;
827
828         /* Notify parent that we're no longer interested in the old VM */
829         tsk = current;
830         old_mm = current->mm;
831         sync_mm_rss(tsk, old_mm);
832         mm_release(tsk, old_mm);
833
834         if (old_mm) {
835                 /*
836                  * Make sure that if there is a core dump in progress
837                  * for the old mm, we get out and die instead of going
838                  * through with the exec.  We must hold mmap_sem around
839                  * checking core_state and changing tsk->mm.
840                  */
841                 down_read(&old_mm->mmap_sem);
842                 if (unlikely(old_mm->core_state)) {
843                         up_read(&old_mm->mmap_sem);
844                         return -EINTR;
845                 }
846         }
847         task_lock(tsk);
848         active_mm = tsk->active_mm;
849         tsk->mm = mm;
850         tsk->active_mm = mm;
851         activate_mm(active_mm, mm);
852         if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
853                 atomic_dec(&old_mm->oom_disable_count);
854                 atomic_inc(&tsk->mm->oom_disable_count);
855         }
856         task_unlock(tsk);
857         arch_pick_mmap_layout(mm);
858         if (old_mm) {
859                 up_read(&old_mm->mmap_sem);
860                 BUG_ON(active_mm != old_mm);
861                 mm_update_next_owner(old_mm);
862                 mmput(old_mm);
863                 return 0;
864         }
865         mmdrop(active_mm);
866         return 0;
867 }
868
869 /*
870  * This function makes sure the current process has its own signal table,
871  * so that flush_signal_handlers can later reset the handlers without
872  * disturbing other processes.  (Other processes might share the signal
873  * table via the CLONE_SIGHAND option to clone().)
874  */
875 static int de_thread(struct task_struct *tsk)
876 {
877         struct signal_struct *sig = tsk->signal;
878         struct sighand_struct *oldsighand = tsk->sighand;
879         spinlock_t *lock = &oldsighand->siglock;
880
881         if (thread_group_empty(tsk))
882                 goto no_thread_group;
883
884         /*
885          * Kill all other threads in the thread group.
886          */
887         spin_lock_irq(lock);
888         if (signal_group_exit(sig)) {
889                 /*
890                  * Another group action in progress, just
891                  * return so that the signal is processed.
892                  */
893                 spin_unlock_irq(lock);
894                 return -EAGAIN;
895         }
896
897         sig->group_exit_task = tsk;
898         sig->notify_count = zap_other_threads(tsk);
899         if (!thread_group_leader(tsk))
900                 sig->notify_count--;
901
902         while (sig->notify_count) {
903                 __set_current_state(TASK_UNINTERRUPTIBLE);
904                 spin_unlock_irq(lock);
905                 schedule();
906                 spin_lock_irq(lock);
907         }
908         spin_unlock_irq(lock);
909
910         /*
911          * At this point all other threads have exited, all we have to
912          * do is to wait for the thread group leader to become inactive,
913          * and to assume its PID:
914          */
915         if (!thread_group_leader(tsk)) {
916                 struct task_struct *leader = tsk->group_leader;
917
918                 sig->notify_count = -1; /* for exit_notify() */
919                 for (;;) {
920                         write_lock_irq(&tasklist_lock);
921                         if (likely(leader->exit_state))
922                                 break;
923                         __set_current_state(TASK_UNINTERRUPTIBLE);
924                         write_unlock_irq(&tasklist_lock);
925                         schedule();
926                 }
927
928                 /*
929                  * The only record we have of the real-time age of a
930                  * process, regardless of execs it's done, is start_time.
931                  * All the past CPU time is accumulated in signal_struct
932                  * from sister threads now dead.  But in this non-leader
933                  * exec, nothing survives from the original leader thread,
934                  * whose birth marks the true age of this process now.
935                  * When we take on its identity by switching to its PID, we
936                  * also take its birthdate (always earlier than our own).
937                  */
938                 tsk->start_time = leader->start_time;
939
940                 BUG_ON(!same_thread_group(leader, tsk));
941                 BUG_ON(has_group_leader_pid(tsk));
942                 /*
943                  * An exec() starts a new thread group with the
944                  * TGID of the previous thread group. Rehash the
945                  * two threads with a switched PID, and release
946                  * the former thread group leader:
947                  */
948
949                 /* Become a process group leader with the old leader's pid.
950                  * The old leader becomes a thread of the this thread group.
951                  * Note: The old leader also uses this pid until release_task
952                  *       is called.  Odd but simple and correct.
953                  */
954                 detach_pid(tsk, PIDTYPE_PID);
955                 tsk->pid = leader->pid;
956                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
957                 transfer_pid(leader, tsk, PIDTYPE_PGID);
958                 transfer_pid(leader, tsk, PIDTYPE_SID);
959
960                 list_replace_rcu(&leader->tasks, &tsk->tasks);
961                 list_replace_init(&leader->sibling, &tsk->sibling);
962
963                 tsk->group_leader = tsk;
964                 leader->group_leader = tsk;
965
966                 tsk->exit_signal = SIGCHLD;
967
968                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
969                 leader->exit_state = EXIT_DEAD;
970                 write_unlock_irq(&tasklist_lock);
971
972                 release_task(leader);
973         }
974
975         sig->group_exit_task = NULL;
976         sig->notify_count = 0;
977
978 no_thread_group:
979         if (current->mm)
980                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
981
982         exit_itimers(sig);
983         flush_itimer_signals();
984
985         if (atomic_read(&oldsighand->count) != 1) {
986                 struct sighand_struct *newsighand;
987                 /*
988                  * This ->sighand is shared with the CLONE_SIGHAND
989                  * but not CLONE_THREAD task, switch to the new one.
990                  */
991                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
992                 if (!newsighand)
993                         return -ENOMEM;
994
995                 atomic_set(&newsighand->count, 1);
996                 memcpy(newsighand->action, oldsighand->action,
997                        sizeof(newsighand->action));
998
999                 write_lock_irq(&tasklist_lock);
1000                 spin_lock(&oldsighand->siglock);
1001                 rcu_assign_pointer(tsk->sighand, newsighand);
1002                 spin_unlock(&oldsighand->siglock);
1003                 write_unlock_irq(&tasklist_lock);
1004
1005                 __cleanup_sighand(oldsighand);
1006         }
1007
1008         BUG_ON(!thread_group_leader(tsk));
1009         return 0;
1010 }
1011
1012 /*
1013  * These functions flushes out all traces of the currently running executable
1014  * so that a new one can be started
1015  */
1016 static void flush_old_files(struct files_struct * files)
1017 {
1018         long j = -1;
1019         struct fdtable *fdt;
1020
1021         spin_lock(&files->file_lock);
1022         for (;;) {
1023                 unsigned long set, i;
1024
1025                 j++;
1026                 i = j * __NFDBITS;
1027                 fdt = files_fdtable(files);
1028                 if (i >= fdt->max_fds)
1029                         break;
1030                 set = fdt->close_on_exec->fds_bits[j];
1031                 if (!set)
1032                         continue;
1033                 fdt->close_on_exec->fds_bits[j] = 0;
1034                 spin_unlock(&files->file_lock);
1035                 for ( ; set ; i++,set >>= 1) {
1036                         if (set & 1) {
1037                                 sys_close(i);
1038                         }
1039                 }
1040                 spin_lock(&files->file_lock);
1041
1042         }
1043         spin_unlock(&files->file_lock);
1044 }
1045
1046 char *get_task_comm(char *buf, struct task_struct *tsk)
1047 {
1048         /* buf must be at least sizeof(tsk->comm) in size */
1049         task_lock(tsk);
1050         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1051         task_unlock(tsk);
1052         return buf;
1053 }
1054 EXPORT_SYMBOL_GPL(get_task_comm);
1055
1056 void set_task_comm(struct task_struct *tsk, char *buf)
1057 {
1058         task_lock(tsk);
1059
1060         /*
1061          * Threads may access current->comm without holding
1062          * the task lock, so write the string carefully.
1063          * Readers without a lock may see incomplete new
1064          * names but are safe from non-terminating string reads.
1065          */
1066         memset(tsk->comm, 0, TASK_COMM_LEN);
1067         wmb();
1068         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1069         task_unlock(tsk);
1070         perf_event_comm(tsk);
1071 }
1072
1073 int flush_old_exec(struct linux_binprm * bprm)
1074 {
1075         int retval;
1076
1077         /*
1078          * Make sure we have a private signal table and that
1079          * we are unassociated from the previous thread group.
1080          */
1081         retval = de_thread(current);
1082         if (retval)
1083                 goto out;
1084
1085         set_mm_exe_file(bprm->mm, bprm->file);
1086
1087         /*
1088          * Release all of the old mmap stuff
1089          */
1090         acct_arg_size(bprm, 0);
1091         retval = exec_mmap(bprm->mm);
1092         if (retval)
1093                 goto out;
1094
1095         bprm->mm = NULL;                /* We're using it now */
1096
1097         current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1098         flush_thread();
1099         current->personality &= ~bprm->per_clear;
1100
1101         return 0;
1102
1103 out:
1104         return retval;
1105 }
1106 EXPORT_SYMBOL(flush_old_exec);
1107
1108 void setup_new_exec(struct linux_binprm * bprm)
1109 {
1110         int i, ch;
1111         const char *name;
1112         char tcomm[sizeof(current->comm)];
1113
1114         arch_pick_mmap_layout(current->mm);
1115
1116         /* This is the point of no return */
1117         current->sas_ss_sp = current->sas_ss_size = 0;
1118
1119         if (current_euid() == current_uid() && current_egid() == current_gid())
1120                 set_dumpable(current->mm, 1);
1121         else
1122                 set_dumpable(current->mm, suid_dumpable);
1123
1124         name = bprm->filename;
1125
1126         /* Copies the binary name from after last slash */
1127         for (i=0; (ch = *(name++)) != '\0';) {
1128                 if (ch == '/')
1129                         i = 0; /* overwrite what we wrote */
1130                 else
1131                         if (i < (sizeof(tcomm) - 1))
1132                                 tcomm[i++] = ch;
1133         }
1134         tcomm[i] = '\0';
1135         set_task_comm(current, tcomm);
1136
1137         /* Set the new mm task size. We have to do that late because it may
1138          * depend on TIF_32BIT which is only updated in flush_thread() on
1139          * some architectures like powerpc
1140          */
1141         current->mm->task_size = TASK_SIZE;
1142
1143         /* install the new credentials */
1144         if (bprm->cred->uid != current_euid() ||
1145             bprm->cred->gid != current_egid()) {
1146                 current->pdeath_signal = 0;
1147         } else if (file_permission(bprm->file, MAY_READ) ||
1148                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1149                 set_dumpable(current->mm, suid_dumpable);
1150         }
1151
1152         /*
1153          * Flush performance counters when crossing a
1154          * security domain:
1155          */
1156         if (!get_dumpable(current->mm))
1157                 perf_event_exit_task(current);
1158
1159         /* An exec changes our domain. We are no longer part of the thread
1160            group */
1161
1162         current->self_exec_id++;
1163                         
1164         flush_signal_handlers(current, 0);
1165         flush_old_files(current->files);
1166 }
1167 EXPORT_SYMBOL(setup_new_exec);
1168
1169 /*
1170  * Prepare credentials and lock ->cred_guard_mutex.
1171  * install_exec_creds() commits the new creds and drops the lock.
1172  * Or, if exec fails before, free_bprm() should release ->cred and
1173  * and unlock.
1174  */
1175 int prepare_bprm_creds(struct linux_binprm *bprm)
1176 {
1177         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1178                 return -ERESTARTNOINTR;
1179
1180         bprm->cred = prepare_exec_creds();
1181         if (likely(bprm->cred))
1182                 return 0;
1183
1184         mutex_unlock(&current->signal->cred_guard_mutex);
1185         return -ENOMEM;
1186 }
1187
1188 void free_bprm(struct linux_binprm *bprm)
1189 {
1190         free_arg_pages(bprm);
1191         if (bprm->cred) {
1192                 mutex_unlock(&current->signal->cred_guard_mutex);
1193                 abort_creds(bprm->cred);
1194         }
1195         kfree(bprm);
1196 }
1197
1198 /*
1199  * install the new credentials for this executable
1200  */
1201 void install_exec_creds(struct linux_binprm *bprm)
1202 {
1203         security_bprm_committing_creds(bprm);
1204
1205         commit_creds(bprm->cred);
1206         bprm->cred = NULL;
1207         /*
1208          * cred_guard_mutex must be held at least to this point to prevent
1209          * ptrace_attach() from altering our determination of the task's
1210          * credentials; any time after this it may be unlocked.
1211          */
1212         security_bprm_committed_creds(bprm);
1213         mutex_unlock(&current->signal->cred_guard_mutex);
1214 }
1215 EXPORT_SYMBOL(install_exec_creds);
1216
1217 /*
1218  * determine how safe it is to execute the proposed program
1219  * - the caller must hold ->cred_guard_mutex to protect against
1220  *   PTRACE_ATTACH
1221  */
1222 int check_unsafe_exec(struct linux_binprm *bprm)
1223 {
1224         struct task_struct *p = current, *t;
1225         unsigned n_fs;
1226         int res = 0;
1227
1228         bprm->unsafe = tracehook_unsafe_exec(p);
1229
1230         n_fs = 1;
1231         spin_lock(&p->fs->lock);
1232         rcu_read_lock();
1233         for (t = next_thread(p); t != p; t = next_thread(t)) {
1234                 if (t->fs == p->fs)
1235                         n_fs++;
1236         }
1237         rcu_read_unlock();
1238
1239         if (p->fs->users > n_fs) {
1240                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1241         } else {
1242                 res = -EAGAIN;
1243                 if (!p->fs->in_exec) {
1244                         p->fs->in_exec = 1;
1245                         res = 1;
1246                 }
1247         }
1248         spin_unlock(&p->fs->lock);
1249
1250         return res;
1251 }
1252
1253 /* 
1254  * Fill the binprm structure from the inode. 
1255  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1256  *
1257  * This may be called multiple times for binary chains (scripts for example).
1258  */
1259 int prepare_binprm(struct linux_binprm *bprm)
1260 {
1261         umode_t mode;
1262         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1263         int retval;
1264
1265         mode = inode->i_mode;
1266         if (bprm->file->f_op == NULL)
1267                 return -EACCES;
1268
1269         /* clear any previous set[ug]id data from a previous binary */
1270         bprm->cred->euid = current_euid();
1271         bprm->cred->egid = current_egid();
1272
1273         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1274                 /* Set-uid? */
1275                 if (mode & S_ISUID) {
1276                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1277                         bprm->cred->euid = inode->i_uid;
1278                 }
1279
1280                 /* Set-gid? */
1281                 /*
1282                  * If setgid is set but no group execute bit then this
1283                  * is a candidate for mandatory locking, not a setgid
1284                  * executable.
1285                  */
1286                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1287                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1288                         bprm->cred->egid = inode->i_gid;
1289                 }
1290         }
1291
1292         /* fill in binprm security blob */
1293         retval = security_bprm_set_creds(bprm);
1294         if (retval)
1295                 return retval;
1296         bprm->cred_prepared = 1;
1297
1298         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1299         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1300 }
1301
1302 EXPORT_SYMBOL(prepare_binprm);
1303
1304 /*
1305  * Arguments are '\0' separated strings found at the location bprm->p
1306  * points to; chop off the first by relocating brpm->p to right after
1307  * the first '\0' encountered.
1308  */
1309 int remove_arg_zero(struct linux_binprm *bprm)
1310 {
1311         int ret = 0;
1312         unsigned long offset;
1313         char *kaddr;
1314         struct page *page;
1315
1316         if (!bprm->argc)
1317                 return 0;
1318
1319         do {
1320                 offset = bprm->p & ~PAGE_MASK;
1321                 page = get_arg_page(bprm, bprm->p, 0);
1322                 if (!page) {
1323                         ret = -EFAULT;
1324                         goto out;
1325                 }
1326                 kaddr = kmap_atomic(page, KM_USER0);
1327
1328                 for (; offset < PAGE_SIZE && kaddr[offset];
1329                                 offset++, bprm->p++)
1330                         ;
1331
1332                 kunmap_atomic(kaddr, KM_USER0);
1333                 put_arg_page(page);
1334
1335                 if (offset == PAGE_SIZE)
1336                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1337         } while (offset == PAGE_SIZE);
1338
1339         bprm->p++;
1340         bprm->argc--;
1341         ret = 0;
1342
1343 out:
1344         return ret;
1345 }
1346 EXPORT_SYMBOL(remove_arg_zero);
1347
1348 /*
1349  * cycle the list of binary formats handler, until one recognizes the image
1350  */
1351 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1352 {
1353         unsigned int depth = bprm->recursion_depth;
1354         int try,retval;
1355         struct linux_binfmt *fmt;
1356
1357         retval = security_bprm_check(bprm);
1358         if (retval)
1359                 return retval;
1360
1361         /* kernel module loader fixup */
1362         /* so we don't try to load run modprobe in kernel space. */
1363         set_fs(USER_DS);
1364
1365         retval = audit_bprm(bprm);
1366         if (retval)
1367                 return retval;
1368
1369         retval = -ENOENT;
1370         for (try=0; try<2; try++) {
1371                 read_lock(&binfmt_lock);
1372                 list_for_each_entry(fmt, &formats, lh) {
1373                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1374                         if (!fn)
1375                                 continue;
1376                         if (!try_module_get(fmt->module))
1377                                 continue;
1378                         read_unlock(&binfmt_lock);
1379                         retval = fn(bprm, regs);
1380                         /*
1381                          * Restore the depth counter to its starting value
1382                          * in this call, so we don't have to rely on every
1383                          * load_binary function to restore it on return.
1384                          */
1385                         bprm->recursion_depth = depth;
1386                         if (retval >= 0) {
1387                                 if (depth == 0)
1388                                         tracehook_report_exec(fmt, bprm, regs);
1389                                 put_binfmt(fmt);
1390                                 allow_write_access(bprm->file);
1391                                 if (bprm->file)
1392                                         fput(bprm->file);
1393                                 bprm->file = NULL;
1394                                 current->did_exec = 1;
1395                                 proc_exec_connector(current);
1396                                 return retval;
1397                         }
1398                         read_lock(&binfmt_lock);
1399                         put_binfmt(fmt);
1400                         if (retval != -ENOEXEC || bprm->mm == NULL)
1401                                 break;
1402                         if (!bprm->file) {
1403                                 read_unlock(&binfmt_lock);
1404                                 return retval;
1405                         }
1406                 }
1407                 read_unlock(&binfmt_lock);
1408                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1409                         break;
1410 #ifdef CONFIG_MODULES
1411                 } else {
1412 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1413                         if (printable(bprm->buf[0]) &&
1414                             printable(bprm->buf[1]) &&
1415                             printable(bprm->buf[2]) &&
1416                             printable(bprm->buf[3]))
1417                                 break; /* -ENOEXEC */
1418                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1419 #endif
1420                 }
1421         }
1422         return retval;
1423 }
1424
1425 EXPORT_SYMBOL(search_binary_handler);
1426
1427 /*
1428  * sys_execve() executes a new program.
1429  */
1430 static int do_execve_common(const char *filename,
1431                                 struct user_arg_ptr argv,
1432                                 struct user_arg_ptr envp,
1433                                 struct pt_regs *regs)
1434 {
1435         struct linux_binprm *bprm;
1436         struct file *file;
1437         struct files_struct *displaced;
1438         bool clear_in_exec;
1439         int retval;
1440
1441         retval = unshare_files(&displaced);
1442         if (retval)
1443                 goto out_ret;
1444
1445         retval = -ENOMEM;
1446         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1447         if (!bprm)
1448                 goto out_files;
1449
1450         retval = prepare_bprm_creds(bprm);
1451         if (retval)
1452                 goto out_free;
1453
1454         retval = check_unsafe_exec(bprm);
1455         if (retval < 0)
1456                 goto out_free;
1457         clear_in_exec = retval;
1458         current->in_execve = 1;
1459
1460         file = open_exec(filename);
1461         retval = PTR_ERR(file);
1462         if (IS_ERR(file))
1463                 goto out_unmark;
1464
1465         sched_exec();
1466
1467         bprm->file = file;
1468         bprm->filename = filename;
1469         bprm->interp = filename;
1470
1471         retval = bprm_mm_init(bprm);
1472         if (retval)
1473                 goto out_file;
1474
1475         bprm->argc = count(argv, MAX_ARG_STRINGS);
1476         if ((retval = bprm->argc) < 0)
1477                 goto out;
1478
1479         bprm->envc = count(envp, MAX_ARG_STRINGS);
1480         if ((retval = bprm->envc) < 0)
1481                 goto out;
1482
1483         retval = prepare_binprm(bprm);
1484         if (retval < 0)
1485                 goto out;
1486
1487         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1488         if (retval < 0)
1489                 goto out;
1490
1491         bprm->exec = bprm->p;
1492         retval = copy_strings(bprm->envc, envp, bprm);
1493         if (retval < 0)
1494                 goto out;
1495
1496         retval = copy_strings(bprm->argc, argv, bprm);
1497         if (retval < 0)
1498                 goto out;
1499
1500         retval = search_binary_handler(bprm,regs);
1501         if (retval < 0)
1502                 goto out;
1503
1504         /* execve succeeded */
1505         current->fs->in_exec = 0;
1506         current->in_execve = 0;
1507         acct_update_integrals(current);
1508         free_bprm(bprm);
1509         if (displaced)
1510                 put_files_struct(displaced);
1511         return retval;
1512
1513 out:
1514         if (bprm->mm) {
1515                 acct_arg_size(bprm, 0);
1516                 mmput(bprm->mm);
1517         }
1518
1519 out_file:
1520         if (bprm->file) {
1521                 allow_write_access(bprm->file);
1522                 fput(bprm->file);
1523         }
1524
1525 out_unmark:
1526         if (clear_in_exec)
1527                 current->fs->in_exec = 0;
1528         current->in_execve = 0;
1529
1530 out_free:
1531         free_bprm(bprm);
1532
1533 out_files:
1534         if (displaced)
1535                 reset_files_struct(displaced);
1536 out_ret:
1537         return retval;
1538 }
1539
1540 int do_execve(const char *filename,
1541         const char __user *const __user *__argv,
1542         const char __user *const __user *__envp,
1543         struct pt_regs *regs)
1544 {
1545         struct user_arg_ptr argv = { .ptr.native = __argv };
1546         struct user_arg_ptr envp = { .ptr.native = __envp };
1547         return do_execve_common(filename, argv, envp, regs);
1548 }
1549
1550 #ifdef CONFIG_COMPAT
1551 int compat_do_execve(char *filename,
1552         compat_uptr_t __user *__argv,
1553         compat_uptr_t __user *__envp,
1554         struct pt_regs *regs)
1555 {
1556         struct user_arg_ptr argv = {
1557                 .is_compat = true,
1558                 .ptr.compat = __argv,
1559         };
1560         struct user_arg_ptr envp = {
1561                 .is_compat = true,
1562                 .ptr.compat = __envp,
1563         };
1564         return do_execve_common(filename, argv, envp, regs);
1565 }
1566 #endif
1567
1568 void set_binfmt(struct linux_binfmt *new)
1569 {
1570         struct mm_struct *mm = current->mm;
1571
1572         if (mm->binfmt)
1573                 module_put(mm->binfmt->module);
1574
1575         mm->binfmt = new;
1576         if (new)
1577                 __module_get(new->module);
1578 }
1579
1580 EXPORT_SYMBOL(set_binfmt);
1581
1582 static int expand_corename(struct core_name *cn)
1583 {
1584         char *old_corename = cn->corename;
1585
1586         cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1587         cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1588
1589         if (!cn->corename) {
1590                 kfree(old_corename);
1591                 return -ENOMEM;
1592         }
1593
1594         return 0;
1595 }
1596
1597 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1598 {
1599         char *cur;
1600         int need;
1601         int ret;
1602         va_list arg;
1603
1604         va_start(arg, fmt);
1605         need = vsnprintf(NULL, 0, fmt, arg);
1606         va_end(arg);
1607
1608         if (likely(need < cn->size - cn->used - 1))
1609                 goto out_printf;
1610
1611         ret = expand_corename(cn);
1612         if (ret)
1613                 goto expand_fail;
1614
1615 out_printf:
1616         cur = cn->corename + cn->used;
1617         va_start(arg, fmt);
1618         vsnprintf(cur, need + 1, fmt, arg);
1619         va_end(arg);
1620         cn->used += need;
1621         return 0;
1622
1623 expand_fail:
1624         return ret;
1625 }
1626
1627 /* format_corename will inspect the pattern parameter, and output a
1628  * name into corename, which must have space for at least
1629  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1630  */
1631 static int format_corename(struct core_name *cn, long signr)
1632 {
1633         const struct cred *cred = current_cred();
1634         const char *pat_ptr = core_pattern;
1635         int ispipe = (*pat_ptr == '|');
1636         int pid_in_pattern = 0;
1637         int err = 0;
1638
1639         cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1640         cn->corename = kmalloc(cn->size, GFP_KERNEL);
1641         cn->used = 0;
1642
1643         if (!cn->corename)
1644                 return -ENOMEM;
1645
1646         /* Repeat as long as we have more pattern to process and more output
1647            space */
1648         while (*pat_ptr) {
1649                 if (*pat_ptr != '%') {
1650                         if (*pat_ptr == 0)
1651                                 goto out;
1652                         err = cn_printf(cn, "%c", *pat_ptr++);
1653                 } else {
1654                         switch (*++pat_ptr) {
1655                         /* single % at the end, drop that */
1656                         case 0:
1657                                 goto out;
1658                         /* Double percent, output one percent */
1659                         case '%':
1660                                 err = cn_printf(cn, "%c", '%');
1661                                 break;
1662                         /* pid */
1663                         case 'p':
1664                                 pid_in_pattern = 1;
1665                                 err = cn_printf(cn, "%d",
1666                                               task_tgid_vnr(current));
1667                                 break;
1668                         /* uid */
1669                         case 'u':
1670                                 err = cn_printf(cn, "%d", cred->uid);
1671                                 break;
1672                         /* gid */
1673                         case 'g':
1674                                 err = cn_printf(cn, "%d", cred->gid);
1675                                 break;
1676                         /* signal that caused the coredump */
1677                         case 's':
1678                                 err = cn_printf(cn, "%ld", signr);
1679                                 break;
1680                         /* UNIX time of coredump */
1681                         case 't': {
1682                                 struct timeval tv;
1683                                 do_gettimeofday(&tv);
1684                                 err = cn_printf(cn, "%lu", tv.tv_sec);
1685                                 break;
1686                         }
1687                         /* hostname */
1688                         case 'h':
1689                                 down_read(&uts_sem);
1690                                 err = cn_printf(cn, "%s",
1691                                               utsname()->nodename);
1692                                 up_read(&uts_sem);
1693                                 break;
1694                         /* executable */
1695                         case 'e':
1696                                 err = cn_printf(cn, "%s", current->comm);
1697                                 break;
1698                         /* core limit size */
1699                         case 'c':
1700                                 err = cn_printf(cn, "%lu",
1701                                               rlimit(RLIMIT_CORE));
1702                                 break;
1703                         default:
1704                                 break;
1705                         }
1706                         ++pat_ptr;
1707                 }
1708
1709                 if (err)
1710                         return err;
1711         }
1712
1713         /* Backward compatibility with core_uses_pid:
1714          *
1715          * If core_pattern does not include a %p (as is the default)
1716          * and core_uses_pid is set, then .%pid will be appended to
1717          * the filename. Do not do this for piped commands. */
1718         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1719                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1720                 if (err)
1721                         return err;
1722         }
1723 out:
1724         return ispipe;
1725 }
1726
1727 static int zap_process(struct task_struct *start, int exit_code)
1728 {
1729         struct task_struct *t;
1730         int nr = 0;
1731
1732         start->signal->flags = SIGNAL_GROUP_EXIT;
1733         start->signal->group_exit_code = exit_code;
1734         start->signal->group_stop_count = 0;
1735
1736         t = start;
1737         do {
1738                 task_clear_group_stop_pending(t);
1739                 if (t != current && t->mm) {
1740                         sigaddset(&t->pending.signal, SIGKILL);
1741                         signal_wake_up(t, 1);
1742                         nr++;
1743                 }
1744         } while_each_thread(start, t);
1745
1746         return nr;
1747 }
1748
1749 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1750                                 struct core_state *core_state, int exit_code)
1751 {
1752         struct task_struct *g, *p;
1753         unsigned long flags;
1754         int nr = -EAGAIN;
1755
1756         spin_lock_irq(&tsk->sighand->siglock);
1757         if (!signal_group_exit(tsk->signal)) {
1758                 mm->core_state = core_state;
1759                 nr = zap_process(tsk, exit_code);
1760         }
1761         spin_unlock_irq(&tsk->sighand->siglock);
1762         if (unlikely(nr < 0))
1763                 return nr;
1764
1765         if (atomic_read(&mm->mm_users) == nr + 1)
1766                 goto done;
1767         /*
1768          * We should find and kill all tasks which use this mm, and we should
1769          * count them correctly into ->nr_threads. We don't take tasklist
1770          * lock, but this is safe wrt:
1771          *
1772          * fork:
1773          *      None of sub-threads can fork after zap_process(leader). All
1774          *      processes which were created before this point should be
1775          *      visible to zap_threads() because copy_process() adds the new
1776          *      process to the tail of init_task.tasks list, and lock/unlock
1777          *      of ->siglock provides a memory barrier.
1778          *
1779          * do_exit:
1780          *      The caller holds mm->mmap_sem. This means that the task which
1781          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1782          *      its ->mm.
1783          *
1784          * de_thread:
1785          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1786          *      we must see either old or new leader, this does not matter.
1787          *      However, it can change p->sighand, so lock_task_sighand(p)
1788          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1789          *      it can't fail.
1790          *
1791          *      Note also that "g" can be the old leader with ->mm == NULL
1792          *      and already unhashed and thus removed from ->thread_group.
1793          *      This is OK, __unhash_process()->list_del_rcu() does not
1794          *      clear the ->next pointer, we will find the new leader via
1795          *      next_thread().
1796          */
1797         rcu_read_lock();
1798         for_each_process(g) {
1799                 if (g == tsk->group_leader)
1800                         continue;
1801                 if (g->flags & PF_KTHREAD)
1802                         continue;
1803                 p = g;
1804                 do {
1805                         if (p->mm) {
1806                                 if (unlikely(p->mm == mm)) {
1807                                         lock_task_sighand(p, &flags);
1808                                         nr += zap_process(p, exit_code);
1809                                         unlock_task_sighand(p, &flags);
1810                                 }
1811                                 break;
1812                         }
1813                 } while_each_thread(g, p);
1814         }
1815         rcu_read_unlock();
1816 done:
1817         atomic_set(&core_state->nr_threads, nr);
1818         return nr;
1819 }
1820
1821 static int coredump_wait(int exit_code, struct core_state *core_state)
1822 {
1823         struct task_struct *tsk = current;
1824         struct mm_struct *mm = tsk->mm;
1825         struct completion *vfork_done;
1826         int core_waiters = -EBUSY;
1827
1828         init_completion(&core_state->startup);
1829         core_state->dumper.task = tsk;
1830         core_state->dumper.next = NULL;
1831
1832         down_write(&mm->mmap_sem);
1833         if (!mm->core_state)
1834                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1835         up_write(&mm->mmap_sem);
1836
1837         if (unlikely(core_waiters < 0))
1838                 goto fail;
1839
1840         /*
1841          * Make sure nobody is waiting for us to release the VM,
1842          * otherwise we can deadlock when we wait on each other
1843          */
1844         vfork_done = tsk->vfork_done;
1845         if (vfork_done) {
1846                 tsk->vfork_done = NULL;
1847                 complete(vfork_done);
1848         }
1849
1850         if (core_waiters)
1851                 wait_for_completion(&core_state->startup);
1852 fail:
1853         return core_waiters;
1854 }
1855
1856 static void coredump_finish(struct mm_struct *mm)
1857 {
1858         struct core_thread *curr, *next;
1859         struct task_struct *task;
1860
1861         next = mm->core_state->dumper.next;
1862         while ((curr = next) != NULL) {
1863                 next = curr->next;
1864                 task = curr->task;
1865                 /*
1866                  * see exit_mm(), curr->task must not see
1867                  * ->task == NULL before we read ->next.
1868                  */
1869                 smp_mb();
1870                 curr->task = NULL;
1871                 wake_up_process(task);
1872         }
1873
1874         mm->core_state = NULL;
1875 }
1876
1877 /*
1878  * set_dumpable converts traditional three-value dumpable to two flags and
1879  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1880  * these bits are not changed atomically.  So get_dumpable can observe the
1881  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1882  * return either old dumpable or new one by paying attention to the order of
1883  * modifying the bits.
1884  *
1885  * dumpable |   mm->flags (binary)
1886  * old  new | initial interim  final
1887  * ---------+-----------------------
1888  *  0    1  |   00      01      01
1889  *  0    2  |   00      10(*)   11
1890  *  1    0  |   01      00      00
1891  *  1    2  |   01      11      11
1892  *  2    0  |   11      10(*)   00
1893  *  2    1  |   11      11      01
1894  *
1895  * (*) get_dumpable regards interim value of 10 as 11.
1896  */
1897 void set_dumpable(struct mm_struct *mm, int value)
1898 {
1899         switch (value) {
1900         case 0:
1901                 clear_bit(MMF_DUMPABLE, &mm->flags);
1902                 smp_wmb();
1903                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1904                 break;
1905         case 1:
1906                 set_bit(MMF_DUMPABLE, &mm->flags);
1907                 smp_wmb();
1908                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1909                 break;
1910         case 2:
1911                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1912                 smp_wmb();
1913                 set_bit(MMF_DUMPABLE, &mm->flags);
1914                 break;
1915         }
1916 }
1917
1918 static int __get_dumpable(unsigned long mm_flags)
1919 {
1920         int ret;
1921
1922         ret = mm_flags & MMF_DUMPABLE_MASK;
1923         return (ret >= 2) ? 2 : ret;
1924 }
1925
1926 int get_dumpable(struct mm_struct *mm)
1927 {
1928         return __get_dumpable(mm->flags);
1929 }
1930
1931 static void wait_for_dump_helpers(struct file *file)
1932 {
1933         struct pipe_inode_info *pipe;
1934
1935         pipe = file->f_path.dentry->d_inode->i_pipe;
1936
1937         pipe_lock(pipe);
1938         pipe->readers++;
1939         pipe->writers--;
1940
1941         while ((pipe->readers > 1) && (!signal_pending(current))) {
1942                 wake_up_interruptible_sync(&pipe->wait);
1943                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1944                 pipe_wait(pipe);
1945         }
1946
1947         pipe->readers--;
1948         pipe->writers++;
1949         pipe_unlock(pipe);
1950
1951 }
1952
1953
1954 /*
1955  * umh_pipe_setup
1956  * helper function to customize the process used
1957  * to collect the core in userspace.  Specifically
1958  * it sets up a pipe and installs it as fd 0 (stdin)
1959  * for the process.  Returns 0 on success, or
1960  * PTR_ERR on failure.
1961  * Note that it also sets the core limit to 1.  This
1962  * is a special value that we use to trap recursive
1963  * core dumps
1964  */
1965 static int umh_pipe_setup(struct subprocess_info *info)
1966 {
1967         struct file *rp, *wp;
1968         struct fdtable *fdt;
1969         struct coredump_params *cp = (struct coredump_params *)info->data;
1970         struct files_struct *cf = current->files;
1971
1972         wp = create_write_pipe(0);
1973         if (IS_ERR(wp))
1974                 return PTR_ERR(wp);
1975
1976         rp = create_read_pipe(wp, 0);
1977         if (IS_ERR(rp)) {
1978                 free_write_pipe(wp);
1979                 return PTR_ERR(rp);
1980         }
1981
1982         cp->file = wp;
1983
1984         sys_close(0);
1985         fd_install(0, rp);
1986         spin_lock(&cf->file_lock);
1987         fdt = files_fdtable(cf);
1988         FD_SET(0, fdt->open_fds);
1989         FD_CLR(0, fdt->close_on_exec);
1990         spin_unlock(&cf->file_lock);
1991
1992         /* and disallow core files too */
1993         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1994
1995         return 0;
1996 }
1997
1998 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1999 {
2000         struct core_state core_state;
2001         struct core_name cn;
2002         struct mm_struct *mm = current->mm;
2003         struct linux_binfmt * binfmt;
2004         const struct cred *old_cred;
2005         struct cred *cred;
2006         int retval = 0;
2007         int flag = 0;
2008         int ispipe;
2009         static atomic_t core_dump_count = ATOMIC_INIT(0);
2010         struct coredump_params cprm = {
2011                 .signr = signr,
2012                 .regs = regs,
2013                 .limit = rlimit(RLIMIT_CORE),
2014                 /*
2015                  * We must use the same mm->flags while dumping core to avoid
2016                  * inconsistency of bit flags, since this flag is not protected
2017                  * by any locks.
2018                  */
2019                 .mm_flags = mm->flags,
2020         };
2021
2022         audit_core_dumps(signr);
2023
2024         binfmt = mm->binfmt;
2025         if (!binfmt || !binfmt->core_dump)
2026                 goto fail;
2027         if (!__get_dumpable(cprm.mm_flags))
2028                 goto fail;
2029
2030         cred = prepare_creds();
2031         if (!cred)
2032                 goto fail;
2033         /*
2034          *      We cannot trust fsuid as being the "true" uid of the
2035          *      process nor do we know its entire history. We only know it
2036          *      was tainted so we dump it as root in mode 2.
2037          */
2038         if (__get_dumpable(cprm.mm_flags) == 2) {
2039                 /* Setuid core dump mode */
2040                 flag = O_EXCL;          /* Stop rewrite attacks */
2041                 cred->fsuid = 0;        /* Dump root private */
2042         }
2043
2044         retval = coredump_wait(exit_code, &core_state);
2045         if (retval < 0)
2046                 goto fail_creds;
2047
2048         old_cred = override_creds(cred);
2049
2050         /*
2051          * Clear any false indication of pending signals that might
2052          * be seen by the filesystem code called to write the core file.
2053          */
2054         clear_thread_flag(TIF_SIGPENDING);
2055
2056         ispipe = format_corename(&cn, signr);
2057
2058         if (ispipe == -ENOMEM) {
2059                 printk(KERN_WARNING "format_corename failed\n");
2060                 printk(KERN_WARNING "Aborting core\n");
2061                 goto fail_corename;
2062         }
2063
2064         if (ispipe) {
2065                 int dump_count;
2066                 char **helper_argv;
2067
2068                 if (cprm.limit == 1) {
2069                         /*
2070                          * Normally core limits are irrelevant to pipes, since
2071                          * we're not writing to the file system, but we use
2072                          * cprm.limit of 1 here as a speacial value. Any
2073                          * non-1 limit gets set to RLIM_INFINITY below, but
2074                          * a limit of 0 skips the dump.  This is a consistent
2075                          * way to catch recursive crashes.  We can still crash
2076                          * if the core_pattern binary sets RLIM_CORE =  !1
2077                          * but it runs as root, and can do lots of stupid things
2078                          * Note that we use task_tgid_vnr here to grab the pid
2079                          * of the process group leader.  That way we get the
2080                          * right pid if a thread in a multi-threaded
2081                          * core_pattern process dies.
2082                          */
2083                         printk(KERN_WARNING
2084                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2085                                 task_tgid_vnr(current), current->comm);
2086                         printk(KERN_WARNING "Aborting core\n");
2087                         goto fail_unlock;
2088                 }
2089                 cprm.limit = RLIM_INFINITY;
2090
2091                 dump_count = atomic_inc_return(&core_dump_count);
2092                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2093                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2094                                task_tgid_vnr(current), current->comm);
2095                         printk(KERN_WARNING "Skipping core dump\n");
2096                         goto fail_dropcount;
2097                 }
2098
2099                 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2100                 if (!helper_argv) {
2101                         printk(KERN_WARNING "%s failed to allocate memory\n",
2102                                __func__);
2103                         goto fail_dropcount;
2104                 }
2105
2106                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2107                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2108                                         NULL, &cprm);
2109                 argv_free(helper_argv);
2110                 if (retval) {
2111                         printk(KERN_INFO "Core dump to %s pipe failed\n",
2112                                cn.corename);
2113                         goto close_fail;
2114                 }
2115         } else {
2116                 struct inode *inode;
2117
2118                 if (cprm.limit < binfmt->min_coredump)
2119                         goto fail_unlock;
2120
2121                 cprm.file = filp_open(cn.corename,
2122                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2123                                  0600);
2124                 if (IS_ERR(cprm.file))
2125                         goto fail_unlock;
2126
2127                 inode = cprm.file->f_path.dentry->d_inode;
2128                 if (inode->i_nlink > 1)
2129                         goto close_fail;
2130                 if (d_unhashed(cprm.file->f_path.dentry))
2131                         goto close_fail;
2132                 /*
2133                  * AK: actually i see no reason to not allow this for named
2134                  * pipes etc, but keep the previous behaviour for now.
2135                  */
2136                 if (!S_ISREG(inode->i_mode))
2137                         goto close_fail;
2138                 /*
2139                  * Dont allow local users get cute and trick others to coredump
2140                  * into their pre-created files.
2141                  */
2142                 if (inode->i_uid != current_fsuid())
2143                         goto close_fail;
2144                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2145                         goto close_fail;
2146                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2147                         goto close_fail;
2148         }
2149
2150         retval = binfmt->core_dump(&cprm);
2151         if (retval)
2152                 current->signal->group_exit_code |= 0x80;
2153
2154         if (ispipe && core_pipe_limit)
2155                 wait_for_dump_helpers(cprm.file);
2156 close_fail:
2157         if (cprm.file)
2158                 filp_close(cprm.file, NULL);
2159 fail_dropcount:
2160         if (ispipe)
2161                 atomic_dec(&core_dump_count);
2162 fail_unlock:
2163         kfree(cn.corename);
2164 fail_corename:
2165         coredump_finish(mm);
2166         revert_creds(old_cred);
2167 fail_creds:
2168         put_cred(cred);
2169 fail:
2170         return;
2171 }
2172
2173 /*
2174  * Core dumping helper functions.  These are the only things you should
2175  * do on a core-file: use only these functions to write out all the
2176  * necessary info.
2177  */
2178 int dump_write(struct file *file, const void *addr, int nr)
2179 {
2180         return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2181 }
2182 EXPORT_SYMBOL(dump_write);
2183
2184 int dump_seek(struct file *file, loff_t off)
2185 {
2186         int ret = 1;
2187
2188         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2189                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2190                         return 0;
2191         } else {
2192                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2193
2194                 if (!buf)
2195                         return 0;
2196                 while (off > 0) {
2197                         unsigned long n = off;
2198
2199                         if (n > PAGE_SIZE)
2200                                 n = PAGE_SIZE;
2201                         if (!dump_write(file, buf, n)) {
2202                                 ret = 0;
2203                                 break;
2204                         }
2205                         off -= n;
2206                 }
2207                 free_page((unsigned long)buf);
2208         }
2209         return ret;
2210 }
2211 EXPORT_SYMBOL(dump_seek);