mm: correctly synchronize rss-counters at exit/exec
[linux-3.10.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/mount.h>
46 #include <linux/security.h>
47 #include <linux/syscalls.h>
48 #include <linux/tsacct_kern.h>
49 #include <linux/cn_proc.h>
50 #include <linux/audit.h>
51 #include <linux/tracehook.h>
52 #include <linux/kmod.h>
53 #include <linux/fsnotify.h>
54 #include <linux/fs_struct.h>
55 #include <linux/pipe_fs_i.h>
56 #include <linux/oom.h>
57 #include <linux/compat.h>
58
59 #include <asm/uaccess.h>
60 #include <asm/mmu_context.h>
61 #include <asm/tlb.h>
62 #include <asm/exec.h>
63
64 #include <trace/events/task.h>
65 #include "internal.h"
66
67 #include <trace/events/sched.h>
68
69 int core_uses_pid;
70 char core_pattern[CORENAME_MAX_SIZE] = "core";
71 unsigned int core_pipe_limit;
72 int suid_dumpable = 0;
73
74 struct core_name {
75         char *corename;
76         int used, size;
77 };
78 static atomic_t call_count = ATOMIC_INIT(1);
79
80 /* The maximal length of core_pattern is also specified in sysctl.c */
81
82 static LIST_HEAD(formats);
83 static DEFINE_RWLOCK(binfmt_lock);
84
85 void __register_binfmt(struct linux_binfmt * fmt, int insert)
86 {
87         BUG_ON(!fmt);
88         write_lock(&binfmt_lock);
89         insert ? list_add(&fmt->lh, &formats) :
90                  list_add_tail(&fmt->lh, &formats);
91         write_unlock(&binfmt_lock);
92 }
93
94 EXPORT_SYMBOL(__register_binfmt);
95
96 void unregister_binfmt(struct linux_binfmt * fmt)
97 {
98         write_lock(&binfmt_lock);
99         list_del(&fmt->lh);
100         write_unlock(&binfmt_lock);
101 }
102
103 EXPORT_SYMBOL(unregister_binfmt);
104
105 static inline void put_binfmt(struct linux_binfmt * fmt)
106 {
107         module_put(fmt->module);
108 }
109
110 /*
111  * Note that a shared library must be both readable and executable due to
112  * security reasons.
113  *
114  * Also note that we take the address to load from from the file itself.
115  */
116 SYSCALL_DEFINE1(uselib, const char __user *, library)
117 {
118         struct file *file;
119         char *tmp = getname(library);
120         int error = PTR_ERR(tmp);
121         static const struct open_flags uselib_flags = {
122                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
123                 .acc_mode = MAY_READ | MAY_EXEC | MAY_OPEN,
124                 .intent = LOOKUP_OPEN
125         };
126
127         if (IS_ERR(tmp))
128                 goto out;
129
130         file = do_filp_open(AT_FDCWD, tmp, &uselib_flags, LOOKUP_FOLLOW);
131         putname(tmp);
132         error = PTR_ERR(file);
133         if (IS_ERR(file))
134                 goto out;
135
136         error = -EINVAL;
137         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
138                 goto exit;
139
140         error = -EACCES;
141         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
142                 goto exit;
143
144         fsnotify_open(file);
145
146         error = -ENOEXEC;
147         if(file->f_op) {
148                 struct linux_binfmt * fmt;
149
150                 read_lock(&binfmt_lock);
151                 list_for_each_entry(fmt, &formats, lh) {
152                         if (!fmt->load_shlib)
153                                 continue;
154                         if (!try_module_get(fmt->module))
155                                 continue;
156                         read_unlock(&binfmt_lock);
157                         error = fmt->load_shlib(file);
158                         read_lock(&binfmt_lock);
159                         put_binfmt(fmt);
160                         if (error != -ENOEXEC)
161                                 break;
162                 }
163                 read_unlock(&binfmt_lock);
164         }
165 exit:
166         fput(file);
167 out:
168         return error;
169 }
170
171 #ifdef CONFIG_MMU
172 /*
173  * The nascent bprm->mm is not visible until exec_mmap() but it can
174  * use a lot of memory, account these pages in current->mm temporary
175  * for oom_badness()->get_mm_rss(). Once exec succeeds or fails, we
176  * change the counter back via acct_arg_size(0).
177  */
178 static void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
179 {
180         struct mm_struct *mm = current->mm;
181         long diff = (long)(pages - bprm->vma_pages);
182
183         if (!mm || !diff)
184                 return;
185
186         bprm->vma_pages = pages;
187         add_mm_counter(mm, MM_ANONPAGES, diff);
188 }
189
190 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
191                 int write)
192 {
193         struct page *page;
194         int ret;
195
196 #ifdef CONFIG_STACK_GROWSUP
197         if (write) {
198                 ret = expand_downwards(bprm->vma, pos);
199                 if (ret < 0)
200                         return NULL;
201         }
202 #endif
203         ret = get_user_pages(current, bprm->mm, pos,
204                         1, write, 1, &page, NULL);
205         if (ret <= 0)
206                 return NULL;
207
208         if (write) {
209                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
210                 struct rlimit *rlim;
211
212                 acct_arg_size(bprm, size / PAGE_SIZE);
213
214                 /*
215                  * We've historically supported up to 32 pages (ARG_MAX)
216                  * of argument strings even with small stacks
217                  */
218                 if (size <= ARG_MAX)
219                         return page;
220
221                 /*
222                  * Limit to 1/4-th the stack size for the argv+env strings.
223                  * This ensures that:
224                  *  - the remaining binfmt code will not run out of stack space,
225                  *  - the program will have a reasonable amount of stack left
226                  *    to work from.
227                  */
228                 rlim = current->signal->rlim;
229                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
230                         put_page(page);
231                         return NULL;
232                 }
233         }
234
235         return page;
236 }
237
238 static void put_arg_page(struct page *page)
239 {
240         put_page(page);
241 }
242
243 static void free_arg_page(struct linux_binprm *bprm, int i)
244 {
245 }
246
247 static void free_arg_pages(struct linux_binprm *bprm)
248 {
249 }
250
251 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
252                 struct page *page)
253 {
254         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
255 }
256
257 static int __bprm_mm_init(struct linux_binprm *bprm)
258 {
259         int err;
260         struct vm_area_struct *vma = NULL;
261         struct mm_struct *mm = bprm->mm;
262
263         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
264         if (!vma)
265                 return -ENOMEM;
266
267         down_write(&mm->mmap_sem);
268         vma->vm_mm = mm;
269
270         /*
271          * Place the stack at the largest stack address the architecture
272          * supports. Later, we'll move this to an appropriate place. We don't
273          * use STACK_TOP because that can depend on attributes which aren't
274          * configured yet.
275          */
276         BUILD_BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
277         vma->vm_end = STACK_TOP_MAX;
278         vma->vm_start = vma->vm_end - PAGE_SIZE;
279         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
280         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
281         INIT_LIST_HEAD(&vma->anon_vma_chain);
282
283         err = insert_vm_struct(mm, vma);
284         if (err)
285                 goto err;
286
287         mm->stack_vm = mm->total_vm = 1;
288         up_write(&mm->mmap_sem);
289         bprm->p = vma->vm_end - sizeof(void *);
290         return 0;
291 err:
292         up_write(&mm->mmap_sem);
293         bprm->vma = NULL;
294         kmem_cache_free(vm_area_cachep, vma);
295         return err;
296 }
297
298 static bool valid_arg_len(struct linux_binprm *bprm, long len)
299 {
300         return len <= MAX_ARG_STRLEN;
301 }
302
303 #else
304
305 static inline void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
306 {
307 }
308
309 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
310                 int write)
311 {
312         struct page *page;
313
314         page = bprm->page[pos / PAGE_SIZE];
315         if (!page && write) {
316                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
317                 if (!page)
318                         return NULL;
319                 bprm->page[pos / PAGE_SIZE] = page;
320         }
321
322         return page;
323 }
324
325 static void put_arg_page(struct page *page)
326 {
327 }
328
329 static void free_arg_page(struct linux_binprm *bprm, int i)
330 {
331         if (bprm->page[i]) {
332                 __free_page(bprm->page[i]);
333                 bprm->page[i] = NULL;
334         }
335 }
336
337 static void free_arg_pages(struct linux_binprm *bprm)
338 {
339         int i;
340
341         for (i = 0; i < MAX_ARG_PAGES; i++)
342                 free_arg_page(bprm, i);
343 }
344
345 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
346                 struct page *page)
347 {
348 }
349
350 static int __bprm_mm_init(struct linux_binprm *bprm)
351 {
352         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
353         return 0;
354 }
355
356 static bool valid_arg_len(struct linux_binprm *bprm, long len)
357 {
358         return len <= bprm->p;
359 }
360
361 #endif /* CONFIG_MMU */
362
363 /*
364  * Create a new mm_struct and populate it with a temporary stack
365  * vm_area_struct.  We don't have enough context at this point to set the stack
366  * flags, permissions, and offset, so we use temporary values.  We'll update
367  * them later in setup_arg_pages().
368  */
369 int bprm_mm_init(struct linux_binprm *bprm)
370 {
371         int err;
372         struct mm_struct *mm = NULL;
373
374         bprm->mm = mm = mm_alloc();
375         err = -ENOMEM;
376         if (!mm)
377                 goto err;
378
379         err = init_new_context(current, mm);
380         if (err)
381                 goto err;
382
383         err = __bprm_mm_init(bprm);
384         if (err)
385                 goto err;
386
387         return 0;
388
389 err:
390         if (mm) {
391                 bprm->mm = NULL;
392                 mmdrop(mm);
393         }
394
395         return err;
396 }
397
398 struct user_arg_ptr {
399 #ifdef CONFIG_COMPAT
400         bool is_compat;
401 #endif
402         union {
403                 const char __user *const __user *native;
404 #ifdef CONFIG_COMPAT
405                 compat_uptr_t __user *compat;
406 #endif
407         } ptr;
408 };
409
410 static const char __user *get_user_arg_ptr(struct user_arg_ptr argv, int nr)
411 {
412         const char __user *native;
413
414 #ifdef CONFIG_COMPAT
415         if (unlikely(argv.is_compat)) {
416                 compat_uptr_t compat;
417
418                 if (get_user(compat, argv.ptr.compat + nr))
419                         return ERR_PTR(-EFAULT);
420
421                 return compat_ptr(compat);
422         }
423 #endif
424
425         if (get_user(native, argv.ptr.native + nr))
426                 return ERR_PTR(-EFAULT);
427
428         return native;
429 }
430
431 /*
432  * count() counts the number of strings in array ARGV.
433  */
434 static int count(struct user_arg_ptr argv, int max)
435 {
436         int i = 0;
437
438         if (argv.ptr.native != NULL) {
439                 for (;;) {
440                         const char __user *p = get_user_arg_ptr(argv, i);
441
442                         if (!p)
443                                 break;
444
445                         if (IS_ERR(p))
446                                 return -EFAULT;
447
448                         if (i++ >= max)
449                                 return -E2BIG;
450
451                         if (fatal_signal_pending(current))
452                                 return -ERESTARTNOHAND;
453                         cond_resched();
454                 }
455         }
456         return i;
457 }
458
459 /*
460  * 'copy_strings()' copies argument/environment strings from the old
461  * processes's memory to the new process's stack.  The call to get_user_pages()
462  * ensures the destination page is created and not swapped out.
463  */
464 static int copy_strings(int argc, struct user_arg_ptr argv,
465                         struct linux_binprm *bprm)
466 {
467         struct page *kmapped_page = NULL;
468         char *kaddr = NULL;
469         unsigned long kpos = 0;
470         int ret;
471
472         while (argc-- > 0) {
473                 const char __user *str;
474                 int len;
475                 unsigned long pos;
476
477                 ret = -EFAULT;
478                 str = get_user_arg_ptr(argv, argc);
479                 if (IS_ERR(str))
480                         goto out;
481
482                 len = strnlen_user(str, MAX_ARG_STRLEN);
483                 if (!len)
484                         goto out;
485
486                 ret = -E2BIG;
487                 if (!valid_arg_len(bprm, len))
488                         goto out;
489
490                 /* We're going to work our way backwords. */
491                 pos = bprm->p;
492                 str += len;
493                 bprm->p -= len;
494
495                 while (len > 0) {
496                         int offset, bytes_to_copy;
497
498                         if (fatal_signal_pending(current)) {
499                                 ret = -ERESTARTNOHAND;
500                                 goto out;
501                         }
502                         cond_resched();
503
504                         offset = pos % PAGE_SIZE;
505                         if (offset == 0)
506                                 offset = PAGE_SIZE;
507
508                         bytes_to_copy = offset;
509                         if (bytes_to_copy > len)
510                                 bytes_to_copy = len;
511
512                         offset -= bytes_to_copy;
513                         pos -= bytes_to_copy;
514                         str -= bytes_to_copy;
515                         len -= bytes_to_copy;
516
517                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
518                                 struct page *page;
519
520                                 page = get_arg_page(bprm, pos, 1);
521                                 if (!page) {
522                                         ret = -E2BIG;
523                                         goto out;
524                                 }
525
526                                 if (kmapped_page) {
527                                         flush_kernel_dcache_page(kmapped_page);
528                                         kunmap(kmapped_page);
529                                         put_arg_page(kmapped_page);
530                                 }
531                                 kmapped_page = page;
532                                 kaddr = kmap(kmapped_page);
533                                 kpos = pos & PAGE_MASK;
534                                 flush_arg_page(bprm, kpos, kmapped_page);
535                         }
536                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
537                                 ret = -EFAULT;
538                                 goto out;
539                         }
540                 }
541         }
542         ret = 0;
543 out:
544         if (kmapped_page) {
545                 flush_kernel_dcache_page(kmapped_page);
546                 kunmap(kmapped_page);
547                 put_arg_page(kmapped_page);
548         }
549         return ret;
550 }
551
552 /*
553  * Like copy_strings, but get argv and its values from kernel memory.
554  */
555 int copy_strings_kernel(int argc, const char *const *__argv,
556                         struct linux_binprm *bprm)
557 {
558         int r;
559         mm_segment_t oldfs = get_fs();
560         struct user_arg_ptr argv = {
561                 .ptr.native = (const char __user *const  __user *)__argv,
562         };
563
564         set_fs(KERNEL_DS);
565         r = copy_strings(argc, argv, bprm);
566         set_fs(oldfs);
567
568         return r;
569 }
570 EXPORT_SYMBOL(copy_strings_kernel);
571
572 #ifdef CONFIG_MMU
573
574 /*
575  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
576  * the binfmt code determines where the new stack should reside, we shift it to
577  * its final location.  The process proceeds as follows:
578  *
579  * 1) Use shift to calculate the new vma endpoints.
580  * 2) Extend vma to cover both the old and new ranges.  This ensures the
581  *    arguments passed to subsequent functions are consistent.
582  * 3) Move vma's page tables to the new range.
583  * 4) Free up any cleared pgd range.
584  * 5) Shrink the vma to cover only the new range.
585  */
586 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
587 {
588         struct mm_struct *mm = vma->vm_mm;
589         unsigned long old_start = vma->vm_start;
590         unsigned long old_end = vma->vm_end;
591         unsigned long length = old_end - old_start;
592         unsigned long new_start = old_start - shift;
593         unsigned long new_end = old_end - shift;
594         struct mmu_gather tlb;
595
596         BUG_ON(new_start > new_end);
597
598         /*
599          * ensure there are no vmas between where we want to go
600          * and where we are
601          */
602         if (vma != find_vma(mm, new_start))
603                 return -EFAULT;
604
605         /*
606          * cover the whole range: [new_start, old_end)
607          */
608         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
609                 return -ENOMEM;
610
611         /*
612          * move the page tables downwards, on failure we rely on
613          * process cleanup to remove whatever mess we made.
614          */
615         if (length != move_page_tables(vma, old_start,
616                                        vma, new_start, length))
617                 return -ENOMEM;
618
619         lru_add_drain();
620         tlb_gather_mmu(&tlb, mm, 0);
621         if (new_end > old_start) {
622                 /*
623                  * when the old and new regions overlap clear from new_end.
624                  */
625                 free_pgd_range(&tlb, new_end, old_end, new_end,
626                         vma->vm_next ? vma->vm_next->vm_start : 0);
627         } else {
628                 /*
629                  * otherwise, clean from old_start; this is done to not touch
630                  * the address space in [new_end, old_start) some architectures
631                  * have constraints on va-space that make this illegal (IA64) -
632                  * for the others its just a little faster.
633                  */
634                 free_pgd_range(&tlb, old_start, old_end, new_end,
635                         vma->vm_next ? vma->vm_next->vm_start : 0);
636         }
637         tlb_finish_mmu(&tlb, new_end, old_end);
638
639         /*
640          * Shrink the vma to just the new range.  Always succeeds.
641          */
642         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
643
644         return 0;
645 }
646
647 /*
648  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
649  * the stack is optionally relocated, and some extra space is added.
650  */
651 int setup_arg_pages(struct linux_binprm *bprm,
652                     unsigned long stack_top,
653                     int executable_stack)
654 {
655         unsigned long ret;
656         unsigned long stack_shift;
657         struct mm_struct *mm = current->mm;
658         struct vm_area_struct *vma = bprm->vma;
659         struct vm_area_struct *prev = NULL;
660         unsigned long vm_flags;
661         unsigned long stack_base;
662         unsigned long stack_size;
663         unsigned long stack_expand;
664         unsigned long rlim_stack;
665
666 #ifdef CONFIG_STACK_GROWSUP
667         /* Limit stack size to 1GB */
668         stack_base = rlimit_max(RLIMIT_STACK);
669         if (stack_base > (1 << 30))
670                 stack_base = 1 << 30;
671
672         /* Make sure we didn't let the argument array grow too large. */
673         if (vma->vm_end - vma->vm_start > stack_base)
674                 return -ENOMEM;
675
676         stack_base = PAGE_ALIGN(stack_top - stack_base);
677
678         stack_shift = vma->vm_start - stack_base;
679         mm->arg_start = bprm->p - stack_shift;
680         bprm->p = vma->vm_end - stack_shift;
681 #else
682         stack_top = arch_align_stack(stack_top);
683         stack_top = PAGE_ALIGN(stack_top);
684
685         if (unlikely(stack_top < mmap_min_addr) ||
686             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
687                 return -ENOMEM;
688
689         stack_shift = vma->vm_end - stack_top;
690
691         bprm->p -= stack_shift;
692         mm->arg_start = bprm->p;
693 #endif
694
695         if (bprm->loader)
696                 bprm->loader -= stack_shift;
697         bprm->exec -= stack_shift;
698
699         down_write(&mm->mmap_sem);
700         vm_flags = VM_STACK_FLAGS;
701
702         /*
703          * Adjust stack execute permissions; explicitly enable for
704          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
705          * (arch default) otherwise.
706          */
707         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
708                 vm_flags |= VM_EXEC;
709         else if (executable_stack == EXSTACK_DISABLE_X)
710                 vm_flags &= ~VM_EXEC;
711         vm_flags |= mm->def_flags;
712         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
713
714         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
715                         vm_flags);
716         if (ret)
717                 goto out_unlock;
718         BUG_ON(prev != vma);
719
720         /* Move stack pages down in memory. */
721         if (stack_shift) {
722                 ret = shift_arg_pages(vma, stack_shift);
723                 if (ret)
724                         goto out_unlock;
725         }
726
727         /* mprotect_fixup is overkill to remove the temporary stack flags */
728         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
729
730         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
731         stack_size = vma->vm_end - vma->vm_start;
732         /*
733          * Align this down to a page boundary as expand_stack
734          * will align it up.
735          */
736         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
737 #ifdef CONFIG_STACK_GROWSUP
738         if (stack_size + stack_expand > rlim_stack)
739                 stack_base = vma->vm_start + rlim_stack;
740         else
741                 stack_base = vma->vm_end + stack_expand;
742 #else
743         if (stack_size + stack_expand > rlim_stack)
744                 stack_base = vma->vm_end - rlim_stack;
745         else
746                 stack_base = vma->vm_start - stack_expand;
747 #endif
748         current->mm->start_stack = bprm->p;
749         ret = expand_stack(vma, stack_base);
750         if (ret)
751                 ret = -EFAULT;
752
753 out_unlock:
754         up_write(&mm->mmap_sem);
755         return ret;
756 }
757 EXPORT_SYMBOL(setup_arg_pages);
758
759 #endif /* CONFIG_MMU */
760
761 struct file *open_exec(const char *name)
762 {
763         struct file *file;
764         int err;
765         static const struct open_flags open_exec_flags = {
766                 .open_flag = O_LARGEFILE | O_RDONLY | __FMODE_EXEC,
767                 .acc_mode = MAY_EXEC | MAY_OPEN,
768                 .intent = LOOKUP_OPEN
769         };
770
771         file = do_filp_open(AT_FDCWD, name, &open_exec_flags, LOOKUP_FOLLOW);
772         if (IS_ERR(file))
773                 goto out;
774
775         err = -EACCES;
776         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
777                 goto exit;
778
779         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
780                 goto exit;
781
782         fsnotify_open(file);
783
784         err = deny_write_access(file);
785         if (err)
786                 goto exit;
787
788 out:
789         return file;
790
791 exit:
792         fput(file);
793         return ERR_PTR(err);
794 }
795 EXPORT_SYMBOL(open_exec);
796
797 int kernel_read(struct file *file, loff_t offset,
798                 char *addr, unsigned long count)
799 {
800         mm_segment_t old_fs;
801         loff_t pos = offset;
802         int result;
803
804         old_fs = get_fs();
805         set_fs(get_ds());
806         /* The cast to a user pointer is valid due to the set_fs() */
807         result = vfs_read(file, (void __user *)addr, count, &pos);
808         set_fs(old_fs);
809         return result;
810 }
811
812 EXPORT_SYMBOL(kernel_read);
813
814 static int exec_mmap(struct mm_struct *mm)
815 {
816         struct task_struct *tsk;
817         struct mm_struct * old_mm, *active_mm;
818
819         /* Notify parent that we're no longer interested in the old VM */
820         tsk = current;
821         old_mm = current->mm;
822         mm_release(tsk, old_mm);
823
824         if (old_mm) {
825                 /*
826                  * Make sure that if there is a core dump in progress
827                  * for the old mm, we get out and die instead of going
828                  * through with the exec.  We must hold mmap_sem around
829                  * checking core_state and changing tsk->mm.
830                  */
831                 down_read(&old_mm->mmap_sem);
832                 if (unlikely(old_mm->core_state)) {
833                         up_read(&old_mm->mmap_sem);
834                         return -EINTR;
835                 }
836         }
837         task_lock(tsk);
838         active_mm = tsk->active_mm;
839         tsk->mm = mm;
840         tsk->active_mm = mm;
841         activate_mm(active_mm, mm);
842         task_unlock(tsk);
843         arch_pick_mmap_layout(mm);
844         if (old_mm) {
845                 up_read(&old_mm->mmap_sem);
846                 BUG_ON(active_mm != old_mm);
847                 setmax_mm_hiwater_rss(&tsk->signal->maxrss, old_mm);
848                 mm_update_next_owner(old_mm);
849                 mmput(old_mm);
850                 return 0;
851         }
852         mmdrop(active_mm);
853         return 0;
854 }
855
856 /*
857  * This function makes sure the current process has its own signal table,
858  * so that flush_signal_handlers can later reset the handlers without
859  * disturbing other processes.  (Other processes might share the signal
860  * table via the CLONE_SIGHAND option to clone().)
861  */
862 static int de_thread(struct task_struct *tsk)
863 {
864         struct signal_struct *sig = tsk->signal;
865         struct sighand_struct *oldsighand = tsk->sighand;
866         spinlock_t *lock = &oldsighand->siglock;
867
868         if (thread_group_empty(tsk))
869                 goto no_thread_group;
870
871         /*
872          * Kill all other threads in the thread group.
873          */
874         spin_lock_irq(lock);
875         if (signal_group_exit(sig)) {
876                 /*
877                  * Another group action in progress, just
878                  * return so that the signal is processed.
879                  */
880                 spin_unlock_irq(lock);
881                 return -EAGAIN;
882         }
883
884         sig->group_exit_task = tsk;
885         sig->notify_count = zap_other_threads(tsk);
886         if (!thread_group_leader(tsk))
887                 sig->notify_count--;
888
889         while (sig->notify_count) {
890                 __set_current_state(TASK_UNINTERRUPTIBLE);
891                 spin_unlock_irq(lock);
892                 schedule();
893                 spin_lock_irq(lock);
894         }
895         spin_unlock_irq(lock);
896
897         /*
898          * At this point all other threads have exited, all we have to
899          * do is to wait for the thread group leader to become inactive,
900          * and to assume its PID:
901          */
902         if (!thread_group_leader(tsk)) {
903                 struct task_struct *leader = tsk->group_leader;
904
905                 sig->notify_count = -1; /* for exit_notify() */
906                 for (;;) {
907                         write_lock_irq(&tasklist_lock);
908                         if (likely(leader->exit_state))
909                                 break;
910                         __set_current_state(TASK_UNINTERRUPTIBLE);
911                         write_unlock_irq(&tasklist_lock);
912                         schedule();
913                 }
914
915                 /*
916                  * The only record we have of the real-time age of a
917                  * process, regardless of execs it's done, is start_time.
918                  * All the past CPU time is accumulated in signal_struct
919                  * from sister threads now dead.  But in this non-leader
920                  * exec, nothing survives from the original leader thread,
921                  * whose birth marks the true age of this process now.
922                  * When we take on its identity by switching to its PID, we
923                  * also take its birthdate (always earlier than our own).
924                  */
925                 tsk->start_time = leader->start_time;
926
927                 BUG_ON(!same_thread_group(leader, tsk));
928                 BUG_ON(has_group_leader_pid(tsk));
929                 /*
930                  * An exec() starts a new thread group with the
931                  * TGID of the previous thread group. Rehash the
932                  * two threads with a switched PID, and release
933                  * the former thread group leader:
934                  */
935
936                 /* Become a process group leader with the old leader's pid.
937                  * The old leader becomes a thread of the this thread group.
938                  * Note: The old leader also uses this pid until release_task
939                  *       is called.  Odd but simple and correct.
940                  */
941                 detach_pid(tsk, PIDTYPE_PID);
942                 tsk->pid = leader->pid;
943                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
944                 transfer_pid(leader, tsk, PIDTYPE_PGID);
945                 transfer_pid(leader, tsk, PIDTYPE_SID);
946
947                 list_replace_rcu(&leader->tasks, &tsk->tasks);
948                 list_replace_init(&leader->sibling, &tsk->sibling);
949
950                 tsk->group_leader = tsk;
951                 leader->group_leader = tsk;
952
953                 tsk->exit_signal = SIGCHLD;
954                 leader->exit_signal = -1;
955
956                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
957                 leader->exit_state = EXIT_DEAD;
958
959                 /*
960                  * We are going to release_task()->ptrace_unlink() silently,
961                  * the tracer can sleep in do_wait(). EXIT_DEAD guarantees
962                  * the tracer wont't block again waiting for this thread.
963                  */
964                 if (unlikely(leader->ptrace))
965                         __wake_up_parent(leader, leader->parent);
966                 write_unlock_irq(&tasklist_lock);
967
968                 release_task(leader);
969         }
970
971         sig->group_exit_task = NULL;
972         sig->notify_count = 0;
973
974 no_thread_group:
975         /* we have changed execution domain */
976         tsk->exit_signal = SIGCHLD;
977
978         exit_itimers(sig);
979         flush_itimer_signals();
980
981         if (atomic_read(&oldsighand->count) != 1) {
982                 struct sighand_struct *newsighand;
983                 /*
984                  * This ->sighand is shared with the CLONE_SIGHAND
985                  * but not CLONE_THREAD task, switch to the new one.
986                  */
987                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
988                 if (!newsighand)
989                         return -ENOMEM;
990
991                 atomic_set(&newsighand->count, 1);
992                 memcpy(newsighand->action, oldsighand->action,
993                        sizeof(newsighand->action));
994
995                 write_lock_irq(&tasklist_lock);
996                 spin_lock(&oldsighand->siglock);
997                 rcu_assign_pointer(tsk->sighand, newsighand);
998                 spin_unlock(&oldsighand->siglock);
999                 write_unlock_irq(&tasklist_lock);
1000
1001                 __cleanup_sighand(oldsighand);
1002         }
1003
1004         BUG_ON(!thread_group_leader(tsk));
1005         return 0;
1006 }
1007
1008 /*
1009  * These functions flushes out all traces of the currently running executable
1010  * so that a new one can be started
1011  */
1012 static void flush_old_files(struct files_struct * files)
1013 {
1014         long j = -1;
1015         struct fdtable *fdt;
1016
1017         spin_lock(&files->file_lock);
1018         for (;;) {
1019                 unsigned long set, i;
1020
1021                 j++;
1022                 i = j * __NFDBITS;
1023                 fdt = files_fdtable(files);
1024                 if (i >= fdt->max_fds)
1025                         break;
1026                 set = fdt->close_on_exec[j];
1027                 if (!set)
1028                         continue;
1029                 fdt->close_on_exec[j] = 0;
1030                 spin_unlock(&files->file_lock);
1031                 for ( ; set ; i++,set >>= 1) {
1032                         if (set & 1) {
1033                                 sys_close(i);
1034                         }
1035                 }
1036                 spin_lock(&files->file_lock);
1037
1038         }
1039         spin_unlock(&files->file_lock);
1040 }
1041
1042 char *get_task_comm(char *buf, struct task_struct *tsk)
1043 {
1044         /* buf must be at least sizeof(tsk->comm) in size */
1045         task_lock(tsk);
1046         strncpy(buf, tsk->comm, sizeof(tsk->comm));
1047         task_unlock(tsk);
1048         return buf;
1049 }
1050 EXPORT_SYMBOL_GPL(get_task_comm);
1051
1052 void set_task_comm(struct task_struct *tsk, char *buf)
1053 {
1054         task_lock(tsk);
1055
1056         trace_task_rename(tsk, buf);
1057
1058         /*
1059          * Threads may access current->comm without holding
1060          * the task lock, so write the string carefully.
1061          * Readers without a lock may see incomplete new
1062          * names but are safe from non-terminating string reads.
1063          */
1064         memset(tsk->comm, 0, TASK_COMM_LEN);
1065         wmb();
1066         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1067         task_unlock(tsk);
1068         perf_event_comm(tsk);
1069 }
1070
1071 static void filename_to_taskname(char *tcomm, const char *fn, unsigned int len)
1072 {
1073         int i, ch;
1074
1075         /* Copies the binary name from after last slash */
1076         for (i = 0; (ch = *(fn++)) != '\0';) {
1077                 if (ch == '/')
1078                         i = 0; /* overwrite what we wrote */
1079                 else
1080                         if (i < len - 1)
1081                                 tcomm[i++] = ch;
1082         }
1083         tcomm[i] = '\0';
1084 }
1085
1086 int flush_old_exec(struct linux_binprm * bprm)
1087 {
1088         int retval;
1089
1090         /*
1091          * Make sure we have a private signal table and that
1092          * we are unassociated from the previous thread group.
1093          */
1094         retval = de_thread(current);
1095         if (retval)
1096                 goto out;
1097
1098         set_mm_exe_file(bprm->mm, bprm->file);
1099
1100         filename_to_taskname(bprm->tcomm, bprm->filename, sizeof(bprm->tcomm));
1101         /*
1102          * Release all of the old mmap stuff
1103          */
1104         acct_arg_size(bprm, 0);
1105         retval = exec_mmap(bprm->mm);
1106         if (retval)
1107                 goto out;
1108
1109         bprm->mm = NULL;                /* We're using it now */
1110
1111         set_fs(USER_DS);
1112         current->flags &= ~(PF_RANDOMIZE | PF_FORKNOEXEC | PF_KTHREAD);
1113         flush_thread();
1114         current->personality &= ~bprm->per_clear;
1115
1116         return 0;
1117
1118 out:
1119         return retval;
1120 }
1121 EXPORT_SYMBOL(flush_old_exec);
1122
1123 void would_dump(struct linux_binprm *bprm, struct file *file)
1124 {
1125         if (inode_permission(file->f_path.dentry->d_inode, MAY_READ) < 0)
1126                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
1127 }
1128 EXPORT_SYMBOL(would_dump);
1129
1130 void setup_new_exec(struct linux_binprm * bprm)
1131 {
1132         arch_pick_mmap_layout(current->mm);
1133
1134         /* This is the point of no return */
1135         current->sas_ss_sp = current->sas_ss_size = 0;
1136
1137         if (uid_eq(current_euid(), current_uid()) && gid_eq(current_egid(), current_gid()))
1138                 set_dumpable(current->mm, 1);
1139         else
1140                 set_dumpable(current->mm, suid_dumpable);
1141
1142         set_task_comm(current, bprm->tcomm);
1143
1144         /* Set the new mm task size. We have to do that late because it may
1145          * depend on TIF_32BIT which is only updated in flush_thread() on
1146          * some architectures like powerpc
1147          */
1148         current->mm->task_size = TASK_SIZE;
1149
1150         /* install the new credentials */
1151         if (!uid_eq(bprm->cred->uid, current_euid()) ||
1152             !gid_eq(bprm->cred->gid, current_egid())) {
1153                 current->pdeath_signal = 0;
1154         } else {
1155                 would_dump(bprm, bprm->file);
1156                 if (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)
1157                         set_dumpable(current->mm, suid_dumpable);
1158         }
1159
1160         /*
1161          * Flush performance counters when crossing a
1162          * security domain:
1163          */
1164         if (!get_dumpable(current->mm))
1165                 perf_event_exit_task(current);
1166
1167         /* An exec changes our domain. We are no longer part of the thread
1168            group */
1169
1170         current->self_exec_id++;
1171                         
1172         flush_signal_handlers(current, 0);
1173         flush_old_files(current->files);
1174 }
1175 EXPORT_SYMBOL(setup_new_exec);
1176
1177 /*
1178  * Prepare credentials and lock ->cred_guard_mutex.
1179  * install_exec_creds() commits the new creds and drops the lock.
1180  * Or, if exec fails before, free_bprm() should release ->cred and
1181  * and unlock.
1182  */
1183 int prepare_bprm_creds(struct linux_binprm *bprm)
1184 {
1185         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1186                 return -ERESTARTNOINTR;
1187
1188         bprm->cred = prepare_exec_creds();
1189         if (likely(bprm->cred))
1190                 return 0;
1191
1192         mutex_unlock(&current->signal->cred_guard_mutex);
1193         return -ENOMEM;
1194 }
1195
1196 void free_bprm(struct linux_binprm *bprm)
1197 {
1198         free_arg_pages(bprm);
1199         if (bprm->cred) {
1200                 mutex_unlock(&current->signal->cred_guard_mutex);
1201                 abort_creds(bprm->cred);
1202         }
1203         kfree(bprm);
1204 }
1205
1206 /*
1207  * install the new credentials for this executable
1208  */
1209 void install_exec_creds(struct linux_binprm *bprm)
1210 {
1211         security_bprm_committing_creds(bprm);
1212
1213         commit_creds(bprm->cred);
1214         bprm->cred = NULL;
1215         /*
1216          * cred_guard_mutex must be held at least to this point to prevent
1217          * ptrace_attach() from altering our determination of the task's
1218          * credentials; any time after this it may be unlocked.
1219          */
1220         security_bprm_committed_creds(bprm);
1221         mutex_unlock(&current->signal->cred_guard_mutex);
1222 }
1223 EXPORT_SYMBOL(install_exec_creds);
1224
1225 /*
1226  * determine how safe it is to execute the proposed program
1227  * - the caller must hold ->cred_guard_mutex to protect against
1228  *   PTRACE_ATTACH
1229  */
1230 static int check_unsafe_exec(struct linux_binprm *bprm)
1231 {
1232         struct task_struct *p = current, *t;
1233         unsigned n_fs;
1234         int res = 0;
1235
1236         if (p->ptrace) {
1237                 if (p->ptrace & PT_PTRACE_CAP)
1238                         bprm->unsafe |= LSM_UNSAFE_PTRACE_CAP;
1239                 else
1240                         bprm->unsafe |= LSM_UNSAFE_PTRACE;
1241         }
1242
1243         /*
1244          * This isn't strictly necessary, but it makes it harder for LSMs to
1245          * mess up.
1246          */
1247         if (current->no_new_privs)
1248                 bprm->unsafe |= LSM_UNSAFE_NO_NEW_PRIVS;
1249
1250         n_fs = 1;
1251         spin_lock(&p->fs->lock);
1252         rcu_read_lock();
1253         for (t = next_thread(p); t != p; t = next_thread(t)) {
1254                 if (t->fs == p->fs)
1255                         n_fs++;
1256         }
1257         rcu_read_unlock();
1258
1259         if (p->fs->users > n_fs) {
1260                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1261         } else {
1262                 res = -EAGAIN;
1263                 if (!p->fs->in_exec) {
1264                         p->fs->in_exec = 1;
1265                         res = 1;
1266                 }
1267         }
1268         spin_unlock(&p->fs->lock);
1269
1270         return res;
1271 }
1272
1273 /* 
1274  * Fill the binprm structure from the inode. 
1275  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1276  *
1277  * This may be called multiple times for binary chains (scripts for example).
1278  */
1279 int prepare_binprm(struct linux_binprm *bprm)
1280 {
1281         umode_t mode;
1282         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1283         int retval;
1284
1285         mode = inode->i_mode;
1286         if (bprm->file->f_op == NULL)
1287                 return -EACCES;
1288
1289         /* clear any previous set[ug]id data from a previous binary */
1290         bprm->cred->euid = current_euid();
1291         bprm->cred->egid = current_egid();
1292
1293         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) &&
1294             !current->no_new_privs) {
1295                 /* Set-uid? */
1296                 if (mode & S_ISUID) {
1297                         if (!kuid_has_mapping(bprm->cred->user_ns, inode->i_uid))
1298                                 return -EPERM;
1299                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1300                         bprm->cred->euid = inode->i_uid;
1301
1302                 }
1303
1304                 /* Set-gid? */
1305                 /*
1306                  * If setgid is set but no group execute bit then this
1307                  * is a candidate for mandatory locking, not a setgid
1308                  * executable.
1309                  */
1310                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1311                         if (!kgid_has_mapping(bprm->cred->user_ns, inode->i_gid))
1312                                 return -EPERM;
1313                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1314                         bprm->cred->egid = inode->i_gid;
1315                 }
1316         }
1317
1318         /* fill in binprm security blob */
1319         retval = security_bprm_set_creds(bprm);
1320         if (retval)
1321                 return retval;
1322         bprm->cred_prepared = 1;
1323
1324         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1325         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1326 }
1327
1328 EXPORT_SYMBOL(prepare_binprm);
1329
1330 /*
1331  * Arguments are '\0' separated strings found at the location bprm->p
1332  * points to; chop off the first by relocating brpm->p to right after
1333  * the first '\0' encountered.
1334  */
1335 int remove_arg_zero(struct linux_binprm *bprm)
1336 {
1337         int ret = 0;
1338         unsigned long offset;
1339         char *kaddr;
1340         struct page *page;
1341
1342         if (!bprm->argc)
1343                 return 0;
1344
1345         do {
1346                 offset = bprm->p & ~PAGE_MASK;
1347                 page = get_arg_page(bprm, bprm->p, 0);
1348                 if (!page) {
1349                         ret = -EFAULT;
1350                         goto out;
1351                 }
1352                 kaddr = kmap_atomic(page);
1353
1354                 for (; offset < PAGE_SIZE && kaddr[offset];
1355                                 offset++, bprm->p++)
1356                         ;
1357
1358                 kunmap_atomic(kaddr);
1359                 put_arg_page(page);
1360
1361                 if (offset == PAGE_SIZE)
1362                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1363         } while (offset == PAGE_SIZE);
1364
1365         bprm->p++;
1366         bprm->argc--;
1367         ret = 0;
1368
1369 out:
1370         return ret;
1371 }
1372 EXPORT_SYMBOL(remove_arg_zero);
1373
1374 /*
1375  * cycle the list of binary formats handler, until one recognizes the image
1376  */
1377 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1378 {
1379         unsigned int depth = bprm->recursion_depth;
1380         int try,retval;
1381         struct linux_binfmt *fmt;
1382         pid_t old_pid, old_vpid;
1383
1384         retval = security_bprm_check(bprm);
1385         if (retval)
1386                 return retval;
1387
1388         retval = audit_bprm(bprm);
1389         if (retval)
1390                 return retval;
1391
1392         /* Need to fetch pid before load_binary changes it */
1393         old_pid = current->pid;
1394         rcu_read_lock();
1395         old_vpid = task_pid_nr_ns(current, task_active_pid_ns(current->parent));
1396         rcu_read_unlock();
1397
1398         retval = -ENOENT;
1399         for (try=0; try<2; try++) {
1400                 read_lock(&binfmt_lock);
1401                 list_for_each_entry(fmt, &formats, lh) {
1402                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1403                         if (!fn)
1404                                 continue;
1405                         if (!try_module_get(fmt->module))
1406                                 continue;
1407                         read_unlock(&binfmt_lock);
1408                         retval = fn(bprm, regs);
1409                         /*
1410                          * Restore the depth counter to its starting value
1411                          * in this call, so we don't have to rely on every
1412                          * load_binary function to restore it on return.
1413                          */
1414                         bprm->recursion_depth = depth;
1415                         if (retval >= 0) {
1416                                 if (depth == 0) {
1417                                         trace_sched_process_exec(current, old_pid, bprm);
1418                                         ptrace_event(PTRACE_EVENT_EXEC, old_vpid);
1419                                 }
1420                                 put_binfmt(fmt);
1421                                 allow_write_access(bprm->file);
1422                                 if (bprm->file)
1423                                         fput(bprm->file);
1424                                 bprm->file = NULL;
1425                                 current->did_exec = 1;
1426                                 proc_exec_connector(current);
1427                                 return retval;
1428                         }
1429                         read_lock(&binfmt_lock);
1430                         put_binfmt(fmt);
1431                         if (retval != -ENOEXEC || bprm->mm == NULL)
1432                                 break;
1433                         if (!bprm->file) {
1434                                 read_unlock(&binfmt_lock);
1435                                 return retval;
1436                         }
1437                 }
1438                 read_unlock(&binfmt_lock);
1439 #ifdef CONFIG_MODULES
1440                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1441                         break;
1442                 } else {
1443 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1444                         if (printable(bprm->buf[0]) &&
1445                             printable(bprm->buf[1]) &&
1446                             printable(bprm->buf[2]) &&
1447                             printable(bprm->buf[3]))
1448                                 break; /* -ENOEXEC */
1449                         if (try)
1450                                 break; /* -ENOEXEC */
1451                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1452                 }
1453 #else
1454                 break;
1455 #endif
1456         }
1457         return retval;
1458 }
1459
1460 EXPORT_SYMBOL(search_binary_handler);
1461
1462 /*
1463  * sys_execve() executes a new program.
1464  */
1465 static int do_execve_common(const char *filename,
1466                                 struct user_arg_ptr argv,
1467                                 struct user_arg_ptr envp,
1468                                 struct pt_regs *regs)
1469 {
1470         struct linux_binprm *bprm;
1471         struct file *file;
1472         struct files_struct *displaced;
1473         bool clear_in_exec;
1474         int retval;
1475         const struct cred *cred = current_cred();
1476
1477         /*
1478          * We move the actual failure in case of RLIMIT_NPROC excess from
1479          * set*uid() to execve() because too many poorly written programs
1480          * don't check setuid() return code.  Here we additionally recheck
1481          * whether NPROC limit is still exceeded.
1482          */
1483         if ((current->flags & PF_NPROC_EXCEEDED) &&
1484             atomic_read(&cred->user->processes) > rlimit(RLIMIT_NPROC)) {
1485                 retval = -EAGAIN;
1486                 goto out_ret;
1487         }
1488
1489         /* We're below the limit (still or again), so we don't want to make
1490          * further execve() calls fail. */
1491         current->flags &= ~PF_NPROC_EXCEEDED;
1492
1493         retval = unshare_files(&displaced);
1494         if (retval)
1495                 goto out_ret;
1496
1497         retval = -ENOMEM;
1498         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1499         if (!bprm)
1500                 goto out_files;
1501
1502         retval = prepare_bprm_creds(bprm);
1503         if (retval)
1504                 goto out_free;
1505
1506         retval = check_unsafe_exec(bprm);
1507         if (retval < 0)
1508                 goto out_free;
1509         clear_in_exec = retval;
1510         current->in_execve = 1;
1511
1512         file = open_exec(filename);
1513         retval = PTR_ERR(file);
1514         if (IS_ERR(file))
1515                 goto out_unmark;
1516
1517         sched_exec();
1518
1519         bprm->file = file;
1520         bprm->filename = filename;
1521         bprm->interp = filename;
1522
1523         retval = bprm_mm_init(bprm);
1524         if (retval)
1525                 goto out_file;
1526
1527         bprm->argc = count(argv, MAX_ARG_STRINGS);
1528         if ((retval = bprm->argc) < 0)
1529                 goto out;
1530
1531         bprm->envc = count(envp, MAX_ARG_STRINGS);
1532         if ((retval = bprm->envc) < 0)
1533                 goto out;
1534
1535         retval = prepare_binprm(bprm);
1536         if (retval < 0)
1537                 goto out;
1538
1539         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1540         if (retval < 0)
1541                 goto out;
1542
1543         bprm->exec = bprm->p;
1544         retval = copy_strings(bprm->envc, envp, bprm);
1545         if (retval < 0)
1546                 goto out;
1547
1548         retval = copy_strings(bprm->argc, argv, bprm);
1549         if (retval < 0)
1550                 goto out;
1551
1552         retval = search_binary_handler(bprm,regs);
1553         if (retval < 0)
1554                 goto out;
1555
1556         /* execve succeeded */
1557         current->fs->in_exec = 0;
1558         current->in_execve = 0;
1559         acct_update_integrals(current);
1560         free_bprm(bprm);
1561         if (displaced)
1562                 put_files_struct(displaced);
1563         return retval;
1564
1565 out:
1566         if (bprm->mm) {
1567                 acct_arg_size(bprm, 0);
1568                 mmput(bprm->mm);
1569         }
1570
1571 out_file:
1572         if (bprm->file) {
1573                 allow_write_access(bprm->file);
1574                 fput(bprm->file);
1575         }
1576
1577 out_unmark:
1578         if (clear_in_exec)
1579                 current->fs->in_exec = 0;
1580         current->in_execve = 0;
1581
1582 out_free:
1583         free_bprm(bprm);
1584
1585 out_files:
1586         if (displaced)
1587                 reset_files_struct(displaced);
1588 out_ret:
1589         return retval;
1590 }
1591
1592 int do_execve(const char *filename,
1593         const char __user *const __user *__argv,
1594         const char __user *const __user *__envp,
1595         struct pt_regs *regs)
1596 {
1597         struct user_arg_ptr argv = { .ptr.native = __argv };
1598         struct user_arg_ptr envp = { .ptr.native = __envp };
1599         return do_execve_common(filename, argv, envp, regs);
1600 }
1601
1602 #ifdef CONFIG_COMPAT
1603 int compat_do_execve(char *filename,
1604         compat_uptr_t __user *__argv,
1605         compat_uptr_t __user *__envp,
1606         struct pt_regs *regs)
1607 {
1608         struct user_arg_ptr argv = {
1609                 .is_compat = true,
1610                 .ptr.compat = __argv,
1611         };
1612         struct user_arg_ptr envp = {
1613                 .is_compat = true,
1614                 .ptr.compat = __envp,
1615         };
1616         return do_execve_common(filename, argv, envp, regs);
1617 }
1618 #endif
1619
1620 void set_binfmt(struct linux_binfmt *new)
1621 {
1622         struct mm_struct *mm = current->mm;
1623
1624         if (mm->binfmt)
1625                 module_put(mm->binfmt->module);
1626
1627         mm->binfmt = new;
1628         if (new)
1629                 __module_get(new->module);
1630 }
1631
1632 EXPORT_SYMBOL(set_binfmt);
1633
1634 static int expand_corename(struct core_name *cn)
1635 {
1636         char *old_corename = cn->corename;
1637
1638         cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1639         cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1640
1641         if (!cn->corename) {
1642                 kfree(old_corename);
1643                 return -ENOMEM;
1644         }
1645
1646         return 0;
1647 }
1648
1649 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1650 {
1651         char *cur;
1652         int need;
1653         int ret;
1654         va_list arg;
1655
1656         va_start(arg, fmt);
1657         need = vsnprintf(NULL, 0, fmt, arg);
1658         va_end(arg);
1659
1660         if (likely(need < cn->size - cn->used - 1))
1661                 goto out_printf;
1662
1663         ret = expand_corename(cn);
1664         if (ret)
1665                 goto expand_fail;
1666
1667 out_printf:
1668         cur = cn->corename + cn->used;
1669         va_start(arg, fmt);
1670         vsnprintf(cur, need + 1, fmt, arg);
1671         va_end(arg);
1672         cn->used += need;
1673         return 0;
1674
1675 expand_fail:
1676         return ret;
1677 }
1678
1679 static void cn_escape(char *str)
1680 {
1681         for (; *str; str++)
1682                 if (*str == '/')
1683                         *str = '!';
1684 }
1685
1686 static int cn_print_exe_file(struct core_name *cn)
1687 {
1688         struct file *exe_file;
1689         char *pathbuf, *path;
1690         int ret;
1691
1692         exe_file = get_mm_exe_file(current->mm);
1693         if (!exe_file) {
1694                 char *commstart = cn->corename + cn->used;
1695                 ret = cn_printf(cn, "%s (path unknown)", current->comm);
1696                 cn_escape(commstart);
1697                 return ret;
1698         }
1699
1700         pathbuf = kmalloc(PATH_MAX, GFP_TEMPORARY);
1701         if (!pathbuf) {
1702                 ret = -ENOMEM;
1703                 goto put_exe_file;
1704         }
1705
1706         path = d_path(&exe_file->f_path, pathbuf, PATH_MAX);
1707         if (IS_ERR(path)) {
1708                 ret = PTR_ERR(path);
1709                 goto free_buf;
1710         }
1711
1712         cn_escape(path);
1713
1714         ret = cn_printf(cn, "%s", path);
1715
1716 free_buf:
1717         kfree(pathbuf);
1718 put_exe_file:
1719         fput(exe_file);
1720         return ret;
1721 }
1722
1723 /* format_corename will inspect the pattern parameter, and output a
1724  * name into corename, which must have space for at least
1725  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1726  */
1727 static int format_corename(struct core_name *cn, long signr)
1728 {
1729         const struct cred *cred = current_cred();
1730         const char *pat_ptr = core_pattern;
1731         int ispipe = (*pat_ptr == '|');
1732         int pid_in_pattern = 0;
1733         int err = 0;
1734
1735         cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1736         cn->corename = kmalloc(cn->size, GFP_KERNEL);
1737         cn->used = 0;
1738
1739         if (!cn->corename)
1740                 return -ENOMEM;
1741
1742         /* Repeat as long as we have more pattern to process and more output
1743            space */
1744         while (*pat_ptr) {
1745                 if (*pat_ptr != '%') {
1746                         if (*pat_ptr == 0)
1747                                 goto out;
1748                         err = cn_printf(cn, "%c", *pat_ptr++);
1749                 } else {
1750                         switch (*++pat_ptr) {
1751                         /* single % at the end, drop that */
1752                         case 0:
1753                                 goto out;
1754                         /* Double percent, output one percent */
1755                         case '%':
1756                                 err = cn_printf(cn, "%c", '%');
1757                                 break;
1758                         /* pid */
1759                         case 'p':
1760                                 pid_in_pattern = 1;
1761                                 err = cn_printf(cn, "%d",
1762                                               task_tgid_vnr(current));
1763                                 break;
1764                         /* uid */
1765                         case 'u':
1766                                 err = cn_printf(cn, "%d", cred->uid);
1767                                 break;
1768                         /* gid */
1769                         case 'g':
1770                                 err = cn_printf(cn, "%d", cred->gid);
1771                                 break;
1772                         /* signal that caused the coredump */
1773                         case 's':
1774                                 err = cn_printf(cn, "%ld", signr);
1775                                 break;
1776                         /* UNIX time of coredump */
1777                         case 't': {
1778                                 struct timeval tv;
1779                                 do_gettimeofday(&tv);
1780                                 err = cn_printf(cn, "%lu", tv.tv_sec);
1781                                 break;
1782                         }
1783                         /* hostname */
1784                         case 'h': {
1785                                 char *namestart = cn->corename + cn->used;
1786                                 down_read(&uts_sem);
1787                                 err = cn_printf(cn, "%s",
1788                                               utsname()->nodename);
1789                                 up_read(&uts_sem);
1790                                 cn_escape(namestart);
1791                                 break;
1792                         }
1793                         /* executable */
1794                         case 'e': {
1795                                 char *commstart = cn->corename + cn->used;
1796                                 err = cn_printf(cn, "%s", current->comm);
1797                                 cn_escape(commstart);
1798                                 break;
1799                         }
1800                         case 'E':
1801                                 err = cn_print_exe_file(cn);
1802                                 break;
1803                         /* core limit size */
1804                         case 'c':
1805                                 err = cn_printf(cn, "%lu",
1806                                               rlimit(RLIMIT_CORE));
1807                                 break;
1808                         default:
1809                                 break;
1810                         }
1811                         ++pat_ptr;
1812                 }
1813
1814                 if (err)
1815                         return err;
1816         }
1817
1818         /* Backward compatibility with core_uses_pid:
1819          *
1820          * If core_pattern does not include a %p (as is the default)
1821          * and core_uses_pid is set, then .%pid will be appended to
1822          * the filename. Do not do this for piped commands. */
1823         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1824                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1825                 if (err)
1826                         return err;
1827         }
1828 out:
1829         return ispipe;
1830 }
1831
1832 static int zap_process(struct task_struct *start, int exit_code)
1833 {
1834         struct task_struct *t;
1835         int nr = 0;
1836
1837         start->signal->flags = SIGNAL_GROUP_EXIT;
1838         start->signal->group_exit_code = exit_code;
1839         start->signal->group_stop_count = 0;
1840
1841         t = start;
1842         do {
1843                 task_clear_jobctl_pending(t, JOBCTL_PENDING_MASK);
1844                 if (t != current && t->mm) {
1845                         sigaddset(&t->pending.signal, SIGKILL);
1846                         signal_wake_up(t, 1);
1847                         nr++;
1848                 }
1849         } while_each_thread(start, t);
1850
1851         return nr;
1852 }
1853
1854 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1855                                 struct core_state *core_state, int exit_code)
1856 {
1857         struct task_struct *g, *p;
1858         unsigned long flags;
1859         int nr = -EAGAIN;
1860
1861         spin_lock_irq(&tsk->sighand->siglock);
1862         if (!signal_group_exit(tsk->signal)) {
1863                 mm->core_state = core_state;
1864                 nr = zap_process(tsk, exit_code);
1865         }
1866         spin_unlock_irq(&tsk->sighand->siglock);
1867         if (unlikely(nr < 0))
1868                 return nr;
1869
1870         if (atomic_read(&mm->mm_users) == nr + 1)
1871                 goto done;
1872         /*
1873          * We should find and kill all tasks which use this mm, and we should
1874          * count them correctly into ->nr_threads. We don't take tasklist
1875          * lock, but this is safe wrt:
1876          *
1877          * fork:
1878          *      None of sub-threads can fork after zap_process(leader). All
1879          *      processes which were created before this point should be
1880          *      visible to zap_threads() because copy_process() adds the new
1881          *      process to the tail of init_task.tasks list, and lock/unlock
1882          *      of ->siglock provides a memory barrier.
1883          *
1884          * do_exit:
1885          *      The caller holds mm->mmap_sem. This means that the task which
1886          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1887          *      its ->mm.
1888          *
1889          * de_thread:
1890          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1891          *      we must see either old or new leader, this does not matter.
1892          *      However, it can change p->sighand, so lock_task_sighand(p)
1893          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1894          *      it can't fail.
1895          *
1896          *      Note also that "g" can be the old leader with ->mm == NULL
1897          *      and already unhashed and thus removed from ->thread_group.
1898          *      This is OK, __unhash_process()->list_del_rcu() does not
1899          *      clear the ->next pointer, we will find the new leader via
1900          *      next_thread().
1901          */
1902         rcu_read_lock();
1903         for_each_process(g) {
1904                 if (g == tsk->group_leader)
1905                         continue;
1906                 if (g->flags & PF_KTHREAD)
1907                         continue;
1908                 p = g;
1909                 do {
1910                         if (p->mm) {
1911                                 if (unlikely(p->mm == mm)) {
1912                                         lock_task_sighand(p, &flags);
1913                                         nr += zap_process(p, exit_code);
1914                                         unlock_task_sighand(p, &flags);
1915                                 }
1916                                 break;
1917                         }
1918                 } while_each_thread(g, p);
1919         }
1920         rcu_read_unlock();
1921 done:
1922         atomic_set(&core_state->nr_threads, nr);
1923         return nr;
1924 }
1925
1926 static int coredump_wait(int exit_code, struct core_state *core_state)
1927 {
1928         struct task_struct *tsk = current;
1929         struct mm_struct *mm = tsk->mm;
1930         int core_waiters = -EBUSY;
1931
1932         init_completion(&core_state->startup);
1933         core_state->dumper.task = tsk;
1934         core_state->dumper.next = NULL;
1935
1936         down_write(&mm->mmap_sem);
1937         if (!mm->core_state)
1938                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1939         up_write(&mm->mmap_sem);
1940
1941         if (core_waiters > 0) {
1942                 struct core_thread *ptr;
1943
1944                 wait_for_completion(&core_state->startup);
1945                 /*
1946                  * Wait for all the threads to become inactive, so that
1947                  * all the thread context (extended register state, like
1948                  * fpu etc) gets copied to the memory.
1949                  */
1950                 ptr = core_state->dumper.next;
1951                 while (ptr != NULL) {
1952                         wait_task_inactive(ptr->task, 0);
1953                         ptr = ptr->next;
1954                 }
1955         }
1956
1957         return core_waiters;
1958 }
1959
1960 static void coredump_finish(struct mm_struct *mm)
1961 {
1962         struct core_thread *curr, *next;
1963         struct task_struct *task;
1964
1965         next = mm->core_state->dumper.next;
1966         while ((curr = next) != NULL) {
1967                 next = curr->next;
1968                 task = curr->task;
1969                 /*
1970                  * see exit_mm(), curr->task must not see
1971                  * ->task == NULL before we read ->next.
1972                  */
1973                 smp_mb();
1974                 curr->task = NULL;
1975                 wake_up_process(task);
1976         }
1977
1978         mm->core_state = NULL;
1979 }
1980
1981 /*
1982  * set_dumpable converts traditional three-value dumpable to two flags and
1983  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1984  * these bits are not changed atomically.  So get_dumpable can observe the
1985  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1986  * return either old dumpable or new one by paying attention to the order of
1987  * modifying the bits.
1988  *
1989  * dumpable |   mm->flags (binary)
1990  * old  new | initial interim  final
1991  * ---------+-----------------------
1992  *  0    1  |   00      01      01
1993  *  0    2  |   00      10(*)   11
1994  *  1    0  |   01      00      00
1995  *  1    2  |   01      11      11
1996  *  2    0  |   11      10(*)   00
1997  *  2    1  |   11      11      01
1998  *
1999  * (*) get_dumpable regards interim value of 10 as 11.
2000  */
2001 void set_dumpable(struct mm_struct *mm, int value)
2002 {
2003         switch (value) {
2004         case 0:
2005                 clear_bit(MMF_DUMPABLE, &mm->flags);
2006                 smp_wmb();
2007                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2008                 break;
2009         case 1:
2010                 set_bit(MMF_DUMPABLE, &mm->flags);
2011                 smp_wmb();
2012                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
2013                 break;
2014         case 2:
2015                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
2016                 smp_wmb();
2017                 set_bit(MMF_DUMPABLE, &mm->flags);
2018                 break;
2019         }
2020 }
2021
2022 static int __get_dumpable(unsigned long mm_flags)
2023 {
2024         int ret;
2025
2026         ret = mm_flags & MMF_DUMPABLE_MASK;
2027         return (ret >= 2) ? 2 : ret;
2028 }
2029
2030 int get_dumpable(struct mm_struct *mm)
2031 {
2032         return __get_dumpable(mm->flags);
2033 }
2034
2035 static void wait_for_dump_helpers(struct file *file)
2036 {
2037         struct pipe_inode_info *pipe;
2038
2039         pipe = file->f_path.dentry->d_inode->i_pipe;
2040
2041         pipe_lock(pipe);
2042         pipe->readers++;
2043         pipe->writers--;
2044
2045         while ((pipe->readers > 1) && (!signal_pending(current))) {
2046                 wake_up_interruptible_sync(&pipe->wait);
2047                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
2048                 pipe_wait(pipe);
2049         }
2050
2051         pipe->readers--;
2052         pipe->writers++;
2053         pipe_unlock(pipe);
2054
2055 }
2056
2057
2058 /*
2059  * umh_pipe_setup
2060  * helper function to customize the process used
2061  * to collect the core in userspace.  Specifically
2062  * it sets up a pipe and installs it as fd 0 (stdin)
2063  * for the process.  Returns 0 on success, or
2064  * PTR_ERR on failure.
2065  * Note that it also sets the core limit to 1.  This
2066  * is a special value that we use to trap recursive
2067  * core dumps
2068  */
2069 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
2070 {
2071         struct file *rp, *wp;
2072         struct fdtable *fdt;
2073         struct coredump_params *cp = (struct coredump_params *)info->data;
2074         struct files_struct *cf = current->files;
2075
2076         wp = create_write_pipe(0);
2077         if (IS_ERR(wp))
2078                 return PTR_ERR(wp);
2079
2080         rp = create_read_pipe(wp, 0);
2081         if (IS_ERR(rp)) {
2082                 free_write_pipe(wp);
2083                 return PTR_ERR(rp);
2084         }
2085
2086         cp->file = wp;
2087
2088         sys_close(0);
2089         fd_install(0, rp);
2090         spin_lock(&cf->file_lock);
2091         fdt = files_fdtable(cf);
2092         __set_open_fd(0, fdt);
2093         __clear_close_on_exec(0, fdt);
2094         spin_unlock(&cf->file_lock);
2095
2096         /* and disallow core files too */
2097         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
2098
2099         return 0;
2100 }
2101
2102 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
2103 {
2104         struct core_state core_state;
2105         struct core_name cn;
2106         struct mm_struct *mm = current->mm;
2107         struct linux_binfmt * binfmt;
2108         const struct cred *old_cred;
2109         struct cred *cred;
2110         int retval = 0;
2111         int flag = 0;
2112         int ispipe;
2113         static atomic_t core_dump_count = ATOMIC_INIT(0);
2114         struct coredump_params cprm = {
2115                 .signr = signr,
2116                 .regs = regs,
2117                 .limit = rlimit(RLIMIT_CORE),
2118                 /*
2119                  * We must use the same mm->flags while dumping core to avoid
2120                  * inconsistency of bit flags, since this flag is not protected
2121                  * by any locks.
2122                  */
2123                 .mm_flags = mm->flags,
2124         };
2125
2126         audit_core_dumps(signr);
2127
2128         binfmt = mm->binfmt;
2129         if (!binfmt || !binfmt->core_dump)
2130                 goto fail;
2131         if (!__get_dumpable(cprm.mm_flags))
2132                 goto fail;
2133
2134         cred = prepare_creds();
2135         if (!cred)
2136                 goto fail;
2137         /*
2138          *      We cannot trust fsuid as being the "true" uid of the
2139          *      process nor do we know its entire history. We only know it
2140          *      was tainted so we dump it as root in mode 2.
2141          */
2142         if (__get_dumpable(cprm.mm_flags) == 2) {
2143                 /* Setuid core dump mode */
2144                 flag = O_EXCL;          /* Stop rewrite attacks */
2145                 cred->fsuid = GLOBAL_ROOT_UID;  /* Dump root private */
2146         }
2147
2148         retval = coredump_wait(exit_code, &core_state);
2149         if (retval < 0)
2150                 goto fail_creds;
2151
2152         old_cred = override_creds(cred);
2153
2154         /*
2155          * Clear any false indication of pending signals that might
2156          * be seen by the filesystem code called to write the core file.
2157          */
2158         clear_thread_flag(TIF_SIGPENDING);
2159
2160         ispipe = format_corename(&cn, signr);
2161
2162         if (ispipe) {
2163                 int dump_count;
2164                 char **helper_argv;
2165
2166                 if (ispipe < 0) {
2167                         printk(KERN_WARNING "format_corename failed\n");
2168                         printk(KERN_WARNING "Aborting core\n");
2169                         goto fail_corename;
2170                 }
2171
2172                 if (cprm.limit == 1) {
2173                         /*
2174                          * Normally core limits are irrelevant to pipes, since
2175                          * we're not writing to the file system, but we use
2176                          * cprm.limit of 1 here as a speacial value. Any
2177                          * non-1 limit gets set to RLIM_INFINITY below, but
2178                          * a limit of 0 skips the dump.  This is a consistent
2179                          * way to catch recursive crashes.  We can still crash
2180                          * if the core_pattern binary sets RLIM_CORE =  !1
2181                          * but it runs as root, and can do lots of stupid things
2182                          * Note that we use task_tgid_vnr here to grab the pid
2183                          * of the process group leader.  That way we get the
2184                          * right pid if a thread in a multi-threaded
2185                          * core_pattern process dies.
2186                          */
2187                         printk(KERN_WARNING
2188                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
2189                                 task_tgid_vnr(current), current->comm);
2190                         printk(KERN_WARNING "Aborting core\n");
2191                         goto fail_unlock;
2192                 }
2193                 cprm.limit = RLIM_INFINITY;
2194
2195                 dump_count = atomic_inc_return(&core_dump_count);
2196                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2197                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2198                                task_tgid_vnr(current), current->comm);
2199                         printk(KERN_WARNING "Skipping core dump\n");
2200                         goto fail_dropcount;
2201                 }
2202
2203                 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2204                 if (!helper_argv) {
2205                         printk(KERN_WARNING "%s failed to allocate memory\n",
2206                                __func__);
2207                         goto fail_dropcount;
2208                 }
2209
2210                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2211                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2212                                         NULL, &cprm);
2213                 argv_free(helper_argv);
2214                 if (retval) {
2215                         printk(KERN_INFO "Core dump to %s pipe failed\n",
2216                                cn.corename);
2217                         goto close_fail;
2218                 }
2219         } else {
2220                 struct inode *inode;
2221
2222                 if (cprm.limit < binfmt->min_coredump)
2223                         goto fail_unlock;
2224
2225                 cprm.file = filp_open(cn.corename,
2226                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2227                                  0600);
2228                 if (IS_ERR(cprm.file))
2229                         goto fail_unlock;
2230
2231                 inode = cprm.file->f_path.dentry->d_inode;
2232                 if (inode->i_nlink > 1)
2233                         goto close_fail;
2234                 if (d_unhashed(cprm.file->f_path.dentry))
2235                         goto close_fail;
2236                 /*
2237                  * AK: actually i see no reason to not allow this for named
2238                  * pipes etc, but keep the previous behaviour for now.
2239                  */
2240                 if (!S_ISREG(inode->i_mode))
2241                         goto close_fail;
2242                 /*
2243                  * Dont allow local users get cute and trick others to coredump
2244                  * into their pre-created files.
2245                  */
2246                 if (!uid_eq(inode->i_uid, current_fsuid()))
2247                         goto close_fail;
2248                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2249                         goto close_fail;
2250                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2251                         goto close_fail;
2252         }
2253
2254         retval = binfmt->core_dump(&cprm);
2255         if (retval)
2256                 current->signal->group_exit_code |= 0x80;
2257
2258         if (ispipe && core_pipe_limit)
2259                 wait_for_dump_helpers(cprm.file);
2260 close_fail:
2261         if (cprm.file)
2262                 filp_close(cprm.file, NULL);
2263 fail_dropcount:
2264         if (ispipe)
2265                 atomic_dec(&core_dump_count);
2266 fail_unlock:
2267         kfree(cn.corename);
2268 fail_corename:
2269         coredump_finish(mm);
2270         revert_creds(old_cred);
2271 fail_creds:
2272         put_cred(cred);
2273 fail:
2274         return;
2275 }
2276
2277 /*
2278  * Core dumping helper functions.  These are the only things you should
2279  * do on a core-file: use only these functions to write out all the
2280  * necessary info.
2281  */
2282 int dump_write(struct file *file, const void *addr, int nr)
2283 {
2284         return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2285 }
2286 EXPORT_SYMBOL(dump_write);
2287
2288 int dump_seek(struct file *file, loff_t off)
2289 {
2290         int ret = 1;
2291
2292         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2293                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2294                         return 0;
2295         } else {
2296                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2297
2298                 if (!buf)
2299                         return 0;
2300                 while (off > 0) {
2301                         unsigned long n = off;
2302
2303                         if (n > PAGE_SIZE)
2304                                 n = PAGE_SIZE;
2305                         if (!dump_write(file, buf, n)) {
2306                                 ret = 0;
2307                                 break;
2308                         }
2309                         off -= n;
2310                 }
2311                 free_page((unsigned long)buf);
2312         }
2313         return ret;
2314 }
2315 EXPORT_SYMBOL(dump_seek);