Switch open_exec() and sys_uselib() to do_open_filp()
[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/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/ima.h>
49 #include <linux/syscalls.h>
50 #include <linux/tsacct_kern.h>
51 #include <linux/cn_proc.h>
52 #include <linux/audit.h>
53 #include <linux/tracehook.h>
54 #include <linux/kmod.h>
55 #include <linux/fsnotify.h>
56 #include <linux/fs_struct.h>
57
58 #include <asm/uaccess.h>
59 #include <asm/mmu_context.h>
60 #include <asm/tlb.h>
61 #include "internal.h"
62
63 int core_uses_pid;
64 char core_pattern[CORENAME_MAX_SIZE] = "core";
65 int suid_dumpable = 0;
66
67 /* The maximal length of core_pattern is also specified in sysctl.c */
68
69 static LIST_HEAD(formats);
70 static DEFINE_RWLOCK(binfmt_lock);
71
72 int __register_binfmt(struct linux_binfmt * fmt, int insert)
73 {
74         if (!fmt)
75                 return -EINVAL;
76         write_lock(&binfmt_lock);
77         insert ? list_add(&fmt->lh, &formats) :
78                  list_add_tail(&fmt->lh, &formats);
79         write_unlock(&binfmt_lock);
80         return 0;       
81 }
82
83 EXPORT_SYMBOL(__register_binfmt);
84
85 void unregister_binfmt(struct linux_binfmt * fmt)
86 {
87         write_lock(&binfmt_lock);
88         list_del(&fmt->lh);
89         write_unlock(&binfmt_lock);
90 }
91
92 EXPORT_SYMBOL(unregister_binfmt);
93
94 static inline void put_binfmt(struct linux_binfmt * fmt)
95 {
96         module_put(fmt->module);
97 }
98
99 /*
100  * Note that a shared library must be both readable and executable due to
101  * security reasons.
102  *
103  * Also note that we take the address to load from from the file itself.
104  */
105 SYSCALL_DEFINE1(uselib, const char __user *, library)
106 {
107         struct file *file;
108         char *tmp = getname(library);
109         int error = PTR_ERR(tmp);
110
111         if (IS_ERR(tmp))
112                 goto out;
113
114         file = do_filp_open(AT_FDCWD, tmp,
115                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
116                                 MAY_READ | MAY_EXEC | MAY_OPEN);
117         putname(tmp);
118         error = PTR_ERR(file);
119         if (IS_ERR(file))
120                 goto out;
121
122         error = -EINVAL;
123         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
124                 goto exit;
125
126         error = -EACCES;
127         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
128                 goto exit;
129
130         fsnotify_open(file->f_path.dentry);
131
132         error = -ENOEXEC;
133         if(file->f_op) {
134                 struct linux_binfmt * fmt;
135
136                 read_lock(&binfmt_lock);
137                 list_for_each_entry(fmt, &formats, lh) {
138                         if (!fmt->load_shlib)
139                                 continue;
140                         if (!try_module_get(fmt->module))
141                                 continue;
142                         read_unlock(&binfmt_lock);
143                         error = fmt->load_shlib(file);
144                         read_lock(&binfmt_lock);
145                         put_binfmt(fmt);
146                         if (error != -ENOEXEC)
147                                 break;
148                 }
149                 read_unlock(&binfmt_lock);
150         }
151 exit:
152         fput(file);
153 out:
154         return error;
155 }
156
157 #ifdef CONFIG_MMU
158
159 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
160                 int write)
161 {
162         struct page *page;
163         int ret;
164
165 #ifdef CONFIG_STACK_GROWSUP
166         if (write) {
167                 ret = expand_stack_downwards(bprm->vma, pos);
168                 if (ret < 0)
169                         return NULL;
170         }
171 #endif
172         ret = get_user_pages(current, bprm->mm, pos,
173                         1, write, 1, &page, NULL);
174         if (ret <= 0)
175                 return NULL;
176
177         if (write) {
178                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
179                 struct rlimit *rlim;
180
181                 /*
182                  * We've historically supported up to 32 pages (ARG_MAX)
183                  * of argument strings even with small stacks
184                  */
185                 if (size <= ARG_MAX)
186                         return page;
187
188                 /*
189                  * Limit to 1/4-th the stack size for the argv+env strings.
190                  * This ensures that:
191                  *  - the remaining binfmt code will not run out of stack space,
192                  *  - the program will have a reasonable amount of stack left
193                  *    to work from.
194                  */
195                 rlim = current->signal->rlim;
196                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
197                         put_page(page);
198                         return NULL;
199                 }
200         }
201
202         return page;
203 }
204
205 static void put_arg_page(struct page *page)
206 {
207         put_page(page);
208 }
209
210 static void free_arg_page(struct linux_binprm *bprm, int i)
211 {
212 }
213
214 static void free_arg_pages(struct linux_binprm *bprm)
215 {
216 }
217
218 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
219                 struct page *page)
220 {
221         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
222 }
223
224 static int __bprm_mm_init(struct linux_binprm *bprm)
225 {
226         int err;
227         struct vm_area_struct *vma = NULL;
228         struct mm_struct *mm = bprm->mm;
229
230         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
231         if (!vma)
232                 return -ENOMEM;
233
234         down_write(&mm->mmap_sem);
235         vma->vm_mm = mm;
236
237         /*
238          * Place the stack at the largest stack address the architecture
239          * supports. Later, we'll move this to an appropriate place. We don't
240          * use STACK_TOP because that can depend on attributes which aren't
241          * configured yet.
242          */
243         vma->vm_end = STACK_TOP_MAX;
244         vma->vm_start = vma->vm_end - PAGE_SIZE;
245         vma->vm_flags = VM_STACK_FLAGS;
246         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
247         err = insert_vm_struct(mm, vma);
248         if (err)
249                 goto err;
250
251         mm->stack_vm = mm->total_vm = 1;
252         up_write(&mm->mmap_sem);
253         bprm->p = vma->vm_end - sizeof(void *);
254         return 0;
255 err:
256         up_write(&mm->mmap_sem);
257         bprm->vma = NULL;
258         kmem_cache_free(vm_area_cachep, vma);
259         return err;
260 }
261
262 static bool valid_arg_len(struct linux_binprm *bprm, long len)
263 {
264         return len <= MAX_ARG_STRLEN;
265 }
266
267 #else
268
269 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
270                 int write)
271 {
272         struct page *page;
273
274         page = bprm->page[pos / PAGE_SIZE];
275         if (!page && write) {
276                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
277                 if (!page)
278                         return NULL;
279                 bprm->page[pos / PAGE_SIZE] = page;
280         }
281
282         return page;
283 }
284
285 static void put_arg_page(struct page *page)
286 {
287 }
288
289 static void free_arg_page(struct linux_binprm *bprm, int i)
290 {
291         if (bprm->page[i]) {
292                 __free_page(bprm->page[i]);
293                 bprm->page[i] = NULL;
294         }
295 }
296
297 static void free_arg_pages(struct linux_binprm *bprm)
298 {
299         int i;
300
301         for (i = 0; i < MAX_ARG_PAGES; i++)
302                 free_arg_page(bprm, i);
303 }
304
305 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
306                 struct page *page)
307 {
308 }
309
310 static int __bprm_mm_init(struct linux_binprm *bprm)
311 {
312         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
313         return 0;
314 }
315
316 static bool valid_arg_len(struct linux_binprm *bprm, long len)
317 {
318         return len <= bprm->p;
319 }
320
321 #endif /* CONFIG_MMU */
322
323 /*
324  * Create a new mm_struct and populate it with a temporary stack
325  * vm_area_struct.  We don't have enough context at this point to set the stack
326  * flags, permissions, and offset, so we use temporary values.  We'll update
327  * them later in setup_arg_pages().
328  */
329 int bprm_mm_init(struct linux_binprm *bprm)
330 {
331         int err;
332         struct mm_struct *mm = NULL;
333
334         bprm->mm = mm = mm_alloc();
335         err = -ENOMEM;
336         if (!mm)
337                 goto err;
338
339         err = init_new_context(current, mm);
340         if (err)
341                 goto err;
342
343         err = __bprm_mm_init(bprm);
344         if (err)
345                 goto err;
346
347         return 0;
348
349 err:
350         if (mm) {
351                 bprm->mm = NULL;
352                 mmdrop(mm);
353         }
354
355         return err;
356 }
357
358 /*
359  * count() counts the number of strings in array ARGV.
360  */
361 static int count(char __user * __user * argv, int max)
362 {
363         int i = 0;
364
365         if (argv != NULL) {
366                 for (;;) {
367                         char __user * p;
368
369                         if (get_user(p, argv))
370                                 return -EFAULT;
371                         if (!p)
372                                 break;
373                         argv++;
374                         if (i++ >= max)
375                                 return -E2BIG;
376                         cond_resched();
377                 }
378         }
379         return i;
380 }
381
382 /*
383  * 'copy_strings()' copies argument/environment strings from the old
384  * processes's memory to the new process's stack.  The call to get_user_pages()
385  * ensures the destination page is created and not swapped out.
386  */
387 static int copy_strings(int argc, char __user * __user * argv,
388                         struct linux_binprm *bprm)
389 {
390         struct page *kmapped_page = NULL;
391         char *kaddr = NULL;
392         unsigned long kpos = 0;
393         int ret;
394
395         while (argc-- > 0) {
396                 char __user *str;
397                 int len;
398                 unsigned long pos;
399
400                 if (get_user(str, argv+argc) ||
401                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
402                         ret = -EFAULT;
403                         goto out;
404                 }
405
406                 if (!valid_arg_len(bprm, len)) {
407                         ret = -E2BIG;
408                         goto out;
409                 }
410
411                 /* We're going to work our way backwords. */
412                 pos = bprm->p;
413                 str += len;
414                 bprm->p -= len;
415
416                 while (len > 0) {
417                         int offset, bytes_to_copy;
418
419                         offset = pos % PAGE_SIZE;
420                         if (offset == 0)
421                                 offset = PAGE_SIZE;
422
423                         bytes_to_copy = offset;
424                         if (bytes_to_copy > len)
425                                 bytes_to_copy = len;
426
427                         offset -= bytes_to_copy;
428                         pos -= bytes_to_copy;
429                         str -= bytes_to_copy;
430                         len -= bytes_to_copy;
431
432                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
433                                 struct page *page;
434
435                                 page = get_arg_page(bprm, pos, 1);
436                                 if (!page) {
437                                         ret = -E2BIG;
438                                         goto out;
439                                 }
440
441                                 if (kmapped_page) {
442                                         flush_kernel_dcache_page(kmapped_page);
443                                         kunmap(kmapped_page);
444                                         put_arg_page(kmapped_page);
445                                 }
446                                 kmapped_page = page;
447                                 kaddr = kmap(kmapped_page);
448                                 kpos = pos & PAGE_MASK;
449                                 flush_arg_page(bprm, kpos, kmapped_page);
450                         }
451                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
452                                 ret = -EFAULT;
453                                 goto out;
454                         }
455                 }
456         }
457         ret = 0;
458 out:
459         if (kmapped_page) {
460                 flush_kernel_dcache_page(kmapped_page);
461                 kunmap(kmapped_page);
462                 put_arg_page(kmapped_page);
463         }
464         return ret;
465 }
466
467 /*
468  * Like copy_strings, but get argv and its values from kernel memory.
469  */
470 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
471 {
472         int r;
473         mm_segment_t oldfs = get_fs();
474         set_fs(KERNEL_DS);
475         r = copy_strings(argc, (char __user * __user *)argv, bprm);
476         set_fs(oldfs);
477         return r;
478 }
479 EXPORT_SYMBOL(copy_strings_kernel);
480
481 #ifdef CONFIG_MMU
482
483 /*
484  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
485  * the binfmt code determines where the new stack should reside, we shift it to
486  * its final location.  The process proceeds as follows:
487  *
488  * 1) Use shift to calculate the new vma endpoints.
489  * 2) Extend vma to cover both the old and new ranges.  This ensures the
490  *    arguments passed to subsequent functions are consistent.
491  * 3) Move vma's page tables to the new range.
492  * 4) Free up any cleared pgd range.
493  * 5) Shrink the vma to cover only the new range.
494  */
495 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
496 {
497         struct mm_struct *mm = vma->vm_mm;
498         unsigned long old_start = vma->vm_start;
499         unsigned long old_end = vma->vm_end;
500         unsigned long length = old_end - old_start;
501         unsigned long new_start = old_start - shift;
502         unsigned long new_end = old_end - shift;
503         struct mmu_gather *tlb;
504
505         BUG_ON(new_start > new_end);
506
507         /*
508          * ensure there are no vmas between where we want to go
509          * and where we are
510          */
511         if (vma != find_vma(mm, new_start))
512                 return -EFAULT;
513
514         /*
515          * cover the whole range: [new_start, old_end)
516          */
517         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
518
519         /*
520          * move the page tables downwards, on failure we rely on
521          * process cleanup to remove whatever mess we made.
522          */
523         if (length != move_page_tables(vma, old_start,
524                                        vma, new_start, length))
525                 return -ENOMEM;
526
527         lru_add_drain();
528         tlb = tlb_gather_mmu(mm, 0);
529         if (new_end > old_start) {
530                 /*
531                  * when the old and new regions overlap clear from new_end.
532                  */
533                 free_pgd_range(tlb, new_end, old_end, new_end,
534                         vma->vm_next ? vma->vm_next->vm_start : 0);
535         } else {
536                 /*
537                  * otherwise, clean from old_start; this is done to not touch
538                  * the address space in [new_end, old_start) some architectures
539                  * have constraints on va-space that make this illegal (IA64) -
540                  * for the others its just a little faster.
541                  */
542                 free_pgd_range(tlb, old_start, old_end, new_end,
543                         vma->vm_next ? vma->vm_next->vm_start : 0);
544         }
545         tlb_finish_mmu(tlb, new_end, old_end);
546
547         /*
548          * shrink the vma to just the new range.
549          */
550         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
551
552         return 0;
553 }
554
555 #define EXTRA_STACK_VM_PAGES    20      /* random */
556
557 /*
558  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
559  * the stack is optionally relocated, and some extra space is added.
560  */
561 int setup_arg_pages(struct linux_binprm *bprm,
562                     unsigned long stack_top,
563                     int executable_stack)
564 {
565         unsigned long ret;
566         unsigned long stack_shift;
567         struct mm_struct *mm = current->mm;
568         struct vm_area_struct *vma = bprm->vma;
569         struct vm_area_struct *prev = NULL;
570         unsigned long vm_flags;
571         unsigned long stack_base;
572
573 #ifdef CONFIG_STACK_GROWSUP
574         /* Limit stack size to 1GB */
575         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
576         if (stack_base > (1 << 30))
577                 stack_base = 1 << 30;
578
579         /* Make sure we didn't let the argument array grow too large. */
580         if (vma->vm_end - vma->vm_start > stack_base)
581                 return -ENOMEM;
582
583         stack_base = PAGE_ALIGN(stack_top - stack_base);
584
585         stack_shift = vma->vm_start - stack_base;
586         mm->arg_start = bprm->p - stack_shift;
587         bprm->p = vma->vm_end - stack_shift;
588 #else
589         stack_top = arch_align_stack(stack_top);
590         stack_top = PAGE_ALIGN(stack_top);
591         stack_shift = vma->vm_end - stack_top;
592
593         bprm->p -= stack_shift;
594         mm->arg_start = bprm->p;
595 #endif
596
597         if (bprm->loader)
598                 bprm->loader -= stack_shift;
599         bprm->exec -= stack_shift;
600
601         down_write(&mm->mmap_sem);
602         vm_flags = VM_STACK_FLAGS;
603
604         /*
605          * Adjust stack execute permissions; explicitly enable for
606          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
607          * (arch default) otherwise.
608          */
609         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
610                 vm_flags |= VM_EXEC;
611         else if (executable_stack == EXSTACK_DISABLE_X)
612                 vm_flags &= ~VM_EXEC;
613         vm_flags |= mm->def_flags;
614
615         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
616                         vm_flags);
617         if (ret)
618                 goto out_unlock;
619         BUG_ON(prev != vma);
620
621         /* Move stack pages down in memory. */
622         if (stack_shift) {
623                 ret = shift_arg_pages(vma, stack_shift);
624                 if (ret) {
625                         up_write(&mm->mmap_sem);
626                         return ret;
627                 }
628         }
629
630 #ifdef CONFIG_STACK_GROWSUP
631         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
632 #else
633         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
634 #endif
635         ret = expand_stack(vma, stack_base);
636         if (ret)
637                 ret = -EFAULT;
638
639 out_unlock:
640         up_write(&mm->mmap_sem);
641         return 0;
642 }
643 EXPORT_SYMBOL(setup_arg_pages);
644
645 #endif /* CONFIG_MMU */
646
647 struct file *open_exec(const char *name)
648 {
649         struct file *file;
650         int err;
651
652         file = do_filp_open(AT_FDCWD, name,
653                                 O_LARGEFILE | O_RDONLY | FMODE_EXEC, 0,
654                                 MAY_EXEC | MAY_OPEN);
655         if (IS_ERR(file))
656                 goto out;
657
658         err = -EACCES;
659         if (!S_ISREG(file->f_path.dentry->d_inode->i_mode))
660                 goto exit;
661
662         if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
663                 goto exit;
664
665         fsnotify_open(file->f_path.dentry);
666
667         err = deny_write_access(file);
668         if (err)
669                 goto exit;
670
671 out:
672         return file;
673
674 exit:
675         fput(file);
676         return ERR_PTR(err);
677 }
678 EXPORT_SYMBOL(open_exec);
679
680 int kernel_read(struct file *file, unsigned long offset,
681         char *addr, unsigned long count)
682 {
683         mm_segment_t old_fs;
684         loff_t pos = offset;
685         int result;
686
687         old_fs = get_fs();
688         set_fs(get_ds());
689         /* The cast to a user pointer is valid due to the set_fs() */
690         result = vfs_read(file, (void __user *)addr, count, &pos);
691         set_fs(old_fs);
692         return result;
693 }
694
695 EXPORT_SYMBOL(kernel_read);
696
697 static int exec_mmap(struct mm_struct *mm)
698 {
699         struct task_struct *tsk;
700         struct mm_struct * old_mm, *active_mm;
701
702         /* Notify parent that we're no longer interested in the old VM */
703         tsk = current;
704         old_mm = current->mm;
705         mm_release(tsk, old_mm);
706
707         if (old_mm) {
708                 /*
709                  * Make sure that if there is a core dump in progress
710                  * for the old mm, we get out and die instead of going
711                  * through with the exec.  We must hold mmap_sem around
712                  * checking core_state and changing tsk->mm.
713                  */
714                 down_read(&old_mm->mmap_sem);
715                 if (unlikely(old_mm->core_state)) {
716                         up_read(&old_mm->mmap_sem);
717                         return -EINTR;
718                 }
719         }
720         task_lock(tsk);
721         active_mm = tsk->active_mm;
722         tsk->mm = mm;
723         tsk->active_mm = mm;
724         activate_mm(active_mm, mm);
725         task_unlock(tsk);
726         arch_pick_mmap_layout(mm);
727         if (old_mm) {
728                 up_read(&old_mm->mmap_sem);
729                 BUG_ON(active_mm != old_mm);
730                 mm_update_next_owner(old_mm);
731                 mmput(old_mm);
732                 return 0;
733         }
734         mmdrop(active_mm);
735         return 0;
736 }
737
738 /*
739  * This function makes sure the current process has its own signal table,
740  * so that flush_signal_handlers can later reset the handlers without
741  * disturbing other processes.  (Other processes might share the signal
742  * table via the CLONE_SIGHAND option to clone().)
743  */
744 static int de_thread(struct task_struct *tsk)
745 {
746         struct signal_struct *sig = tsk->signal;
747         struct sighand_struct *oldsighand = tsk->sighand;
748         spinlock_t *lock = &oldsighand->siglock;
749         int count;
750
751         if (thread_group_empty(tsk))
752                 goto no_thread_group;
753
754         /*
755          * Kill all other threads in the thread group.
756          */
757         spin_lock_irq(lock);
758         if (signal_group_exit(sig)) {
759                 /*
760                  * Another group action in progress, just
761                  * return so that the signal is processed.
762                  */
763                 spin_unlock_irq(lock);
764                 return -EAGAIN;
765         }
766         sig->group_exit_task = tsk;
767         zap_other_threads(tsk);
768
769         /* Account for the thread group leader hanging around: */
770         count = thread_group_leader(tsk) ? 1 : 2;
771         sig->notify_count = count;
772         while (atomic_read(&sig->count) > count) {
773                 __set_current_state(TASK_UNINTERRUPTIBLE);
774                 spin_unlock_irq(lock);
775                 schedule();
776                 spin_lock_irq(lock);
777         }
778         spin_unlock_irq(lock);
779
780         /*
781          * At this point all other threads have exited, all we have to
782          * do is to wait for the thread group leader to become inactive,
783          * and to assume its PID:
784          */
785         if (!thread_group_leader(tsk)) {
786                 struct task_struct *leader = tsk->group_leader;
787
788                 sig->notify_count = -1; /* for exit_notify() */
789                 for (;;) {
790                         write_lock_irq(&tasklist_lock);
791                         if (likely(leader->exit_state))
792                                 break;
793                         __set_current_state(TASK_UNINTERRUPTIBLE);
794                         write_unlock_irq(&tasklist_lock);
795                         schedule();
796                 }
797
798                 /*
799                  * The only record we have of the real-time age of a
800                  * process, regardless of execs it's done, is start_time.
801                  * All the past CPU time is accumulated in signal_struct
802                  * from sister threads now dead.  But in this non-leader
803                  * exec, nothing survives from the original leader thread,
804                  * whose birth marks the true age of this process now.
805                  * When we take on its identity by switching to its PID, we
806                  * also take its birthdate (always earlier than our own).
807                  */
808                 tsk->start_time = leader->start_time;
809
810                 BUG_ON(!same_thread_group(leader, tsk));
811                 BUG_ON(has_group_leader_pid(tsk));
812                 /*
813                  * An exec() starts a new thread group with the
814                  * TGID of the previous thread group. Rehash the
815                  * two threads with a switched PID, and release
816                  * the former thread group leader:
817                  */
818
819                 /* Become a process group leader with the old leader's pid.
820                  * The old leader becomes a thread of the this thread group.
821                  * Note: The old leader also uses this pid until release_task
822                  *       is called.  Odd but simple and correct.
823                  */
824                 detach_pid(tsk, PIDTYPE_PID);
825                 tsk->pid = leader->pid;
826                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
827                 transfer_pid(leader, tsk, PIDTYPE_PGID);
828                 transfer_pid(leader, tsk, PIDTYPE_SID);
829                 list_replace_rcu(&leader->tasks, &tsk->tasks);
830
831                 tsk->group_leader = tsk;
832                 leader->group_leader = tsk;
833
834                 tsk->exit_signal = SIGCHLD;
835
836                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
837                 leader->exit_state = EXIT_DEAD;
838                 write_unlock_irq(&tasklist_lock);
839
840                 release_task(leader);
841         }
842
843         sig->group_exit_task = NULL;
844         sig->notify_count = 0;
845
846 no_thread_group:
847         exit_itimers(sig);
848         flush_itimer_signals();
849
850         if (atomic_read(&oldsighand->count) != 1) {
851                 struct sighand_struct *newsighand;
852                 /*
853                  * This ->sighand is shared with the CLONE_SIGHAND
854                  * but not CLONE_THREAD task, switch to the new one.
855                  */
856                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
857                 if (!newsighand)
858                         return -ENOMEM;
859
860                 atomic_set(&newsighand->count, 1);
861                 memcpy(newsighand->action, oldsighand->action,
862                        sizeof(newsighand->action));
863
864                 write_lock_irq(&tasklist_lock);
865                 spin_lock(&oldsighand->siglock);
866                 rcu_assign_pointer(tsk->sighand, newsighand);
867                 spin_unlock(&oldsighand->siglock);
868                 write_unlock_irq(&tasklist_lock);
869
870                 __cleanup_sighand(oldsighand);
871         }
872
873         BUG_ON(!thread_group_leader(tsk));
874         return 0;
875 }
876
877 /*
878  * These functions flushes out all traces of the currently running executable
879  * so that a new one can be started
880  */
881 static void flush_old_files(struct files_struct * files)
882 {
883         long j = -1;
884         struct fdtable *fdt;
885
886         spin_lock(&files->file_lock);
887         for (;;) {
888                 unsigned long set, i;
889
890                 j++;
891                 i = j * __NFDBITS;
892                 fdt = files_fdtable(files);
893                 if (i >= fdt->max_fds)
894                         break;
895                 set = fdt->close_on_exec->fds_bits[j];
896                 if (!set)
897                         continue;
898                 fdt->close_on_exec->fds_bits[j] = 0;
899                 spin_unlock(&files->file_lock);
900                 for ( ; set ; i++,set >>= 1) {
901                         if (set & 1) {
902                                 sys_close(i);
903                         }
904                 }
905                 spin_lock(&files->file_lock);
906
907         }
908         spin_unlock(&files->file_lock);
909 }
910
911 char *get_task_comm(char *buf, struct task_struct *tsk)
912 {
913         /* buf must be at least sizeof(tsk->comm) in size */
914         task_lock(tsk);
915         strncpy(buf, tsk->comm, sizeof(tsk->comm));
916         task_unlock(tsk);
917         return buf;
918 }
919
920 void set_task_comm(struct task_struct *tsk, char *buf)
921 {
922         task_lock(tsk);
923         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
924         task_unlock(tsk);
925 }
926
927 int flush_old_exec(struct linux_binprm * bprm)
928 {
929         char * name;
930         int i, ch, retval;
931         char tcomm[sizeof(current->comm)];
932
933         /*
934          * Make sure we have a private signal table and that
935          * we are unassociated from the previous thread group.
936          */
937         retval = de_thread(current);
938         if (retval)
939                 goto out;
940
941         set_mm_exe_file(bprm->mm, bprm->file);
942
943         /*
944          * Release all of the old mmap stuff
945          */
946         retval = exec_mmap(bprm->mm);
947         if (retval)
948                 goto out;
949
950         bprm->mm = NULL;                /* We're using it now */
951
952         /* This is the point of no return */
953         current->sas_ss_sp = current->sas_ss_size = 0;
954
955         if (current_euid() == current_uid() && current_egid() == current_gid())
956                 set_dumpable(current->mm, 1);
957         else
958                 set_dumpable(current->mm, suid_dumpable);
959
960         name = bprm->filename;
961
962         /* Copies the binary name from after last slash */
963         for (i=0; (ch = *(name++)) != '\0';) {
964                 if (ch == '/')
965                         i = 0; /* overwrite what we wrote */
966                 else
967                         if (i < (sizeof(tcomm) - 1))
968                                 tcomm[i++] = ch;
969         }
970         tcomm[i] = '\0';
971         set_task_comm(current, tcomm);
972
973         current->flags &= ~PF_RANDOMIZE;
974         flush_thread();
975
976         /* Set the new mm task size. We have to do that late because it may
977          * depend on TIF_32BIT which is only updated in flush_thread() on
978          * some architectures like powerpc
979          */
980         current->mm->task_size = TASK_SIZE;
981
982         /* install the new credentials */
983         if (bprm->cred->uid != current_euid() ||
984             bprm->cred->gid != current_egid()) {
985                 current->pdeath_signal = 0;
986         } else if (file_permission(bprm->file, MAY_READ) ||
987                    bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP) {
988                 set_dumpable(current->mm, suid_dumpable);
989         }
990
991         current->personality &= ~bprm->per_clear;
992
993         /* An exec changes our domain. We are no longer part of the thread
994            group */
995
996         current->self_exec_id++;
997                         
998         flush_signal_handlers(current, 0);
999         flush_old_files(current->files);
1000
1001         return 0;
1002
1003 out:
1004         return retval;
1005 }
1006
1007 EXPORT_SYMBOL(flush_old_exec);
1008
1009 /*
1010  * install the new credentials for this executable
1011  */
1012 void install_exec_creds(struct linux_binprm *bprm)
1013 {
1014         security_bprm_committing_creds(bprm);
1015
1016         commit_creds(bprm->cred);
1017         bprm->cred = NULL;
1018
1019         /* cred_exec_mutex must be held at least to this point to prevent
1020          * ptrace_attach() from altering our determination of the task's
1021          * credentials; any time after this it may be unlocked */
1022
1023         security_bprm_committed_creds(bprm);
1024 }
1025 EXPORT_SYMBOL(install_exec_creds);
1026
1027 /*
1028  * determine how safe it is to execute the proposed program
1029  * - the caller must hold current->cred_exec_mutex to protect against
1030  *   PTRACE_ATTACH
1031  */
1032 int check_unsafe_exec(struct linux_binprm *bprm)
1033 {
1034         struct task_struct *p = current, *t;
1035         unsigned n_fs;
1036         int res = 0;
1037
1038         bprm->unsafe = tracehook_unsafe_exec(p);
1039
1040         n_fs = 1;
1041         write_lock(&p->fs->lock);
1042         rcu_read_lock();
1043         for (t = next_thread(p); t != p; t = next_thread(t)) {
1044                 if (t->fs == p->fs)
1045                         n_fs++;
1046         }
1047         rcu_read_unlock();
1048
1049         if (p->fs->users > n_fs) {
1050                 bprm->unsafe |= LSM_UNSAFE_SHARE;
1051         } else {
1052                 res = -EAGAIN;
1053                 if (!p->fs->in_exec) {
1054                         p->fs->in_exec = 1;
1055                         res = 1;
1056                 }
1057         }
1058         write_unlock(&p->fs->lock);
1059
1060         return res;
1061 }
1062
1063 /* 
1064  * Fill the binprm structure from the inode. 
1065  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1066  *
1067  * This may be called multiple times for binary chains (scripts for example).
1068  */
1069 int prepare_binprm(struct linux_binprm *bprm)
1070 {
1071         umode_t mode;
1072         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1073         int retval;
1074
1075         mode = inode->i_mode;
1076         if (bprm->file->f_op == NULL)
1077                 return -EACCES;
1078
1079         /* clear any previous set[ug]id data from a previous binary */
1080         bprm->cred->euid = current_euid();
1081         bprm->cred->egid = current_egid();
1082
1083         if (!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1084                 /* Set-uid? */
1085                 if (mode & S_ISUID) {
1086                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1087                         bprm->cred->euid = inode->i_uid;
1088                 }
1089
1090                 /* Set-gid? */
1091                 /*
1092                  * If setgid is set but no group execute bit then this
1093                  * is a candidate for mandatory locking, not a setgid
1094                  * executable.
1095                  */
1096                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1097                         bprm->per_clear |= PER_CLEAR_ON_SETID;
1098                         bprm->cred->egid = inode->i_gid;
1099                 }
1100         }
1101
1102         /* fill in binprm security blob */
1103         retval = security_bprm_set_creds(bprm);
1104         if (retval)
1105                 return retval;
1106         bprm->cred_prepared = 1;
1107
1108         memset(bprm->buf, 0, BINPRM_BUF_SIZE);
1109         return kernel_read(bprm->file, 0, bprm->buf, BINPRM_BUF_SIZE);
1110 }
1111
1112 EXPORT_SYMBOL(prepare_binprm);
1113
1114 /*
1115  * Arguments are '\0' separated strings found at the location bprm->p
1116  * points to; chop off the first by relocating brpm->p to right after
1117  * the first '\0' encountered.
1118  */
1119 int remove_arg_zero(struct linux_binprm *bprm)
1120 {
1121         int ret = 0;
1122         unsigned long offset;
1123         char *kaddr;
1124         struct page *page;
1125
1126         if (!bprm->argc)
1127                 return 0;
1128
1129         do {
1130                 offset = bprm->p & ~PAGE_MASK;
1131                 page = get_arg_page(bprm, bprm->p, 0);
1132                 if (!page) {
1133                         ret = -EFAULT;
1134                         goto out;
1135                 }
1136                 kaddr = kmap_atomic(page, KM_USER0);
1137
1138                 for (; offset < PAGE_SIZE && kaddr[offset];
1139                                 offset++, bprm->p++)
1140                         ;
1141
1142                 kunmap_atomic(kaddr, KM_USER0);
1143                 put_arg_page(page);
1144
1145                 if (offset == PAGE_SIZE)
1146                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1147         } while (offset == PAGE_SIZE);
1148
1149         bprm->p++;
1150         bprm->argc--;
1151         ret = 0;
1152
1153 out:
1154         return ret;
1155 }
1156 EXPORT_SYMBOL(remove_arg_zero);
1157
1158 /*
1159  * cycle the list of binary formats handler, until one recognizes the image
1160  */
1161 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1162 {
1163         unsigned int depth = bprm->recursion_depth;
1164         int try,retval;
1165         struct linux_binfmt *fmt;
1166
1167         retval = security_bprm_check(bprm);
1168         if (retval)
1169                 return retval;
1170         retval = ima_bprm_check(bprm);
1171         if (retval)
1172                 return retval;
1173
1174         /* kernel module loader fixup */
1175         /* so we don't try to load run modprobe in kernel space. */
1176         set_fs(USER_DS);
1177
1178         retval = audit_bprm(bprm);
1179         if (retval)
1180                 return retval;
1181
1182         retval = -ENOENT;
1183         for (try=0; try<2; try++) {
1184                 read_lock(&binfmt_lock);
1185                 list_for_each_entry(fmt, &formats, lh) {
1186                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1187                         if (!fn)
1188                                 continue;
1189                         if (!try_module_get(fmt->module))
1190                                 continue;
1191                         read_unlock(&binfmt_lock);
1192                         retval = fn(bprm, regs);
1193                         /*
1194                          * Restore the depth counter to its starting value
1195                          * in this call, so we don't have to rely on every
1196                          * load_binary function to restore it on return.
1197                          */
1198                         bprm->recursion_depth = depth;
1199                         if (retval >= 0) {
1200                                 if (depth == 0)
1201                                         tracehook_report_exec(fmt, bprm, regs);
1202                                 put_binfmt(fmt);
1203                                 allow_write_access(bprm->file);
1204                                 if (bprm->file)
1205                                         fput(bprm->file);
1206                                 bprm->file = NULL;
1207                                 current->did_exec = 1;
1208                                 proc_exec_connector(current);
1209                                 return retval;
1210                         }
1211                         read_lock(&binfmt_lock);
1212                         put_binfmt(fmt);
1213                         if (retval != -ENOEXEC || bprm->mm == NULL)
1214                                 break;
1215                         if (!bprm->file) {
1216                                 read_unlock(&binfmt_lock);
1217                                 return retval;
1218                         }
1219                 }
1220                 read_unlock(&binfmt_lock);
1221                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1222                         break;
1223 #ifdef CONFIG_MODULES
1224                 } else {
1225 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1226                         if (printable(bprm->buf[0]) &&
1227                             printable(bprm->buf[1]) &&
1228                             printable(bprm->buf[2]) &&
1229                             printable(bprm->buf[3]))
1230                                 break; /* -ENOEXEC */
1231                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1232 #endif
1233                 }
1234         }
1235         return retval;
1236 }
1237
1238 EXPORT_SYMBOL(search_binary_handler);
1239
1240 void free_bprm(struct linux_binprm *bprm)
1241 {
1242         free_arg_pages(bprm);
1243         if (bprm->cred)
1244                 abort_creds(bprm->cred);
1245         kfree(bprm);
1246 }
1247
1248 /*
1249  * sys_execve() executes a new program.
1250  */
1251 int do_execve(char * filename,
1252         char __user *__user *argv,
1253         char __user *__user *envp,
1254         struct pt_regs * regs)
1255 {
1256         struct linux_binprm *bprm;
1257         struct file *file;
1258         struct files_struct *displaced;
1259         bool clear_in_exec;
1260         int retval;
1261
1262         retval = unshare_files(&displaced);
1263         if (retval)
1264                 goto out_ret;
1265
1266         retval = -ENOMEM;
1267         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1268         if (!bprm)
1269                 goto out_files;
1270
1271         retval = mutex_lock_interruptible(&current->cred_exec_mutex);
1272         if (retval < 0)
1273                 goto out_free;
1274         current->in_execve = 1;
1275
1276         retval = -ENOMEM;
1277         bprm->cred = prepare_exec_creds();
1278         if (!bprm->cred)
1279                 goto out_unlock;
1280
1281         retval = check_unsafe_exec(bprm);
1282         if (retval < 0)
1283                 goto out_unlock;
1284         clear_in_exec = retval;
1285
1286         file = open_exec(filename);
1287         retval = PTR_ERR(file);
1288         if (IS_ERR(file))
1289                 goto out_unmark;
1290
1291         sched_exec();
1292
1293         bprm->file = file;
1294         bprm->filename = filename;
1295         bprm->interp = filename;
1296
1297         retval = bprm_mm_init(bprm);
1298         if (retval)
1299                 goto out_file;
1300
1301         bprm->argc = count(argv, MAX_ARG_STRINGS);
1302         if ((retval = bprm->argc) < 0)
1303                 goto out;
1304
1305         bprm->envc = count(envp, MAX_ARG_STRINGS);
1306         if ((retval = bprm->envc) < 0)
1307                 goto out;
1308
1309         retval = prepare_binprm(bprm);
1310         if (retval < 0)
1311                 goto out;
1312
1313         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1314         if (retval < 0)
1315                 goto out;
1316
1317         bprm->exec = bprm->p;
1318         retval = copy_strings(bprm->envc, envp, bprm);
1319         if (retval < 0)
1320                 goto out;
1321
1322         retval = copy_strings(bprm->argc, argv, bprm);
1323         if (retval < 0)
1324                 goto out;
1325
1326         current->flags &= ~PF_KTHREAD;
1327         retval = search_binary_handler(bprm,regs);
1328         if (retval < 0)
1329                 goto out;
1330
1331         /* execve succeeded */
1332         current->fs->in_exec = 0;
1333         current->in_execve = 0;
1334         mutex_unlock(&current->cred_exec_mutex);
1335         acct_update_integrals(current);
1336         free_bprm(bprm);
1337         if (displaced)
1338                 put_files_struct(displaced);
1339         return retval;
1340
1341 out:
1342         if (bprm->mm)
1343                 mmput (bprm->mm);
1344
1345 out_file:
1346         if (bprm->file) {
1347                 allow_write_access(bprm->file);
1348                 fput(bprm->file);
1349         }
1350
1351 out_unmark:
1352         if (clear_in_exec)
1353                 current->fs->in_exec = 0;
1354
1355 out_unlock:
1356         current->in_execve = 0;
1357         mutex_unlock(&current->cred_exec_mutex);
1358
1359 out_free:
1360         free_bprm(bprm);
1361
1362 out_files:
1363         if (displaced)
1364                 reset_files_struct(displaced);
1365 out_ret:
1366         return retval;
1367 }
1368
1369 int set_binfmt(struct linux_binfmt *new)
1370 {
1371         struct linux_binfmt *old = current->binfmt;
1372
1373         if (new) {
1374                 if (!try_module_get(new->module))
1375                         return -1;
1376         }
1377         current->binfmt = new;
1378         if (old)
1379                 module_put(old->module);
1380         return 0;
1381 }
1382
1383 EXPORT_SYMBOL(set_binfmt);
1384
1385 /* format_corename will inspect the pattern parameter, and output a
1386  * name into corename, which must have space for at least
1387  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1388  */
1389 static int format_corename(char *corename, long signr)
1390 {
1391         const struct cred *cred = current_cred();
1392         const char *pat_ptr = core_pattern;
1393         int ispipe = (*pat_ptr == '|');
1394         char *out_ptr = corename;
1395         char *const out_end = corename + CORENAME_MAX_SIZE;
1396         int rc;
1397         int pid_in_pattern = 0;
1398
1399         /* Repeat as long as we have more pattern to process and more output
1400            space */
1401         while (*pat_ptr) {
1402                 if (*pat_ptr != '%') {
1403                         if (out_ptr == out_end)
1404                                 goto out;
1405                         *out_ptr++ = *pat_ptr++;
1406                 } else {
1407                         switch (*++pat_ptr) {
1408                         case 0:
1409                                 goto out;
1410                         /* Double percent, output one percent */
1411                         case '%':
1412                                 if (out_ptr == out_end)
1413                                         goto out;
1414                                 *out_ptr++ = '%';
1415                                 break;
1416                         /* pid */
1417                         case 'p':
1418                                 pid_in_pattern = 1;
1419                                 rc = snprintf(out_ptr, out_end - out_ptr,
1420                                               "%d", task_tgid_vnr(current));
1421                                 if (rc > out_end - out_ptr)
1422                                         goto out;
1423                                 out_ptr += rc;
1424                                 break;
1425                         /* uid */
1426                         case 'u':
1427                                 rc = snprintf(out_ptr, out_end - out_ptr,
1428                                               "%d", cred->uid);
1429                                 if (rc > out_end - out_ptr)
1430                                         goto out;
1431                                 out_ptr += rc;
1432                                 break;
1433                         /* gid */
1434                         case 'g':
1435                                 rc = snprintf(out_ptr, out_end - out_ptr,
1436                                               "%d", cred->gid);
1437                                 if (rc > out_end - out_ptr)
1438                                         goto out;
1439                                 out_ptr += rc;
1440                                 break;
1441                         /* signal that caused the coredump */
1442                         case 's':
1443                                 rc = snprintf(out_ptr, out_end - out_ptr,
1444                                               "%ld", signr);
1445                                 if (rc > out_end - out_ptr)
1446                                         goto out;
1447                                 out_ptr += rc;
1448                                 break;
1449                         /* UNIX time of coredump */
1450                         case 't': {
1451                                 struct timeval tv;
1452                                 do_gettimeofday(&tv);
1453                                 rc = snprintf(out_ptr, out_end - out_ptr,
1454                                               "%lu", tv.tv_sec);
1455                                 if (rc > out_end - out_ptr)
1456                                         goto out;
1457                                 out_ptr += rc;
1458                                 break;
1459                         }
1460                         /* hostname */
1461                         case 'h':
1462                                 down_read(&uts_sem);
1463                                 rc = snprintf(out_ptr, out_end - out_ptr,
1464                                               "%s", utsname()->nodename);
1465                                 up_read(&uts_sem);
1466                                 if (rc > out_end - out_ptr)
1467                                         goto out;
1468                                 out_ptr += rc;
1469                                 break;
1470                         /* executable */
1471                         case 'e':
1472                                 rc = snprintf(out_ptr, out_end - out_ptr,
1473                                               "%s", current->comm);
1474                                 if (rc > out_end - out_ptr)
1475                                         goto out;
1476                                 out_ptr += rc;
1477                                 break;
1478                         /* core limit size */
1479                         case 'c':
1480                                 rc = snprintf(out_ptr, out_end - out_ptr,
1481                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1482                                 if (rc > out_end - out_ptr)
1483                                         goto out;
1484                                 out_ptr += rc;
1485                                 break;
1486                         default:
1487                                 break;
1488                         }
1489                         ++pat_ptr;
1490                 }
1491         }
1492         /* Backward compatibility with core_uses_pid:
1493          *
1494          * If core_pattern does not include a %p (as is the default)
1495          * and core_uses_pid is set, then .%pid will be appended to
1496          * the filename. Do not do this for piped commands. */
1497         if (!ispipe && !pid_in_pattern && core_uses_pid) {
1498                 rc = snprintf(out_ptr, out_end - out_ptr,
1499                               ".%d", task_tgid_vnr(current));
1500                 if (rc > out_end - out_ptr)
1501                         goto out;
1502                 out_ptr += rc;
1503         }
1504 out:
1505         *out_ptr = 0;
1506         return ispipe;
1507 }
1508
1509 static int zap_process(struct task_struct *start)
1510 {
1511         struct task_struct *t;
1512         int nr = 0;
1513
1514         start->signal->flags = SIGNAL_GROUP_EXIT;
1515         start->signal->group_stop_count = 0;
1516
1517         t = start;
1518         do {
1519                 if (t != current && t->mm) {
1520                         sigaddset(&t->pending.signal, SIGKILL);
1521                         signal_wake_up(t, 1);
1522                         nr++;
1523                 }
1524         } while_each_thread(start, t);
1525
1526         return nr;
1527 }
1528
1529 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1530                                 struct core_state *core_state, int exit_code)
1531 {
1532         struct task_struct *g, *p;
1533         unsigned long flags;
1534         int nr = -EAGAIN;
1535
1536         spin_lock_irq(&tsk->sighand->siglock);
1537         if (!signal_group_exit(tsk->signal)) {
1538                 mm->core_state = core_state;
1539                 tsk->signal->group_exit_code = exit_code;
1540                 nr = zap_process(tsk);
1541         }
1542         spin_unlock_irq(&tsk->sighand->siglock);
1543         if (unlikely(nr < 0))
1544                 return nr;
1545
1546         if (atomic_read(&mm->mm_users) == nr + 1)
1547                 goto done;
1548         /*
1549          * We should find and kill all tasks which use this mm, and we should
1550          * count them correctly into ->nr_threads. We don't take tasklist
1551          * lock, but this is safe wrt:
1552          *
1553          * fork:
1554          *      None of sub-threads can fork after zap_process(leader). All
1555          *      processes which were created before this point should be
1556          *      visible to zap_threads() because copy_process() adds the new
1557          *      process to the tail of init_task.tasks list, and lock/unlock
1558          *      of ->siglock provides a memory barrier.
1559          *
1560          * do_exit:
1561          *      The caller holds mm->mmap_sem. This means that the task which
1562          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1563          *      its ->mm.
1564          *
1565          * de_thread:
1566          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1567          *      we must see either old or new leader, this does not matter.
1568          *      However, it can change p->sighand, so lock_task_sighand(p)
1569          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1570          *      it can't fail.
1571          *
1572          *      Note also that "g" can be the old leader with ->mm == NULL
1573          *      and already unhashed and thus removed from ->thread_group.
1574          *      This is OK, __unhash_process()->list_del_rcu() does not
1575          *      clear the ->next pointer, we will find the new leader via
1576          *      next_thread().
1577          */
1578         rcu_read_lock();
1579         for_each_process(g) {
1580                 if (g == tsk->group_leader)
1581                         continue;
1582                 if (g->flags & PF_KTHREAD)
1583                         continue;
1584                 p = g;
1585                 do {
1586                         if (p->mm) {
1587                                 if (unlikely(p->mm == mm)) {
1588                                         lock_task_sighand(p, &flags);
1589                                         nr += zap_process(p);
1590                                         unlock_task_sighand(p, &flags);
1591                                 }
1592                                 break;
1593                         }
1594                 } while_each_thread(g, p);
1595         }
1596         rcu_read_unlock();
1597 done:
1598         atomic_set(&core_state->nr_threads, nr);
1599         return nr;
1600 }
1601
1602 static int coredump_wait(int exit_code, struct core_state *core_state)
1603 {
1604         struct task_struct *tsk = current;
1605         struct mm_struct *mm = tsk->mm;
1606         struct completion *vfork_done;
1607         int core_waiters;
1608
1609         init_completion(&core_state->startup);
1610         core_state->dumper.task = tsk;
1611         core_state->dumper.next = NULL;
1612         core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1613         up_write(&mm->mmap_sem);
1614
1615         if (unlikely(core_waiters < 0))
1616                 goto fail;
1617
1618         /*
1619          * Make sure nobody is waiting for us to release the VM,
1620          * otherwise we can deadlock when we wait on each other
1621          */
1622         vfork_done = tsk->vfork_done;
1623         if (vfork_done) {
1624                 tsk->vfork_done = NULL;
1625                 complete(vfork_done);
1626         }
1627
1628         if (core_waiters)
1629                 wait_for_completion(&core_state->startup);
1630 fail:
1631         return core_waiters;
1632 }
1633
1634 static void coredump_finish(struct mm_struct *mm)
1635 {
1636         struct core_thread *curr, *next;
1637         struct task_struct *task;
1638
1639         next = mm->core_state->dumper.next;
1640         while ((curr = next) != NULL) {
1641                 next = curr->next;
1642                 task = curr->task;
1643                 /*
1644                  * see exit_mm(), curr->task must not see
1645                  * ->task == NULL before we read ->next.
1646                  */
1647                 smp_mb();
1648                 curr->task = NULL;
1649                 wake_up_process(task);
1650         }
1651
1652         mm->core_state = NULL;
1653 }
1654
1655 /*
1656  * set_dumpable converts traditional three-value dumpable to two flags and
1657  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1658  * these bits are not changed atomically.  So get_dumpable can observe the
1659  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1660  * return either old dumpable or new one by paying attention to the order of
1661  * modifying the bits.
1662  *
1663  * dumpable |   mm->flags (binary)
1664  * old  new | initial interim  final
1665  * ---------+-----------------------
1666  *  0    1  |   00      01      01
1667  *  0    2  |   00      10(*)   11
1668  *  1    0  |   01      00      00
1669  *  1    2  |   01      11      11
1670  *  2    0  |   11      10(*)   00
1671  *  2    1  |   11      11      01
1672  *
1673  * (*) get_dumpable regards interim value of 10 as 11.
1674  */
1675 void set_dumpable(struct mm_struct *mm, int value)
1676 {
1677         switch (value) {
1678         case 0:
1679                 clear_bit(MMF_DUMPABLE, &mm->flags);
1680                 smp_wmb();
1681                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1682                 break;
1683         case 1:
1684                 set_bit(MMF_DUMPABLE, &mm->flags);
1685                 smp_wmb();
1686                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1687                 break;
1688         case 2:
1689                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1690                 smp_wmb();
1691                 set_bit(MMF_DUMPABLE, &mm->flags);
1692                 break;
1693         }
1694 }
1695
1696 int get_dumpable(struct mm_struct *mm)
1697 {
1698         int ret;
1699
1700         ret = mm->flags & 0x3;
1701         return (ret >= 2) ? 2 : ret;
1702 }
1703
1704 void do_coredump(long signr, int exit_code, struct pt_regs *regs)
1705 {
1706         struct core_state core_state;
1707         char corename[CORENAME_MAX_SIZE + 1];
1708         struct mm_struct *mm = current->mm;
1709         struct linux_binfmt * binfmt;
1710         struct inode * inode;
1711         struct file * file;
1712         const struct cred *old_cred;
1713         struct cred *cred;
1714         int retval = 0;
1715         int flag = 0;
1716         int ispipe = 0;
1717         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1718         char **helper_argv = NULL;
1719         int helper_argc = 0;
1720         char *delimit;
1721
1722         audit_core_dumps(signr);
1723
1724         binfmt = current->binfmt;
1725         if (!binfmt || !binfmt->core_dump)
1726                 goto fail;
1727
1728         cred = prepare_creds();
1729         if (!cred) {
1730                 retval = -ENOMEM;
1731                 goto fail;
1732         }
1733
1734         down_write(&mm->mmap_sem);
1735         /*
1736          * If another thread got here first, or we are not dumpable, bail out.
1737          */
1738         if (mm->core_state || !get_dumpable(mm)) {
1739                 up_write(&mm->mmap_sem);
1740                 put_cred(cred);
1741                 goto fail;
1742         }
1743
1744         /*
1745          *      We cannot trust fsuid as being the "true" uid of the
1746          *      process nor do we know its entire history. We only know it
1747          *      was tainted so we dump it as root in mode 2.
1748          */
1749         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1750                 flag = O_EXCL;          /* Stop rewrite attacks */
1751                 cred->fsuid = 0;        /* Dump root private */
1752         }
1753
1754         retval = coredump_wait(exit_code, &core_state);
1755         if (retval < 0) {
1756                 put_cred(cred);
1757                 goto fail;
1758         }
1759
1760         old_cred = override_creds(cred);
1761
1762         /*
1763          * Clear any false indication of pending signals that might
1764          * be seen by the filesystem code called to write the core file.
1765          */
1766         clear_thread_flag(TIF_SIGPENDING);
1767
1768         /*
1769          * lock_kernel() because format_corename() is controlled by sysctl, which
1770          * uses lock_kernel()
1771          */
1772         lock_kernel();
1773         ispipe = format_corename(corename, signr);
1774         unlock_kernel();
1775         /*
1776          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1777          * to a pipe.  Since we're not writing directly to the filesystem
1778          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1779          * created unless the pipe reader choses to write out the core file
1780          * at which point file size limits and permissions will be imposed
1781          * as it does with any other process
1782          */
1783         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1784                 goto fail_unlock;
1785
1786         if (ispipe) {
1787                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1788                 if (!helper_argv) {
1789                         printk(KERN_WARNING "%s failed to allocate memory\n",
1790                                __func__);
1791                         goto fail_unlock;
1792                 }
1793                 /* Terminate the string before the first option */
1794                 delimit = strchr(corename, ' ');
1795                 if (delimit)
1796                         *delimit = '\0';
1797                 delimit = strrchr(helper_argv[0], '/');
1798                 if (delimit)
1799                         delimit++;
1800                 else
1801                         delimit = helper_argv[0];
1802                 if (!strcmp(delimit, current->comm)) {
1803                         printk(KERN_NOTICE "Recursive core dump detected, "
1804                                         "aborting\n");
1805                         goto fail_unlock;
1806                 }
1807
1808                 core_limit = RLIM_INFINITY;
1809
1810                 /* SIGPIPE can happen, but it's just never processed */
1811                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1812                                 &file)) {
1813                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1814                                corename);
1815                         goto fail_unlock;
1816                 }
1817         } else
1818                 file = filp_open(corename,
1819                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1820                                  0600);
1821         if (IS_ERR(file))
1822                 goto fail_unlock;
1823         inode = file->f_path.dentry->d_inode;
1824         if (inode->i_nlink > 1)
1825                 goto close_fail;        /* multiple links - don't dump */
1826         if (!ispipe && d_unhashed(file->f_path.dentry))
1827                 goto close_fail;
1828
1829         /* AK: actually i see no reason to not allow this for named pipes etc.,
1830            but keep the previous behaviour for now. */
1831         if (!ispipe && !S_ISREG(inode->i_mode))
1832                 goto close_fail;
1833         /*
1834          * Dont allow local users get cute and trick others to coredump
1835          * into their pre-created files:
1836          */
1837         if (inode->i_uid != current_fsuid())
1838                 goto close_fail;
1839         if (!file->f_op)
1840                 goto close_fail;
1841         if (!file->f_op->write)
1842                 goto close_fail;
1843         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1844                 goto close_fail;
1845
1846         retval = binfmt->core_dump(signr, regs, file, core_limit);
1847
1848         if (retval)
1849                 current->signal->group_exit_code |= 0x80;
1850 close_fail:
1851         filp_close(file, NULL);
1852 fail_unlock:
1853         if (helper_argv)
1854                 argv_free(helper_argv);
1855
1856         revert_creds(old_cred);
1857         put_cred(cred);
1858         coredump_finish(mm);
1859 fail:
1860         return;
1861 }