exec: copy-and-paste the fixes into compat_do_execve() paths
[linux-2.6.git] / fs / exec.c
1 /*
2  *  linux/fs/exec.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  * #!-checking implemented by tytso.
9  */
10 /*
11  * Demand-loading implemented 01.12.91 - no need to read anything but
12  * the header into memory. The inode of the executable is put into
13  * "current->executable", and page faults do the actual loading. Clean.
14  *
15  * Once more I can proudly say that linux stood up to being changed: it
16  * was less than 2 hours work to get demand-loading completely implemented.
17  *
18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
19  * current->executable is only used by the procfs.  This allows a dispatch
20  * table to check for several different types  of binary formats.  We keep
21  * trying until we recognize the file or we run out of supported binary
22  * formats. 
23  */
24
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
28 #include <linux/mm.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/swap.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.h>
35 #include <linux/perf_event.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52 #include <linux/tracehook.h>
53 #include <linux/kmod.h>
54 #include <linux/fsnotify.h>
55 #include <linux/fs_struct.h>
56 #include <linux/pipe_fs_i.h>
57 #include <linux/oom.h>
58
59 #include <asm/uaccess.h>
60 #include <asm/mmu_context.h>
61 #include <asm/tlb.h>
62 #include "internal.h"
63
64 int core_uses_pid;
65 char core_pattern[CORENAME_MAX_SIZE] = "core";
66 unsigned int core_pipe_limit;
67 int suid_dumpable = 0;
68
69 struct core_name {
70         char *corename;
71         int used, size;
72 };
73 static atomic_t call_count = ATOMIC_INIT(1);
74
75 /* The maximal length of core_pattern is also specified in sysctl.c */
76
77 static LIST_HEAD(formats);
78 static DEFINE_RWLOCK(binfmt_lock);
79
80 int __register_binfmt(struct linux_binfmt * fmt, int insert)
81 {
82         if (!fmt)
83                 return -EINVAL;
84         write_lock(&binfmt_lock);
85         insert ? list_add(&fmt->lh, &formats) :
86                  list_add_tail(&fmt->lh, &formats);
87         write_unlock(&binfmt_lock);
88         return 0;       
89 }
90
91 EXPORT_SYMBOL(__register_binfmt);
92
93 void unregister_binfmt(struct linux_binfmt * fmt)
94 {
95         write_lock(&binfmt_lock);
96         list_del(&fmt->lh);
97         write_unlock(&binfmt_lock);
98 }
99
100 EXPORT_SYMBOL(unregister_binfmt);
101
102 static inline void put_binfmt(struct linux_binfmt * fmt)
103 {
104         module_put(fmt->module);
105 }
106
107 /*
108  * Note that a shared library must be both readable and executable due to
109  * security reasons.
110  *
111  * Also note that we take the address to load from from the file itself.
112  */
113 SYSCALL_DEFINE1(uselib, const char __user *, library)
114 {
115         struct file *file;
116         char *tmp = getname(library);
117         int error = PTR_ERR(tmp);
118
119         if (IS_ERR(tmp))
120                 goto out;
121
122         file = do_filp_open(AT_FDCWD, tmp,
123                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
124                                 MAY_READ | MAY_EXEC | MAY_OPEN);
125         putname(tmp);
126         error = PTR_ERR(file);
127         if (IS_ERR(file))
128                 goto out;
129
130         error = -EINVAL;
131         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
132                 goto exit;
133
134         error = -EACCES;
135         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
136                 goto exit;
137
138         fsnotify_open(file);
139
140         error = -ENOEXEC;
141         if(file->f_op) {
142                 struct linux_binfmt * fmt;
143
144                 read_lock(&binfmt_lock);
145                 list_for_each_entry(fmt, &formats, lh) {
146                         if (!fmt->load_shlib)
147                                 continue;
148                         if (!try_module_get(fmt->module))
149                                 continue;
150                         read_unlock(&binfmt_lock);
151                         error = fmt->load_shlib(file);
152                         read_lock(&binfmt_lock);
153                         put_binfmt(fmt);
154                         if (error != -ENOEXEC)
155                                 break;
156                 }
157                 read_unlock(&binfmt_lock);
158         }
159 exit:
160         fput(file);
161 out:
162         return error;
163 }
164
165 #ifdef CONFIG_MMU
166
167 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
168 {
169         struct mm_struct *mm = current->mm;
170         long diff = (long)(pages - bprm->vma_pages);
171
172         if (!mm || !diff)
173                 return;
174
175         bprm->vma_pages = pages;
176
177 #ifdef SPLIT_RSS_COUNTING
178         add_mm_counter(mm, MM_ANONPAGES, diff);
179 #else
180         spin_lock(&mm->page_table_lock);
181         add_mm_counter(mm, MM_ANONPAGES, diff);
182         spin_unlock(&mm->page_table_lock);
183 #endif
184 }
185
186 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
187                 int write)
188 {
189         struct page *page;
190         int ret;
191
192 #ifdef CONFIG_STACK_GROWSUP
193         if (write) {
194                 ret = expand_stack_downwards(bprm->vma, pos);
195                 if (ret < 0)
196                         return NULL;
197         }
198 #endif
199         ret = get_user_pages(current, bprm->mm, pos,
200                         1, write, 1, &page, NULL);
201         if (ret <= 0)
202                 return NULL;
203
204         if (write) {
205                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
206                 struct rlimit *rlim;
207
208                 acct_arg_size(bprm, size / PAGE_SIZE);
209
210                 /*
211                  * We've historically supported up to 32 pages (ARG_MAX)
212                  * of argument strings even with small stacks
213                  */
214                 if (size <= ARG_MAX)
215                         return page;
216
217                 /*
218                  * Limit to 1/4-th the stack size for the argv+env strings.
219                  * This ensures that:
220                  *  - the remaining binfmt code will not run out of stack space,
221                  *  - the program will have a reasonable amount of stack left
222                  *    to work from.
223                  */
224                 rlim = current->signal->rlim;
225                 if (size > ACCESS_ONCE(rlim[RLIMIT_STACK].rlim_cur) / 4) {
226                         put_page(page);
227                         return NULL;
228                 }
229         }
230
231         return page;
232 }
233
234 static void put_arg_page(struct page *page)
235 {
236         put_page(page);
237 }
238
239 static void free_arg_page(struct linux_binprm *bprm, int i)
240 {
241 }
242
243 static void free_arg_pages(struct linux_binprm *bprm)
244 {
245 }
246
247 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
248                 struct page *page)
249 {
250         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
251 }
252
253 static int __bprm_mm_init(struct linux_binprm *bprm)
254 {
255         int err;
256         struct vm_area_struct *vma = NULL;
257         struct mm_struct *mm = bprm->mm;
258
259         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
260         if (!vma)
261                 return -ENOMEM;
262
263         down_write(&mm->mmap_sem);
264         vma->vm_mm = mm;
265
266         /*
267          * Place the stack at the largest stack address the architecture
268          * supports. Later, we'll move this to an appropriate place. We don't
269          * use STACK_TOP because that can depend on attributes which aren't
270          * configured yet.
271          */
272         BUG_ON(VM_STACK_FLAGS & VM_STACK_INCOMPLETE_SETUP);
273         vma->vm_end = STACK_TOP_MAX;
274         vma->vm_start = vma->vm_end - PAGE_SIZE;
275         vma->vm_flags = VM_STACK_FLAGS | VM_STACK_INCOMPLETE_SETUP;
276         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
277         INIT_LIST_HEAD(&vma->anon_vma_chain);
278         err = insert_vm_struct(mm, vma);
279         if (err)
280                 goto err;
281
282         mm->stack_vm = mm->total_vm = 1;
283         up_write(&mm->mmap_sem);
284         bprm->p = vma->vm_end - sizeof(void *);
285         return 0;
286 err:
287         up_write(&mm->mmap_sem);
288         bprm->vma = NULL;
289         kmem_cache_free(vm_area_cachep, vma);
290         return err;
291 }
292
293 static bool valid_arg_len(struct linux_binprm *bprm, long len)
294 {
295         return len <= MAX_ARG_STRLEN;
296 }
297
298 #else
299
300 void acct_arg_size(struct linux_binprm *bprm, unsigned long pages)
301 {
302 }
303
304 struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
305                 int write)
306 {
307         struct page *page;
308
309         page = bprm->page[pos / PAGE_SIZE];
310         if (!page && write) {
311                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
312                 if (!page)
313                         return NULL;
314                 bprm->page[pos / PAGE_SIZE] = page;
315         }
316
317         return page;
318 }
319
320 static void put_arg_page(struct page *page)
321 {
322 }
323
324 static void free_arg_page(struct linux_binprm *bprm, int i)
325 {
326         if (bprm->page[i]) {
327                 __free_page(bprm->page[i]);
328                 bprm->page[i] = NULL;
329         }
330 }
331
332 static void free_arg_pages(struct linux_binprm *bprm)
333 {
334         int i;
335
336         for (i = 0; i < MAX_ARG_PAGES; i++)
337                 free_arg_page(bprm, i);
338 }
339
340 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
341                 struct page *page)
342 {
343 }
344
345 static int __bprm_mm_init(struct linux_binprm *bprm)
346 {
347         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
348         return 0;
349 }
350
351 static bool valid_arg_len(struct linux_binprm *bprm, long len)
352 {
353         return len <= bprm->p;
354 }
355
356 #endif /* CONFIG_MMU */
357
358 /*
359  * Create a new mm_struct and populate it with a temporary stack
360  * vm_area_struct.  We don't have enough context at this point to set the stack
361  * flags, permissions, and offset, so we use temporary values.  We'll update
362  * them later in setup_arg_pages().
363  */
364 int bprm_mm_init(struct linux_binprm *bprm)
365 {
366         int err;
367         struct mm_struct *mm = NULL;
368
369         bprm->mm = mm = mm_alloc();
370         err = -ENOMEM;
371         if (!mm)
372                 goto err;
373
374         err = init_new_context(current, mm);
375         if (err)
376                 goto err;
377
378         err = __bprm_mm_init(bprm);
379         if (err)
380                 goto err;
381
382         return 0;
383
384 err:
385         if (mm) {
386                 bprm->mm = NULL;
387                 mmdrop(mm);
388         }
389
390         return err;
391 }
392
393 /*
394  * count() counts the number of strings in array ARGV.
395  */
396 static int count(const char __user * const __user * argv, int max)
397 {
398         int i = 0;
399
400         if (argv != NULL) {
401                 for (;;) {
402                         const char __user * p;
403
404                         if (get_user(p, argv))
405                                 return -EFAULT;
406                         if (!p)
407                                 break;
408                         argv++;
409                         if (i++ >= max)
410                                 return -E2BIG;
411
412                         if (fatal_signal_pending(current))
413                                 return -ERESTARTNOHAND;
414                         cond_resched();
415                 }
416         }
417         return i;
418 }
419
420 /*
421  * 'copy_strings()' copies argument/environment strings from the old
422  * processes's memory to the new process's stack.  The call to get_user_pages()
423  * ensures the destination page is created and not swapped out.
424  */
425 static int copy_strings(int argc, const char __user *const __user *argv,
426                         struct linux_binprm *bprm)
427 {
428         struct page *kmapped_page = NULL;
429         char *kaddr = NULL;
430         unsigned long kpos = 0;
431         int ret;
432
433         while (argc-- > 0) {
434                 const char __user *str;
435                 int len;
436                 unsigned long pos;
437
438                 if (get_user(str, argv+argc) ||
439                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
440                         ret = -EFAULT;
441                         goto out;
442                 }
443
444                 if (!valid_arg_len(bprm, len)) {
445                         ret = -E2BIG;
446                         goto out;
447                 }
448
449                 /* We're going to work our way backwords. */
450                 pos = bprm->p;
451                 str += len;
452                 bprm->p -= len;
453
454                 while (len > 0) {
455                         int offset, bytes_to_copy;
456
457                         if (fatal_signal_pending(current)) {
458                                 ret = -ERESTARTNOHAND;
459                                 goto out;
460                         }
461                         cond_resched();
462
463                         offset = pos % PAGE_SIZE;
464                         if (offset == 0)
465                                 offset = PAGE_SIZE;
466
467                         bytes_to_copy = offset;
468                         if (bytes_to_copy > len)
469                                 bytes_to_copy = len;
470
471                         offset -= bytes_to_copy;
472                         pos -= bytes_to_copy;
473                         str -= bytes_to_copy;
474                         len -= bytes_to_copy;
475
476                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
477                                 struct page *page;
478
479                                 page = get_arg_page(bprm, pos, 1);
480                                 if (!page) {
481                                         ret = -E2BIG;
482                                         goto out;
483                                 }
484
485                                 if (kmapped_page) {
486                                         flush_kernel_dcache_page(kmapped_page);
487                                         kunmap(kmapped_page);
488                                         put_arg_page(kmapped_page);
489                                 }
490                                 kmapped_page = page;
491                                 kaddr = kmap(kmapped_page);
492                                 kpos = pos & PAGE_MASK;
493                                 flush_arg_page(bprm, kpos, kmapped_page);
494                         }
495                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
496                                 ret = -EFAULT;
497                                 goto out;
498                         }
499                 }
500         }
501         ret = 0;
502 out:
503         if (kmapped_page) {
504                 flush_kernel_dcache_page(kmapped_page);
505                 kunmap(kmapped_page);
506                 put_arg_page(kmapped_page);
507         }
508         return ret;
509 }
510
511 /*
512  * Like copy_strings, but get argv and its values from kernel memory.
513  */
514 int copy_strings_kernel(int argc, const char *const *argv,
515                         struct linux_binprm *bprm)
516 {
517         int r;
518         mm_segment_t oldfs = get_fs();
519         set_fs(KERNEL_DS);
520         r = copy_strings(argc, (const char __user *const  __user *)argv, bprm);
521         set_fs(oldfs);
522         return r;
523 }
524 EXPORT_SYMBOL(copy_strings_kernel);
525
526 #ifdef CONFIG_MMU
527
528 /*
529  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
530  * the binfmt code determines where the new stack should reside, we shift it to
531  * its final location.  The process proceeds as follows:
532  *
533  * 1) Use shift to calculate the new vma endpoints.
534  * 2) Extend vma to cover both the old and new ranges.  This ensures the
535  *    arguments passed to subsequent functions are consistent.
536  * 3) Move vma's page tables to the new range.
537  * 4) Free up any cleared pgd range.
538  * 5) Shrink the vma to cover only the new range.
539  */
540 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
541 {
542         struct mm_struct *mm = vma->vm_mm;
543         unsigned long old_start = vma->vm_start;
544         unsigned long old_end = vma->vm_end;
545         unsigned long length = old_end - old_start;
546         unsigned long new_start = old_start - shift;
547         unsigned long new_end = old_end - shift;
548         struct mmu_gather *tlb;
549
550         BUG_ON(new_start > new_end);
551
552         /*
553          * ensure there are no vmas between where we want to go
554          * and where we are
555          */
556         if (vma != find_vma(mm, new_start))
557                 return -EFAULT;
558
559         /*
560          * cover the whole range: [new_start, old_end)
561          */
562         if (vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL))
563                 return -ENOMEM;
564
565         /*
566          * move the page tables downwards, on failure we rely on
567          * process cleanup to remove whatever mess we made.
568          */
569         if (length != move_page_tables(vma, old_start,
570                                        vma, new_start, length))
571                 return -ENOMEM;
572
573         lru_add_drain();
574         tlb = tlb_gather_mmu(mm, 0);
575         if (new_end > old_start) {
576                 /*
577                  * when the old and new regions overlap clear from new_end.
578                  */
579                 free_pgd_range(tlb, new_end, old_end, new_end,
580                         vma->vm_next ? vma->vm_next->vm_start : 0);
581         } else {
582                 /*
583                  * otherwise, clean from old_start; this is done to not touch
584                  * the address space in [new_end, old_start) some architectures
585                  * have constraints on va-space that make this illegal (IA64) -
586                  * for the others its just a little faster.
587                  */
588                 free_pgd_range(tlb, old_start, old_end, new_end,
589                         vma->vm_next ? vma->vm_next->vm_start : 0);
590         }
591         tlb_finish_mmu(tlb, new_end, old_end);
592
593         /*
594          * Shrink the vma to just the new range.  Always succeeds.
595          */
596         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
597
598         return 0;
599 }
600
601 /*
602  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
603  * the stack is optionally relocated, and some extra space is added.
604  */
605 int setup_arg_pages(struct linux_binprm *bprm,
606                     unsigned long stack_top,
607                     int executable_stack)
608 {
609         unsigned long ret;
610         unsigned long stack_shift;
611         struct mm_struct *mm = current->mm;
612         struct vm_area_struct *vma = bprm->vma;
613         struct vm_area_struct *prev = NULL;
614         unsigned long vm_flags;
615         unsigned long stack_base;
616         unsigned long stack_size;
617         unsigned long stack_expand;
618         unsigned long rlim_stack;
619
620 #ifdef CONFIG_STACK_GROWSUP
621         /* Limit stack size to 1GB */
622         stack_base = rlimit_max(RLIMIT_STACK);
623         if (stack_base > (1 << 30))
624                 stack_base = 1 << 30;
625
626         /* Make sure we didn't let the argument array grow too large. */
627         if (vma->vm_end - vma->vm_start > stack_base)
628                 return -ENOMEM;
629
630         stack_base = PAGE_ALIGN(stack_top - stack_base);
631
632         stack_shift = vma->vm_start - stack_base;
633         mm->arg_start = bprm->p - stack_shift;
634         bprm->p = vma->vm_end - stack_shift;
635 #else
636         stack_top = arch_align_stack(stack_top);
637         stack_top = PAGE_ALIGN(stack_top);
638
639         if (unlikely(stack_top < mmap_min_addr) ||
640             unlikely(vma->vm_end - vma->vm_start >= stack_top - mmap_min_addr))
641                 return -ENOMEM;
642
643         stack_shift = vma->vm_end - stack_top;
644
645         bprm->p -= stack_shift;
646         mm->arg_start = bprm->p;
647 #endif
648
649         if (bprm->loader)
650                 bprm->loader -= stack_shift;
651         bprm->exec -= stack_shift;
652
653         down_write(&mm->mmap_sem);
654         vm_flags = VM_STACK_FLAGS;
655
656         /*
657          * Adjust stack execute permissions; explicitly enable for
658          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
659          * (arch default) otherwise.
660          */
661         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
662                 vm_flags |= VM_EXEC;
663         else if (executable_stack == EXSTACK_DISABLE_X)
664                 vm_flags &= ~VM_EXEC;
665         vm_flags |= mm->def_flags;
666         vm_flags |= VM_STACK_INCOMPLETE_SETUP;
667
668         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
669                         vm_flags);
670         if (ret)
671                 goto out_unlock;
672         BUG_ON(prev != vma);
673
674         /* Move stack pages down in memory. */
675         if (stack_shift) {
676                 ret = shift_arg_pages(vma, stack_shift);
677                 if (ret)
678                         goto out_unlock;
679         }
680
681         /* mprotect_fixup is overkill to remove the temporary stack flags */
682         vma->vm_flags &= ~VM_STACK_INCOMPLETE_SETUP;
683
684         stack_expand = 131072UL; /* randomly 32*4k (or 2*64k) pages */
685         stack_size = vma->vm_end - vma->vm_start;
686         /*
687          * Align this down to a page boundary as expand_stack
688          * will align it up.
689          */
690         rlim_stack = rlimit(RLIMIT_STACK) & PAGE_MASK;
691 #ifdef CONFIG_STACK_GROWSUP
692         if (stack_size + stack_expand > rlim_stack)
693                 stack_base = vma->vm_start + rlim_stack;
694         else
695                 stack_base = vma->vm_end + stack_expand;
696 #else
697         if (stack_size + stack_expand > rlim_stack)
698                 stack_base = vma->vm_end - rlim_stack;
699         else
700                 stack_base = vma->vm_start - stack_expand;
701 #endif
702         current->mm->start_stack = bprm->p;
703         ret = expand_stack(vma, stack_base);
704         if (ret)
705                 ret = -EFAULT;
706
707 out_unlock:
708         up_write(&mm->mmap_sem);
709         return ret;
710 }
711 EXPORT_SYMBOL(setup_arg_pages);
712
713 #endif /* CONFIG_MMU */
714
715 struct file *open_exec(const char *name)
716 {
717         struct file *file;
718         int err;
719
720         file = do_filp_open(AT_FDCWD, name,
721                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
722                                 MAY_EXEC | MAY_OPEN);
723         if (IS_ERR(file))
724                 goto out;
725
726         err = -EACCES;
727         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
728                 goto exit;
729
730         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
731                 goto exit;
732
733         fsnotify_open(file);
734
735         err = deny_write_access(file);
736         if (err)
737                 goto exit;
738
739 out:
740         return file;
741
742 exit:
743         fput(file);
744         return ERR_PTR(err);
745 }
746 EXPORT_SYMBOL(open_exec);
747
748 int kernel_read(struct file *file, loff_t offset,
749                 char *addr, unsigned long count)
750 {
751         mm_segment_t old_fs;
752         loff_t pos = offset;
753         int result;
754
755         old_fs = get_fs();
756         set_fs(get_ds());
757         /* The cast to a user pointer is valid due to the set_fs() */
758         result = vfs_read(file, (void __user *)addr, count, &pos);
759         set_fs(old_fs);
760         return result;
761 }
762
763 EXPORT_SYMBOL(kernel_read);
764
765 static int exec_mmap(struct mm_struct *mm)
766 {
767         struct task_struct *tsk;
768         struct mm_struct * old_mm, *active_mm;
769
770         /* Notify parent that we're no longer interested in the old VM */
771         tsk = current;
772         old_mm = current->mm;
773         sync_mm_rss(tsk, old_mm);
774         mm_release(tsk, old_mm);
775
776         if (old_mm) {
777                 /*
778                  * Make sure that if there is a core dump in progress
779                  * for the old mm, we get out and die instead of going
780                  * through with the exec.  We must hold mmap_sem around
781                  * checking core_state and changing tsk->mm.
782                  */
783                 down_read(&old_mm->mmap_sem);
784                 if (unlikely(old_mm->core_state)) {
785                         up_read(&old_mm->mmap_sem);
786                         return -EINTR;
787                 }
788         }
789         task_lock(tsk);
790         active_mm = tsk->active_mm;
791         tsk->mm = mm;
792         tsk->active_mm = mm;
793         activate_mm(active_mm, mm);
794         if (old_mm && tsk->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
795                 atomic_dec(&old_mm->oom_disable_count);
796                 atomic_inc(&tsk->mm->oom_disable_count);
797         }
798         task_unlock(tsk);
799         arch_pick_mmap_layout(mm);
800         if (old_mm) {
801                 up_read(&old_mm->mmap_sem);
802                 BUG_ON(active_mm != old_mm);
803                 mm_update_next_owner(old_mm);
804                 mmput(old_mm);
805                 return 0;
806         }
807         mmdrop(active_mm);
808         return 0;
809 }
810
811 /*
812  * This function makes sure the current process has its own signal table,
813  * so that flush_signal_handlers can later reset the handlers without
814  * disturbing other processes.  (Other processes might share the signal
815  * table via the CLONE_SIGHAND option to clone().)
816  */
817 static int de_thread(struct task_struct *tsk)
818 {
819         struct signal_struct *sig = tsk->signal;
820         struct sighand_struct *oldsighand = tsk->sighand;
821         spinlock_t *lock = &oldsighand->siglock;
822
823         if (thread_group_empty(tsk))
824                 goto no_thread_group;
825
826         /*
827          * Kill all other threads in the thread group.
828          */
829         spin_lock_irq(lock);
830         if (signal_group_exit(sig)) {
831                 /*
832                  * Another group action in progress, just
833                  * return so that the signal is processed.
834                  */
835                 spin_unlock_irq(lock);
836                 return -EAGAIN;
837         }
838
839         sig->group_exit_task = tsk;
840         sig->notify_count = zap_other_threads(tsk);
841         if (!thread_group_leader(tsk))
842                 sig->notify_count--;
843
844         while (sig->notify_count) {
845                 __set_current_state(TASK_UNINTERRUPTIBLE);
846                 spin_unlock_irq(lock);
847                 schedule();
848                 spin_lock_irq(lock);
849         }
850         spin_unlock_irq(lock);
851
852         /*
853          * At this point all other threads have exited, all we have to
854          * do is to wait for the thread group leader to become inactive,
855          * and to assume its PID:
856          */
857         if (!thread_group_leader(tsk)) {
858                 struct task_struct *leader = tsk->group_leader;
859
860                 sig->notify_count = -1; /* for exit_notify() */
861                 for (;;) {
862                         write_lock_irq(&tasklist_lock);
863                         if (likely(leader->exit_state))
864                                 break;
865                         __set_current_state(TASK_UNINTERRUPTIBLE);
866                         write_unlock_irq(&tasklist_lock);
867                         schedule();
868                 }
869
870                 /*
871                  * The only record we have of the real-time age of a
872                  * process, regardless of execs it's done, is start_time.
873                  * All the past CPU time is accumulated in signal_struct
874                  * from sister threads now dead.  But in this non-leader
875                  * exec, nothing survives from the original leader thread,
876                  * whose birth marks the true age of this process now.
877                  * When we take on its identity by switching to its PID, we
878                  * also take its birthdate (always earlier than our own).
879                  */
880                 tsk->start_time = leader->start_time;
881
882                 BUG_ON(!same_thread_group(leader, tsk));
883                 BUG_ON(has_group_leader_pid(tsk));
884                 /*
885                  * An exec() starts a new thread group with the
886                  * TGID of the previous thread group. Rehash the
887                  * two threads with a switched PID, and release
888                  * the former thread group leader:
889                  */
890
891                 /* Become a process group leader with the old leader's pid.
892                  * The old leader becomes a thread of the this thread group.
893                  * Note: The old leader also uses this pid until release_task
894                  *       is called.  Odd but simple and correct.
895                  */
896                 detach_pid(tsk, PIDTYPE_PID);
897                 tsk->pid = leader->pid;
898                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
899                 transfer_pid(leader, tsk, PIDTYPE_PGID);
900                 transfer_pid(leader, tsk, PIDTYPE_SID);
901
902                 list_replace_rcu(&leader->tasks, &tsk->tasks);
903                 list_replace_init(&leader->sibling, &tsk->sibling);
904
905                 tsk->group_leader = tsk;
906                 leader->group_leader = tsk;
907
908                 tsk->exit_signal = SIGCHLD;
909
910                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
911                 leader->exit_state = EXIT_DEAD;
912                 write_unlock_irq(&tasklist_lock);
913
914                 release_task(leader);
915         }
916
917         sig->group_exit_task = NULL;
918         sig->notify_count = 0;
919
920 no_thread_group:
921         if (current->mm)
922                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
923
924         exit_itimers(sig);
925         flush_itimer_signals();
926
927         if (atomic_read(&oldsighand->count) != 1) {
928                 struct sighand_struct *newsighand;
929                 /*
930                  * This ->sighand is shared with the CLONE_SIGHAND
931                  * but not CLONE_THREAD task, switch to the new one.
932                  */
933                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
934                 if (!newsighand)
935                         return -ENOMEM;
936
937                 atomic_set(&newsighand->count, 1);
938                 memcpy(newsighand->action, oldsighand->action,
939                        sizeof(newsighand->action));
940
941                 write_lock_irq(&tasklist_lock);
942                 spin_lock(&oldsighand->siglock);
943                 rcu_assign_pointer(tsk->sighand, newsighand);
944                 spin_unlock(&oldsighand->siglock);
945                 write_unlock_irq(&tasklist_lock);
946
947                 __cleanup_sighand(oldsighand);
948         }
949
950         BUG_ON(!thread_group_leader(tsk));
951         return 0;
952 }
953
954 /*
955  * These functions flushes out all traces of the currently running executable
956  * so that a new one can be started
957  */
958 static void flush_old_files(struct files_struct * files)
959 {
960         long j = -1;
961         struct fdtable *fdt;
962
963         spin_lock(&files->file_lock);
964         for (;;) {
965                 unsigned long set, i;
966
967                 j++;
968                 i = j * __NFDBITS;
969                 fdt = files_fdtable(files);
970                 if (i >= fdt->max_fds)
971                         break;
972                 set = fdt->close_on_exec->fds_bits[j];
973                 if (!set)
974                         continue;
975                 fdt->close_on_exec->fds_bits[j] = 0;
976                 spin_unlock(&files->file_lock);
977                 for ( ; set ; i++,set >>= 1) {
978                         if (set & 1) {
979                                 sys_close(i);
980                         }
981                 }
982                 spin_lock(&files->file_lock);
983
984         }
985         spin_unlock(&files->file_lock);
986 }
987
988 char *get_task_comm(char *buf, struct task_struct *tsk)
989 {
990         /* buf must be at least sizeof(tsk->comm) in size */
991         task_lock(tsk);
992         strncpy(buf, tsk->comm, sizeof(tsk->comm));
993         task_unlock(tsk);
994         return buf;
995 }
996
997 void set_task_comm(struct task_struct *tsk, char *buf)
998 {
999         task_lock(tsk);
1000
1001         /*
1002          * Threads may access current->comm without holding
1003          * the task lock, so write the string carefully.
1004          * Readers without a lock may see incomplete new
1005          * names but are safe from non-terminating string reads.
1006          */
1007         memset(tsk->comm, 0, TASK_COMM_LEN);
1008         wmb();
1009         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
1010         task_unlock(tsk);
1011         perf_event_comm(tsk);
1012 }
1013
1014 int flush_old_exec(struct linux_binprm * bprm)
1015 {
1016         int retval;
1017
1018         /*
1019          * Make sure we have a private signal table and that
1020          * we are unassociated from the previous thread group.
1021          */
1022         retval = de_thread(current);
1023         if (retval)
1024                 goto out;
1025
1026         set_mm_exe_file(bprm->mm, bprm->file);
1027
1028         /*
1029          * Release all of the old mmap stuff
1030          */
1031         acct_arg_size(bprm, 0);
1032         retval = exec_mmap(bprm->mm);
1033         if (retval)
1034                 goto out;
1035
1036         bprm->mm = NULL;                /* We're using it now */
1037
1038         current->flags &= ~(PF_RANDOMIZE | PF_KTHREAD);
1039         flush_thread();
1040         current->personality &= ~bprm->per_clear;
1041
1042         return 0;
1043
1044 out:
1045         return retval;
1046 }
1047 EXPORT_SYMBOL(flush_old_exec);
1048
1049 void setup_new_exec(struct linux_binprm * bprm)
1050 {
1051         int i, ch;
1052         const char *name;
1053         char tcomm[sizeof(current->comm)];
1054
1055         arch_pick_mmap_layout(current->mm);
1056
1057         /* This is the point of no return */
1058         current->sas_ss_sp = current->sas_ss_size = 0;
1059
1060         if (current_euid() == current_uid() && current_egid() == current_gid())
1061                 set_dumpable(current->mm, 1);
1062         else
1063                 set_dumpable(current->mm, suid_dumpable);
1064
1065         name = bprm->filename;
1066
1067         /* Copies the binary name from after last slash */
1068         for (i=0; (ch = *(name++)) != '\0';) {
1069                 if (ch == '/')
1070                         i = 0; /* overwrite what we wrote */
1071                 else
1072                         if (i < (sizeof(tcomm) - 1))
1073                                 tcomm[i++] = ch;
1074         }
1075         tcomm[i] = '\0';
1076         set_task_comm(current, tcomm);
1077
1078         /* Set the new mm task size. We have to do that late because it may
1079          * depend on TIF_32BIT which is only updated in flush_thread() on
1080          * some architectures like powerpc
1081          */
1082         current->mm->task_size = TASK_SIZE;
1083
1084         /* install the new credentials */
1085         if (bprm->cred->uid != current_euid() ||
1086             bprm->cred->gid != current_egid()) {
1087                 current->pdeath_signal = 0;
1088         } else if (file_permission(bprm->file, MAY_READ) ||
1089                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1090                 set_dumpable(current->mm, suid_dumpable);
1091         }
1092
1093         /*
1094          * Flush performance counters when crossing a
1095          * security domain:
1096          */
1097         if (!get_dumpable(current->mm))
1098                 perf_event_exit_task(current);
1099
1100         /* An exec changes our domain. We are no longer part of the thread
1101            group */
1102
1103         current->self_exec_id++;
1104                         
1105         flush_signal_handlers(current, 0);
1106         flush_old_files(current->files);
1107 }
1108 EXPORT_SYMBOL(setup_new_exec);
1109
1110 /*
1111  * Prepare credentials and lock ->cred_guard_mutex.
1112  * install_exec_creds() commits the new creds and drops the lock.
1113  * Or, if exec fails before, free_bprm() should release ->cred and
1114  * and unlock.
1115  */
1116 int prepare_bprm_creds(struct linux_binprm *bprm)
1117 {
1118         if (mutex_lock_interruptible(&current->signal->cred_guard_mutex))
1119                 return -ERESTARTNOINTR;
1120
1121         bprm->cred = prepare_exec_creds();
1122         if (likely(bprm->cred))
1123                 return 0;
1124
1125         mutex_unlock(&current->signal->cred_guard_mutex);
1126         return -ENOMEM;
1127 }
1128
1129 void free_bprm(struct linux_binprm *bprm)
1130 {
1131         free_arg_pages(bprm);
1132         if (bprm->cred) {
1133                 mutex_unlock(&current->signal->cred_guard_mutex);
1134                 abort_creds(bprm->cred);
1135         }
1136         kfree(bprm);
1137 }
1138
1139 /*
1140  * install the new credentials for this executable
1141  */
1142 void install_exec_creds(struct linux_binprm *bprm)
1143 {
1144         security_bprm_committing_creds(bprm);
1145
1146         commit_creds(bprm->cred);
1147         bprm->cred = NULL;
1148         /*
1149          * cred_guard_mutex must be held at least to this point to prevent
1150          * ptrace_attach() from altering our determination of the task's
1151          * credentials; any time after this it may be unlocked.
1152          */
1153         security_bprm_committed_creds(bprm);
1154         mutex_unlock(&current->signal->cred_guard_mutex);
1155 }
1156 EXPORT_SYMBOL(install_exec_creds);
1157
1158 /*
1159  * determine how safe it is to execute the proposed program
1160  * - the caller must hold ->cred_guard_mutex to protect against
1161  *   PTRACE_ATTACH
1162  */
1163 int check_unsafe_exec(struct linux_binprm *bprm)
1164 {
1165         struct task_struct *p = current, *t;
1166         unsigned n_fs;
1167         int res = 0;
1168
1169         bprm->unsafe = tracehook_unsafe_exec(p);
1170
1171         n_fs = 1;
1172         spin_lock(&p->fs->lock);
1173         rcu_read_lock();
1174         for (t = next_thread(p); t != p; t = next_thread(t)) {
1175                 if (t->fs == p->fs)
1176                         n_fs++;
1177         }
1178         rcu_read_unlock();
1179
1180         if (p->fs->users > n_fs) {
1181                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1182         } else {
1183                 res = -EAGAIN;
1184                 if (!p->fs->in_exec) {
1185                         p->fs->in_exec = 1;
1186                         res = 1;
1187                 }
1188         }
1189         spin_unlock(&p->fs->lock);
1190
1191         return res;
1192 }
1193
1194 /* 
1195  * Fill the binprm structure from the inode. 
1196  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1197  *
1198  * This may be called multiple times for binary chains (scripts for example).
1199  */
1200 int prepare_binprm(struct linux_binprm *bprm)
1201 {
1202         umode_t mode;
1203         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1204         int retval;
1205
1206         mode = inode->i_mode;
1207         if (bprm->file->f_op == NULL)
1208                 return -EACCES;
1209
1210         /* clear any previous set[ug]id data from a previous binary */
1211         bprm->cred->euid = current_euid();
1212         bprm->cred->egid = current_egid();
1213
1214         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1215                 /* Set-uid? */
1216                 if (mode & S_ISUID) {
1217                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1218                         bprm->cred->euid = inode->i_uid;
1219                 }
1220
1221                 /* Set-gid? */
1222                 /*
1223                  * If setgid is set but no group execute bit then this
1224                  * is a candidate for mandatory locking, not a setgid
1225                  * executable.
1226                  */
1227                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1228                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1229                         bprm->cred->egid = inode->i_gid;
1230                 }
1231         }
1232
1233         /* fill in binprm security blob */
1234         retval = security_bprm_set_creds(bprm);
1235         if (retval)
1236                 return retval;
1237         bprm->cred_prepared = 1;
1238
1239         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1240         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1241 }
1242
1243 EXPORT_SYMBOL(prepare_binprm);
1244
1245 /*
1246  * Arguments are '\0' separated strings found at the location bprm->p
1247  * points to; chop off the first by relocating brpm->p to right after
1248  * the first '\0' encountered.
1249  */
1250 int remove_arg_zero(struct linux_binprm *bprm)
1251 {
1252         int ret = 0;
1253         unsigned long offset;
1254         char *kaddr;
1255         struct page *page;
1256
1257         if (!bprm->argc)
1258                 return 0;
1259
1260         do {
1261                 offset = bprm->p & ~PAGE_MASK;
1262                 page = get_arg_page(bprm, bprm->p, 0);
1263                 if (!page) {
1264                         ret = -EFAULT;
1265                         goto out;
1266                 }
1267                 kaddr = kmap_atomic(page, KM_USER0);
1268
1269                 for (; offset < PAGE_SIZE && kaddr[offset];
1270                                 offset++, bprm->p++)
1271                         ;
1272
1273                 kunmap_atomic(kaddr, KM_USER0);
1274                 put_arg_page(page);
1275
1276                 if (offset == PAGE_SIZE)
1277                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1278         } while (offset == PAGE_SIZE);
1279
1280         bprm->p++;
1281         bprm->argc--;
1282         ret = 0;
1283
1284 out:
1285         return ret;
1286 }
1287 EXPORT_SYMBOL(remove_arg_zero);
1288
1289 /*
1290  * cycle the list of binary formats handler, until one recognizes the image
1291  */
1292 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1293 {
1294         unsigned int depth = bprm->recursion_depth;
1295         int try,retval;
1296         struct linux_binfmt *fmt;
1297
1298         retval = security_bprm_check(bprm);
1299         if (retval)
1300                 return retval;
1301
1302         /* kernel module loader fixup */
1303         /* so we don't try to load run modprobe in kernel space. */
1304         set_fs(USER_DS);
1305
1306         retval = audit_bprm(bprm);
1307         if (retval)
1308                 return retval;
1309
1310         retval = -ENOENT;
1311         for (try=0; try<2; try++) {
1312                 read_lock(&binfmt_lock);
1313                 list_for_each_entry(fmt, &formats, lh) {
1314                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1315                         if (!fn)
1316                                 continue;
1317                         if (!try_module_get(fmt->module))
1318                                 continue;
1319                         read_unlock(&binfmt_lock);
1320                         retval = fn(bprm, regs);
1321                         /*
1322                          * Restore the depth counter to its starting value
1323                          * in this call, so we don't have to rely on every
1324                          * load_binary function to restore it on return.
1325                          */
1326                         bprm->recursion_depth = depth;
1327                         if (retval >= 0) {
1328                                 if (depth == 0)
1329                                         tracehook_report_exec(fmt, bprm, regs);
1330                                 put_binfmt(fmt);
1331                                 allow_write_access(bprm->file);
1332                                 if (bprm->file)
1333                                         fput(bprm->file);
1334                                 bprm->file = NULL;
1335                                 current->did_exec = 1;
1336                                 proc_exec_connector(current);
1337                                 return retval;
1338                         }
1339                         read_lock(&binfmt_lock);
1340                         put_binfmt(fmt);
1341                         if (retval != -ENOEXEC || bprm->mm == NULL)
1342                                 break;
1343                         if (!bprm->file) {
1344                                 read_unlock(&binfmt_lock);
1345                                 return retval;
1346                         }
1347                 }
1348                 read_unlock(&binfmt_lock);
1349                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1350                         break;
1351 #ifdef CONFIG_MODULES
1352                 } else {
1353 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1354                         if (printable(bprm->buf[0]) &&
1355                             printable(bprm->buf[1]) &&
1356                             printable(bprm->buf[2]) &&
1357                             printable(bprm->buf[3]))
1358                                 break; /* -ENOEXEC */
1359                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1360 #endif
1361                 }
1362         }
1363         return retval;
1364 }
1365
1366 EXPORT_SYMBOL(search_binary_handler);
1367
1368 /*
1369  * sys_execve() executes a new program.
1370  */
1371 int do_execve(const char * filename,
1372         const char __user *const __user *argv,
1373         const char __user *const __user *envp,
1374         struct pt_regs * regs)
1375 {
1376         struct linux_binprm *bprm;
1377         struct file *file;
1378         struct files_struct *displaced;
1379         bool clear_in_exec;
1380         int retval;
1381
1382         retval = unshare_files(&displaced);
1383         if (retval)
1384                 goto out_ret;
1385
1386         retval = -ENOMEM;
1387         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1388         if (!bprm)
1389                 goto out_files;
1390
1391         retval = prepare_bprm_creds(bprm);
1392         if (retval)
1393                 goto out_free;
1394
1395         retval = check_unsafe_exec(bprm);
1396         if (retval < 0)
1397                 goto out_free;
1398         clear_in_exec = retval;
1399         current->in_execve = 1;
1400
1401         file = open_exec(filename);
1402         retval = PTR_ERR(file);
1403         if (IS_ERR(file))
1404                 goto out_unmark;
1405
1406         sched_exec();
1407
1408         bprm->file = file;
1409         bprm->filename = filename;
1410         bprm->interp = filename;
1411
1412         retval = bprm_mm_init(bprm);
1413         if (retval)
1414                 goto out_file;
1415
1416         bprm->argc = count(argv, MAX_ARG_STRINGS);
1417         if ((retval = bprm->argc) < 0)
1418                 goto out;
1419
1420         bprm->envc = count(envp, MAX_ARG_STRINGS);
1421         if ((retval = bprm->envc) < 0)
1422                 goto out;
1423
1424         retval = prepare_binprm(bprm);
1425         if (retval < 0)
1426                 goto out;
1427
1428         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1429         if (retval < 0)
1430                 goto out;
1431
1432         bprm->exec = bprm->p;
1433         retval = copy_strings(bprm->envc, envp, bprm);
1434         if (retval < 0)
1435                 goto out;
1436
1437         retval = copy_strings(bprm->argc, argv, bprm);
1438         if (retval < 0)
1439                 goto out;
1440
1441         retval = search_binary_handler(bprm,regs);
1442         if (retval < 0)
1443                 goto out;
1444
1445         /* execve succeeded */
1446         current->fs->in_exec = 0;
1447         current->in_execve = 0;
1448         acct_update_integrals(current);
1449         free_bprm(bprm);
1450         if (displaced)
1451                 put_files_struct(displaced);
1452         return retval;
1453
1454 out:
1455         if (bprm->mm) {
1456                 acct_arg_size(bprm, 0);
1457                 mmput(bprm->mm);
1458         }
1459
1460 out_file:
1461         if (bprm->file) {
1462                 allow_write_access(bprm->file);
1463                 fput(bprm->file);
1464         }
1465
1466 out_unmark:
1467         if (clear_in_exec)
1468                 current->fs->in_exec = 0;
1469         current->in_execve = 0;
1470
1471 out_free:
1472         free_bprm(bprm);
1473
1474 out_files:
1475         if (displaced)
1476                 reset_files_struct(displaced);
1477 out_ret:
1478         return retval;
1479 }
1480
1481 void set_binfmt(struct linux_binfmt *new)
1482 {
1483         struct mm_struct *mm = current->mm;
1484
1485         if (mm->binfmt)
1486                 module_put(mm->binfmt->module);
1487
1488         mm->binfmt = new;
1489         if (new)
1490                 __module_get(new->module);
1491 }
1492
1493 EXPORT_SYMBOL(set_binfmt);
1494
1495 static int expand_corename(struct core_name *cn)
1496 {
1497         char *old_corename = cn->corename;
1498
1499         cn->size = CORENAME_MAX_SIZE * atomic_inc_return(&call_count);
1500         cn->corename = krealloc(old_corename, cn->size, GFP_KERNEL);
1501
1502         if (!cn->corename) {
1503                 kfree(old_corename);
1504                 return -ENOMEM;
1505         }
1506
1507         return 0;
1508 }
1509
1510 static int cn_printf(struct core_name *cn, const char *fmt, ...)
1511 {
1512         char *cur;
1513         int need;
1514         int ret;
1515         va_list arg;
1516
1517         va_start(arg, fmt);
1518         need = vsnprintf(NULL, 0, fmt, arg);
1519         va_end(arg);
1520
1521         if (likely(need < cn->size - cn->used - 1))
1522                 goto out_printf;
1523
1524         ret = expand_corename(cn);
1525         if (ret)
1526                 goto expand_fail;
1527
1528 out_printf:
1529         cur = cn->corename + cn->used;
1530         va_start(arg, fmt);
1531         vsnprintf(cur, need + 1, fmt, arg);
1532         va_end(arg);
1533         cn->used += need;
1534         return 0;
1535
1536 expand_fail:
1537         return ret;
1538 }
1539
1540 /* format_corename will inspect the pattern parameter, and output a
1541  * name into corename, which must have space for at least
1542  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1543  */
1544 static int format_corename(struct core_name *cn, long signr)
1545 {
1546         const struct cred *cred = current_cred();
1547         const char *pat_ptr = core_pattern;
1548         int ispipe = (*pat_ptr == '|');
1549         int pid_in_pattern = 0;
1550         int err = 0;
1551
1552         cn->size = CORENAME_MAX_SIZE * atomic_read(&call_count);
1553         cn->corename = kmalloc(cn->size, GFP_KERNEL);
1554         cn->used = 0;
1555
1556         if (!cn->corename)
1557                 return -ENOMEM;
1558
1559         /* Repeat as long as we have more pattern to process and more output
1560            space */
1561         while (*pat_ptr) {
1562                 if (*pat_ptr != '%') {
1563                         if (*pat_ptr == 0)
1564                                 goto out;
1565                         err = cn_printf(cn, "%c", *pat_ptr++);
1566                 } else {
1567                         switch (*++pat_ptr) {
1568                         /* single % at the end, drop that */
1569                         case 0:
1570                                 goto out;
1571                         /* Double percent, output one percent */
1572                         case '%':
1573                                 err = cn_printf(cn, "%c", '%');
1574                                 break;
1575                         /* pid */
1576                         case 'p':
1577                                 pid_in_pattern = 1;
1578                                 err = cn_printf(cn, "%d",
1579                                               task_tgid_vnr(current));
1580                                 break;
1581                         /* uid */
1582                         case 'u':
1583                                 err = cn_printf(cn, "%d", cred->uid);
1584                                 break;
1585                         /* gid */
1586                         case 'g':
1587                                 err = cn_printf(cn, "%d", cred->gid);
1588                                 break;
1589                         /* signal that caused the coredump */
1590                         case 's':
1591                                 err = cn_printf(cn, "%ld", signr);
1592                                 break;
1593                         /* UNIX time of coredump */
1594                         case 't': {
1595                                 struct timeval tv;
1596                                 do_gettimeofday(&tv);
1597                                 err = cn_printf(cn, "%lu", tv.tv_sec);
1598                                 break;
1599                         }
1600                         /* hostname */
1601                         case 'h':
1602                                 down_read(&uts_sem);
1603                                 err = cn_printf(cn, "%s",
1604                                               utsname()->nodename);
1605                                 up_read(&uts_sem);
1606                                 break;
1607                         /* executable */
1608                         case 'e':
1609                                 err = cn_printf(cn, "%s", current->comm);
1610                                 break;
1611                         /* core limit size */
1612                         case 'c':
1613                                 err = cn_printf(cn, "%lu",
1614                                               rlimit(RLIMIT_CORE));
1615                                 break;
1616                         default:
1617                                 break;
1618                         }
1619                         ++pat_ptr;
1620                 }
1621
1622                 if (err)
1623                         return err;
1624         }
1625
1626         /* Backward compatibility with core_uses_pid:
1627          *
1628          * If core_pattern does not include a %p (as is the default)
1629          * and core_uses_pid is set, then .%pid will be appended to
1630          * the filename. Do not do this for piped commands. */
1631         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1632                 err = cn_printf(cn, ".%d", task_tgid_vnr(current));
1633                 if (err)
1634                         return err;
1635         }
1636 out:
1637         return ispipe;
1638 }
1639
1640 static int zap_process(struct task_struct *start, int exit_code)
1641 {
1642         struct task_struct *t;
1643         int nr = 0;
1644
1645         start->signal->flags = SIGNAL_GROUP_EXIT;
1646         start->signal->group_exit_code = exit_code;
1647         start->signal->group_stop_count = 0;
1648
1649         t = start;
1650         do {
1651                 if (t != current && t->mm) {
1652                         sigaddset(&t->pending.signal, SIGKILL);
1653                         signal_wake_up(t, 1);
1654                         nr++;
1655                 }
1656         } while_each_thread(start, t);
1657
1658         return nr;
1659 }
1660
1661 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1662                                 struct core_state *core_state, int exit_code)
1663 {
1664         struct task_struct *g, *p;
1665         unsigned long flags;
1666         int nr = -EAGAIN;
1667
1668         spin_lock_irq(&tsk->sighand->siglock);
1669         if (!signal_group_exit(tsk->signal)) {
1670                 mm->core_state = core_state;
1671                 nr = zap_process(tsk, exit_code);
1672         }
1673         spin_unlock_irq(&tsk->sighand->siglock);
1674         if (unlikely(nr < 0))
1675                 return nr;
1676
1677         if (atomic_read(&mm->mm_users) == nr + 1)
1678                 goto done;
1679         /*
1680          * We should find and kill all tasks which use this mm, and we should
1681          * count them correctly into ->nr_threads. We don't take tasklist
1682          * lock, but this is safe wrt:
1683          *
1684          * fork:
1685          *      None of sub-threads can fork after zap_process(leader). All
1686          *      processes which were created before this point should be
1687          *      visible to zap_threads() because copy_process() adds the new
1688          *      process to the tail of init_task.tasks list, and lock/unlock
1689          *      of ->siglock provides a memory barrier.
1690          *
1691          * do_exit:
1692          *      The caller holds mm->mmap_sem. This means that the task which
1693          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1694          *      its ->mm.
1695          *
1696          * de_thread:
1697          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1698          *      we must see either old or new leader, this does not matter.
1699          *      However, it can change p->sighand, so lock_task_sighand(p)
1700          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1701          *      it can't fail.
1702          *
1703          *      Note also that "g" can be the old leader with ->mm == NULL
1704          *      and already unhashed and thus removed from ->thread_group.
1705          *      This is OK, __unhash_process()->list_del_rcu() does not
1706          *      clear the ->next pointer, we will find the new leader via
1707          *      next_thread().
1708          */
1709         rcu_read_lock();
1710         for_each_process(g) {
1711                 if (g == tsk->group_leader)
1712                         continue;
1713                 if (g->flags & PF_KTHREAD)
1714                         continue;
1715                 p = g;
1716                 do {
1717                         if (p->mm) {
1718                                 if (unlikely(p->mm == mm)) {
1719                                         lock_task_sighand(p, &flags);
1720                                         nr += zap_process(p, exit_code);
1721                                         unlock_task_sighand(p, &flags);
1722                                 }
1723                                 break;
1724                         }
1725                 } while_each_thread(g, p);
1726         }
1727         rcu_read_unlock();
1728 done:
1729         atomic_set(&core_state->nr_threads, nr);
1730         return nr;
1731 }
1732
1733 static int coredump_wait(int exit_code, struct core_state *core_state)
1734 {
1735         struct task_struct *tsk = current;
1736         struct mm_struct *mm = tsk->mm;
1737         struct completion *vfork_done;
1738         int core_waiters = -EBUSY;
1739
1740         init_completion(&core_state->startup);
1741         core_state->dumper.task = tsk;
1742         core_state->dumper.next = NULL;
1743
1744         down_write(&mm->mmap_sem);
1745         if (!mm->core_state)
1746                 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1747         up_write(&mm->mmap_sem);
1748
1749         if (unlikely(core_waiters < 0))
1750                 goto fail;
1751
1752         /*
1753          * Make sure nobody is waiting for us to release the VM,
1754          * otherwise we can deadlock when we wait on each other
1755          */
1756         vfork_done = tsk->vfork_done;
1757         if (vfork_done) {
1758                 tsk->vfork_done = NULL;
1759                 complete(vfork_done);
1760         }
1761
1762         if (core_waiters)
1763                 wait_for_completion(&core_state->startup);
1764 fail:
1765         return core_waiters;
1766 }
1767
1768 static void coredump_finish(struct mm_struct *mm)
1769 {
1770         struct core_thread *curr, *next;
1771         struct task_struct *task;
1772
1773         next = mm->core_state->dumper.next;
1774         while ((curr = next) != NULL) {
1775                 next = curr->next;
1776                 task = curr->task;
1777                 /*
1778                  * see exit_mm(), curr->task must not see
1779                  * ->task == NULL before we read ->next.
1780                  */
1781                 smp_mb();
1782                 curr->task = NULL;
1783                 wake_up_process(task);
1784         }
1785
1786         mm->core_state = NULL;
1787 }
1788
1789 /*
1790  * set_dumpable converts traditional three-value dumpable to two flags and
1791  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1792  * these bits are not changed atomically.  So get_dumpable can observe the
1793  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1794  * return either old dumpable or new one by paying attention to the order of
1795  * modifying the bits.
1796  *
1797  * dumpable |   mm->flags (binary)
1798  * old  new | initial interim  final
1799  * ---------+-----------------------
1800  *  0    1  |   00      01      01
1801  *  0    2  |   00      10(*)   11
1802  *  1    0  |   01      00      00
1803  *  1    2  |   01      11      11
1804  *  2    0  |   11      10(*)   00
1805  *  2    1  |   11      11      01
1806  *
1807  * (*) get_dumpable regards interim value of 10 as 11.
1808  */
1809 void set_dumpable(struct mm_struct *mm, int value)
1810 {
1811         switch (value) {
1812         case 0:
1813                 clear_bit(MMF_DUMPABLE, &mm->flags);
1814                 smp_wmb();
1815                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1816                 break;
1817         case 1:
1818                 set_bit(MMF_DUMPABLE, &mm->flags);
1819                 smp_wmb();
1820                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1821                 break;
1822         case 2:
1823                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1824                 smp_wmb();
1825                 set_bit(MMF_DUMPABLE, &mm->flags);
1826                 break;
1827         }
1828 }
1829
1830 static int __get_dumpable(unsigned long mm_flags)
1831 {
1832         int ret;
1833
1834         ret = mm_flags & MMF_DUMPABLE_MASK;
1835         return (ret >= 2) ? 2 : ret;
1836 }
1837
1838 int get_dumpable(struct mm_struct *mm)
1839 {
1840         return __get_dumpable(mm->flags);
1841 }
1842
1843 static void wait_for_dump_helpers(struct file *file)
1844 {
1845         struct pipe_inode_info *pipe;
1846
1847         pipe = file->f_path.dentry->d_inode->i_pipe;
1848
1849         pipe_lock(pipe);
1850         pipe->readers++;
1851         pipe->writers--;
1852
1853         while ((pipe->readers > 1) && (!signal_pending(current))) {
1854                 wake_up_interruptible_sync(&pipe->wait);
1855                 kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
1856                 pipe_wait(pipe);
1857         }
1858
1859         pipe->readers--;
1860         pipe->writers++;
1861         pipe_unlock(pipe);
1862
1863 }
1864
1865
1866 /*
1867  * uhm_pipe_setup
1868  * helper function to customize the process used
1869  * to collect the core in userspace.  Specifically
1870  * it sets up a pipe and installs it as fd 0 (stdin)
1871  * for the process.  Returns 0 on success, or
1872  * PTR_ERR on failure.
1873  * Note that it also sets the core limit to 1.  This
1874  * is a special value that we use to trap recursive
1875  * core dumps
1876  */
1877 static int umh_pipe_setup(struct subprocess_info *info)
1878 {
1879         struct file *rp, *wp;
1880         struct fdtable *fdt;
1881         struct coredump_params *cp = (struct coredump_params *)info->data;
1882         struct files_struct *cf = current->files;
1883
1884         wp = create_write_pipe(0);
1885         if (IS_ERR(wp))
1886                 return PTR_ERR(wp);
1887
1888         rp = create_read_pipe(wp, 0);
1889         if (IS_ERR(rp)) {
1890                 free_write_pipe(wp);
1891                 return PTR_ERR(rp);
1892         }
1893
1894         cp->file = wp;
1895
1896         sys_close(0);
1897         fd_install(0, rp);
1898         spin_lock(&cf->file_lock);
1899         fdt = files_fdtable(cf);
1900         FD_SET(0, fdt->open_fds);
1901         FD_CLR(0, fdt->close_on_exec);
1902         spin_unlock(&cf->file_lock);
1903
1904         /* and disallow core files too */
1905         current->signal->rlim[RLIMIT_CORE] = (struct rlimit){1, 1};
1906
1907         return 0;
1908 }
1909
1910 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1911 {
1912         struct core_state core_state;
1913         struct core_name cn;
1914         struct mm_struct *mm = current->mm;
1915         struct linux_binfmt * binfmt;
1916         const struct cred *old_cred;
1917         struct cred *cred;
1918         int retval = 0;
1919         int flag = 0;
1920         int ispipe;
1921         static atomic_t core_dump_count = ATOMIC_INIT(0);
1922         struct coredump_params cprm = {
1923                 .signr = signr,
1924                 .regs = regs,
1925                 .limit = rlimit(RLIMIT_CORE),
1926                 /*
1927                  * We must use the same mm->flags while dumping core to avoid
1928                  * inconsistency of bit flags, since this flag is not protected
1929                  * by any locks.
1930                  */
1931                 .mm_flags = mm->flags,
1932         };
1933
1934         audit_core_dumps(signr);
1935
1936         binfmt = mm->binfmt;
1937         if (!binfmt || !binfmt->core_dump)
1938                 goto fail;
1939         if (!__get_dumpable(cprm.mm_flags))
1940                 goto fail;
1941
1942         cred = prepare_creds();
1943         if (!cred)
1944                 goto fail;
1945         /*
1946          *      We cannot trust fsuid as being the "true" uid of the
1947          *      process nor do we know its entire history. We only know it
1948          *      was tainted so we dump it as root in mode 2.
1949          */
1950         if (__get_dumpable(cprm.mm_flags) == 2) {
1951                 /* Setuid core dump mode */
1952                 flag = O_EXCL;          /* Stop rewrite attacks */
1953                 cred->fsuid = 0;        /* Dump root private */
1954         }
1955
1956         retval = coredump_wait(exit_code, &core_state);
1957         if (retval < 0)
1958                 goto fail_creds;
1959
1960         old_cred = override_creds(cred);
1961
1962         /*
1963          * Clear any false indication of pending signals that might
1964          * be seen by the filesystem code called to write the core file.
1965          */
1966         clear_thread_flag(TIF_SIGPENDING);
1967
1968         ispipe = format_corename(&cn, signr);
1969
1970         if (ispipe == -ENOMEM) {
1971                 printk(KERN_WARNING "format_corename failed\n");
1972                 printk(KERN_WARNING "Aborting core\n");
1973                 goto fail_corename;
1974         }
1975
1976         if (ispipe) {
1977                 int dump_count;
1978                 char **helper_argv;
1979
1980                 if (cprm.limit == 1) {
1981                         /*
1982                          * Normally core limits are irrelevant to pipes, since
1983                          * we're not writing to the file system, but we use
1984                          * cprm.limit of 1 here as a speacial value. Any
1985                          * non-1 limit gets set to RLIM_INFINITY below, but
1986                          * a limit of 0 skips the dump.  This is a consistent
1987                          * way to catch recursive crashes.  We can still crash
1988                          * if the core_pattern binary sets RLIM_CORE =  !1
1989                          * but it runs as root, and can do lots of stupid things
1990                          * Note that we use task_tgid_vnr here to grab the pid
1991                          * of the process group leader.  That way we get the
1992                          * right pid if a thread in a multi-threaded
1993                          * core_pattern process dies.
1994                          */
1995                         printk(KERN_WARNING
1996                                 "Process %d(%s) has RLIMIT_CORE set to 1\n",
1997                                 task_tgid_vnr(current), current->comm);
1998                         printk(KERN_WARNING "Aborting core\n");
1999                         goto fail_unlock;
2000                 }
2001                 cprm.limit = RLIM_INFINITY;
2002
2003                 dump_count = atomic_inc_return(&core_dump_count);
2004                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
2005                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
2006                                task_tgid_vnr(current), current->comm);
2007                         printk(KERN_WARNING "Skipping core dump\n");
2008                         goto fail_dropcount;
2009                 }
2010
2011                 helper_argv = argv_split(GFP_KERNEL, cn.corename+1, NULL);
2012                 if (!helper_argv) {
2013                         printk(KERN_WARNING "%s failed to allocate memory\n",
2014                                __func__);
2015                         goto fail_dropcount;
2016                 }
2017
2018                 retval = call_usermodehelper_fns(helper_argv[0], helper_argv,
2019                                         NULL, UMH_WAIT_EXEC, umh_pipe_setup,
2020                                         NULL, &cprm);
2021                 argv_free(helper_argv);
2022                 if (retval) {
2023                         printk(KERN_INFO "Core dump to %s pipe failed\n",
2024                                cn.corename);
2025                         goto close_fail;
2026                 }
2027         } else {
2028                 struct inode *inode;
2029
2030                 if (cprm.limit < binfmt->min_coredump)
2031                         goto fail_unlock;
2032
2033                 cprm.file = filp_open(cn.corename,
2034                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
2035                                  0600);
2036                 if (IS_ERR(cprm.file))
2037                         goto fail_unlock;
2038
2039                 inode = cprm.file->f_path.dentry->d_inode;
2040                 if (inode->i_nlink > 1)
2041                         goto close_fail;
2042                 if (d_unhashed(cprm.file->f_path.dentry))
2043                         goto close_fail;
2044                 /*
2045                  * AK: actually i see no reason to not allow this for named
2046                  * pipes etc, but keep the previous behaviour for now.
2047                  */
2048                 if (!S_ISREG(inode->i_mode))
2049                         goto close_fail;
2050                 /*
2051                  * Dont allow local users get cute and trick others to coredump
2052                  * into their pre-created files.
2053                  */
2054                 if (inode->i_uid != current_fsuid())
2055                         goto close_fail;
2056                 if (!cprm.file->f_op || !cprm.file->f_op->write)
2057                         goto close_fail;
2058                 if (do_truncate(cprm.file->f_path.dentry, 0, 0, cprm.file))
2059                         goto close_fail;
2060         }
2061
2062         retval = binfmt->core_dump(&cprm);
2063         if (retval)
2064                 current->signal->group_exit_code |= 0x80;
2065
2066         if (ispipe && core_pipe_limit)
2067                 wait_for_dump_helpers(cprm.file);
2068 close_fail:
2069         if (cprm.file)
2070                 filp_close(cprm.file, NULL);
2071 fail_dropcount:
2072         if (ispipe)
2073                 atomic_dec(&core_dump_count);
2074 fail_unlock:
2075         kfree(cn.corename);
2076 fail_corename:
2077         coredump_finish(mm);
2078         revert_creds(old_cred);
2079 fail_creds:
2080         put_cred(cred);
2081 fail:
2082         return;
2083 }
2084
2085 /*
2086  * Core dumping helper functions.  These are the only things you should
2087  * do on a core-file: use only these functions to write out all the
2088  * necessary info.
2089  */
2090 int dump_write(struct file *file, const void *addr, int nr)
2091 {
2092         return access_ok(VERIFY_READ, addr, nr) && file->f_op->write(file, addr, nr, &file->f_pos) == nr;
2093 }
2094 EXPORT_SYMBOL(dump_write);
2095
2096 int dump_seek(struct file *file, loff_t off)
2097 {
2098         int ret = 1;
2099
2100         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
2101                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
2102                         return 0;
2103         } else {
2104                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
2105
2106                 if (!buf)
2107                         return 0;
2108                 while (off > 0) {
2109                         unsigned long n = off;
2110
2111                         if (n > PAGE_SIZE)
2112                                 n = PAGE_SIZE;
2113                         if (!dump_write(file, buf, n)) {
2114                                 ret = 0;
2115                                 break;
2116                         }
2117                         off -= n;
2118                 }
2119                 free_page((unsigned long)buf);
2120         }
2121         return ret;
2122 }
2123 EXPORT_SYMBOL(dump_seek);