Embed a struct path into struct nameidata instead of nd->{dentry,mnt}
[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/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 /* The maximal length of core_pattern is also specified in sysctl.c */
68
69 static LIST_HEAD(formats);
70 static DEFINE_RWLOCK(binfmt_lock);
71
72 int register_binfmt(struct linux_binfmt * fmt)
73 {
74         if (!fmt)
75                 return -EINVAL;
76         write_lock(&binfmt_lock);
77         list_add(&fmt->lh, &formats);
78         write_unlock(&binfmt_lock);
79         return 0;       
80 }
81
82 EXPORT_SYMBOL(register_binfmt);
83
84 void unregister_binfmt(struct linux_binfmt * fmt)
85 {
86         write_lock(&binfmt_lock);
87         list_del(&fmt->lh);
88         write_unlock(&binfmt_lock);
89 }
90
91 EXPORT_SYMBOL(unregister_binfmt);
92
93 static inline void put_binfmt(struct linux_binfmt * fmt)
94 {
95         module_put(fmt->module);
96 }
97
98 /*
99  * Note that a shared library must be both readable and executable due to
100  * security reasons.
101  *
102  * Also note that we take the address to load from from the file itself.
103  */
104 asmlinkage long sys_uselib(const char __user * library)
105 {
106         struct file * file;
107         struct nameidata nd;
108         int error;
109
110         error = __user_path_lookup_open(library, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
111         if (error)
112                 goto out;
113
114         error = -EINVAL;
115         if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
116                 goto exit;
117
118         error = vfs_permission(&nd, MAY_READ | MAY_EXEC);
119         if (error)
120                 goto exit;
121
122         file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
123         error = PTR_ERR(file);
124         if (IS_ERR(file))
125                 goto out;
126
127         error = -ENOEXEC;
128         if(file->f_op) {
129                 struct linux_binfmt * fmt;
130
131                 read_lock(&binfmt_lock);
132                 list_for_each_entry(fmt, &formats, lh) {
133                         if (!fmt->load_shlib)
134                                 continue;
135                         if (!try_module_get(fmt->module))
136                                 continue;
137                         read_unlock(&binfmt_lock);
138                         error = fmt->load_shlib(file);
139                         read_lock(&binfmt_lock);
140                         put_binfmt(fmt);
141                         if (error != -ENOEXEC)
142                                 break;
143                 }
144                 read_unlock(&binfmt_lock);
145         }
146         fput(file);
147 out:
148         return error;
149 exit:
150         release_open_intent(&nd);
151         path_release(&nd);
152         goto out;
153 }
154
155 #ifdef CONFIG_MMU
156
157 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
158                 int write)
159 {
160         struct page *page;
161         int ret;
162
163 #ifdef CONFIG_STACK_GROWSUP
164         if (write) {
165                 ret = expand_stack_downwards(bprm->vma, pos);
166                 if (ret < 0)
167                         return NULL;
168         }
169 #endif
170         ret = get_user_pages(current, bprm->mm, pos,
171                         1, write, 1, &page, NULL);
172         if (ret <= 0)
173                 return NULL;
174
175         if (write) {
176                 struct rlimit *rlim = current->signal->rlim;
177                 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
178
179                 /*
180                  * Limit to 1/4-th the stack size for the argv+env strings.
181                  * This ensures that:
182                  *  - the remaining binfmt code will not run out of stack space,
183                  *  - the program will have a reasonable amount of stack left
184                  *    to work from.
185                  */
186                 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
187                         put_page(page);
188                         return NULL;
189                 }
190         }
191
192         return page;
193 }
194
195 static void put_arg_page(struct page *page)
196 {
197         put_page(page);
198 }
199
200 static void free_arg_page(struct linux_binprm *bprm, int i)
201 {
202 }
203
204 static void free_arg_pages(struct linux_binprm *bprm)
205 {
206 }
207
208 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
209                 struct page *page)
210 {
211         flush_cache_page(bprm->vma, pos, page_to_pfn(page));
212 }
213
214 static int __bprm_mm_init(struct linux_binprm *bprm)
215 {
216         int err = -ENOMEM;
217         struct vm_area_struct *vma = NULL;
218         struct mm_struct *mm = bprm->mm;
219
220         bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
221         if (!vma)
222                 goto err;
223
224         down_write(&mm->mmap_sem);
225         vma->vm_mm = mm;
226
227         /*
228          * Place the stack at the largest stack address the architecture
229          * supports. Later, we'll move this to an appropriate place. We don't
230          * use STACK_TOP because that can depend on attributes which aren't
231          * configured yet.
232          */
233         vma->vm_end = STACK_TOP_MAX;
234         vma->vm_start = vma->vm_end - PAGE_SIZE;
235
236         vma->vm_flags = VM_STACK_FLAGS;
237         vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
238         err = insert_vm_struct(mm, vma);
239         if (err) {
240                 up_write(&mm->mmap_sem);
241                 goto err;
242         }
243
244         mm->stack_vm = mm->total_vm = 1;
245         up_write(&mm->mmap_sem);
246
247         bprm->p = vma->vm_end - sizeof(void *);
248
249         return 0;
250
251 err:
252         if (vma) {
253                 bprm->vma = NULL;
254                 kmem_cache_free(vm_area_cachep, vma);
255         }
256
257         return err;
258 }
259
260 static bool valid_arg_len(struct linux_binprm *bprm, long len)
261 {
262         return len <= MAX_ARG_STRLEN;
263 }
264
265 #else
266
267 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
268                 int write)
269 {
270         struct page *page;
271
272         page = bprm->page[pos / PAGE_SIZE];
273         if (!page && write) {
274                 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
275                 if (!page)
276                         return NULL;
277                 bprm->page[pos / PAGE_SIZE] = page;
278         }
279
280         return page;
281 }
282
283 static void put_arg_page(struct page *page)
284 {
285 }
286
287 static void free_arg_page(struct linux_binprm *bprm, int i)
288 {
289         if (bprm->page[i]) {
290                 __free_page(bprm->page[i]);
291                 bprm->page[i] = NULL;
292         }
293 }
294
295 static void free_arg_pages(struct linux_binprm *bprm)
296 {
297         int i;
298
299         for (i = 0; i < MAX_ARG_PAGES; i++)
300                 free_arg_page(bprm, i);
301 }
302
303 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
304                 struct page *page)
305 {
306 }
307
308 static int __bprm_mm_init(struct linux_binprm *bprm)
309 {
310         bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
311         return 0;
312 }
313
314 static bool valid_arg_len(struct linux_binprm *bprm, long len)
315 {
316         return len <= bprm->p;
317 }
318
319 #endif /* CONFIG_MMU */
320
321 /*
322  * Create a new mm_struct and populate it with a temporary stack
323  * vm_area_struct.  We don't have enough context at this point to set the stack
324  * flags, permissions, and offset, so we use temporary values.  We'll update
325  * them later in setup_arg_pages().
326  */
327 int bprm_mm_init(struct linux_binprm *bprm)
328 {
329         int err;
330         struct mm_struct *mm = NULL;
331
332         bprm->mm = mm = mm_alloc();
333         err = -ENOMEM;
334         if (!mm)
335                 goto err;
336
337         err = init_new_context(current, mm);
338         if (err)
339                 goto err;
340
341         err = __bprm_mm_init(bprm);
342         if (err)
343                 goto err;
344
345         return 0;
346
347 err:
348         if (mm) {
349                 bprm->mm = NULL;
350                 mmdrop(mm);
351         }
352
353         return err;
354 }
355
356 /*
357  * count() counts the number of strings in array ARGV.
358  */
359 static int count(char __user * __user * argv, int max)
360 {
361         int i = 0;
362
363         if (argv != NULL) {
364                 for (;;) {
365                         char __user * p;
366
367                         if (get_user(p, argv))
368                                 return -EFAULT;
369                         if (!p)
370                                 break;
371                         argv++;
372                         if(++i > max)
373                                 return -E2BIG;
374                         cond_resched();
375                 }
376         }
377         return i;
378 }
379
380 /*
381  * 'copy_strings()' copies argument/environment strings from the old
382  * processes's memory to the new process's stack.  The call to get_user_pages()
383  * ensures the destination page is created and not swapped out.
384  */
385 static int copy_strings(int argc, char __user * __user * argv,
386                         struct linux_binprm *bprm)
387 {
388         struct page *kmapped_page = NULL;
389         char *kaddr = NULL;
390         unsigned long kpos = 0;
391         int ret;
392
393         while (argc-- > 0) {
394                 char __user *str;
395                 int len;
396                 unsigned long pos;
397
398                 if (get_user(str, argv+argc) ||
399                                 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
400                         ret = -EFAULT;
401                         goto out;
402                 }
403
404                 if (!valid_arg_len(bprm, len)) {
405                         ret = -E2BIG;
406                         goto out;
407                 }
408
409                 /* We're going to work our way backwords. */
410                 pos = bprm->p;
411                 str += len;
412                 bprm->p -= len;
413
414                 while (len > 0) {
415                         int offset, bytes_to_copy;
416
417                         offset = pos % PAGE_SIZE;
418                         if (offset == 0)
419                                 offset = PAGE_SIZE;
420
421                         bytes_to_copy = offset;
422                         if (bytes_to_copy > len)
423                                 bytes_to_copy = len;
424
425                         offset -= bytes_to_copy;
426                         pos -= bytes_to_copy;
427                         str -= bytes_to_copy;
428                         len -= bytes_to_copy;
429
430                         if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
431                                 struct page *page;
432
433                                 page = get_arg_page(bprm, pos, 1);
434                                 if (!page) {
435                                         ret = -E2BIG;
436                                         goto out;
437                                 }
438
439                                 if (kmapped_page) {
440                                         flush_kernel_dcache_page(kmapped_page);
441                                         kunmap(kmapped_page);
442                                         put_arg_page(kmapped_page);
443                                 }
444                                 kmapped_page = page;
445                                 kaddr = kmap(kmapped_page);
446                                 kpos = pos & PAGE_MASK;
447                                 flush_arg_page(bprm, kpos, kmapped_page);
448                         }
449                         if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
450                                 ret = -EFAULT;
451                                 goto out;
452                         }
453                 }
454         }
455         ret = 0;
456 out:
457         if (kmapped_page) {
458                 flush_kernel_dcache_page(kmapped_page);
459                 kunmap(kmapped_page);
460                 put_arg_page(kmapped_page);
461         }
462         return ret;
463 }
464
465 /*
466  * Like copy_strings, but get argv and its values from kernel memory.
467  */
468 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
469 {
470         int r;
471         mm_segment_t oldfs = get_fs();
472         set_fs(KERNEL_DS);
473         r = copy_strings(argc, (char __user * __user *)argv, bprm);
474         set_fs(oldfs);
475         return r;
476 }
477 EXPORT_SYMBOL(copy_strings_kernel);
478
479 #ifdef CONFIG_MMU
480
481 /*
482  * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX.  Once
483  * the binfmt code determines where the new stack should reside, we shift it to
484  * its final location.  The process proceeds as follows:
485  *
486  * 1) Use shift to calculate the new vma endpoints.
487  * 2) Extend vma to cover both the old and new ranges.  This ensures the
488  *    arguments passed to subsequent functions are consistent.
489  * 3) Move vma's page tables to the new range.
490  * 4) Free up any cleared pgd range.
491  * 5) Shrink the vma to cover only the new range.
492  */
493 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
494 {
495         struct mm_struct *mm = vma->vm_mm;
496         unsigned long old_start = vma->vm_start;
497         unsigned long old_end = vma->vm_end;
498         unsigned long length = old_end - old_start;
499         unsigned long new_start = old_start - shift;
500         unsigned long new_end = old_end - shift;
501         struct mmu_gather *tlb;
502
503         BUG_ON(new_start > new_end);
504
505         /*
506          * ensure there are no vmas between where we want to go
507          * and where we are
508          */
509         if (vma != find_vma(mm, new_start))
510                 return -EFAULT;
511
512         /*
513          * cover the whole range: [new_start, old_end)
514          */
515         vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
516
517         /*
518          * move the page tables downwards, on failure we rely on
519          * process cleanup to remove whatever mess we made.
520          */
521         if (length != move_page_tables(vma, old_start,
522                                        vma, new_start, length))
523                 return -ENOMEM;
524
525         lru_add_drain();
526         tlb = tlb_gather_mmu(mm, 0);
527         if (new_end > old_start) {
528                 /*
529                  * when the old and new regions overlap clear from new_end.
530                  */
531                 free_pgd_range(&tlb, new_end, old_end, new_end,
532                         vma->vm_next ? vma->vm_next->vm_start : 0);
533         } else {
534                 /*
535                  * otherwise, clean from old_start; this is done to not touch
536                  * the address space in [new_end, old_start) some architectures
537                  * have constraints on va-space that make this illegal (IA64) -
538                  * for the others its just a little faster.
539                  */
540                 free_pgd_range(&tlb, old_start, old_end, new_end,
541                         vma->vm_next ? vma->vm_next->vm_start : 0);
542         }
543         tlb_finish_mmu(tlb, new_end, old_end);
544
545         /*
546          * shrink the vma to just the new range.
547          */
548         vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
549
550         return 0;
551 }
552
553 #define EXTRA_STACK_VM_PAGES    20      /* random */
554
555 /*
556  * Finalizes the stack vm_area_struct. The flags and permissions are updated,
557  * the stack is optionally relocated, and some extra space is added.
558  */
559 int setup_arg_pages(struct linux_binprm *bprm,
560                     unsigned long stack_top,
561                     int executable_stack)
562 {
563         unsigned long ret;
564         unsigned long stack_shift;
565         struct mm_struct *mm = current->mm;
566         struct vm_area_struct *vma = bprm->vma;
567         struct vm_area_struct *prev = NULL;
568         unsigned long vm_flags;
569         unsigned long stack_base;
570
571 #ifdef CONFIG_STACK_GROWSUP
572         /* Limit stack size to 1GB */
573         stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
574         if (stack_base > (1 << 30))
575                 stack_base = 1 << 30;
576
577         /* Make sure we didn't let the argument array grow too large. */
578         if (vma->vm_end - vma->vm_start > stack_base)
579                 return -ENOMEM;
580
581         stack_base = PAGE_ALIGN(stack_top - stack_base);
582
583         stack_shift = vma->vm_start - stack_base;
584         mm->arg_start = bprm->p - stack_shift;
585         bprm->p = vma->vm_end - stack_shift;
586 #else
587         stack_top = arch_align_stack(stack_top);
588         stack_top = PAGE_ALIGN(stack_top);
589         stack_shift = vma->vm_end - stack_top;
590
591         bprm->p -= stack_shift;
592         mm->arg_start = bprm->p;
593 #endif
594
595         if (bprm->loader)
596                 bprm->loader -= stack_shift;
597         bprm->exec -= stack_shift;
598
599         down_write(&mm->mmap_sem);
600         vm_flags = vma->vm_flags;
601
602         /*
603          * Adjust stack execute permissions; explicitly enable for
604          * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
605          * (arch default) otherwise.
606          */
607         if (unlikely(executable_stack == EXSTACK_ENABLE_X))
608                 vm_flags |= VM_EXEC;
609         else if (executable_stack == EXSTACK_DISABLE_X)
610                 vm_flags &= ~VM_EXEC;
611         vm_flags |= mm->def_flags;
612
613         ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
614                         vm_flags);
615         if (ret)
616                 goto out_unlock;
617         BUG_ON(prev != vma);
618
619         /* Move stack pages down in memory. */
620         if (stack_shift) {
621                 ret = shift_arg_pages(vma, stack_shift);
622                 if (ret) {
623                         up_write(&mm->mmap_sem);
624                         return ret;
625                 }
626         }
627
628 #ifdef CONFIG_STACK_GROWSUP
629         stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
630 #else
631         stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
632 #endif
633         ret = expand_stack(vma, stack_base);
634         if (ret)
635                 ret = -EFAULT;
636
637 out_unlock:
638         up_write(&mm->mmap_sem);
639         return 0;
640 }
641 EXPORT_SYMBOL(setup_arg_pages);
642
643 #endif /* CONFIG_MMU */
644
645 struct file *open_exec(const char *name)
646 {
647         struct nameidata nd;
648         int err;
649         struct file *file;
650
651         err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd, FMODE_READ|FMODE_EXEC);
652         file = ERR_PTR(err);
653
654         if (!err) {
655                 struct inode *inode = nd.path.dentry->d_inode;
656                 file = ERR_PTR(-EACCES);
657                 if (S_ISREG(inode->i_mode)) {
658                         int err = vfs_permission(&nd, MAY_EXEC);
659                         file = ERR_PTR(err);
660                         if (!err) {
661                                 file = nameidata_to_filp(&nd,
662                                                         O_RDONLY|O_LARGEFILE);
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 (signal_group_exit(sig)) {
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 == task_child_reaper(tsk)))
780                 task_active_pid_ns(tsk)->child_reaper = tsk;
781
782         sig->group_exit_task = tsk;
783         zap_other_threads(tsk);
784         read_unlock(&tasklist_lock);
785
786         /* Account for the thread group leader hanging around: */
787         count = thread_group_leader(tsk) ? 1 : 2;
788         sig->notify_count = count;
789         while (atomic_read(&sig->count) > count) {
790                 __set_current_state(TASK_UNINTERRUPTIBLE);
791                 spin_unlock_irq(lock);
792                 schedule();
793                 spin_lock_irq(lock);
794         }
795         spin_unlock_irq(lock);
796
797         /*
798          * At this point all other threads have exited, all we have to
799          * do is to wait for the thread group leader to become inactive,
800          * and to assume its PID:
801          */
802         if (!thread_group_leader(tsk)) {
803                 leader = tsk->group_leader;
804
805                 sig->notify_count = -1;
806                 for (;;) {
807                         write_lock_irq(&tasklist_lock);
808                         if (likely(leader->exit_state))
809                                 break;
810                         __set_current_state(TASK_UNINTERRUPTIBLE);
811                         write_unlock_irq(&tasklist_lock);
812                         schedule();
813                 }
814
815                 /*
816                  * The only record we have of the real-time age of a
817                  * process, regardless of execs it's done, is start_time.
818                  * All the past CPU time is accumulated in signal_struct
819                  * from sister threads now dead.  But in this non-leader
820                  * exec, nothing survives from the original leader thread,
821                  * whose birth marks the true age of this process now.
822                  * When we take on its identity by switching to its PID, we
823                  * also take its birthdate (always earlier than our own).
824                  */
825                 tsk->start_time = leader->start_time;
826
827                 BUG_ON(!same_thread_group(leader, tsk));
828                 BUG_ON(has_group_leader_pid(tsk));
829                 /*
830                  * An exec() starts a new thread group with the
831                  * TGID of the previous thread group. Rehash the
832                  * two threads with a switched PID, and release
833                  * the former thread group leader:
834                  */
835
836                 /* Become a process group leader with the old leader's pid.
837                  * The old leader becomes a thread of the this thread group.
838                  * Note: The old leader also uses this pid until release_task
839                  *       is called.  Odd but simple and correct.
840                  */
841                 detach_pid(tsk, PIDTYPE_PID);
842                 tsk->pid = leader->pid;
843                 attach_pid(tsk, PIDTYPE_PID,  task_pid(leader));
844                 transfer_pid(leader, tsk, PIDTYPE_PGID);
845                 transfer_pid(leader, tsk, PIDTYPE_SID);
846                 list_replace_rcu(&leader->tasks, &tsk->tasks);
847
848                 tsk->group_leader = tsk;
849                 leader->group_leader = tsk;
850
851                 tsk->exit_signal = SIGCHLD;
852
853                 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
854                 leader->exit_state = EXIT_DEAD;
855
856                 write_unlock_irq(&tasklist_lock);
857         }
858
859         sig->group_exit_task = NULL;
860         sig->notify_count = 0;
861
862 no_thread_group:
863         exit_itimers(sig);
864         if (leader)
865                 release_task(leader);
866
867         if (atomic_read(&oldsighand->count) != 1) {
868                 struct sighand_struct *newsighand;
869                 /*
870                  * This ->sighand is shared with the CLONE_SIGHAND
871                  * but not CLONE_THREAD task, switch to the new one.
872                  */
873                 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
874                 if (!newsighand)
875                         return -ENOMEM;
876
877                 atomic_set(&newsighand->count, 1);
878                 memcpy(newsighand->action, oldsighand->action,
879                        sizeof(newsighand->action));
880
881                 write_lock_irq(&tasklist_lock);
882                 spin_lock(&oldsighand->siglock);
883                 rcu_assign_pointer(tsk->sighand, newsighand);
884                 spin_unlock(&oldsighand->siglock);
885                 write_unlock_irq(&tasklist_lock);
886
887                 __cleanup_sighand(oldsighand);
888         }
889
890         BUG_ON(!thread_group_leader(tsk));
891         return 0;
892 }
893
894 /*
895  * These functions flushes out all traces of the currently running executable
896  * so that a new one can be started
897  */
898 static void flush_old_files(struct files_struct * files)
899 {
900         long j = -1;
901         struct fdtable *fdt;
902
903         spin_lock(&files->file_lock);
904         for (;;) {
905                 unsigned long set, i;
906
907                 j++;
908                 i = j * __NFDBITS;
909                 fdt = files_fdtable(files);
910                 if (i >= fdt->max_fds)
911                         break;
912                 set = fdt->close_on_exec->fds_bits[j];
913                 if (!set)
914                         continue;
915                 fdt->close_on_exec->fds_bits[j] = 0;
916                 spin_unlock(&files->file_lock);
917                 for ( ; set ; i++,set >>= 1) {
918                         if (set & 1) {
919                                 sys_close(i);
920                         }
921                 }
922                 spin_lock(&files->file_lock);
923
924         }
925         spin_unlock(&files->file_lock);
926 }
927
928 char *get_task_comm(char *buf, struct task_struct *tsk)
929 {
930         /* buf must be at least sizeof(tsk->comm) in size */
931         task_lock(tsk);
932         strncpy(buf, tsk->comm, sizeof(tsk->comm));
933         task_unlock(tsk);
934         return buf;
935 }
936
937 void set_task_comm(struct task_struct *tsk, char *buf)
938 {
939         task_lock(tsk);
940         strlcpy(tsk->comm, buf, sizeof(tsk->comm));
941         task_unlock(tsk);
942 }
943
944 int flush_old_exec(struct linux_binprm * bprm)
945 {
946         char * name;
947         int i, ch, retval;
948         struct files_struct *files;
949         char tcomm[sizeof(current->comm)];
950
951         /*
952          * Make sure we have a private signal table and that
953          * we are unassociated from the previous thread group.
954          */
955         retval = de_thread(current);
956         if (retval)
957                 goto out;
958
959         /*
960          * Make sure we have private file handles. Ask the
961          * fork helper to do the work for us and the exit
962          * helper to do the cleanup of the old one.
963          */
964         files = current->files;         /* refcounted so safe to hold */
965         retval = unshare_files();
966         if (retval)
967                 goto out;
968         /*
969          * Release all of the old mmap stuff
970          */
971         retval = exec_mmap(bprm->mm);
972         if (retval)
973                 goto mmap_failed;
974
975         bprm->mm = NULL;                /* We're using it now */
976
977         /* This is the point of no return */
978         put_files_struct(files);
979
980         current->sas_ss_sp = current->sas_ss_size = 0;
981
982         if (current->euid == current->uid && current->egid == current->gid)
983                 set_dumpable(current->mm, 1);
984         else
985                 set_dumpable(current->mm, suid_dumpable);
986
987         name = bprm->filename;
988
989         /* Copies the binary name from after last slash */
990         for (i=0; (ch = *(name++)) != '\0';) {
991                 if (ch == '/')
992                         i = 0; /* overwrite what we wrote */
993                 else
994                         if (i < (sizeof(tcomm) - 1))
995                                 tcomm[i++] = ch;
996         }
997         tcomm[i] = '\0';
998         set_task_comm(current, tcomm);
999
1000         current->flags &= ~PF_RANDOMIZE;
1001         flush_thread();
1002
1003         /* Set the new mm task size. We have to do that late because it may
1004          * depend on TIF_32BIT which is only updated in flush_thread() on
1005          * some architectures like powerpc
1006          */
1007         current->mm->task_size = TASK_SIZE;
1008
1009         if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1010                 suid_keys(current);
1011                 set_dumpable(current->mm, suid_dumpable);
1012                 current->pdeath_signal = 0;
1013         } else if (file_permission(bprm->file, MAY_READ) ||
1014                         (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1015                 suid_keys(current);
1016                 set_dumpable(current->mm, suid_dumpable);
1017         }
1018
1019         /* An exec changes our domain. We are no longer part of the thread
1020            group */
1021
1022         current->self_exec_id++;
1023                         
1024         flush_signal_handlers(current, 0);
1025         flush_old_files(current->files);
1026
1027         return 0;
1028
1029 mmap_failed:
1030         reset_files_struct(current, files);
1031 out:
1032         return retval;
1033 }
1034
1035 EXPORT_SYMBOL(flush_old_exec);
1036
1037 /* 
1038  * Fill the binprm structure from the inode. 
1039  * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1040  */
1041 int prepare_binprm(struct linux_binprm *bprm)
1042 {
1043         int mode;
1044         struct inode * inode = bprm->file->f_path.dentry->d_inode;
1045         int retval;
1046
1047         mode = inode->i_mode;
1048         if (bprm->file->f_op == NULL)
1049                 return -EACCES;
1050
1051         bprm->e_uid = current->euid;
1052         bprm->e_gid = current->egid;
1053
1054         if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1055                 /* Set-uid? */
1056                 if (mode & S_ISUID) {
1057                         current->personality &= ~PER_CLEAR_ON_SETID;
1058                         bprm->e_uid = inode->i_uid;
1059                 }
1060
1061                 /* Set-gid? */
1062                 /*
1063                  * If setgid is set but no group execute bit then this
1064                  * is a candidate for mandatory locking, not a setgid
1065                  * executable.
1066                  */
1067                 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1068                         current->personality &= ~PER_CLEAR_ON_SETID;
1069                         bprm->e_gid = inode->i_gid;
1070                 }
1071         }
1072
1073         /* fill in binprm security blob */
1074         retval = security_bprm_set(bprm);
1075         if (retval)
1076                 return retval;
1077
1078         memset(bprm->buf,0,BINPRM_BUF_SIZE);
1079         return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1080 }
1081
1082 EXPORT_SYMBOL(prepare_binprm);
1083
1084 static int unsafe_exec(struct task_struct *p)
1085 {
1086         int unsafe = 0;
1087         if (p->ptrace & PT_PTRACED) {
1088                 if (p->ptrace & PT_PTRACE_CAP)
1089                         unsafe |= LSM_UNSAFE_PTRACE_CAP;
1090                 else
1091                         unsafe |= LSM_UNSAFE_PTRACE;
1092         }
1093         if (atomic_read(&p->fs->count) > 1 ||
1094             atomic_read(&p->files->count) > 1 ||
1095             atomic_read(&p->sighand->count) > 1)
1096                 unsafe |= LSM_UNSAFE_SHARE;
1097
1098         return unsafe;
1099 }
1100
1101 void compute_creds(struct linux_binprm *bprm)
1102 {
1103         int unsafe;
1104
1105         if (bprm->e_uid != current->uid) {
1106                 suid_keys(current);
1107                 current->pdeath_signal = 0;
1108         }
1109         exec_keys(current);
1110
1111         task_lock(current);
1112         unsafe = unsafe_exec(current);
1113         security_bprm_apply_creds(bprm, unsafe);
1114         task_unlock(current);
1115         security_bprm_post_apply_creds(bprm);
1116 }
1117 EXPORT_SYMBOL(compute_creds);
1118
1119 /*
1120  * Arguments are '\0' separated strings found at the location bprm->p
1121  * points to; chop off the first by relocating brpm->p to right after
1122  * the first '\0' encountered.
1123  */
1124 int remove_arg_zero(struct linux_binprm *bprm)
1125 {
1126         int ret = 0;
1127         unsigned long offset;
1128         char *kaddr;
1129         struct page *page;
1130
1131         if (!bprm->argc)
1132                 return 0;
1133
1134         do {
1135                 offset = bprm->p & ~PAGE_MASK;
1136                 page = get_arg_page(bprm, bprm->p, 0);
1137                 if (!page) {
1138                         ret = -EFAULT;
1139                         goto out;
1140                 }
1141                 kaddr = kmap_atomic(page, KM_USER0);
1142
1143                 for (; offset < PAGE_SIZE && kaddr[offset];
1144                                 offset++, bprm->p++)
1145                         ;
1146
1147                 kunmap_atomic(kaddr, KM_USER0);
1148                 put_arg_page(page);
1149
1150                 if (offset == PAGE_SIZE)
1151                         free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1152         } while (offset == PAGE_SIZE);
1153
1154         bprm->p++;
1155         bprm->argc--;
1156         ret = 0;
1157
1158 out:
1159         return ret;
1160 }
1161 EXPORT_SYMBOL(remove_arg_zero);
1162
1163 /*
1164  * cycle the list of binary formats handler, until one recognizes the image
1165  */
1166 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1167 {
1168         int try,retval;
1169         struct linux_binfmt *fmt;
1170 #if defined(__alpha__) && defined(CONFIG_ARCH_SUPPORTS_AOUT)
1171         /* handle /sbin/loader.. */
1172         {
1173             struct exec * eh = (struct exec *) bprm->buf;
1174
1175             if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1176                 (eh->fh.f_flags & 0x3000) == 0x3000)
1177             {
1178                 struct file * file;
1179                 unsigned long loader;
1180
1181                 allow_write_access(bprm->file);
1182                 fput(bprm->file);
1183                 bprm->file = NULL;
1184
1185                 loader = bprm->vma->vm_end - sizeof(void *);
1186
1187                 file = open_exec("/sbin/loader");
1188                 retval = PTR_ERR(file);
1189                 if (IS_ERR(file))
1190                         return retval;
1191
1192                 /* Remember if the application is TASO.  */
1193                 bprm->sh_bang = eh->ah.entry < 0x100000000UL;
1194
1195                 bprm->file = file;
1196                 bprm->loader = loader;
1197                 retval = prepare_binprm(bprm);
1198                 if (retval<0)
1199                         return retval;
1200                 /* should call search_binary_handler recursively here,
1201                    but it does not matter */
1202             }
1203         }
1204 #endif
1205         retval = security_bprm_check(bprm);
1206         if (retval)
1207                 return retval;
1208
1209         /* kernel module loader fixup */
1210         /* so we don't try to load run modprobe in kernel space. */
1211         set_fs(USER_DS);
1212
1213         retval = audit_bprm(bprm);
1214         if (retval)
1215                 return retval;
1216
1217         retval = -ENOENT;
1218         for (try=0; try<2; try++) {
1219                 read_lock(&binfmt_lock);
1220                 list_for_each_entry(fmt, &formats, lh) {
1221                         int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1222                         if (!fn)
1223                                 continue;
1224                         if (!try_module_get(fmt->module))
1225                                 continue;
1226                         read_unlock(&binfmt_lock);
1227                         retval = fn(bprm, regs);
1228                         if (retval >= 0) {
1229                                 put_binfmt(fmt);
1230                                 allow_write_access(bprm->file);
1231                                 if (bprm->file)
1232                                         fput(bprm->file);
1233                                 bprm->file = NULL;
1234                                 current->did_exec = 1;
1235                                 proc_exec_connector(current);
1236                                 return retval;
1237                         }
1238                         read_lock(&binfmt_lock);
1239                         put_binfmt(fmt);
1240                         if (retval != -ENOEXEC || bprm->mm == NULL)
1241                                 break;
1242                         if (!bprm->file) {
1243                                 read_unlock(&binfmt_lock);
1244                                 return retval;
1245                         }
1246                 }
1247                 read_unlock(&binfmt_lock);
1248                 if (retval != -ENOEXEC || bprm->mm == NULL) {
1249                         break;
1250 #ifdef CONFIG_KMOD
1251                 }else{
1252 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1253                         if (printable(bprm->buf[0]) &&
1254                             printable(bprm->buf[1]) &&
1255                             printable(bprm->buf[2]) &&
1256                             printable(bprm->buf[3]))
1257                                 break; /* -ENOEXEC */
1258                         request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1259 #endif
1260                 }
1261         }
1262         return retval;
1263 }
1264
1265 EXPORT_SYMBOL(search_binary_handler);
1266
1267 /*
1268  * sys_execve() executes a new program.
1269  */
1270 int do_execve(char * filename,
1271         char __user *__user *argv,
1272         char __user *__user *envp,
1273         struct pt_regs * regs)
1274 {
1275         struct linux_binprm *bprm;
1276         struct file *file;
1277         unsigned long env_p;
1278         int retval;
1279
1280         retval = -ENOMEM;
1281         bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1282         if (!bprm)
1283                 goto out_ret;
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         env_p = bprm->p;
1326         retval = copy_strings(bprm->argc, argv, bprm);
1327         if (retval < 0)
1328                 goto out;
1329         bprm->argv_len = env_p - bprm->p;
1330
1331         retval = search_binary_handler(bprm,regs);
1332         if (retval >= 0) {
1333                 /* execve success */
1334                 free_arg_pages(bprm);
1335                 security_bprm_free(bprm);
1336                 acct_update_integrals(current);
1337                 kfree(bprm);
1338                 return retval;
1339         }
1340
1341 out:
1342         free_arg_pages(bprm);
1343         if (bprm->security)
1344                 security_bprm_free(bprm);
1345
1346 out_mm:
1347         if (bprm->mm)
1348                 mmput (bprm->mm);
1349
1350 out_file:
1351         if (bprm->file) {
1352                 allow_write_access(bprm->file);
1353                 fput(bprm->file);
1354         }
1355 out_kfree:
1356         kfree(bprm);
1357
1358 out_ret:
1359         return retval;
1360 }
1361
1362 int set_binfmt(struct linux_binfmt *new)
1363 {
1364         struct linux_binfmt *old = current->binfmt;
1365
1366         if (new) {
1367                 if (!try_module_get(new->module))
1368                         return -1;
1369         }
1370         current->binfmt = new;
1371         if (old)
1372                 module_put(old->module);
1373         return 0;
1374 }
1375
1376 EXPORT_SYMBOL(set_binfmt);
1377
1378 /* format_corename will inspect the pattern parameter, and output a
1379  * name into corename, which must have space for at least
1380  * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1381  */
1382 static int format_corename(char *corename, const char *pattern, long signr)
1383 {
1384         const char *pat_ptr = pattern;
1385         char *out_ptr = corename;
1386         char *const out_end = corename + CORENAME_MAX_SIZE;
1387         int rc;
1388         int pid_in_pattern = 0;
1389         int ispipe = 0;
1390
1391         if (*pattern == '|')
1392                 ispipe = 1;
1393
1394         /* Repeat as long as we have more pattern to process and more output
1395            space */
1396         while (*pat_ptr) {
1397                 if (*pat_ptr != '%') {
1398                         if (out_ptr == out_end)
1399                                 goto out;
1400                         *out_ptr++ = *pat_ptr++;
1401                 } else {
1402                         switch (*++pat_ptr) {
1403                         case 0:
1404                                 goto out;
1405                         /* Double percent, output one percent */
1406                         case '%':
1407                                 if (out_ptr == out_end)
1408                                         goto out;
1409                                 *out_ptr++ = '%';
1410                                 break;
1411                         /* pid */
1412                         case 'p':
1413                                 pid_in_pattern = 1;
1414                                 rc = snprintf(out_ptr, out_end - out_ptr,
1415                                               "%d", task_tgid_vnr(current));
1416                                 if (rc > out_end - out_ptr)
1417                                         goto out;
1418                                 out_ptr += rc;
1419                                 break;
1420                         /* uid */
1421                         case 'u':
1422                                 rc = snprintf(out_ptr, out_end - out_ptr,
1423                                               "%d", current->uid);
1424                                 if (rc > out_end - out_ptr)
1425                                         goto out;
1426                                 out_ptr += rc;
1427                                 break;
1428                         /* gid */
1429                         case 'g':
1430                                 rc = snprintf(out_ptr, out_end - out_ptr,
1431                                               "%d", current->gid);
1432                                 if (rc > out_end - out_ptr)
1433                                         goto out;
1434                                 out_ptr += rc;
1435                                 break;
1436                         /* signal that caused the coredump */
1437                         case 's':
1438                                 rc = snprintf(out_ptr, out_end - out_ptr,
1439                                               "%ld", signr);
1440                                 if (rc > out_end - out_ptr)
1441                                         goto out;
1442                                 out_ptr += rc;
1443                                 break;
1444                         /* UNIX time of coredump */
1445                         case 't': {
1446                                 struct timeval tv;
1447                                 do_gettimeofday(&tv);
1448                                 rc = snprintf(out_ptr, out_end - out_ptr,
1449                                               "%lu", tv.tv_sec);
1450                                 if (rc > out_end - out_ptr)
1451                                         goto out;
1452                                 out_ptr += rc;
1453                                 break;
1454                         }
1455                         /* hostname */
1456                         case 'h':
1457                                 down_read(&uts_sem);
1458                                 rc = snprintf(out_ptr, out_end - out_ptr,
1459                                               "%s", utsname()->nodename);
1460                                 up_read(&uts_sem);
1461                                 if (rc > out_end - out_ptr)
1462                                         goto out;
1463                                 out_ptr += rc;
1464                                 break;
1465                         /* executable */
1466                         case 'e':
1467                                 rc = snprintf(out_ptr, out_end - out_ptr,
1468                                               "%s", current->comm);
1469                                 if (rc > out_end - out_ptr)
1470                                         goto out;
1471                                 out_ptr += rc;
1472                                 break;
1473                         /* core limit size */
1474                         case 'c':
1475                                 rc = snprintf(out_ptr, out_end - out_ptr,
1476                                               "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1477                                 if (rc > out_end - out_ptr)
1478                                         goto out;
1479                                 out_ptr += rc;
1480                                 break;
1481                         default:
1482                                 break;
1483                         }
1484                         ++pat_ptr;
1485                 }
1486         }
1487         /* Backward compatibility with core_uses_pid:
1488          *
1489          * If core_pattern does not include a %p (as is the default)
1490          * and core_uses_pid is set, then .%pid will be appended to
1491          * the filename. Do not do this for piped commands. */
1492         if (!ispipe && !pid_in_pattern
1493             && (core_uses_pid || atomic_read(&current->mm->mm_users) != 1)) {
1494                 rc = snprintf(out_ptr, out_end - out_ptr,
1495                               ".%d", task_tgid_vnr(current));
1496                 if (rc > out_end - out_ptr)
1497                         goto out;
1498                 out_ptr += rc;
1499         }
1500 out:
1501         *out_ptr = 0;
1502         return ispipe;
1503 }
1504
1505 static void zap_process(struct task_struct *start)
1506 {
1507         struct task_struct *t;
1508
1509         start->signal->flags = SIGNAL_GROUP_EXIT;
1510         start->signal->group_stop_count = 0;
1511
1512         t = start;
1513         do {
1514                 if (t != current && t->mm) {
1515                         t->mm->core_waiters++;
1516                         sigaddset(&t->pending.signal, SIGKILL);
1517                         signal_wake_up(t, 1);
1518                 }
1519         } while ((t = next_thread(t)) != start);
1520 }
1521
1522 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1523                                 int exit_code)
1524 {
1525         struct task_struct *g, *p;
1526         unsigned long flags;
1527         int err = -EAGAIN;
1528
1529         spin_lock_irq(&tsk->sighand->siglock);
1530         if (!signal_group_exit(tsk->signal)) {
1531                 tsk->signal->group_exit_code = exit_code;
1532                 zap_process(tsk);
1533                 err = 0;
1534         }
1535         spin_unlock_irq(&tsk->sighand->siglock);
1536         if (err)
1537                 return err;
1538
1539         if (atomic_read(&mm->mm_users) == mm->core_waiters + 1)
1540                 goto done;
1541
1542         rcu_read_lock();
1543         for_each_process(g) {
1544                 if (g == tsk->group_leader)
1545                         continue;
1546
1547                 p = g;
1548                 do {
1549                         if (p->mm) {
1550                                 if (p->mm == mm) {
1551                                         /*
1552                                          * p->sighand can't disappear, but
1553                                          * may be changed by de_thread()
1554                                          */
1555                                         lock_task_sighand(p, &flags);
1556                                         zap_process(p);
1557                                         unlock_task_sighand(p, &flags);
1558                                 }
1559                                 break;
1560                         }
1561                 } while ((p = next_thread(p)) != g);
1562         }
1563         rcu_read_unlock();
1564 done:
1565         return mm->core_waiters;
1566 }
1567
1568 static int coredump_wait(int exit_code)
1569 {
1570         struct task_struct *tsk = current;
1571         struct mm_struct *mm = tsk->mm;
1572         struct completion startup_done;
1573         struct completion *vfork_done;
1574         int core_waiters;
1575
1576         init_completion(&mm->core_done);
1577         init_completion(&startup_done);
1578         mm->core_startup_done = &startup_done;
1579
1580         core_waiters = zap_threads(tsk, mm, exit_code);
1581         up_write(&mm->mmap_sem);
1582
1583         if (unlikely(core_waiters < 0))
1584                 goto fail;
1585
1586         /*
1587          * Make sure nobody is waiting for us to release the VM,
1588          * otherwise we can deadlock when we wait on each other
1589          */
1590         vfork_done = tsk->vfork_done;
1591         if (vfork_done) {
1592                 tsk->vfork_done = NULL;
1593                 complete(vfork_done);
1594         }
1595
1596         if (core_waiters)
1597                 wait_for_completion(&startup_done);
1598 fail:
1599         BUG_ON(mm->core_waiters);
1600         return core_waiters;
1601 }
1602
1603 /*
1604  * set_dumpable converts traditional three-value dumpable to two flags and
1605  * stores them into mm->flags.  It modifies lower two bits of mm->flags, but
1606  * these bits are not changed atomically.  So get_dumpable can observe the
1607  * intermediate state.  To avoid doing unexpected behavior, get get_dumpable
1608  * return either old dumpable or new one by paying attention to the order of
1609  * modifying the bits.
1610  *
1611  * dumpable |   mm->flags (binary)
1612  * old  new | initial interim  final
1613  * ---------+-----------------------
1614  *  0    1  |   00      01      01
1615  *  0    2  |   00      10(*)   11
1616  *  1    0  |   01      00      00
1617  *  1    2  |   01      11      11
1618  *  2    0  |   11      10(*)   00
1619  *  2    1  |   11      11      01
1620  *
1621  * (*) get_dumpable regards interim value of 10 as 11.
1622  */
1623 void set_dumpable(struct mm_struct *mm, int value)
1624 {
1625         switch (value) {
1626         case 0:
1627                 clear_bit(MMF_DUMPABLE, &mm->flags);
1628                 smp_wmb();
1629                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1630                 break;
1631         case 1:
1632                 set_bit(MMF_DUMPABLE, &mm->flags);
1633                 smp_wmb();
1634                 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1635                 break;
1636         case 2:
1637                 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1638                 smp_wmb();
1639                 set_bit(MMF_DUMPABLE, &mm->flags);
1640                 break;
1641         }
1642 }
1643
1644 int get_dumpable(struct mm_struct *mm)
1645 {
1646         int ret;
1647
1648         ret = mm->flags & 0x3;
1649         return (ret >= 2) ? 2 : ret;
1650 }
1651
1652 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1653 {
1654         char corename[CORENAME_MAX_SIZE + 1];
1655         struct mm_struct *mm = current->mm;
1656         struct linux_binfmt * binfmt;
1657         struct inode * inode;
1658         struct file * file;
1659         int retval = 0;
1660         int fsuid = current->fsuid;
1661         int flag = 0;
1662         int ispipe = 0;
1663         unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1664         char **helper_argv = NULL;
1665         int helper_argc = 0;
1666         char *delimit;
1667
1668         audit_core_dumps(signr);
1669
1670         binfmt = current->binfmt;
1671         if (!binfmt || !binfmt->core_dump)
1672                 goto fail;
1673         down_write(&mm->mmap_sem);
1674         /*
1675          * If another thread got here first, or we are not dumpable, bail out.
1676          */
1677         if (mm->core_waiters || !get_dumpable(mm)) {
1678                 up_write(&mm->mmap_sem);
1679                 goto fail;
1680         }
1681
1682         /*
1683          *      We cannot trust fsuid as being the "true" uid of the
1684          *      process nor do we know its entire history. We only know it
1685          *      was tainted so we dump it as root in mode 2.
1686          */
1687         if (get_dumpable(mm) == 2) {    /* Setuid core dump mode */
1688                 flag = O_EXCL;          /* Stop rewrite attacks */
1689                 current->fsuid = 0;     /* Dump root private */
1690         }
1691
1692         retval = coredump_wait(exit_code);
1693         if (retval < 0)
1694                 goto fail;
1695
1696         /*
1697          * Clear any false indication of pending signals that might
1698          * be seen by the filesystem code called to write the core file.
1699          */
1700         clear_thread_flag(TIF_SIGPENDING);
1701
1702         /*
1703          * lock_kernel() because format_corename() is controlled by sysctl, which
1704          * uses lock_kernel()
1705          */
1706         lock_kernel();
1707         ispipe = format_corename(corename, core_pattern, signr);
1708         unlock_kernel();
1709         /*
1710          * Don't bother to check the RLIMIT_CORE value if core_pattern points
1711          * to a pipe.  Since we're not writing directly to the filesystem
1712          * RLIMIT_CORE doesn't really apply, as no actual core file will be
1713          * created unless the pipe reader choses to write out the core file
1714          * at which point file size limits and permissions will be imposed
1715          * as it does with any other process
1716          */
1717         if ((!ispipe) && (core_limit < binfmt->min_coredump))
1718                 goto fail_unlock;
1719
1720         if (ispipe) {
1721                 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1722                 /* Terminate the string before the first option */
1723                 delimit = strchr(corename, ' ');
1724                 if (delimit)
1725                         *delimit = '\0';
1726                 delimit = strrchr(helper_argv[0], '/');
1727                 if (delimit)
1728                         delimit++;
1729                 else
1730                         delimit = helper_argv[0];
1731                 if (!strcmp(delimit, current->comm)) {
1732                         printk(KERN_NOTICE "Recursive core dump detected, "
1733                                         "aborting\n");
1734                         goto fail_unlock;
1735                 }
1736
1737                 core_limit = RLIM_INFINITY;
1738
1739                 /* SIGPIPE can happen, but it's just never processed */
1740                 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1741                                 &file)) {
1742                         printk(KERN_INFO "Core dump to %s pipe failed\n",
1743                                corename);
1744                         goto fail_unlock;
1745                 }
1746         } else
1747                 file = filp_open(corename,
1748                                  O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1749                                  0600);
1750         if (IS_ERR(file))
1751                 goto fail_unlock;
1752         inode = file->f_path.dentry->d_inode;
1753         if (inode->i_nlink > 1)
1754                 goto close_fail;        /* multiple links - don't dump */
1755         if (!ispipe && d_unhashed(file->f_path.dentry))
1756                 goto close_fail;
1757
1758         /* AK: actually i see no reason to not allow this for named pipes etc.,
1759            but keep the previous behaviour for now. */
1760         if (!ispipe && !S_ISREG(inode->i_mode))
1761                 goto close_fail;
1762         /*
1763          * Dont allow local users get cute and trick others to coredump
1764          * into their pre-created files:
1765          */
1766         if (inode->i_uid != current->fsuid)
1767                 goto close_fail;
1768         if (!file->f_op)
1769                 goto close_fail;
1770         if (!file->f_op->write)
1771                 goto close_fail;
1772         if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1773                 goto close_fail;
1774
1775         retval = binfmt->core_dump(signr, regs, file, core_limit);
1776
1777         if (retval)
1778                 current->signal->group_exit_code |= 0x80;
1779 close_fail:
1780         filp_close(file, NULL);
1781 fail_unlock:
1782         if (helper_argv)
1783                 argv_free(helper_argv);
1784
1785         current->fsuid = fsuid;
1786         complete_all(&mm->core_done);
1787 fail:
1788         return retval;
1789 }