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