fuse: fix permission checking on sticky directories
[linux-2.6] / 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/mman.h>
28 #include <linux/a.out.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/string.h>
33 #include <linux/init.h>
34 #include <linux/pagemap.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/swap.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/ptrace.h>
47 #include <linux/mount.h>
48 #include <linux/security.h>
49 #include <linux/syscalls.h>
50 #include <linux/rmap.h>
51 #include <linux/tsacct_kern.h>
52 #include <linux/cn_proc.h>
53 #include <linux/audit.h>
54
55 #include <asm/uaccess.h>
56 #include <asm/mmu_context.h>
57 #include <asm/tlb.h>
58
59 #ifdef CONFIG_KMOD
60 #include <linux/kmod.h>
61 #endif
62
63 int core_uses_pid;
64 char core_pattern[CORENAME_MAX_SIZE] = "core";
65 int suid_dumpable = 0;
66
67 EXPORT_SYMBOL(suid_dumpable);
68 /* The maximal length of core_pattern is also specified in sysctl.c */
69
70 static LIST_HEAD(formats);
71 static DEFINE_RWLOCK(binfmt_lock);
72
73 int register_binfmt(struct linux_binfmt * fmt)
74 {
75         if (!fmt)
76                 return -EINVAL;
77         write_lock(&binfmt_lock);
78         list_add(&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 asmlinkage long sys_uselib(const char __user * library)
106 {
107         struct file * file;
108         struct nameidata nd;
109         int error;
110
111         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
112         if (error)
113                 goto out;
114
115         error = -EINVAL;
116         if (!S_ISREG(nd.dentry->d_inode->i_mode))
117                 goto exit;
118
119         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
120         if (error)
121                 goto exit;
122
123         file = nameidata_to_filp(&nd, O_RDONLY);
124         error = PTR_ERR(file);
125         if (IS_ERR(file))
126                 goto out;
127
128         error = -ENOEXEC;
129         if(file->f_op) {
130                 struct linux_binfmt * fmt;
131
132                 read_lock(&binfmt_lock);
133                 list_for_each_entry(fmt, &formats, lh) {
134                         if (!fmt->load_shlib)
135                                 continue;
136                         if (!try_module_get(fmt->module))
137                                 continue;
138                         read_unlock(&binfmt_lock);
139                         error = fmt->load_shlib(file);
140                         read_lock(&binfmt_lock);
141                         put_binfmt(fmt);
142                         if (error != -ENOEXEC)
143                                 break;
144                 }
145                 read_unlock(&binfmt_lock);
146         }
147         fput(file);
148 out:
149         return error;
150 exit:
151         release_open_intent(&nd);
152         path_release(&nd);
153         goto out;
154 }
155
156 #ifdef CONFIG_MMU
157
158 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
159                 int write)
160 {
161         struct page *page;
162         int ret;
163
164 #ifdef CONFIG_STACK_GROWSUP
165         if (write) {
166                 ret = expand_stack_downwards(bprm->vma, pos);
167                 if (ret < 0)
168                         return NULL;
169         }
170 #endif
171         ret = get_user_pages(current, bprm->mm, pos,
172                         1, write, 1, &page, NULL);
173         if (ret <= 0)
174                 return NULL;
175
176         if (write) {
177                 struct rlimit *rlim = current->signal->rlim;
178                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
179
180                 /*
181                  * Limit to 1/4-th the stack size for the argv+env strings.
182                  * This ensures that:
183                  *  - the remaining binfmt code will not run out of stack space,
184                  *  - the program will have a reasonable amount of stack left
185                  *    to work from.
186                  */
187                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
188                         put_page(page);
189                         return NULL;
190                 }
191         }
192
193         return page;
194 }
195
196 static void put_arg_page(struct page *page)
197 {
198         put_page(page);
199 }
200
201 static void free_arg_page(struct linux_binprm *bprm, int i)
202 {
203 }
204
205 static void free_arg_pages(struct linux_binprm *bprm)
206 {
207 }
208
209 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
210                 struct page *page)
211 {
212         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
213 }
214
215 static int __bprm_mm_init(struct linux_binprm *bprm)
216 {
217         int err = -ENOMEM;
218         struct vm_area_struct *vma = NULL;
219         struct mm_struct *mm = bprm->mm;
220
221         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
222         if (!vma)
223                 goto err;
224
225         down_write(&mm->mmap_sem);
226         vma->vm_mm = mm;
227
228         /*
229          * Place the stack at the largest stack address the architecture
230          * supports. Later, we'll move this to an appropriate place. We don't
231          * use STACK_TOP because that can depend on attributes which aren't
232          * configured yet.
233          */
234         vma->vm_end = STACK_TOP_MAX;
235         vma->vm_start = vma->vm_end - PAGE_SIZE;
236
237         vma->vm_flags = VM_STACK_FLAGS;
238         vma->vm_page_prot = protection_map[vma->vm_flags & 0x7];
239         err = insert_vm_struct(mm, vma);
240         if (err) {
241                 up_write(&mm->mmap_sem);
242                 goto err;
243         }
244
245         mm->stack_vm = mm->total_vm = 1;
246         up_write(&mm->mmap_sem);
247
248         bprm->p = vma->vm_end - sizeof(void *);
249
250         return 0;
251
252 err:
253         if (vma) {
254                 bprm->vma = NULL;
255                 kmem_cache_free(vm_area_cachep, vma);
256         }
257
258         return err;
259 }
260
261 static bool valid_arg_len(struct linux_binprm *bprm, long len)
262 {
263         return len <= MAX_ARG_STRLEN;
264 }
265
266 #else
267
268 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
269                 int write)
270 {
271         struct page *page;
272
273         page = bprm->page[pos / PAGE_SIZE];
274         if (!page && write) {
275                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
276                 if (!page)
277                         return NULL;
278                 bprm->page[pos / PAGE_SIZE] = page;
279         }
280
281         return page;
282 }
283
284 static void put_arg_page(struct page *page)
285 {
286 }
287
288 static void free_arg_page(struct linux_binprm *bprm, int i)
289 {
290         if (bprm->page[i]) {
291                 __free_page(bprm->page[i]);
292                 bprm->page[i] = NULL;
293         }
294 }
295
296 static void free_arg_pages(struct linux_binprm *bprm)
297 {
298         int i;
299
300         for (i = 0; i < MAX_ARG_PAGES; i++)
301                 free_arg_page(bprm, i);
302 }
303
304 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
305                 struct page *page)
306 {
307 }
308
309 static int __bprm_mm_init(struct linux_binprm *bprm)
310 {
311         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
312         return 0;
313 }
314
315 static bool valid_arg_len(struct linux_binprm *bprm, long len)
316 {
317         return len <= bprm->p;
318 }
319
320 #endif /* CONFIG_MMU */
321
322 /*
323  * Create a new mm_struct and populate it with a temporary stack
324  * vm_area_struct.  We don't have enough context at this point to set the stack
325  * flags, permissions, and offset, so we use temporary values.  We'll update
326  * them later in setup_arg_pages().
327  */
328 int bprm_mm_init(struct linux_binprm *bprm)
329 {
330         int err;
331         struct mm_struct *mm = NULL;
332
333         bprm->mm = mm = mm_alloc();
334         err = -ENOMEM;
335         if (!mm)
336                 goto err;
337
338         err = init_new_context(current, mm);
339         if (err)
340                 goto err;
341
342         err = __bprm_mm_init(bprm);
343         if (err)
344                 goto err;
345
346         return 0;
347
348 err:
349         if (mm) {
350                 bprm->mm = NULL;
351                 mmdrop(mm);
352         }
353
354         return err;
355 }
356
357 /*
358  * count() counts the number of strings in array ARGV.
359  */
360 static int count(char __user * __user * argv, int max)
361 {
362         int i = 0;
363
364         if (argv != NULL) {
365                 for (;;) {
366                         char __user * p;
367
368                         if (get_user(p, argv))
369                                 return -EFAULT;
370                         if (!p)
371                                 break;
372                         argv++;
373                         if(++i > max)
374                                 return -E2BIG;
375                         cond_resched();
376                 }
377         }
378         return i;
379 }
380
381 /*
382  * 'copy_strings()' copies argument/environment strings from the old
383  * processes's memory to the new process's stack.  The call to get_user_pages()
384  * ensures the destination page is created and not swapped out.
385  */
386 static int copy_strings(int argc, char __user * __user * argv,
387                         struct linux_binprm *bprm)
388 {
389         struct page *kmapped_page = NULL;
390         char *kaddr = NULL;
391         unsigned long kpos = 0;
392         int ret;
393
394         while (argc-- > 0) {
395                 char __user *str;
396                 int len;
397                 unsigned long pos;
398
399                 if (get_user(str, argv+argc) ||
400                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
401                         ret = -EFAULT;
402                         goto out;
403                 }
404
405                 if (!valid_arg_len(bprm, len)) {
406                         ret = -E2BIG;
407                         goto out;
408                 }
409
410                 /* We're going to work our way backwords. */
411                 pos = bprm->p;
412                 str += len;
413                 bprm->p -= len;
414
415                 while (len > 0) {
416                         int offset, bytes_to_copy;
417
418                         offset = pos % PAGE_SIZE;
419                         if (offset == 0)
420                                 offset = PAGE_SIZE;
421
422                         bytes_to_copy = offset;
423                         if (bytes_to_copy > len)
424                                 bytes_to_copy = len;
425
426                         offset -= bytes_to_copy;
427                         pos -= bytes_to_copy;
428                         str -= bytes_to_copy;
429                         len -= bytes_to_copy;
430
431                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
432                                 struct page *page;
433
434                                 page = get_arg_page(bprm, pos, 1);
435                                 if (!page) {
436                                         ret = -E2BIG;
437                                         goto out;
438                                 }
439
440                                 if (kmapped_page) {
441                                         flush_kernel_dcache_page(kmapped_page);
442                                         kunmap(kmapped_page);
443                                         put_arg_page(kmapped_page);
444                                 }
445                                 kmapped_page = page;
446                                 kaddr = kmap(kmapped_page);
447                                 kpos = pos & PAGE_MASK;
448                                 flush_arg_page(bprm, kpos, kmapped_page);
449                         }
450                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
451                                 ret = -EFAULT;
452                                 goto out;
453                         }
454                 }
455         }
456         ret = 0;
457 out:
458         if (kmapped_page) {
459                 flush_kernel_dcache_page(kmapped_page);
460                 kunmap(kmapped_page);
461                 put_arg_page(kmapped_page);
462         }
463         return ret;
464 }
465
466 /*
467  * Like copy_strings, but get argv and its values from kernel memory.
468  */
469 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
470 {
471         int r;
472         mm_segment_t oldfs = get_fs();
473         set_fs(KERNEL_DS);
474         r = copy_strings(argc, (char __user * __user *)argv, bprm);
475         set_fs(oldfs);
476         return r;
477 }
478 EXPORT_SYMBOL(copy_strings_kernel);
479
480 #ifdef CONFIG_MMU
481
482 /*
483  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
484  * the binfmt code determines where the new stack should reside, we shift it to
485  * its final location.  The process proceeds as follows:
486  *
487  * 1) Use shift to calculate the new vma endpoints.
488  * 2) Extend vma to cover both the old and new ranges.  This ensures the
489  *    arguments passed to subsequent functions are consistent.
490  * 3) Move vma's page tables to the new range.
491  * 4) Free up any cleared pgd range.
492  * 5) Shrink the vma to cover only the new range.
493  */
494 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
495 {
496         struct mm_struct *mm = vma->vm_mm;
497         unsigned long old_start = vma->vm_start;
498         unsigned long old_end = vma->vm_end;
499         unsigned long length = old_end - old_start;
500         unsigned long new_start = old_start - shift;
501         unsigned long new_end = old_end - shift;
502         struct mmu_gather *tlb;
503
504         BUG_ON(new_start > new_end);
505
506         /*
507          * ensure there are no vmas between where we want to go
508          * and where we are
509          */
510         if (vma != find_vma(mm, new_start))
511                 return -EFAULT;
512
513         /*
514          * cover the whole range: [new_start, old_end)
515          */
516         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
517
518         /*
519          * move the page tables downwards, on failure we rely on
520          * process cleanup to remove whatever mess we made.
521          */
522         if (length != move_page_tables(vma, old_start,
523                                        vma, new_start, length))
524                 return -ENOMEM;
525
526         lru_add_drain();
527         tlb = tlb_gather_mmu(mm, 0);
528         if (new_end > old_start) {
529                 /*
530                  * when the old and new regions overlap clear from new_end.
531                  */
532                 free_pgd_range(&tlb, new_end, old_end, new_end,
533                         vma->vm_next ? vma->vm_next->vm_start : 0);
534         } else {
535                 /*
536                  * otherwise, clean from old_start; this is done to not touch
537                  * the address space in [new_end, old_start) some architectures
538                  * have constraints on va-space that make this illegal (IA64) -
539                  * for the others its just a little faster.
540                  */
541                 free_pgd_range(&tlb, old_start, old_end, new_end,
542                         vma->vm_next ? vma->vm_next->vm_start : 0);
543         }
544         tlb_finish_mmu(tlb, new_end, old_end);
545
546         /*
547          * shrink the vma to just the new range.
548          */
549         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
550
551         return 0;
552 }
553
554 #define EXTRA_STACK_VM_PAGES    20      /* random */
555
556 /*
557  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
558  * the stack is optionally relocated, and some extra space is added.
559  */
560 int setup_arg_pages(struct linux_binprm *bprm,
561                     unsigned long stack_top,
562                     int executable_stack)
563 {
564         unsigned long ret;
565         unsigned long stack_shift;
566         struct mm_struct *mm = current->mm;
567         struct vm_area_struct *vma = bprm->vma;
568         struct vm_area_struct *prev = NULL;
569         unsigned long vm_flags;
570         unsigned long stack_base;
571
572 #ifdef CONFIG_STACK_GROWSUP
573         /* Limit stack size to 1GB */
574         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
575         if (stack_base > (1 << 30))
576                 stack_base = 1 << 30;
577
578         /* Make sure we didn't let the argument array grow too large. */
579         if (vma->vm_end - vma->vm_start > stack_base)
580                 return -ENOMEM;
581
582         stack_base = PAGE_ALIGN(stack_top - stack_base);
583
584         stack_shift = vma->vm_start - stack_base;
585         mm->arg_start = bprm->p - stack_shift;
586         bprm->p = vma->vm_end - stack_shift;
587 #else
588         stack_top = arch_align_stack(stack_top);
589         stack_top = PAGE_ALIGN(stack_top);
590         stack_shift = vma->vm_end - stack_top;
591
592         bprm->p -= stack_shift;
593         mm->arg_start = bprm->p;
594 #endif
595
596         if (bprm->loader)
597                 bprm->loader -= stack_shift;
598         bprm->exec -= stack_shift;
599
600         down_write(&mm->mmap_sem);
601         vm_flags = vma->vm_flags;
602
603         /*
604          * Adjust stack execute permissions; explicitly enable for
605          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
606          * (arch default) otherwise.
607          */
608         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
609                 vm_flags |= VM_EXEC;
610         else if (executable_stack == EXSTACK_DISABLE_X)
611                 vm_flags &= ~VM_EXEC;
612         vm_flags |= mm->def_flags;
613
614         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
615                         vm_flags);
616         if (ret)
617                 goto out_unlock;
618         BUG_ON(prev != vma);
619
620         /* Move stack pages down in memory. */
621         if (stack_shift) {
622                 ret = shift_arg_pages(vma, stack_shift);
623                 if (ret) {
624                         up_write(&mm->mmap_sem);
625                         return ret;
626                 }
627         }
628
629 #ifdef CONFIG_STACK_GROWSUP
630         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
631 #else
632         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
633 #endif
634         ret = expand_stack(vma, stack_base);
635         if (ret)
636                 ret = -EFAULT;
637
638 out_unlock:
639         up_write(&mm->mmap_sem);
640         return 0;
641 }
642 EXPORT_SYMBOL(setup_arg_pages);
643
644 #endif /* CONFIG_MMU */
645
646 struct file *open_exec(const char *name)
647 {
648         struct nameidata nd;
649         int err;
650         struct file *file;
651
652         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
653         file = ERR_PTR(err);
654
655         if (!err) {
656                 struct inode *inode = nd.dentry->d_inode;
657                 file = ERR_PTR(-EACCES);
658                 if (S_ISREG(inode->i_mode)) {
659                         int err = vfs_permission(&nd, MAY_EXEC);
660                         file = ERR_PTR(err);
661                         if (!err) {
662                                 file = nameidata_to_filp(&nd, O_RDONLY);
663                                 if (!IS_ERR(file)) {
664                                         err = deny_write_access(file);
665                                         if (err) {
666                                                 fput(file);
667                                                 file = ERR_PTR(err);
668                                         }
669                                 }
670 out:
671                                 return file;
672                         }
673                 }
674                 release_open_intent(&nd);
675                 path_release(&nd);
676         }
677         goto out;
678 }
679
680 EXPORT_SYMBOL(open_exec);
681
682 int kernel_read(struct file *file, unsigned long offset,
683         char *addr, unsigned long count)
684 {
685         mm_segment_t old_fs;
686         loff_t pos = offset;
687         int result;
688
689         old_fs = get_fs();
690         set_fs(get_ds());
691         /* The cast to a user pointer is valid due to the set_fs() */
692         result = vfs_read(file, (void __user *)addr, count, &pos);
693         set_fs(old_fs);
694         return result;
695 }
696
697 EXPORT_SYMBOL(kernel_read);
698
699 static int exec_mmap(struct mm_struct *mm)
700 {
701         struct task_struct *tsk;
702         struct mm_struct * old_mm, *active_mm;
703
704         /* Notify parent that we're no longer interested in the old VM */
705         tsk = current;
706         old_mm = current->mm;
707         mm_release(tsk, old_mm);
708
709         if (old_mm) {
710                 /*
711                  * Make sure that if there is a core dump in progress
712                  * for the old mm, we get out and die instead of going
713                  * through with the exec.  We must hold mmap_sem around
714                  * checking core_waiters and changing tsk->mm.  The
715                  * core-inducing thread will increment core_waiters for
716                  * each thread whose ->mm == old_mm.
717                  */
718                 down_read(&old_mm->mmap_sem);
719                 if (unlikely(old_mm->core_waiters)) {
720                         up_read(&old_mm->mmap_sem);
721                         return -EINTR;
722                 }
723         }
724         task_lock(tsk);
725         active_mm = tsk->active_mm;
726         tsk->mm = mm;
727         tsk->active_mm = mm;
728         activate_mm(active_mm, mm);
729         task_unlock(tsk);
730         arch_pick_mmap_layout(mm);
731         if (old_mm) {
732                 up_read(&old_mm->mmap_sem);
733                 BUG_ON(active_mm != old_mm);
734                 mmput(old_mm);
735                 return 0;
736         }
737         mmdrop(active_mm);
738         return 0;
739 }
740
741 /*
742  * This function makes sure the current process has its own signal table,
743  * so that flush_signal_handlers can later reset the handlers without
744  * disturbing other processes.  (Other processes might share the signal
745  * table via the CLONE_SIGHAND option to clone().)
746  */
747 static int de_thread(struct task_struct *tsk)
748 {
749         struct signal_struct *sig = tsk->signal;
750         struct sighand_struct *oldsighand = tsk->sighand;
751         spinlock_t *lock = &oldsighand->siglock;
752         struct task_struct *leader = NULL;
753         int count;
754
755         if (thread_group_empty(tsk))
756                 goto no_thread_group;
757
758         /*
759          * Kill all other threads in the thread group.
760          * We must hold tasklist_lock to call zap_other_threads.
761          */
762         read_lock(&tasklist_lock);
763         spin_lock_irq(lock);
764         if (sig->flags & SIGNAL_GROUP_EXIT) {
765                 /*
766                  * Another group action in progress, just
767                  * return so that the signal is processed.
768                  */
769                 spin_unlock_irq(lock);
770                 read_unlock(&tasklist_lock);
771                 return -EAGAIN;
772         }
773
774         /*
775          * child_reaper ignores SIGKILL, change it now.
776          * Reparenting needs write_lock on tasklist_lock,
777          * so it is safe to do it under read_lock.
778          */
779         if (unlikely(tsk->group_leader == child_reaper(tsk)))
780                 tsk->nsproxy->pid_ns->child_reaper = tsk;
781
782         zap_other_threads(tsk);
783         read_unlock(&tasklist_lock);
784
785         /*
786          * Account for the thread group leader hanging around:
787          */
788         count = 1;
789         if (!thread_group_leader(tsk)) {
790                 count = 2;
791                 /*
792                  * The SIGALRM timer survives the exec, but needs to point
793                  * at us as the new group leader now.  We have a race with
794                  * a timer firing now getting the old leader, so we need to
795                  * synchronize with any firing (by calling del_timer_sync)
796                  * before we can safely let the old group leader die.
797                  */
798                 sig->tsk = tsk;
799                 spin_unlock_irq(lock);
800                 if (hrtimer_cancel(&sig->real_timer))
801                         hrtimer_restart(&sig->real_timer);
802                 spin_lock_irq(lock);
803         }
804
805         sig->notify_count = count;
806         sig->group_exit_task = tsk;
807         while (atomic_read(&sig->count) > count) {
808                 __set_current_state(TASK_UNINTERRUPTIBLE);
809                 spin_unlock_irq(lock);
810                 schedule();
811                 spin_lock_irq(lock);
812         }
813         spin_unlock_irq(lock);
814
815         /*
816          * At this point all other threads have exited, all we have to
817          * do is to wait for the thread group leader to become inactive,
818          * and to assume its PID:
819          */
820         if (!thread_group_leader(tsk)) {
821                 leader = tsk->group_leader;
822
823                 sig->notify_count = -1;
824                 for (;;) {
825                         write_lock_irq(&tasklist_lock);
826                         if (likely(leader->exit_state))
827                                 break;
828                         __set_current_state(TASK_UNINTERRUPTIBLE);
829                         write_unlock_irq(&tasklist_lock);
830                         schedule();
831                 }
832
833                 /*
834                  * The only record we have of the real-time age of a
835                  * process, regardless of execs it's done, is start_time.
836                  * All the past CPU time is accumulated in signal_struct
837                  * from sister threads now dead.  But in this non-leader
838                  * exec, nothing survives from the original leader thread,
839                  * whose birth marks the true age of this process now.
840                  * When we take on its identity by switching to its PID, we
841                  * also take its birthdate (always earlier than our own).
842                  */
843                 tsk->start_time = leader->start_time;
844
845                 BUG_ON(leader->tgid != tsk->tgid);
846                 BUG_ON(tsk->pid == tsk->tgid);
847                 /*
848                  * An exec() starts a new thread group with the
849                  * TGID of the previous thread group. Rehash the
850                  * two threads with a switched PID, and release
851                  * the former thread group leader:
852                  */
853
854                 /* Become a process group leader with the old leader's pid.
855                  * The old leader becomes a thread of the this thread group.
856                  * Note: The old leader also uses this pid until release_task
857                  *       is called.  Odd but simple and correct.
858                  */
859                 detach_pid(tsk, PIDTYPE_PID);
860                 tsk->pid = leader->pid;
861                 attach_pid(tsk, PIDTYPE_PID,  find_pid(tsk->pid));
862                 transfer_pid(leader, tsk, PIDTYPE_PGID);
863                 transfer_pid(leader, tsk, PIDTYPE_SID);
864                 list_replace_rcu(&leader->tasks, &tsk->tasks);
865
866                 tsk->group_leader = tsk;
867                 leader->group_leader = tsk;
868
869                 tsk->exit_signal = SIGCHLD;
870
871                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
872                 leader->exit_state = EXIT_DEAD;
873
874                 write_unlock_irq(&tasklist_lock);
875         }
876
877         sig->group_exit_task = NULL;
878         sig->notify_count = 0;
879         /*
880          * There may be one thread left which is just exiting,
881          * but it's safe to stop telling the group to kill themselves.
882          */
883         sig->flags = 0;
884
885 no_thread_group:
886         exit_itimers(sig);
887         if (leader)
888                 release_task(leader);
889
890         if (atomic_read(&oldsighand->count) != 1) {
891                 struct sighand_struct *newsighand;
892                 /*
893                  * This ->sighand is shared with the CLONE_SIGHAND
894                  * but not CLONE_THREAD task, switch to the new one.
895                  */
896                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
897                 if (!newsighand)
898                         return -ENOMEM;
899
900                 atomic_set(&newsighand->count, 1);
901                 memcpy(newsighand->action, oldsighand->action,
902                        sizeof(newsighand->action));
903
904                 write_lock_irq(&tasklist_lock);
905                 spin_lock(&oldsighand->siglock);
906                 rcu_assign_pointer(tsk->sighand, newsighand);
907                 spin_unlock(&oldsighand->siglock);
908                 write_unlock_irq(&tasklist_lock);
909
910                 __cleanup_sighand(oldsighand);
911         }
912
913         BUG_ON(!thread_group_leader(tsk));
914         return 0;
915 }
916
917 /*
918  * These functions flushes out all traces of the currently running executable
919  * so that a new one can be started
920  */
921 static void flush_old_files(struct files_struct * files)
922 {
923         long j = -1;
924         struct fdtable *fdt;
925
926         spin_lock(&files->file_lock);
927         for (;;) {
928                 unsigned long set, i;
929
930                 j++;
931                 i = j * __NFDBITS;
932                 fdt = files_fdtable(files);
933                 if (i >= fdt->max_fds)
934                         break;
935                 set = fdt->close_on_exec->fds_bits[j];
936                 if (!set)
937                         continue;
938                 fdt->close_on_exec->fds_bits[j] = 0;
939                 spin_unlock(&files->file_lock);
940                 for ( ; set ; i++,set >>= 1) {
941                         if (set & 1) {
942                                 sys_close(i);
943                         }
944                 }
945                 spin_lock(&files->file_lock);
946
947         }
948         spin_unlock(&files->file_lock);
949 }
950
951 void get_task_comm(char *buf, struct task_struct *tsk)
952 {
953         /* buf must be at least sizeof(tsk->comm) in size */
954         task_lock(tsk);
955         strncpy(buf, tsk->comm, sizeof(tsk->comm));
956         task_unlock(tsk);
957 }
958
959 void set_task_comm(struct task_struct *tsk, char *buf)
960 {
961         task_lock(tsk);
962         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
963         task_unlock(tsk);
964 }
965
966 int flush_old_exec(struct linux_binprm * bprm)
967 {
968         char * name;
969         int i, ch, retval;
970         struct files_struct *files;
971         char tcomm[sizeof(current->comm)];
972
973         /*
974          * Make sure we have a private signal table and that
975          * we are unassociated from the previous thread group.
976          */
977         retval = de_thread(current);
978         if (retval)
979                 goto out;
980
981         /*
982          * Make sure we have private file handles. Ask the
983          * fork helper to do the work for us and the exit
984          * helper to do the cleanup of the old one.
985          */
986         files = current->files;         /* refcounted so safe to hold */
987         retval = unshare_files();
988         if (retval)
989                 goto out;
990         /*
991          * Release all of the old mmap stuff
992          */
993         retval = exec_mmap(bprm->mm);
994         if (retval)
995                 goto mmap_failed;
996
997         bprm->mm = NULL;                /* We're using it now */
998
999         /* This is the point of no return */
1000         put_files_struct(files);
1001
1002         current->sas_ss_sp = current->sas_ss_size = 0;
1003
1004         if (current->euid == current->uid && current->egid == current->gid)
1005                 set_dumpable(current->mm, 1);
1006         else
1007                 set_dumpable(current->mm, suid_dumpable);
1008
1009         name = bprm->filename;
1010
1011         /* Copies the binary name from after last slash */
1012         for (i=0; (ch = *(name++)) != '\0';) {
1013                 if (ch == '/')
1014                         i = 0; /* overwrite what we wrote */
1015                 else
1016                         if (i < (sizeof(tcomm) - 1))
1017                                 tcomm[i++] = ch;
1018         }
1019         tcomm[i] = '\0';
1020         set_task_comm(current, tcomm);
1021
1022         current->flags &= ~PF_RANDOMIZE;
1023         flush_thread();
1024
1025         /* Set the new mm task size. We have to do that late because it may
1026          * depend on TIF_32BIT which is only updated in flush_thread() on
1027          * some architectures like powerpc
1028          */
1029         current->mm->task_size = TASK_SIZE;
1030
1031         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1032                 suid_keys(current);
1033                 set_dumpable(current->mm, suid_dumpable);
1034                 current->pdeath_signal = 0;
1035         } else if (file_permission(bprm->file, MAY_READ) ||
1036                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1037                 suid_keys(current);
1038                 set_dumpable(current->mm, suid_dumpable);
1039         }
1040
1041         /* An exec changes our domain. We are no longer part of the thread
1042            group */
1043
1044         current->self_exec_id++;
1045                         
1046         flush_signal_handlers(current, 0);
1047         flush_old_files(current->files);
1048
1049         return 0;
1050
1051 mmap_failed:
1052         reset_files_struct(current, files);
1053 out:
1054         return retval;
1055 }
1056
1057 EXPORT_SYMBOL(flush_old_exec);
1058
1059 /* 
1060  * Fill the binprm structure from the inode. 
1061  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1062  */
1063 int prepare_binprm(struct linux_binprm *bprm)
1064 {
1065         int mode;
1066         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1067         int retval;
1068
1069         mode = inode->i_mode;
1070         if (bprm->file->f_op == NULL)
1071                 return -EACCES;
1072
1073         bprm->e_uid = current->euid;
1074         bprm->e_gid = current->egid;
1075
1076         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1077                 /* Set-uid? */
1078                 if (mode & S_ISUID) {
1079                         current->personality &= ~PER_CLEAR_ON_SETID;
1080                         bprm->e_uid = inode->i_uid;
1081                 }
1082
1083                 /* Set-gid? */
1084                 /*
1085                  * If setgid is set but no group execute bit then this
1086                  * is a candidate for mandatory locking, not a setgid
1087                  * executable.
1088                  */
1089                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1090                         current->personality &= ~PER_CLEAR_ON_SETID;
1091                         bprm->e_gid = inode->i_gid;
1092                 }
1093         }
1094
1095         /* fill in binprm security blob */
1096         retval = security_bprm_set(bprm);
1097         if (retval)
1098                 return retval;
1099
1100         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1101         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1102 }
1103
1104 EXPORT_SYMBOL(prepare_binprm);
1105
1106 static int unsafe_exec(struct task_struct *p)
1107 {
1108         int unsafe = 0;
1109         if (p->ptrace & PT_PTRACED) {
1110                 if (p->ptrace & PT_PTRACE_CAP)
1111                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1112                 else
1113                         unsafe |= LSM_UNSAFE_PTRACE;
1114         }
1115         if (atomic_read(&p->fs->count) > 1 ||
1116             atomic_read(&p->files->count) > 1 ||
1117             atomic_read(&p->sighand->count) > 1)
1118                 unsafe |= LSM_UNSAFE_SHARE;
1119
1120         return unsafe;
1121 }
1122
1123 void compute_creds(struct linux_binprm *bprm)
1124 {
1125         int unsafe;
1126
1127         if (bprm->e_uid != current->uid) {
1128                 suid_keys(current);
1129                 current->pdeath_signal = 0;
1130         }
1131         exec_keys(current);
1132
1133         task_lock(current);
1134         unsafe = unsafe_exec(current);
1135         security_bprm_apply_creds(bprm, unsafe);
1136         task_unlock(current);
1137         security_bprm_post_apply_creds(bprm);
1138 }
1139 EXPORT_SYMBOL(compute_creds);
1140
1141 /*
1142  * Arguments are '\0' separated strings found at the location bprm->p
1143  * points to; chop off the first by relocating brpm->p to right after
1144  * the first '\0' encountered.
1145  */
1146 int remove_arg_zero(struct linux_binprm *bprm)
1147 {
1148         int ret = 0;
1149         unsigned long offset;
1150         char *kaddr;
1151         struct page *page;
1152
1153         if (!bprm->argc)
1154                 return 0;
1155
1156         do {
1157                 offset = bprm->p & ~PAGE_MASK;
1158                 page = get_arg_page(bprm, bprm->p, 0);
1159                 if (!page) {
1160                         ret = -EFAULT;
1161                         goto out;
1162                 }
1163                 kaddr = kmap_atomic(page, KM_USER0);
1164
1165                 for (; offset < PAGE_SIZE && kaddr[offset];
1166                                 offset++, bprm->p++)
1167                         ;
1168
1169                 kunmap_atomic(kaddr, KM_USER0);
1170                 put_arg_page(page);
1171
1172                 if (offset == PAGE_SIZE)
1173                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1174         } while (offset == PAGE_SIZE);
1175
1176         bprm->p++;
1177         bprm->argc--;
1178         ret = 0;
1179
1180 out:
1181         return ret;
1182 }
1183 EXPORT_SYMBOL(remove_arg_zero);
1184
1185 /*
1186  * cycle the list of binary formats handler, until one recognizes the image
1187  */
1188 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1189 {
1190         int try,retval;
1191         struct linux_binfmt *fmt;
1192 #ifdef __alpha__
1193         /* handle /sbin/loader.. */
1194         {
1195             struct exec * eh = (struct exec *) bprm->buf;
1196
1197             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1198                 (eh->fh.f_flags & 0x3000) == 0x3000)
1199             {
1200                 struct file * file;
1201                 unsigned long loader;
1202
1203                 allow_write_access(bprm->file);
1204                 fput(bprm->file);
1205                 bprm->file = NULL;
1206
1207                 loader = bprm->vma->vm_end - sizeof(void *);
1208
1209                 file = open_exec("/sbin/loader");
1210                 retval = PTR_ERR(file);
1211                 if (IS_ERR(file))
1212                         return retval;
1213
1214                 /* Remember if the application is TASO.  */
1215                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1216
1217                 bprm->file = file;
1218                 bprm->loader = loader;
1219                 retval = prepare_binprm(bprm);
1220                 if (retval<0)
1221                         return retval;
1222                 /* should call search_binary_handler recursively here,
1223                    but it does not matter */
1224             }
1225         }
1226 #endif
1227         retval = security_bprm_check(bprm);
1228         if (retval)
1229                 return retval;
1230
1231         /* kernel module loader fixup */
1232         /* so we don't try to load run modprobe in kernel space. */
1233         set_fs(USER_DS);
1234
1235         retval = audit_bprm(bprm);
1236         if (retval)
1237                 return retval;
1238
1239         retval = -ENOENT;
1240         for (try=0; try<2; try++) {
1241                 read_lock(&binfmt_lock);
1242                 list_for_each_entry(fmt, &formats, lh) {
1243                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1244                         if (!fn)
1245                                 continue;
1246                         if (!try_module_get(fmt->module))
1247                                 continue;
1248                         read_unlock(&binfmt_lock);
1249                         retval = fn(bprm, regs);
1250                         if (retval >= 0) {
1251                                 put_binfmt(fmt);
1252                                 allow_write_access(bprm->file);
1253                                 if (bprm->file)
1254                                         fput(bprm->file);
1255                                 bprm->file = NULL;
1256                                 current->did_exec = 1;
1257                                 proc_exec_connector(current);
1258                                 return retval;
1259                         }
1260                         read_lock(&binfmt_lock);
1261                         put_binfmt(fmt);
1262                         if (retval != -ENOEXEC || bprm->mm == NULL)
1263                                 break;
1264                         if (!bprm->file) {
1265                                 read_unlock(&binfmt_lock);
1266                                 return retval;
1267                         }
1268                 }
1269                 read_unlock(&binfmt_lock);
1270                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1271                         break;
1272 #ifdef CONFIG_KMOD
1273                 }else{
1274 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1275                         if (printable(bprm->buf[0]) &&
1276                             printable(bprm->buf[1]) &&
1277                             printable(bprm->buf[2]) &&
1278                             printable(bprm->buf[3]))
1279                                 break; /* -ENOEXEC */
1280                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1281 #endif
1282                 }
1283         }
1284         return retval;
1285 }
1286
1287 EXPORT_SYMBOL(search_binary_handler);
1288
1289 /*
1290  * sys_execve() executes a new program.
1291  */
1292 int do_execve(char * filename,
1293         char __user *__user *argv,
1294         char __user *__user *envp,
1295         struct pt_regs * regs)
1296 {
1297         struct linux_binprm *bprm;
1298         struct file *file;
1299         unsigned long env_p;
1300         int retval;
1301
1302         retval = -ENOMEM;
1303         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1304         if (!bprm)
1305                 goto out_ret;
1306
1307         file = open_exec(filename);
1308         retval = PTR_ERR(file);
1309         if (IS_ERR(file))
1310                 goto out_kfree;
1311
1312         sched_exec();
1313
1314         bprm->file = file;
1315         bprm->filename = filename;
1316         bprm->interp = filename;
1317
1318         retval = bprm_mm_init(bprm);
1319         if (retval)
1320                 goto out_file;
1321
1322         bprm->argc = count(argv, MAX_ARG_STRINGS);
1323         if ((retval = bprm->argc) < 0)
1324                 goto out_mm;
1325
1326         bprm->envc = count(envp, MAX_ARG_STRINGS);
1327         if ((retval = bprm->envc) < 0)
1328                 goto out_mm;
1329
1330         retval = security_bprm_alloc(bprm);
1331         if (retval)
1332                 goto out;
1333
1334         retval = prepare_binprm(bprm);
1335         if (retval < 0)
1336                 goto out;
1337
1338         retval = copy_strings_kernel(1, &bprm->filename, bprm);
1339         if (retval < 0)
1340                 goto out;
1341
1342         bprm->exec = bprm->p;
1343         retval = copy_strings(bprm->envc, envp, bprm);
1344         if (retval < 0)
1345                 goto out;
1346
1347         env_p = bprm->p;
1348         retval = copy_strings(bprm->argc, argv, bprm);
1349         if (retval < 0)
1350                 goto out;
1351         bprm->argv_len = env_p - bprm->p;
1352
1353         retval = search_binary_handler(bprm,regs);
1354         if (retval >= 0) {
1355                 /* execve success */
1356                 free_arg_pages(bprm);
1357                 security_bprm_free(bprm);
1358                 acct_update_integrals(current);
1359                 kfree(bprm);
1360                 return retval;
1361         }
1362
1363 out:
1364         free_arg_pages(bprm);
1365         if (bprm->security)
1366                 security_bprm_free(bprm);
1367
1368 out_mm:
1369         if (bprm->mm)
1370                 mmput (bprm->mm);
1371
1372 out_file:
1373         if (bprm->file) {
1374                 allow_write_access(bprm->file);
1375                 fput(bprm->file);
1376         }
1377 out_kfree:
1378         kfree(bprm);
1379
1380 out_ret:
1381         return retval;
1382 }
1383
1384 int set_binfmt(struct linux_binfmt *new)
1385 {
1386         struct linux_binfmt *old = current->binfmt;
1387
1388         if (new) {
1389                 if (!try_module_get(new->module))
1390                         return -1;
1391         }
1392         current->binfmt = new;
1393         if (old)
1394                 module_put(old->module);
1395         return 0;
1396 }
1397
1398 EXPORT_SYMBOL(set_binfmt);
1399
1400 /* format_corename will inspect the pattern parameter, and output a
1401  * name into corename, which must have space for at least
1402  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1403  */
1404 static int format_corename(char *corename, const char *pattern, long signr)
1405 {
1406         const char *pat_ptr = pattern;
1407         char *out_ptr = corename;
1408         char *const out_end = corename + CORENAME_MAX_SIZE;
1409         int rc;
1410         int pid_in_pattern = 0;
1411         int ispipe = 0;
1412
1413         if (*pattern == '|')
1414                 ispipe = 1;
1415
1416         /* Repeat as long as we have more pattern to process and more output
1417            space */
1418         while (*pat_ptr) {
1419                 if (*pat_ptr != '%') {
1420                         if (out_ptr == out_end)
1421                                 goto out;
1422                         *out_ptr++ = *pat_ptr++;
1423                 } else {
1424                         switch (*++pat_ptr) {
1425                         case 0:
1426                                 goto out;
1427                         /* Double percent, output one percent */
1428                         case '%':
1429                                 if (out_ptr == out_end)
1430                                         goto out;
1431                                 *out_ptr++ = '%';
1432                                 break;
1433                         /* pid */
1434                         case 'p':
1435                                 pid_in_pattern = 1;
1436                                 rc = snprintf(out_ptr, out_end - out_ptr,
1437                                               "%d", current->tgid);
1438                                 if (rc > out_end - out_ptr)
1439                                         goto out;
1440                                 out_ptr += rc;
1441                                 break;
1442                         /* uid */
1443                         case 'u':
1444                                 rc = snprintf(out_ptr, out_end - out_ptr,
1445                                               "%d", current->uid);
1446                                 if (rc > out_end - out_ptr)
1447                                         goto out;
1448                                 out_ptr += rc;
1449                                 break;
1450                         /* gid */
1451                         case 'g':
1452                                 rc = snprintf(out_ptr, out_end - out_ptr,
1453                                               "%d", current->gid);
1454                                 if (rc > out_end - out_ptr)
1455                                         goto out;
1456                                 out_ptr += rc;
1457                                 break;
1458                         /* signal that caused the coredump */
1459                         case 's':
1460                                 rc = snprintf(out_ptr, out_end - out_ptr,
1461                                               "%ld", signr);
1462                                 if (rc > out_end - out_ptr)
1463                                         goto out;
1464                                 out_ptr += rc;
1465                                 break;
1466                         /* UNIX time of coredump */
1467                         case 't': {
1468                                 struct timeval tv;
1469                                 do_gettimeofday(&tv);
1470                                 rc = snprintf(out_ptr, out_end - out_ptr,
1471                                               "%lu", tv.tv_sec);
1472                                 if (rc > out_end - out_ptr)
1473                                         goto out;
1474                                 out_ptr += rc;
1475                                 break;
1476                         }
1477                         /* hostname */
1478                         case 'h':
1479                                 down_read(&uts_sem);
1480                                 rc = snprintf(out_ptr, out_end - out_ptr,
1481                                               "%s", utsname()->nodename);
1482                                 up_read(&uts_sem);
1483                                 if (rc > out_end - out_ptr)
1484                                         goto out;
1485                                 out_ptr += rc;
1486                                 break;
1487                         /* executable */
1488                         case 'e':
1489                                 rc = snprintf(out_ptr, out_end - out_ptr,
1490                                               "%s", current->comm);
1491                                 if (rc > out_end - out_ptr)
1492                                         goto out;
1493                                 out_ptr += rc;
1494                                 break;
1495                         /* core limit size */
1496                         case 'c':
1497                                 rc = snprintf(out_ptr, out_end - out_ptr,
1498                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1499                                 if (rc > out_end - out_ptr)
1500                                         goto out;
1501                                 out_ptr += rc;
1502                                 break;
1503                         default:
1504                                 break;
1505                         }
1506                         ++pat_ptr;
1507                 }
1508         }
1509         /* Backward compatibility with core_uses_pid:
1510          *
1511          * If core_pattern does not include a %p (as is the default)
1512          * and core_uses_pid is set, then .%pid will be appended to
1513          * the filename. Do not do this for piped commands. */
1514         if (!ispipe && !pid_in_pattern
1515             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1516                 rc = snprintf(out_ptr, out_end - out_ptr,
1517                               ".%d", current->tgid);
1518                 if (rc > out_end - out_ptr)
1519                         goto out;
1520                 out_ptr += rc;
1521         }
1522 out:
1523         *out_ptr = 0;
1524         return ispipe;
1525 }
1526
1527 static void zap_process(struct task_struct *start)
1528 {
1529         struct task_struct *t;
1530
1531         start->signal->flags = SIGNAL_GROUP_EXIT;
1532         start->signal->group_stop_count = 0;
1533
1534         t = start;
1535         do {
1536                 if (t != current && t->mm) {
1537                         t->mm->core_waiters++;
1538                         sigaddset(&t->pending.signal, SIGKILL);
1539                         signal_wake_up(t, 1);
1540                 }
1541         } while ((t = next_thread(t)) != start);
1542 }
1543
1544 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1545                                 int exit_code)
1546 {
1547         struct task_struct *g, *p;
1548         unsigned long flags;
1549         int err = -EAGAIN;
1550
1551         spin_lock_irq(&tsk->sighand->siglock);
1552         if (!(tsk->signal->flags & SIGNAL_GROUP_EXIT)) {
1553                 tsk->signal->group_exit_code = exit_code;
1554                 zap_process(tsk);
1555                 err = 0;
1556         }
1557         spin_unlock_irq(&tsk->sighand->siglock);
1558         if (err)
1559                 return err;
1560
1561         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1562                 goto done;
1563
1564         rcu_read_lock();
1565         for_each_process(g) {
1566                 if (g == tsk->group_leader)
1567                         continue;
1568
1569                 p = g;
1570                 do {
1571                         if (p->mm) {
1572                                 if (p->mm == mm) {
1573                                         /*
1574                                          * p->sighand can't disappear, but
1575                                          * may be changed by de_thread()
1576                                          */
1577                                         lock_task_sighand(p, &flags);
1578                                         zap_process(p);
1579                                         unlock_task_sighand(p, &flags);
1580                                 }
1581                                 break;
1582                         }
1583                 } while ((p = next_thread(p)) != g);
1584         }
1585         rcu_read_unlock();
1586 done:
1587         return mm->core_waiters;
1588 }
1589
1590 static int coredump_wait(int exit_code)
1591 {
1592         struct task_struct *tsk = current;
1593         struct mm_struct *mm = tsk->mm;
1594         struct completion startup_done;
1595         struct completion *vfork_done;
1596         int core_waiters;
1597
1598         init_completion(&mm->core_done);
1599         init_completion(&startup_done);
1600         mm->core_startup_done = &startup_done;
1601
1602         core_waiters = zap_threads(tsk, mm, exit_code);
1603         up_write(&mm->mmap_sem);
1604
1605         if (unlikely(core_waiters < 0))
1606                 goto fail;
1607
1608         /*
1609          * Make sure nobody is waiting for us to release the VM,
1610          * otherwise we can deadlock when we wait on each other
1611          */
1612         vfork_done = tsk->vfork_done;
1613         if (vfork_done) {
1614                 tsk->vfork_done = NULL;
1615                 complete(vfork_done);
1616         }
1617
1618         if (core_waiters)
1619                 wait_for_completion(&startup_done);
1620 fail:
1621         BUG_ON(mm->core_waiters);
1622         return core_waiters;
1623 }
1624
1625 /*
1626  * set_dumpable converts traditional three-value dumpable to two flags and
1627  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1628  * these bits are not changed atomically.  So get_dumpable can observe the
1629  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1630  * return either old dumpable or new one by paying attention to the order of
1631  * modifying the bits.
1632  *
1633  * dumpable |   mm->flags (binary)
1634  * old  new | initial interim  final
1635  * ---------+-----------------------
1636  *  0    1  |   00      01      01
1637  *  0    2  |   00      10(*)   11
1638  *  1    0  |   01      00      00
1639  *  1    2  |   01      11      11
1640  *  2    0  |   11      10(*)   00
1641  *  2    1  |   11      11      01
1642  *
1643  * (*) get_dumpable regards interim value of 10 as 11.
1644  */
1645 void set_dumpable(struct mm_struct *mm, int value)
1646 {
1647         switch (value) {
1648         case 0:
1649                 clear_bit(MMF_DUMPABLE, &mm->flags);
1650                 smp_wmb();
1651                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1652                 break;
1653         case 1:
1654                 set_bit(MMF_DUMPABLE, &mm->flags);
1655                 smp_wmb();
1656                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1657                 break;
1658         case 2:
1659                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1660                 smp_wmb();
1661                 set_bit(MMF_DUMPABLE, &mm->flags);
1662                 break;
1663         }
1664 }
1665 EXPORT_SYMBOL_GPL(set_dumpable);
1666
1667 int get_dumpable(struct mm_struct *mm)
1668 {
1669         int ret;
1670
1671         ret = mm->flags & 0x3;
1672         return (ret >= 2) ? 2 : ret;
1673 }
1674
1675 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1676 {
1677         char corename[CORENAME_MAX_SIZE + 1];
1678         struct mm_struct *mm = current->mm;
1679         struct linux_binfmt * binfmt;
1680         struct inode * inode;
1681         struct file * file;
1682         int retval = 0;
1683         int fsuid = current->fsuid;
1684         int flag = 0;
1685         int ispipe = 0;
1686         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1687         char **helper_argv = NULL;
1688         int helper_argc = 0;
1689         char *delimit;
1690
1691         audit_core_dumps(signr);
1692
1693         binfmt = current->binfmt;
1694         if (!binfmt || !binfmt->core_dump)
1695                 goto fail;
1696         down_write(&mm->mmap_sem);
1697         if (!get_dumpable(mm)) {
1698                 up_write(&mm->mmap_sem);
1699                 goto fail;
1700         }
1701
1702         /*
1703          *      We cannot trust fsuid as being the "true" uid of the
1704          *      process nor do we know its entire history. We only know it
1705          *      was tainted so we dump it as root in mode 2.
1706          */
1707         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1708                 flag = O_EXCL;          /* Stop rewrite attacks */
1709                 current->fsuid = 0;     /* Dump root private */
1710         }
1711         set_dumpable(mm, 0);
1712
1713         retval = coredump_wait(exit_code);
1714         if (retval < 0)
1715                 goto fail;
1716
1717         /*
1718          * Clear any false indication of pending signals that might
1719          * be seen by the filesystem code called to write the core file.
1720          */
1721         clear_thread_flag(TIF_SIGPENDING);
1722
1723         /*
1724          * lock_kernel() because format_corename() is controlled by sysctl, which
1725          * uses lock_kernel()
1726          */
1727         lock_kernel();
1728         ispipe = format_corename(corename, core_pattern, signr);
1729         unlock_kernel();
1730         /*
1731          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1732          * to a pipe.  Since we're not writing directly to the filesystem
1733          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1734          * created unless the pipe reader choses to write out the core file
1735          * at which point file size limits and permissions will be imposed
1736          * as it does with any other process
1737          */
1738         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1739                 goto fail_unlock;
1740
1741         if (ispipe) {
1742                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1743                 /* Terminate the string before the first option */
1744                 delimit = strchr(corename, ' ');
1745                 if (delimit)
1746                         *delimit = '\0';
1747                 delimit = strrchr(helper_argv[0], '/');
1748                 if (delimit)
1749                         delimit++;
1750                 else
1751                         delimit = helper_argv[0];
1752                 if (!strcmp(delimit, current->comm)) {
1753                         printk(KERN_NOTICE "Recursive core dump detected, "
1754                                         "aborting\n");
1755                         goto fail_unlock;
1756                 }
1757
1758                 core_limit = RLIM_INFINITY;
1759
1760                 /* SIGPIPE can happen, but it's just never processed */
1761                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1762                                 &file)) {
1763                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1764                                corename);
1765                         goto fail_unlock;
1766                 }
1767         } else
1768                 file = filp_open(corename,
1769                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1770                                  0600);
1771         if (IS_ERR(file))
1772                 goto fail_unlock;
1773         inode = file->f_path.dentry->d_inode;
1774         if (inode->i_nlink > 1)
1775                 goto close_fail;        /* multiple links - don't dump */
1776         if (!ispipe && d_unhashed(file->f_path.dentry))
1777                 goto close_fail;
1778
1779         /* AK: actually i see no reason to not allow this for named pipes etc.,
1780            but keep the previous behaviour for now. */
1781         if (!ispipe && !S_ISREG(inode->i_mode))
1782                 goto close_fail;
1783         if (!file->f_op)
1784                 goto close_fail;
1785         if (!file->f_op->write)
1786                 goto close_fail;
1787         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1788                 goto close_fail;
1789
1790         retval = binfmt->core_dump(signr, regs, file, core_limit);
1791
1792         if (retval)
1793                 current->signal->group_exit_code |= 0x80;
1794 close_fail:
1795         filp_close(file, NULL);
1796 fail_unlock:
1797         if (helper_argv)
1798                 argv_free(helper_argv);
1799
1800         current->fsuid = fsuid;
1801         complete_all(&mm->core_done);
1802 fail:
1803         return retval;
1804 }