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