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