perf: Do the big rename: Performance Counters -> Performance Events
[linux-2.6.git] / fs / exec.c
1 /*
2  *  linux/fs/exec.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6
7 /*
8  * #!-checking implemented by tytso.
9  */
10 /*
11  * Demand-loading implemented 01.12.91 - no need to read anything but
12  * the header into memory. The inode of the executable is put into
13  * "current->executable", and page faults do the actual loading. Clean.
14  *
15  * Once more I can proudly say that linux stood up to being changed: it
16  * was less than 2 hours work to get demand-loading completely implemented.
17  *
18  * Demand loading changed July 1993 by Eric Youngdale.   Use mmap instead,
19  * current->executable is only used by the procfs.  This allows a dispatch
20  * table to check for several different types  of binary formats.  We keep
21  * trying until we recognize the file or we run out of supported binary
22  * formats. 
23  */
24
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
28 #include <linux/mm.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/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 int suid_dumpable = 0;
67
68 /* The maximal length of core_pattern is also specified in sysctl.c */
69
70 static LIST_HEAD(formats);
71 static DEFINE_RWLOCK(binfmt_lock);
72
73 int __register_binfmt(struct linux_binfmt * fmt, int insert)
74 {
75         if (!fmt)
76                 return -EINVAL;
77         write_lock(&binfmt_lock);
78         insert ? list_add(&fmt->lh, &formats) :
79                  list_add_tail(&fmt->lh, &formats);
80         write_unlock(&binfmt_lock);
81         return 0;       
82 }
83
84 EXPORT_SYMBOL(__register_binfmt);
85
86 void unregister_binfmt(struct linux_binfmt * fmt)
87 {
88         write_lock(&binfmt_lock);
89         list_del(&fmt->lh);
90         write_unlock(&binfmt_lock);
91 }
92
93 EXPORT_SYMBOL(unregister_binfmt);
94
95 static inline void put_binfmt(struct linux_binfmt * fmt)
96 {
97         module_put(fmt->module);
98 }
99
100 /*
101  * Note that a shared library must be both readable and executable due to
102  * security reasons.
103  *
104  * Also note that we take the address to load from from the file itself.
105  */
106 SYSCALL_DEFINE1(uselib, const char __user *, library)
107 {
108         struct file *file;
109         char *tmp = getname(library);
110         int error = PTR_ERR(tmp);
111
112         if (IS_ERR(tmp))
113                 goto out;
114
115         file = do_filp_open(AT_FDCWD, tmp,
116                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
117                                 MAY_READ | MAY_EXEC | MAY_OPEN);
118         putname(tmp);
119         error = PTR_ERR(file);
120         if (IS_ERR(file))
121                 goto out;
122
123         error = -EINVAL;
124         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
125                 goto exit;
126
127         error = -EACCES;
128         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
129                 goto exit;
130
131         fsnotify_open(file->f_path.dentry);
132
133         error = -ENOEXEC;
134         if(file->f_op) {
135                 struct linux_binfmt * fmt;
136
137                 read_lock(&binfmt_lock);
138                 list_for_each_entry(fmt, &formats, lh) {
139                         if (!fmt->load_shlib)
140                                 continue;
141                         if (!try_module_get(fmt->module))
142                                 continue;
143                         read_unlock(&binfmt_lock);
144                         error = fmt->load_shlib(file);
145                         read_lock(&binfmt_lock);
146                         put_binfmt(fmt);
147                         if (error != -ENOEXEC)
148                                 break;
149                 }
150                 read_unlock(&binfmt_lock);
151         }
152 exit:
153         fput(file);
154 out:
155         return error;
156 }
157
158 #ifdef CONFIG_MMU
159
160 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
161                 int write)
162 {
163         struct page *page;
164         int ret;
165
166 #ifdef CONFIG_STACK_GROWSUP
167         if (write) {
168                 ret = expand_stack_downwards(bprm->vma, pos);
169                 if (ret < 0)
170                         return NULL;
171         }
172 #endif
173         ret = get_user_pages(current, bprm->mm, pos,
174                         1, write, 1, &page, NULL);
175         if (ret <= 0)
176                 return NULL;
177
178         if (write) {
179                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
180                 struct rlimit *rlim;
181
182                 /*
183                  * We've historically supported up to 32 pages (ARG_MAX)
184                  * of argument strings even with small stacks
185                  */
186                 if (size <= ARG_MAX)
187                         return page;
188
189                 /*
190                  * Limit to 1/4-th the stack size for the argv+env strings.
191                  * This ensures that:
192                  *  - the remaining binfmt code will not run out of stack space,
193                  *  - the program will have a reasonable amount of stack left
194                  *    to work from.
195                  */
196                 rlim = current->signal->rlim;
197                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
198                         put_page(page);
199                         return NULL;
200                 }
201         }
202
203         return page;
204 }
205
206 static void put_arg_page(struct page *page)
207 {
208         put_page(page);
209 }
210
211 static void free_arg_page(struct linux_binprm *bprm, int i)
212 {
213 }
214
215 static void free_arg_pages(struct linux_binprm *bprm)
216 {
217 }
218
219 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
220                 struct page *page)
221 {
222         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
223 }
224
225 static int __bprm_mm_init(struct linux_binprm *bprm)
226 {
227         int err;
228         struct vm_area_struct *vma = NULL;
229         struct mm_struct *mm = bprm->mm;
230
231         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
232         if (!vma)
233                 return -ENOMEM;
234
235         down_write(&mm->mmap_sem);
236         vma->vm_mm = mm;
237
238         /*
239          * Place the stack at the largest stack address the architecture
240          * supports. Later, we'll move this to an appropriate place. We don't
241          * use STACK_TOP because that can depend on attributes which aren't
242          * configured yet.
243          */
244         vma->vm_end = STACK_TOP_MAX;
245         vma->vm_start = vma->vm_end - PAGE_SIZE;
246         vma->vm_flags = VM_STACK_FLAGS;
247         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
248         err = insert_vm_struct(mm, vma);
249         if (err)
250                 goto err;
251
252         mm->stack_vm = mm->total_vm = 1;
253         up_write(&mm->mmap_sem);
254         bprm->p = vma->vm_end - sizeof(void *);
255         return 0;
256 err:
257         up_write(&mm->mmap_sem);
258         bprm->vma = NULL;
259         kmem_cache_free(vm_area_cachep, vma);
260         return err;
261 }
262
263 static bool valid_arg_len(struct linux_binprm *bprm, long len)
264 {
265         return len <= MAX_ARG_STRLEN;
266 }
267
268 #else
269
270 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
271                 int write)
272 {
273         struct page *page;
274
275         page = bprm->page[pos / PAGE_SIZE];
276         if (!page && write) {
277                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
278                 if (!page)
279                         return NULL;
280                 bprm->page[pos / PAGE_SIZE] = page;
281         }
282
283         return page;
284 }
285
286 static void put_arg_page(struct page *page)
287 {
288 }
289
290 static void free_arg_page(struct linux_binprm *bprm, int i)
291 {
292         if (bprm->page[i]) {
293                 __free_page(bprm->page[i]);
294                 bprm->page[i] = NULL;
295         }
296 }
297
298 static void free_arg_pages(struct linux_binprm *bprm)
299 {
300         int i;
301
302         for (i = 0; i < MAX_ARG_PAGES; i++)
303                 free_arg_page(bprm, i);
304 }
305
306 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
307                 struct page *page)
308 {
309 }
310
311 static int __bprm_mm_init(struct linux_binprm *bprm)
312 {
313         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
314         return 0;
315 }
316
317 static bool valid_arg_len(struct linux_binprm *bprm, long len)
318 {
319         return len <= bprm->p;
320 }
321
322 #endif /* CONFIG_MMU */
323
324 /*
325  * Create a new mm_struct and populate it with a temporary stack
326  * vm_area_struct.  We don't have enough context at this point to set the stack
327  * flags, permissions, and offset, so we use temporary values.  We'll update
328  * them later in setup_arg_pages().
329  */
330 int bprm_mm_init(struct linux_binprm *bprm)
331 {
332         int err;
333         struct mm_struct *mm = NULL;
334
335         bprm->mm = mm = mm_alloc();
336         err = -ENOMEM;
337         if (!mm)
338                 goto err;
339
340         err = init_new_context(current, mm);
341         if (err)
342                 goto err;
343
344         err = __bprm_mm_init(bprm);
345         if (err)
346                 goto err;
347
348         return 0;
349
350 err:
351         if (mm) {
352                 bprm->mm = NULL;
353                 mmdrop(mm);
354         }
355
356         return err;
357 }
358
359 /*
360  * count() counts the number of strings in array ARGV.
361  */
362 static int count(char __user * __user * argv, int max)
363 {
364         int i = 0;
365
366         if (argv != NULL) {
367                 for (;;) {
368                         char __user * p;
369
370                         if (get_user(p, argv))
371                                 return -EFAULT;
372                         if (!p)
373                                 break;
374                         argv++;
375                         if (i++ >= max)
376                                 return -E2BIG;
377                         cond_resched();
378                 }
379         }
380         return i;
381 }
382
383 /*
384  * 'copy_strings()' copies argument/environment strings from the old
385  * processes's memory to the new process's stack.  The call to get_user_pages()
386  * ensures the destination page is created and not swapped out.
387  */
388 static int copy_strings(int argc, char __user * __user * argv,
389                         struct linux_binprm *bprm)
390 {
391         struct page *kmapped_page = NULL;
392         char *kaddr = NULL;
393         unsigned long kpos = 0;
394         int ret;
395
396         while (argc-- > 0) {
397                 char __user *str;
398                 int len;
399                 unsigned long pos;
400
401                 if (get_user(str, argv+argc) ||
402                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
403                         ret = -EFAULT;
404                         goto out;
405                 }
406
407                 if (!valid_arg_len(bprm, len)) {
408                         ret = -E2BIG;
409                         goto out;
410                 }
411
412                 /* We're going to work our way backwords. */
413                 pos = bprm->p;
414                 str += len;
415                 bprm->p -= len;
416
417                 while (len > 0) {
418                         int offset, bytes_to_copy;
419
420                         offset = pos % PAGE_SIZE;
421                         if (offset == 0)
422                                 offset = PAGE_SIZE;
423
424                         bytes_to_copy = offset;
425                         if (bytes_to_copy > len)
426                                 bytes_to_copy = len;
427
428                         offset -= bytes_to_copy;
429                         pos -= bytes_to_copy;
430                         str -= bytes_to_copy;
431                         len -= bytes_to_copy;
432
433                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
434                                 struct page *page;
435
436                                 page = get_arg_page(bprm, pos, 1);
437                                 if (!page) {
438                                         ret = -E2BIG;
439                                         goto out;
440                                 }
441
442                                 if (kmapped_page) {
443                                         flush_kernel_dcache_page(kmapped_page);
444                                         kunmap(kmapped_page);
445                                         put_arg_page(kmapped_page);
446                                 }
447                                 kmapped_page = page;
448                                 kaddr = kmap(kmapped_page);
449                                 kpos = pos & PAGE_MASK;
450                                 flush_arg_page(bprm, kpos, kmapped_page);
451                         }
452                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
453                                 ret = -EFAULT;
454                                 goto out;
455                         }
456                 }
457         }
458         ret = 0;
459 out:
460         if (kmapped_page) {
461                 flush_kernel_dcache_page(kmapped_page);
462                 kunmap(kmapped_page);
463                 put_arg_page(kmapped_page);
464         }
465         return ret;
466 }
467
468 /*
469  * Like copy_strings, but get argv and its values from kernel memory.
470  */
471 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
472 {
473         int r;
474         mm_segment_t oldfs = get_fs();
475         set_fs(KERNEL_DS);
476         r = copy_strings(argc, (char __user * __user *)argv, bprm);
477         set_fs(oldfs);
478         return r;
479 }
480 EXPORT_SYMBOL(copy_strings_kernel);
481
482 #ifdef CONFIG_MMU
483
484 /*
485  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
486  * the binfmt code determines where the new stack should reside, we shift it to
487  * its final location.  The process proceeds as follows:
488  *
489  * 1) Use shift to calculate the new vma endpoints.
490  * 2) Extend vma to cover both the old and new ranges.  This ensures the
491  *    arguments passed to subsequent functions are consistent.
492  * 3) Move vma's page tables to the new range.
493  * 4) Free up any cleared pgd range.
494  * 5) Shrink the vma to cover only the new range.
495  */
496 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
497 {
498         struct mm_struct *mm = vma->vm_mm;
499         unsigned long old_start = vma->vm_start;
500         unsigned long old_end = vma->vm_end;
501         unsigned long length = old_end - old_start;
502         unsigned long new_start = old_start - shift;
503         unsigned long new_end = old_end - shift;
504         struct mmu_gather *tlb;
505
506         BUG_ON(new_start > new_end);
507
508         /*
509          * ensure there are no vmas between where we want to go
510          * and where we are
511          */
512         if (vma != find_vma(mm, new_start))
513                 return -EFAULT;
514
515         /*
516          * cover the whole range: [new_start, old_end)
517          */
518         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
519
520         /*
521          * move the page tables downwards, on failure we rely on
522          * process cleanup to remove whatever mess we made.
523          */
524         if (length != move_page_tables(vma, old_start,
525                                        vma, new_start, length))
526                 return -ENOMEM;
527
528         lru_add_drain();
529         tlb = tlb_gather_mmu(mm, 0);
530         if (new_end > old_start) {
531                 /*
532                  * when the old and new regions overlap clear from new_end.
533                  */
534                 free_pgd_range(tlb, new_end, old_end, new_end,
535                         vma->vm_next ? vma->vm_next->vm_start : 0);
536         } else {
537                 /*
538                  * otherwise, clean from old_start; this is done to not touch
539                  * the address space in [new_end, old_start) some architectures
540                  * have constraints on va-space that make this illegal (IA64) -
541                  * for the others its just a little faster.
542                  */
543                 free_pgd_range(tlb, old_start, old_end, new_end,
544                         vma->vm_next ? vma->vm_next->vm_start : 0);
545         }
546         tlb_finish_mmu(tlb, new_end, old_end);
547
548         /*
549          * shrink the vma to just the new range.
550          */
551         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
552
553         return 0;
554 }
555
556 #define EXTRA_STACK_VM_PAGES    20      /* random */
557
558 /*
559  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
560  * the stack is optionally relocated, and some extra space is added.
561  */
562 int setup_arg_pages(struct linux_binprm *bprm,
563                     unsigned long stack_top,
564                     int executable_stack)
565 {
566         unsigned long ret;
567         unsigned long stack_shift;
568         struct mm_struct *mm = current->mm;
569         struct vm_area_struct *vma = bprm->vma;
570         struct vm_area_struct *prev = NULL;
571         unsigned long vm_flags;
572         unsigned long stack_base;
573
574 #ifdef CONFIG_STACK_GROWSUP
575         /* Limit stack size to 1GB */
576         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
577         if (stack_base > (1 << 30))
578                 stack_base = 1 << 30;
579
580         /* Make sure we didn't let the argument array grow too large. */
581         if (vma->vm_end - vma->vm_start > stack_base)
582                 return -ENOMEM;
583
584         stack_base = PAGE_ALIGN(stack_top - stack_base);
585
586         stack_shift = vma->vm_start - stack_base;
587         mm->arg_start = bprm->p - stack_shift;
588         bprm->p = vma->vm_end - stack_shift;
589 #else
590         stack_top = arch_align_stack(stack_top);
591         stack_top = PAGE_ALIGN(stack_top);
592         stack_shift = vma->vm_end - stack_top;
593
594         bprm->p -= stack_shift;
595         mm->arg_start = bprm->p;
596 #endif
597
598         if (bprm->loader)
599                 bprm->loader -= stack_shift;
600         bprm->exec -= stack_shift;
601
602         down_write(&mm->mmap_sem);
603         vm_flags = VM_STACK_FLAGS;
604
605         /*
606          * Adjust stack execute permissions; explicitly enable for
607          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
608          * (arch default) otherwise.
609          */
610         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
611                 vm_flags |= VM_EXEC;
612         else if (executable_stack == EXSTACK_DISABLE_X)
613                 vm_flags &= ~VM_EXEC;
614         vm_flags |= mm->def_flags;
615
616         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
617                         vm_flags);
618         if (ret)
619                 goto out_unlock;
620         BUG_ON(prev != vma);
621
622         /* Move stack pages down in memory. */
623         if (stack_shift) {
624                 ret = shift_arg_pages(vma, stack_shift);
625                 if (ret) {
626                         up_write(&mm->mmap_sem);
627                         return ret;
628                 }
629         }
630
631 #ifdef CONFIG_STACK_GROWSUP
632         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
633 #else
634         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
635 #endif
636         ret = expand_stack(vma, stack_base);
637         if (ret)
638                 ret = -EFAULT;
639
640 out_unlock:
641         up_write(&mm->mmap_sem);
642         return 0;
643 }
644 EXPORT_SYMBOL(setup_arg_pages);
645
646 #endif /* CONFIG_MMU */
647
648 struct file *open_exec(const char *name)
649 {
650         struct file *file;
651         int err;
652
653         file = do_filp_open(AT_FDCWD, name,
654                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
655                                 MAY_EXEC | MAY_OPEN);
656         if (IS_ERR(file))
657                 goto out;
658
659         err = -EACCES;
660         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
661                 goto exit;
662
663         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
664                 goto exit;
665
666         fsnotify_open(file->f_path.dentry);
667
668         err = deny_write_access(file);
669         if (err)
670                 goto exit;
671
672 out:
673         return file;
674
675 exit:
676         fput(file);
677         return ERR_PTR(err);
678 }
679 EXPORT_SYMBOL(open_exec);
680
681 int kernel_read(struct file *file, loff_t offset,
682                 char *addr, unsigned long count)
683 {
684         mm_segment_t old_fs;
685         loff_t pos = offset;
686         int result;
687
688         old_fs = get_fs();
689         set_fs(get_ds());
690         /* The cast to a user pointer is valid due to the set_fs() */
691         result = vfs_read(file, (void __user *)addr, count, &pos);
692         set_fs(old_fs);
693         return result;
694 }
695
696 EXPORT_SYMBOL(kernel_read);
697
698 static int exec_mmap(struct mm_struct *mm)
699 {
700         struct task_struct *tsk;
701         struct mm_struct * old_mm, *active_mm;
702
703         /* Notify parent that we're no longer interested in the old VM */
704         tsk = current;
705         old_mm = current->mm;
706         mm_release(tsk, old_mm);
707
708         if (old_mm) {
709                 /*
710                  * Make sure that if there is a core dump in progress
711                  * for the old mm, we get out and die instead of going
712                  * through with the exec.  We must hold mmap_sem around
713                  * checking core_state and changing tsk->mm.
714                  */
715                 down_read(&old_mm->mmap_sem);
716                 if (unlikely(old_mm->core_state)) {
717                         up_read(&old_mm->mmap_sem);
718                         return -EINTR;
719                 }
720         }
721         task_lock(tsk);
722         active_mm = tsk->active_mm;
723         tsk->mm = mm;
724         tsk->active_mm = mm;
725         activate_mm(active_mm, mm);
726         task_unlock(tsk);
727         arch_pick_mmap_layout(mm);
728         if (old_mm) {
729                 up_read(&old_mm->mmap_sem);
730                 BUG_ON(active_mm != old_mm);
731                 mm_update_next_owner(old_mm);
732                 mmput(old_mm);
733                 return 0;
734         }
735         mmdrop(active_mm);
736         return 0;
737 }
738
739 /*
740  * This function makes sure the current process has its own signal table,
741  * so that flush_signal_handlers can later reset the handlers without
742  * disturbing other processes.  (Other processes might share the signal
743  * table via the CLONE_SIGHAND option to clone().)
744  */
745 static int de_thread(struct task_struct *tsk)
746 {
747         struct signal_struct *sig = tsk->signal;
748         struct sighand_struct *oldsighand = tsk->sighand;
749         spinlock_t *lock = &oldsighand->siglock;
750         int count;
751
752         if (thread_group_empty(tsk))
753                 goto no_thread_group;
754
755         /*
756          * Kill all other threads in the thread group.
757          */
758         spin_lock_irq(lock);
759         if (signal_group_exit(sig)) {
760                 /*
761                  * Another group action in progress, just
762                  * return so that the signal is processed.
763                  */
764                 spin_unlock_irq(lock);
765                 return -EAGAIN;
766         }
767         sig->group_exit_task = tsk;
768         zap_other_threads(tsk);
769
770         /* Account for the thread group leader hanging around: */
771         count = thread_group_leader(tsk) ? 1 : 2;
772         sig->notify_count = count;
773         while (atomic_read(&sig->count) > count) {
774                 __set_current_state(TASK_UNINTERRUPTIBLE);
775                 spin_unlock_irq(lock);
776                 schedule();
777                 spin_lock_irq(lock);
778         }
779         spin_unlock_irq(lock);
780
781         /*
782          * At this point all other threads have exited, all we have to
783          * do is to wait for the thread group leader to become inactive,
784          * and to assume its PID:
785          */
786         if (!thread_group_leader(tsk)) {
787                 struct task_struct *leader = tsk->group_leader;
788
789                 sig->notify_count = -1; /* for exit_notify() */
790                 for (;;) {
791                         write_lock_irq(&tasklist_lock);
792                         if (likely(leader->exit_state))
793                                 break;
794                         __set_current_state(TASK_UNINTERRUPTIBLE);
795                         write_unlock_irq(&tasklist_lock);
796                         schedule();
797                 }
798
799                 /*
800                  * The only record we have of the real-time age of a
801                  * process, regardless of execs it's done, is start_time.
802                  * All the past CPU time is accumulated in signal_struct
803                  * from sister threads now dead.  But in this non-leader
804                  * exec, nothing survives from the original leader thread,
805                  * whose birth marks the true age of this process now.
806                  * When we take on its identity by switching to its PID, we
807                  * also take its birthdate (always earlier than our own).
808                  */
809                 tsk->start_time = leader->start_time;
810
811                 BUG_ON(!same_thread_group(leader, tsk));
812                 BUG_ON(has_group_leader_pid(tsk));
813                 /*
814                  * An exec() starts a new thread group with the
815                  * TGID of the previous thread group. Rehash the
816                  * two threads with a switched PID, and release
817                  * the former thread group leader:
818                  */
819
820                 /* Become a process group leader with the old leader's pid.
821                  * The old leader becomes a thread of the this thread group.
822                  * Note: The old leader also uses this pid until release_task
823                  *       is called.  Odd but simple and correct.
824                  */
825                 detach_pid(tsk, PIDTYPE_PID);
826                 tsk->pid = leader->pid;
827                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
828                 transfer_pid(leader, tsk, PIDTYPE_PGID);
829                 transfer_pid(leader, tsk, PIDTYPE_SID);
830                 list_replace_rcu(&leader->tasks, &tsk->tasks);
831
832                 tsk->group_leader = tsk;
833                 leader->group_leader = tsk;
834
835                 tsk->exit_signal = SIGCHLD;
836
837                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
838                 leader->exit_state = EXIT_DEAD;
839                 write_unlock_irq(&tasklist_lock);
840
841                 release_task(leader);
842         }
843
844         sig->group_exit_task = NULL;
845         sig->notify_count = 0;
846
847 no_thread_group:
848         exit_itimers(sig);
849         flush_itimer_signals();
850
851         if (atomic_read(&oldsighand->count) != 1) {
852                 struct sighand_struct *newsighand;
853                 /*
854                  * This ->sighand is shared with the CLONE_SIGHAND
855                  * but not CLONE_THREAD task, switch to the new one.
856                  */
857                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
858                 if (!newsighand)
859                         return -ENOMEM;
860
861                 atomic_set(&newsighand->count, 1);
862                 memcpy(newsighand->action, oldsighand->action,
863                        sizeof(newsighand->action));
864
865                 write_lock_irq(&tasklist_lock);
866                 spin_lock(&oldsighand->siglock);
867                 rcu_assign_pointer(tsk->sighand, newsighand);
868                 spin_unlock(&oldsighand->siglock);
869                 write_unlock_irq(&tasklist_lock);
870
871                 __cleanup_sighand(oldsighand);
872         }
873
874         BUG_ON(!thread_group_leader(tsk));
875         return 0;
876 }
877
878 /*
879  * These functions flushes out all traces of the currently running executable
880  * so that a new one can be started
881  */
882 static void flush_old_files(struct files_struct * files)
883 {
884         long j = -1;
885         struct fdtable *fdt;
886
887         spin_lock(&files->file_lock);
888         for (;;) {
889                 unsigned long set, i;
890
891                 j++;
892                 i = j * __NFDBITS;
893                 fdt = files_fdtable(files);
894                 if (i >= fdt->max_fds)
895                         break;
896                 set = fdt->close_on_exec->fds_bits[j];
897                 if (!set)
898                         continue;
899                 fdt->close_on_exec->fds_bits[j] = 0;
900                 spin_unlock(&files->file_lock);
901                 for ( ; set ; i++,set >>= 1) {
902                         if (set & 1) {
903                                 sys_close(i);
904                         }
905                 }
906                 spin_lock(&files->file_lock);
907
908         }
909         spin_unlock(&files->file_lock);
910 }
911
912 char *get_task_comm(char *buf, struct task_struct *tsk)
913 {
914         /* buf must be at least sizeof(tsk->comm) in size */
915         task_lock(tsk);
916         strncpy(buf, tsk->comm, sizeof(tsk->comm));
917         task_unlock(tsk);
918         return buf;
919 }
920
921 void set_task_comm(struct task_struct *tsk, char *buf)
922 {
923         task_lock(tsk);
924         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
925         task_unlock(tsk);
926         perf_event_comm(tsk);
927 }
928
929 int flush_old_exec(struct linux_binprm * bprm)
930 {
931         char * name;
932         int i, ch, retval;
933         char tcomm[sizeof(current->comm)];
934
935         /*
936          * Make sure we have a private signal table and that
937          * we are unassociated from the previous thread group.
938          */
939         retval = de_thread(current);
940         if (retval)
941                 goto out;
942
943         set_mm_exe_file(bprm->mm, bprm->file);
944
945         /*
946          * Release all of the old mmap stuff
947          */
948         retval = exec_mmap(bprm->mm);
949         if (retval)
950                 goto out;
951
952         bprm->mm = NULL;                /* We're using it now */
953
954         /* This is the point of no return */
955         current->sas_ss_sp = current->sas_ss_size = 0;
956
957         if (current_euid() == current_uid() && current_egid() == current_gid())
958                 set_dumpable(current->mm, 1);
959         else
960                 set_dumpable(current->mm, suid_dumpable);
961
962         name = bprm->filename;
963
964         /* Copies the binary name from after last slash */
965         for (i=0; (ch = *(name++)) != '\0';) {
966                 if (ch == '/')
967                         i = 0; /* overwrite what we wrote */
968                 else
969                         if (i < (sizeof(tcomm) - 1))
970                                 tcomm[i++] = ch;
971         }
972         tcomm[i] = '\0';
973         set_task_comm(current, tcomm);
974
975         current->flags &= ~PF_RANDOMIZE;
976         flush_thread();
977
978         /* Set the new mm task size. We have to do that late because it may
979          * depend on TIF_32BIT which is only updated in flush_thread() on
980          * some architectures like powerpc
981          */
982         current->mm->task_size = TASK_SIZE;
983
984         /* install the new credentials */
985         if (bprm->cred->uid != current_euid() ||
986             bprm->cred->gid != current_egid()) {
987                 current->pdeath_signal = 0;
988         } else if (file_permission(bprm->file, MAY_READ) ||
989                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
990                 set_dumpable(current->mm, suid_dumpable);
991         }
992
993         current->personality &= ~bprm->per_clear;
994
995         /*
996          * Flush performance counters when crossing a
997          * security domain:
998          */
999         if (!get_dumpable(current->mm))
1000                 perf_event_exit_task(current);
1001
1002         /* An exec changes our domain. We are no longer part of the thread
1003            group */
1004
1005         current->self_exec_id++;
1006                         
1007         flush_signal_handlers(current, 0);
1008         flush_old_files(current->files);
1009
1010         return 0;
1011
1012 out:
1013         return retval;
1014 }
1015
1016 EXPORT_SYMBOL(flush_old_exec);
1017
1018 /*
1019  * Prepare credentials and lock ->cred_guard_mutex.
1020  * install_exec_creds() commits the new creds and drops the lock.
1021  * Or, if exec fails before, free_bprm() should release ->cred and
1022  * and unlock.
1023  */
1024 int prepare_bprm_creds(struct linux_binprm *bprm)
1025 {
1026         if (mutex_lock_interruptible(&current->cred_guard_mutex))
1027                 return -ERESTARTNOINTR;
1028
1029         bprm->cred = prepare_exec_creds();
1030         if (likely(bprm->cred))
1031                 return 0;
1032
1033         mutex_unlock(&current->cred_guard_mutex);
1034         return -ENOMEM;
1035 }
1036
1037 void free_bprm(struct linux_binprm *bprm)
1038 {
1039         free_arg_pages(bprm);
1040         if (bprm->cred) {
1041                 mutex_unlock(&current->cred_guard_mutex);
1042                 abort_creds(bprm->cred);
1043         }
1044         kfree(bprm);
1045 }
1046
1047 /*
1048  * install the new credentials for this executable
1049  */
1050 void install_exec_creds(struct linux_binprm *bprm)
1051 {
1052         security_bprm_committing_creds(bprm);
1053
1054         commit_creds(bprm->cred);
1055         bprm->cred = NULL;
1056         /*
1057          * cred_guard_mutex must be held at least to this point to prevent
1058          * ptrace_attach() from altering our determination of the task's
1059          * credentials; any time after this it may be unlocked.
1060          */
1061         security_bprm_committed_creds(bprm);
1062         mutex_unlock(&current->cred_guard_mutex);
1063 }
1064 EXPORT_SYMBOL(install_exec_creds);
1065
1066 /*
1067  * determine how safe it is to execute the proposed program
1068  * - the caller must hold current->cred_guard_mutex to protect against
1069  *   PTRACE_ATTACH
1070  */
1071 int check_unsafe_exec(struct linux_binprm *bprm)
1072 {
1073         struct task_struct *p = current, *t;
1074         unsigned n_fs;
1075         int res = 0;
1076
1077         bprm->unsafe = tracehook_unsafe_exec(p);
1078
1079         n_fs = 1;
1080         write_lock(&p->fs->lock);
1081         rcu_read_lock();
1082         for (t = next_thread(p); t != p; t = next_thread(t)) {
1083                 if (t->fs == p->fs)
1084                         n_fs++;
1085         }
1086         rcu_read_unlock();
1087
1088         if (p->fs->users > n_fs) {
1089                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1090         } else {
1091                 res = -EAGAIN;
1092                 if (!p->fs->in_exec) {
1093                         p->fs->in_exec = 1;
1094                         res = 1;
1095                 }
1096         }
1097         write_unlock(&p->fs->lock);
1098
1099         return res;
1100 }
1101
1102 /* 
1103  * Fill the binprm structure from the inode. 
1104  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1105  *
1106  * This may be called multiple times for binary chains (scripts for example).
1107  */
1108 int prepare_binprm(struct linux_binprm *bprm)
1109 {
1110         umode_t mode;
1111         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1112         int retval;
1113
1114         mode = inode->i_mode;
1115         if (bprm->file->f_op == NULL)
1116                 return -EACCES;
1117
1118         /* clear any previous set[ug]id data from a previous binary */
1119         bprm->cred->euid = current_euid();
1120         bprm->cred->egid = current_egid();
1121
1122         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1123                 /* Set-uid? */
1124                 if (mode & S_ISUID) {
1125                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1126                         bprm->cred->euid = inode->i_uid;
1127                 }
1128
1129                 /* Set-gid? */
1130                 /*
1131                  * If setgid is set but no group execute bit then this
1132                  * is a candidate for mandatory locking, not a setgid
1133                  * executable.
1134                  */
1135                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1136                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1137                         bprm->cred->egid = inode->i_gid;
1138                 }
1139         }
1140
1141         /* fill in binprm security blob */
1142         retval = security_bprm_set_creds(bprm);
1143         if (retval)
1144                 return retval;
1145         bprm->cred_prepared = 1;
1146
1147         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1148         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1149 }
1150
1151 EXPORT_SYMBOL(prepare_binprm);
1152
1153 /*
1154  * Arguments are '\0' separated strings found at the location bprm->p
1155  * points to; chop off the first by relocating brpm->p to right after
1156  * the first '\0' encountered.
1157  */
1158 int remove_arg_zero(struct linux_binprm *bprm)
1159 {
1160         int ret = 0;
1161         unsigned long offset;
1162         char *kaddr;
1163         struct page *page;
1164
1165         if (!bprm->argc)
1166                 return 0;
1167
1168         do {
1169                 offset = bprm->p & ~PAGE_MASK;
1170                 page = get_arg_page(bprm, bprm->p, 0);
1171                 if (!page) {
1172                         ret = -EFAULT;
1173                         goto out;
1174                 }
1175                 kaddr = kmap_atomic(page, KM_USER0);
1176
1177                 for (; offset < PAGE_SIZE && kaddr[offset];
1178                                 offset++, bprm->p++)
1179                         ;
1180
1181                 kunmap_atomic(kaddr, KM_USER0);
1182                 put_arg_page(page);
1183
1184                 if (offset == PAGE_SIZE)
1185                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1186         } while (offset == PAGE_SIZE);
1187
1188         bprm->p++;
1189         bprm->argc--;
1190         ret = 0;
1191
1192 out:
1193         return ret;
1194 }
1195 EXPORT_SYMBOL(remove_arg_zero);
1196
1197 /*
1198  * cycle the list of binary formats handler, until one recognizes the image
1199  */
1200 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1201 {
1202         unsigned int depth = bprm->recursion_depth;
1203         int try,retval;
1204         struct linux_binfmt *fmt;
1205
1206         retval = security_bprm_check(bprm);
1207         if (retval)
1208                 return retval;
1209         retval = ima_bprm_check(bprm);
1210         if (retval)
1211                 return retval;
1212
1213         /* kernel module loader fixup */
1214         /* so we don't try to load run modprobe in kernel space. */
1215         set_fs(USER_DS);
1216
1217         retval = audit_bprm(bprm);
1218         if (retval)
1219                 return retval;
1220
1221         retval = -ENOENT;
1222         for (try=0; try<2; try++) {
1223                 read_lock(&binfmt_lock);
1224                 list_for_each_entry(fmt, &formats, lh) {
1225                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1226                         if (!fn)
1227                                 continue;
1228                         if (!try_module_get(fmt->module))
1229                                 continue;
1230                         read_unlock(&binfmt_lock);
1231                         retval = fn(bprm, regs);
1232                         /*
1233                          * Restore the depth counter to its starting value
1234                          * in this call, so we don't have to rely on every
1235                          * load_binary function to restore it on return.
1236                          */
1237                         bprm->recursion_depth = depth;
1238                         if (retval >= 0) {
1239                                 if (depth == 0)
1240                                         tracehook_report_exec(fmt, bprm, regs);
1241                                 put_binfmt(fmt);
1242                                 allow_write_access(bprm->file);
1243                                 if (bprm->file)
1244                                         fput(bprm->file);
1245                                 bprm->file = NULL;
1246                                 current->did_exec = 1;
1247                                 proc_exec_connector(current);
1248                                 return retval;
1249                         }
1250                         read_lock(&binfmt_lock);
1251                         put_binfmt(fmt);
1252                         if (retval != -ENOEXEC || bprm->mm == NULL)
1253                                 break;
1254                         if (!bprm->file) {
1255                                 read_unlock(&binfmt_lock);
1256                                 return retval;
1257                         }
1258                 }
1259                 read_unlock(&binfmt_lock);
1260                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1261                         break;
1262 #ifdef CONFIG_MODULES
1263                 } else {
1264 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1265                         if (printable(bprm->buf[0]) &&
1266                             printable(bprm->buf[1]) &&
1267                             printable(bprm->buf[2]) &&
1268                             printable(bprm->buf[3]))
1269                                 break; /* -ENOEXEC */
1270                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1271 #endif
1272                 }
1273         }
1274         return retval;
1275 }
1276
1277 EXPORT_SYMBOL(search_binary_handler);
1278
1279 /*
1280  * sys_execve() executes a new program.
1281  */
1282 int do_execve(char * filename,
1283         char __user *__user *argv,
1284         char __user *__user *envp,
1285         struct pt_regs * regs)
1286 {
1287         struct linux_binprm *bprm;
1288         struct file *file;
1289         struct files_struct *displaced;
1290         bool clear_in_exec;
1291         int retval;
1292
1293         retval = unshare_files(&displaced);
1294         if (retval)
1295                 goto out_ret;
1296
1297         retval = -ENOMEM;
1298         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1299         if (!bprm)
1300                 goto out_files;
1301
1302         retval = prepare_bprm_creds(bprm);
1303         if (retval)
1304                 goto out_free;
1305
1306         retval = check_unsafe_exec(bprm);
1307         if (retval < 0)
1308                 goto out_free;
1309         clear_in_exec = retval;
1310         current->in_execve = 1;
1311
1312         file = open_exec(filename);
1313         retval = PTR_ERR(file);
1314         if (IS_ERR(file))
1315                 goto out_unmark;
1316
1317         sched_exec();
1318
1319         bprm->file = file;
1320         bprm->filename = filename;
1321         bprm->interp = filename;
1322
1323         retval = bprm_mm_init(bprm);
1324         if (retval)
1325                 goto out_file;
1326
1327         bprm->argc = count(argv, MAX_ARG_STRINGS);
1328         if ((retval = bprm->argc) < 0)
1329                 goto out;
1330
1331         bprm->envc = count(envp, MAX_ARG_STRINGS);
1332         if ((retval = bprm->envc) < 0)
1333                 goto out;
1334
1335         retval = prepare_binprm(bprm);
1336         if (retval < 0)
1337                 goto out;
1338
1339         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1340         if (retval < 0)
1341                 goto out;
1342
1343         bprm->exec = bprm->p;
1344         retval = copy_strings(bprm->envc, envp, bprm);
1345         if (retval < 0)
1346                 goto out;
1347
1348         retval = copy_strings(bprm->argc, argv, bprm);
1349         if (retval < 0)
1350                 goto out;
1351
1352         current->flags &= ~PF_KTHREAD;
1353         retval = search_binary_handler(bprm,regs);
1354         if (retval < 0)
1355                 goto out;
1356
1357         /* execve succeeded */
1358         current->fs->in_exec = 0;
1359         current->in_execve = 0;
1360         acct_update_integrals(current);
1361         free_bprm(bprm);
1362         if (displaced)
1363                 put_files_struct(displaced);
1364         return retval;
1365
1366 out:
1367         if (bprm->mm)
1368                 mmput (bprm->mm);
1369
1370 out_file:
1371         if (bprm->file) {
1372                 allow_write_access(bprm->file);
1373                 fput(bprm->file);
1374         }
1375
1376 out_unmark:
1377         if (clear_in_exec)
1378                 current->fs->in_exec = 0;
1379         current->in_execve = 0;
1380
1381 out_free:
1382         free_bprm(bprm);
1383
1384 out_files:
1385         if (displaced)
1386                 reset_files_struct(displaced);
1387 out_ret:
1388         return retval;
1389 }
1390
1391 int set_binfmt(struct linux_binfmt *new)
1392 {
1393         struct linux_binfmt *old = current->binfmt;
1394
1395         if (new) {
1396                 if (!try_module_get(new->module))
1397                         return -1;
1398         }
1399         current->binfmt = new;
1400         if (old)
1401                 module_put(old->module);
1402         return 0;
1403 }
1404
1405 EXPORT_SYMBOL(set_binfmt);
1406
1407 /* format_corename will inspect the pattern parameter, and output a
1408  * name into corename, which must have space for at least
1409  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1410  */
1411 static int format_corename(char *corename, long signr)
1412 {
1413         const struct cred *cred = current_cred();
1414         const char *pat_ptr = core_pattern;
1415         int ispipe = (*pat_ptr == '|');
1416         char *out_ptr = corename;
1417         char *const out_end = corename + CORENAME_MAX_SIZE;
1418         int rc;
1419         int pid_in_pattern = 0;
1420
1421         /* Repeat as long as we have more pattern to process and more output
1422            space */
1423         while (*pat_ptr) {
1424                 if (*pat_ptr != '%') {
1425                         if (out_ptr == out_end)
1426                                 goto out;
1427                         *out_ptr++ = *pat_ptr++;
1428                 } else {
1429                         switch (*++pat_ptr) {
1430                         case 0:
1431                                 goto out;
1432                         /* Double percent, output one percent */
1433                         case '%':
1434                                 if (out_ptr == out_end)
1435                                         goto out;
1436                                 *out_ptr++ = '%';
1437                                 break;
1438                         /* pid */
1439                         case 'p':
1440                                 pid_in_pattern = 1;
1441                                 rc = snprintf(out_ptr, out_end - out_ptr,
1442                                               "%d", task_tgid_vnr(current));
1443                                 if (rc > out_end - out_ptr)
1444                                         goto out;
1445                                 out_ptr += rc;
1446                                 break;
1447                         /* uid */
1448                         case 'u':
1449                                 rc = snprintf(out_ptr, out_end - out_ptr,
1450                                               "%d", cred->uid);
1451                                 if (rc > out_end - out_ptr)
1452                                         goto out;
1453                                 out_ptr += rc;
1454                                 break;
1455                         /* gid */
1456                         case 'g':
1457                                 rc = snprintf(out_ptr, out_end - out_ptr,
1458                                               "%d", cred->gid);
1459                                 if (rc > out_end - out_ptr)
1460                                         goto out;
1461                                 out_ptr += rc;
1462                                 break;
1463                         /* signal that caused the coredump */
1464                         case 's':
1465                                 rc = snprintf(out_ptr, out_end - out_ptr,
1466                                               "%ld", signr);
1467                                 if (rc > out_end - out_ptr)
1468                                         goto out;
1469                                 out_ptr += rc;
1470                                 break;
1471                         /* UNIX time of coredump */
1472                         case 't': {
1473                                 struct timeval tv;
1474                                 do_gettimeofday(&tv);
1475                                 rc = snprintf(out_ptr, out_end - out_ptr,
1476                                               "%lu", tv.tv_sec);
1477                                 if (rc > out_end - out_ptr)
1478                                         goto out;
1479                                 out_ptr += rc;
1480                                 break;
1481                         }
1482                         /* hostname */
1483                         case 'h':
1484                                 down_read(&uts_sem);
1485                                 rc = snprintf(out_ptr, out_end - out_ptr,
1486                                               "%s", utsname()->nodename);
1487                                 up_read(&uts_sem);
1488                                 if (rc > out_end - out_ptr)
1489                                         goto out;
1490                                 out_ptr += rc;
1491                                 break;
1492                         /* executable */
1493                         case 'e':
1494                                 rc = snprintf(out_ptr, out_end - out_ptr,
1495                                               "%s", current->comm);
1496                                 if (rc > out_end - out_ptr)
1497                                         goto out;
1498                                 out_ptr += rc;
1499                                 break;
1500                         /* core limit size */
1501                         case 'c':
1502                                 rc = snprintf(out_ptr, out_end - out_ptr,
1503                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1504                                 if (rc > out_end - out_ptr)
1505                                         goto out;
1506                                 out_ptr += rc;
1507                                 break;
1508                         default:
1509                                 break;
1510                         }
1511                         ++pat_ptr;
1512                 }
1513         }
1514         /* Backward compatibility with core_uses_pid:
1515          *
1516          * If core_pattern does not include a %p (as is the default)
1517          * and core_uses_pid is set, then .%pid will be appended to
1518          * the filename. Do not do this for piped commands. */
1519         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1520                 rc = snprintf(out_ptr, out_end - out_ptr,
1521                               ".%d", task_tgid_vnr(current));
1522                 if (rc > out_end - out_ptr)
1523                         goto out;
1524                 out_ptr += rc;
1525         }
1526 out:
1527         *out_ptr = 0;
1528         return ispipe;
1529 }
1530
1531 static int zap_process(struct task_struct *start)
1532 {
1533         struct task_struct *t;
1534         int nr = 0;
1535
1536         start->signal->flags = SIGNAL_GROUP_EXIT;
1537         start->signal->group_stop_count = 0;
1538
1539         t = start;
1540         do {
1541                 if (t != current && t->mm) {
1542                         sigaddset(&t->pending.signal, SIGKILL);
1543                         signal_wake_up(t, 1);
1544                         nr++;
1545                 }
1546         } while_each_thread(start, t);
1547
1548         return nr;
1549 }
1550
1551 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1552                                 struct core_state *core_state, int exit_code)
1553 {
1554         struct task_struct *g, *p;
1555         unsigned long flags;
1556         int nr = -EAGAIN;
1557
1558         spin_lock_irq(&tsk->sighand->siglock);
1559         if (!signal_group_exit(tsk->signal)) {
1560                 mm->core_state = core_state;
1561                 tsk->signal->group_exit_code = exit_code;
1562                 nr = zap_process(tsk);
1563         }
1564         spin_unlock_irq(&tsk->sighand->siglock);
1565         if (unlikely(nr < 0))
1566                 return nr;
1567
1568         if (atomic_read(&mm->mm_users) == nr + 1)
1569                 goto done;
1570         /*
1571          * We should find and kill all tasks which use this mm, and we should
1572          * count them correctly into ->nr_threads. We don't take tasklist
1573          * lock, but this is safe wrt:
1574          *
1575          * fork:
1576          *      None of sub-threads can fork after zap_process(leader). All
1577          *      processes which were created before this point should be
1578          *      visible to zap_threads() because copy_process() adds the new
1579          *      process to the tail of init_task.tasks list, and lock/unlock
1580          *      of ->siglock provides a memory barrier.
1581          *
1582          * do_exit:
1583          *      The caller holds mm->mmap_sem. This means that the task which
1584          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1585          *      its ->mm.
1586          *
1587          * de_thread:
1588          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1589          *      we must see either old or new leader, this does not matter.
1590          *      However, it can change p->sighand, so lock_task_sighand(p)
1591          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1592          *      it can't fail.
1593          *
1594          *      Note also that "g" can be the old leader with ->mm == NULL
1595          *      and already unhashed and thus removed from ->thread_group.
1596          *      This is OK, __unhash_process()->list_del_rcu() does not
1597          *      clear the ->next pointer, we will find the new leader via
1598          *      next_thread().
1599          */
1600         rcu_read_lock();
1601         for_each_process(g) {
1602                 if (g == tsk->group_leader)
1603                         continue;
1604                 if (g->flags & PF_KTHREAD)
1605                         continue;
1606                 p = g;
1607                 do {
1608                         if (p->mm) {
1609                                 if (unlikely(p->mm == mm)) {
1610                                         lock_task_sighand(p, &flags);
1611                                         nr += zap_process(p);
1612                                         unlock_task_sighand(p, &flags);
1613                                 }
1614                                 break;
1615                         }
1616                 } while_each_thread(g, p);
1617         }
1618         rcu_read_unlock();
1619 done:
1620         atomic_set(&core_state->nr_threads, nr);
1621         return nr;
1622 }
1623
1624 static int coredump_wait(int exit_code, struct core_state *core_state)
1625 {
1626         struct task_struct *tsk = current;
1627         struct mm_struct *mm = tsk->mm;
1628         struct completion *vfork_done;
1629         int core_waiters;
1630
1631         init_completion(&core_state->startup);
1632         core_state->dumper.task = tsk;
1633         core_state->dumper.next = NULL;
1634         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1635         up_write(&mm->mmap_sem);
1636
1637         if (unlikely(core_waiters < 0))
1638                 goto fail;
1639
1640         /*
1641          * Make sure nobody is waiting for us to release the VM,
1642          * otherwise we can deadlock when we wait on each other
1643          */
1644         vfork_done = tsk->vfork_done;
1645         if (vfork_done) {
1646                 tsk->vfork_done = NULL;
1647                 complete(vfork_done);
1648         }
1649
1650         if (core_waiters)
1651                 wait_for_completion(&core_state->startup);
1652 fail:
1653         return core_waiters;
1654 }
1655
1656 static void coredump_finish(struct mm_struct *mm)
1657 {
1658         struct core_thread *curr, *next;
1659         struct task_struct *task;
1660
1661         next = mm->core_state->dumper.next;
1662         while ((curr = next) != NULL) {
1663                 next = curr->next;
1664                 task = curr->task;
1665                 /*
1666                  * see exit_mm(), curr->task must not see
1667                  * ->task == NULL before we read ->next.
1668                  */
1669                 smp_mb();
1670                 curr->task = NULL;
1671                 wake_up_process(task);
1672         }
1673
1674         mm->core_state = NULL;
1675 }
1676
1677 /*
1678  * set_dumpable converts traditional three-value dumpable to two flags and
1679  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1680  * these bits are not changed atomically.  So get_dumpable can observe the
1681  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1682  * return either old dumpable or new one by paying attention to the order of
1683  * modifying the bits.
1684  *
1685  * dumpable |   mm->flags (binary)
1686  * old  new | initial interim  final
1687  * ---------+-----------------------
1688  *  0    1  |   00      01      01
1689  *  0    2  |   00      10(*)   11
1690  *  1    0  |   01      00      00
1691  *  1    2  |   01      11      11
1692  *  2    0  |   11      10(*)   00
1693  *  2    1  |   11      11      01
1694  *
1695  * (*) get_dumpable regards interim value of 10 as 11.
1696  */
1697 void set_dumpable(struct mm_struct *mm, int value)
1698 {
1699         switch (value) {
1700         case 0:
1701                 clear_bit(MMF_DUMPABLE, &mm->flags);
1702                 smp_wmb();
1703                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1704                 break;
1705         case 1:
1706                 set_bit(MMF_DUMPABLE, &mm->flags);
1707                 smp_wmb();
1708                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1709                 break;
1710         case 2:
1711                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1712                 smp_wmb();
1713                 set_bit(MMF_DUMPABLE, &mm->flags);
1714                 break;
1715         }
1716 }
1717
1718 int get_dumpable(struct mm_struct *mm)
1719 {
1720         int ret;
1721
1722         ret = mm->flags & 0x3;
1723         return (ret >= 2) ? 2 : ret;
1724 }
1725
1726 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1727 {
1728         struct core_state core_state;
1729         char corename[CORENAME_MAX_SIZE + 1];
1730         struct mm_struct *mm = current->mm;
1731         struct linux_binfmt * binfmt;
1732         struct inode * inode;
1733         struct file * file;
1734         const struct cred *old_cred;
1735         struct cred *cred;
1736         int retval = 0;
1737         int flag = 0;
1738         int ispipe = 0;
1739         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1740         char **helper_argv = NULL;
1741         int helper_argc = 0;
1742         char *delimit;
1743
1744         audit_core_dumps(signr);
1745
1746         binfmt = current->binfmt;
1747         if (!binfmt || !binfmt->core_dump)
1748                 goto fail;
1749
1750         cred = prepare_creds();
1751         if (!cred) {
1752                 retval = -ENOMEM;
1753                 goto fail;
1754         }
1755
1756         down_write(&mm->mmap_sem);
1757         /*
1758          * If another thread got here first, or we are not dumpable, bail out.
1759          */
1760         if (mm->core_state || !get_dumpable(mm)) {
1761                 up_write(&mm->mmap_sem);
1762                 put_cred(cred);
1763                 goto fail;
1764         }
1765
1766         /*
1767          *      We cannot trust fsuid as being the "true" uid of the
1768          *      process nor do we know its entire history. We only know it
1769          *      was tainted so we dump it as root in mode 2.
1770          */
1771         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1772                 flag = O_EXCL;          /* Stop rewrite attacks */
1773                 cred->fsuid = 0;        /* Dump root private */
1774         }
1775
1776         retval = coredump_wait(exit_code, &core_state);
1777         if (retval < 0) {
1778                 put_cred(cred);
1779                 goto fail;
1780         }
1781
1782         old_cred = override_creds(cred);
1783
1784         /*
1785          * Clear any false indication of pending signals that might
1786          * be seen by the filesystem code called to write the core file.
1787          */
1788         clear_thread_flag(TIF_SIGPENDING);
1789
1790         /*
1791          * lock_kernel() because format_corename() is controlled by sysctl, which
1792          * uses lock_kernel()
1793          */
1794         lock_kernel();
1795         ispipe = format_corename(corename, signr);
1796         unlock_kernel();
1797         /*
1798          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1799          * to a pipe.  Since we're not writing directly to the filesystem
1800          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1801          * created unless the pipe reader choses to write out the core file
1802          * at which point file size limits and permissions will be imposed
1803          * as it does with any other process
1804          */
1805         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1806                 goto fail_unlock;
1807
1808         if (ispipe) {
1809                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1810                 if (!helper_argv) {
1811                         printk(KERN_WARNING "%s failed to allocate memory\n",
1812                                __func__);
1813                         goto fail_unlock;
1814                 }
1815                 /* Terminate the string before the first option */
1816                 delimit = strchr(corename, ' ');
1817                 if (delimit)
1818                         *delimit = '\0';
1819                 delimit = strrchr(helper_argv[0], '/');
1820                 if (delimit)
1821                         delimit++;
1822                 else
1823                         delimit = helper_argv[0];
1824                 if (!strcmp(delimit, current->comm)) {
1825                         printk(KERN_NOTICE "Recursive core dump detected, "
1826                                         "aborting\n");
1827                         goto fail_unlock;
1828                 }
1829
1830                 core_limit = RLIM_INFINITY;
1831
1832                 /* SIGPIPE can happen, but it's just never processed */
1833                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1834                                 &file)) {
1835                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1836                                corename);
1837                         goto fail_unlock;
1838                 }
1839         } else
1840                 file = filp_open(corename,
1841                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1842                                  0600);
1843         if (IS_ERR(file))
1844                 goto fail_unlock;
1845         inode = file->f_path.dentry->d_inode;
1846         if (inode->i_nlink > 1)
1847                 goto close_fail;        /* multiple links - don't dump */
1848         if (!ispipe && d_unhashed(file->f_path.dentry))
1849                 goto close_fail;
1850
1851         /* AK: actually i see no reason to not allow this for named pipes etc.,
1852            but keep the previous behaviour for now. */
1853         if (!ispipe && !S_ISREG(inode->i_mode))
1854                 goto close_fail;
1855         /*
1856          * Dont allow local users get cute and trick others to coredump
1857          * into their pre-created files:
1858          */
1859         if (inode->i_uid != current_fsuid())
1860                 goto close_fail;
1861         if (!file->f_op)
1862                 goto close_fail;
1863         if (!file->f_op->write)
1864                 goto close_fail;
1865         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1866                 goto close_fail;
1867
1868         retval = binfmt->core_dump(signr, regs, file, core_limit);
1869
1870         if (retval)
1871                 current->signal->group_exit_code |= 0x80;
1872 close_fail:
1873         filp_close(file, NULL);
1874 fail_unlock:
1875         if (helper_argv)
1876                 argv_free(helper_argv);
1877
1878         revert_creds(old_cred);
1879         put_cred(cred);
1880         coredump_finish(mm);
1881 fail:
1882         return;
1883 }