exec: introduce struct user_arg_ptr
[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
59 #include <asm/uaccess.h>
60 #include <asm/mmu_context.h>
61 #include <asm/tlb.h>
62 #include "internal.h"
63
64 int core_uses_pid;
65 char core_pattern[CORENAME_MAX_SIZE] = "core";
66 unsigned int core_pipe_limit;
67 int suid_dumpable = 0;
68
69 struct core_name {
70         char *corename;
71         int used, size;
72 };
73 static atomic_t call_count = ATOMIC_INIT(1);
74
75 /* The maximal length of core_pattern is also specified in sysctl.c */
76
77 static LIST_HEAD(formats);
78 static DEFINE_RWLOCK(binfmt_lock);
79
80 int __register_binfmt(struct linux_binfmt * fmt, int insert)
81 {
82         if (!fmt)
83                 return -EINVAL;
84         write_lock(&binfmt_lock);
85         insert ? list_add(&fmt->lh, &formats) :
86                  list_add_tail(&fmt->lh, &formats);
87         write_unlock(&binfmt_lock);
88         return 0;       
89 }
90
91 EXPORT_SYMBOL(__register_binfmt);
92
93 void unregister_binfmt(struct linux_binfmt * fmt)
94 {
95         write_lock(&binfmt_lock);
96         list_del(&fmt->lh);
97         write_unlock(&binfmt_lock);
98 }
99
100 EXPORT_SYMBOL(unregister_binfmt);
101
102 static inline void put_binfmt(struct linux_binfmt * fmt)
103 {
104         module_put(fmt->module);
105 }
106
107 /*
108  * Note that a shared library must be both readable and executable due to
109  * security reasons.
110  *
111  * Also note that we take the address to load from from the file itself.
112  */
113 SYSCALL_DEFINE1(uselib, const char __user *, library)
114 {
115         struct file *file;
116         char *tmp = getname(library);
117         int error = PTR_ERR(tmp);
118         static const struct open_flags uselib_flags = {
119                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
120                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
121                 .intent = LOOKUP_OPEN
122         };
123
124         if (IS_ERR(tmp))
125                 goto out;
126
127         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
128         putname(tmp);
129         error = PTR_ERR(file);
130         if (IS_ERR(file))
131                 goto out;
132
133         error = -EINVAL;
134         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
135                 goto exit;
136
137         error = -EACCES;
138         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
139                 goto exit;
140
141         fsnotify_open(file);
142
143         error = -ENOEXEC;
144         if(file->f_op) {
145                 struct linux_binfmt * fmt;
146
147                 read_lock(&binfmt_lock);
148                 list_for_each_entry(fmt, &formats, lh) {
149                         if (!fmt->load_shlib)
150                                 continue;
151                         if (!try_module_get(fmt->module))
152                                 continue;
153                         read_unlock(&binfmt_lock);
154                         error = fmt->load_shlib(file);
155                         read_lock(&binfmt_lock);
156                         put_binfmt(fmt);
157                         if (error != -ENOEXEC)
158                                 break;
159                 }
160                 read_unlock(&binfmt_lock);
161         }
162 exit:
163         fput(file);
164 out:
165         return error;
166 }
167
168 #ifdef CONFIG_MMU
169
170 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
171 {
172         struct mm_struct *mm = current->mm;
173         long diff = (long)(pages - bprm->vma_pages);
174
175         if (!mm || !diff)
176                 return;
177
178         bprm->vma_pages = pages;
179
180 #ifdef SPLIT_RSS_COUNTING
181         add_mm_counter(mm, MM_ANONPAGES, diff);
182 #else
183         spin_lock(&mm->page_table_lock);
184         add_mm_counter(mm, MM_ANONPAGES, diff);
185         spin_unlock(&mm->page_table_lock);
186 #endif
187 }
188
189 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
190                 int write)
191 {
192         struct page *page;
193         int ret;
194
195 #ifdef CONFIG_STACK_GROWSUP
196         if (write) {
197                 ret = expand_stack_downwards(bprm->vma, pos);
198                 if (ret < 0)
199                         return NULL;
200         }
201 #endif
202         ret = get_user_pages(current, bprm->mm, pos,
203                         1, write, 1, &page, NULL);
204         if (ret <= 0)
205                 return NULL;
206
207         if (write) {
208                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
209                 struct rlimit *rlim;
210
211                 acct_arg_size(bprm, size / PAGE_SIZE);
212
213                 /*
214                  * We've historically supported up to 32 pages (ARG_MAX)
215                  * of argument strings even with small stacks
216                  */
217                 if (size <= ARG_MAX)
218                         return page;
219
220                 /*
221                  * Limit to 1/4-th the stack size for the argv+env strings.
222                  * This ensures that:
223                  *  - the remaining binfmt code will not run out of stack space,
224                  *  - the program will have a reasonable amount of stack left
225                  *    to work from.
226                  */
227                 rlim = current->signal->rlim;
228                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
229                         put_page(page);
230                         return NULL;
231                 }
232         }
233
234         return page;
235 }
236
237 static void put_arg_page(struct page *page)
238 {
239         put_page(page);
240 }
241
242 static void free_arg_page(struct linux_binprm *bprm, int i)
243 {
244 }
245
246 static void free_arg_pages(struct linux_binprm *bprm)
247 {
248 }
249
250 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
251                 struct page *page)
252 {
253         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
254 }
255
256 static int __bprm_mm_init(struct linux_binprm *bprm)
257 {
258         int err;
259         struct vm_area_struct *vma = NULL;
260         struct mm_struct *mm = bprm->mm;
261
262         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
263         if (!vma)
264                 return -ENOMEM;
265
266         down_write(&mm->mmap_sem);
267         vma->vm_mm = mm;
268
269         /*
270          * Place the stack at the largest stack address the architecture
271          * supports. Later, we'll move this to an appropriate place. We don't
272          * use STACK_TOP because that can depend on attributes which aren't
273          * configured yet.
274          */
275         BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
276         vma->vm_end = STACK_TOP_MAX;
277         vma->vm_start = vma->vm_end - PAGE_SIZE;
278         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
279         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
280         INIT_LIST_HEAD(&vma->anon_vma_chain);
281
282         err = security_file_mmap(NULL, 0, 0, 0, vma->vm_start, 1);
283         if (err)
284                 goto err;
285
286         err = insert_vm_struct(mm, vma);
287         if (err)
288                 goto err;
289
290         mm->stack_vm = mm->total_vm = 1;
291         up_write(&mm->mmap_sem);
292         bprm->p = vma->vm_end - sizeof(void *);
293         return 0;
294 err:
295         up_write(&mm->mmap_sem);
296         bprm->vma = NULL;
297         kmem_cache_free(vm_area_cachep, vma);
298         return err;
299 }
300
301 static bool valid_arg_len(struct linux_binprm *bprm, long len)
302 {
303         return len <= MAX_ARG_STRLEN;
304 }
305
306 #else
307
308 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
309 {
310 }
311
312 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
313                 int write)
314 {
315         struct page *page;
316
317         page = bprm->page[pos / PAGE_SIZE];
318         if (!page && write) {
319                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
320                 if (!page)
321                         return NULL;
322                 bprm->page[pos / PAGE_SIZE] = page;
323         }
324
325         return page;
326 }
327
328 static void put_arg_page(struct page *page)
329 {
330 }
331
332 static void free_arg_page(struct linux_binprm *bprm, int i)
333 {
334         if (bprm->page[i]) {
335                 __free_page(bprm->page[i]);
336                 bprm->page[i] = NULL;
337         }
338 }
339
340 static void free_arg_pages(struct linux_binprm *bprm)
341 {
342         int i;
343
344         for (i = 0; i < MAX_ARG_PAGES; i++)
345                 free_arg_page(bprm, i);
346 }
347
348 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
349                 struct page *page)
350 {
351 }
352
353 static int __bprm_mm_init(struct linux_binprm *bprm)
354 {
355         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
356         return 0;
357 }
358
359 static bool valid_arg_len(struct linux_binprm *bprm, long len)
360 {
361         return len <= bprm->p;
362 }
363
364 #endif /* CONFIG_MMU */
365
366 /*
367  * Create a new mm_struct and populate it with a temporary stack
368  * vm_area_struct.  We don't have enough context at this point to set the stack
369  * flags, permissions, and offset, so we use temporary values.  We'll update
370  * them later in setup_arg_pages().
371  */
372 int bprm_mm_init(struct linux_binprm *bprm)
373 {
374         int err;
375         struct mm_struct *mm = NULL;
376
377         bprm->mm = mm = mm_alloc();
378         err = -ENOMEM;
379         if (!mm)
380                 goto err;
381
382         err = init_new_context(current, mm);
383         if (err)
384                 goto err;
385
386         err = __bprm_mm_init(bprm);
387         if (err)
388                 goto err;
389
390         return 0;
391
392 err:
393         if (mm) {
394                 bprm->mm = NULL;
395                 mmdrop(mm);
396         }
397
398         return err;
399 }
400
401 struct user_arg_ptr {
402         const char __user *const __user *native;
403 };
404
405 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
406 {
407         const char __user *ptr;
408
409         if (get_user(ptr, argv.native + nr))
410                 return ERR_PTR(-EFAULT);
411
412         return ptr;
413 }
414
415 /*
416  * count() counts the number of strings in array ARGV.
417  */
418 static int count(struct user_arg_ptr argv, int max)
419 {
420         int i = 0;
421
422         if (argv.native != NULL) {
423                 for (;;) {
424                         const char __user *p = get_user_arg_ptr(argv, i);
425
426                         if (!p)
427                                 break;
428
429                         if (IS_ERR(p))
430                                 return -EFAULT;
431
432                         if (i++ >= max)
433                                 return -E2BIG;
434
435                         if (fatal_signal_pending(current))
436                                 return -ERESTARTNOHAND;
437                         cond_resched();
438                 }
439         }
440         return i;
441 }
442
443 /*
444  * 'copy_strings()' copies argument/environment strings from the old
445  * processes's memory to the new process's stack.  The call to get_user_pages()
446  * ensures the destination page is created and not swapped out.
447  */
448 static int copy_strings(int argc, struct user_arg_ptr argv,
449                         struct linux_binprm *bprm)
450 {
451         struct page *kmapped_page = NULL;
452         char *kaddr = NULL;
453         unsigned long kpos = 0;
454         int ret;
455
456         while (argc-- > 0) {
457                 const char __user *str;
458                 int len;
459                 unsigned long pos;
460
461                 ret = -EFAULT;
462                 str = get_user_arg_ptr(argv, argc);
463                 if (IS_ERR(str))
464                         goto out;
465
466                 len = strnlen_user(str, MAX_ARG_STRLEN);
467                 if (!len)
468                         goto out;
469
470                 ret = -E2BIG;
471                 if (!valid_arg_len(bprm, len))
472                         goto out;
473
474                 /* We're going to work our way backwords. */
475                 pos = bprm->p;
476                 str += len;
477                 bprm->p -= len;
478
479                 while (len > 0) {
480                         int offset, bytes_to_copy;
481
482                         if (fatal_signal_pending(current)) {
483                                 ret = -ERESTARTNOHAND;
484                                 goto out;
485                         }
486                         cond_resched();
487
488                         offset = pos % PAGE_SIZE;
489                         if (offset == 0)
490                                 offset = PAGE_SIZE;
491
492                         bytes_to_copy = offset;
493                         if (bytes_to_copy > len)
494                                 bytes_to_copy = len;
495
496                         offset -= bytes_to_copy;
497                         pos -= bytes_to_copy;
498                         str -= bytes_to_copy;
499                         len -= bytes_to_copy;
500
501                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
502                                 struct page *page;
503
504                                 page = get_arg_page(bprm, pos, 1);
505                                 if (!page) {
506                                         ret = -E2BIG;
507                                         goto out;
508                                 }
509
510                                 if (kmapped_page) {
511                                         flush_kernel_dcache_page(kmapped_page);
512                                         kunmap(kmapped_page);
513                                         put_arg_page(kmapped_page);
514                                 }
515                                 kmapped_page = page;
516                                 kaddr = kmap(kmapped_page);
517                                 kpos = pos & PAGE_MASK;
518                                 flush_arg_page(bprm, kpos, kmapped_page);
519                         }
520                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
521                                 ret = -EFAULT;
522                                 goto out;
523                         }
524                 }
525         }
526         ret = 0;
527 out:
528         if (kmapped_page) {
529                 flush_kernel_dcache_page(kmapped_page);
530                 kunmap(kmapped_page);
531                 put_arg_page(kmapped_page);
532         }
533         return ret;
534 }
535
536 /*
537  * Like copy_strings, but get argv and its values from kernel memory.
538  */
539 int copy_strings_kernel(int argc, const char *const *__argv,
540                         struct linux_binprm *bprm)
541 {
542         int r;
543         mm_segment_t oldfs = get_fs();
544         struct user_arg_ptr argv = {
545                 .native = (const char __user *const  __user *)__argv,
546         };
547
548         set_fs(KERNEL_DS);
549         r = copy_strings(argc, argv, bprm);
550         set_fs(oldfs);
551
552         return r;
553 }
554 EXPORT_SYMBOL(copy_strings_kernel);
555
556 #ifdef CONFIG_MMU
557
558 /*
559  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
560  * the binfmt code determines where the new stack should reside, we shift it to
561  * its final location.  The process proceeds as follows:
562  *
563  * 1) Use shift to calculate the new vma endpoints.
564  * 2) Extend vma to cover both the old and new ranges.  This ensures the
565  *    arguments passed to subsequent functions are consistent.
566  * 3) Move vma's page tables to the new range.
567  * 4) Free up any cleared pgd range.
568  * 5) Shrink the vma to cover only the new range.
569  */
570 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
571 {
572         struct mm_struct *mm = vma->vm_mm;
573         unsigned long old_start = vma->vm_start;
574         unsigned long old_end = vma->vm_end;
575         unsigned long length = old_end - old_start;
576         unsigned long new_start = old_start - shift;
577         unsigned long new_end = old_end - shift;
578         struct mmu_gather *tlb;
579
580         BUG_ON(new_start > new_end);
581
582         /*
583          * ensure there are no vmas between where we want to go
584          * and where we are
585          */
586         if (vma != find_vma(mm, new_start))
587                 return -EFAULT;
588
589         /*
590          * cover the whole range: [new_start, old_end)
591          */
592         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
593                 return -ENOMEM;
594
595         /*
596          * move the page tables downwards, on failure we rely on
597          * process cleanup to remove whatever mess we made.
598          */
599         if (length != move_page_tables(vma, old_start,
600                                        vma, new_start, length))
601                 return -ENOMEM;
602
603         lru_add_drain();
604         tlb = tlb_gather_mmu(mm, 0);
605         if (new_end > old_start) {
606                 /*
607                  * when the old and new regions overlap clear from new_end.
608                  */
609                 free_pgd_range(tlb, new_end, old_end, new_end,
610                         vma->vm_next ? vma->vm_next->vm_start : 0);
611         } else {
612                 /*
613                  * otherwise, clean from old_start; this is done to not touch
614                  * the address space in [new_end, old_start) some architectures
615                  * have constraints on va-space that make this illegal (IA64) -
616                  * for the others its just a little faster.
617                  */
618                 free_pgd_range(tlb, old_start, old_end, new_end,
619                         vma->vm_next ? vma->vm_next->vm_start : 0);
620         }
621         tlb_finish_mmu(tlb, new_end, old_end);
622
623         /*
624          * Shrink the vma to just the new range.  Always succeeds.
625          */
626         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
627
628         return 0;
629 }
630
631 /*
632  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
633  * the stack is optionally relocated, and some extra space is added.
634  */
635 int setup_arg_pages(struct linux_binprm *bprm,
636                     unsigned long stack_top,
637                     int executable_stack)
638 {
639         unsigned long ret;
640         unsigned long stack_shift;
641         struct mm_struct *mm = current->mm;
642         struct vm_area_struct *vma = bprm->vma;
643         struct vm_area_struct *prev = NULL;
644         unsigned long vm_flags;
645         unsigned long stack_base;
646         unsigned long stack_size;
647         unsigned long stack_expand;
648         unsigned long rlim_stack;
649
650 #ifdef CONFIG_STACK_GROWSUP
651         /* Limit stack size to 1GB */
652         stack_base = rlimit_max(RLIMIT_STACK);
653         if (stack_base > (1 << 30))
654                 stack_base = 1 << 30;
655
656         /* Make sure we didn't let the argument array grow too large. */
657         if (vma->vm_end - vma->vm_start > stack_base)
658                 return -ENOMEM;
659
660         stack_base = PAGE_ALIGN(stack_top - stack_base);
661
662         stack_shift = vma->vm_start - stack_base;
663         mm->arg_start = bprm->p - stack_shift;
664         bprm->p = vma->vm_end - stack_shift;
665 #else
666         stack_top = arch_align_stack(stack_top);
667         stack_top = PAGE_ALIGN(stack_top);
668
669         if (unlikely(stack_top < mmap_min_addr) ||
670             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
671                 return -ENOMEM;
672
673         stack_shift = vma->vm_end - stack_top;
674
675         bprm->p -= stack_shift;
676         mm->arg_start = bprm->p;
677 #endif
678
679         if (bprm->loader)
680                 bprm->loader -= stack_shift;
681         bprm->exec -= stack_shift;
682
683         down_write(&mm->mmap_sem);
684         vm_flags = VM_STACK_FLAGS;
685
686         /*
687          * Adjust stack execute permissions; explicitly enable for
688          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
689          * (arch default) otherwise.
690          */
691         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
692                 vm_flags |= VM_EXEC;
693         else if (executable_stack == EXSTACK_DISABLE_X)
694                 vm_flags &= ~VM_EXEC;
695         vm_flags |= mm->def_flags;
696         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
697
698         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
699                         vm_flags);
700         if (ret)
701                 goto out_unlock;
702         BUG_ON(prev != vma);
703
704         /* Move stack pages down in memory. */
705         if (stack_shift) {
706                 ret = shift_arg_pages(vma, stack_shift);
707                 if (ret)
708                         goto out_unlock;
709         }
710
711         /* mprotect_fixup is overkill to remove the temporary stack flags */
712         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
713
714         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
715         stack_size = vma->vm_end - vma->vm_start;
716         /*
717          * Align this down to a page boundary as expand_stack
718          * will align it up.
719          */
720         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
721 #ifdef CONFIG_STACK_GROWSUP
722         if (stack_size + stack_expand > rlim_stack)
723                 stack_base = vma->vm_start + rlim_stack;
724         else
725                 stack_base = vma->vm_end + stack_expand;
726 #else
727         if (stack_size + stack_expand > rlim_stack)
728                 stack_base = vma->vm_end - rlim_stack;
729         else
730                 stack_base = vma->vm_start - stack_expand;
731 #endif
732         current->mm->start_stack = bprm->p;
733         ret = expand_stack(vma, stack_base);
734         if (ret)
735                 ret = -EFAULT;
736
737 out_unlock:
738         up_write(&mm->mmap_sem);
739         return ret;
740 }
741 EXPORT_SYMBOL(setup_arg_pages);
742
743 #endif /* CONFIG_MMU */
744
745 struct file *open_exec(const char *name)
746 {
747         struct file *file;
748         int err;
749         static const struct open_flags open_exec_flags = {
750                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
751                 .acc_mode = MAY_EXEC | MAY_OPEN,
752                 .intent = LOOKUP_OPEN
753         };
754
755         file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
756         if (IS_ERR(file))
757                 goto out;
758
759         err = -EACCES;
760         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
761                 goto exit;
762
763         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
764                 goto exit;
765
766         fsnotify_open(file);
767
768         err = deny_write_access(file);
769         if (err)
770                 goto exit;
771
772 out:
773         return file;
774
775 exit:
776         fput(file);
777         return ERR_PTR(err);
778 }
779 EXPORT_SYMBOL(open_exec);
780
781 int kernel_read(struct file *file, loff_t offset,
782                 char *addr, unsigned long count)
783 {
784         mm_segment_t old_fs;
785         loff_t pos = offset;
786         int result;
787
788         old_fs = get_fs();
789         set_fs(get_ds());
790         /* The cast to a user pointer is valid due to the set_fs() */
791         result = vfs_read(file, (void __user *)addr, count, &pos);
792         set_fs(old_fs);
793         return result;
794 }
795
796 EXPORT_SYMBOL(kernel_read);
797
798 static int exec_mmap(struct mm_struct *mm)
799 {
800         struct task_struct *tsk;
801         struct mm_struct * old_mm, *active_mm;
802
803         /* Notify parent that we're no longer interested in the old VM */
804         tsk = current;
805         old_mm = current->mm;
806         sync_mm_rss(tsk, old_mm);
807         mm_release(tsk, old_mm);
808
809         if (old_mm) {
810                 /*
811                  * Make sure that if there is a core dump in progress
812                  * for the old mm, we get out and die instead of going
813                  * through with the exec.  We must hold mmap_sem around
814                  * checking core_state and changing tsk->mm.
815                  */
816                 down_read(&old_mm->mmap_sem);
817                 if (unlikely(old_mm->core_state)) {
818                         up_read(&old_mm->mmap_sem);
819                         return -EINTR;
820                 }
821         }
822         task_lock(tsk);
823         active_mm = tsk->active_mm;
824         tsk->mm = mm;
825         tsk->active_mm = mm;
826         activate_mm(active_mm, mm);
827         if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
828                 atomic_dec(&old_mm->oom_disable_count);
829                 atomic_inc(&tsk->mm->oom_disable_count);
830         }
831         task_unlock(tsk);
832         arch_pick_mmap_layout(mm);
833         if (old_mm) {
834                 up_read(&old_mm->mmap_sem);
835                 BUG_ON(active_mm != old_mm);
836                 mm_update_next_owner(old_mm);
837                 mmput(old_mm);
838                 return 0;
839         }
840         mmdrop(active_mm);
841         return 0;
842 }
843
844 /*
845  * This function makes sure the current process has its own signal table,
846  * so that flush_signal_handlers can later reset the handlers without
847  * disturbing other processes.  (Other processes might share the signal
848  * table via the CLONE_SIGHAND option to clone().)
849  */
850 static int de_thread(struct task_struct *tsk)
851 {
852         struct signal_struct *sig = tsk->signal;
853         struct sighand_struct *oldsighand = tsk->sighand;
854         spinlock_t *lock = &oldsighand->siglock;
855
856         if (thread_group_empty(tsk))
857                 goto no_thread_group;
858
859         /*
860          * Kill all other threads in the thread group.
861          */
862         spin_lock_irq(lock);
863         if (signal_group_exit(sig)) {
864                 /*
865                  * Another group action in progress, just
866                  * return so that the signal is processed.
867                  */
868                 spin_unlock_irq(lock);
869                 return -EAGAIN;
870         }
871
872         sig->group_exit_task = tsk;
873         sig->notify_count = zap_other_threads(tsk);
874         if (!thread_group_leader(tsk))
875                 sig->notify_count--;
876
877         while (sig->notify_count) {
878                 __set_current_state(TASK_UNINTERRUPTIBLE);
879                 spin_unlock_irq(lock);
880                 schedule();
881                 spin_lock_irq(lock);
882         }
883         spin_unlock_irq(lock);
884
885         /*
886          * At this point all other threads have exited, all we have to
887          * do is to wait for the thread group leader to become inactive,
888          * and to assume its PID:
889          */
890         if (!thread_group_leader(tsk)) {
891                 struct task_struct *leader = tsk->group_leader;
892
893                 sig->notify_count = -1; /* for exit_notify() */
894                 for (;;) {
895                         write_lock_irq(&tasklist_lock);
896                         if (likely(leader->exit_state))
897                                 break;
898                         __set_current_state(TASK_UNINTERRUPTIBLE);
899                         write_unlock_irq(&tasklist_lock);
900                         schedule();
901                 }
902
903                 /*
904                  * The only record we have of the real-time age of a
905                  * process, regardless of execs it's done, is start_time.
906                  * All the past CPU time is accumulated in signal_struct
907                  * from sister threads now dead.  But in this non-leader
908                  * exec, nothing survives from the original leader thread,
909                  * whose birth marks the true age of this process now.
910                  * When we take on its identity by switching to its PID, we
911                  * also take its birthdate (always earlier than our own).
912                  */
913                 tsk->start_time = leader->start_time;
914
915                 BUG_ON(!same_thread_group(leader, tsk));
916                 BUG_ON(has_group_leader_pid(tsk));
917                 /*
918                  * An exec() starts a new thread group with the
919                  * TGID of the previous thread group. Rehash the
920                  * two threads with a switched PID, and release
921                  * the former thread group leader:
922                  */
923
924                 /* Become a process group leader with the old leader's pid.
925                  * The old leader becomes a thread of the this thread group.
926                  * Note: The old leader also uses this pid until release_task
927                  *       is called.  Odd but simple and correct.
928                  */
929                 detach_pid(tsk, PIDTYPE_PID);
930                 tsk->pid = leader->pid;
931                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
932                 transfer_pid(leader, tsk, PIDTYPE_PGID);
933                 transfer_pid(leader, tsk, PIDTYPE_SID);
934
935                 list_replace_rcu(&leader->tasks, &tsk->tasks);
936                 list_replace_init(&leader->sibling, &tsk->sibling);
937
938                 tsk->group_leader = tsk;
939                 leader->group_leader = tsk;
940
941                 tsk->exit_signal = SIGCHLD;
942
943                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
944                 leader->exit_state = EXIT_DEAD;
945                 write_unlock_irq(&tasklist_lock);
946
947                 release_task(leader);
948         }
949
950         sig->group_exit_task = NULL;
951         sig->notify_count = 0;
952
953 no_thread_group:
954         if (current->mm)
955                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
956
957         exit_itimers(sig);
958         flush_itimer_signals();
959
960         if (atomic_read(&oldsighand->count) != 1) {
961                 struct sighand_struct *newsighand;
962                 /*
963                  * This ->sighand is shared with the CLONE_SIGHAND
964                  * but not CLONE_THREAD task, switch to the new one.
965                  */
966                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
967                 if (!newsighand)
968                         return -ENOMEM;
969
970                 atomic_set(&newsighand->count, 1);
971                 memcpy(newsighand->action, oldsighand->action,
972                        sizeof(newsighand->action));
973
974                 write_lock_irq(&tasklist_lock);
975                 spin_lock(&oldsighand->siglock);
976                 rcu_assign_pointer(tsk->sighand, newsighand);
977                 spin_unlock(&oldsighand->siglock);
978                 write_unlock_irq(&tasklist_lock);
979
980                 __cleanup_sighand(oldsighand);
981         }
982
983         BUG_ON(!thread_group_leader(tsk));
984         return 0;
985 }
986
987 /*
988  * These functions flushes out all traces of the currently running executable
989  * so that a new one can be started
990  */
991 static void flush_old_files(struct files_struct * files)
992 {
993         long j = -1;
994         struct fdtable *fdt;
995
996         spin_lock(&files->file_lock);
997         for (;;) {
998                 unsigned long set, i;
999
1000                 j++;
1001                 i = j * __NFDBITS;
1002                 fdt = files_fdtable(files);
1003                 if (i >= fdt->max_fds)
1004                         break;
1005                 set = fdt->close_on_exec->fds_bits[j];
1006                 if (!set)
1007                         continue;
1008                 fdt->close_on_exec->fds_bits[j] = 0;
1009                 spin_unlock(&files->file_lock);
1010                 for ( ; set ; i++,set >>= 1) {
1011                         if (set & 1) {
1012                                 sys_close(i);
1013                         }
1014                 }
1015                 spin_lock(&files->file_lock);
1016
1017         }
1018         spin_unlock(&files->file_lock);
1019 }
1020
1021 char *get_task_comm(char *buf, struct task_struct *tsk)
1022 {
1023         /* buf must be at least sizeof(tsk->comm) in size */
1024         task_lock(tsk);
1025         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1026         task_unlock(tsk);
1027         return buf;
1028 }
1029
1030 void set_task_comm(struct task_struct *tsk, char *buf)
1031 {
1032         task_lock(tsk);
1033
1034         /*
1035          * Threads may access current->comm without holding
1036          * the task lock, so write the string carefully.
1037          * Readers without a lock may see incomplete new
1038          * names but are safe from non-terminating string reads.
1039          */
1040         memset(tsk->comm, 0, TASK_COMM_LEN);
1041         wmb();
1042         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1043         task_unlock(tsk);
1044         perf_event_comm(tsk);
1045 }
1046
1047 int flush_old_exec(struct linux_binprm * bprm)
1048 {
1049         int retval;
1050
1051         /*
1052          * Make sure we have a private signal table and that
1053          * we are unassociated from the previous thread group.
1054          */
1055         retval = de_thread(current);
1056         if (retval)
1057                 goto out;
1058
1059         set_mm_exe_file(bprm->mm, bprm->file);
1060
1061         /*
1062          * Release all of the old mmap stuff
1063          */
1064         acct_arg_size(bprm, 0);
1065         retval = exec_mmap(bprm->mm);
1066         if (retval)
1067                 goto out;
1068
1069         bprm->mm = NULL;                /* We're using it now */
1070
1071         current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1072         flush_thread();
1073         current->personality &= ~bprm->per_clear;
1074
1075         return 0;
1076
1077 out:
1078         return retval;
1079 }
1080 EXPORT_SYMBOL(flush_old_exec);
1081
1082 void setup_new_exec(struct linux_binprm * bprm)
1083 {
1084         int i, ch;
1085         const char *name;
1086         char tcomm[sizeof(current->comm)];
1087
1088         arch_pick_mmap_layout(current->mm);
1089
1090         /* This is the point of no return */
1091         current->sas_ss_sp = current->sas_ss_size = 0;
1092
1093         if (current_euid() == current_uid() && current_egid() == current_gid())
1094                 set_dumpable(current->mm, 1);
1095         else
1096                 set_dumpable(current->mm, suid_dumpable);
1097
1098         name = bprm->filename;
1099
1100         /* Copies the binary name from after last slash */
1101         for (i=0; (ch = *(name++)) != '\0';) {
1102                 if (ch == '/')
1103                         i = 0; /* overwrite what we wrote */
1104                 else
1105                         if (i < (sizeof(tcomm) - 1))
1106                                 tcomm[i++] = ch;
1107         }
1108         tcomm[i] = '\0';
1109         set_task_comm(current, tcomm);
1110
1111         /* Set the new mm task size. We have to do that late because it may
1112          * depend on TIF_32BIT which is only updated in flush_thread() on
1113          * some architectures like powerpc
1114          */
1115         current->mm->task_size = TASK_SIZE;
1116
1117         /* install the new credentials */
1118         if (bprm->cred->uid != current_euid() ||
1119             bprm->cred->gid != current_egid()) {
1120                 current->pdeath_signal = 0;
1121         } else if (file_permission(bprm->file, MAY_READ) ||
1122                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1123                 set_dumpable(current->mm, suid_dumpable);
1124         }
1125
1126         /*
1127          * Flush performance counters when crossing a
1128          * security domain:
1129          */
1130         if (!get_dumpable(current->mm))
1131                 perf_event_exit_task(current);
1132
1133         /* An exec changes our domain. We are no longer part of the thread
1134            group */
1135
1136         current->self_exec_id++;
1137                         
1138         flush_signal_handlers(current, 0);
1139         flush_old_files(current->files);
1140 }
1141 EXPORT_SYMBOL(setup_new_exec);
1142
1143 /*
1144  * Prepare credentials and lock ->cred_guard_mutex.
1145  * install_exec_creds() commits the new creds and drops the lock.
1146  * Or, if exec fails before, free_bprm() should release ->cred and
1147  * and unlock.
1148  */
1149 int prepare_bprm_creds(struct linux_binprm *bprm)
1150 {
1151         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1152                 return -ERESTARTNOINTR;
1153
1154         bprm->cred = prepare_exec_creds();
1155         if (likely(bprm->cred))
1156                 return 0;
1157
1158         mutex_unlock(&current->signal->cred_guard_mutex);
1159         return -ENOMEM;
1160 }
1161
1162 void free_bprm(struct linux_binprm *bprm)
1163 {
1164         free_arg_pages(bprm);
1165         if (bprm->cred) {
1166                 mutex_unlock(&current->signal->cred_guard_mutex);
1167                 abort_creds(bprm->cred);
1168         }
1169         kfree(bprm);
1170 }
1171
1172 /*
1173  * install the new credentials for this executable
1174  */
1175 void install_exec_creds(struct linux_binprm *bprm)
1176 {
1177         security_bprm_committing_creds(bprm);
1178
1179         commit_creds(bprm->cred);
1180         bprm->cred = NULL;
1181         /*
1182          * cred_guard_mutex must be held at least to this point to prevent
1183          * ptrace_attach() from altering our determination of the task's
1184          * credentials; any time after this it may be unlocked.
1185          */
1186         security_bprm_committed_creds(bprm);
1187         mutex_unlock(&current->signal->cred_guard_mutex);
1188 }
1189 EXPORT_SYMBOL(install_exec_creds);
1190
1191 /*
1192  * determine how safe it is to execute the proposed program
1193  * - the caller must hold ->cred_guard_mutex to protect against
1194  *   PTRACE_ATTACH
1195  */
1196 int check_unsafe_exec(struct linux_binprm *bprm)
1197 {
1198         struct task_struct *p = current, *t;
1199         unsigned n_fs;
1200         int res = 0;
1201
1202         bprm->unsafe = tracehook_unsafe_exec(p);
1203
1204         n_fs = 1;
1205         spin_lock(&p->fs->lock);
1206         rcu_read_lock();
1207         for (t = next_thread(p); t != p; t = next_thread(t)) {
1208                 if (t->fs == p->fs)
1209                         n_fs++;
1210         }
1211         rcu_read_unlock();
1212
1213         if (p->fs->users > n_fs) {
1214                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1215         } else {
1216                 res = -EAGAIN;
1217                 if (!p->fs->in_exec) {
1218                         p->fs->in_exec = 1;
1219                         res = 1;
1220                 }
1221         }
1222         spin_unlock(&p->fs->lock);
1223
1224         return res;
1225 }
1226
1227 /* 
1228  * Fill the binprm structure from the inode. 
1229  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1230  *
1231  * This may be called multiple times for binary chains (scripts for example).
1232  */
1233 int prepare_binprm(struct linux_binprm *bprm)
1234 {
1235         umode_t mode;
1236         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1237         int retval;
1238
1239         mode = inode->i_mode;
1240         if (bprm->file->f_op == NULL)
1241                 return -EACCES;
1242
1243         /* clear any previous set[ug]id data from a previous binary */
1244         bprm->cred->euid = current_euid();
1245         bprm->cred->egid = current_egid();
1246
1247         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1248                 /* Set-uid? */
1249                 if (mode & S_ISUID) {
1250                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1251                         bprm->cred->euid = inode->i_uid;
1252                 }
1253
1254                 /* Set-gid? */
1255                 /*
1256                  * If setgid is set but no group execute bit then this
1257                  * is a candidate for mandatory locking, not a setgid
1258                  * executable.
1259                  */
1260                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1261                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1262                         bprm->cred->egid = inode->i_gid;
1263                 }
1264         }
1265
1266         /* fill in binprm security blob */
1267         retval = security_bprm_set_creds(bprm);
1268         if (retval)
1269                 return retval;
1270         bprm->cred_prepared = 1;
1271
1272         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1273         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1274 }
1275
1276 EXPORT_SYMBOL(prepare_binprm);
1277
1278 /*
1279  * Arguments are '\0' separated strings found at the location bprm->p
1280  * points to; chop off the first by relocating brpm->p to right after
1281  * the first '\0' encountered.
1282  */
1283 int remove_arg_zero(struct linux_binprm *bprm)
1284 {
1285         int ret = 0;
1286         unsigned long offset;
1287         char *kaddr;
1288         struct page *page;
1289
1290         if (!bprm->argc)
1291                 return 0;
1292
1293         do {
1294                 offset = bprm->p & ~PAGE_MASK;
1295                 page = get_arg_page(bprm, bprm->p, 0);
1296                 if (!page) {
1297                         ret = -EFAULT;
1298                         goto out;
1299                 }
1300                 kaddr = kmap_atomic(page, KM_USER0);
1301
1302                 for (; offset < PAGE_SIZE && kaddr[offset];
1303                                 offset++, bprm->p++)
1304                         ;
1305
1306                 kunmap_atomic(kaddr, KM_USER0);
1307                 put_arg_page(page);
1308
1309                 if (offset == PAGE_SIZE)
1310                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1311         } while (offset == PAGE_SIZE);
1312
1313         bprm->p++;
1314         bprm->argc--;
1315         ret = 0;
1316
1317 out:
1318         return ret;
1319 }
1320 EXPORT_SYMBOL(remove_arg_zero);
1321
1322 /*
1323  * cycle the list of binary formats handler, until one recognizes the image
1324  */
1325 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1326 {
1327         unsigned int depth = bprm->recursion_depth;
1328         int try,retval;
1329         struct linux_binfmt *fmt;
1330
1331         retval = security_bprm_check(bprm);
1332         if (retval)
1333                 return retval;
1334
1335         /* kernel module loader fixup */
1336         /* so we don't try to load run modprobe in kernel space. */
1337         set_fs(USER_DS);
1338
1339         retval = audit_bprm(bprm);
1340         if (retval)
1341                 return retval;
1342
1343         retval = -ENOENT;
1344         for (try=0; try<2; try++) {
1345                 read_lock(&binfmt_lock);
1346                 list_for_each_entry(fmt, &formats, lh) {
1347                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1348                         if (!fn)
1349                                 continue;
1350                         if (!try_module_get(fmt->module))
1351                                 continue;
1352                         read_unlock(&binfmt_lock);
1353                         retval = fn(bprm, regs);
1354                         /*
1355                          * Restore the depth counter to its starting value
1356                          * in this call, so we don't have to rely on every
1357                          * load_binary function to restore it on return.
1358                          */
1359                         bprm->recursion_depth = depth;
1360                         if (retval >= 0) {
1361                                 if (depth == 0)
1362                                         tracehook_report_exec(fmt, bprm, regs);
1363                                 put_binfmt(fmt);
1364                                 allow_write_access(bprm->file);
1365                                 if (bprm->file)
1366                                         fput(bprm->file);
1367                                 bprm->file = NULL;
1368                                 current->did_exec = 1;
1369                                 proc_exec_connector(current);
1370                                 return retval;
1371                         }
1372                         read_lock(&binfmt_lock);
1373                         put_binfmt(fmt);
1374                         if (retval != -ENOEXEC || bprm->mm == NULL)
1375                                 break;
1376                         if (!bprm->file) {
1377                                 read_unlock(&binfmt_lock);
1378                                 return retval;
1379                         }
1380                 }
1381                 read_unlock(&binfmt_lock);
1382                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1383                         break;
1384 #ifdef CONFIG_MODULES
1385                 } else {
1386 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1387                         if (printable(bprm->buf[0]) &&
1388                             printable(bprm->buf[1]) &&
1389                             printable(bprm->buf[2]) &&
1390                             printable(bprm->buf[3]))
1391                                 break; /* -ENOEXEC */
1392                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1393 #endif
1394                 }
1395         }
1396         return retval;
1397 }
1398
1399 EXPORT_SYMBOL(search_binary_handler);
1400
1401 /*
1402  * sys_execve() executes a new program.
1403  */
1404 static int do_execve_common(const char *filename,
1405                                 struct user_arg_ptr argv,
1406                                 struct user_arg_ptr envp,
1407                                 struct pt_regs *regs)
1408 {
1409         struct linux_binprm *bprm;
1410         struct file *file;
1411         struct files_struct *displaced;
1412         bool clear_in_exec;
1413         int retval;
1414
1415         retval = unshare_files(&displaced);
1416         if (retval)
1417                 goto out_ret;
1418
1419         retval = -ENOMEM;
1420         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1421         if (!bprm)
1422                 goto out_files;
1423
1424         retval = prepare_bprm_creds(bprm);
1425         if (retval)
1426                 goto out_free;
1427
1428         retval = check_unsafe_exec(bprm);
1429         if (retval < 0)
1430                 goto out_free;
1431         clear_in_exec = retval;
1432         current->in_execve = 1;
1433
1434         file = open_exec(filename);
1435         retval = PTR_ERR(file);
1436         if (IS_ERR(file))
1437                 goto out_unmark;
1438
1439         sched_exec();
1440
1441         bprm->file = file;
1442         bprm->filename = filename;
1443         bprm->interp = filename;
1444
1445         retval = bprm_mm_init(bprm);
1446         if (retval)
1447                 goto out_file;
1448
1449         bprm->argc = count(argv, MAX_ARG_STRINGS);
1450         if ((retval = bprm->argc) < 0)
1451                 goto out;
1452
1453         bprm->envc = count(envp, MAX_ARG_STRINGS);
1454         if ((retval = bprm->envc) < 0)
1455                 goto out;
1456
1457         retval = prepare_binprm(bprm);
1458         if (retval < 0)
1459                 goto out;
1460
1461         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1462         if (retval < 0)
1463                 goto out;
1464
1465         bprm->exec = bprm->p;
1466         retval = copy_strings(bprm->envc, envp, bprm);
1467         if (retval < 0)
1468                 goto out;
1469
1470         retval = copy_strings(bprm->argc, argv, bprm);
1471         if (retval < 0)
1472                 goto out;
1473
1474         retval = search_binary_handler(bprm,regs);
1475         if (retval < 0)
1476                 goto out;
1477
1478         /* execve succeeded */
1479         current->fs->in_exec = 0;
1480         current->in_execve = 0;
1481         acct_update_integrals(current);
1482         free_bprm(bprm);
1483         if (displaced)
1484                 put_files_struct(displaced);
1485         return retval;
1486
1487 out:
1488         if (bprm->mm) {
1489                 acct_arg_size(bprm, 0);
1490                 mmput(bprm->mm);
1491         }
1492
1493 out_file:
1494         if (bprm->file) {
1495                 allow_write_access(bprm->file);
1496                 fput(bprm->file);
1497         }
1498
1499 out_unmark:
1500         if (clear_in_exec)
1501                 current->fs->in_exec = 0;
1502         current->in_execve = 0;
1503
1504 out_free:
1505         free_bprm(bprm);
1506
1507 out_files:
1508         if (displaced)
1509                 reset_files_struct(displaced);
1510 out_ret:
1511         return retval;
1512 }
1513
1514 int do_execve(const char *filename,
1515         const char __user *const __user *__argv,
1516         const char __user *const __user *__envp,
1517         struct pt_regs *regs)
1518 {
1519         struct user_arg_ptr argv = { .native = __argv };
1520         struct user_arg_ptr envp = { .native = __envp };
1521         return do_execve_common(filename, argv, envp, regs);
1522 }
1523
1524 void set_binfmt(struct linux_binfmt *new)
1525 {
1526         struct mm_struct *mm = current->mm;
1527
1528         if (mm->binfmt)
1529                 module_put(mm->binfmt->module);
1530
1531         mm->binfmt = new;
1532         if (new)
1533                 __module_get(new->module);
1534 }
1535
1536 EXPORT_SYMBOL(set_binfmt);
1537
1538 static int expand_corename(struct core_name *cn)
1539 {
1540         char *old_corename = cn->corename;
1541
1542         cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1543         cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1544
1545         if (!cn->corename) {
1546                 kfree(old_corename);
1547                 return -ENOMEM;
1548         }
1549
1550         return 0;
1551 }
1552
1553 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1554 {
1555         char *cur;
1556         int need;
1557         int ret;
1558         va_list arg;
1559
1560         va_start(arg, fmt);
1561         need = vsnprintf(NULL, 0, fmt, arg);
1562         va_end(arg);
1563
1564         if (likely(need < cn->size - cn->used - 1))
1565                 goto out_printf;
1566
1567         ret = expand_corename(cn);
1568         if (ret)
1569                 goto expand_fail;
1570
1571 out_printf:
1572         cur = cn->corename + cn->used;
1573         va_start(arg, fmt);
1574         vsnprintf(cur, need + 1, fmt, arg);
1575         va_end(arg);
1576         cn->used += need;
1577         return 0;
1578
1579 expand_fail:
1580         return ret;
1581 }
1582
1583 /* format_corename will inspect the pattern parameter, and output a
1584  * name into corename, which must have space for at least
1585  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1586  */
1587 static int format_corename(struct core_name *cn, long signr)
1588 {
1589         const struct cred *cred = current_cred();
1590         const char *pat_ptr = core_pattern;
1591         int ispipe = (*pat_ptr == '|');
1592         int pid_in_pattern = 0;
1593         int err = 0;
1594
1595         cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1596         cn->corename = kmalloc(cn->size, GFP_KERNEL);
1597         cn->used = 0;
1598
1599         if (!cn->corename)
1600                 return -ENOMEM;
1601
1602         /* Repeat as long as we have more pattern to process and more output
1603            space */
1604         while (*pat_ptr) {
1605                 if (*pat_ptr != '%') {
1606                         if (*pat_ptr == 0)
1607                                 goto out;
1608                         err = cn_printf(cn, "%c", *pat_ptr++);
1609                 } else {
1610                         switch (*++pat_ptr) {
1611                         /* single % at the end, drop that */
1612                         case 0:
1613                                 goto out;
1614                         /* Double percent, output one percent */
1615                         case '%':
1616                                 err = cn_printf(cn, "%c", '%');
1617                                 break;
1618                         /* pid */
1619                         case 'p':
1620                                 pid_in_pattern = 1;
1621                                 err = cn_printf(cn, "%d",
1622                                               task_tgid_vnr(current));
1623                                 break;
1624                         /* uid */
1625                         case 'u':
1626                                 err = cn_printf(cn, "%d", cred->uid);
1627                                 break;
1628                         /* gid */
1629                         case 'g':
1630                                 err = cn_printf(cn, "%d", cred->gid);
1631                                 break;
1632                         /* signal that caused the coredump */
1633                         case 's':
1634                                 err = cn_printf(cn, "%ld", signr);
1635                                 break;
1636                         /* UNIX time of coredump */
1637                         case 't': {
1638                                 struct timeval tv;
1639                                 do_gettimeofday(&tv);
1640                                 err = cn_printf(cn, "%lu", tv.tv_sec);
1641                                 break;
1642                         }
1643                         /* hostname */
1644                         case 'h':
1645                                 down_read(&uts_sem);
1646                                 err = cn_printf(cn, "%s",
1647                                               utsname()->nodename);
1648                                 up_read(&uts_sem);
1649                                 break;
1650                         /* executable */
1651                         case 'e':
1652                                 err = cn_printf(cn, "%s", current->comm);
1653                                 break;
1654                         /* core limit size */
1655                         case 'c':
1656                                 err = cn_printf(cn, "%lu",
1657                                               rlimit(RLIMIT_CORE));
1658                                 break;
1659                         default:
1660                                 break;
1661                         }
1662                         ++pat_ptr;
1663                 }
1664
1665                 if (err)
1666                         return err;
1667         }
1668
1669         /* Backward compatibility with core_uses_pid:
1670          *
1671          * If core_pattern does not include a %p (as is the default)
1672          * and core_uses_pid is set, then .%pid will be appended to
1673          * the filename. Do not do this for piped commands. */
1674         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1675                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1676                 if (err)
1677                         return err;
1678         }
1679 out:
1680         return ispipe;
1681 }
1682
1683 static int zap_process(struct task_struct *start, int exit_code)
1684 {
1685         struct task_struct *t;
1686         int nr = 0;
1687
1688         start->signal->flags = SIGNAL_GROUP_EXIT;
1689         start->signal->group_exit_code = exit_code;
1690         start->signal->group_stop_count = 0;
1691
1692         t = start;
1693         do {
1694                 if (t != current && t->mm) {
1695                         sigaddset(&t->pending.signal, SIGKILL);
1696                         signal_wake_up(t, 1);
1697                         nr++;
1698                 }
1699         } while_each_thread(start, t);
1700
1701         return nr;
1702 }
1703
1704 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1705                                 struct core_state *core_state, int exit_code)
1706 {
1707         struct task_struct *g, *p;
1708         unsigned long flags;
1709         int nr = -EAGAIN;
1710
1711         spin_lock_irq(&tsk->sighand->siglock);
1712         if (!signal_group_exit(tsk->signal)) {
1713                 mm->core_state = core_state;
1714                 nr = zap_process(tsk, exit_code);
1715         }
1716         spin_unlock_irq(&tsk->sighand->siglock);
1717         if (unlikely(nr < 0))
1718                 return nr;
1719
1720         if (atomic_read(&mm->mm_users) == nr + 1)
1721                 goto done;
1722         /*
1723          * We should find and kill all tasks which use this mm, and we should
1724          * count them correctly into ->nr_threads. We don't take tasklist
1725          * lock, but this is safe wrt:
1726          *
1727          * fork:
1728          *      None of sub-threads can fork after zap_process(leader). All
1729          *      processes which were created before this point should be
1730          *      visible to zap_threads() because copy_process() adds the new
1731          *      process to the tail of init_task.tasks list, and lock/unlock
1732          *      of ->siglock provides a memory barrier.
1733          *
1734          * do_exit:
1735          *      The caller holds mm->mmap_sem. This means that the task which
1736          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1737          *      its ->mm.
1738          *
1739          * de_thread:
1740          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1741          *      we must see either old or new leader, this does not matter.
1742          *      However, it can change p->sighand, so lock_task_sighand(p)
1743          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1744          *      it can't fail.
1745          *
1746          *      Note also that "g" can be the old leader with ->mm == NULL
1747          *      and already unhashed and thus removed from ->thread_group.
1748          *      This is OK, __unhash_process()->list_del_rcu() does not
1749          *      clear the ->next pointer, we will find the new leader via
1750          *      next_thread().
1751          */
1752         rcu_read_lock();
1753         for_each_process(g) {
1754                 if (g == tsk->group_leader)
1755                         continue;
1756                 if (g->flags & PF_KTHREAD)
1757                         continue;
1758                 p = g;
1759                 do {
1760                         if (p->mm) {
1761                                 if (unlikely(p->mm == mm)) {
1762                                         lock_task_sighand(p, &flags);
1763                                         nr += zap_process(p, exit_code);
1764                                         unlock_task_sighand(p, &flags);
1765                                 }
1766                                 break;
1767                         }
1768                 } while_each_thread(g, p);
1769         }
1770         rcu_read_unlock();
1771 done:
1772         atomic_set(&core_state->nr_threads, nr);
1773         return nr;
1774 }
1775
1776 static int coredump_wait(int exit_code, struct core_state *core_state)
1777 {
1778         struct task_struct *tsk = current;
1779         struct mm_struct *mm = tsk->mm;
1780         struct completion *vfork_done;
1781         int core_waiters = -EBUSY;
1782
1783         init_completion(&core_state->startup);
1784         core_state->dumper.task = tsk;
1785         core_state->dumper.next = NULL;
1786
1787         down_write(&mm->mmap_sem);
1788         if (!mm->core_state)
1789                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1790         up_write(&mm->mmap_sem);
1791
1792         if (unlikely(core_waiters < 0))
1793                 goto fail;
1794
1795         /*
1796          * Make sure nobody is waiting for us to release the VM,
1797          * otherwise we can deadlock when we wait on each other
1798          */
1799         vfork_done = tsk->vfork_done;
1800         if (vfork_done) {
1801                 tsk->vfork_done = NULL;
1802                 complete(vfork_done);
1803         }
1804
1805         if (core_waiters)
1806                 wait_for_completion(&core_state->startup);
1807 fail:
1808         return core_waiters;
1809 }
1810
1811 static void coredump_finish(struct mm_struct *mm)
1812 {
1813         struct core_thread *curr, *next;
1814         struct task_struct *task;
1815
1816         next = mm->core_state->dumper.next;
1817         while ((curr = next) != NULL) {
1818                 next = curr->next;
1819                 task = curr->task;
1820                 /*
1821                  * see exit_mm(), curr->task must not see
1822                  * ->task == NULL before we read ->next.
1823                  */
1824                 smp_mb();
1825                 curr->task = NULL;
1826                 wake_up_process(task);
1827         }
1828
1829         mm->core_state = NULL;
1830 }
1831
1832 /*
1833  * set_dumpable converts traditional three-value dumpable to two flags and
1834  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1835  * these bits are not changed atomically.  So get_dumpable can observe the
1836  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1837  * return either old dumpable or new one by paying attention to the order of
1838  * modifying the bits.
1839  *
1840  * dumpable |   mm->flags (binary)
1841  * old  new | initial interim  final
1842  * ---------+-----------------------
1843  *  0    1  |   00      01      01
1844  *  0    2  |   00      10(*)   11
1845  *  1    0  |   01      00      00
1846  *  1    2  |   01      11      11
1847  *  2    0  |   11      10(*)   00
1848  *  2    1  |   11      11      01
1849  *
1850  * (*) get_dumpable regards interim value of 10 as 11.
1851  */
1852 void set_dumpable(struct mm_struct *mm, int value)
1853 {
1854         switch (value) {
1855         case 0:
1856                 clear_bit(MMF_DUMPABLE, &mm->flags);
1857                 smp_wmb();
1858                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1859                 break;
1860         case 1:
1861                 set_bit(MMF_DUMPABLE, &mm->flags);
1862                 smp_wmb();
1863                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1864                 break;
1865         case 2:
1866                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1867                 smp_wmb();
1868                 set_bit(MMF_DUMPABLE, &mm->flags);
1869                 break;
1870         }
1871 }
1872
1873 static int __get_dumpable(unsigned long mm_flags)
1874 {
1875         int ret;
1876
1877         ret = mm_flags & MMF_DUMPABLE_MASK;
1878         return (ret >= 2) ? 2 : ret;
1879 }
1880
1881 int get_dumpable(struct mm_struct *mm)
1882 {
1883         return __get_dumpable(mm->flags);
1884 }
1885
1886 static void wait_for_dump_helpers(struct file *file)
1887 {
1888         struct pipe_inode_info *pipe;
1889
1890         pipe = file->f_path.dentry->d_inode->i_pipe;
1891
1892         pipe_lock(pipe);
1893         pipe->readers++;
1894         pipe->writers--;
1895
1896         while ((pipe->readers > 1) && (!signal_pending(current))) {
1897                 wake_up_interruptible_sync(&pipe->wait);
1898                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1899                 pipe_wait(pipe);
1900         }
1901
1902         pipe->readers--;
1903         pipe->writers++;
1904         pipe_unlock(pipe);
1905
1906 }
1907
1908
1909 /*
1910  * umh_pipe_setup
1911  * helper function to customize the process used
1912  * to collect the core in userspace.  Specifically
1913  * it sets up a pipe and installs it as fd 0 (stdin)
1914  * for the process.  Returns 0 on success, or
1915  * PTR_ERR on failure.
1916  * Note that it also sets the core limit to 1.  This
1917  * is a special value that we use to trap recursive
1918  * core dumps
1919  */
1920 static int umh_pipe_setup(struct subprocess_info *info)
1921 {
1922         struct file *rp, *wp;
1923         struct fdtable *fdt;
1924         struct coredump_params *cp = (struct coredump_params *)info->data;
1925         struct files_struct *cf = current->files;
1926
1927         wp = create_write_pipe(0);
1928         if (IS_ERR(wp))
1929                 return PTR_ERR(wp);
1930
1931         rp = create_read_pipe(wp, 0);
1932         if (IS_ERR(rp)) {
1933                 free_write_pipe(wp);
1934                 return PTR_ERR(rp);
1935         }
1936
1937         cp->file = wp;
1938
1939         sys_close(0);
1940         fd_install(0, rp);
1941         spin_lock(&cf->file_lock);
1942         fdt = files_fdtable(cf);
1943         FD_SET(0, fdt->open_fds);
1944         FD_CLR(0, fdt->close_on_exec);
1945         spin_unlock(&cf->file_lock);
1946
1947         /* and disallow core files too */
1948         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1949
1950         return 0;
1951 }
1952
1953 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1954 {
1955         struct core_state core_state;
1956         struct core_name cn;
1957         struct mm_struct *mm = current->mm;
1958         struct linux_binfmt * binfmt;
1959         const struct cred *old_cred;
1960         struct cred *cred;
1961         int retval = 0;
1962         int flag = 0;
1963         int ispipe;
1964         static atomic_t core_dump_count = ATOMIC_INIT(0);
1965         struct coredump_params cprm = {
1966                 .signr = signr,
1967                 .regs = regs,
1968                 .limit = rlimit(RLIMIT_CORE),
1969                 /*
1970                  * We must use the same mm->flags while dumping core to avoid
1971                  * inconsistency of bit flags, since this flag is not protected
1972                  * by any locks.
1973                  */
1974                 .mm_flags = mm->flags,
1975         };
1976
1977         audit_core_dumps(signr);
1978
1979         binfmt = mm->binfmt;
1980         if (!binfmt || !binfmt->core_dump)
1981                 goto fail;
1982         if (!__get_dumpable(cprm.mm_flags))
1983                 goto fail;
1984
1985         cred = prepare_creds();
1986         if (!cred)
1987                 goto fail;
1988         /*
1989          *      We cannot trust fsuid as being the "true" uid of the
1990          *      process nor do we know its entire history. We only know it
1991          *      was tainted so we dump it as root in mode 2.
1992          */
1993         if (__get_dumpable(cprm.mm_flags) == 2) {
1994                 /* Setuid core dump mode */
1995                 flag = O_EXCL;          /* Stop rewrite attacks */
1996                 cred->fsuid = 0;        /* Dump root private */
1997         }
1998
1999         retval = coredump_wait(exit_code, &core_state);
2000         if (retval < 0)
2001                 goto fail_creds;
2002
2003         old_cred = override_creds(cred);
2004
2005         /*
2006          * Clear any false indication of pending signals that might
2007          * be seen by the filesystem code called to write the core file.
2008          */
2009         clear_thread_flag(TIF_SIGPENDING);
2010
2011         ispipe = format_corename(&cn, signr);
2012
2013         if (ispipe == -ENOMEM) {
2014                 printk(KERN_WARNING "format_corename failed\n");
2015                 printk(KERN_WARNING "Aborting core\n");
2016                 goto fail_corename;
2017         }
2018
2019         if (ispipe) {
2020                 int dump_count;
2021                 char **helper_argv;
2022
2023                 if (cprm.limit == 1) {
2024                         /*
2025                          * Normally core limits are irrelevant to pipes, since
2026                          * we're not writing to the file system, but we use
2027                          * cprm.limit of 1 here as a speacial value. Any
2028                          * non-1 limit gets set to RLIM_INFINITY below, but
2029                          * a limit of 0 skips the dump.  This is a consistent
2030                          * way to catch recursive crashes.  We can still crash
2031                          * if the core_pattern binary sets RLIM_CORE =  !1
2032                          * but it runs as root, and can do lots of stupid things
2033                          * Note that we use task_tgid_vnr here to grab the pid
2034                          * of the process group leader.  That way we get the
2035                          * right pid if a thread in a multi-threaded
2036                          * core_pattern process dies.
2037                          */
2038                         printk(KERN_WARNING
2039                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2040                                 task_tgid_vnr(current), current->comm);
2041                         printk(KERN_WARNING "Aborting core\n");
2042                         goto fail_unlock;
2043                 }
2044                 cprm.limit = RLIM_INFINITY;
2045
2046                 dump_count = atomic_inc_return(&core_dump_count);
2047                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2048                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2049                                task_tgid_vnr(current), current->comm);
2050                         printk(KERN_WARNING "Skipping core dump\n");
2051                         goto fail_dropcount;
2052                 }
2053
2054                 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2055                 if (!helper_argv) {
2056                         printk(KERN_WARNING "%s failed to allocate memory\n",
2057                                __func__);
2058                         goto fail_dropcount;
2059                 }
2060
2061                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2062                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2063                                         NULL, &cprm);
2064                 argv_free(helper_argv);
2065                 if (retval) {
2066                         printk(KERN_INFO "Core dump to %s pipe failed\n",
2067                                cn.corename);
2068                         goto close_fail;
2069                 }
2070         } else {
2071                 struct inode *inode;
2072
2073                 if (cprm.limit < binfmt->min_coredump)
2074                         goto fail_unlock;
2075
2076                 cprm.file = filp_open(cn.corename,
2077                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2078                                  0600);
2079                 if (IS_ERR(cprm.file))
2080                         goto fail_unlock;
2081
2082                 inode = cprm.file->f_path.dentry->d_inode;
2083                 if (inode->i_nlink > 1)
2084                         goto close_fail;
2085                 if (d_unhashed(cprm.file->f_path.dentry))
2086                         goto close_fail;
2087                 /*
2088                  * AK: actually i see no reason to not allow this for named
2089                  * pipes etc, but keep the previous behaviour for now.
2090                  */
2091                 if (!S_ISREG(inode->i_mode))
2092                         goto close_fail;
2093                 /*
2094                  * Dont allow local users get cute and trick others to coredump
2095                  * into their pre-created files.
2096                  */
2097                 if (inode->i_uid != current_fsuid())
2098                         goto close_fail;
2099                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2100                         goto close_fail;
2101                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2102                         goto close_fail;
2103         }
2104
2105         retval = binfmt->core_dump(&cprm);
2106         if (retval)
2107                 current->signal->group_exit_code |= 0x80;
2108
2109         if (ispipe && core_pipe_limit)
2110                 wait_for_dump_helpers(cprm.file);
2111 close_fail:
2112         if (cprm.file)
2113                 filp_close(cprm.file, NULL);
2114 fail_dropcount:
2115         if (ispipe)
2116                 atomic_dec(&core_dump_count);
2117 fail_unlock:
2118         kfree(cn.corename);
2119 fail_corename:
2120         coredump_finish(mm);
2121         revert_creds(old_cred);
2122 fail_creds:
2123         put_cred(cred);
2124 fail:
2125         return;
2126 }
2127
2128 /*
2129  * Core dumping helper functions.  These are the only things you should
2130  * do on a core-file: use only these functions to write out all the
2131  * necessary info.
2132  */
2133 int dump_write(struct file *file, const void *addr, int nr)
2134 {
2135         return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2136 }
2137 EXPORT_SYMBOL(dump_write);
2138
2139 int dump_seek(struct file *file, loff_t off)
2140 {
2141         int ret = 1;
2142
2143         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2144                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2145                         return 0;
2146         } else {
2147                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2148
2149                 if (!buf)
2150                         return 0;
2151                 while (off > 0) {
2152                         unsigned long n = off;
2153
2154                         if (n > PAGE_SIZE)
2155                                 n = PAGE_SIZE;
2156                         if (!dump_write(file, buf, n)) {
2157                                 ret = 0;
2158                                 break;
2159                         }
2160                         off -= n;
2161                 }
2162                 free_page((unsigned long)buf);
2163         }
2164         return ret;
2165 }
2166 EXPORT_SYMBOL(dump_seek);