Merge branch 'linus' into perfcounters/core
[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/smp_lock.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/perf_counter.h>
37 #include <linux/highmem.h>
38 #include <linux/spinlock.h>
39 #include <linux/key.h>
40 #include <linux/personality.h>
41 #include <linux/binfmts.h>
42 #include <linux/utsname.h>
43 #include <linux/pid_namespace.h>
44 #include <linux/module.h>
45 #include <linux/namei.h>
46 #include <linux/proc_fs.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/ima.h>
50 #include <linux/syscalls.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
54 #include <linux/tracehook.h>
55 #include <linux/kmod.h>
56 #include <linux/fsnotify.h>
57 #include <linux/fs_struct.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 int suid_dumpable = 0;
67
68 /* The maximal length of core_pattern is also specified in sysctl.c */
69
70 static LIST_HEAD(formats);
71 static DEFINE_RWLOCK(binfmt_lock);
72
73 int register_binfmt(struct linux_binfmt * fmt)
74 {
75         if (!fmt)
76                 return -EINVAL;
77         write_lock(&binfmt_lock);
78         list_add(&fmt->lh, &formats);
79         write_unlock(&binfmt_lock);
80         return 0;       
81 }
82
83 EXPORT_SYMBOL(register_binfmt);
84
85 void unregister_binfmt(struct linux_binfmt * fmt)
86 {
87         write_lock(&binfmt_lock);
88         list_del(&fmt->lh);
89         write_unlock(&binfmt_lock);
90 }
91
92 EXPORT_SYMBOL(unregister_binfmt);
93
94 static inline void put_binfmt(struct linux_binfmt * fmt)
95 {
96         module_put(fmt->module);
97 }
98
99 /*
100  * Note that a shared library must be both readable and executable due to
101  * security reasons.
102  *
103  * Also note that we take the address to load from from the file itself.
104  */
105 SYSCALL_DEFINE1(uselib, const char __user *, library)
106 {
107         struct file *file;
108         struct nameidata nd;
109         char *tmp = getname(library);
110         int error = PTR_ERR(tmp);
111
112         if (!IS_ERR(tmp)) {
113                 error = path_lookup_open(AT_FDCWD, tmp,
114                                          LOOKUP_FOLLOW, &nd,
115                                          FMODE_READ|FMODE_EXEC);
116                 putname(tmp);
117         }
118         if (error)
119                 goto out;
120
121         error = -EINVAL;
122         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
123                 goto exit;
124
125         error = -EACCES;
126         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
127                 goto exit;
128
129         error = inode_permission(nd.path.dentry->d_inode,
130                                  MAY_READ | MAY_EXEC | MAY_OPEN);
131         if (error)
132                 goto exit;
133         error = ima_path_check(&nd.path, MAY_READ | MAY_EXEC | MAY_OPEN);
134         if (error)
135                 goto exit;
136
137         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
138         error = PTR_ERR(file);
139         if (IS_ERR(file))
140                 goto out;
141
142         fsnotify_open(file->f_path.dentry);
143
144         error = -ENOEXEC;
145         if(file->f_op) {
146                 struct linux_binfmt * fmt;
147
148                 read_lock(&binfmt_lock);
149                 list_for_each_entry(fmt, &formats, lh) {
150                         if (!fmt->load_shlib)
151                                 continue;
152                         if (!try_module_get(fmt->module))
153                                 continue;
154                         read_unlock(&binfmt_lock);
155                         error = fmt->load_shlib(file);
156                         read_lock(&binfmt_lock);
157                         put_binfmt(fmt);
158                         if (error != -ENOEXEC)
159                                 break;
160                 }
161                 read_unlock(&binfmt_lock);
162         }
163         fput(file);
164 out:
165         return error;
166 exit:
167         release_open_intent(&nd);
168         path_put(&nd.path);
169         goto out;
170 }
171
172 #ifdef CONFIG_MMU
173
174 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
175                 int write)
176 {
177         struct page *page;
178         int ret;
179
180 #ifdef CONFIG_STACK_GROWSUP
181         if (write) {
182                 ret = expand_stack_downwards(bprm->vma, pos);
183                 if (ret < 0)
184                         return NULL;
185         }
186 #endif
187         ret = get_user_pages(current, bprm->mm, pos,
188                         1, write, 1, &page, NULL);
189         if (ret <= 0)
190                 return NULL;
191
192         if (write) {
193                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
194                 struct rlimit *rlim;
195
196                 /*
197                  * We've historically supported up to 32 pages (ARG_MAX)
198                  * of argument strings even with small stacks
199                  */
200                 if (size <= ARG_MAX)
201                         return page;
202
203                 /*
204                  * Limit to 1/4-th the stack size for the argv+env strings.
205                  * This ensures that:
206                  *  - the remaining binfmt code will not run out of stack space,
207                  *  - the program will have a reasonable amount of stack left
208                  *    to work from.
209                  */
210                 rlim = current->signal->rlim;
211                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
212                         put_page(page);
213                         return NULL;
214                 }
215         }
216
217         return page;
218 }
219
220 static void put_arg_page(struct page *page)
221 {
222         put_page(page);
223 }
224
225 static void free_arg_page(struct linux_binprm *bprm, int i)
226 {
227 }
228
229 static void free_arg_pages(struct linux_binprm *bprm)
230 {
231 }
232
233 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
234                 struct page *page)
235 {
236         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
237 }
238
239 static int __bprm_mm_init(struct linux_binprm *bprm)
240 {
241         int err;
242         struct vm_area_struct *vma = NULL;
243         struct mm_struct *mm = bprm->mm;
244
245         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
246         if (!vma)
247                 return -ENOMEM;
248
249         down_write(&mm->mmap_sem);
250         vma->vm_mm = mm;
251
252         /*
253          * Place the stack at the largest stack address the architecture
254          * supports. Later, we'll move this to an appropriate place. We don't
255          * use STACK_TOP because that can depend on attributes which aren't
256          * configured yet.
257          */
258         vma->vm_end = STACK_TOP_MAX;
259         vma->vm_start = vma->vm_end - PAGE_SIZE;
260         vma->vm_flags = VM_STACK_FLAGS;
261         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
262         err = insert_vm_struct(mm, vma);
263         if (err)
264                 goto err;
265
266         mm->stack_vm = mm->total_vm = 1;
267         up_write(&mm->mmap_sem);
268         bprm->p = vma->vm_end - sizeof(void *);
269         return 0;
270 err:
271         up_write(&mm->mmap_sem);
272         bprm->vma = NULL;
273         kmem_cache_free(vm_area_cachep, vma);
274         return err;
275 }
276
277 static bool valid_arg_len(struct linux_binprm *bprm, long len)
278 {
279         return len <= MAX_ARG_STRLEN;
280 }
281
282 #else
283
284 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
285                 int write)
286 {
287         struct page *page;
288
289         page = bprm->page[pos / PAGE_SIZE];
290         if (!page && write) {
291                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
292                 if (!page)
293                         return NULL;
294                 bprm->page[pos / PAGE_SIZE] = page;
295         }
296
297         return page;
298 }
299
300 static void put_arg_page(struct page *page)
301 {
302 }
303
304 static void free_arg_page(struct linux_binprm *bprm, int i)
305 {
306         if (bprm->page[i]) {
307                 __free_page(bprm->page[i]);
308                 bprm->page[i] = NULL;
309         }
310 }
311
312 static void free_arg_pages(struct linux_binprm *bprm)
313 {
314         int i;
315
316         for (i = 0; i < MAX_ARG_PAGES; i++)
317                 free_arg_page(bprm, i);
318 }
319
320 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
321                 struct page *page)
322 {
323 }
324
325 static int __bprm_mm_init(struct linux_binprm *bprm)
326 {
327         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
328         return 0;
329 }
330
331 static bool valid_arg_len(struct linux_binprm *bprm, long len)
332 {
333         return len <= bprm->p;
334 }
335
336 #endif /* CONFIG_MMU */
337
338 /*
339  * Create a new mm_struct and populate it with a temporary stack
340  * vm_area_struct.  We don't have enough context at this point to set the stack
341  * flags, permissions, and offset, so we use temporary values.  We'll update
342  * them later in setup_arg_pages().
343  */
344 int bprm_mm_init(struct linux_binprm *bprm)
345 {
346         int err;
347         struct mm_struct *mm = NULL;
348
349         bprm->mm = mm = mm_alloc();
350         err = -ENOMEM;
351         if (!mm)
352                 goto err;
353
354         err = init_new_context(current, mm);
355         if (err)
356                 goto err;
357
358         err = __bprm_mm_init(bprm);
359         if (err)
360                 goto err;
361
362         return 0;
363
364 err:
365         if (mm) {
366                 bprm->mm = NULL;
367                 mmdrop(mm);
368         }
369
370         return err;
371 }
372
373 /*
374  * count() counts the number of strings in array ARGV.
375  */
376 static int count(char __user * __user * argv, int max)
377 {
378         int i = 0;
379
380         if (argv != NULL) {
381                 for (;;) {
382                         char __user * p;
383
384                         if (get_user(p, argv))
385                                 return -EFAULT;
386                         if (!p)
387                                 break;
388                         argv++;
389                         if (i++ >= max)
390                                 return -E2BIG;
391                         cond_resched();
392                 }
393         }
394         return i;
395 }
396
397 /*
398  * 'copy_strings()' copies argument/environment strings from the old
399  * processes's memory to the new process's stack.  The call to get_user_pages()
400  * ensures the destination page is created and not swapped out.
401  */
402 static int copy_strings(int argc, char __user * __user * argv,
403                         struct linux_binprm *bprm)
404 {
405         struct page *kmapped_page = NULL;
406         char *kaddr = NULL;
407         unsigned long kpos = 0;
408         int ret;
409
410         while (argc-- > 0) {
411                 char __user *str;
412                 int len;
413                 unsigned long pos;
414
415                 if (get_user(str, argv+argc) ||
416                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
417                         ret = -EFAULT;
418                         goto out;
419                 }
420
421                 if (!valid_arg_len(bprm, len)) {
422                         ret = -E2BIG;
423                         goto out;
424                 }
425
426                 /* We're going to work our way backwords. */
427                 pos = bprm->p;
428                 str += len;
429                 bprm->p -= len;
430
431                 while (len > 0) {
432                         int offset, bytes_to_copy;
433
434                         offset = pos % PAGE_SIZE;
435                         if (offset == 0)
436                                 offset = PAGE_SIZE;
437
438                         bytes_to_copy = offset;
439                         if (bytes_to_copy > len)
440                                 bytes_to_copy = len;
441
442                         offset -= bytes_to_copy;
443                         pos -= bytes_to_copy;
444                         str -= bytes_to_copy;
445                         len -= bytes_to_copy;
446
447                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
448                                 struct page *page;
449
450                                 page = get_arg_page(bprm, pos, 1);
451                                 if (!page) {
452                                         ret = -E2BIG;
453                                         goto out;
454                                 }
455
456                                 if (kmapped_page) {
457                                         flush_kernel_dcache_page(kmapped_page);
458                                         kunmap(kmapped_page);
459                                         put_arg_page(kmapped_page);
460                                 }
461                                 kmapped_page = page;
462                                 kaddr = kmap(kmapped_page);
463                                 kpos = pos & PAGE_MASK;
464                                 flush_arg_page(bprm, kpos, kmapped_page);
465                         }
466                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
467                                 ret = -EFAULT;
468                                 goto out;
469                         }
470                 }
471         }
472         ret = 0;
473 out:
474         if (kmapped_page) {
475                 flush_kernel_dcache_page(kmapped_page);
476                 kunmap(kmapped_page);
477                 put_arg_page(kmapped_page);
478         }
479         return ret;
480 }
481
482 /*
483  * Like copy_strings, but get argv and its values from kernel memory.
484  */
485 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
486 {
487         int r;
488         mm_segment_t oldfs = get_fs();
489         set_fs(KERNEL_DS);
490         r = copy_strings(argc, (char __user * __user *)argv, bprm);
491         set_fs(oldfs);
492         return r;
493 }
494 EXPORT_SYMBOL(copy_strings_kernel);
495
496 #ifdef CONFIG_MMU
497
498 /*
499  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
500  * the binfmt code determines where the new stack should reside, we shift it to
501  * its final location.  The process proceeds as follows:
502  *
503  * 1) Use shift to calculate the new vma endpoints.
504  * 2) Extend vma to cover both the old and new ranges.  This ensures the
505  *    arguments passed to subsequent functions are consistent.
506  * 3) Move vma's page tables to the new range.
507  * 4) Free up any cleared pgd range.
508  * 5) Shrink the vma to cover only the new range.
509  */
510 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
511 {
512         struct mm_struct *mm = vma->vm_mm;
513         unsigned long old_start = vma->vm_start;
514         unsigned long old_end = vma->vm_end;
515         unsigned long length = old_end - old_start;
516         unsigned long new_start = old_start - shift;
517         unsigned long new_end = old_end - shift;
518         struct mmu_gather *tlb;
519
520         BUG_ON(new_start > new_end);
521
522         /*
523          * ensure there are no vmas between where we want to go
524          * and where we are
525          */
526         if (vma != find_vma(mm, new_start))
527                 return -EFAULT;
528
529         /*
530          * cover the whole range: [new_start, old_end)
531          */
532         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
533
534         /*
535          * move the page tables downwards, on failure we rely on
536          * process cleanup to remove whatever mess we made.
537          */
538         if (length != move_page_tables(vma, old_start,
539                                        vma, new_start, length))
540                 return -ENOMEM;
541
542         lru_add_drain();
543         tlb = tlb_gather_mmu(mm, 0);
544         if (new_end > old_start) {
545                 /*
546                  * when the old and new regions overlap clear from new_end.
547                  */
548                 free_pgd_range(tlb, new_end, old_end, new_end,
549                         vma->vm_next ? vma->vm_next->vm_start : 0);
550         } else {
551                 /*
552                  * otherwise, clean from old_start; this is done to not touch
553                  * the address space in [new_end, old_start) some architectures
554                  * have constraints on va-space that make this illegal (IA64) -
555                  * for the others its just a little faster.
556                  */
557                 free_pgd_range(tlb, old_start, old_end, new_end,
558                         vma->vm_next ? vma->vm_next->vm_start : 0);
559         }
560         tlb_finish_mmu(tlb, new_end, old_end);
561
562         /*
563          * shrink the vma to just the new range.
564          */
565         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
566
567         return 0;
568 }
569
570 #define EXTRA_STACK_VM_PAGES    20      /* random */
571
572 /*
573  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
574  * the stack is optionally relocated, and some extra space is added.
575  */
576 int setup_arg_pages(struct linux_binprm *bprm,
577                     unsigned long stack_top,
578                     int executable_stack)
579 {
580         unsigned long ret;
581         unsigned long stack_shift;
582         struct mm_struct *mm = current->mm;
583         struct vm_area_struct *vma = bprm->vma;
584         struct vm_area_struct *prev = NULL;
585         unsigned long vm_flags;
586         unsigned long stack_base;
587
588 #ifdef CONFIG_STACK_GROWSUP
589         /* Limit stack size to 1GB */
590         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
591         if (stack_base > (1 << 30))
592                 stack_base = 1 << 30;
593
594         /* Make sure we didn't let the argument array grow too large. */
595         if (vma->vm_end - vma->vm_start > stack_base)
596                 return -ENOMEM;
597
598         stack_base = PAGE_ALIGN(stack_top - stack_base);
599
600         stack_shift = vma->vm_start - stack_base;
601         mm->arg_start = bprm->p - stack_shift;
602         bprm->p = vma->vm_end - stack_shift;
603 #else
604         stack_top = arch_align_stack(stack_top);
605         stack_top = PAGE_ALIGN(stack_top);
606         stack_shift = vma->vm_end - stack_top;
607
608         bprm->p -= stack_shift;
609         mm->arg_start = bprm->p;
610 #endif
611
612         if (bprm->loader)
613                 bprm->loader -= stack_shift;
614         bprm->exec -= stack_shift;
615
616         down_write(&mm->mmap_sem);
617         vm_flags = VM_STACK_FLAGS;
618
619         /*
620          * Adjust stack execute permissions; explicitly enable for
621          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
622          * (arch default) otherwise.
623          */
624         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
625                 vm_flags |= VM_EXEC;
626         else if (executable_stack == EXSTACK_DISABLE_X)
627                 vm_flags &= ~VM_EXEC;
628         vm_flags |= mm->def_flags;
629
630         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
631                         vm_flags);
632         if (ret)
633                 goto out_unlock;
634         BUG_ON(prev != vma);
635
636         /* Move stack pages down in memory. */
637         if (stack_shift) {
638                 ret = shift_arg_pages(vma, stack_shift);
639                 if (ret) {
640                         up_write(&mm->mmap_sem);
641                         return ret;
642                 }
643         }
644
645 #ifdef CONFIG_STACK_GROWSUP
646         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
647 #else
648         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
649 #endif
650         ret = expand_stack(vma, stack_base);
651         if (ret)
652                 ret = -EFAULT;
653
654 out_unlock:
655         up_write(&mm->mmap_sem);
656         return 0;
657 }
658 EXPORT_SYMBOL(setup_arg_pages);
659
660 #endif /* CONFIG_MMU */
661
662 struct file *open_exec(const char *name)
663 {
664         struct nameidata nd;
665         struct file *file;
666         int err;
667
668         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
669                                 FMODE_READ|FMODE_EXEC);
670         if (err)
671                 goto out;
672
673         err = -EACCES;
674         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
675                 goto out_path_put;
676
677         if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
678                 goto out_path_put;
679
680         err = inode_permission(nd.path.dentry->d_inode, MAY_EXEC | MAY_OPEN);
681         if (err)
682                 goto out_path_put;
683         err = ima_path_check(&nd.path, MAY_EXEC | MAY_OPEN);
684         if (err)
685                 goto out_path_put;
686
687         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
688         if (IS_ERR(file))
689                 return file;
690
691         fsnotify_open(file->f_path.dentry);
692
693         err = deny_write_access(file);
694         if (err) {
695                 fput(file);
696                 goto out;
697         }
698
699         return file;
700
701  out_path_put:
702         release_open_intent(&nd);
703         path_put(&nd.path);
704  out:
705         return ERR_PTR(err);
706 }
707 EXPORT_SYMBOL(open_exec);
708
709 int kernel_read(struct file *file, unsigned long offset,
710         char *addr, unsigned long count)
711 {
712         mm_segment_t old_fs;
713         loff_t pos = offset;
714         int result;
715
716         old_fs = get_fs();
717         set_fs(get_ds());
718         /* The cast to a user pointer is valid due to the set_fs() */
719         result = vfs_read(file, (void __user *)addr, count, &pos);
720         set_fs(old_fs);
721         return result;
722 }
723
724 EXPORT_SYMBOL(kernel_read);
725
726 static int exec_mmap(struct mm_struct *mm)
727 {
728         struct task_struct *tsk;
729         struct mm_struct * old_mm, *active_mm;
730
731         /* Notify parent that we're no longer interested in the old VM */
732         tsk = current;
733         old_mm = current->mm;
734         mm_release(tsk, old_mm);
735
736         if (old_mm) {
737                 /*
738                  * Make sure that if there is a core dump in progress
739                  * for the old mm, we get out and die instead of going
740                  * through with the exec.  We must hold mmap_sem around
741                  * checking core_state and changing tsk->mm.
742                  */
743                 down_read(&old_mm->mmap_sem);
744                 if (unlikely(old_mm->core_state)) {
745                         up_read(&old_mm->mmap_sem);
746                         return -EINTR;
747                 }
748         }
749         task_lock(tsk);
750         active_mm = tsk->active_mm;
751         tsk->mm = mm;
752         tsk->active_mm = mm;
753         activate_mm(active_mm, mm);
754         task_unlock(tsk);
755         arch_pick_mmap_layout(mm);
756         if (old_mm) {
757                 up_read(&old_mm->mmap_sem);
758                 BUG_ON(active_mm != old_mm);
759                 mm_update_next_owner(old_mm);
760                 mmput(old_mm);
761                 return 0;
762         }
763         mmdrop(active_mm);
764         return 0;
765 }
766
767 /*
768  * This function makes sure the current process has its own signal table,
769  * so that flush_signal_handlers can later reset the handlers without
770  * disturbing other processes.  (Other processes might share the signal
771  * table via the CLONE_SIGHAND option to clone().)
772  */
773 static int de_thread(struct task_struct *tsk)
774 {
775         struct signal_struct *sig = tsk->signal;
776         struct sighand_struct *oldsighand = tsk->sighand;
777         spinlock_t *lock = &oldsighand->siglock;
778         int count;
779
780         if (thread_group_empty(tsk))
781                 goto no_thread_group;
782
783         /*
784          * Kill all other threads in the thread group.
785          */
786         spin_lock_irq(lock);
787         if (signal_group_exit(sig)) {
788                 /*
789                  * Another group action in progress, just
790                  * return so that the signal is processed.
791                  */
792                 spin_unlock_irq(lock);
793                 return -EAGAIN;
794         }
795         sig->group_exit_task = tsk;
796         zap_other_threads(tsk);
797
798         /* Account for the thread group leader hanging around: */
799         count = thread_group_leader(tsk) ? 1 : 2;
800         sig->notify_count = count;
801         while (atomic_read(&sig->count) > count) {
802                 __set_current_state(TASK_UNINTERRUPTIBLE);
803                 spin_unlock_irq(lock);
804                 schedule();
805                 spin_lock_irq(lock);
806         }
807         spin_unlock_irq(lock);
808
809         /*
810          * At this point all other threads have exited, all we have to
811          * do is to wait for the thread group leader to become inactive,
812          * and to assume its PID:
813          */
814         if (!thread_group_leader(tsk)) {
815                 struct task_struct *leader = tsk->group_leader;
816
817                 sig->notify_count = -1; /* for exit_notify() */
818                 for (;;) {
819                         write_lock_irq(&tasklist_lock);
820                         if (likely(leader->exit_state))
821                                 break;
822                         __set_current_state(TASK_UNINTERRUPTIBLE);
823                         write_unlock_irq(&tasklist_lock);
824                         schedule();
825                 }
826
827                 /*
828                  * The only record we have of the real-time age of a
829                  * process, regardless of execs it's done, is start_time.
830                  * All the past CPU time is accumulated in signal_struct
831                  * from sister threads now dead.  But in this non-leader
832                  * exec, nothing survives from the original leader thread,
833                  * whose birth marks the true age of this process now.
834                  * When we take on its identity by switching to its PID, we
835                  * also take its birthdate (always earlier than our own).
836                  */
837                 tsk->start_time = leader->start_time;
838
839                 BUG_ON(!same_thread_group(leader, tsk));
840                 BUG_ON(has_group_leader_pid(tsk));
841                 /*
842                  * An exec() starts a new thread group with the
843                  * TGID of the previous thread group. Rehash the
844                  * two threads with a switched PID, and release
845                  * the former thread group leader:
846                  */
847
848                 /* Become a process group leader with the old leader's pid.
849                  * The old leader becomes a thread of the this thread group.
850                  * Note: The old leader also uses this pid until release_task
851                  *       is called.  Odd but simple and correct.
852                  */
853                 detach_pid(tsk, PIDTYPE_PID);
854                 tsk->pid = leader->pid;
855                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
856                 transfer_pid(leader, tsk, PIDTYPE_PGID);
857                 transfer_pid(leader, tsk, PIDTYPE_SID);
858                 list_replace_rcu(&leader->tasks, &tsk->tasks);
859
860                 tsk->group_leader = tsk;
861                 leader->group_leader = tsk;
862
863                 tsk->exit_signal = SIGCHLD;
864
865                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
866                 leader->exit_state = EXIT_DEAD;
867                 write_unlock_irq(&tasklist_lock);
868
869                 release_task(leader);
870         }
871
872         sig->group_exit_task = NULL;
873         sig->notify_count = 0;
874
875 no_thread_group:
876         exit_itimers(sig);
877         flush_itimer_signals();
878
879         if (atomic_read(&oldsighand->count) != 1) {
880                 struct sighand_struct *newsighand;
881                 /*
882                  * This ->sighand is shared with the CLONE_SIGHAND
883                  * but not CLONE_THREAD task, switch to the new one.
884                  */
885                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
886                 if (!newsighand)
887                         return -ENOMEM;
888
889                 atomic_set(&newsighand->count, 1);
890                 memcpy(newsighand->action, oldsighand->action,
891                        sizeof(newsighand->action));
892
893                 write_lock_irq(&tasklist_lock);
894                 spin_lock(&oldsighand->siglock);
895                 rcu_assign_pointer(tsk->sighand, newsighand);
896                 spin_unlock(&oldsighand->siglock);
897                 write_unlock_irq(&tasklist_lock);
898
899                 __cleanup_sighand(oldsighand);
900         }
901
902         BUG_ON(!thread_group_leader(tsk));
903         return 0;
904 }
905
906 /*
907  * These functions flushes out all traces of the currently running executable
908  * so that a new one can be started
909  */
910 static void flush_old_files(struct files_struct * files)
911 {
912         long j = -1;
913         struct fdtable *fdt;
914
915         spin_lock(&files->file_lock);
916         for (;;) {
917                 unsigned long set, i;
918
919                 j++;
920                 i = j * __NFDBITS;
921                 fdt = files_fdtable(files);
922                 if (i >= fdt->max_fds)
923                         break;
924                 set = fdt->close_on_exec->fds_bits[j];
925                 if (!set)
926                         continue;
927                 fdt->close_on_exec->fds_bits[j] = 0;
928                 spin_unlock(&files->file_lock);
929                 for ( ; set ; i++,set >>= 1) {
930                         if (set & 1) {
931                                 sys_close(i);
932                         }
933                 }
934                 spin_lock(&files->file_lock);
935
936         }
937         spin_unlock(&files->file_lock);
938 }
939
940 char *get_task_comm(char *buf, struct task_struct *tsk)
941 {
942         /* buf must be at least sizeof(tsk->comm) in size */
943         task_lock(tsk);
944         strncpy(buf, tsk->comm, sizeof(tsk->comm));
945         task_unlock(tsk);
946         return buf;
947 }
948
949 void set_task_comm(struct task_struct *tsk, char *buf)
950 {
951         task_lock(tsk);
952         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
953         task_unlock(tsk);
954         perf_counter_comm(tsk);
955 }
956
957 int flush_old_exec(struct linux_binprm * bprm)
958 {
959         char * name;
960         int i, ch, retval;
961         char tcomm[sizeof(current->comm)];
962
963         /*
964          * Make sure we have a private signal table and that
965          * we are unassociated from the previous thread group.
966          */
967         retval = de_thread(current);
968         if (retval)
969                 goto out;
970
971         set_mm_exe_file(bprm->mm, bprm->file);
972
973         /*
974          * Release all of the old mmap stuff
975          */
976         retval = exec_mmap(bprm->mm);
977         if (retval)
978                 goto out;
979
980         bprm->mm = NULL;                /* We're using it now */
981
982         /* This is the point of no return */
983         current->sas_ss_sp = current->sas_ss_size = 0;
984
985         if (current_euid() == current_uid() && current_egid() == current_gid())
986                 set_dumpable(current->mm, 1);
987         else
988                 set_dumpable(current->mm, suid_dumpable);
989
990         name = bprm->filename;
991
992         /* Copies the binary name from after last slash */
993         for (i=0; (ch = *(name++)) != '\0';) {
994                 if (ch == '/')
995                         i = 0; /* overwrite what we wrote */
996                 else
997                         if (i < (sizeof(tcomm) - 1))
998                                 tcomm[i++] = ch;
999         }
1000         tcomm[i] = '\0';
1001         set_task_comm(current, tcomm);
1002
1003         current->flags &= ~PF_RANDOMIZE;
1004         flush_thread();
1005
1006         /* Set the new mm task size. We have to do that late because it may
1007          * depend on TIF_32BIT which is only updated in flush_thread() on
1008          * some architectures like powerpc
1009          */
1010         current->mm->task_size = TASK_SIZE;
1011
1012         /* install the new credentials */
1013         if (bprm->cred->uid != current_euid() ||
1014             bprm->cred->gid != current_egid()) {
1015                 current->pdeath_signal = 0;
1016         } else if (file_permission(bprm->file, MAY_READ) ||
1017                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
1018                 set_dumpable(current->mm, suid_dumpable);
1019         }
1020
1021         current->personality &= ~bprm->per_clear;
1022
1023         /*
1024          * Flush performance counters when crossing a
1025          * security domain:
1026          */
1027         if (!get_dumpable(current->mm))
1028                 perf_counter_exit_task(current);
1029
1030         /* An exec changes our domain. We are no longer part of the thread
1031            group */
1032
1033         current->self_exec_id++;
1034                         
1035         flush_signal_handlers(current, 0);
1036         flush_old_files(current->files);
1037
1038         return 0;
1039
1040 out:
1041         return retval;
1042 }
1043
1044 EXPORT_SYMBOL(flush_old_exec);
1045
1046 /*
1047  * install the new credentials for this executable
1048  */
1049 void install_exec_creds(struct linux_binprm *bprm)
1050 {
1051         security_bprm_committing_creds(bprm);
1052
1053         commit_creds(bprm->cred);
1054         bprm->cred = NULL;
1055
1056         /* cred_exec_mutex must be held at least to this point to prevent
1057          * ptrace_attach() from altering our determination of the task's
1058          * credentials; any time after this it may be unlocked */
1059
1060         security_bprm_committed_creds(bprm);
1061 }
1062 EXPORT_SYMBOL(install_exec_creds);
1063
1064 /*
1065  * determine how safe it is to execute the proposed program
1066  * - the caller must hold current->cred_exec_mutex to protect against
1067  *   PTRACE_ATTACH
1068  */
1069 int check_unsafe_exec(struct linux_binprm *bprm)
1070 {
1071         struct task_struct *p = current, *t;
1072         unsigned n_fs;
1073         int res = 0;
1074
1075         bprm->unsafe = tracehook_unsafe_exec(p);
1076
1077         n_fs = 1;
1078         write_lock(&p->fs->lock);
1079         rcu_read_lock();
1080         for (t = next_thread(p); t != p; t = next_thread(t)) {
1081                 if (t->fs == p->fs)
1082                         n_fs++;
1083         }
1084         rcu_read_unlock();
1085
1086         if (p->fs->users > n_fs) {
1087                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1088         } else {
1089                 res = -EAGAIN;
1090                 if (!p->fs->in_exec) {
1091                         p->fs->in_exec = 1;
1092                         res = 1;
1093                 }
1094         }
1095         write_unlock(&p->fs->lock);
1096
1097         return res;
1098 }
1099
1100 /* 
1101  * Fill the binprm structure from the inode. 
1102  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1103  *
1104  * This may be called multiple times for binary chains (scripts for example).
1105  */
1106 int prepare_binprm(struct linux_binprm *bprm)
1107 {
1108         umode_t mode;
1109         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1110         int retval;
1111
1112         mode = inode->i_mode;
1113         if (bprm->file->f_op == NULL)
1114                 return -EACCES;
1115
1116         /* clear any previous set[ug]id data from a previous binary */
1117         bprm->cred->euid = current_euid();
1118         bprm->cred->egid = current_egid();
1119
1120         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1121                 /* Set-uid? */
1122                 if (mode & S_ISUID) {
1123                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1124                         bprm->cred->euid = inode->i_uid;
1125                 }
1126
1127                 /* Set-gid? */
1128                 /*
1129                  * If setgid is set but no group execute bit then this
1130                  * is a candidate for mandatory locking, not a setgid
1131                  * executable.
1132                  */
1133                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1134                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1135                         bprm->cred->egid = inode->i_gid;
1136                 }
1137         }
1138
1139         /* fill in binprm security blob */
1140         retval = security_bprm_set_creds(bprm);
1141         if (retval)
1142                 return retval;
1143         bprm->cred_prepared = 1;
1144
1145         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1146         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1147 }
1148
1149 EXPORT_SYMBOL(prepare_binprm);
1150
1151 /*
1152  * Arguments are '\0' separated strings found at the location bprm->p
1153  * points to; chop off the first by relocating brpm->p to right after
1154  * the first '\0' encountered.
1155  */
1156 int remove_arg_zero(struct linux_binprm *bprm)
1157 {
1158         int ret = 0;
1159         unsigned long offset;
1160         char *kaddr;
1161         struct page *page;
1162
1163         if (!bprm->argc)
1164                 return 0;
1165
1166         do {
1167                 offset = bprm->p & ~PAGE_MASK;
1168                 page = get_arg_page(bprm, bprm->p, 0);
1169                 if (!page) {
1170                         ret = -EFAULT;
1171                         goto out;
1172                 }
1173                 kaddr = kmap_atomic(page, KM_USER0);
1174
1175                 for (; offset < PAGE_SIZE && kaddr[offset];
1176                                 offset++, bprm->p++)
1177                         ;
1178
1179                 kunmap_atomic(kaddr, KM_USER0);
1180                 put_arg_page(page);
1181
1182                 if (offset == PAGE_SIZE)
1183                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1184         } while (offset == PAGE_SIZE);
1185
1186         bprm->p++;
1187         bprm->argc--;
1188         ret = 0;
1189
1190 out:
1191         return ret;
1192 }
1193 EXPORT_SYMBOL(remove_arg_zero);
1194
1195 /*
1196  * cycle the list of binary formats handler, until one recognizes the image
1197  */
1198 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1199 {
1200         unsigned int depth = bprm->recursion_depth;
1201         int try,retval;
1202         struct linux_binfmt *fmt;
1203
1204         retval = security_bprm_check(bprm);
1205         if (retval)
1206                 return retval;
1207         retval = ima_bprm_check(bprm);
1208         if (retval)
1209                 return retval;
1210
1211         /* kernel module loader fixup */
1212         /* so we don't try to load run modprobe in kernel space. */
1213         set_fs(USER_DS);
1214
1215         retval = audit_bprm(bprm);
1216         if (retval)
1217                 return retval;
1218
1219         retval = -ENOENT;
1220         for (try=0; try<2; try++) {
1221                 read_lock(&binfmt_lock);
1222                 list_for_each_entry(fmt, &formats, lh) {
1223                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1224                         if (!fn)
1225                                 continue;
1226                         if (!try_module_get(fmt->module))
1227                                 continue;
1228                         read_unlock(&binfmt_lock);
1229                         retval = fn(bprm, regs);
1230                         /*
1231                          * Restore the depth counter to its starting value
1232                          * in this call, so we don't have to rely on every
1233                          * load_binary function to restore it on return.
1234                          */
1235                         bprm->recursion_depth = depth;
1236                         if (retval >= 0) {
1237                                 if (depth == 0)
1238                                         tracehook_report_exec(fmt, bprm, regs);
1239                                 put_binfmt(fmt);
1240                                 allow_write_access(bprm->file);
1241                                 if (bprm->file)
1242                                         fput(bprm->file);
1243                                 bprm->file = NULL;
1244                                 current->did_exec = 1;
1245                                 proc_exec_connector(current);
1246                                 return retval;
1247                         }
1248                         read_lock(&binfmt_lock);
1249                         put_binfmt(fmt);
1250                         if (retval != -ENOEXEC || bprm->mm == NULL)
1251                                 break;
1252                         if (!bprm->file) {
1253                                 read_unlock(&binfmt_lock);
1254                                 return retval;
1255                         }
1256                 }
1257                 read_unlock(&binfmt_lock);
1258                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1259                         break;
1260 #ifdef CONFIG_MODULES
1261                 } else {
1262 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1263                         if (printable(bprm->buf[0]) &&
1264                             printable(bprm->buf[1]) &&
1265                             printable(bprm->buf[2]) &&
1266                             printable(bprm->buf[3]))
1267                                 break; /* -ENOEXEC */
1268                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1269 #endif
1270                 }
1271         }
1272         return retval;
1273 }
1274
1275 EXPORT_SYMBOL(search_binary_handler);
1276
1277 void free_bprm(struct linux_binprm *bprm)
1278 {
1279         free_arg_pages(bprm);
1280         if (bprm->cred)
1281                 abort_creds(bprm->cred);
1282         kfree(bprm);
1283 }
1284
1285 /*
1286  * sys_execve() executes a new program.
1287  */
1288 int do_execve(char * filename,
1289         char __user *__user *argv,
1290         char __user *__user *envp,
1291         struct pt_regs * regs)
1292 {
1293         struct linux_binprm *bprm;
1294         struct file *file;
1295         struct files_struct *displaced;
1296         bool clear_in_exec;
1297         int retval;
1298
1299         retval = unshare_files(&displaced);
1300         if (retval)
1301                 goto out_ret;
1302
1303         retval = -ENOMEM;
1304         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1305         if (!bprm)
1306                 goto out_files;
1307
1308         retval = mutex_lock_interruptible(&current->cred_exec_mutex);
1309         if (retval < 0)
1310                 goto out_free;
1311         current->in_execve = 1;
1312
1313         retval = -ENOMEM;
1314         bprm->cred = prepare_exec_creds();
1315         if (!bprm->cred)
1316                 goto out_unlock;
1317
1318         retval = check_unsafe_exec(bprm);
1319         if (retval < 0)
1320                 goto out_unlock;
1321         clear_in_exec = retval;
1322
1323         file = open_exec(filename);
1324         retval = PTR_ERR(file);
1325         if (IS_ERR(file))
1326                 goto out_unmark;
1327
1328         sched_exec();
1329
1330         bprm->file = file;
1331         bprm->filename = filename;
1332         bprm->interp = filename;
1333
1334         retval = bprm_mm_init(bprm);
1335         if (retval)
1336                 goto out_file;
1337
1338         bprm->argc = count(argv, MAX_ARG_STRINGS);
1339         if ((retval = bprm->argc) < 0)
1340                 goto out;
1341
1342         bprm->envc = count(envp, MAX_ARG_STRINGS);
1343         if ((retval = bprm->envc) < 0)
1344                 goto out;
1345
1346         retval = prepare_binprm(bprm);
1347         if (retval < 0)
1348                 goto out;
1349
1350         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1351         if (retval < 0)
1352                 goto out;
1353
1354         bprm->exec = bprm->p;
1355         retval = copy_strings(bprm->envc, envp, bprm);
1356         if (retval < 0)
1357                 goto out;
1358
1359         retval = copy_strings(bprm->argc, argv, bprm);
1360         if (retval < 0)
1361                 goto out;
1362
1363         current->flags &= ~PF_KTHREAD;
1364         retval = search_binary_handler(bprm,regs);
1365         if (retval < 0)
1366                 goto out;
1367
1368         /* execve succeeded */
1369         current->fs->in_exec = 0;
1370         current->in_execve = 0;
1371         mutex_unlock(&current->cred_exec_mutex);
1372         acct_update_integrals(current);
1373         free_bprm(bprm);
1374         if (displaced)
1375                 put_files_struct(displaced);
1376         return retval;
1377
1378 out:
1379         if (bprm->mm)
1380                 mmput (bprm->mm);
1381
1382 out_file:
1383         if (bprm->file) {
1384                 allow_write_access(bprm->file);
1385                 fput(bprm->file);
1386         }
1387
1388 out_unmark:
1389         if (clear_in_exec)
1390                 current->fs->in_exec = 0;
1391
1392 out_unlock:
1393         current->in_execve = 0;
1394         mutex_unlock(&current->cred_exec_mutex);
1395
1396 out_free:
1397         free_bprm(bprm);
1398
1399 out_files:
1400         if (displaced)
1401                 reset_files_struct(displaced);
1402 out_ret:
1403         return retval;
1404 }
1405
1406 int set_binfmt(struct linux_binfmt *new)
1407 {
1408         struct linux_binfmt *old = current->binfmt;
1409
1410         if (new) {
1411                 if (!try_module_get(new->module))
1412                         return -1;
1413         }
1414         current->binfmt = new;
1415         if (old)
1416                 module_put(old->module);
1417         return 0;
1418 }
1419
1420 EXPORT_SYMBOL(set_binfmt);
1421
1422 /* format_corename will inspect the pattern parameter, and output a
1423  * name into corename, which must have space for at least
1424  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1425  */
1426 static int format_corename(char *corename, long signr)
1427 {
1428         const struct cred *cred = current_cred();
1429         const char *pat_ptr = core_pattern;
1430         int ispipe = (*pat_ptr == '|');
1431         char *out_ptr = corename;
1432         char *const out_end = corename + CORENAME_MAX_SIZE;
1433         int rc;
1434         int pid_in_pattern = 0;
1435
1436         /* Repeat as long as we have more pattern to process and more output
1437            space */
1438         while (*pat_ptr) {
1439                 if (*pat_ptr != '%') {
1440                         if (out_ptr == out_end)
1441                                 goto out;
1442                         *out_ptr++ = *pat_ptr++;
1443                 } else {
1444                         switch (*++pat_ptr) {
1445                         case 0:
1446                                 goto out;
1447                         /* Double percent, output one percent */
1448                         case '%':
1449                                 if (out_ptr == out_end)
1450                                         goto out;
1451                                 *out_ptr++ = '%';
1452                                 break;
1453                         /* pid */
1454                         case 'p':
1455                                 pid_in_pattern = 1;
1456                                 rc = snprintf(out_ptr, out_end - out_ptr,
1457                                               "%d", task_tgid_vnr(current));
1458                                 if (rc > out_end - out_ptr)
1459                                         goto out;
1460                                 out_ptr += rc;
1461                                 break;
1462                         /* uid */
1463                         case 'u':
1464                                 rc = snprintf(out_ptr, out_end - out_ptr,
1465                                               "%d", cred->uid);
1466                                 if (rc > out_end - out_ptr)
1467                                         goto out;
1468                                 out_ptr += rc;
1469                                 break;
1470                         /* gid */
1471                         case 'g':
1472                                 rc = snprintf(out_ptr, out_end - out_ptr,
1473                                               "%d", cred->gid);
1474                                 if (rc > out_end - out_ptr)
1475                                         goto out;
1476                                 out_ptr += rc;
1477                                 break;
1478                         /* signal that caused the coredump */
1479                         case 's':
1480                                 rc = snprintf(out_ptr, out_end - out_ptr,
1481                                               "%ld", signr);
1482                                 if (rc > out_end - out_ptr)
1483                                         goto out;
1484                                 out_ptr += rc;
1485                                 break;
1486                         /* UNIX time of coredump */
1487                         case 't': {
1488                                 struct timeval tv;
1489                                 do_gettimeofday(&tv);
1490                                 rc = snprintf(out_ptr, out_end - out_ptr,
1491                                               "%lu", tv.tv_sec);
1492                                 if (rc > out_end - out_ptr)
1493                                         goto out;
1494                                 out_ptr += rc;
1495                                 break;
1496                         }
1497                         /* hostname */
1498                         case 'h':
1499                                 down_read(&uts_sem);
1500                                 rc = snprintf(out_ptr, out_end - out_ptr,
1501                                               "%s", utsname()->nodename);
1502                                 up_read(&uts_sem);
1503                                 if (rc > out_end - out_ptr)
1504                                         goto out;
1505                                 out_ptr += rc;
1506                                 break;
1507                         /* executable */
1508                         case 'e':
1509                                 rc = snprintf(out_ptr, out_end - out_ptr,
1510                                               "%s", current->comm);
1511                                 if (rc > out_end - out_ptr)
1512                                         goto out;
1513                                 out_ptr += rc;
1514                                 break;
1515                         /* core limit size */
1516                         case 'c':
1517                                 rc = snprintf(out_ptr, out_end - out_ptr,
1518                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1519                                 if (rc > out_end - out_ptr)
1520                                         goto out;
1521                                 out_ptr += rc;
1522                                 break;
1523                         default:
1524                                 break;
1525                         }
1526                         ++pat_ptr;
1527                 }
1528         }
1529         /* Backward compatibility with core_uses_pid:
1530          *
1531          * If core_pattern does not include a %p (as is the default)
1532          * and core_uses_pid is set, then .%pid will be appended to
1533          * the filename. Do not do this for piped commands. */
1534         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1535                 rc = snprintf(out_ptr, out_end - out_ptr,
1536                               ".%d", task_tgid_vnr(current));
1537                 if (rc > out_end - out_ptr)
1538                         goto out;
1539                 out_ptr += rc;
1540         }
1541 out:
1542         *out_ptr = 0;
1543         return ispipe;
1544 }
1545
1546 static int zap_process(struct task_struct *start)
1547 {
1548         struct task_struct *t;
1549         int nr = 0;
1550
1551         start->signal->flags = SIGNAL_GROUP_EXIT;
1552         start->signal->group_stop_count = 0;
1553
1554         t = start;
1555         do {
1556                 if (t != current && t->mm) {
1557                         sigaddset(&t->pending.signal, SIGKILL);
1558                         signal_wake_up(t, 1);
1559                         nr++;
1560                 }
1561         } while_each_thread(start, t);
1562
1563         return nr;
1564 }
1565
1566 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1567                                 struct core_state *core_state, int exit_code)
1568 {
1569         struct task_struct *g, *p;
1570         unsigned long flags;
1571         int nr = -EAGAIN;
1572
1573         spin_lock_irq(&tsk->sighand->siglock);
1574         if (!signal_group_exit(tsk->signal)) {
1575                 mm->core_state = core_state;
1576                 tsk->signal->group_exit_code = exit_code;
1577                 nr = zap_process(tsk);
1578         }
1579         spin_unlock_irq(&tsk->sighand->siglock);
1580         if (unlikely(nr < 0))
1581                 return nr;
1582
1583         if (atomic_read(&mm->mm_users) == nr + 1)
1584                 goto done;
1585         /*
1586          * We should find and kill all tasks which use this mm, and we should
1587          * count them correctly into ->nr_threads. We don't take tasklist
1588          * lock, but this is safe wrt:
1589          *
1590          * fork:
1591          *      None of sub-threads can fork after zap_process(leader). All
1592          *      processes which were created before this point should be
1593          *      visible to zap_threads() because copy_process() adds the new
1594          *      process to the tail of init_task.tasks list, and lock/unlock
1595          *      of ->siglock provides a memory barrier.
1596          *
1597          * do_exit:
1598          *      The caller holds mm->mmap_sem. This means that the task which
1599          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1600          *      its ->mm.
1601          *
1602          * de_thread:
1603          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1604          *      we must see either old or new leader, this does not matter.
1605          *      However, it can change p->sighand, so lock_task_sighand(p)
1606          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1607          *      it can't fail.
1608          *
1609          *      Note also that "g" can be the old leader with ->mm == NULL
1610          *      and already unhashed and thus removed from ->thread_group.
1611          *      This is OK, __unhash_process()->list_del_rcu() does not
1612          *      clear the ->next pointer, we will find the new leader via
1613          *      next_thread().
1614          */
1615         rcu_read_lock();
1616         for_each_process(g) {
1617                 if (g == tsk->group_leader)
1618                         continue;
1619                 if (g->flags & PF_KTHREAD)
1620                         continue;
1621                 p = g;
1622                 do {
1623                         if (p->mm) {
1624                                 if (unlikely(p->mm == mm)) {
1625                                         lock_task_sighand(p, &flags);
1626                                         nr += zap_process(p);
1627                                         unlock_task_sighand(p, &flags);
1628                                 }
1629                                 break;
1630                         }
1631                 } while_each_thread(g, p);
1632         }
1633         rcu_read_unlock();
1634 done:
1635         atomic_set(&core_state->nr_threads, nr);
1636         return nr;
1637 }
1638
1639 static int coredump_wait(int exit_code, struct core_state *core_state)
1640 {
1641         struct task_struct *tsk = current;
1642         struct mm_struct *mm = tsk->mm;
1643         struct completion *vfork_done;
1644         int core_waiters;
1645
1646         init_completion(&core_state->startup);
1647         core_state->dumper.task = tsk;
1648         core_state->dumper.next = NULL;
1649         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1650         up_write(&mm->mmap_sem);
1651
1652         if (unlikely(core_waiters < 0))
1653                 goto fail;
1654
1655         /*
1656          * Make sure nobody is waiting for us to release the VM,
1657          * otherwise we can deadlock when we wait on each other
1658          */
1659         vfork_done = tsk->vfork_done;
1660         if (vfork_done) {
1661                 tsk->vfork_done = NULL;
1662                 complete(vfork_done);
1663         }
1664
1665         if (core_waiters)
1666                 wait_for_completion(&core_state->startup);
1667 fail:
1668         return core_waiters;
1669 }
1670
1671 static void coredump_finish(struct mm_struct *mm)
1672 {
1673         struct core_thread *curr, *next;
1674         struct task_struct *task;
1675
1676         next = mm->core_state->dumper.next;
1677         while ((curr = next) != NULL) {
1678                 next = curr->next;
1679                 task = curr->task;
1680                 /*
1681                  * see exit_mm(), curr->task must not see
1682                  * ->task == NULL before we read ->next.
1683                  */
1684                 smp_mb();
1685                 curr->task = NULL;
1686                 wake_up_process(task);
1687         }
1688
1689         mm->core_state = NULL;
1690 }
1691
1692 /*
1693  * set_dumpable converts traditional three-value dumpable to two flags and
1694  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1695  * these bits are not changed atomically.  So get_dumpable can observe the
1696  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1697  * return either old dumpable or new one by paying attention to the order of
1698  * modifying the bits.
1699  *
1700  * dumpable |   mm->flags (binary)
1701  * old  new | initial interim  final
1702  * ---------+-----------------------
1703  *  0    1  |   00      01      01
1704  *  0    2  |   00      10(*)   11
1705  *  1    0  |   01      00      00
1706  *  1    2  |   01      11      11
1707  *  2    0  |   11      10(*)   00
1708  *  2    1  |   11      11      01
1709  *
1710  * (*) get_dumpable regards interim value of 10 as 11.
1711  */
1712 void set_dumpable(struct mm_struct *mm, int value)
1713 {
1714         switch (value) {
1715         case 0:
1716                 clear_bit(MMF_DUMPABLE, &mm->flags);
1717                 smp_wmb();
1718                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1719                 break;
1720         case 1:
1721                 set_bit(MMF_DUMPABLE, &mm->flags);
1722                 smp_wmb();
1723                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1724                 break;
1725         case 2:
1726                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1727                 smp_wmb();
1728                 set_bit(MMF_DUMPABLE, &mm->flags);
1729                 break;
1730         }
1731 }
1732
1733 int get_dumpable(struct mm_struct *mm)
1734 {
1735         int ret;
1736
1737         ret = mm->flags & 0x3;
1738         return (ret >= 2) ? 2 : ret;
1739 }
1740
1741 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1742 {
1743         struct core_state core_state;
1744         char corename[CORENAME_MAX_SIZE + 1];
1745         struct mm_struct *mm = current->mm;
1746         struct linux_binfmt * binfmt;
1747         struct inode * inode;
1748         struct file * file;
1749         const struct cred *old_cred;
1750         struct cred *cred;
1751         int retval = 0;
1752         int flag = 0;
1753         int ispipe = 0;
1754         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1755         char **helper_argv = NULL;
1756         int helper_argc = 0;
1757         char *delimit;
1758
1759         audit_core_dumps(signr);
1760
1761         binfmt = current->binfmt;
1762         if (!binfmt || !binfmt->core_dump)
1763                 goto fail;
1764
1765         cred = prepare_creds();
1766         if (!cred) {
1767                 retval = -ENOMEM;
1768                 goto fail;
1769         }
1770
1771         down_write(&mm->mmap_sem);
1772         /*
1773          * If another thread got here first, or we are not dumpable, bail out.
1774          */
1775         if (mm->core_state || !get_dumpable(mm)) {
1776                 up_write(&mm->mmap_sem);
1777                 put_cred(cred);
1778                 goto fail;
1779         }
1780
1781         /*
1782          *      We cannot trust fsuid as being the "true" uid of the
1783          *      process nor do we know its entire history. We only know it
1784          *      was tainted so we dump it as root in mode 2.
1785          */
1786         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1787                 flag = O_EXCL;          /* Stop rewrite attacks */
1788                 cred->fsuid = 0;        /* Dump root private */
1789         }
1790
1791         retval = coredump_wait(exit_code, &core_state);
1792         if (retval < 0) {
1793                 put_cred(cred);
1794                 goto fail;
1795         }
1796
1797         old_cred = override_creds(cred);
1798
1799         /*
1800          * Clear any false indication of pending signals that might
1801          * be seen by the filesystem code called to write the core file.
1802          */
1803         clear_thread_flag(TIF_SIGPENDING);
1804
1805         /*
1806          * lock_kernel() because format_corename() is controlled by sysctl, which
1807          * uses lock_kernel()
1808          */
1809         lock_kernel();
1810         ispipe = format_corename(corename, signr);
1811         unlock_kernel();
1812         /*
1813          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1814          * to a pipe.  Since we're not writing directly to the filesystem
1815          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1816          * created unless the pipe reader choses to write out the core file
1817          * at which point file size limits and permissions will be imposed
1818          * as it does with any other process
1819          */
1820         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1821                 goto fail_unlock;
1822
1823         if (ispipe) {
1824                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1825                 if (!helper_argv) {
1826                         printk(KERN_WARNING "%s failed to allocate memory\n",
1827                                __func__);
1828                         goto fail_unlock;
1829                 }
1830                 /* Terminate the string before the first option */
1831                 delimit = strchr(corename, ' ');
1832                 if (delimit)
1833                         *delimit = '\0';
1834                 delimit = strrchr(helper_argv[0], '/');
1835                 if (delimit)
1836                         delimit++;
1837                 else
1838                         delimit = helper_argv[0];
1839                 if (!strcmp(delimit, current->comm)) {
1840                         printk(KERN_NOTICE "Recursive core dump detected, "
1841                                         "aborting\n");
1842                         goto fail_unlock;
1843                 }
1844
1845                 core_limit = RLIM_INFINITY;
1846
1847                 /* SIGPIPE can happen, but it's just never processed */
1848                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1849                                 &file)) {
1850                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1851                                corename);
1852                         goto fail_unlock;
1853                 }
1854         } else
1855                 file = filp_open(corename,
1856                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1857                                  0600);
1858         if (IS_ERR(file))
1859                 goto fail_unlock;
1860         inode = file->f_path.dentry->d_inode;
1861         if (inode->i_nlink > 1)
1862                 goto close_fail;        /* multiple links - don't dump */
1863         if (!ispipe && d_unhashed(file->f_path.dentry))
1864                 goto close_fail;
1865
1866         /* AK: actually i see no reason to not allow this for named pipes etc.,
1867            but keep the previous behaviour for now. */
1868         if (!ispipe && !S_ISREG(inode->i_mode))
1869                 goto close_fail;
1870         /*
1871          * Dont allow local users get cute and trick others to coredump
1872          * into their pre-created files:
1873          */
1874         if (inode->i_uid != current_fsuid())
1875                 goto close_fail;
1876         if (!file->f_op)
1877                 goto close_fail;
1878         if (!file->f_op->write)
1879                 goto close_fail;
1880         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1881                 goto close_fail;
1882
1883         retval = binfmt->core_dump(signr, regs, file, core_limit);
1884
1885         if (retval)
1886                 current->signal->group_exit_code |= 0x80;
1887 close_fail:
1888         filp_close(file, NULL);
1889 fail_unlock:
1890         if (helper_argv)
1891                 argv_free(helper_argv);
1892
1893         revert_creds(old_cred);
1894         put_cred(cred);
1895         coredump_finish(mm);
1896 fail:
1897         return;
1898 }