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