dc022dd15d51222e9ff159c39e93adb28f0682cd
[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/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 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 > 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         vma->vm_end = STACK_TOP_MAX;
246         vma->vm_start = vma->vm_end - PAGE_SIZE;
247         vma->vm_flags = VM_STACK_FLAGS;
248         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
249         err = insert_vm_struct(mm, vma);
250         if (err)
251                 goto err;
252
253         mm->stack_vm = mm->total_vm = 1;
254         up_write(&mm->mmap_sem);
255         bprm->p = vma->vm_end - sizeof(void *);
256         return 0;
257 err:
258         up_write(&mm->mmap_sem);
259         bprm->vma = NULL;
260         kmem_cache_free(vm_area_cachep, vma);
261         return err;
262 }
263
264 static bool valid_arg_len(struct linux_binprm *bprm, long len)
265 {
266         return len <= MAX_ARG_STRLEN;
267 }
268
269 #else
270
271 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
272                 int write)
273 {
274         struct page *page;
275
276         page = bprm->page[pos / PAGE_SIZE];
277         if (!page && write) {
278                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
279                 if (!page)
280                         return NULL;
281                 bprm->page[pos / PAGE_SIZE] = page;
282         }
283
284         return page;
285 }
286
287 static void put_arg_page(struct page *page)
288 {
289 }
290
291 static void free_arg_page(struct linux_binprm *bprm, int i)
292 {
293         if (bprm->page[i]) {
294                 __free_page(bprm->page[i]);
295                 bprm->page[i] = NULL;
296         }
297 }
298
299 static void free_arg_pages(struct linux_binprm *bprm)
300 {
301         int i;
302
303         for (i = 0; i < MAX_ARG_PAGES; i++)
304                 free_arg_page(bprm, i);
305 }
306
307 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
308                 struct page *page)
309 {
310 }
311
312 static int __bprm_mm_init(struct linux_binprm *bprm)
313 {
314         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
315         return 0;
316 }
317
318 static bool valid_arg_len(struct linux_binprm *bprm, long len)
319 {
320         return len <= bprm->p;
321 }
322
323 #endif /* CONFIG_MMU */
324
325 /*
326  * Create a new mm_struct and populate it with a temporary stack
327  * vm_area_struct.  We don't have enough context at this point to set the stack
328  * flags, permissions, and offset, so we use temporary values.  We'll update
329  * them later in setup_arg_pages().
330  */
331 int bprm_mm_init(struct linux_binprm *bprm)
332 {
333         int err;
334         struct mm_struct *mm = NULL;
335
336         bprm->mm = mm = mm_alloc();
337         err = -ENOMEM;
338         if (!mm)
339                 goto err;
340
341         err = init_new_context(current, mm);
342         if (err)
343                 goto err;
344
345         err = __bprm_mm_init(bprm);
346         if (err)
347                 goto err;
348
349         return 0;
350
351 err:
352         if (mm) {
353                 bprm->mm = NULL;
354                 mmdrop(mm);
355         }
356
357         return err;
358 }
359
360 /*
361  * count() counts the number of strings in array ARGV.
362  */
363 static int count(char __user * __user * argv, int max)
364 {
365         int i = 0;
366
367         if (argv != NULL) {
368                 for (;;) {
369                         char __user * p;
370
371                         if (get_user(p, argv))
372                                 return -EFAULT;
373                         if (!p)
374                                 break;
375                         argv++;
376                         if (i++ >= max)
377                                 return -E2BIG;
378                         cond_resched();
379                 }
380         }
381         return i;
382 }
383
384 /*
385  * 'copy_strings()' copies argument/environment strings from the old
386  * processes's memory to the new process's stack.  The call to get_user_pages()
387  * ensures the destination page is created and not swapped out.
388  */
389 static int copy_strings(int argc, char __user * __user * argv,
390                         struct linux_binprm *bprm)
391 {
392         struct page *kmapped_page = NULL;
393         char *kaddr = NULL;
394         unsigned long kpos = 0;
395         int ret;
396
397         while (argc-- > 0) {
398                 char __user *str;
399                 int len;
400                 unsigned long pos;
401
402                 if (get_user(str, argv+argc) ||
403                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
404                         ret = -EFAULT;
405                         goto out;
406                 }
407
408                 if (!valid_arg_len(bprm, len)) {
409                         ret = -E2BIG;
410                         goto out;
411                 }
412
413                 /* We're going to work our way backwords. */
414                 pos = bprm->p;
415                 str += len;
416                 bprm->p -= len;
417
418                 while (len > 0) {
419                         int offset, bytes_to_copy;
420
421                         offset = pos % PAGE_SIZE;
422                         if (offset == 0)
423                                 offset = PAGE_SIZE;
424
425                         bytes_to_copy = offset;
426                         if (bytes_to_copy > len)
427                                 bytes_to_copy = len;
428
429                         offset -= bytes_to_copy;
430                         pos -= bytes_to_copy;
431                         str -= bytes_to_copy;
432                         len -= bytes_to_copy;
433
434                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
435                                 struct page *page;
436
437                                 page = get_arg_page(bprm, pos, 1);
438                                 if (!page) {
439                                         ret = -E2BIG;
440                                         goto out;
441                                 }
442
443                                 if (kmapped_page) {
444                                         flush_kernel_dcache_page(kmapped_page);
445                                         kunmap(kmapped_page);
446                                         put_arg_page(kmapped_page);
447                                 }
448                                 kmapped_page = page;
449                                 kaddr = kmap(kmapped_page);
450                                 kpos = pos & PAGE_MASK;
451                                 flush_arg_page(bprm, kpos, kmapped_page);
452                         }
453                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
454                                 ret = -EFAULT;
455                                 goto out;
456                         }
457                 }
458         }
459         ret = 0;
460 out:
461         if (kmapped_page) {
462                 flush_kernel_dcache_page(kmapped_page);
463                 kunmap(kmapped_page);
464                 put_arg_page(kmapped_page);
465         }
466         return ret;
467 }
468
469 /*
470  * Like copy_strings, but get argv and its values from kernel memory.
471  */
472 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
473 {
474         int r;
475         mm_segment_t oldfs = get_fs();
476         set_fs(KERNEL_DS);
477         r = copy_strings(argc, (char __user * __user *)argv, bprm);
478         set_fs(oldfs);
479         return r;
480 }
481 EXPORT_SYMBOL(copy_strings_kernel);
482
483 #ifdef CONFIG_MMU
484
485 /*
486  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
487  * the binfmt code determines where the new stack should reside, we shift it to
488  * its final location.  The process proceeds as follows:
489  *
490  * 1) Use shift to calculate the new vma endpoints.
491  * 2) Extend vma to cover both the old and new ranges.  This ensures the
492  *    arguments passed to subsequent functions are consistent.
493  * 3) Move vma's page tables to the new range.
494  * 4) Free up any cleared pgd range.
495  * 5) Shrink the vma to cover only the new range.
496  */
497 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
498 {
499         struct mm_struct *mm = vma->vm_mm;
500         unsigned long old_start = vma->vm_start;
501         unsigned long old_end = vma->vm_end;
502         unsigned long length = old_end - old_start;
503         unsigned long new_start = old_start - shift;
504         unsigned long new_end = old_end - shift;
505         struct mmu_gather *tlb;
506
507         BUG_ON(new_start > new_end);
508
509         /*
510          * ensure there are no vmas between where we want to go
511          * and where we are
512          */
513         if (vma != find_vma(mm, new_start))
514                 return -EFAULT;
515
516         /*
517          * cover the whole range: [new_start, old_end)
518          */
519         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
520
521         /*
522          * move the page tables downwards, on failure we rely on
523          * process cleanup to remove whatever mess we made.
524          */
525         if (length != move_page_tables(vma, old_start,
526                                        vma, new_start, length))
527                 return -ENOMEM;
528
529         lru_add_drain();
530         tlb = tlb_gather_mmu(mm, 0);
531         if (new_end > old_start) {
532                 /*
533                  * when the old and new regions overlap clear from new_end.
534                  */
535                 free_pgd_range(tlb, new_end, old_end, new_end,
536                         vma->vm_next ? vma->vm_next->vm_start : 0);
537         } else {
538                 /*
539                  * otherwise, clean from old_start; this is done to not touch
540                  * the address space in [new_end, old_start) some architectures
541                  * have constraints on va-space that make this illegal (IA64) -
542                  * for the others its just a little faster.
543                  */
544                 free_pgd_range(tlb, old_start, old_end, new_end,
545                         vma->vm_next ? vma->vm_next->vm_start : 0);
546         }
547         tlb_finish_mmu(tlb, new_end, old_end);
548
549         /*
550          * shrink the vma to just the new range.
551          */
552         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
553
554         return 0;
555 }
556
557 #define EXTRA_STACK_VM_PAGES    20      /* random */
558
559 /*
560  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
561  * the stack is optionally relocated, and some extra space is added.
562  */
563 int setup_arg_pages(struct linux_binprm *bprm,
564                     unsigned long stack_top,
565                     int executable_stack)
566 {
567         unsigned long ret;
568         unsigned long stack_shift;
569         struct mm_struct *mm = current->mm;
570         struct vm_area_struct *vma = bprm->vma;
571         struct vm_area_struct *prev = NULL;
572         unsigned long vm_flags;
573         unsigned long stack_base;
574
575 #ifdef CONFIG_STACK_GROWSUP
576         /* Limit stack size to 1GB */
577         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
578         if (stack_base > (1 << 30))
579                 stack_base = 1 << 30;
580
581         /* Make sure we didn't let the argument array grow too large. */
582         if (vma->vm_end - vma->vm_start > stack_base)
583                 return -ENOMEM;
584
585         stack_base = PAGE_ALIGN(stack_top - stack_base);
586
587         stack_shift = vma->vm_start - stack_base;
588         mm->arg_start = bprm->p - stack_shift;
589         bprm->p = vma->vm_end - stack_shift;
590 #else
591         stack_top = arch_align_stack(stack_top);
592         stack_top = PAGE_ALIGN(stack_top);
593         stack_shift = vma->vm_end - stack_top;
594
595         bprm->p -= stack_shift;
596         mm->arg_start = bprm->p;
597 #endif
598
599         if (bprm->loader)
600                 bprm->loader -= stack_shift;
601         bprm->exec -= stack_shift;
602
603         down_write(&mm->mmap_sem);
604         vm_flags = VM_STACK_FLAGS;
605
606         /*
607          * Adjust stack execute permissions; explicitly enable for
608          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
609          * (arch default) otherwise.
610          */
611         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
612                 vm_flags |= VM_EXEC;
613         else if (executable_stack == EXSTACK_DISABLE_X)
614                 vm_flags &= ~VM_EXEC;
615         vm_flags |= mm->def_flags;
616
617         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
618                         vm_flags);
619         if (ret)
620                 goto out_unlock;
621         BUG_ON(prev != vma);
622
623         /* Move stack pages down in memory. */
624         if (stack_shift) {
625                 ret = shift_arg_pages(vma, stack_shift);
626                 if (ret) {
627                         up_write(&mm->mmap_sem);
628                         return ret;
629                 }
630         }
631
632 #ifdef CONFIG_STACK_GROWSUP
633         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
634 #else
635         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
636 #endif
637         ret = expand_stack(vma, stack_base);
638         if (ret)
639                 ret = -EFAULT;
640
641 out_unlock:
642         up_write(&mm->mmap_sem);
643         return 0;
644 }
645 EXPORT_SYMBOL(setup_arg_pages);
646
647 #endif /* CONFIG_MMU */
648
649 struct file *open_exec(const char *name)
650 {
651         struct file *file;
652         int err;
653
654         file = do_filp_open(AT_FDCWD, name,
655                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
656                                 MAY_EXEC | MAY_OPEN);
657         if (IS_ERR(file))
658                 goto out;
659
660         err = -EACCES;
661         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
662                 goto exit;
663
664         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
665                 goto exit;
666
667         fsnotify_open(file->f_path.dentry);
668
669         err = deny_write_access(file);
670         if (err)
671                 goto exit;
672
673 out:
674         return file;
675
676 exit:
677         fput(file);
678         return ERR_PTR(err);
679 }
680 EXPORT_SYMBOL(open_exec);
681
682 int kernel_read(struct file *file, loff_t offset,
683                 char *addr, unsigned long count)
684 {
685         mm_segment_t old_fs;
686         loff_t pos = offset;
687         int result;
688
689         old_fs = get_fs();
690         set_fs(get_ds());
691         /* The cast to a user pointer is valid due to the set_fs() */
692         result = vfs_read(file, (void __user *)addr, count, &pos);
693         set_fs(old_fs);
694         return result;
695 }
696
697 EXPORT_SYMBOL(kernel_read);
698
699 static int exec_mmap(struct mm_struct *mm)
700 {
701         struct task_struct *tsk;
702         struct mm_struct * old_mm, *active_mm;
703
704         /* Notify parent that we're no longer interested in the old VM */
705         tsk = current;
706         old_mm = current->mm;
707         mm_release(tsk, old_mm);
708
709         if (old_mm) {
710                 /*
711                  * Make sure that if there is a core dump in progress
712                  * for the old mm, we get out and die instead of going
713                  * through with the exec.  We must hold mmap_sem around
714                  * checking core_state and changing tsk->mm.
715                  */
716                 down_read(&old_mm->mmap_sem);
717                 if (unlikely(old_mm->core_state)) {
718                         up_read(&old_mm->mmap_sem);
719                         return -EINTR;
720                 }
721         }
722         task_lock(tsk);
723         active_mm = tsk->active_mm;
724         tsk->mm = mm;
725         tsk->active_mm = mm;
726         activate_mm(active_mm, mm);
727         task_unlock(tsk);
728         arch_pick_mmap_layout(mm);
729         if (old_mm) {
730                 up_read(&old_mm->mmap_sem);
731                 BUG_ON(active_mm != old_mm);
732                 mm_update_next_owner(old_mm);
733                 mmput(old_mm);
734                 return 0;
735         }
736         mmdrop(active_mm);
737         return 0;
738 }
739
740 /*
741  * This function makes sure the current process has its own signal table,
742  * so that flush_signal_handlers can later reset the handlers without
743  * disturbing other processes.  (Other processes might share the signal
744  * table via the CLONE_SIGHAND option to clone().)
745  */
746 static int de_thread(struct task_struct *tsk)
747 {
748         struct signal_struct *sig = tsk->signal;
749         struct sighand_struct *oldsighand = tsk->sighand;
750         spinlock_t *lock = &oldsighand->siglock;
751         int count;
752
753         if (thread_group_empty(tsk))
754                 goto no_thread_group;
755
756         /*
757          * Kill all other threads in the thread group.
758          */
759         spin_lock_irq(lock);
760         if (signal_group_exit(sig)) {
761                 /*
762                  * Another group action in progress, just
763                  * return so that the signal is processed.
764                  */
765                 spin_unlock_irq(lock);
766                 return -EAGAIN;
767         }
768         sig->group_exit_task = tsk;
769         zap_other_threads(tsk);
770
771         /* Account for the thread group leader hanging around: */
772         count = thread_group_leader(tsk) ? 1 : 2;
773         sig->notify_count = count;
774         while (atomic_read(&sig->count) > count) {
775                 __set_current_state(TASK_UNINTERRUPTIBLE);
776                 spin_unlock_irq(lock);
777                 schedule();
778                 spin_lock_irq(lock);
779         }
780         spin_unlock_irq(lock);
781
782         /*
783          * At this point all other threads have exited, all we have to
784          * do is to wait for the thread group leader to become inactive,
785          * and to assume its PID:
786          */
787         if (!thread_group_leader(tsk)) {
788                 struct task_struct *leader = tsk->group_leader;
789
790                 sig->notify_count = -1; /* for exit_notify() */
791                 for (;;) {
792                         write_lock_irq(&tasklist_lock);
793                         if (likely(leader->exit_state))
794                                 break;
795                         __set_current_state(TASK_UNINTERRUPTIBLE);
796                         write_unlock_irq(&tasklist_lock);
797                         schedule();
798                 }
799
800                 /*
801                  * The only record we have of the real-time age of a
802                  * process, regardless of execs it's done, is start_time.
803                  * All the past CPU time is accumulated in signal_struct
804                  * from sister threads now dead.  But in this non-leader
805                  * exec, nothing survives from the original leader thread,
806                  * whose birth marks the true age of this process now.
807                  * When we take on its identity by switching to its PID, we
808                  * also take its birthdate (always earlier than our own).
809                  */
810                 tsk->start_time = leader->start_time;
811
812                 BUG_ON(!same_thread_group(leader, tsk));
813                 BUG_ON(has_group_leader_pid(tsk));
814                 /*
815                  * An exec() starts a new thread group with the
816                  * TGID of the previous thread group. Rehash the
817                  * two threads with a switched PID, and release
818                  * the former thread group leader:
819                  */
820
821                 /* Become a process group leader with the old leader's pid.
822                  * The old leader becomes a thread of the this thread group.
823                  * Note: The old leader also uses this pid until release_task
824                  *       is called.  Odd but simple and correct.
825                  */
826                 detach_pid(tsk, PIDTYPE_PID);
827                 tsk->pid = leader->pid;
828                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
829                 transfer_pid(leader, tsk, PIDTYPE_PGID);
830                 transfer_pid(leader, tsk, PIDTYPE_SID);
831                 list_replace_rcu(&leader->tasks, &tsk->tasks);
832
833                 tsk->group_leader = tsk;
834                 leader->group_leader = tsk;
835
836                 tsk->exit_signal = SIGCHLD;
837
838                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
839                 leader->exit_state = EXIT_DEAD;
840                 write_unlock_irq(&tasklist_lock);
841
842                 release_task(leader);
843         }
844
845         sig->group_exit_task = NULL;
846         sig->notify_count = 0;
847
848 no_thread_group:
849         if (current->mm)
850                 setmax_mm_hiwater_rss(&sig->maxrss, current->mm);
851
852         exit_itimers(sig);
853         flush_itimer_signals();
854
855         if (atomic_read(&oldsighand->count) != 1) {
856                 struct sighand_struct *newsighand;
857                 /*
858                  * This ->sighand is shared with the CLONE_SIGHAND
859                  * but not CLONE_THREAD task, switch to the new one.
860                  */
861                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
862                 if (!newsighand)
863                         return -ENOMEM;
864
865                 atomic_set(&newsighand->count, 1);
866                 memcpy(newsighand->action, oldsighand->action,
867                        sizeof(newsighand->action));
868
869                 write_lock_irq(&tasklist_lock);
870                 spin_lock(&oldsighand->siglock);
871                 rcu_assign_pointer(tsk->sighand, newsighand);
872                 spin_unlock(&oldsighand->siglock);
873                 write_unlock_irq(&tasklist_lock);
874
875                 __cleanup_sighand(oldsighand);
876         }
877
878         BUG_ON(!thread_group_leader(tsk));
879         return 0;
880 }
881
882 /*
883  * These functions flushes out all traces of the currently running executable
884  * so that a new one can be started
885  */
886 static void flush_old_files(struct files_struct * files)
887 {
888         long j = -1;
889         struct fdtable *fdt;
890
891         spin_lock(&files->file_lock);
892         for (;;) {
893                 unsigned long set, i;
894
895                 j++;
896                 i = j * __NFDBITS;
897                 fdt = files_fdtable(files);
898                 if (i >= fdt->max_fds)
899                         break;
900                 set = fdt->close_on_exec->fds_bits[j];
901                 if (!set)
902                         continue;
903                 fdt->close_on_exec->fds_bits[j] = 0;
904                 spin_unlock(&files->file_lock);
905                 for ( ; set ; i++,set >>= 1) {
906                         if (set & 1) {
907                                 sys_close(i);
908                         }
909                 }
910                 spin_lock(&files->file_lock);
911
912         }
913         spin_unlock(&files->file_lock);
914 }
915
916 char *get_task_comm(char *buf, struct task_struct *tsk)
917 {
918         /* buf must be at least sizeof(tsk->comm) in size */
919         task_lock(tsk);
920         strncpy(buf, tsk->comm, sizeof(tsk->comm));
921         task_unlock(tsk);
922         return buf;
923 }
924
925 void set_task_comm(struct task_struct *tsk, char *buf)
926 {
927         task_lock(tsk);
928         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
929         task_unlock(tsk);
930         perf_event_comm(tsk);
931 }
932
933 int flush_old_exec(struct linux_binprm * bprm)
934 {
935         char * name;
936         int i, ch, retval;
937         char tcomm[sizeof(current->comm)];
938
939         /*
940          * Make sure we have a private signal table and that
941          * we are unassociated from the previous thread group.
942          */
943         retval = de_thread(current);
944         if (retval)
945                 goto out;
946
947         set_mm_exe_file(bprm->mm, bprm->file);
948
949         /*
950          * Release all of the old mmap stuff
951          */
952         retval = exec_mmap(bprm->mm);
953         if (retval)
954                 goto out;
955
956         bprm->mm = NULL;                /* We're using it now */
957
958         /* This is the point of no return */
959         current->sas_ss_sp = current->sas_ss_size = 0;
960
961         if (current_euid() == current_uid() && current_egid() == current_gid())
962                 set_dumpable(current->mm, 1);
963         else
964                 set_dumpable(current->mm, suid_dumpable);
965
966         name = bprm->filename;
967
968         /* Copies the binary name from after last slash */
969         for (i=0; (ch = *(name++)) != '\0';) {
970                 if (ch == '/')
971                         i = 0; /* overwrite what we wrote */
972                 else
973                         if (i < (sizeof(tcomm) - 1))
974                                 tcomm[i++] = ch;
975         }
976         tcomm[i] = '\0';
977         set_task_comm(current, tcomm);
978
979         current->flags &= ~PF_RANDOMIZE;
980         flush_thread();
981
982         /* Set the new mm task size. We have to do that late because it may
983          * depend on TIF_32BIT which is only updated in flush_thread() on
984          * some architectures like powerpc
985          */
986         current->mm->task_size = TASK_SIZE;
987
988         /* install the new credentials */
989         if (bprm->cred->uid != current_euid() ||
990             bprm->cred->gid != current_egid()) {
991                 current->pdeath_signal = 0;
992         } else if (file_permission(bprm->file, MAY_READ) ||
993                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
994                 set_dumpable(current->mm, suid_dumpable);
995         }
996
997         current->personality &= ~bprm->per_clear;
998
999         /*
1000          * Flush performance counters when crossing a
1001          * security domain:
1002          */
1003         if (!get_dumpable(current->mm))
1004                 perf_event_exit_task(current);
1005
1006         /* An exec changes our domain. We are no longer part of the thread
1007            group */
1008
1009         current->self_exec_id++;
1010                         
1011         flush_signal_handlers(current, 0);
1012         flush_old_files(current->files);
1013
1014         return 0;
1015
1016 out:
1017         return retval;
1018 }
1019
1020 EXPORT_SYMBOL(flush_old_exec);
1021
1022 /*
1023  * Prepare credentials and lock ->cred_guard_mutex.
1024  * install_exec_creds() commits the new creds and drops the lock.
1025  * Or, if exec fails before, free_bprm() should release ->cred and
1026  * and unlock.
1027  */
1028 int prepare_bprm_creds(struct linux_binprm *bprm)
1029 {
1030         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1031                 return -ERESTARTNOINTR;
1032
1033         bprm->cred = prepare_exec_creds();
1034         if (likely(bprm->cred))
1035                 return 0;
1036
1037         mutex_unlock(&current->cred_guard_mutex);
1038         return -ENOMEM;
1039 }
1040
1041 void free_bprm(struct linux_binprm *bprm)
1042 {
1043         free_arg_pages(bprm);
1044         if (bprm->cred) {
1045                 mutex_unlock(&current->cred_guard_mutex);
1046                 abort_creds(bprm->cred);
1047         }
1048         kfree(bprm);
1049 }
1050
1051 /*
1052  * install the new credentials for this executable
1053  */
1054 void install_exec_creds(struct linux_binprm *bprm)
1055 {
1056         security_bprm_committing_creds(bprm);
1057
1058         commit_creds(bprm->cred);
1059         bprm->cred = NULL;
1060         /*
1061          * cred_guard_mutex must be held at least to this point to prevent
1062          * ptrace_attach() from altering our determination of the task's
1063          * credentials; any time after this it may be unlocked.
1064          */
1065         security_bprm_committed_creds(bprm);
1066         mutex_unlock(&current->cred_guard_mutex);
1067 }
1068 EXPORT_SYMBOL(install_exec_creds);
1069
1070 /*
1071  * determine how safe it is to execute the proposed program
1072  * - the caller must hold current->cred_guard_mutex to protect against
1073  *   PTRACE_ATTACH
1074  */
1075 int check_unsafe_exec(struct linux_binprm *bprm)
1076 {
1077         struct task_struct *p = current, *t;
1078         unsigned n_fs;
1079         int res = 0;
1080
1081         bprm->unsafe = tracehook_unsafe_exec(p);
1082
1083         n_fs = 1;
1084         write_lock(&p->fs->lock);
1085         rcu_read_lock();
1086         for (t = next_thread(p); t != p; t = next_thread(t)) {
1087                 if (t->fs == p->fs)
1088                         n_fs++;
1089         }
1090         rcu_read_unlock();
1091
1092         if (p->fs->users > n_fs) {
1093                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1094         } else {
1095                 res = -EAGAIN;
1096                 if (!p->fs->in_exec) {
1097                         p->fs->in_exec = 1;
1098                         res = 1;
1099                 }
1100         }
1101         write_unlock(&p->fs->lock);
1102
1103         return res;
1104 }
1105
1106 /* 
1107  * Fill the binprm structure from the inode. 
1108  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1109  *
1110  * This may be called multiple times for binary chains (scripts for example).
1111  */
1112 int prepare_binprm(struct linux_binprm *bprm)
1113 {
1114         umode_t mode;
1115         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1116         int retval;
1117
1118         mode = inode->i_mode;
1119         if (bprm->file->f_op == NULL)
1120                 return -EACCES;
1121
1122         /* clear any previous set[ug]id data from a previous binary */
1123         bprm->cred->euid = current_euid();
1124         bprm->cred->egid = current_egid();
1125
1126         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1127                 /* Set-uid? */
1128                 if (mode & S_ISUID) {
1129                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1130                         bprm->cred->euid = inode->i_uid;
1131                 }
1132
1133                 /* Set-gid? */
1134                 /*
1135                  * If setgid is set but no group execute bit then this
1136                  * is a candidate for mandatory locking, not a setgid
1137                  * executable.
1138                  */
1139                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1140                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1141                         bprm->cred->egid = inode->i_gid;
1142                 }
1143         }
1144
1145         /* fill in binprm security blob */
1146         retval = security_bprm_set_creds(bprm);
1147         if (retval)
1148                 return retval;
1149         bprm->cred_prepared = 1;
1150
1151         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1152         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1153 }
1154
1155 EXPORT_SYMBOL(prepare_binprm);
1156
1157 /*
1158  * Arguments are '\0' separated strings found at the location bprm->p
1159  * points to; chop off the first by relocating brpm->p to right after
1160  * the first '\0' encountered.
1161  */
1162 int remove_arg_zero(struct linux_binprm *bprm)
1163 {
1164         int ret = 0;
1165         unsigned long offset;
1166         char *kaddr;
1167         struct page *page;
1168
1169         if (!bprm->argc)
1170                 return 0;
1171
1172         do {
1173                 offset = bprm->p & ~PAGE_MASK;
1174                 page = get_arg_page(bprm, bprm->p, 0);
1175                 if (!page) {
1176                         ret = -EFAULT;
1177                         goto out;
1178                 }
1179                 kaddr = kmap_atomic(page, KM_USER0);
1180
1181                 for (; offset < PAGE_SIZE && kaddr[offset];
1182                                 offset++, bprm->p++)
1183                         ;
1184
1185                 kunmap_atomic(kaddr, KM_USER0);
1186                 put_arg_page(page);
1187
1188                 if (offset == PAGE_SIZE)
1189                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1190         } while (offset == PAGE_SIZE);
1191
1192         bprm->p++;
1193         bprm->argc--;
1194         ret = 0;
1195
1196 out:
1197         return ret;
1198 }
1199 EXPORT_SYMBOL(remove_arg_zero);
1200
1201 /*
1202  * cycle the list of binary formats handler, until one recognizes the image
1203  */
1204 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1205 {
1206         unsigned int depth = bprm->recursion_depth;
1207         int try,retval;
1208         struct linux_binfmt *fmt;
1209
1210         retval = security_bprm_check(bprm);
1211         if (retval)
1212                 return retval;
1213         retval = ima_bprm_check(bprm);
1214         if (retval)
1215                 return retval;
1216
1217         /* kernel module loader fixup */
1218         /* so we don't try to load run modprobe in kernel space. */
1219         set_fs(USER_DS);
1220
1221         retval = audit_bprm(bprm);
1222         if (retval)
1223                 return retval;
1224
1225         retval = -ENOENT;
1226         for (try=0; try<2; try++) {
1227                 read_lock(&binfmt_lock);
1228                 list_for_each_entry(fmt, &formats, lh) {
1229                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1230                         if (!fn)
1231                                 continue;
1232                         if (!try_module_get(fmt->module))
1233                                 continue;
1234                         read_unlock(&binfmt_lock);
1235                         retval = fn(bprm, regs);
1236                         /*
1237                          * Restore the depth counter to its starting value
1238                          * in this call, so we don't have to rely on every
1239                          * load_binary function to restore it on return.
1240                          */
1241                         bprm->recursion_depth = depth;
1242                         if (retval >= 0) {
1243                                 if (depth == 0)
1244                                         tracehook_report_exec(fmt, bprm, regs);
1245                                 put_binfmt(fmt);
1246                                 allow_write_access(bprm->file);
1247                                 if (bprm->file)
1248                                         fput(bprm->file);
1249                                 bprm->file = NULL;
1250                                 current->did_exec = 1;
1251                                 proc_exec_connector(current);
1252                                 return retval;
1253                         }
1254                         read_lock(&binfmt_lock);
1255                         put_binfmt(fmt);
1256                         if (retval != -ENOEXEC || bprm->mm == NULL)
1257                                 break;
1258                         if (!bprm->file) {
1259                                 read_unlock(&binfmt_lock);
1260                                 return retval;
1261                         }
1262                 }
1263                 read_unlock(&binfmt_lock);
1264                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1265                         break;
1266 #ifdef CONFIG_MODULES
1267                 } else {
1268 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1269                         if (printable(bprm->buf[0]) &&
1270                             printable(bprm->buf[1]) &&
1271                             printable(bprm->buf[2]) &&
1272                             printable(bprm->buf[3]))
1273                                 break; /* -ENOEXEC */
1274                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1275 #endif
1276                 }
1277         }
1278         return retval;
1279 }
1280
1281 EXPORT_SYMBOL(search_binary_handler);
1282
1283 /*
1284  * sys_execve() executes a new program.
1285  */
1286 int do_execve(char * filename,
1287         char __user *__user *argv,
1288         char __user *__user *envp,
1289         struct pt_regs * regs)
1290 {
1291         struct linux_binprm *bprm;
1292         struct file *file;
1293         struct files_struct *displaced;
1294         bool clear_in_exec;
1295         int retval;
1296
1297         retval = unshare_files(&displaced);
1298         if (retval)
1299                 goto out_ret;
1300
1301         retval = -ENOMEM;
1302         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1303         if (!bprm)
1304                 goto out_files;
1305
1306         retval = prepare_bprm_creds(bprm);
1307         if (retval)
1308                 goto out_free;
1309
1310         retval = check_unsafe_exec(bprm);
1311         if (retval < 0)
1312                 goto out_free;
1313         clear_in_exec = retval;
1314         current->in_execve = 1;
1315
1316         file = open_exec(filename);
1317         retval = PTR_ERR(file);
1318         if (IS_ERR(file))
1319                 goto out_unmark;
1320
1321         sched_exec();
1322
1323         bprm->file = file;
1324         bprm->filename = filename;
1325         bprm->interp = filename;
1326
1327         retval = bprm_mm_init(bprm);
1328         if (retval)
1329                 goto out_file;
1330
1331         bprm->argc = count(argv, MAX_ARG_STRINGS);
1332         if ((retval = bprm->argc) < 0)
1333                 goto out;
1334
1335         bprm->envc = count(envp, MAX_ARG_STRINGS);
1336         if ((retval = bprm->envc) < 0)
1337                 goto out;
1338
1339         retval = prepare_binprm(bprm);
1340         if (retval < 0)
1341                 goto out;
1342
1343         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1344         if (retval < 0)
1345                 goto out;
1346
1347         bprm->exec = bprm->p;
1348         retval = copy_strings(bprm->envc, envp, bprm);
1349         if (retval < 0)
1350                 goto out;
1351
1352         retval = copy_strings(bprm->argc, argv, bprm);
1353         if (retval < 0)
1354                 goto out;
1355
1356         current->flags &= ~PF_KTHREAD;
1357         retval = search_binary_handler(bprm,regs);
1358         if (retval < 0)
1359                 goto out;
1360
1361         current->stack_start = current->mm->start_stack;
1362
1363         /* execve succeeded */
1364         current->fs->in_exec = 0;
1365         current->in_execve = 0;
1366         acct_update_integrals(current);
1367         free_bprm(bprm);
1368         if (displaced)
1369                 put_files_struct(displaced);
1370         return retval;
1371
1372 out:
1373         if (bprm->mm)
1374                 mmput (bprm->mm);
1375
1376 out_file:
1377         if (bprm->file) {
1378                 allow_write_access(bprm->file);
1379                 fput(bprm->file);
1380         }
1381
1382 out_unmark:
1383         if (clear_in_exec)
1384                 current->fs->in_exec = 0;
1385         current->in_execve = 0;
1386
1387 out_free:
1388         free_bprm(bprm);
1389
1390 out_files:
1391         if (displaced)
1392                 reset_files_struct(displaced);
1393 out_ret:
1394         return retval;
1395 }
1396
1397 int set_binfmt(struct linux_binfmt *new)
1398 {
1399         struct linux_binfmt *old = current->binfmt;
1400
1401         if (new) {
1402                 if (!try_module_get(new->module))
1403                         return -1;
1404         }
1405         current->binfmt = new;
1406         if (old)
1407                 module_put(old->module);
1408         return 0;
1409 }
1410
1411 EXPORT_SYMBOL(set_binfmt);
1412
1413 /* format_corename will inspect the pattern parameter, and output a
1414  * name into corename, which must have space for at least
1415  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1416  */
1417 static int format_corename(char *corename, long signr)
1418 {
1419         const struct cred *cred = current_cred();
1420         const char *pat_ptr = core_pattern;
1421         int ispipe = (*pat_ptr == '|');
1422         char *out_ptr = corename;
1423         char *const out_end = corename + CORENAME_MAX_SIZE;
1424         int rc;
1425         int pid_in_pattern = 0;
1426
1427         /* Repeat as long as we have more pattern to process and more output
1428            space */
1429         while (*pat_ptr) {
1430                 if (*pat_ptr != '%') {
1431                         if (out_ptr == out_end)
1432                                 goto out;
1433                         *out_ptr++ = *pat_ptr++;
1434                 } else {
1435                         switch (*++pat_ptr) {
1436                         case 0:
1437                                 goto out;
1438                         /* Double percent, output one percent */
1439                         case '%':
1440                                 if (out_ptr == out_end)
1441                                         goto out;
1442                                 *out_ptr++ = '%';
1443                                 break;
1444                         /* pid */
1445                         case 'p':
1446                                 pid_in_pattern = 1;
1447                                 rc = snprintf(out_ptr, out_end - out_ptr,
1448                                               "%d", task_tgid_vnr(current));
1449                                 if (rc > out_end - out_ptr)
1450                                         goto out;
1451                                 out_ptr += rc;
1452                                 break;
1453                         /* uid */
1454                         case 'u':
1455                                 rc = snprintf(out_ptr, out_end - out_ptr,
1456                                               "%d", cred->uid);
1457                                 if (rc > out_end - out_ptr)
1458                                         goto out;
1459                                 out_ptr += rc;
1460                                 break;
1461                         /* gid */
1462                         case 'g':
1463                                 rc = snprintf(out_ptr, out_end - out_ptr,
1464                                               "%d", cred->gid);
1465                                 if (rc > out_end - out_ptr)
1466                                         goto out;
1467                                 out_ptr += rc;
1468                                 break;
1469                         /* signal that caused the coredump */
1470                         case 's':
1471                                 rc = snprintf(out_ptr, out_end - out_ptr,
1472                                               "%ld", signr);
1473                                 if (rc > out_end - out_ptr)
1474                                         goto out;
1475                                 out_ptr += rc;
1476                                 break;
1477                         /* UNIX time of coredump */
1478                         case 't': {
1479                                 struct timeval tv;
1480                                 do_gettimeofday(&tv);
1481                                 rc = snprintf(out_ptr, out_end - out_ptr,
1482                                               "%lu", tv.tv_sec);
1483                                 if (rc > out_end - out_ptr)
1484                                         goto out;
1485                                 out_ptr += rc;
1486                                 break;
1487                         }
1488                         /* hostname */
1489                         case 'h':
1490                                 down_read(&uts_sem);
1491                                 rc = snprintf(out_ptr, out_end - out_ptr,
1492                                               "%s", utsname()->nodename);
1493                                 up_read(&uts_sem);
1494                                 if (rc > out_end - out_ptr)
1495                                         goto out;
1496                                 out_ptr += rc;
1497                                 break;
1498                         /* executable */
1499                         case 'e':
1500                                 rc = snprintf(out_ptr, out_end - out_ptr,
1501                                               "%s", current->comm);
1502                                 if (rc > out_end - out_ptr)
1503                                         goto out;
1504                                 out_ptr += rc;
1505                                 break;
1506                         /* core limit size */
1507                         case 'c':
1508                                 rc = snprintf(out_ptr, out_end - out_ptr,
1509                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1510                                 if (rc > out_end - out_ptr)
1511                                         goto out;
1512                                 out_ptr += rc;
1513                                 break;
1514                         default:
1515                                 break;
1516                         }
1517                         ++pat_ptr;
1518                 }
1519         }
1520         /* Backward compatibility with core_uses_pid:
1521          *
1522          * If core_pattern does not include a %p (as is the default)
1523          * and core_uses_pid is set, then .%pid will be appended to
1524          * the filename. Do not do this for piped commands. */
1525         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1526                 rc = snprintf(out_ptr, out_end - out_ptr,
1527                               ".%d", task_tgid_vnr(current));
1528                 if (rc > out_end - out_ptr)
1529                         goto out;
1530                 out_ptr += rc;
1531         }
1532 out:
1533         *out_ptr = 0;
1534         return ispipe;
1535 }
1536
1537 static int zap_process(struct task_struct *start)
1538 {
1539         struct task_struct *t;
1540         int nr = 0;
1541
1542         start->signal->flags = SIGNAL_GROUP_EXIT;
1543         start->signal->group_stop_count = 0;
1544
1545         t = start;
1546         do {
1547                 if (t != current && t->mm) {
1548                         sigaddset(&t->pending.signal, SIGKILL);
1549                         signal_wake_up(t, 1);
1550                         nr++;
1551                 }
1552         } while_each_thread(start, t);
1553
1554         return nr;
1555 }
1556
1557 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1558                                 struct core_state *core_state, int exit_code)
1559 {
1560         struct task_struct *g, *p;
1561         unsigned long flags;
1562         int nr = -EAGAIN;
1563
1564         spin_lock_irq(&tsk->sighand->siglock);
1565         if (!signal_group_exit(tsk->signal)) {
1566                 mm->core_state = core_state;
1567                 tsk->signal->group_exit_code = exit_code;
1568                 nr = zap_process(tsk);
1569         }
1570         spin_unlock_irq(&tsk->sighand->siglock);
1571         if (unlikely(nr < 0))
1572                 return nr;
1573
1574         if (atomic_read(&mm->mm_users) == nr + 1)
1575                 goto done;
1576         /*
1577          * We should find and kill all tasks which use this mm, and we should
1578          * count them correctly into ->nr_threads. We don't take tasklist
1579          * lock, but this is safe wrt:
1580          *
1581          * fork:
1582          *      None of sub-threads can fork after zap_process(leader). All
1583          *      processes which were created before this point should be
1584          *      visible to zap_threads() because copy_process() adds the new
1585          *      process to the tail of init_task.tasks list, and lock/unlock
1586          *      of ->siglock provides a memory barrier.
1587          *
1588          * do_exit:
1589          *      The caller holds mm->mmap_sem. This means that the task which
1590          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1591          *      its ->mm.
1592          *
1593          * de_thread:
1594          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1595          *      we must see either old or new leader, this does not matter.
1596          *      However, it can change p->sighand, so lock_task_sighand(p)
1597          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1598          *      it can't fail.
1599          *
1600          *      Note also that "g" can be the old leader with ->mm == NULL
1601          *      and already unhashed and thus removed from ->thread_group.
1602          *      This is OK, __unhash_process()->list_del_rcu() does not
1603          *      clear the ->next pointer, we will find the new leader via
1604          *      next_thread().
1605          */
1606         rcu_read_lock();
1607         for_each_process(g) {
1608                 if (g == tsk->group_leader)
1609                         continue;
1610                 if (g->flags & PF_KTHREAD)
1611                         continue;
1612                 p = g;
1613                 do {
1614                         if (p->mm) {
1615                                 if (unlikely(p->mm == mm)) {
1616                                         lock_task_sighand(p, &flags);
1617                                         nr += zap_process(p);
1618                                         unlock_task_sighand(p, &flags);
1619                                 }
1620                                 break;
1621                         }
1622                 } while_each_thread(g, p);
1623         }
1624         rcu_read_unlock();
1625 done:
1626         atomic_set(&core_state->nr_threads, nr);
1627         return nr;
1628 }
1629
1630 static int coredump_wait(int exit_code, struct core_state *core_state)
1631 {
1632         struct task_struct *tsk = current;
1633         struct mm_struct *mm = tsk->mm;
1634         struct completion *vfork_done;
1635         int core_waiters;
1636
1637         init_completion(&core_state->startup);
1638         core_state->dumper.task = tsk;
1639         core_state->dumper.next = NULL;
1640         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1641         up_write(&mm->mmap_sem);
1642
1643         if (unlikely(core_waiters < 0))
1644                 goto fail;
1645
1646         /*
1647          * Make sure nobody is waiting for us to release the VM,
1648          * otherwise we can deadlock when we wait on each other
1649          */
1650         vfork_done = tsk->vfork_done;
1651         if (vfork_done) {
1652                 tsk->vfork_done = NULL;
1653                 complete(vfork_done);
1654         }
1655
1656         if (core_waiters)
1657                 wait_for_completion(&core_state->startup);
1658 fail:
1659         return core_waiters;
1660 }
1661
1662 static void coredump_finish(struct mm_struct *mm)
1663 {
1664         struct core_thread *curr, *next;
1665         struct task_struct *task;
1666
1667         next = mm->core_state->dumper.next;
1668         while ((curr = next) != NULL) {
1669                 next = curr->next;
1670                 task = curr->task;
1671                 /*
1672                  * see exit_mm(), curr->task must not see
1673                  * ->task == NULL before we read ->next.
1674                  */
1675                 smp_mb();
1676                 curr->task = NULL;
1677                 wake_up_process(task);
1678         }
1679
1680         mm->core_state = NULL;
1681 }
1682
1683 /*
1684  * set_dumpable converts traditional three-value dumpable to two flags and
1685  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1686  * these bits are not changed atomically.  So get_dumpable can observe the
1687  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1688  * return either old dumpable or new one by paying attention to the order of
1689  * modifying the bits.
1690  *
1691  * dumpable |   mm->flags (binary)
1692  * old  new | initial interim  final
1693  * ---------+-----------------------
1694  *  0    1  |   00      01      01
1695  *  0    2  |   00      10(*)   11
1696  *  1    0  |   01      00      00
1697  *  1    2  |   01      11      11
1698  *  2    0  |   11      10(*)   00
1699  *  2    1  |   11      11      01
1700  *
1701  * (*) get_dumpable regards interim value of 10 as 11.
1702  */
1703 void set_dumpable(struct mm_struct *mm, int value)
1704 {
1705         switch (value) {
1706         case 0:
1707                 clear_bit(MMF_DUMPABLE, &mm->flags);
1708                 smp_wmb();
1709                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1710                 break;
1711         case 1:
1712                 set_bit(MMF_DUMPABLE, &mm->flags);
1713                 smp_wmb();
1714                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1715                 break;
1716         case 2:
1717                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1718                 smp_wmb();
1719                 set_bit(MMF_DUMPABLE, &mm->flags);
1720                 break;
1721         }
1722 }
1723
1724 int get_dumpable(struct mm_struct *mm)
1725 {
1726         int ret;
1727
1728         ret = mm->flags & 0x3;
1729         return (ret >= 2) ? 2 : ret;
1730 }
1731
1732 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1733 {
1734         struct core_state core_state;
1735         char corename[CORENAME_MAX_SIZE + 1];
1736         struct mm_struct *mm = current->mm;
1737         struct linux_binfmt * binfmt;
1738         struct inode * inode;
1739         struct file * file;
1740         const struct cred *old_cred;
1741         struct cred *cred;
1742         int retval = 0;
1743         int flag = 0;
1744         int ispipe = 0;
1745         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1746         char **helper_argv = NULL;
1747         int helper_argc = 0;
1748         int dump_count = 0;
1749         static atomic_t core_dump_count = ATOMIC_INIT(0);
1750
1751         audit_core_dumps(signr);
1752
1753         binfmt = current->binfmt;
1754         if (!binfmt || !binfmt->core_dump)
1755                 goto fail;
1756
1757         cred = prepare_creds();
1758         if (!cred) {
1759                 retval = -ENOMEM;
1760                 goto fail;
1761         }
1762
1763         down_write(&mm->mmap_sem);
1764         /*
1765          * If another thread got here first, or we are not dumpable, bail out.
1766          */
1767         if (mm->core_state || !get_dumpable(mm)) {
1768                 up_write(&mm->mmap_sem);
1769                 put_cred(cred);
1770                 goto fail;
1771         }
1772
1773         /*
1774          *      We cannot trust fsuid as being the "true" uid of the
1775          *      process nor do we know its entire history. We only know it
1776          *      was tainted so we dump it as root in mode 2.
1777          */
1778         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1779                 flag = O_EXCL;          /* Stop rewrite attacks */
1780                 cred->fsuid = 0;        /* Dump root private */
1781         }
1782
1783         retval = coredump_wait(exit_code, &core_state);
1784         if (retval < 0) {
1785                 put_cred(cred);
1786                 goto fail;
1787         }
1788
1789         old_cred = override_creds(cred);
1790
1791         /*
1792          * Clear any false indication of pending signals that might
1793          * be seen by the filesystem code called to write the core file.
1794          */
1795         clear_thread_flag(TIF_SIGPENDING);
1796
1797         /*
1798          * lock_kernel() because format_corename() is controlled by sysctl, which
1799          * uses lock_kernel()
1800          */
1801         lock_kernel();
1802         ispipe = format_corename(corename, signr);
1803         unlock_kernel();
1804
1805         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1806                 goto fail_unlock;
1807
1808         if (ispipe) {
1809                 if (core_limit == 0) {
1810                         /*
1811                          * Normally core limits are irrelevant to pipes, since
1812                          * we're not writing to the file system, but we use
1813                          * core_limit of 0 here as a speacial value. Any
1814                          * non-zero limit gets set to RLIM_INFINITY below, but
1815                          * a limit of 0 skips the dump.  This is a consistent
1816                          * way to catch recursive crashes.  We can still crash
1817                          * if the core_pattern binary sets RLIM_CORE =  !0
1818                          * but it runs as root, and can do lots of stupid things
1819                          * Note that we use task_tgid_vnr here to grab the pid
1820                          * of the process group leader.  That way we get the
1821                          * right pid if a thread in a multi-threaded
1822                          * core_pattern process dies.
1823                          */
1824                         printk(KERN_WARNING
1825                                 "Process %d(%s) has RLIMIT_CORE set to 0\n",
1826                                 task_tgid_vnr(current), current->comm);
1827                         printk(KERN_WARNING "Aborting core\n");
1828                         goto fail_unlock;
1829                 }
1830
1831                 dump_count = atomic_inc_return(&core_dump_count);
1832                 if (core_pipe_limit && (core_pipe_limit < dump_count)) {
1833                         printk(KERN_WARNING "Pid %d(%s) over core_pipe_limit\n",
1834                                task_tgid_vnr(current), current->comm);
1835                         printk(KERN_WARNING "Skipping core dump\n");
1836                         goto fail_dropcount;
1837                 }
1838
1839                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1840                 if (!helper_argv) {
1841                         printk(KERN_WARNING "%s failed to allocate memory\n",
1842                                __func__);
1843                         goto fail_dropcount;
1844                 }
1845
1846                 core_limit = RLIM_INFINITY;
1847
1848                 /* SIGPIPE can happen, but it's just never processed */
1849                 if (call_usermodehelper_pipe(helper_argv[0], helper_argv, NULL,
1850                                 &file)) {
1851                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1852                                corename);
1853                         goto fail_dropcount;
1854                 }
1855         } else
1856                 file = filp_open(corename,
1857                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1858                                  0600);
1859         if (IS_ERR(file))
1860                 goto fail_dropcount;
1861         inode = file->f_path.dentry->d_inode;
1862         if (inode->i_nlink > 1)
1863                 goto close_fail;        /* multiple links - don't dump */
1864         if (!ispipe && d_unhashed(file->f_path.dentry))
1865                 goto close_fail;
1866
1867         /* AK: actually i see no reason to not allow this for named pipes etc.,
1868            but keep the previous behaviour for now. */
1869         if (!ispipe && !S_ISREG(inode->i_mode))
1870                 goto close_fail;
1871         /*
1872          * Dont allow local users get cute and trick others to coredump
1873          * into their pre-created files:
1874          */
1875         if (inode->i_uid != current_fsuid())
1876                 goto close_fail;
1877         if (!file->f_op)
1878                 goto close_fail;
1879         if (!file->f_op->write)
1880                 goto close_fail;
1881         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1882                 goto close_fail;
1883
1884         retval = binfmt->core_dump(signr, regs, file, core_limit);
1885
1886         if (retval)
1887                 current->signal->group_exit_code |= 0x80;
1888 close_fail:
1889         filp_close(file, NULL);
1890 fail_dropcount:
1891         if (dump_count)
1892                 atomic_dec(&core_dump_count);
1893 fail_unlock:
1894         if (helper_argv)
1895                 argv_free(helper_argv);
1896
1897         revert_creds(old_cred);
1898         put_cred(cred);
1899         coredump_finish(mm);
1900 fail:
1901         return;
1902 }