71734568f0188f1df9cb7f21349980c4df45395e
[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/highmem.h>
36 #include <linux/spinlock.h>
37 #include <linux/key.h>
38 #include <linux/personality.h>
39 #include <linux/binfmts.h>
40 #include <linux/utsname.h>
41 #include <linux/pid_namespace.h>
42 #include <linux/module.h>
43 #include <linux/namei.h>
44 #include <linux/proc_fs.h>
45 #include <linux/ptrace.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52
53 #include <asm/uaccess.h>
54 #include <asm/mmu_context.h>
55 #include <asm/tlb.h>
56
57 #ifdef CONFIG_KMOD
58 #include <linux/kmod.h>
59 #endif
60
61 #ifdef __alpha__
62 /* for /sbin/loader handling in search_binary_handler() */
63 #include <linux/a.out.h>
64 #endif
65
66 int core_uses_pid;
67 char core_pattern[CORENAME_MAX_SIZE] = "core";
68 int suid_dumpable = 0;
69
70 /* The maximal length of core_pattern is also specified in sysctl.c */
71
72 static LIST_HEAD(formats);
73 static DEFINE_RWLOCK(binfmt_lock);
74
75 int register_binfmt(struct linux_binfmt * fmt)
76 {
77         if (!fmt)
78                 return -EINVAL;
79         write_lock(&binfmt_lock);
80         list_add(&fmt->lh, &formats);
81         write_unlock(&binfmt_lock);
82         return 0;       
83 }
84
85 EXPORT_SYMBOL(register_binfmt);
86
87 void unregister_binfmt(struct linux_binfmt * fmt)
88 {
89         write_lock(&binfmt_lock);
90         list_del(&fmt->lh);
91         write_unlock(&binfmt_lock);
92 }
93
94 EXPORT_SYMBOL(unregister_binfmt);
95
96 static inline void put_binfmt(struct linux_binfmt * fmt)
97 {
98         module_put(fmt->module);
99 }
100
101 /*
102  * Note that a shared library must be both readable and executable due to
103  * security reasons.
104  *
105  * Also note that we take the address to load from from the file itself.
106  */
107 asmlinkage long sys_uselib(const char __user * library)
108 {
109         struct file * file;
110         struct nameidata nd;
111         int error;
112
113         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
114         if (error)
115                 goto out;
116
117         error = -EINVAL;
118         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
119                 goto exit;
120
121         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
122         if (error)
123                 goto exit;
124
125         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
126         error = PTR_ERR(file);
127         if (IS_ERR(file))
128                 goto out;
129
130         error = -ENOEXEC;
131         if(file->f_op) {
132                 struct linux_binfmt * fmt;
133
134                 read_lock(&binfmt_lock);
135                 list_for_each_entry(fmt, &formats, lh) {
136                         if (!fmt->load_shlib)
137                                 continue;
138                         if (!try_module_get(fmt->module))
139                                 continue;
140                         read_unlock(&binfmt_lock);
141                         error = fmt->load_shlib(file);
142                         read_lock(&binfmt_lock);
143                         put_binfmt(fmt);
144                         if (error != -ENOEXEC)
145                                 break;
146                 }
147                 read_unlock(&binfmt_lock);
148         }
149         fput(file);
150 out:
151         return error;
152 exit:
153         release_open_intent(&nd);
154         path_put(&nd.path);
155         goto out;
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 = -ENOMEM;
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                 goto err;
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
247         vma->vm_flags = VM_STACK_FLAGS;
248         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
249         err = insert_vm_struct(mm, vma);
250         if (err) {
251                 up_write(&mm->mmap_sem);
252                 goto err;
253         }
254
255         mm->stack_vm = mm->total_vm = 1;
256         up_write(&mm->mmap_sem);
257
258         bprm->p = vma->vm_end - sizeof(void *);
259
260         return 0;
261
262 err:
263         if (vma) {
264                 bprm->vma = NULL;
265                 kmem_cache_free(vm_area_cachep, vma);
266         }
267
268         return err;
269 }
270
271 static bool valid_arg_len(struct linux_binprm *bprm, long len)
272 {
273         return len <= MAX_ARG_STRLEN;
274 }
275
276 #else
277
278 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
279                 int write)
280 {
281         struct page *page;
282
283         page = bprm->page[pos / PAGE_SIZE];
284         if (!page && write) {
285                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
286                 if (!page)
287                         return NULL;
288                 bprm->page[pos / PAGE_SIZE] = page;
289         }
290
291         return page;
292 }
293
294 static void put_arg_page(struct page *page)
295 {
296 }
297
298 static void free_arg_page(struct linux_binprm *bprm, int i)
299 {
300         if (bprm->page[i]) {
301                 __free_page(bprm->page[i]);
302                 bprm->page[i] = NULL;
303         }
304 }
305
306 static void free_arg_pages(struct linux_binprm *bprm)
307 {
308         int i;
309
310         for (i = 0; i < MAX_ARG_PAGES; i++)
311                 free_arg_page(bprm, i);
312 }
313
314 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
315                 struct page *page)
316 {
317 }
318
319 static int __bprm_mm_init(struct linux_binprm *bprm)
320 {
321         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
322         return 0;
323 }
324
325 static bool valid_arg_len(struct linux_binprm *bprm, long len)
326 {
327         return len <= bprm->p;
328 }
329
330 #endif /* CONFIG_MMU */
331
332 /*
333  * Create a new mm_struct and populate it with a temporary stack
334  * vm_area_struct.  We don't have enough context at this point to set the stack
335  * flags, permissions, and offset, so we use temporary values.  We'll update
336  * them later in setup_arg_pages().
337  */
338 int bprm_mm_init(struct linux_binprm *bprm)
339 {
340         int err;
341         struct mm_struct *mm = NULL;
342
343         bprm->mm = mm = mm_alloc();
344         err = -ENOMEM;
345         if (!mm)
346                 goto err;
347
348         err = init_new_context(current, mm);
349         if (err)
350                 goto err;
351
352         err = __bprm_mm_init(bprm);
353         if (err)
354                 goto err;
355
356         return 0;
357
358 err:
359         if (mm) {
360                 bprm->mm = NULL;
361                 mmdrop(mm);
362         }
363
364         return err;
365 }
366
367 /*
368  * count() counts the number of strings in array ARGV.
369  */
370 static int count(char __user * __user * argv, int max)
371 {
372         int i = 0;
373
374         if (argv != NULL) {
375                 for (;;) {
376                         char __user * p;
377
378                         if (get_user(p, argv))
379                                 return -EFAULT;
380                         if (!p)
381                                 break;
382                         argv++;
383                         if(++i > max)
384                                 return -E2BIG;
385                         cond_resched();
386                 }
387         }
388         return i;
389 }
390
391 /*
392  * 'copy_strings()' copies argument/environment strings from the old
393  * processes's memory to the new process's stack.  The call to get_user_pages()
394  * ensures the destination page is created and not swapped out.
395  */
396 static int copy_strings(int argc, char __user * __user * argv,
397                         struct linux_binprm *bprm)
398 {
399         struct page *kmapped_page = NULL;
400         char *kaddr = NULL;
401         unsigned long kpos = 0;
402         int ret;
403
404         while (argc-- > 0) {
405                 char __user *str;
406                 int len;
407                 unsigned long pos;
408
409                 if (get_user(str, argv+argc) ||
410                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
411                         ret = -EFAULT;
412                         goto out;
413                 }
414
415                 if (!valid_arg_len(bprm, len)) {
416                         ret = -E2BIG;
417                         goto out;
418                 }
419
420                 /* We're going to work our way backwords. */
421                 pos = bprm->p;
422                 str += len;
423                 bprm->p -= len;
424
425                 while (len > 0) {
426                         int offset, bytes_to_copy;
427
428                         offset = pos % PAGE_SIZE;
429                         if (offset == 0)
430                                 offset = PAGE_SIZE;
431
432                         bytes_to_copy = offset;
433                         if (bytes_to_copy > len)
434                                 bytes_to_copy = len;
435
436                         offset -= bytes_to_copy;
437                         pos -= bytes_to_copy;
438                         str -= bytes_to_copy;
439                         len -= bytes_to_copy;
440
441                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
442                                 struct page *page;
443
444                                 page = get_arg_page(bprm, pos, 1);
445                                 if (!page) {
446                                         ret = -E2BIG;
447                                         goto out;
448                                 }
449
450                                 if (kmapped_page) {
451                                         flush_kernel_dcache_page(kmapped_page);
452                                         kunmap(kmapped_page);
453                                         put_arg_page(kmapped_page);
454                                 }
455                                 kmapped_page = page;
456                                 kaddr = kmap(kmapped_page);
457                                 kpos = pos & PAGE_MASK;
458                                 flush_arg_page(bprm, kpos, kmapped_page);
459                         }
460                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
461                                 ret = -EFAULT;
462                                 goto out;
463                         }
464                 }
465         }
466         ret = 0;
467 out:
468         if (kmapped_page) {
469                 flush_kernel_dcache_page(kmapped_page);
470                 kunmap(kmapped_page);
471                 put_arg_page(kmapped_page);
472         }
473         return ret;
474 }
475
476 /*
477  * Like copy_strings, but get argv and its values from kernel memory.
478  */
479 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
480 {
481         int r;
482         mm_segment_t oldfs = get_fs();
483         set_fs(KERNEL_DS);
484         r = copy_strings(argc, (char __user * __user *)argv, bprm);
485         set_fs(oldfs);
486         return r;
487 }
488 EXPORT_SYMBOL(copy_strings_kernel);
489
490 #ifdef CONFIG_MMU
491
492 /*
493  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
494  * the binfmt code determines where the new stack should reside, we shift it to
495  * its final location.  The process proceeds as follows:
496  *
497  * 1) Use shift to calculate the new vma endpoints.
498  * 2) Extend vma to cover both the old and new ranges.  This ensures the
499  *    arguments passed to subsequent functions are consistent.
500  * 3) Move vma's page tables to the new range.
501  * 4) Free up any cleared pgd range.
502  * 5) Shrink the vma to cover only the new range.
503  */
504 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
505 {
506         struct mm_struct *mm = vma->vm_mm;
507         unsigned long old_start = vma->vm_start;
508         unsigned long old_end = vma->vm_end;
509         unsigned long length = old_end - old_start;
510         unsigned long new_start = old_start - shift;
511         unsigned long new_end = old_end - shift;
512         struct mmu_gather *tlb;
513
514         BUG_ON(new_start > new_end);
515
516         /*
517          * ensure there are no vmas between where we want to go
518          * and where we are
519          */
520         if (vma != find_vma(mm, new_start))
521                 return -EFAULT;
522
523         /*
524          * cover the whole range: [new_start, old_end)
525          */
526         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
527
528         /*
529          * move the page tables downwards, on failure we rely on
530          * process cleanup to remove whatever mess we made.
531          */
532         if (length != move_page_tables(vma, old_start,
533                                        vma, new_start, length))
534                 return -ENOMEM;
535
536         lru_add_drain();
537         tlb = tlb_gather_mmu(mm, 0);
538         if (new_end > old_start) {
539                 /*
540                  * when the old and new regions overlap clear from new_end.
541                  */
542                 free_pgd_range(tlb, new_end, old_end, new_end,
543                         vma->vm_next ? vma->vm_next->vm_start : 0);
544         } else {
545                 /*
546                  * otherwise, clean from old_start; this is done to not touch
547                  * the address space in [new_end, old_start) some architectures
548                  * have constraints on va-space that make this illegal (IA64) -
549                  * for the others its just a little faster.
550                  */
551                 free_pgd_range(tlb, old_start, old_end, new_end,
552                         vma->vm_next ? vma->vm_next->vm_start : 0);
553         }
554         tlb_finish_mmu(tlb, new_end, old_end);
555
556         /*
557          * shrink the vma to just the new range.
558          */
559         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
560
561         return 0;
562 }
563
564 #define EXTRA_STACK_VM_PAGES    20      /* random */
565
566 /*
567  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
568  * the stack is optionally relocated, and some extra space is added.
569  */
570 int setup_arg_pages(struct linux_binprm *bprm,
571                     unsigned long stack_top,
572                     int executable_stack)
573 {
574         unsigned long ret;
575         unsigned long stack_shift;
576         struct mm_struct *mm = current->mm;
577         struct vm_area_struct *vma = bprm->vma;
578         struct vm_area_struct *prev = NULL;
579         unsigned long vm_flags;
580         unsigned long stack_base;
581
582 #ifdef CONFIG_STACK_GROWSUP
583         /* Limit stack size to 1GB */
584         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
585         if (stack_base > (1 << 30))
586                 stack_base = 1 << 30;
587
588         /* Make sure we didn't let the argument array grow too large. */
589         if (vma->vm_end - vma->vm_start > stack_base)
590                 return -ENOMEM;
591
592         stack_base = PAGE_ALIGN(stack_top - stack_base);
593
594         stack_shift = vma->vm_start - stack_base;
595         mm->arg_start = bprm->p - stack_shift;
596         bprm->p = vma->vm_end - stack_shift;
597 #else
598         stack_top = arch_align_stack(stack_top);
599         stack_top = PAGE_ALIGN(stack_top);
600         stack_shift = vma->vm_end - stack_top;
601
602         bprm->p -= stack_shift;
603         mm->arg_start = bprm->p;
604 #endif
605
606         if (bprm->loader)
607                 bprm->loader -= stack_shift;
608         bprm->exec -= stack_shift;
609
610         down_write(&mm->mmap_sem);
611         vm_flags = VM_STACK_FLAGS;
612
613         /*
614          * Adjust stack execute permissions; explicitly enable for
615          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
616          * (arch default) otherwise.
617          */
618         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
619                 vm_flags |= VM_EXEC;
620         else if (executable_stack == EXSTACK_DISABLE_X)
621                 vm_flags &= ~VM_EXEC;
622         vm_flags |= mm->def_flags;
623
624         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
625                         vm_flags);
626         if (ret)
627                 goto out_unlock;
628         BUG_ON(prev != vma);
629
630         /* Move stack pages down in memory. */
631         if (stack_shift) {
632                 ret = shift_arg_pages(vma, stack_shift);
633                 if (ret) {
634                         up_write(&mm->mmap_sem);
635                         return ret;
636                 }
637         }
638
639 #ifdef CONFIG_STACK_GROWSUP
640         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
641 #else
642         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
643 #endif
644         ret = expand_stack(vma, stack_base);
645         if (ret)
646                 ret = -EFAULT;
647
648 out_unlock:
649         up_write(&mm->mmap_sem);
650         return 0;
651 }
652 EXPORT_SYMBOL(setup_arg_pages);
653
654 #endif /* CONFIG_MMU */
655
656 struct file *open_exec(const char *name)
657 {
658         struct nameidata nd;
659         int err;
660         struct file *file;
661
662         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
663         file = ERR_PTR(err);
664
665         if (!err) {
666                 struct inode *inode = nd.path.dentry->d_inode;
667                 file = ERR_PTR(-EACCES);
668                 if (S_ISREG(inode->i_mode)) {
669                         int err = vfs_permission(&nd, MAY_EXEC);
670                         file = ERR_PTR(err);
671                         if (!err) {
672                                 file = nameidata_to_filp(&nd,
673                                                         O_RDONLY|O_LARGEFILE);
674                                 if (!IS_ERR(file)) {
675                                         err = deny_write_access(file);
676                                         if (err) {
677                                                 fput(file);
678                                                 file = ERR_PTR(err);
679                                         }
680                                 }
681 out:
682                                 return file;
683                         }
684                 }
685                 release_open_intent(&nd);
686                 path_put(&nd.path);
687         }
688         goto out;
689 }
690
691 EXPORT_SYMBOL(open_exec);
692
693 int kernel_read(struct file *file, unsigned long offset,
694         char *addr, unsigned long count)
695 {
696         mm_segment_t old_fs;
697         loff_t pos = offset;
698         int result;
699
700         old_fs = get_fs();
701         set_fs(get_ds());
702         /* The cast to a user pointer is valid due to the set_fs() */
703         result = vfs_read(file, (void __user *)addr, count, &pos);
704         set_fs(old_fs);
705         return result;
706 }
707
708 EXPORT_SYMBOL(kernel_read);
709
710 static int exec_mmap(struct mm_struct *mm)
711 {
712         struct task_struct *tsk;
713         struct mm_struct * old_mm, *active_mm;
714
715         /* Notify parent that we're no longer interested in the old VM */
716         tsk = current;
717         old_mm = current->mm;
718         mm_release(tsk, old_mm);
719
720         if (old_mm) {
721                 /*
722                  * Make sure that if there is a core dump in progress
723                  * for the old mm, we get out and die instead of going
724                  * through with the exec.  We must hold mmap_sem around
725                  * checking core_waiters and changing tsk->mm.  The
726                  * core-inducing thread will increment core_waiters for
727                  * each thread whose ->mm == old_mm.
728                  */
729                 down_read(&old_mm->mmap_sem);
730                 if (unlikely(old_mm->core_waiters)) {
731                         up_read(&old_mm->mmap_sem);
732                         return -EINTR;
733                 }
734         }
735         task_lock(tsk);
736         active_mm = tsk->active_mm;
737         tsk->mm = mm;
738         tsk->active_mm = mm;
739         activate_mm(active_mm, mm);
740         task_unlock(tsk);
741         mm_update_next_owner(old_mm);
742         arch_pick_mmap_layout(mm);
743         if (old_mm) {
744                 up_read(&old_mm->mmap_sem);
745                 BUG_ON(active_mm != old_mm);
746                 mmput(old_mm);
747                 return 0;
748         }
749         mmdrop(active_mm);
750         return 0;
751 }
752
753 /*
754  * This function makes sure the current process has its own signal table,
755  * so that flush_signal_handlers can later reset the handlers without
756  * disturbing other processes.  (Other processes might share the signal
757  * table via the CLONE_SIGHAND option to clone().)
758  */
759 static int de_thread(struct task_struct *tsk)
760 {
761         struct signal_struct *sig = tsk->signal;
762         struct sighand_struct *oldsighand = tsk->sighand;
763         spinlock_t *lock = &oldsighand->siglock;
764         struct task_struct *leader = NULL;
765         int count;
766
767         if (thread_group_empty(tsk))
768                 goto no_thread_group;
769
770         /*
771          * Kill all other threads in the thread group.
772          */
773         spin_lock_irq(lock);
774         if (signal_group_exit(sig)) {
775                 /*
776                  * Another group action in progress, just
777                  * return so that the signal is processed.
778                  */
779                 spin_unlock_irq(lock);
780                 return -EAGAIN;
781         }
782         sig->group_exit_task = tsk;
783         zap_other_threads(tsk);
784
785         /* Account for the thread group leader hanging around: */
786         count = thread_group_leader(tsk) ? 1 : 2;
787         sig->notify_count = count;
788         while (atomic_read(&sig->count) > count) {
789                 __set_current_state(TASK_UNINTERRUPTIBLE);
790                 spin_unlock_irq(lock);
791                 schedule();
792                 spin_lock_irq(lock);
793         }
794         spin_unlock_irq(lock);
795
796         /*
797          * At this point all other threads have exited, all we have to
798          * do is to wait for the thread group leader to become inactive,
799          * and to assume its PID:
800          */
801         if (!thread_group_leader(tsk)) {
802                 leader = tsk->group_leader;
803
804                 sig->notify_count = -1; /* for exit_notify() */
805                 for (;;) {
806                         write_lock_irq(&tasklist_lock);
807                         if (likely(leader->exit_state))
808                                 break;
809                         __set_current_state(TASK_UNINTERRUPTIBLE);
810                         write_unlock_irq(&tasklist_lock);
811                         schedule();
812                 }
813
814                 if (unlikely(task_child_reaper(tsk) == leader))
815                         task_active_pid_ns(tsk)->child_reaper = tsk;
816                 /*
817                  * The only record we have of the real-time age of a
818                  * process, regardless of execs it's done, is start_time.
819                  * All the past CPU time is accumulated in signal_struct
820                  * from sister threads now dead.  But in this non-leader
821                  * exec, nothing survives from the original leader thread,
822                  * whose birth marks the true age of this process now.
823                  * When we take on its identity by switching to its PID, we
824                  * also take its birthdate (always earlier than our own).
825                  */
826                 tsk->start_time = leader->start_time;
827
828                 BUG_ON(!same_thread_group(leader, tsk));
829                 BUG_ON(has_group_leader_pid(tsk));
830                 /*
831                  * An exec() starts a new thread group with the
832                  * TGID of the previous thread group. Rehash the
833                  * two threads with a switched PID, and release
834                  * the former thread group leader:
835                  */
836
837                 /* Become a process group leader with the old leader's pid.
838                  * The old leader becomes a thread of the this thread group.
839                  * Note: The old leader also uses this pid until release_task
840                  *       is called.  Odd but simple and correct.
841                  */
842                 detach_pid(tsk, PIDTYPE_PID);
843                 tsk->pid = leader->pid;
844                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
845                 transfer_pid(leader, tsk, PIDTYPE_PGID);
846                 transfer_pid(leader, tsk, PIDTYPE_SID);
847                 list_replace_rcu(&leader->tasks, &tsk->tasks);
848
849                 tsk->group_leader = tsk;
850                 leader->group_leader = tsk;
851
852                 tsk->exit_signal = SIGCHLD;
853
854                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
855                 leader->exit_state = EXIT_DEAD;
856
857                 write_unlock_irq(&tasklist_lock);
858         }
859
860         sig->group_exit_task = NULL;
861         sig->notify_count = 0;
862
863 no_thread_group:
864         exit_itimers(sig);
865         flush_itimer_signals();
866         if (leader)
867                 release_task(leader);
868
869         if (atomic_read(&oldsighand->count) != 1) {
870                 struct sighand_struct *newsighand;
871                 /*
872                  * This ->sighand is shared with the CLONE_SIGHAND
873                  * but not CLONE_THREAD task, switch to the new one.
874                  */
875                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
876                 if (!newsighand)
877                         return -ENOMEM;
878
879                 atomic_set(&newsighand->count, 1);
880                 memcpy(newsighand->action, oldsighand->action,
881                        sizeof(newsighand->action));
882
883                 write_lock_irq(&tasklist_lock);
884                 spin_lock(&oldsighand->siglock);
885                 rcu_assign_pointer(tsk->sighand, newsighand);
886                 spin_unlock(&oldsighand->siglock);
887                 write_unlock_irq(&tasklist_lock);
888
889                 __cleanup_sighand(oldsighand);
890         }
891
892         BUG_ON(!thread_group_leader(tsk));
893         return 0;
894 }
895
896 /*
897  * These functions flushes out all traces of the currently running executable
898  * so that a new one can be started
899  */
900 static void flush_old_files(struct files_struct * files)
901 {
902         long j = -1;
903         struct fdtable *fdt;
904
905         spin_lock(&files->file_lock);
906         for (;;) {
907                 unsigned long set, i;
908
909                 j++;
910                 i = j * __NFDBITS;
911                 fdt = files_fdtable(files);
912                 if (i >= fdt->max_fds)
913                         break;
914                 set = fdt->close_on_exec->fds_bits[j];
915                 if (!set)
916                         continue;
917                 fdt->close_on_exec->fds_bits[j] = 0;
918                 spin_unlock(&files->file_lock);
919                 for ( ; set ; i++,set >>= 1) {
920                         if (set & 1) {
921                                 sys_close(i);
922                         }
923                 }
924                 spin_lock(&files->file_lock);
925
926         }
927         spin_unlock(&files->file_lock);
928 }
929
930 char *get_task_comm(char *buf, struct task_struct *tsk)
931 {
932         /* buf must be at least sizeof(tsk->comm) in size */
933         task_lock(tsk);
934         strncpy(buf, tsk->comm, sizeof(tsk->comm));
935         task_unlock(tsk);
936         return buf;
937 }
938
939 void set_task_comm(struct task_struct *tsk, char *buf)
940 {
941         task_lock(tsk);
942         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
943         task_unlock(tsk);
944 }
945
946 int flush_old_exec(struct linux_binprm * bprm)
947 {
948         char * name;
949         int i, ch, retval;
950         char tcomm[sizeof(current->comm)];
951
952         /*
953          * Make sure we have a private signal table and that
954          * we are unassociated from the previous thread group.
955          */
956         retval = de_thread(current);
957         if (retval)
958                 goto out;
959
960         set_mm_exe_file(bprm->mm, bprm->file);
961
962         /*
963          * Release all of the old mmap stuff
964          */
965         retval = exec_mmap(bprm->mm);
966         if (retval)
967                 goto out;
968
969         bprm->mm = NULL;                /* We're using it now */
970
971         /* This is the point of no return */
972         current->sas_ss_sp = current->sas_ss_size = 0;
973
974         if (current->euid == current->uid && current->egid == current->gid)
975                 set_dumpable(current->mm, 1);
976         else
977                 set_dumpable(current->mm, suid_dumpable);
978
979         name = bprm->filename;
980
981         /* Copies the binary name from after last slash */
982         for (i=0; (ch = *(name++)) != '\0';) {
983                 if (ch == '/')
984                         i = 0; /* overwrite what we wrote */
985                 else
986                         if (i < (sizeof(tcomm) - 1))
987                                 tcomm[i++] = ch;
988         }
989         tcomm[i] = '\0';
990         set_task_comm(current, tcomm);
991
992         current->flags &= ~PF_RANDOMIZE;
993         flush_thread();
994
995         /* Set the new mm task size. We have to do that late because it may
996          * depend on TIF_32BIT which is only updated in flush_thread() on
997          * some architectures like powerpc
998          */
999         current->mm->task_size = TASK_SIZE;
1000
1001         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1002                 suid_keys(current);
1003                 set_dumpable(current->mm, suid_dumpable);
1004                 current->pdeath_signal = 0;
1005         } else if (file_permission(bprm->file, MAY_READ) ||
1006                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1007                 suid_keys(current);
1008                 set_dumpable(current->mm, suid_dumpable);
1009         }
1010
1011         /* An exec changes our domain. We are no longer part of the thread
1012            group */
1013
1014         current->self_exec_id++;
1015                         
1016         flush_signal_handlers(current, 0);
1017         flush_old_files(current->files);
1018
1019         return 0;
1020
1021 out:
1022         return retval;
1023 }
1024
1025 EXPORT_SYMBOL(flush_old_exec);
1026
1027 /* 
1028  * Fill the binprm structure from the inode. 
1029  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1030  */
1031 int prepare_binprm(struct linux_binprm *bprm)
1032 {
1033         int mode;
1034         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1035         int retval;
1036
1037         mode = inode->i_mode;
1038         if (bprm->file->f_op == NULL)
1039                 return -EACCES;
1040
1041         bprm->e_uid = current->euid;
1042         bprm->e_gid = current->egid;
1043
1044         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1045                 /* Set-uid? */
1046                 if (mode & S_ISUID) {
1047                         current->personality &= ~PER_CLEAR_ON_SETID;
1048                         bprm->e_uid = inode->i_uid;
1049                 }
1050
1051                 /* Set-gid? */
1052                 /*
1053                  * If setgid is set but no group execute bit then this
1054                  * is a candidate for mandatory locking, not a setgid
1055                  * executable.
1056                  */
1057                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1058                         current->personality &= ~PER_CLEAR_ON_SETID;
1059                         bprm->e_gid = inode->i_gid;
1060                 }
1061         }
1062
1063         /* fill in binprm security blob */
1064         retval = security_bprm_set(bprm);
1065         if (retval)
1066                 return retval;
1067
1068         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1069         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1070 }
1071
1072 EXPORT_SYMBOL(prepare_binprm);
1073
1074 static int unsafe_exec(struct task_struct *p)
1075 {
1076         int unsafe = 0;
1077         if (p->ptrace & PT_PTRACED) {
1078                 if (p->ptrace & PT_PTRACE_CAP)
1079                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1080                 else
1081                         unsafe |= LSM_UNSAFE_PTRACE;
1082         }
1083         if (atomic_read(&p->fs->count) > 1 ||
1084             atomic_read(&p->files->count) > 1 ||
1085             atomic_read(&p->sighand->count) > 1)
1086                 unsafe |= LSM_UNSAFE_SHARE;
1087
1088         return unsafe;
1089 }
1090
1091 void compute_creds(struct linux_binprm *bprm)
1092 {
1093         int unsafe;
1094
1095         if (bprm->e_uid != current->uid) {
1096                 suid_keys(current);
1097                 current->pdeath_signal = 0;
1098         }
1099         exec_keys(current);
1100
1101         task_lock(current);
1102         unsafe = unsafe_exec(current);
1103         security_bprm_apply_creds(bprm, unsafe);
1104         task_unlock(current);
1105         security_bprm_post_apply_creds(bprm);
1106 }
1107 EXPORT_SYMBOL(compute_creds);
1108
1109 /*
1110  * Arguments are '\0' separated strings found at the location bprm->p
1111  * points to; chop off the first by relocating brpm->p to right after
1112  * the first '\0' encountered.
1113  */
1114 int remove_arg_zero(struct linux_binprm *bprm)
1115 {
1116         int ret = 0;
1117         unsigned long offset;
1118         char *kaddr;
1119         struct page *page;
1120
1121         if (!bprm->argc)
1122                 return 0;
1123
1124         do {
1125                 offset = bprm->p & ~PAGE_MASK;
1126                 page = get_arg_page(bprm, bprm->p, 0);
1127                 if (!page) {
1128                         ret = -EFAULT;
1129                         goto out;
1130                 }
1131                 kaddr = kmap_atomic(page, KM_USER0);
1132
1133                 for (; offset < PAGE_SIZE && kaddr[offset];
1134                                 offset++, bprm->p++)
1135                         ;
1136
1137                 kunmap_atomic(kaddr, KM_USER0);
1138                 put_arg_page(page);
1139
1140                 if (offset == PAGE_SIZE)
1141                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1142         } while (offset == PAGE_SIZE);
1143
1144         bprm->p++;
1145         bprm->argc--;
1146         ret = 0;
1147
1148 out:
1149         return ret;
1150 }
1151 EXPORT_SYMBOL(remove_arg_zero);
1152
1153 /*
1154  * cycle the list of binary formats handler, until one recognizes the image
1155  */
1156 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1157 {
1158         int try,retval;
1159         struct linux_binfmt *fmt;
1160 #ifdef __alpha__
1161         /* handle /sbin/loader.. */
1162         {
1163             struct exec * eh = (struct exec *) bprm->buf;
1164
1165             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1166                 (eh->fh.f_flags & 0x3000) == 0x3000)
1167             {
1168                 struct file * file;
1169                 unsigned long loader;
1170
1171                 allow_write_access(bprm->file);
1172                 fput(bprm->file);
1173                 bprm->file = NULL;
1174
1175                 loader = bprm->vma->vm_end - sizeof(void *);
1176
1177                 file = open_exec("/sbin/loader");
1178                 retval = PTR_ERR(file);
1179                 if (IS_ERR(file))
1180                         return retval;
1181
1182                 /* Remember if the application is TASO.  */
1183                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1184
1185                 bprm->file = file;
1186                 bprm->loader = loader;
1187                 retval = prepare_binprm(bprm);
1188                 if (retval<0)
1189                         return retval;
1190                 /* should call search_binary_handler recursively here,
1191                    but it does not matter */
1192             }
1193         }
1194 #endif
1195         retval = security_bprm_check(bprm);
1196         if (retval)
1197                 return retval;
1198
1199         /* kernel module loader fixup */
1200         /* so we don't try to load run modprobe in kernel space. */
1201         set_fs(USER_DS);
1202
1203         retval = audit_bprm(bprm);
1204         if (retval)
1205                 return retval;
1206
1207         retval = -ENOENT;
1208         for (try=0; try<2; try++) {
1209                 read_lock(&binfmt_lock);
1210                 list_for_each_entry(fmt, &formats, lh) {
1211                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1212                         if (!fn)
1213                                 continue;
1214                         if (!try_module_get(fmt->module))
1215                                 continue;
1216                         read_unlock(&binfmt_lock);
1217                         retval = fn(bprm, regs);
1218                         if (retval >= 0) {
1219                                 put_binfmt(fmt);
1220                                 allow_write_access(bprm->file);
1221                                 if (bprm->file)
1222                                         fput(bprm->file);
1223                                 bprm->file = NULL;
1224                                 current->did_exec = 1;
1225                                 proc_exec_connector(current);
1226                                 return retval;
1227                         }
1228                         read_lock(&binfmt_lock);
1229                         put_binfmt(fmt);
1230                         if (retval != -ENOEXEC || bprm->mm == NULL)
1231                                 break;
1232                         if (!bprm->file) {
1233                                 read_unlock(&binfmt_lock);
1234                                 return retval;
1235                         }
1236                 }
1237                 read_unlock(&binfmt_lock);
1238                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1239                         break;
1240 #ifdef CONFIG_KMOD
1241                 }else{
1242 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1243                         if (printable(bprm->buf[0]) &&
1244                             printable(bprm->buf[1]) &&
1245                             printable(bprm->buf[2]) &&
1246                             printable(bprm->buf[3]))
1247                                 break; /* -ENOEXEC */
1248                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1249 #endif
1250                 }
1251         }
1252         return retval;
1253 }
1254
1255 EXPORT_SYMBOL(search_binary_handler);
1256
1257 void free_bprm(struct linux_binprm *bprm)
1258 {
1259         free_arg_pages(bprm);
1260         kfree(bprm);
1261 }
1262
1263 /*
1264  * sys_execve() executes a new program.
1265  */
1266 int do_execve(char * filename,
1267         char __user *__user *argv,
1268         char __user *__user *envp,
1269         struct pt_regs * regs)
1270 {
1271         struct linux_binprm *bprm;
1272         struct file *file;
1273         struct files_struct *displaced;
1274         int retval;
1275
1276         retval = unshare_files(&displaced);
1277         if (retval)
1278                 goto out_ret;
1279
1280         retval = -ENOMEM;
1281         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1282         if (!bprm)
1283                 goto out_files;
1284
1285         file = open_exec(filename);
1286         retval = PTR_ERR(file);
1287         if (IS_ERR(file))
1288                 goto out_kfree;
1289
1290         sched_exec();
1291
1292         bprm->file = file;
1293         bprm->filename = filename;
1294         bprm->interp = filename;
1295
1296         retval = bprm_mm_init(bprm);
1297         if (retval)
1298                 goto out_file;
1299
1300         bprm->argc = count(argv, MAX_ARG_STRINGS);
1301         if ((retval = bprm->argc) < 0)
1302                 goto out_mm;
1303
1304         bprm->envc = count(envp, MAX_ARG_STRINGS);
1305         if ((retval = bprm->envc) < 0)
1306                 goto out_mm;
1307
1308         retval = security_bprm_alloc(bprm);
1309         if (retval)
1310                 goto out;
1311
1312         retval = prepare_binprm(bprm);
1313         if (retval < 0)
1314                 goto out;
1315
1316         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1317         if (retval < 0)
1318                 goto out;
1319
1320         bprm->exec = bprm->p;
1321         retval = copy_strings(bprm->envc, envp, bprm);
1322         if (retval < 0)
1323                 goto out;
1324
1325         retval = copy_strings(bprm->argc, argv, bprm);
1326         if (retval < 0)
1327                 goto out;
1328
1329         current->flags &= ~PF_KTHREAD;
1330         retval = search_binary_handler(bprm,regs);
1331         if (retval >= 0) {
1332                 /* execve success */
1333                 security_bprm_free(bprm);
1334                 acct_update_integrals(current);
1335                 free_bprm(bprm);
1336                 if (displaced)
1337                         put_files_struct(displaced);
1338                 return retval;
1339         }
1340
1341 out:
1342         if (bprm->security)
1343                 security_bprm_free(bprm);
1344
1345 out_mm:
1346         if (bprm->mm)
1347                 mmput (bprm->mm);
1348
1349 out_file:
1350         if (bprm->file) {
1351                 allow_write_access(bprm->file);
1352                 fput(bprm->file);
1353         }
1354 out_kfree:
1355         free_bprm(bprm);
1356
1357 out_files:
1358         if (displaced)
1359                 reset_files_struct(displaced);
1360 out_ret:
1361         return retval;
1362 }
1363
1364 int set_binfmt(struct linux_binfmt *new)
1365 {
1366         struct linux_binfmt *old = current->binfmt;
1367
1368         if (new) {
1369                 if (!try_module_get(new->module))
1370                         return -1;
1371         }
1372         current->binfmt = new;
1373         if (old)
1374                 module_put(old->module);
1375         return 0;
1376 }
1377
1378 EXPORT_SYMBOL(set_binfmt);
1379
1380 /* format_corename will inspect the pattern parameter, and output a
1381  * name into corename, which must have space for at least
1382  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1383  */
1384 static int format_corename(char *corename, const char *pattern, long signr)
1385 {
1386         const char *pat_ptr = pattern;
1387         char *out_ptr = corename;
1388         char *const out_end = corename + CORENAME_MAX_SIZE;
1389         int rc;
1390         int pid_in_pattern = 0;
1391         int ispipe = 0;
1392
1393         if (*pattern == '|')
1394                 ispipe = 1;
1395
1396         /* Repeat as long as we have more pattern to process and more output
1397            space */
1398         while (*pat_ptr) {
1399                 if (*pat_ptr != '%') {
1400                         if (out_ptr == out_end)
1401                                 goto out;
1402                         *out_ptr++ = *pat_ptr++;
1403                 } else {
1404                         switch (*++pat_ptr) {
1405                         case 0:
1406                                 goto out;
1407                         /* Double percent, output one percent */
1408                         case '%':
1409                                 if (out_ptr == out_end)
1410                                         goto out;
1411                                 *out_ptr++ = '%';
1412                                 break;
1413                         /* pid */
1414                         case 'p':
1415                                 pid_in_pattern = 1;
1416                                 rc = snprintf(out_ptr, out_end - out_ptr,
1417                                               "%d", task_tgid_vnr(current));
1418                                 if (rc > out_end - out_ptr)
1419                                         goto out;
1420                                 out_ptr += rc;
1421                                 break;
1422                         /* uid */
1423                         case 'u':
1424                                 rc = snprintf(out_ptr, out_end - out_ptr,
1425                                               "%d", current->uid);
1426                                 if (rc > out_end - out_ptr)
1427                                         goto out;
1428                                 out_ptr += rc;
1429                                 break;
1430                         /* gid */
1431                         case 'g':
1432                                 rc = snprintf(out_ptr, out_end - out_ptr,
1433                                               "%d", current->gid);
1434                                 if (rc > out_end - out_ptr)
1435                                         goto out;
1436                                 out_ptr += rc;
1437                                 break;
1438                         /* signal that caused the coredump */
1439                         case 's':
1440                                 rc = snprintf(out_ptr, out_end - out_ptr,
1441                                               "%ld", signr);
1442                                 if (rc > out_end - out_ptr)
1443                                         goto out;
1444                                 out_ptr += rc;
1445                                 break;
1446                         /* UNIX time of coredump */
1447                         case 't': {
1448                                 struct timeval tv;
1449                                 do_gettimeofday(&tv);
1450                                 rc = snprintf(out_ptr, out_end - out_ptr,
1451                                               "%lu", tv.tv_sec);
1452                                 if (rc > out_end - out_ptr)
1453                                         goto out;
1454                                 out_ptr += rc;
1455                                 break;
1456                         }
1457                         /* hostname */
1458                         case 'h':
1459                                 down_read(&uts_sem);
1460                                 rc = snprintf(out_ptr, out_end - out_ptr,
1461                                               "%s", utsname()->nodename);
1462                                 up_read(&uts_sem);
1463                                 if (rc > out_end - out_ptr)
1464                                         goto out;
1465                                 out_ptr += rc;
1466                                 break;
1467                         /* executable */
1468                         case 'e':
1469                                 rc = snprintf(out_ptr, out_end - out_ptr,
1470                                               "%s", current->comm);
1471                                 if (rc > out_end - out_ptr)
1472                                         goto out;
1473                                 out_ptr += rc;
1474                                 break;
1475                         /* core limit size */
1476                         case 'c':
1477                                 rc = snprintf(out_ptr, out_end - out_ptr,
1478                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1479                                 if (rc > out_end - out_ptr)
1480                                         goto out;
1481                                 out_ptr += rc;
1482                                 break;
1483                         default:
1484                                 break;
1485                         }
1486                         ++pat_ptr;
1487                 }
1488         }
1489         /* Backward compatibility with core_uses_pid:
1490          *
1491          * If core_pattern does not include a %p (as is the default)
1492          * and core_uses_pid is set, then .%pid will be appended to
1493          * the filename. Do not do this for piped commands. */
1494         if (!ispipe && !pid_in_pattern
1495             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1496                 rc = snprintf(out_ptr, out_end - out_ptr,
1497                               ".%d", task_tgid_vnr(current));
1498                 if (rc > out_end - out_ptr)
1499                         goto out;
1500                 out_ptr += rc;
1501         }
1502 out:
1503         *out_ptr = 0;
1504         return ispipe;
1505 }
1506
1507 static void zap_process(struct task_struct *start)
1508 {
1509         struct task_struct *t;
1510
1511         start->signal->flags = SIGNAL_GROUP_EXIT;
1512         start->signal->group_stop_count = 0;
1513
1514         t = start;
1515         do {
1516                 if (t != current && t->mm) {
1517                         t->mm->core_waiters++;
1518                         sigaddset(&t->pending.signal, SIGKILL);
1519                         signal_wake_up(t, 1);
1520                 }
1521         } while_each_thread(start, t);
1522 }
1523
1524 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1525                                 int exit_code)
1526 {
1527         struct task_struct *g, *p;
1528         unsigned long flags;
1529         int err = -EAGAIN;
1530
1531         spin_lock_irq(&tsk->sighand->siglock);
1532         if (!signal_group_exit(tsk->signal)) {
1533                 tsk->signal->group_exit_code = exit_code;
1534                 zap_process(tsk);
1535                 err = 0;
1536         }
1537         spin_unlock_irq(&tsk->sighand->siglock);
1538         if (err)
1539                 return err;
1540
1541         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1542                 goto done;
1543         /*
1544          * We should find and kill all tasks which use this mm, and we should
1545          * count them correctly into mm->core_waiters. We don't take tasklist
1546          * lock, but this is safe wrt:
1547          *
1548          * fork:
1549          *      None of sub-threads can fork after zap_process(leader). All
1550          *      processes which were created before this point should be
1551          *      visible to zap_threads() because copy_process() adds the new
1552          *      process to the tail of init_task.tasks list, and lock/unlock
1553          *      of ->siglock provides a memory barrier.
1554          *
1555          * do_exit:
1556          *      The caller holds mm->mmap_sem. This means that the task which
1557          *      uses this mm can't pass exit_mm(), so it can't exit or clear
1558          *      its ->mm.
1559          *
1560          * de_thread:
1561          *      It does list_replace_rcu(&leader->tasks, &current->tasks),
1562          *      we must see either old or new leader, this does not matter.
1563          *      However, it can change p->sighand, so lock_task_sighand(p)
1564          *      must be used. Since p->mm != NULL and we hold ->mmap_sem
1565          *      it can't fail.
1566          *
1567          *      Note also that "g" can be the old leader with ->mm == NULL
1568          *      and already unhashed and thus removed from ->thread_group.
1569          *      This is OK, __unhash_process()->list_del_rcu() does not
1570          *      clear the ->next pointer, we will find the new leader via
1571          *      next_thread().
1572          */
1573         rcu_read_lock();
1574         for_each_process(g) {
1575                 if (g == tsk->group_leader)
1576                         continue;
1577                 if (g->flags & PF_KTHREAD)
1578                         continue;
1579                 p = g;
1580                 do {
1581                         if (p->mm) {
1582                                 if (unlikely(p->mm == mm)) {
1583                                         lock_task_sighand(p, &flags);
1584                                         zap_process(p);
1585                                         unlock_task_sighand(p, &flags);
1586                                 }
1587                                 break;
1588                         }
1589                 } while_each_thread(g, p);
1590         }
1591         rcu_read_unlock();
1592 done:
1593         return mm->core_waiters;
1594 }
1595
1596 static int coredump_wait(int exit_code)
1597 {
1598         struct task_struct *tsk = current;
1599         struct mm_struct *mm = tsk->mm;
1600         struct core_state core_state;
1601         struct completion *vfork_done;
1602         int core_waiters;
1603
1604         init_completion(&mm->core_done);
1605         init_completion(&core_state.startup);
1606         mm->core_state = &core_state;
1607
1608         core_waiters = zap_threads(tsk, mm, exit_code);
1609         up_write(&mm->mmap_sem);
1610
1611         if (unlikely(core_waiters < 0))
1612                 goto fail;
1613
1614         /*
1615          * Make sure nobody is waiting for us to release the VM,
1616          * otherwise we can deadlock when we wait on each other
1617          */
1618         vfork_done = tsk->vfork_done;
1619         if (vfork_done) {
1620                 tsk->vfork_done = NULL;
1621                 complete(vfork_done);
1622         }
1623
1624         if (core_waiters)
1625                 wait_for_completion(&core_state.startup);
1626 fail:
1627         BUG_ON(mm->core_waiters);
1628         return core_waiters;
1629 }
1630
1631 /*
1632  * set_dumpable converts traditional three-value dumpable to two flags and
1633  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1634  * these bits are not changed atomically.  So get_dumpable can observe the
1635  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1636  * return either old dumpable or new one by paying attention to the order of
1637  * modifying the bits.
1638  *
1639  * dumpable |   mm->flags (binary)
1640  * old  new | initial interim  final
1641  * ---------+-----------------------
1642  *  0    1  |   00      01      01
1643  *  0    2  |   00      10(*)   11
1644  *  1    0  |   01      00      00
1645  *  1    2  |   01      11      11
1646  *  2    0  |   11      10(*)   00
1647  *  2    1  |   11      11      01
1648  *
1649  * (*) get_dumpable regards interim value of 10 as 11.
1650  */
1651 void set_dumpable(struct mm_struct *mm, int value)
1652 {
1653         switch (value) {
1654         case 0:
1655                 clear_bit(MMF_DUMPABLE, &mm->flags);
1656                 smp_wmb();
1657                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1658                 break;
1659         case 1:
1660                 set_bit(MMF_DUMPABLE, &mm->flags);
1661                 smp_wmb();
1662                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1663                 break;
1664         case 2:
1665                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1666                 smp_wmb();
1667                 set_bit(MMF_DUMPABLE, &mm->flags);
1668                 break;
1669         }
1670 }
1671
1672 int get_dumpable(struct mm_struct *mm)
1673 {
1674         int ret;
1675
1676         ret = mm->flags & 0x3;
1677         return (ret >= 2) ? 2 : ret;
1678 }
1679
1680 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1681 {
1682         char corename[CORENAME_MAX_SIZE + 1];
1683         struct mm_struct *mm = current->mm;
1684         struct linux_binfmt * binfmt;
1685         struct inode * inode;
1686         struct file * file;
1687         int retval = 0;
1688         int fsuid = current->fsuid;
1689         int flag = 0;
1690         int ispipe = 0;
1691         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1692         char **helper_argv = NULL;
1693         int helper_argc = 0;
1694         char *delimit;
1695
1696         audit_core_dumps(signr);
1697
1698         binfmt = current->binfmt;
1699         if (!binfmt || !binfmt->core_dump)
1700                 goto fail;
1701         down_write(&mm->mmap_sem);
1702         /*
1703          * If another thread got here first, or we are not dumpable, bail out.
1704          */
1705         if (mm->core_waiters || !get_dumpable(mm)) {
1706                 up_write(&mm->mmap_sem);
1707                 goto fail;
1708         }
1709
1710         /*
1711          *      We cannot trust fsuid as being the "true" uid of the
1712          *      process nor do we know its entire history. We only know it
1713          *      was tainted so we dump it as root in mode 2.
1714          */
1715         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1716                 flag = O_EXCL;          /* Stop rewrite attacks */
1717                 current->fsuid = 0;     /* Dump root private */
1718         }
1719
1720         retval = coredump_wait(exit_code);
1721         if (retval < 0)
1722                 goto fail;
1723
1724         /*
1725          * Clear any false indication of pending signals that might
1726          * be seen by the filesystem code called to write the core file.
1727          */
1728         clear_thread_flag(TIF_SIGPENDING);
1729
1730         /*
1731          * lock_kernel() because format_corename() is controlled by sysctl, which
1732          * uses lock_kernel()
1733          */
1734         lock_kernel();
1735         ispipe = format_corename(corename, core_pattern, signr);
1736         unlock_kernel();
1737         /*
1738          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1739          * to a pipe.  Since we're not writing directly to the filesystem
1740          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1741          * created unless the pipe reader choses to write out the core file
1742          * at which point file size limits and permissions will be imposed
1743          * as it does with any other process
1744          */
1745         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1746                 goto fail_unlock;
1747
1748         if (ispipe) {
1749                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1750                 /* Terminate the string before the first option */
1751                 delimit = strchr(corename, ' ');
1752                 if (delimit)
1753                         *delimit = '\0';
1754                 delimit = strrchr(helper_argv[0], '/');
1755                 if (delimit)
1756                         delimit++;
1757                 else
1758                         delimit = helper_argv[0];
1759                 if (!strcmp(delimit, current->comm)) {
1760                         printk(KERN_NOTICE "Recursive core dump detected, "
1761                                         "aborting\n");
1762                         goto fail_unlock;
1763                 }
1764
1765                 core_limit = RLIM_INFINITY;
1766
1767                 /* SIGPIPE can happen, but it's just never processed */
1768                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1769                                 &file)) {
1770                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1771                                corename);
1772                         goto fail_unlock;
1773                 }
1774         } else
1775                 file = filp_open(corename,
1776                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1777                                  0600);
1778         if (IS_ERR(file))
1779                 goto fail_unlock;
1780         inode = file->f_path.dentry->d_inode;
1781         if (inode->i_nlink > 1)
1782                 goto close_fail;        /* multiple links - don't dump */
1783         if (!ispipe && d_unhashed(file->f_path.dentry))
1784                 goto close_fail;
1785
1786         /* AK: actually i see no reason to not allow this for named pipes etc.,
1787            but keep the previous behaviour for now. */
1788         if (!ispipe && !S_ISREG(inode->i_mode))
1789                 goto close_fail;
1790         /*
1791          * Dont allow local users get cute and trick others to coredump
1792          * into their pre-created files:
1793          */
1794         if (inode->i_uid != current->fsuid)
1795                 goto close_fail;
1796         if (!file->f_op)
1797                 goto close_fail;
1798         if (!file->f_op->write)
1799                 goto close_fail;
1800         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1801                 goto close_fail;
1802
1803         retval = binfmt->core_dump(signr, regs, file, core_limit);
1804
1805         if (retval)
1806                 current->signal->group_exit_code |= 0x80;
1807 close_fail:
1808         filp_close(file, NULL);
1809 fail_unlock:
1810         if (helper_argv)
1811                 argv_free(helper_argv);
1812
1813         current->fsuid = fsuid;
1814         complete_all(&mm->core_done);
1815 fail:
1816         return retval;
1817 }